Dissertation presented at Uppsala University to be publicly examined in B115, Ekonomikum,
Kyrkogårdsgatan 10 B, Uppsala, Monday, 16 December 2013 at 10:15 for the degree of
Doctor of Philosophy. The examination will be conducted in English. Faculty examiner:
Professor Matthew Lindqvist (Swedish Institute for Social Research (SOFI), Stockholm
University).
Abstract
Erixson, O. 2013. Economic Decisions and Social Norms in Life and Death Situations.
Economic studies 141. 180 pp. Uppsala: Nationalekonomiska institutionen, Uppsala
universitet. ISBN 978-91-85519-48-4.
Essay 1: (with Mikael Elinder) Since the sinking of the Titanic, there has been a widespread
belief that the social norm of “women and children first” (WCF) gives women a survival
advantage over men in maritime disasters, and that captains and crew members give priority to
passengers. We analyze a database of 18 maritime disasters spanning three centuries, covering
the fate of over 15,000 individuals of more than 30 nationalities. Our results provide a unique
picture of maritime disasters. Women have a distinct survival disadvantage compared with men.
Captains and crew survive at a significantly higher rate than passengers. We also find that: the
captain has the power to enforce normative behavior; there seems to be no association between
duration of a disaster and the impact of social norms; women fare no better when they constitute
a small share of the ship’s complement; the length of the voyage before the disaster appears to
have no impact on women’s relative survival rate; the sex gap in survival rates has declined since
World War I; and women have a larger disadvantage in British shipwrecks. Taken together, our
findings show that human behavior in life-and-death situations is best captured by the expression
“every man for himself.”
Essay 2: (with Henry Ohlsson) The objective of this essay is to study to what extent parents
divide their estates unequally between their children and the determinants of this decision. We
use a new dataset based on the estate reports for almost 70,000 Swedish widows, widowers,
divorcees and unmarried individuals who died with positive estates and at least two children.
Unequal sharing is unusual; depending on definitions only 2–12 percent of the estates are
unequally divided. Previous studies for other countries, particularly from the US, find that
around 20–40 percent of parents divide their estates unequally. We argue that the relatively low
frequency of unequal sharing in Sweden might be explained by contextual factors such as the
inheritance law, the transfer tax system, the income distribution, and the welfare state. We also
estimate models with family fixed effects to study how the characteristics of children to parents
who choose unequal division affect the size of the transfer. The empirical estimates show that
bequests are not used to compensate for income differences between children, suggesting that
bequests are not guided by altruistic motives. Children who are likely to have provided services
to the parent receive more than their siblings however. This suggests that, at least some bequests
are guided by exchange motives.
Essay 3: (with Mikael Elinder and Henry Ohlsson) The objective of this essay is to study
when and how much labor and capital income of heirs respond to inheritances. We estimate
fixed effects models following direct heirs, inheriting in 2004, during the years 2000–2008 using
Swedish panel data. Our first main result is that the more the heir inherits, the lower her labor
income becomes. This labor income effect appears in the years after the heir had inherited and
is stronger for old heirs than for young heirs. We also find evidence of anticipation effects that
occur before the actual transfer. Our second main result is that the more the heir inherits, the
higher her capital income becomes. This effect only appears in the years after receiving the
inheritance. It seems to be dissipating after a couple of years.
Essay 4: This essay contributes in two ways to the literature on the effects of economic
circumstances on health. First, it deals with reverse causality and omitted variable bias by
exploiting previously exogenous variation in inherited wealth generated by the unexpected
repeal of the Swedish inheritance tax. Second, it analyzes responses in health outcomes from
administrative registers. The results show that increased wealth has limited impacts on objective
adult health over a period of six years. This is in line with what has been documented previously
regarding subjective health outcomes. If anything, it appears as if the wealth shock resulting
from the tax reform leads people to seek care for symptoms of disease, which result in that cancer
is detected and possibly treated earlier. One possible explanation for this preventive response
is that good health is needed for enjoying the improved consumption prospects generated by
the wealth shock.
Keywords: mortality, social norms, dicrimination, maritime disasters, altruism, estate division,
wills, equal sharing, bequests motives, inheritances, windfall gains, tax reform, labor income,
capital income, health
Oscar Erixson, Department of Economics, Box 513, Uppsala University, SE-75120 Uppsala,
Sweden.
© Oscar Erixson 2013
ISSN 0283-7668
ISBN 978-91-85519-48-4
urn:nbn:se:uu:diva-209981 (http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-209981)
To Didi
Acknowledgements
Writing a thesis in economics is surely not a simple task. In my case it would
not have been possible without the great support from colleagues, friends
and family.
First of all I want to send my deepest thanks to my main advisor Henry
Ohlsson. Henry, your wisdom, support, encouragement and not least, your
optimism has been invaluable to me. I am so glad that you did such a great
job in promoting bequest research! I also want to send my sincere gratitude
to my assistant advisor Mikael Elinder. Micke, you have been my knight in
shining armor in many life and death situations. Your great knowledge about
a pretty much everything between heaven and earth is truly remarkable and I
am honored that you have shared it with me.
I want to extend my thanks to Jukka Pirttilä and Katarina Nordblom for
their very helpful comments at my licentiate and final seminars. I am also
indebted to Wojciech Kopczuk for inviting me to Columbia University during my third year of the PhD program. Moreover, Håkan Selin and Mikael
Lindahl deserve acknowledgement for sharing their great understanding on
important aspects of empirical research. Sven-Åke Carlsson and Per Johansson should receive many thanks for opening my eyes to economic writing
during my undergraduate studies. I also want to thank Marcus Vingren at
SCB for helping me with various data issues and the Jan Wallander and Tom
Hedelius Foundations for their generous financial support throughout my
PhD studies.
It feels like I have spent an entire lifetime at Neken. Although I have to
admit that, walking them stairs up to the fourth floor has not always been
utility enhancing (especially not in the middle of night) I want to emphasize
that my life has been truly interesting and inspiring, and importantly, in very
good hands. There is no doubt that social norms of helping behavior and
friendship dominates at the department and I am indebted to all of you who
prove this every day.
My special thanks go to my roomie and homie Erik Spector. Having
Spector behind my back has never felt scary (although my snarling may have
been a good reason) but rather enjoyable and interesting, especially due to
his remarkable understanding for the difficult questions in economics. I also
want to thank Susanne for our friendship and for making my life more joyful. Sebastian should receive many thanks, not only for being a good friend,
but also for being a great companion during endless merging and appending
exercises.
My gratitude is also directed to Anna, Arizo, Karolina Mattias, and Martin for being the best peers. We have done a great journey together and I
hope it will keep us connected for many years to come! Moreover, Adrian,
Spencer, and Mattias Nordin deserve big thanks, not only for providing me
with excellent answers to research related questions, but also for being great
friends and especially, for always laughing to my latest puns. I also want to
thank Grip for being an inspiring friend who over and over again proves that
nothing is impossible. Jonas and Teodora deserve acknowledgment for keeping my office chair and my apartment warm when I was in the States. My
thanks are also directed to Tove and Lovisa for being great co-organizers of
the UCFS Brown Bag seminar series. Gabriella, Jon, Glenn, Vesna, Pia,
Jonas K, Niklas, Che, Ulrika, and Heléne have also contributed to this work
by making my seemingly boring pasta lunches so much better. I am also
thankful to Johannes, Per, Christian, Matz, and Laurent for sharing my passion for endurance sports. Olle should feel proud of himself for being a great
friend in the City. I would also like to thank Tomas and Javad for showing
me what excellent teaching really is. Katarina, Åke, Ann-Sofie, Nina, Berit,
P-A, and Stina have kindly provided me with superb help in all types of
practical matters, for which I am very thankful. I am also grateful to Luca for
our discussions on everything from childcare to hiking in the Dolomites, to
Jan for keeping me updated on the situation north of Orsasjön, and to Daniel
for his advices regarding my present and future doings.
My childhood friends Erik, Andreas, Danne, Johan, Per, Fredrik, and Jens
deserve a special mention. Thank you guys for always being by my side! I
am also grateful to my dear NYC neighbors Lucia and Dominic for their
support and friendship.
To my beloved family. Your everlasting support means everything to me.
Thank you. I also want to send a special thank to my father Gösta for his
incredible fighting spirit.
And finally, to Maria, my inamorata and life partner. I cannot describe
how highly I value all you done for me, just take my word that you are the
best thing that ever happened to me!
Uppsala, October 2013
Oscar Erixson
Contents
Introduction ................................................................................................... 13
1 Human behavior in life-and-death situations ........................................ 13
2 Financial decision making in the terminal stage of life ......................... 15
3 Consequences of inherited wealth ......................................................... 16
3.1 The Nobel and Carnegie conjecture............................................... 17
3.2 The wealth-health nexus ................................................................ 18
References ................................................................................................ 20
Essay 1: Gender, social norms, and survival in maritime disasters .............. 23
Essay 2: Estate division: Equal sharing as choice, social norm, and legal
requirement ................................................................................................... 57
1 Introduction ........................................................................................... 58
2 Data and descriptive facts...................................................................... 62
2.1 Data ................................................................................................ 62
2.2 The parents and the estates ............................................................ 64
2.3 The children and the inheritances .................................................. 67
3 Descriptive evidence on the frequency of unequal sharing ................... 68
4 Econometric evidence ........................................................................... 73
4.1 The probability of writing wills ..................................................... 73
4.2 The probability of unequal sharing ................................................ 78
5 Inherited amounts and the characteristics of heirs ................................ 87
6 Discussion and concluding remarks ...................................................... 95
Appendix .................................................................................................. 98
Appendix A: Descriptive statistics ...................................................... 98
References .............................................................................................. 100
Essay 3: The impact of inheritances on heirs’ labor and capital income .... 103
Essay 4: Health responses to a wealth shock: Evidence from a Swedish
tax reform .................................................................................................... 141
1 Introduction ......................................................................................... 142
2 Review of related literature ................................................................. 145
2.1 Theoretical arguments for causal effects of wealth on health...... 145
2.2 Findings in the previous literature ............................................... 146
3 The Swedish inheritance tax and how it was unexpectedly repealed .. 149
3.1 Taxation on inheritances before the reform ................................. 149
3.2 The unexpected reform ................................................................ 149
4 Data ..................................................................................................... 151
4.1 The sample and approximation of tax status ............................... 151
4.2 Health outcomes .......................................................................... 153
5 Empirical strategies ............................................................................. 156
6 Exogeneity of the wealth shock and test of identifying assumptions .. 159
6.1 Test for differences in pre-determined characteristics between
treated and controls ............................................................................ 160
6.2 Test for parallel trends in health .................................................. 161
7 Results ................................................................................................. 163
7.1 The effect of the wealth shock on Hospitalization ...................... 163
7.2 The effect of the wealth shock on Sick leave and Mortality........ 169
8 Concluding discussion......................................................................... 170
Appendix A: Description of diagnose variables..................................... 172
Appendix B: Cross-sectional evidence of the wealth-health gradient .... 174
Appendix C: Sample distribution of wealth shock, treated subjects,
Main sample ........................................................................................... 175
Appendix D: Sample characteristics, Placebo sample. .......................... 176
Appendix E: DID estimates, heterogeneous effects, Hospitalization..... 177
Appendix F: DID estimates of the effect of the wealth shock on Sick
leave and Mortality................................................................................. 179
10 References ......................................................................................... 180
Introduction
This thesis consists of four self-contained essays, all of which are in some
way related to death. Many people would probably consider this morbid, but
the fact is that death – although a very tragic event – can contribute to our
knowledge about human behavior in a variety of contexts. The four essays
capture both decision-making in the terminal stage of life and some of the
consequences that death may have for the survivors. The first essay (1) investigates to what extent behavior is governed by social norms in life-anddeath situations. The second essay (2) investigates the decisions parents
make regarding their financial assets before they die and how these decisions
affect the distribution of bequests among their offspring. The two final essays (3 and 4) investigate how the bequests impact economic behavior and
health of the receiving offspring.
1 Human behavior in life-and-death situations
Imagine the following. It is close to midnight and you are sitting in the bar
onboard a cruise ship, listening to the band. Suddenly, you hear a loud bang
and the ground shakes. It gets dark and the ship takes on a heavy list. An
alarm starts and all passengers are told to head for the lifeboats. You praise
yourself for not having gone to bed. From the bar, it is close to the boat deck
where the lifeboats are located. When you reach the deck, you see how the
lifeboats fill up quickly. The heavy list only allows for loading the lifeboats
on one side of the ship. It is dark out and big waves are breaking against the
ship’s hull. The water is freezing. You estimate that the ship will sink in just
a few minutes. A crewmember announces that only one more lifeboat will be
lowered. Luckily, you find yourself standing next to it. Behind you men and
women, boys and girls are striving to reach the lifeboat. What would you
1
do?
Economists tend to make the assumption that people are rational and act
according to their best self-interest. In the situation described above, there is
an immense threat to life, and helping other people will most likely reduce
1
This depiction is largely inspired by the introduction (in Swedish) in Elinder and Erixson
(2012).
13
your survival chances even more. It is not even certain whether or not those
who are helped will survive in the end. It would therefore be rational for
most people to save themselves rather than helping others.
At the same time we know from empirical studies that social norms are
important and that people indeed do engage in behaviors which are not always in their best self-interest (Elster, 1989). It has even been argued that
social norms are upheld also in life-and-death situations, such as maritime
disasters, where the cost of compliance is substantial (Frey et al., 2010).
Traditionally, the assumption has been that women and children are saved
first in shipwrecks (Delap 2006). Men are expected to give women and children priority to the lifeboats and, similarly, captain and crew are expected to
rescue the passengers before they save themselves. By studying the survival
patterns from the sinking of the Titanic, researchers have documented that
women and children survive at a higher degree than men, and that the passengers have a higher survival probability than the crew (see for example
Hall, 1986 and Frey et al., 2010).
In Essay 1 (with Mikael Elinder), we contrast the general view on the upholding of social norms in maritime disasters. Our analysis of survival patterns from 18 of the most notable shipwrecks over the last three centuries
show that the survival rate of women is, on average, only about half that of
men. Given that men have an advantage over women in characteristics essential for survival in shipwrecks - physical strength, competitiveness, etc. this finding suggests that men have not complied with the ’women and children first’ protocol. Instead, they have tried to save themselves. Our analysis
also shows that crew and captains have substantially better survival prospects than passengers. Due to the fact that the crew is likely to have an a
priori survival advantage over the passengers since they are familiar with the
ship, often have emergency training and are likely to receive early information about the severity of the situation, this finding indicates that they have
saved themselves instead of having followed their procedures.
We interpret these results as evidence that compliance with the “women
and children first” norm is the exception in maritime disasters and that human behavior in these kind of situations is better characterized by the expression “every man for himself”. Individual ability seems to be the most
important determinant of survival.
The essay contributes to our general knowledge on human behavior by
showing that the simple model of a selfish homo economicus dominates social norms in life-and-death situations. Even though the essay focuses on
maritime disasters, it should be remembered that disasters involving large
number of people, like earthquakes, tsunamis and terrorist attacks, occur
regularly. The results indicate what we should expect regarding human behavior in these types of disasters as well. This should be of particular interest
for government agencies and rescue organizations which plan and carry out
evacuations.
14
2 Financial decision making in the terminal stage of life
You are 80 years old. Life is coming to an end. You live in the house that you
and your late husband bought 40 years ago. The consistent rise in real estate
prices over the last decades has made it worth quite a lot. You also have
savings in the bank. Now you are thinking about how you want these assets
to be distributed between your son and your daughter when you die. Your
daughter has assisted you considerably in the last years and you think it is
reasonable that she gets financial compensation for her efforts. Establishing
a will stipulating a larger bequest to your daughter would, however, imply
that your son will receive less. This may make him feel disfavored. What
would you do?
Economists’ interest in how people make decisions about their financial
assets in the terminal stage of life has primarily been driven by their desire to
explain the motivations underlying transfers between generations. Knowledge about transfer motives is important for understanding how policies,
such as intergenerational government redistribution and taxation, affect savings and wealth accumulation and also whether private transfers equalize or
enhance inequalities within generations. That people do have deliberate motive for their transfers is generally acknowledged. Bequests constitute too
large a share of wealth in society to simply be the result of peoples’ inability
to consume all their wealth before they die (see for example Kotlikoff and
Summers, 1981 and Kotlikoff, 1988). Also, the existence of estate planning
and avoidance behavior with respect to inheritance and estate taxation suggests that transfers are intended (Joulfaian, 2004; Nordblom and Ohlsson,
2006; Kopczuk, 2007; Eliasson and Ohlsson, 2010).
In particular, there are two models explaining why people leave bequests:
the altruistic model (Barro, 1974; Becker, 1974) and the exchange model
(Cox, 1987). The altruistic model is predicated on the idea that the parent
leaves bequests because he or she cares about the welfare of his/her children.
The parent is assumed to transfer more to the less well-off children in order
to equalize consumption possibilities within the family. The exchange
model, on the other hand, assumes that bequests are payments (quid pro quo)
for services provided by the children to the parent. In an exchange regime,
the parent transfers larger amounts to children who have provided them with
more services. Accordingly, as long as there is heterogeneity across children
in economic circumstances or in care provision, these two models predict
that the estate will be unequally divided.
In Essay 2 (with Henry Ohlsson), we explore to what extent parents divide their estates unequally between their children, in addition to whether or
not individual characteristics determine if heirs receive more or less than
their siblings. For this purpose, we use a new administrative dataset on the
15
estate reports for 70,000 deceased Swedish parents with positive estates and
two or more children.
We find that only 2–12 percent of the parents divide their estates unequally between their children. For the few estates where the amounts differ
between the children, the children’s economic position does not significantly
affect the differences in inherited amounts. This suggests that parents do not
use transfers at death to equalize differences in consumption possibilities
within the family. We do find, however, that heirs who are more likely to
have provided services to the parent receive larger bequests than their siblings. This could be interpreted as if, at least for some parents, post mortem
transfers are motivated by exchange.
This is not the first study to reject the two traditional transfer theories on
the basis of sharing patterns. Several studies, particularly from the US, have
also reported that that equal sharing is the norm (see for example Menchik,
1980, Wilhelm, 1996). A noticeable difference between our results and the
results in the earlier literature lies in the extent to which parents adhere to
equal sharing. Previous studies commonly report frequencies of unequal
sharing in the interval of 20–40 percent. We propose that differences between the countries in contextual factors, such as the inheritance law, the tax
treatment of transfers, the income distribution and the welfare state, may
explain the discrepancy.
3 Consequences of inherited wealth
Your mother passed away two months ago. There has been a lot to do since
then: arranging the funeral, selling your mother’s apartment and sorting
through all her belongings. However, you and your brother hired a lawyer
to establish the estate inventory report and now the two of you are at the law
firm to divide the property. It turns out that your mother had a will. It stipulates that your brother should inherit the summer house. This seems fair as
he has been the one taking care of it during the last decade. The will stipulates, however, that you should receive the cash equivalent of the value of
the house. You suddenly realize that this is a substantial amount. What
would you do with the money?
Economists are interested in the answers to this question, since knowledge
about behavioral effects of improved financial resources is essential for a
wide range of economic questions. The effects of tax cuts, transfers payments and other policies leaving people with more money in their pockets
depend on how increases in wealth affect labor supply, consumption and
savings decisions. How people use their newly obtained wealth could also
inform us about the most efficient strategies to tackle inequalities in society.
Cash grants, for example, could be considered a reasonable strategy to re-
16
duce the observed socioeconomic disparities in health if improved wealth in
fact leads people to engage in health enhancing behaviors.
The final two essays (Essay 3 and Essay 4) are natural extensions of Essay 2 in that they consider the impacts bequests have on the recipients. In
Essay 3 (Section 3.1), the impacts on labor supply and savings decisions are
investigated, whereas the health consequences are investigated in Essay 4
(Section 3.2).
Inheritances are good for studying how improved financial resources affect economic behavior and well-being, because, regardless of the motivation for the transfer, they often greatly affect the recipient’s wealth. Unlike
other wealth shocks, such as lottery winnings and stock market gains, inheritances are also widespread in the population. One must keep in mind, however, that it is not straightforward to interpret behavioral responses to inheritances as wealth effects. There are, in particular, two issues which need to be
taken into account. First, the inheritance is likely to be correlated with individual characteristics which are themselves correlated with the outcome
variable, implying that what appears to be a wealth effect may just be a spurious correlation. Second, the inheritance may be expected. Failure to distinguish between inheritances which are expected and those which are not introduces a potential downward bias in estimates of wealth effects, if behavioral adjustments to an expected inheritance have taken place already before
the receipt. Different empirical strategies to address these methodological
challenges are used in Essay 3 and Essay 4 respectively.
3.1 The Nobel and Carnegie conjecture
“Experience has taught me that great fortunes acquired by inheritance never
bring happiness, they only dull the faculties. Any man possessing a large fortune ought not to leave more than a small part of it to his heirs not even his
direct heirs - just enough to make their way in the world.” [Swedish inventor
and industrialist Alfred Nobel, in Chester 1998, p. 31]
“The parent who leaves his son enormous wealth generally deadens the talents and energies of the son, and tempts him to lead a less useful and less
worthy life than he otherwise would . . .” [American industrialist Andrew
Carnegie, in Carnegie 1962, p. 56]
In Essay 3 (with Mikael Elinder and Henry Ohlsson), we test the merit of the
conjecture that inheritances depress work effort and encourage spendthrift
behavior by investigating how inheritances affect the recipients’ labor and
capital incomes.
The essay contributes to the previous literature on labor supply and consumption responses to inheritances, as well as to a more general literature on
the marginal propensity to earn and consume out of wealth (see for example
17
Holtz-Eakin et al., 1993, Poterba, 2000, and Imbens et al., 2001). This is
achieved by using Swedish register based panel data to estimate the dynamics of responses over a longer time period than what has previously been
done. Empirical estimates of the wealth effects are obtained from individual
fixed effects models, exploiting variation in inherited amounts across heirs
of deceased parents who died in Stockholm in 2004.
We find that inheritances, amounting to about one average yearly salary,
have persistent and considerable negative effects on labor income. The estimates suggest that labor income decreases by an amount corresponding to 4–
9 percent of the wealth increase. This is larger than what has been reported
in previous studies and contrasts with the belief that wealth shocks need to
be significant to overcome frictions that may intrude on labor supply decisions (see for example Blundell and MaCurdy 1999). Another finding is that
capital income increases substantially in the years immediately after inheriting. This suggests that capital gains on inherited assets have been realized.
Taken together, our estimates indicate that the heirs make themselves better
off in terms of leisure as well as consumption possibilities, suggesting that
the fears of Nobel and Carnegie are valid even when it comes to comparatively small inheritances.
The empirical estimates in the essay are likely to be of interest to policy
makers who want to account for behavioral responses when designing optimal estate or inheritance tax schedules. One important implication is that
inheritance taxes are likely to increase revenue from labor income taxes
whereas revenue from capital taxes may decrease.
3.2 The wealth-health nexus
Does higher wealth leads to better health? A substantial number of studies
report evidence of a large positive association between many measures of
economic resources, including wealth and income, and most measures of
health (see for example Marmot, 1999, and Smith, 1999). These crosssectional associations, however, seem to be less persuasive than they first
appear. Empirical estimates from a growing literature which attempts to
isolate the causal effects of wealth on health with quasi-experimental techniques suggest that wealth has a limited impact on health, commonly measured by self-reported general health status, and mortality. This has led to the
conclusion that there is no causal effect of wealth on adult health and that the
cross-sectional association is more likely to be due to other factors, such as
economic circumstances in childhood or education (see for example Deaton,
2003 and Cutler et al., 2011).
Nevertheless, in order to draw any conclusions regarding, for example,
the most effective strategy to reduce socioeconomic disparities in health, and
whether or not tax and welfare policies need to factor in their potential con-
18
sequences for population health, it is important to understand how wealth
affects other aspects of health.
In Essay 4, I investigate how increased wealth affects objective health
outcomes commonly found in administrative registers. As far as I am aware,
this essay is the first study that uses objective measures of health other than
mortality for this purpose. To identify the casual effect of wealth on health, I
exploit the fact that the unexpected repeal of the Swedish inheritance tax
generated a windfall inheritance, corresponding to around seven percent of
initial wealth, for heirs receiving taxable inheritance from parents who died
after the reform, but not for comparable heirs who inherited prior to the reform.
The empirical analysis shows that the positive wealth shock resulting
from the tax reform increases the likelihood of hospital admission by five
percent over a period of six years. Since health care in Sweden is universal,
this finding may at a first be interpreted as if the wealth shock is harmful to
health. An investigation of responses in the diagnoses underlying the hospital admissions suggests, however, that the increase in hospitalization may be
the result of preventative actions against future morbidity. This behavioral
response could potentially be explained by people seeking good health to
fully benefit from improved future consumption prospects generated by the
wealth shock. The conclusion that the wealth shock has limited consequences for individual objective health is further confirmed by statistically
insignificant wealth effects on health outcomes that are likely to capture
health events which are both less severe (publically insured sick leave) and
more severe (mortality) than those resulting in hospital admissions.
The results in the essay should be of interest to policy makers, since they
suggest that wealth changes which may be expected from tax reforms of
similar magnitudes as the repeal of the Swedish inheritance tax are unlikely
to have any short or medium run consequences for health. The results, moreover, suggest that policies targeted at reducing socioeconomic inequalities in
health are likely to be more usefully channeled toward interventions that
directly improve health.
19
References
Barro, R. J. (1974). “Are government bonds net wealth?” Journal of Political Economy, 82, 1095–1117.
Becker, G. S. (1974). “A theory of social interactions,” Journal of Political Economy, 82, 1063–1093.
Blundell, R. and T. MaCurdy (1999). “Labor supply: a review of alternative approaches,” in O. Ashenfelter and D. Card (eds.), Handbook of Labor Economics. North-Holland.
Carnegie, A. (1891/1962). “The advantages of poverty,” in E. C. Kirkland (ed.), The
Gospel Wealth and other Timely Essays. The Belknap Press of Harvard University Press.
Chester, R. (1998). “Inheritance in American legal thought,” in R. Miller and S.
McNamee (eds.), Inheritance and Wealth in America. Plenum Press.
Cox, D. (1987). “Motives for private income transfers,” Journal of Political Economy, 95, 508–546.
Cutler, D. M., A. Deaton, and A. Lleras-Muney (2011). “Socioeconomic status and
health: dimensions and mechanisms,” in S. Glied and P. C. Smith (eds.), The
Oxford Handbook of Health Economics. Oxford University Press.
Deaton, A. (2003). “Health, inequality, and economic development,” Journal of
Economic Literature, 41, 113–158.
Delap, L. (2006). “’Thus does man prove his fitness to be the master of things’:
Shipwrecks, chivalry and masculinities in nineteenth- and twentieth-century
Britain” Cultural and Social History, 3, 45–74.
Eliason, M. and H. Ohlsson (2010). “Timing of death and the repeal of the Swedish
inheritance tax,” Working Paper 2010:2, Uppsala Center for Fiscal Studies, Department of Economics, Uppsala University.
Elinder, M. and O. Erixson (2012). “Rädde sig den som kan!” Ekonomisk Debatt,
40, 26–35.
Elster, J. (1989). ”Social norms and economic theory,” Journal of Economic Perspectives, 3, 99–117.
Frey, B. S., Savage, D. A., and B. Torgler (2010). “Interaction of natural survival
instincts and internalized social norms exploring the Titanic and Lusitania disasters,” Proceedings of the National Academy of Science of the United States of
America, 107, 4862–4865.
Hall, W (1986). “Social class and survival on the S.S. Titanic,” Social Science &
Medicine, 22, 687–690.
Holtz-Eakin, D., D. Joulfaian, and H. S. Rosen (1993). “The Carnegie conjecture:
Some empirical evidence,” Quarterly Journal of Economics, 108, 413–435.
Imbens, G., D. Rubin, and B. Sacerdote (2001). “Estimating the effect of unearned
income on labor earnings, savings, and consumption: Evidence from a survey of
lottery players,” American Economic Review, 91, 778–794.
Joulfaian, D. (2004). “Gift taxes and lifetime transfers: time series evidence,”
Journal of Public Economics, 88, 1917–1929.
Kopczuk, W. (2007). “Bequest and tax planning: Evidence from estate tax returns,”
Quarterly Journal of Economics, 122, 1801–1854.
Kotlikoff, L. and L. Summers (1981). “The role of intergenerational transfers in
aggregate capital accumulation,” Journal of Political Economy, 89, 706–732.
Kotlikof, L. (1988). “Intergenerational transfers and savings,” The Journal of Economic Perspectives, 2, 41–58.
20
Marmot, M. (1999). “Multi-level approaches to understanding social determinants,”
in L. Berkmanand and I. Kawachi (eds.), Social Epidemiology. Oxford University Press.
Menchik, P. L. (1980). “Primogeniture, equal sharing and the U.S. distribution of
wealth,” Quarterly Journal of Economics, 94, 299–316.
Nordblom, K. and H. Ohlsson (2006). “Tax avoidance and intra-family transfers,”
Journal of Public Economics, 90,1669–1680.
Poterba, J. M. (2000). “Stock market wealth and consumption,” Journal of Economic Perspectives, 14, 99–118.
Smith, J.P. (1999). “Healthy bodies and thick wallets: the dual relation between
health and economic status,” Journal of Economic Perspectives, 13, 145–166.
Wilhelm, M. O. (1996). “Bequest behavior and the effect of heirs’ earnings: Testing
the altruistic model of bequests,” American Economic Review, 86, 874–892.
21
Essay 1: Gender, social norms, and survival in
maritime disasters
♥
Co-authored with Mikael Elinder
♥
This essay is published as “Gender, Social Norms, and Survival in Maritime Disasters,”
Proceedings of the National Academy of Science of the United States of America, Vol. 109,
No. 33, pp. 13220–13224 (2012). We thank Sebastian Escobar for excellent research assistance. We also thank Niclas Berggren, Henrik Jordahl, Wojciech Kopczuk, Che-Yuan Liang,
Henry Ohlsson, Jukka Pirttilä, Johanna Rickne, Erik Spector, Benno Torgler, Daniel Waldenström, the editor, and two anonymous reviewers, as well as seminar participants at Uppsala
University, the Research Institute of Industrial Economics, and the Berlin Network of Labor
Market Research for valuable comments and suggestions. Financial support from the Jan
Wallander and Tom Hedelius Foundations and the Swedish Council for Working Life and
Social Research is acknowledged.
23
Gender, social norms, and survival in
maritime disasters
Mikael Elindera,b,1 and Oscar Erixsona
a
Department of Economics, Uppsala University, SE-751 20 Uppsala, Sweden; and bThe Research Institute of Industrial Economics, SE-102 15 Stockholm, Sweden
Edited by Kenneth Wachter, University of California, Berkeley, CA, and approved June 29, 2012 (received for review May 2, 2012)
Since the sinking of the Titanic, there has been a widespread belief
that the social norm of “women and children first” (WCF) gives
women a survival advantage over men in maritime disasters, and
that captains and crew members give priority to passengers. We
analyze a database of 18 maritime disasters spanning three centuries, covering the fate of over 15,000 individuals of more than 30
nationalities. Our results provide a unique picture of maritime disasters. Women have a distinct survival disadvantage compared
with men. Captains and crew survive at a significantly higher rate
than passengers. We also find that: the captain has the power to
enforce normative behavior; there seems to be no association between duration of a disaster and the impact of social norms;
women fare no better when they constitute a small share of the
ship’s complement; the length of the voyage before the disaster
appears to have no impact on women’s relative survival rate; the
sex gap in survival rates has declined since World War I; and women
have a larger disadvantage in British shipwrecks. Taken together,
our findings show that human behavior in life-and-death situations
is best captured by the expression “every man for himself.”
altruism
| discrimination | homo economicus | leadership | mortality
O
n April 15, 2012, a century had passed since RMS Titanic
sank in the North Atlantic Ocean. The Titanic disaster has
generated immense public and scholarly interest and, as one of
the most extensively covered events in history, obtained an almost mythological status. The evacuation of the Titanic serves as
the prime example of chivalry at sea. Men stood back, while
women and children were given priority to board the lifeboats. In
the end, 70% of the women and children were saved compared
with only 20% of the men (1). The social norm of saving “women
and children first” (WCF) in shipwrecks has often been referred
to as the “unwritten law of the sea.”
It is well known that social norms of fairness and cooperation
influence human behavior in a wide range of situations (2, 3). For
instance, charitable giving and donation of blood and organs is
widespread (4–6). Men and women are, however, subject to different norms of helping behavior (7, 8). Men are in general
expected to help people in emergencies, whereas women are, to
a higher degree, expected to engage in care over the long term.
The expectation of men to display chivalry and heroism in maritime disasters can be seen as an archetypal example of sex differences in social norms of helping behavior. Men displaying
extreme altruism in disasters contrasts the picture from economic
experiments in which men tend to be more selfish than women (9).
Rational individuals, whether with self-regarding or other-regarding preferences, compare the benefits and costs of helping.
When helping substantially increases the risk of dying, it would
be rational for most individuals to save themselves rather than
helping others. This cost–benefit logic is fundamental in economic models of human behavior, including models in which
individuals choose to comply with or violate social norms, for
instance by committing crimes (10).
Maritime disasters provide a valuable context in which it is
possible to empirically investigate how people act and organize
behavior in life-and-death situations and, in particular, if social
norms of helping behavior are being upheld. However, so far,
13220–13224 | PNAS | August 14, 2012 | vol. 109 | no. 33
only the shipwrecks of the Titanic and the Lusitania have been
analyzed with respect to sex and survival (1, 11–14). It has been
concluded that the men on board the Titanic followed the norm
of WCF (11, 12). Based on a comparison of the Titanic and the
Lusitania (where the former sank in 160 min and the latter in less
than 20 min), a conjecture has been suggested to the effect that
norm compliance is more pronounced in disasters that evolve
slowly (11, 12).
Do women normally have a survival advantage in maritime
disasters or was the evacuation of the Titanic an exception? What
situational and cultural conditions determine who survives and
who dies? And what role does the captain play?
To address these questions, we have compiled and analyzed
a database of 18 maritime disasters over the period 1852–2011
(Table 1). Our data cover the fate of over 15,000 passengers and
crew members of more than 30 different nationalities.
Eight hypotheses are tested. The first and main hypothesis (H1)
is that women have a survival advantage over men in maritime
disasters. Previous research on the Titanic has found, in line with
the notion of WCF, that women have a survival advantage over
men, whereas evidence from the Lusitania disaster indicates no
difference in survival rates between men and women (11, 12).
There are, however, several reasons to believe that men have
better survival prospects than women, if they do not engage in selfsacrificing helping behavior. The most important argument would
be that men are physically stronger than women. In the evacuation
of a sinking ship, success is typically determined by the ability to
move fast through corridors and stairs, which is often made difficult by heavy list, congestion, and debris. Other traits that may
enhance survival prospects, such as aggressiveness, competitiveness, and swimming ability, are also more prevalent in men (9, 15–
17), whereas for example resistance to cold water may benefit either sex (18–20). Accordingly, if men try to save themselves, we
expect women to have a relative survival disadvantage. We would,
however, expect women’s survival chances to improve if men
comply with the norm of WCF. Hence, an observed survival advantage of women is regarded as supporting evidence of behavior
being governed by the WCF norm. A small survival disadvantage
for women is difficult to interpret, as it can either indicate that the
WCF norm has helped women from a potentially larger disadvantage or that the norm has not been upheld. However, if we
observe a substantial survival disadvantage of women we regard it
as evidence that compliance with the WCF norm is exceptional in
maritime disasters.
As a second hypothesis (H2), we posit that crew members have
a survival advantage over passengers. According to maritime conventions, it is the duty of crew members—and in particular the
captain—to conduct a safe evacuation of the ship (21). If the crew
Author contributions: M.E. and O.E. designed research, performed research, analyzed
data, and wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
1
To whom correspondence should be addressed. E-mail: [email protected].
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
1073/pnas.1207156109/-/DCSupplemental.
www.pnas.org/cgi/doi/10.1073/pnas.1207156109
24
Table 1. Maritime disasters from 1852 to 2011
Name of ship
Nationality
Year
Cause of
disaster
HMS
Birkenhead
SS Arctic
British
1852
Grounding
US
1854
Collision
SS Golden Gate
US
1862
Fire
SS Northfleet
British
1873
Collision
RMS Atlantic
British
1873
Grounding
SS Princess Alice
British
1878
Collision
SS Norge
Danish
1904
Grounding
RMS Titanic
British
1912
Collision
RMS Empress
of Ireland
RMS Lusitania
British
1914
Collision
British
1915
Torpedoed
SS Principessa
Mafalda
SS Vestris
Italian
1927
Technical
British
1928
Weather
SS Morro Castle
US
1934
Fire
MV Princess
Victoria
SS Admiral
Nakhimov
MS Estonia
MS Princess of
the Stars
MV Bulgaria
British
1953
Weather
Russian
1986
Collision
Estonian
Philippine
1994
2008
Technical
Weather
Russian
2011
Weather
Water
Duration
Indian Ocean,
South Africa
North Atlantic,
Canada
Pacific Ocean,
Mexico
English Channel,
United Kingdom
North Atlantic,
Canada
River Thames,
United Kingdom
North Atlantic,
United Kingdom
North Atlantic,
Canada
St. Lawrence River,
Canada
North Atlantic,
United Kingdom
Atlantic Ocean,
Brazil
Atlantic Ocean,
United States
Atlantic Ocean,
United States
North Channel,
United Kingdom
Black Sea, Ukraine
Quick
Baltic Sea, Finland
Philippine Sea,
Philippines
Volga, Russia
WCF order
Yes
Voyage,
days
Women, % of
passengers
21
1.4
Casualties
365
Survivors
191
Slow
Yes
6
39.7
227
41
Slow
No
6
16.3
206
172
Slow
Yes
9
22.8
287
80
Slow
No
12
29.6
538
330
Quick
No
1
56.8
697
140
Quick
No
6
51.0
635
160
Slow
Yes
5
35.2
1,496
712
Quick
No
2
38.2
983
465
Quick
Yes
6
39.0
1,190
768
Slow
No
7
27.0
309
877
Slow
No
2
33.6
125
183
Slow
No
3
60.4
130
412
Slow
No
1
20.2
135
44
Quick
No
1
47.9
423
820
Slow
Slow
No
Not known
2
2
47.4
49.6
852
791
137
59
Quick
Not known
1
47.7
110
76
follow procedures and leave the ship after the passengers, we expect them to suffer a survival disadvantage compared with passengers. However, crew members are familiar with the ship, often
have emergency training, and are likely to receive early information
about the severity of the situation. We, therefore, expect the crew
to have a relative survival advantage if they try to save themselves
rather than assisting the passengers. Evidence from the Titanic
suggests that crew members indeed have a significant survival advantage over passengers (11).
The third hypothesis (H3) is that the survival rate of women,
relative to that of men, improves when the captain orders WCF.
The potentially important role of the captain has largely been
overlooked in previous studies. Evidence of people helping each
other is not necessarily evidence of other-regarding preferences,
or social norms, governing behavior. It has been shown, both
theoretically and experimentally that people, who would not
otherwise do so, may comply with a social norm if violation is
threatened with punishment (22–24). Unlike other types of catastrophes, e.g., earthquakes, tsunamis, and terrorist attacks, a
maritime disaster is characterized by the presence of a well-defined leader. On board a ship, the captain is the commanding
officer with the supreme power to give and enforce orders. In the
evacuation of the Titanic, the captain ordered WCF (25) and
officers were reported to have shot at men who disobeyed the
order (26). The situation on the Titanic resonates with the
situation in a third-party punishment game (TPPG), in which
threat of punishment is necessary for self-regarding players to
transfer resources to other players (22). Similar to the TPPG, in
which punishment is costly, the WCF order comes at a cost for the
captain because with the order he agrees to remain on board the ship
until all women and children have been rescued. When the captain does not order priority to women, the situation resembles the
allocation problem of a standard dictator game (27, 28), in which
self-regarding players comply with norms only if the cost of the
social stigma of violation exceeds the cost of compliance.
The fourth hypothesis (H4) is that women fare worse, relative to
men, when the ship sinks quickly. It has been suggested that time
is of critical importance for norms to guide behavior (11). When a
ship sinks quickly, human actions are driven by hormonal reactions, such as a rapid increase of adrenaline, and selfish behavior
should dominate. Evidence in favor of this argument rests on
a comparison of the slowly sinking Titanic and the quickly sinking
Lusitania. If a shipwreck is to be considered quick or slow depends
on the size of the ship as well as the number of people on board the
ship. Consequently, we define a shipwreck as quick if the ship sank
in less than X minutes, where we let X be proportional to the size
of the ship’s complement. For a ship of the average size in our
sample (686 passengers and crew) X = 30 min. See SI Appendix, A
for a detailed description of how quick is defined.
PNAS | August 14, 2012 | vol. 109 | no. 33 | 13221
Elinder and Erixson
25
ECONOMIC
SCIENCES
Duration indicates whether the ship sank quickly or slowly. WCF order indicates whether the captain gave the WCF order. (In the analysis, no and not
known are treated as if the order was not given.) Voyage refers to the number of calendar days between departure and the sinking.
The share of women among the passengers may have important implications for helping behavior among men. Giving
priority to women comes at a cost for the men, as they lose
valuable time in abandoning the ship and securing a lifeboat seat.
This cost is lower when there are fewer women on board the
ship, suggesting that behavior in line with the WCF norm will be
more prevalent in shipwrecks with relatively few women. On the
other hand, men have been shown to be more inclined to take
risk in the presence of women (29), suggesting that the presence
of relatively few women may make men less inclined to display
chivalry. As the fifth hypothesis (H5), we posit that the survival
rate of women improves, relative to that of men, when they constitute a comparably small share of the total number of passengers
(below the sample mean of 36.8%).
The sixth hypothesis (H6) is that the survival rate of women
improves, relative to that of men, if the voyage lasted for more
than 1 d before the disaster. The premise is that longer time on
board the ship will lead to more social interactions and increase
social proximity by reducing anonymity between people, formation
of networks, and strengthening of group cohesion. This, in turn,
increases the likelihood that helping behavior is governed by social
norms (30–32). Similarly, social proximity is likely to be higher on
ships with a more intimate atmosphere. We, therefore, also test an
alternative formulation of H6, H6.1, that the relative survival rate
of women is higher when the ship is small (carrying fewer people
than the average-sized ship in the sample, 686 people).
Whereas norms vary over time and space, it has been a grand
challenge for scientists to understand when, where, or how
norms develop, strengthen, or wane (33–35). It is possible that
chivalry at sea was a common phenomenon in the 19th and early
20th century and that the fates of women were determined by
men. With the rise of more sex-equal societies, however, women
may have become more capable of surviving on their own. For
instance, improved swimming skills as well as less restrictive
clothing may have increased the survival prospects of women.
World War I has been seen as a paradigmatic shift in the general
view of manliness and the role of women in society (36). If H1 is
true, but the strength of the WCF norm has weakened over time,
we expect the survival advantage of women to be lower after
World War I. However, if H1 is false, and women have a survival
disadvantage compared with men, we expect the disadvantage to
be smaller after World War I, as women have become more
capable of surviving on their own. In both cases, we expect the
survival rates of men and women to have converged. The seventh
hypothesis (H7) is that the survival difference between men and
women is lower after World War I.
Helping behaviors differ between cultures (34). Such differences may be present in maritime disasters involving ships with
passengers and crew of different nationalities. Previous research
on sex differences in survival has focused solely on British
shipwrecks. Chivalry at sea has been seen as a defining characteristic of Britishness (36). If the expected stigma of norm violation is more severe for British men than for men of other
nationalities, we expect higher compliance with the WCF norm
on board British ships. The captains are British on all British
ships in our sample; likewise crew and passengers are dominated
by Britons on these ships. Our eighth and final hypothesis (H8) is
that women fare better, relative to men, in maritime disasters
involving British ships than in shipwrecks of other nationalities.
Results
Because the hypotheses have been derived mainly from evidence
from the Titanic disaster (and to some extent from the Lusitania), we focus primarily on the 16 previously uninvestigated
shipwrecks, data that we label as our main sample (MS). We
denote the full sample including all shipwrecks in our data FS.
Fig. 1 displays that, in the MS, crew members have the highest
survival rate, followed by captains and male passengers, whereas
13222 | www.pnas.org/cgi/doi/10.1073/pnas.1207156109
Fig. 1. Survival rates of passengers and crew. Survival rates of children are
only available for nine shipwrecks in MS. See SI Appendix, B, Tables S2 and S3
for the statistics underlying this figure.
the lowest survival rates are observed for women and children.
This pattern stands in sharp contrast to the pattern observed for
the Titanic.
We use regression analysis to study determinants of survival in
shipwrecks. The shipwrecks are analyzed both in separate regressions and in regressions based on pooled data including all
of the shipwrecks. The separate analyses of the shipwrecks allow
us to test only H1 and H2. The advantage of these tests, however, is that they are methodologically comparable to previous
tests conducted on data from the Titanic and the Lusitania. The
regression analyses of the pooled data make it possible to control
for unobservable shipwreck-specific circumstances and to test all
eight hypotheses.
The first hypothesis (H1) is that women have a survival advantage over men in maritime disasters. In the separate analyses
of all of the shipwrecks (FS) we find that women have a survival
advantage (P < 0.01) over men in only 2 of the 18 disasters: the
Birkenhead and the Titanic. For 11 of the shipwrecks, we find that
women have a survival disadvantage (P < 0.01) compared with
men. For the remaining 5 shipwrecks, we find no clear evidence
of survival differences between men and women.
If crew members try to save themselves rather than assisting
the passengers, we expect them to have a survival advantage over
passengers (H2). Indeed, we find that crew members have a relative survival advantage (P < 0.01) in 9 of the 18 disasters. For
the remaining 9 shipwrecks, we find no clear evidence of survival
differences between crew and passengers. In addition to the female and crew variables, we augment the regressions with control
variables for characteristics that are likely to affect the individual’s chances of surviving in a shipwreck, such as age, ticket class,
etc. The estimated impacts of those characteristics show that
prime aged adults have a survival advantage over children and
older persons and that there is a class gradient in survival
benefitting first class passengers. Moreover, we find a survival
disadvantage for passengers traveling as part of a group and that
passengers and crew of the same nationality as the ship have no
survival advantage over persons of other nationalities. (For detailed results, see SI Appendix, B, Tables S4 and S5.)
To take full advantage of the data, we present results from
analyses, including all shipwrecks of the MS in each regression.
To control for unobservable factors that vary between ships, but
affect the survival chances of everybody on board each ship
equally, such as e.g., severity of the disaster and weather conditions, we estimate regressions that include shipwreck-specific
fixed effects (37). Table 2 reports the tests of each of the eight
hypotheses (columns 1–8) as well as a joint test of all of the hypotheses together in one regression (column 9). For results of
Elinder and Erixson
26
Table 2. Determinants of survival in maritime disasters
Main hypothesis tested
Female
H1
H2
H3*
H4*
H5*
H6*
H7*
H8*
H1–H8*
1
−0.167
(<0.001)
2
−0.126
(<0.001)
0.187
(<0.001)
3
−0.151
(<0.001)
0.157
(<0.001)
4
−0.151
(<0.001)
0.157
(<0.001)
5
−0.116
(<0.001)
0.157
(<0.001)
6
−0.154
(<0.001)
0.157
(<0.001)
7
−0.195
(<0.001)
0.158
(<0.001)
8
−0.093
(<0.001)
0.159
(<0.001)
9
−0.179
(0.009)
0.161
(<0.001)
−0.153
(<0.001)
0.435
(<0.001)
10,976
0.247
0.096
(0.019)
0.032
(0.452)
−0.050
(0.104)
0.026
(0.443)
0.073
(0.074)
−0.101
(0.002)
0.471
(<0.001)
10,976
0.248
Crew
Female interacted with
WCF order
0.019
(0.477)
Quick
0.005
(0.806)
−0.109
(<0.001)
Small share of women
More than one day voyage
0.006
(0.807)
Post World War I
0.085
(<0.001)
British ship
Constant
Observations
R2
0.346
(<0.001)
10,978
0.249
0.325
(<0.001)
10,976
0.270
0.244
(<0.001)
10,976
0.242
0.237
(<0.001)
10,976
0.242
0.111
(<0.001)
10,976
0.244
0.229
(<0.001)
10,976
0.242
0.329
(<0.001)
10,976
0.244
Linear probability models. The dependent variable (survival) is binary and equals 1 if the person survived the disaster and 0 if the person died. Coefficients
are followed by P values, based on robust SEs, in parentheses. All models include shipwreck-specific fixed effects. Because WCF order, quick, small share of
women, more than one day voyage, post World War I, and British ship do not vary within ships, observations in these regressions are weighted by the inverse
of the number of individuals on board the ship to give all ships equal weight. Complete regression results, as well as results from unweighted regressions and
regressions including the Lusitania and the Titanic can be found in SI Appendix, B, Table S6–S14.
*These regressions also include the binary indicators, which the female variable is interacted with.
relatively better in shipwrecks involving ships with comparably
few people on board (SI Appendix, B, Table S14).
The results in columns 7 and 9 indicate that the survival rate of
women, compared to that of men, is 8.5 and 7.3 percentage
points higher after World War I. The finding that the relative
survival rate of women has improved after World War I holds
also with the inclusion of the Lusitania and the Titanic.
In contrast to H8, the results show that women fare relatively
worse, not better, in shipwrecks involving British ships. The average survival rate of women on board British ships is estimated
to be 15.3 (column 8) and 10.1 (column 9) percentage points
lower than in disasters involving ships of other nationalities.
Although being less strong, the effect remains also with the inclusion of data from the Lusitania and the Titanic. We note that
the WCF order is given more often on board British ships.
However, we find a larger survival disadvantage for women on
British-dominated ships even when controlling for whether or
not the WCF order has been given (column 9).
Discussion
Our results provide unique insights about human behavior in lifeand-death situations. On the Titanic, the survival rate of women
was more than three times higher than the survival rate of men
(11). By investigating a much larger sample of maritime disasters
than what has previously been done, we show that the survival rate
of women is, on average, only about half that of men. We interpret this as evidence that compliance with the WCF norm is
exceptional in maritime disasters. That women fare worse than
men has also been documented for natural disasters (38–42). We
also find that crew members have a higher survival rate than
passengers and that only 9 of 16 captains went down with their
ships. Children appear to have the lowest survival rate.
Moreover, we shed light on some common perceptions of how
situational and cultural conditions affect the survival of women.
Most notably, it seems as if it is the policy of the captain, rather
PNAS | August 14, 2012 | vol. 109 | no. 33 | 13223
Elinder and Erixson
27
ECONOMIC
SCIENCES
samples including the Lusitania and the Titanic, see SI Appendix, B.
The results in column 1 show that the survival rate of women is
16.7 percentage points lower than, or about half of (17.9 vs.
34.6%) that of men. The results in column 2 show that crew
members are 18.7 percentage points more likely to survive than
passengers. The finding that women have a large survival disadvantage compared with men, and that crew members have a survival advantage over passengers, holds true throughout the
specifications in columns 3–9, and also with the inclusion of data
from the Lusitania and the Titanic.
We find some evidence that the survival rate of women, relative to that of men, improves when the captain orders WCF.
Because the WCF order was given on only five ships, including
the Lusitania and the Titanic, MS is not ideal for testing this
hypothesis. Nevertheless, the joint, and most reliable, test (column 9) indicates that the relative survival rate of women
improves by 9.6 percentage points when the captain orders WCF.
The result is strengthened when the Lusitania and the Titanic are
included in the analysis.
The results give no support for H4 (that women fare worse,
relative to men, when the ship sinks quickly, compared with when
the disaster evolves more slowly). Women have a disadvantage
independently of whether the ship sinks quickly or slowly.
The separate test of H5 (column 5) suggests that women fare
worse rather than better, relative to men, when there are comparably few women among the passengers. However, the coefficient is statistically insignificant in the joint test (column 9)
and when we include the Lusitania and the Titanic.
Contrary to H6, we do not find evidence that the relative
survival rate of women improves if the voyage lasts for more than
1 d before the disaster. The coefficient estimates are close to
zero and statistically insignificant in both specifications (columns
6 and 9). This finding also holds true for the alternative test of
this hypothesis (H6.1), i.e., when we test whether women fare
than the moral sentiments of men, that determines whether
women are given preferential treatment in shipwrecks. This
finding suggests an important role for leaders in disasters. Preferences of leaders seem to have affected survival patterns also in
the evacuations of civilians during the Balkan Wars (43). In
contrast to previous studies, we find no association between duration of the disaster and the influence of social norms. Furthermore, women do not appear to benefit from constituting
a small share of the passengers. Neither do we find that contextual
factors, which are likely to reduce social distance on board the
ship, such as the length of the voyage and the size of the complement, influence the survival rate of women. Moreover, we find
that the sex gap in survival rates has decreased since World War I.
This supports previous findings that higher status of women in
society improves their relative survival rate in disasters (41). We
also show that women fare worse, rather than better, relative to
men in maritime disasters involving British ships. This contrasts
with the notion of British men being more gallant than men of
other nationalities. On the basis of our analysis, it becomes evident that the sinking of the Titanic was exceptional in many
dimensions and that what happened on the Titanic seems to have
spurred misconceptions about human behavior in disasters.
information on the sex of survivors and descendants separately. We limit the
sample to shipwrecks involving at least 100 persons and in which at least 5%
survived and 5% died. We have added data for one shipwreck occurring
before 1854, HMS Birkenhead (1852), because it is often referred to as giving
rise to the expression, women and children first: a notion that first became
widespread after the sinking of the Titanic (36). Data for two shipwrecks
that have taken place after 2006 are added: MS Princess of the Stars (2008)
and MV Bulgaria (2011). Despite it being a wartime disaster, we also include
data from the Lusitania (1915) in the sample, as it has been investigated in
previous research. For details about the data, see SI Appendix, A. The data
reported in this paper are available in Dataset S1.
Analytic Method. We test the hypotheses (H1–H8) by estimating linear
probability models. The unit of analysis is the individual passenger or crew
member. The dependent variable (survival) is binary and equals 1 if the
person survived the disaster and 0 if the person died. The independent variable of main interest is the binary variable, female (females = 1, males = 0).
A positive (negative) coefficient implies that women have a higher (lower)
survival rate than men. Crew status is indicated by the binary variable crew
(crew = 1, passengers = 0). For details on coding of variables, see SI Appendix,
A and for model specification see SI Appendix, B.
Data. Starting from the list Some Notable Shipwrecks since 1854, published in
the 140th Edition of The World Almanac and the Book of Facts (44), we have
selected shipwrecks involving passenger ships that have occurred in times
of peace, and for which there are passenger and crew lists containing
ACKNOWLEDGMENTS. We thank Sebastian Escobar for excellent research
assistance. We also thank Niclas Berggren, Henrik Jordahl, Wojciech Kopczuk,
Che-Yuan Liang, Henry Ohlsson, Jukka Pirttilä, Johanna Rickne, Erik Spector,
Benno Torgler, Daniel Waldenström, the editor, and two anonymous
reviewers, as well as seminar participants at Uppsala University, the Research
Institute of Industrial Economics, and the Berlin Network of Labor Market
Research for valuable comments and suggestions. Financial support from
the Jan Wallander and Tom Hedelius Foundations and the Swedish Council
for Working Life and Social Research is acknowledged.
1. Hall W (1986) Social class and survival on the S.S. Titanic. Soc Sci Med 22:687–690.
2. Elster J (1989) Social norms and economic theory. J Econ Perspect 3:99–117.
3. Ostrom E (2000) Collective action and the evolution of social norms. J Econ Perspect
14:137–158.
4. Thorne ED (2006) The economics of organ transplantation. Handbook on the Economics of Giving, Reciprocity and Altruism, eds Kolm S-C, Ythier JM (Elsevier, New
York), Vol 2, pp 1335–1370.
5. Andreoni J (2006) Philanthropy. Handbook on the Economics of Giving, Reciprocity
and Altruism, eds Kolm S-C, Ythier JM (Elsevier, New York), Vol 2, pp 1201–1269.
6. List JA (2011) The market for charitable giving. J Econ Perspect 25:157–180.
7. Eagly AH, Crowley M (1986) Gender and helping behavior: A meta-analytic review of
the social psychological literature. Psychol Bull 100:283–308.
8. Becker SW, Eagly AH (2004) The heroism of women and men. Am Psychol 59:163–178.
9. Croson R, Gneezy U (2009) Gender differences in preferences. J Econ Lit 47:448–474.
10. Becker GS (1968) Crime and punishment: An economic approach. J Polit Econ 76:
169–217.
11. Frey BS, Savage DA, Torgler B (2011) Behavior under extreme conditions: The Titanic
disaster. J Econ Perspect 25:209–222.
12. Frey BS, Savage DA, Torgler B (2010) Interaction of natural survival instincts and internalized social norms exploring the Titanic and Lusitania disasters. Proc Natl Acad
Sci USA 107:4862–4865.
13. Gleicher D, Stevans LK (2004) Who survived Titanic? A logistic regression analysis.
International Journal of Maritime History 16:61–94.
14. Folkesson P (2005) Katastrofer och män: Explorativa undersökningar av ett komplext
förhållande. PhD thesis (Univ of Gothenburg, Gothenburg, Sweden).
15. Niederle M, Vesterlund L (2007) Do women shy away from competition? Do men
compete too much? Q J Econ 122:1067–1101.
16. Taylor SE, et al. (2000) Biobehavioral responses to stress in females: Tend-and-befriend, not fight-or-flight. Psychol Rev 107:411–429.
17. Mazur A, Booth A (1998) Testosterone and dominance in men. Behav Brain Sci 21:
353–363, discussion 363–397.
18. McArdle WD, Toner MM, Magel JR, Spina RJ, Pandolf KB (1992) Thermal responses of
men and women during cold-water immersion: Influence of exercise intensity. Eur J
Appl Physiol Occup Physiol 65:265–270.
19. McArdle WD, Magel JR, Gergley TJ, Spina RJ, Toner MM (1984) Thermal adjustment to
cold-water exposure in resting men and women. J Appl Physiol 56:1565–1571.
20. Rivers JPW (1982) Women and children last: An essay on sex-discrimination in disasters. Disasters 6:256–267.
21. International Maritime Organization (2004) SOLAS: Consolidated edition, 2004:
Consolidated text of the International Convention for the Safety of Life at Sea, 1974,
and its Protocol of 1988: Articles, annexes and certificates: Incorporating all amendments in effect from January 1, 2004 (International Maritime Organization, London).
22. Fehr E, Fischbacher U (2004) Third-party punishment and social norms. Evol Hum
Behav 25:63–87.
23. Fehr E, Schmidt KM (1999) A theory of fairness, competition, and cooperation. Q J
Econ 114:817–868.
24. Fehr E, Fischbacher U (2003) The nature of human altruism. Nature 425:785–791.
25. Great Britain. Commissioner of Wrecks (1912) Formal Investigation into the Loss of
the S. S. “Titanic”: Evidence, appendices, and index (Her Majesty’s Stationery Office,
London).
26. Chapman E (2001) Gunshots on the Titanic. Available at http://www.encyclopediatitanica.org/gunshots-on-titanic.html. Accessed June 27, 2012.
27. Kahneman D, Knetsch JL, Thaler R (1986) Fairness as a constraint on profit seeking:
Entitlements in the market. Am Econ Rev 76:728–741.
28. Forsythe R, Horowitz JL, Savin NE, Sefton M (1994) Fairness in simple bargaining
experiments. Games Econ Behav 6:347–369.
29. Ronay R, von Hippel W (2010) The presence of an attractive woman elevates testosterone and physical risk taking in young men. Social Psychological and Personality
Science 1:57–64.
30. Apicella CL, Marlowe FW, Fowler JH, Christakis NA (2012) Social networks and cooperation in hunter-gatherers. Nature 481:497–501.
31. Fowler JH, Christakis NA (2010) Cooperative behavior cascades in human social networks. Proc Natl Acad Sci 107: 5334–5338.
32. Hoffman E, McCabe K, Smith VL (1996) Social distance and other-regarding behavior
in dictator games. Am Econ Rev 86:653–660.
33. Camerer CF, Fehr E (2006) When does “economic man” dominate social behavior?
Science 311:47–52.
34. Henrich J, et al. (2001) In search of Homo economicus: Behavioral experiments in 15
small-scale societies. Am Econ Rev 91:73–78.
35. Ostrom E, Walker J, Gardner R (1992) Covenants with and without a sword: Selfgovernance is possible. Am Polit Sci Rev 86:404–417.
36. Delap L (2006) ’Thus does man prove his fitness to be the master of things’: Shipwrecks, chivalry and masculinities in nineteenth- and twentieth-century Britain. Cult
Soc Hist 3:45–74.
37. Wooldridge JM (2002) Econometric Analysis of Cross Section and Panel Data (MIT
Press, Cambridge, MA).
38. Frankenberg E, Gillespie T, Preston S, Sikoki B, Thomas D (2011) Mortality, the family
and the Indian Ocean tsunami. Econ J 121:F162–F182.
39. Ikeda K (1995) Gender differences in human loss and vulnerability in natural disasters:
A case study from Bangladesh. Indian J Gend Stud 2:171–193.
40. MacDonald R (2005) How women were affected by the tsunami: A perspective from
Oxfam. PLoS Med 2:e178.
41. Neumayer E, Plümper T (2007) The gendered nature of natural disasters: The impact
of catastrophic events on the gender gap in life expectancy, 1981–2002. Ann Assoc
Am Geogr 97:551–566.
42. Oxfam International (2005) The tsunami’s impact on women. Briefing Note (Oxfam
International, Oxford, UK).
43. Carpenter RC (2003) ’Women and children first’: Gender, norms, and humanitarian
evacuation in the Balkans 1991–95. Int Organ 57:661–694.
44. Joyce CA ed (2007) The World Almanac and Book of Facts 2008 (World Almanac
Books, New York).
Methods
13224 | www.pnas.org/cgi/doi/10.1073/pnas.1207156109
Elinder and Erixson
28
Supplementary Information Appendix for
Gender, Social Norms and Survival in Maritime Disasters
Mikael Elinder: Department of Economics, Uppsala University, and the Research Institute of
Industrial Economics (IFN), Stockholm
Oscar Erixson: Department of Economics, Uppsala University
Corresponding author: Mikael Elinder, Department of Economics, Uppsala University, P.O.
Box 513, SE-751 20 Uppsala, Sweden, email: [email protected], phone: +46 18 471
15 65, fax: +46 18 471 14 78
1
29
Table of Contents
Appendix A ................................................................................................................................ 4
Selection of shipwrecks .......................................................................................................... 4
Data from passenger and crew lists ........................................................................................ 5
Shipwreck characteristics ....................................................................................................... 8
Appendix B ................................................................................................................................ 9
Data underlying Figure 1........................................................................................................ 9
Tests of H1 for individual shipwrecks: linear probability models ....................................... 11
Tests of H1 for individual shipwrecks: probit models ......................................................... 13
Tests of H2 for individual shipwrecks: linear probability models ....................................... 13
Tests of H2 for individual shipwrecks: probit models ......................................................... 14
Regression results for: MS, MS+Lusitania, MS+Titanic, and FS ....................................... 14
Results from unweighted regressions ................................................................................... 20
Results from regressions with respect to H6.1 ..................................................................... 25
References ............................................................................................................................ 27
2
30
List of Tables
Table S1. Availability of individual level data........................................................................... 6
Table S2. Casualty statistics of MS ............................................................................................ 9
Table S3. Casualty statistics of the Titanic ................................................................................ 9
Table S4. Regression results for each shipwreck in FS ........................................................... 10
Table S5. Detailed regression results for the specifications in column 3 of Table S4. ............ 11
Table S6. Regression results for MS ........................................................................................ 16
Table S7. Regression results for MS augmented with the Lusitania ....................................... 17
Table S8. Regression results for MS augmented with the Titanic ........................................... 18
Table S9. Regression results for FS ......................................................................................... 19
Table S10. Results from unweighted regressions on MS......................................................... 21
Table S11. Results from unweighted regressions on MS augmented with the Lusitania ........ 22
Table S12. Results from unweighted regressions on MS augmented with the Titanic ............ 23
Table S13. Results from unweighted regressions on FS .......................................................... 24
Table S14. Regression result from an alternative test of H6 .................................................... 26
3
31
Appendix A
This section provides a description of the database of maritime disasters used in the
article Gender, Social Norms and Survival in Maritime Disasters. It consists of three parts: In
the first part, we describe the selection of shipwrecks. In the second part, we discuss the data
obtained from passenger and crew lists. In the third part, we discuss the shipwreck
characteristics that we use in the analysis. The complete database of shipwrecks is available as
SI Material, Dataset S1.
Selection of shipwrecks
Every year, hundreds or even thousands of accidents occur at sea. Fortunately, only a
few cause substantial loss of life. No official list of the most severe maritime disasters exists.
To select shipwrecks for the analysis, we therefore started off from the list Some Notable
Shipwrecks since 1854 in the 140th Edition of The World Almanac and the Book of Facts (1).
The list contains a total of 152 shipwrecks over the period 1854Ȃ2006. Although the list is
comprehensive and covers maritime disasters globally, it is likely that disasters of the Western
world and disasters that have gained much media attention are overrepresented. It is, however,
the most extensive list we are aware of.
We have imposed four criteria that need to be fulfilled for the shipwreck to be included
in our database: First, the disaster should have occurred in peacetime. Second, the shipwreck
in question should involve a passenger ship. Third, we only include shipwrecks that involved
more than 100 people and in which at least 5 percent survived and 5 percent died.
Fourth, data (individual or aggregate) on survival rates of men and women separately
should be available. The two first criteria can be seen as limiting the population of interest,
while the latter two renders the sample somewhat unrepresentative. It should be mentioned
that information about the shipwrecks and passenger lists are very difficult to obtain for
disasters involving ships from many developing countries. This is unfortunate, since several
of the deadliest disasters have involved such ships. For instance the sinking of Philippine
registered MV Doña Paz and the Senegalese registered MV Le Joola are estimated to have
resulted in more than 4,000 and 1,800 lives lost but are not included in our sample.
Furthermore, language barriers have made it difficult to find extensive information about
some shipwrecks. As a consequence, British and American ships are likely to be
overrepresented in our sample.
Applying the above sample criteria leaves us with a sample of 14 shipwrecks. We have
added one shipwreck occurring before 1854, HMS Birkenhead (1852), as it is often referred
to as giving rise to the expression ‘women and children first’. Moreover, we have added two
shipwrecks that have taken place after year 2006: MS Princess of the Stars (2008) and MV
Bulgaria (2011). We have also added RMS Lusitania, despite occurring in wartime, since it
has been analyzed in previous research. In total, we have a sample consisting of 18
shipwrecks, whereof only RMS Titanic and RMS Lusitania have previously been
systematically investigated with respect to individual and social determinants of survival.
Individual level data for each shipwreck have been collected from the ship’s passenger
and crew lists. 5 of the lists are obtained from books, 3 from official sources, such as e.g.
inquiry commissions or government authorities, 8 from web sites, and 2 are collected from
newspaper articles. It difficult to say which source is the most reliable. Logbooks and ship
records have often been lost in the wreck, especially in earlier years. Moreover, it takes time
to establish accounts of a maritime disaster. As a consequence, we have used the latest source
available. The main sources have been cross-checked with other sources whenever possible.
Reference to each source is provided in the excel file MartimeDisasters.xlsx.
We only include persons who have been confirmed to have been on board the ship at the
time of the accident, or put differently, only those persons appearing in the particular
4
32
passenger and crew lists. As a consequence, the total number of passengers, as well as the
number of survivors and deceased, sometimes differs from the numbers appearing in other
references.
We have individual level data for 17 of the shipwrecks. For the Admiral Nakhimov there
are aggregate data on the number of male and female passengers and crew. Accordingly, we
use the aggregate statistics to construct individual level data.
Data from passenger and crew lists
Below follow details about how the variables obtained from passenger and crew lists are
coded and for which shipwrecks these variables are available. Table S1 reports which
variables are available for each shipwreck.
Survival. Some passenger lists discriminate between deceased and missing persons. In
the majority of cases ‘missing’ implies that the body has not been recovered, but that the
person is presumed dead. For our analysis we have grouped the two categories and created a
binary variable which takes the value one (=1) if the individual survived the disaster and zero
(=0) if the individual died (either confirmed dead or missing). We have compared our
statistics with the casualty figures appearing in other sources and can conclude that there are
only minor discrepancies.
Female. Gender is the individual characteristic of primary interest to us. Only a few
passenger lists provide explicit information about the gender of the persons on board the ship.
For most ships we have used the individual’s name to determine gender. When there are
uncertainties regarding the gender associated with a particular name we have used online
name dictionaries that provide information on the origin of the name and informative statistics
on whether it is typically a male name or a female name. In some passenger lists, especially
those dating back to the 19th century, the classification is simplified by the presence of gender
based prefixes such as MRS (if married female), MISS (if unmarried female) or MR (if adult
male). Professional titles such as Dr, Professor, Stewardess, Captain, etc., have also been
helpful for determining the gender of passengers and crew members. We have been unable to
determine the gender of some individuals, when they are stated with initials instead of
forenames in the passenger manifest. This appears primarily for shipwrecks in the 19th
century. Other difficulties come from misspellings in transcription of names. This occurs
especially among East European emigrants travelling on American and British ships. The
observations, which remained inconclusive after applying the above methods, were left out
from the empirical analysis. In many cases, we cannot discriminate between women and girls
and men and boys. Hence, we use the terms female and male. Gender enters our empirical
analyses as a binary variable, Female, taking the value one (=1) for females (women and
girls) and zero (=0) for males (men and boys).
Crew. All the passenger lists we have gathered provide some sort of indictor of whether
listed person is a passenger or member of the crew. In some passenger lists there is more
detailed information about the crew such as e.g. in which department (i.e., deck, engineering,
or steward) the crew member worked, and in some cases even the specific title. The amount,
quality and type of crew characteristics vary substantially between the ships. We therefore
treat the crew as a homogenous entity. For most shipwrecks the great majority of crew
members are men. This implies that the information on crew membership is not only
important in the test of H2 but also that crew membership is an important control variable in
the other tests as well. Accordingly, we have constructed a binary variable Crew taking the
value one (=1) for crew members and zero (=0) for passengers. The captain is included in the
crew.
5
33
34
X
SS Principessa Malfalda
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Gender
3
X
X1
X
X
X2
X
X
X1
X
Age
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Crew
X
X
X
X5
X
X
X4
X
Passenger class
X
6
X
X
Nationality
X
X
X
Companionship
SS Admiral Nakhimova
X
X
X
X
X
X
X
X
MS Estonia
X
X
X
MS Princess of the Stars
MV Bulgaria
X
X
X
X
Notes. aData are compiled from aggregate statistics. 1Only indicator for whether the individual is a child or adult. 2Indicator for whether the individual is a child
or adult and age for adults and some (presumably older) children. 3The data are incomplete for the crew and therefore not controlled for in the regression
models. 4Passengers are categorized as saloon (1st class) passengers and steerage (2nd and 3rd class) passengers. 5One person appears as first class passenger in
the passenger list and ten persons appear as second class passengers. Because of the relatively small numbers we do not include controls for passenger class in
the analysis of SS Norway. 6The nationalities are: English, Welsh, Scottish, Northern Irish, and Irish. We do not use this information in the analysis of MV
Princess Victoria.
X
X
RMS Lusitania
MV Princess Victoria
X
X
X
X
SS Princess Alice
SS Norge
RMS Titanic
RMS Empress of Ireland
X
X
X
X
X
SS Golden Gate
SS Northfleet
SS Atlantic
SS Vestris
SS Morro Castle
X
X
Survival
HMS Birkenhead
SS Arctic
Ship/Variable
Table S1. Availability of individual level data
6
Age. Physical strength and mobility are likely to be important determinants of survival in
a shipwreck. Unfortunately, the passenger lists do not provide us with this kind of
information. A person’s age may however capture these characteristics fairly well. For
instance, prime aged individuals are likely to be both physically stronger and more mobile
than children and older adults. 9 passenger lists contain information on age. In some cases it
seems as if the availability is systematic. For example, the emigrant ships tend to have more
extensive documentation of the age of the first class (saloon) passengers than the third class
(steerage) passengers. For two of the ships (the Estonia and the Bulgaria) age is not given
explicitly but in the form of year-of-birth. We have then calculated age as the year of the
disaster minus the person’s year-of-birth. Age enters the empirical specifications in the form
of categorical variables, namely: persons younger than 16 (Age<16); persons 16–50 years old
(Age 16–50); and persons older than 50 (Age>50), with Age 16–50 being the reference group.
Similar age groups have been used in previous studies (2, 3). Two passenger lists (the Golden
Gate and the Vestris) do not contain information on age but make a distinction between adults
and children. We create a binary variable Child which equals one (=1) if the person is a child
and zero (=0) if an adult. When we analyze children explicitly we denote persons for whom
Age<16=1 or Child=1 as children.
Passenger class. Another individual characteristic that may correlate with survival is
passenger class. First and second class passengers may have a survival advantage over third
class passengers as their cabins are often located further up in the ship, close to the lifeboats,
while the third class compartments are often located at the lower decks, away from the
lifeboats. Also, in the case of a collision or grounding the ship’s hull, beneath, or just above,
the water level often takes the initial strike with the consequence that third class
compartments are flooded quicker than the first and second class decks. Previous studies on
the loss of the Titanic and the Lusitania report that first and second class passengers had a
significantly better chance to survive than third class passengers (3, 4). 8 passenger lists,
especially those dating back in time, separate passengers into different classes: often first
class, second class and third class, or saloon (first class) and steerage (second and third class).
We have constructed two binary variables: 1st class and 2nd class (3rd class being the reference
group), each taking the value one (=1) if the passenger belongs to the particular class and zero
(=0) otherwise.
Nationality. Another potentially important individual characteristic that could influence
survival chances is nationality. For instance, speaking the same language as the crew may be
important in order to absorb information about safety equipment and also for understanding
directions about where to go during the evacuation. 4 passenger lists contain information on
the nationalities of the passengers and crew members. We create a binary variable,
Nationality, which takes the value one (=1) if the individual is of the same nationality as the
ship and zero (=0) otherwise.
Companionship. The social attachment model of human behavior in disasters (5)
predicts that the presence of familiar persons affect peoples’ perceptions of, and responses to,
danger. A general finding is that people want to keep proximity to attachment figures, such as
family and friends (6). Accordingly, we may see differences in survival rates between persons
traveling alone and those traveling as a part of a social entity. It is, however, not obvious
whether the effect of traveling with family or friends on survival probability is positive or
negative. On the one hand, the social attachment model suggest that group membership could
act as a constraint on survival if the member is slowed down by the search for and help
directed to weaker members. On the other hand a social entity can provide information and
physical help which in turn may increase the survival chances of its members. 3 passenger
lists provide some sort of indicator of the social relationships between the passengers, e.g.
information on whether people were married or whether they shared cabins. We create a
7
35
binary variable Companionship which takes the value one (=1) if the individual traveled in a
group and zero (=0) otherwise.
Shipwreck characteristics
We complement the data obtained from the passenger and crew lists with shipwreck
specific characteristics. The information underlying these variables has been collected from
the key references for each shipwreck, and whenever possible crosschecked against
alternative sources.
WCF order. We have searched the shipwreck accounts for evidence of whether the
captain, or any other officer, gave the order ‘women and children first’ at some point during
the evacuation. For 5 of the shipwrecks we have found supporting evidence of the order while
for 9 cases there is no indication of the order been given. For 2 shipwrecks (the Princess of
the Stars and the Bulgaria) the documentation of the evacuation is too brief to conclude
whether or not the order was given. For the empirical analysis we create a binary variable
WCF order equal to one (=1) if the order was given and zero (=0) if it was not given, or if it is
not known to us whether it was given.
Quick. We define sinking time as the duration between the first indication of distress and
the sinking. For ease of interpretation we classify the disasters into two categories: ‘Quick’
and ‘Slow’. Whether a ship is defined as ‘Quick’ depends on the time period between the first
indication of distress and the sinking and the number of people on board. A ship of average
size in our sample (686 passengers and crew) is defined as ‘Quick’ if it sinks in less than 30
minutes. The threshold time for a ship being categorized as ‘Quick’ is defined as follows:
threshold time= ship size/22.86. If the actual sinking time is lower than the threshold time it is
categorized as ‘Quick’ and ‘Slow’ otherwise. A ship with a complement of 229 people is thus
categorized as ‘Quick’ if it sinks in less than 10 minutes, while a ship with a complement of
2,286 persons is categorized as quick if it sinks in less than 100 minutes. 7 disasters in FS are
‘Quick’ according to this definition. In the econometrical specifications we include a binary
variable Quick, which equals one (=1) if the disaster was ‘Quick’ and zero (=0) if it was
‘Slow’.
Small share of women. We have information on the number of passengers on board each
ship in the sample, as wells as information on how many of these were men and women. For
the empirical analysis we define a binary variable, Small share of women, which equals one
(=1) if the share of women passengers of the total number of passengers on board the ship is
below the mean share in the sample (0.368), and zero (=0) otherwise.
More than one day voyage. For each ship we have information on the date of final
departure and the date of the disaster. This allows us to calculate the length of the voyage (in
calendar days). We use this information to construct a binary variable, More than one day
voyage, which equal one (=1) if the final voyage lasted for more than one day, and zero (=0)
otherwise. The number of days at sea prior to the disaster varies between 1 and 21 in the
sample. 4 ships wrecked on the day of departure.
Small ship. We use information on the total number of persons (passengers and crew) on
board the ships to construct a binary variable which takes the value one (=1) if the ship
carried less people than the average ship in the sample (686), and zero (=0) if the ship carried
more than 686 persons.
Post WWI. The sample spans the period 1852Ȃ2011. For the empirical analysis we
define a binary variable, Post WWI, which equals one (=1) if the disaster took place after
World War I and zero (=0) if it took place before, or during the war. The only shipwreck in
our sample that took place during World War I is the Lusitania disaster in 1915. The first
shipwreck after the World War I, in our sample, is the Principessa Mafalda in 1927.
8
36
British ship. Refers to the country in which the ship was registered at the time of the
accident. In all cases, but three (the Titanic, the Empress of Ireland and the Estonia), there is
an exact match between the ship’s flag and the nationality of the ship owner. Also, all
captains have the same nationality as their respective ship. In the empirical analysis we
discriminate between British ships and vessels of other nationalities. There are 8 British ships
in FS. We create a binary variable (=1) if the ship is British and (=0) otherwise.
Appendix B
In this section we present more detailed results than we provide in the main text. We also
discuss results from supplementary analyses intended to show the robustness of the
conclusions presented in the main text.
Data underlying Figure 1.
Here we present the statistics that are used to construct Fig 1, in the main text. Table S2
displays the casualty statistics for MS. Note that the information about children is based on 9
shipwrecks only. This means that, for these 9 shipwrecks, the sub-groups Men and Women
exclude boys and girls. For the remaining 7 shipwrecks, however, boys and girls are included
in Men and Women. Table S3 shows the casualty statistics for the Titanic disaster.
Table S2. Casualty statistics of MS
Survivors
Deceased
Total
Passengers
Men
1,802 (37.4)
3,010 (62.6)
4,812
Women
849 (26.7)
2,335 (73.3)
3,184
Children
95 (15.3)
526 (84.7)
621
1,441 (61.1)
918 (38.9)
2,359
7 (43.8)
9 (76.2)
16
Crew
Captain
Notes. Survival rates (in percent) are in parentheses. Crew includes captains.
Table S3. Casualty statistics of the Titanic
Survivors
Deceased
Total
782
Passengers
Men
132 (16.9)
650 (83.1)
Women
300 (74.6)
102 (25.4)
402
Children
68 (51.1)
65 (48.9)
133
212 (23.8)
679 (76.2)
891
0 (0)
1 (100)
1
Crew
Captain
Notes. Survival rates (in percent) are in parentheses. Crew includes the captain.
9
37
Table S4. Regression results for each shipwreck in FS
Estimates of the coefficient on Female
Shipwreck
HMS Birkenhead
LPM
(1)
(2)
0.665
0.729
(<0.001) (<0.001)
SS Arctic
-0.199
-0.188
Probit
(4)
(5)
(6)
n.a.
n.a.
n.a.
n.a.
n.a.
n.a.
n.a.
(<0.001) (<0.001) (<0.001)
SS Golden Gate
SS Northfleet
RMS Atlantic
(8)
0.546
0.557
(<0.001) (<0.001)
n.a.
(10)
(11)
0.561
0.567
(<0.001) (<0.001)
(12)
n.a.
0.093
0.021
-0.034
0.090
0.022
-0.059
(0.043)
(0.697)
(0.589)
(0.039)
(0.697)
(0.325)
0.240
0.229
0.295
0.245
0.234
0.303
-0.075
-0.120
-0.139
-0.079
-0.141
(0.334)
(0.136)
(0.091)
(0.343)
(0.124) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
-0.227
-0.233
-0.184
-0.312
-0.317
-0.244
-0.012
-0.065
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (0.879)
(0.419)
-0.521
-0.469
-0.469
-0.096
-0.087
-0.192
-0.183
0.586
0.526
0.542
0.460
n.a.
0.465
-0.012
-0.058
(0.880)
(0.446)
0.692
0.640
n.a.
0.648
n.a.
n.a.
n.a.
-0.095
-0.086
-0.032
0.174
0.136
0.082
0.141
0.103
0.041
(0.076) (<0.001) (0.001)
(0.047)
(0.030)
(0.085)
(0.233)
(0.011)
(0.059)
(0.158)
-0.180
-0.181
0.150
0.062
0.026
0.130
0.047
0.018
(0.314)
(0.716)
(0.005)
(0.304)
(0.745)
0.035
0.141
-0.145
0.042
0.192
-0.032
-0.197
-0.190
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (0.012)
RMS Titanic
N
Probit
(9)
-0.136
(<0.001) (0.001)
SS Norge
(7)
(0.082)
(<0.001) (<0.001) (<0.001)
SS Princess Alice
LPM
(3)
-0.168
Estimates of the coefficient on Crew
0.499
0.506
0.526
0.527
-0.142
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (0.067) (<0.001) (<0.001) (0.069) (<0.001)
RMS Empress of Ireland
-0.288
-0.165
-0.171
-0.335
-0.215
-0.229
0.388
0.330
0.354
0.372
0.303
0.332
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
RMS Lusitania
SS Principessa Mafalda
SS Vestris
-0.029
-0.013
-0.029
-0.023
-0.013
-0.030
0.044
0.040
0.031
0.044
0.039
0.031
(0.238)
(0.633)
(0.327)
(0.240)
(0.634)
(0.324)
(0.057)
(0.114)
(0.501)
(0.056)
(0.114)
(0.497)
-0.008
0.036
0.053
-0.008
0.034
0.050
0.167
0.175
0.154
0.187
0.195
0.175
(0.803)
(0.270)
(0.105)
(0.802)
(0.284)
-0.404
-0.269
-0.254
-0.416
-0.287
(<0.001) (0.002)
SS Morro Castle
MV Princess Victoria
MS Estonia
-0.287
(0.002) (<0.001) (0.003)
0.295
0.079
0.073
-0.022
0.071
0.068
0.12
0.166
0.308
(0.121)
(0.148)
(0.566)
(0.121)
(0.142)
(0.001)
(0.001)
(0.001)
-0.297
-0.311
-0.441
n.a.
n.a.
n.a.
-0.055
0.002
(0.041)
(0.933)
-0.167
-0.172
-0.073
-0.085
(<0.001) (<0.001)
MV Bulgaria
0.168
n.a.
-0.165
-0.055
-0.001
(0.042)
(0.972)
-0.166
-0.171
n.a.
-0.153
-0.334
-0.257
(<0.001) (0.001)
n.a.
-0.216
-0.080
-0.087
(<0.001) (<0.001)
-0.339
-0.271
(0.006) (<0.001) (0.001)
n.a.
-0.231
0.296
0.214
(0.012) (<0.001) (0.001)
-0.022
0.123
0.163
0.161
(0.016)
0.281
(0.001) (<0.001) (<0.001)
-0.064
-0.095
-0.199
-0.066
-0.112
-0.229
(0.358)
(0.169)
(0.095)
(0.373)
(0.185)
(0.074)
0.215
0.216
0.234
0.234
(<0.001) (<0.001)
0.079
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (0.012)
MS Princess of the Stars
0.209
(0.004) (<0.001) (0.002)
(0.569)
(<0.001) (<0.001) (<0.001)
SS Admiral Nakhimov
(0.119) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
n.a.
(<0.001) (<0.001)
n.a.
0.094
0.042
0.071
0.083
0.036
(0.002)
(0.259)
(0.006)
(0.001)
(0.150)
-0.028
-0.061
(0.201)
(0.012)
0.412
0.300
n.a.
0.514
-0.032
-0.053
(0.269)
(0.041)
0.425
0.322
547-554
206-268
356
338-367
633-868
578-837
795
2,198-2,208
1,448
1,287-1,958
1186
296-308
542
93-179
1,243
989
n.a.
850
n.a.
148-186
(0.005) (<0.001) (0.002) (<0.001) (<0.001) (0.003)
Notes. Linear probability and probit models. The dependent variable (Survival) is binary and equals one if the person survived the disaster
and zero if the person died. p-values, based on robust standard errors, in parentheses below the coefficients (marginal effects for probit). N
refers to the number of observations over which the models have been estimated. N varies within some shipwrecks. This is because, for
some shipwrecks, the information underlying the regressor(s) is not available for everybody in the shipwreck.
38
Table S5. Detailed regression results for the specifications in column 3 of Table S4.
Ship\Variable
HMS Birkenhead
SS Arctic
SS Golden Gate
SS Northfleet
RMS Atlantic
SS Princess Alice
SS Norge
RMS Titanic
RMS Empress of Ireland
RMS Lusitania
SS Principessa Mafalda
SS Vestris
SS Morro Castle
MV Princess Victoria
SS Admiral Nakhimov
MS Estonia
MS Princess of the Stars
MV Bulgaria
Age <161
Age >501
Child2
Nationality3
1st class4
2nd class4
Female
Crew
0.729
0.557
Companionship5
Constant
0.271
(<0.001)
(<0.001)
(<0.001)
-0.168
-0.034
-0.134
0.242
(<0.001)
(0.589)
(0.003)
(<0.001)
-0.120
0.295
0.516
0.191
0.238
-0.202
0.364
(0.136)
(<0.001)
(<0.001)
(0.029)
(0.003)
(0.029)
(<0.001)
-0.184
-0.132
0.283
(<0.001)
(<0.001)
(<0.001)
-0.469
0.465
0.119
0.460
(<0.001)
(<0.001)
(0.100)
(<0.001)
-0.032
0.082
-0.027
0.004
0.071
(0.076)
(0.233)
(0.126)
(0.891)
(<0.001)
-0.180
0.026
-0.128
-0.204
-0.005
0.335
(<0.001)
(0.716)
(<0.001)
(<0.001)
(0.913)
(<0.001)
0.499
0.141
0.140
-0.133
0.341
0.142
0.087
(<0.001)
(<0.001)
(0.002)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
-0.171
0.354
0.242
0.023
0.244
(<0.001)
(<0.001)
(<0.001)
(0.425)
(<0.001)
-0.029
0.031
-0.131
-0.161
0.033
0.051
0.098
0.376
(0.327)
(0.501)
(0.002)
(<0.001)
(0.328)
(0.198)
(0.006)
(<0.001)
0.053
0.154
-0.250
-0.080
0.710
(0.105)
(<0.001)
(<0.001)
(0.110)
(<0.001)
-0.254
0.168
-0.433
0.538
(0.002)
(0.012)
(<0.001)
(<0.001)
0.074
0.311
-0.088
0.012
0.517
(0.146)
(0.001)
(0.532)
(0.804)
(<0.001)
-0.441
-0.199
-0.463
-0.324
0.463
(<0.001)
(0.065)
(<0.001)
(0.005)
(<0.001)
0.002
0.216
0.598
(0.933)
(<0.001)
(<0.001)
-0.165
0.042
-0.093
-0.151
-0.016
0.274
(<0.001)
(0.259)
(0.264)
(<0.001)
(0.599)
(<0.001)
-0.085
-0.061
0.110
(<0.001)
(0.012)
(<0.001)
-0.216
0.514
-0.096
-0.144
0.541
(0.006)
(<0.001)
(0.263)
(0.108)
(<0.001)
N
554
268
356
338
868
578
795
2,198
1,448
1,287
1,186
308
542
93
1,243
989
850
169
Notes. Linear probability models. The dependent variable (Survival) is binary and equals one if the person survived the disaster and zero if the person
died. p-values, based on robust standard errors, in parentheses below the coefficient. N refers to the number of observations over which the model has
been estimated. 1The reference group is people aged 16−50. 2The reference group is Adult. 3The reference group consists of persons of nationalities
other than that of the ship. 4The reference group is 3rd class. 5The reference group is people traveling alone.
11
39
Tests of H1 for individual shipwrecks: linear probability models
In this section we provide additional support to the discussion surrounding the separate
analyses of the shipwrecks of FS in the main text. In all models, the dependent variable is
Survival.
Table S4, column 1–3, reports the coefficient for Female from a set of linear probability
models (LPM). These models serve as tests of H1: that women have a survival advantage
over men in maritime disasters. The baseline model (Model 1) is: ܵ‫݈ܽݒ݅ݒݎݑ‬௜ ൌ ܿ‫ ݐ݊ܽݐݏ݊݋‬൅
ߚଵ ‫݈݁ܽ݉݁ܨ‬௜ ൅ ߝ௜ . The subscript i indicates that the variables are estimated at the individual
level and ߝ௜ is an error term. We estimate Model 1 separately for each shipwreck.
Column 1 shows the regression estimates of ߚଵ , which we denote ߚመଵ. We note that ߚመଵ is
negative and statistically significant (p<0.01) for 11 of the 18 shipwrecks in FS. In the case of
both the Admiral Nakhimov and the Golden Gate, ߚመଵ is negative but the corresponding pvalues are somewhat higher. In three cases ߚመଵ is statistically insignificant (p>0.10). And in
two cases (the Titanic and the Birkenhead) ߚመଵ is statistically significant (p<0.01) and positive.
Next, we augment Model 1 with the variable Crew. The results, reported in column 2, do
not change much; ߚመଵ becomes positive and statistically insignificant (p>0.10) for the Admiral
Nakhimov. Also, ߚመଵ remains statistically significant (p<0.01) and positive for the Birkenhead
and the Titanic.
Column 3 reports the results from Model 1 augmented with Crew as well as the
additional control variables. The set of control variables differ between the shipwrecks
depending on which variables are available for the particular shipwreck (see Table S1). For
three shipwrecks (Birkenhead, Admiral Nakhimov, Princess of the Stars) we lack variables
other than Female and Crew. Accordingly, the cells in column 3 are denoted by n.a. (not
available) for these shipwrecks. For the remaining shipwrecks we use all available
information.
The results with respect to the Female and Crew variables are similar in terms of
statistical significance to those in column 2. In one case (the Princess Alice), the
corresponding p-value increases (p<0.076).
In Table S5 we present the results with respect to the additional control variables from
the regressions reported in Table S4, column 3. For 5 out of 10 shipwrecks we find that
children, defined as being younger than 16, or indicated as a child (Child), have a lower
survival probability than persons aged 16–50 (Adult). These results support the discussion
surrounding Fig. 1. in the main text. In fact, the only shipwrecks in which children have a
survival advantage over prime aged adults (Adults) are the Titanic and the Golden Gate.
Moreover, the results with respect to age are also in line with our prior regarding older adults
(Age>50). In 5 shipwrecks the survival rate of this subgroup is between 13 and 32 percentage
points lower than that of individuals aged 16–50. Taken together the results are in line with
the hypothesis that physical strength and mobility are important characteristics in shipwrecks.
We find no statistically significant effect of Nationality suggesting that people who
share nationality with the ship are no more likely to survive than persons of other
nationalities.
Similar to studies based on the Titanic we find evidence of a class gradient in survival
rates (3). For two ships, besides the Titanic, we find that first class passengers (1st class) have
a statistically significant (p<0.01) survival advantage over third class passenger. Being a first
class passenger rather than third class passenger increases the probability of surviving by
between 19 and 34 percentage points. Notably, for one shipwreck (Principessa Mafalda) we
find that first class passengers in fact have lower survival probability than third class
passengers.
Regarding Companionship we find that, in two out of three instances, the coefficient
estimate is negative and statistically significant (p<0.05). These findings indicate that
12
40
traveling together with someone, rather than travelling alone, is associated with a lower
survival probability (between 13 and 20 percentage points).
Tests of H1 for individual shipwrecks: probit models
Next, we show that our results are insensitive to the choice between the linear
probability model and the (non-linear) probit model, which has been used in previous studies
of the Titanic and the Lusitania. We estimate probit models (Model 2) of the form:
ܵ‫݈ܽݒ݅ݒݎݑ‬௜ ൌ ߶ሺܿ‫ ݐ݊ܽݐݏ݊݋‬൅ ߚଵ ‫݈݁ܽ݉݁ܨ‬௜ ൅ ߝ௜ ሻ, where ߶ is the cumulative standard normal
distribution function.
The ߚመଵ’s are obtained using a Maximum Likelihood estimator. However, for ease of
interpretation, as well as to make the results comparable with the results from the linear
models we present the marginal effects. The impact of a change in a regressor on the
dependent variable is calculated with the finite difference method (7). Table S4, column 4–6,
reports the marginal effects for Crew from models augmented with the same control variables
as for the linear probability model. One caveat with the probit model is that it falls short when
all or no women survive. This is the case for four shipwrecks: the Birkenhead (all of the
women on board survived), the Arctic, the Princess Victoria, and the Atlantic (all women
perished). Accordingly, the cells corresponding to these shipwrecks are denoted n.a. (not
available).
We note that the probit results with respect to Female are very similar in terms of
statistical significance to those obtained from the linear models. Likewise, the marginal
effects are similar in size to the ߚመଵ’s in column 1–3.
Tests of H2 for individual shipwrecks: linear probability models
The hypothesis that crew members have a survival advantage over passengers (H2) is
tested using the same approach as we used to test for gender differences. Table S4, column 7–
9, reports the coefficient for Crew from a set of linear probability models (LPM). We start by
estimating the following model (Model 3): ܵ‫݈ܽݒ݅ݒݎݑ‬௜ ൌ ܿ‫ ݐ݊ܽݐݏ݊݋‬൅ ߚଶ ‫ݓ݁ݎܥ‬௜ ൅ ߝ௜ . ߚመଶ is the
estimate of ߚଶ. Table S4 shows that ߚመଶ is statistically significant (p<0.01) and positive for 9
of the 18 shipwrecks. In four cases ߚመଶ is positive, but p-values slightly higher (p<0.05). These
results show that, being a crew member, compared to a passenger, is associated with higher
probability of survival. This is in line with the hypothesis that the crew members have
informational advantages over the passengers, e.g. in knowledge about escape routes. In fact,
Titanic is the only shipwreck where ߚመଶ is statistically significant (p<0.01) and negative.
To control for the influence of gender on the relationship between crew membership and
survival we augment the Model 3 with Female (this model is equivalent to Model 1
augmented with Crew). The results, reported in column 8, are very similar to those in column
7. However, for the Titanic, the coefficient changes sign. The p-value increases somewhat for
the Princess Alice. For the Norge and the Lusitania the p-values increase and becomes
statistically insignificant at conventional levels (p>0.1), when we control for Female.
Noteworthy, ߚመଶ is negative (p<0.05) for the Princess of the Stars, when we control for
Female, suggesting that crew members have a survival disadvantage compared to passengers.
We continue by estimating Model 3 with additional individual level controls. The results
with respect to Crew are presented in column 9. The general conclusion from this exercise is
that the inclusion of additional controls does not change the precision of the ߚመଶs. The
exceptions are: the Titanic, for which ߚመଶ increases and becomes statistically significant
(p<0.01), and the Princess Alice and the Estonia for which we now observe statistically
insignificant ߚመଶ’s
13
41
Tests of H2 for individual shipwrecks: probit models
Next, we switch to a probit model of the form: ܵ‫݈ܽݒ݅ݒݎݑ‬௜ ൌ ߶ሺܿ‫ ݐ݊ܽݐݏ݊݋‬൅ ߚଶ ‫ݓ݁ݎܥ‬௜ ൅
ߝ௜ ሻ. We denote this model: Model 4. We augment the model in the same way as before. The
results are reported in column 10–12. We can conclude that the probit results are similar, in
terms of statistical significance, to the results from the linear model. The marginal effects are
also very similar to the corresponding coefficient estimates in column 7Ȃ9.
Regression results for: MS, MS+Lusitania, MS+Titanic, and FS
This section supports the results reported in Table 2 in the main text. We also show how
the results change when we augment MS with the data from the Lusitania and the Titanic
separately, and together (FS). Table S6–S9 reports the results. Moreover, we report the
regression results from a set of unweighted models. See Table S10–S13.
Table S6 reports the full results of Table 2 in the main text. The results in column 1 are
generated by the following model (Model 5): ܵ‫݈ܽݒ݅ݒݎݑ‬௜ǡ௦ ൌ ܿ‫ ݐ݊ܽݐݏ݊݋‬൅ ߚଵ ‫݈݁ܽ݉݁ܨ‬௜ǡ௦ ൅
ࢾ௦ ൅ ߝ௜ǡ௦ . The subscript i indicates that the variable is measured at the individual level. ࢾ௦ is a
vector of shipwreck specific fixed effects, which is included as a control for unobservable
differences that vary between the ships but do not vary between persons within the ship. We
let ߚመଵ denote the regression estimate of ߚଵ.
The results in column 2 are generated by the model (Model 6): ܵ‫݈ܽݒ݅ݒݎݑ‬௜ǡ௦ ൌ
ܿ‫ ݐ݊ܽݐݏ݊݋‬൅ ߚଵ ‫݈݁ܽ݉݁ܨ‬௜ǡ௦ ൅ ߚଶ ‫ݓ݁ݎܥ‬௜ǡ௦ ൅ ࢾ௦ ൅ ߝ௜ǡ௦ . We let ߚመଶ denote the regression estimate
of ߚଶ.
Column 3–8 reports the separate tests of the hypotheses H3–H8, and column 9 reports
the joint test. The results are generated by the model (Model 7): ܵ‫݈ܽݒ݅ݒݎݑ‬௜ǡ௦ ൌ ܿ‫ ݐ݊ܽݐݏ݊݋‬൅
ߚଵ ‫݈݁ܽ݉݁ܨ‬௜ǡ௦ ൅ ߚଶ ‫ݓ݁ݎܥ‬௜ǡ௦ ൅ ࢽ૚ ࢄ௦ ൅ ࢽ૛ ൫ࢄ௦ ‫݈݁ܽ݉݁ܨ‬௜ǡ௦ ൯ ൅ ࢾ௦ ൅ ߝ௜ǡ௦ . In the separate tests of
H3–H8 (column 3–8) ࢄ௦ is a binary variable: WCF order, Quick, Small share of women,
More than one day voyage, Post WWI, or British ship. ࢄ௦ ‫݈݁ܽ݉݁ܨ‬௜ is the interaction between
the hypothesis specific binary variable and Female. In the joint test (column 9) ࢄ௦ is a vector
including all hypothesis specific dummies (i.e. WCF order, Quick, Small share of women,
ෝ૛ are vectors of regression
ෝ૚ and ࢽ
More than one day voyage, Post WWI, British ship). ࢽ
estimates of ࢽ૚ and ࢽ૛ .
Table S7 displays the regression results when we augment MS with the Lusitania data.
We note that the results are largely similar to those for MS. However, four coefficients
relating to our hypotheses change. First of all, we note that the coefficient for WCF
order*Female becomes statistically significant (p<0.001) in the separate test (column 3).
Second, the coefficient for Quick*Female (column 9) becomes statistically significant
(p<0.05). The sign suggest that women onboard quickly sinking ships have a survival
advantage over women onboard slowly sinking ships. This result is probably due to the fact
that the Lusitania sank in only 18 minutes and that the survival rate of women was relatively
high. Third, the coefficient for Small share of women *Female in column 9 becomes
statistically significant (p<0.10). Fourth, the coefficient for More than one day
voyage*Female also becomes statistically significant (p<0.05) in the joint test.
Table S8 displays the results when we augment MS with the Titanic data. A few things
happen. The gender gap and the crew-passenger gap decrease somewhat. The p-values
remain the same implying that the coefficients are statistically significant at the 1% level. The
result that the WCF order benefits women is strengthened, compared to for the MS.
Inspecting column 4 and 9 it becomes apparent that the coefficient on Quick*Female is
sensitive to the inclusion of the Titanic: it is statistically significant (p<0.01) negative in the
separate test and significant (p<0.01) positive in the joint test. The impact of Small share of
women on women’s survival weakens when we augment MS with the Titanic. Moreover, the
results indicate that More than one day voyage have a positive impact on the survival rate of
14
42
women in this sample: the coefficient on More than one day voyage *Female is statistically
significant both in the separate and in the joint test. Moreover, the coefficient on British
ship*Female changes sign from negative to positive suggesting that women have a relative
survival advantage on board British ships. This is probably an artifact of the Titanic being
British with a comparably high survival rate of women.
Table S9 presents the regression results for FS, i.e., the sample including data on all 18
shipwrecks. We note that our previous findings with respect to Female and Crew hold also
for this sample: the coefficients on Female and Crew are negative (p<0.001) and positive
(p<0.001), respectively. Regarding the results of the separate tests of H3–H8, we see that they
differ somewhat from those obtained for MS. The most notable discrepancy is that
Quick*Female is negative (p<0.10) and positive (p<0.05) in the separate and joints tests,
respectively. In contrast to what we found for MS, Small share of women *Female is
statistically insignificant in the separate test. The p-value increases also in the joint test.
Another apparent difference compared to the results for MS is that More than one day
voyage*Female is statistically significant in the separate test. The coefficient implies that
women have a relative survival advantage on board ships that have been on sea for more than
one day before the disaster. In the joint test, however, the corresponding estimate is
statistically insignificant at all conventional levels (p>0.10). Another discrepancy, compared
to MS, is that Post WWI*Female is negative and statistically insignificant on conventional
levels (p>0.10). In the joint test (column 9) it is however similar to Post WWI*Female for MS
(Table S6, column 9) in terms of sign and p-value. Moreover, we note that British
ship*Female is statistically insignificant (p>0.10), in both the separate and the joint tests.
We can conclude that our findings in the main text i.e., that women have a distinct
survival disadvantage compared to men and that the crew survive at a significantly higher
rate than passengers, are robust to the inclusion of data from the Lusitania and the Titanic.
Although our findings with respect to the separate tests of H3–H8 are somewhat sensitive to
the inclusion of Lusitania and the Titanic the joint, and most reliable, tests of the 8
hypotheses is robust, except for the hypotheses with respect to Quick and Small share of
women.
15
43
Table S6. Regression results for MS
VARIABLES
Female
H1
H2
H3
H4
H5
H6
H7
H8
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
-0.167
-0.126
-0.151
-0.151
-0.116
-0.154
-0.195
-0.093
-0.179
(0.008)
(0.009)
(0.011)
(0.012)
(0.012)
(0.023)
(0.013)
(0.014)
(0.068)
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
(0.009)
Crew
0.187
0.157
0.157
0.157
0.157
0.158
0.159
0.161
(0.011)
(0.014)
(0.014)
(0.014)
(0.014)
(0.014)
(0.014)
(0.014)
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
(<0.001)
WCF order
WCF order*Female
0.085
-0.187
(0.024)
(0.022)
(0.001)
(<0.001)
0.019
0.096
(0.026)
(0.041)
(0.477)
Quick
Quick*Female
(0.019)
0.091
0.321
(0.029)
(0.025)
(0.002)
(<0.001)
0.005
0.032
(0.021)
(0.043)
(0.806)
Small share of women
Small share of women*Female
(0.452)
0.218
0.181
(0.030)
(0.022)
(<0.001)
(<0.001)
-0.109
-0.050
(0.022)
(0.031)
(<0.001)
More than one day voyage
More than one day voyage*Female
(0.104)
0.100
-0.188
(0.039)
(0.030)
(0.010)
(<0.001)
0.006
0.026
(0.025)
(0.034)
(0.807)
Post WWI
Post WWI*Female
(0.443)
0.116
-0.136
(0.041)
(0.026)
(0.004)
(<0.001)
0.085
0.073
(0.019)
(0.041)
(<0.001)
British ship
British ship*Female
Constant
Observations
H1–H8
(0.074)
-0.106
-0.270
(0.041)
(0.024)
(0.009)
(<0.001)
-0.153
-0.101
(0.019)
(0.033)
(<0.001)
(0.002)
0.346
0.325
0.244
0.237
0.111
0.229
0.329
0.435
0.471
(0.020)
(0.020)
(0.015)
(0.022)
(0.023)
(0.034)
(0.020)
(0.036)
(0.044)
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
(<0.001)
10,978
10,976
10,976
10,976
10,976
10,976
10,976
10,976
10,976
R-squared
0.249
0.270
0.242
0.242
0.244
0.242
0.244
0.247
0.248
Notes. Linear probability models. The dependent variable (Survival) is binary and equals one if the person survived the disaster and zero if the
person died. Robust standard errors and p-values in parentheses below the coefficients. All specifications include controls for shipwreck
specific fixed effects. Observations in regressions in column 3−9 are weighted by the inverse of the number of individuals on board the ship.
16
44
Table S7. Regression results for MS augmented with the Lusitania
VARIABLES
Female
H1
H2
H3
H4
H5
H6
H7
H8
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
-0.147
-0.109
-0.154
-0.154
-0.106
-0.156
-0.169
-0.097
-0.261
(0.008)
(0.008)
(0.011)
(0.012)
(0.011)
(0.023)
(0.012)
(0.014)
(0.049)
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
(<0.001)
Crew
0.153
0.147
0.145
0.146
0.146
0.146
0.146
0.151
(0.010)
(0.013)
(0.013)
(0.013)
(0.013)
(0.013)
(0.013)
(0.013)
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
(<0.001)
WCF order
WCF order*Female
0.083
-0.188
(0.024)
(0.022)
(0.001)
(<0.001)
0.077
0.157
(0.021)
(0.025)
(<0.001)
Quick
Quick*Female
(<0.001)
0.144
0.319
(0.024)
(0.025)
(<0.001)
(<0.001)
0.031
0.081
(0.019)
(0.032)
(0.113)
Small share of women
Small share of women*Female
(0.012)
0.112
0.188
(0.024)
(0.022)
(<0.001)
(<0.001)
-0.122
-0.053
(0.021)
(0.031)
(<0.001)
More than one day voyage
More than one day voyage*Female
(0.082)
0.083
0.028
(0.027)
(0.023)
(0.002)
(0.218)
0.021
0.065
(0.025)
(0.028)
(0.407)
Post WWI
Post WWI*Female
(0.021)
0.119
0.082
(0.041)
(0.029)
(0.003)
(0.004)
0.056
0.111
(0.019)
(0.032)
(0.003)
British ship
British ship*Female
Constant
Observations
H1–H8
(0.001)
-0.109
-0.274
(0.041)
(0.024)
(0.007)
(<0.001)
-0.110
-0.072
(0.018)
(0.029)
(<0.001)
(0.012)
0.345
0.329
0.245
0.186
0.219
0.246
0.330
0.440
0.255
(0.020)
(0.020)
(0.015)
(0.013)
(0.015)
(0.019)
(0.020)
(0.036)
(0.031)
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
(<0.001)
12,936
12,934
12,934
12,934
12,934
12,934
12,934
12,934
12,934
R-squared
0.209
0.224
0.227
0.226
0.228
0.226
0.226
0.228
0.232
Notes. Linear probability models. The dependent variable (Survival) is binary and equals one if the person survived the disaster and zero if the
person died. Robust standard errors and p-values in parentheses below the coefficients. All specifications include controls for shipwreck
specific fixed effects. Observations in regressions in column 3−9 are weighted by the inverse of the number of individuals on board the ship.
17
45
Table S8. Regression results for MS augmented with the Titanic
VARIABLES
Female
H1
H2
H3
H4
H5
H6
H7
H8
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
H1–H8
(9)
-0.065
-0.034
-0.155
-0.091
-0.121
-0.157
-0.112
-0.099
-0.350
(0.008)
(0.009)
(0.011)
(0.012)
(0.012)
(0.023)
(0.012)
(0.014)
(0.062)
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
Crew
0.125
0.142
0.138
0.137
0.138
0.137
0.137
0.146
(0.010)
(0.013)
(0.013)
(0.013)
(0.013)
(0.013)
(0.013)
(0.013)
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
WCF order
WCF order*Female
-0.198
-0.205
(0.034)
(0.022)
(<0.001)
(<0.001)
0.254
0.371
(0.024)
(0.034)
(<0.001)
Quick
Quick*Female
(<0.001)
-0.188
0.342
(0.034)
(0.025)
(<0.001)
(<0.001)
-0.059
0.119
(0.021)
(0.038)
(0.005)
Small share of women
Small share of women*Female
(0.002)
0.209
0.202
(0.030)
(0.022)
(<0.001)
(<0.001)
0.021
0.028
(0.020)
(0.029)
(0.306)
More than one day voyage
More than one day voyage*Female
(0.341)
0.096
-0.155
(0.039)
(0.030)
(0.014)
(<0.001)
0.060
0.083
(0.025)
(0.032)
(0.016)
Post WWI
Post WWI*Female
(0.010)
0.193
-0.102
(0.034)
(0.025)
(<0.001)
(<0.001)
-0.004
0.148
(0.019)
(0.039)
(0.813)
British ship
(<0.001)
-0.127
-0.283
(0.033)
(0.024)
(<0.001) (<0.001)
British ship*Female
Constant
-0.037
-0.037
(0.019)
(0.033)
(0.050)
(0.259)
0.344
0.331
0.525
0.519
0.121
0.234
0.330
0.457
0.426
(0.020)
(0.020)
(0.028)
(0.028)
(0.023)
(0.034)
(0.020)
(0.026)
(0.043)
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
Observations
13,186
13,184
13,184
13,184
13,184
13,184
13,184
13,184
13,184
R-squared
0.195
0.205
0.225
0.219
0.218
0.219
0.218
0.218
0.228
Notes. Linear probability models. The dependent variable (Survival) is binary and equals one if the person survived the disaster and zero if
the person died. Robust standard errors and p-values in parentheses below the coefficients. All specifications include controls for shipwreck
specific fixed effects. Observations in regressions in column 3−9 are weighted by the inverse of the number of individuals on board the
ship.
18
46
Table S9. Regression results for FS
VARIABLES
Female
H1
H2
H3
H4
H5
H6
H7
H8
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
H1–H8
(9)
-0.061
-0.030
-0.156
-0.094
-0.111
-0.159
-0.098
-0.101
-0.273
(0.008)
(0.008)
(0.011)
(0.012)
(0.011)
(0.023)
(0.011)
(0.014)
(0.049)
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
Crew
0.111
0.135
0.129
0.128
0.130
0.128
0.128
0.138
(0.009)
(0.012)
(0.012)
(0.012)
(0.012)
(0.012)
(0.012)
(0.012)
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
WCF order
WCF order*Female
0.081
-0.206
(0.024)
(0.022)
(0.001)
(<0.001)
0.235
0.320
(0.020)
(0.024)
(<0.001)
Quick
Quick*Female
(<0.001)
-0.194
0.343
(0.034)
(0.025)
(<0.001)
(<0.001)
-0.034
0.071
(0.019)
(0.032)
(0.078)
Small share of women
Small share of women*Female
(0.025)
0.109
0.191
(0.024)
(0.022)
(<0.001)
(<0.001)
0.008
0.036
(0.020)
(0.029)
(0.679)
More than one day voyage
More than one day voyage*Female
(0.214)
0.082
0.034
(0.027)
(0.023)
(0.002)
(0.143)
0.069
0.043
(0.025)
(0.028)
(0.005)
Post WWI
Post WWI*Female
(0.125)
0.124
0.085
(0.041)
(0.029)
(0.002)
(0.003)
-0.020
0.110
(0.018)
(0.032)
(0.273)
British ship
British ship*Female
Constant
(0.001)
-0.114
-0.279
(0.041)
(0.024)
(0.005)
(<0.001)
-0.015
-0.065
(0.018)
(0.029)
(0.405)
(0.025)
0.344
0.332
0.247
0.525
0.223
0.249
0.331
0.445
0.244
(0.020)
(0.020)
(0.015)
(0.028)
(0.015)
(0.019)
(0.020)
(0.036)
(0.031)
(<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001) (<0.001)
Observations
15,144
15,142
15,142
15,142
15,142
15,142
15,142
15,142
15,142
R-squared
0.169
0.178
0.212
0.205
0.205
0.205
0.205
0.205
0.214
Notes. Linear probability models. The dependent variable (Survival) is binary and equals one if the person survived the disaster and zero if
the person died. Robust standard errors and p-values in parentheses below the coefficients. All specifications include controls for shipwreck
specific fixed effects. Observations in regressions in column 3−9 are weighted by the inverse of the number of individuals on board the
ship.
19
47
Results from unweighted regressions
In Table S10 we present the results of the separate tests of H3ȂH8 (column 1-6) , as well
as the results of the joint test (column 7), from models estimated without sample weights.
Since the tests of H1 and H2 in Table 2 in the main text are generated without sample
weights we omit them in the following representation. Regarding the results with respect to
MS we note from looking at column 1 that the coefficient on WCF order*Female is similar in
sign, size, and p-value to its equivalent in Table 2. This is the case also for the joint test
(column 7). In fact, the p-value for WCF order*Female is smaller than the corresponding pvalue in Table 2 (<0.001 vs. 0.019). The coefficient on Quick*Female (column 2) is, similar
to its equivalent in Table 2, statistically insignificant on all conventional levels (p>0.10).
Notable however is that the coefficient is statistically significant (p<0.01) in the joint test.
This result indicates that the longer it takes between the first indication of distress and the
sinking the higher women’s survival rate becomes. The results of the tests with respect to
Small share of women are similar in magnitude and statistical significance to those obtained
from the weighted regressions. The coefficient on More than one day voyage*Female is
statistically significant (p<0.001) in the separate test. Although this result contrasts the
corresponding result obtained from the weighted regression (Table 2) the joint tests yield
similar results, i.e. a statistically insignificant coefficient (p>0.10).
Regarding Post WWI*Female (column 5) we note that the coefficient is, still, positive
(p<0.001) in the separate test but almost twice as large as the corresponding coefficient in
Table 2 (p<0.001). The coefficient obtained in the joint test (column 7) is also relatively
large. Noteworthy, the p-value is smaller than the corresponding p-value in Table 2.
Moreover, we note that the coefficient on British ship*Female (p<0.001) in column 6 is
somewhat larger than the corresponding coefficient in Table 2 (p<0.001). This finding
remains also for the joint test (column 7).
Table S11–S13 report the results from unweighted regressions for MS+Lusitania,
MS+Titanic, and for FS. We note that these results are very similar to the results in Table S7–
S9. The most notable difference is that the coefficient for Quick*Female vary in terms of
sign, size and statistical significance. Furthermore the significance of the coefficient for
British ship*Female is somewhat sensitive to the inclusion of the Titanic and that More than
one day voyage*Female is negative and statistically significant (p<0.001) in the joint test
with respect to FS.
20
48
Table S10. Results from unweighted regressions on MS
VARIABLES
Female
Crew
WCF order
WCF order*Female
H3
H4
H5
H6
H7
H8
(1)
(2)
(3)
(4)
(5)
(6)
(7)
-0.127
-0.135
-0.103
-0.066
-0.210
-0.062
-0.231
(0.009)
(0.011)
(0.010)
(0.018)
(0.012)
(0.011)
(0.057)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
0.187
0.187
0.187
0.186
0.188
0.187
0.190
(0.011)
(0.011)
(0.011)
(0.011)
(0.011)
(0.011)
(0.011)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
0.109
-0.190
(0.038)
(0.022)
(0.004)
(<0.001)
0.017
0.180
(0.030)
(0.038)
(0.574)
Quick
Quick*Female
(<0.001)
0.108
0.186
(0.038)
(0.030)
(0.005)
(<0.001)
0.019
0.115
(0.017)
(0.034)
(0.262)
Small share of women
Small share of women*Female
(0.001)
-0.111
0.187
(0.040)
(0.022)
(0.005)
(<0.001)
-0.094
-0.003
(0.020)
(0.026)
(<0.001)
More than one day voyage
More than one day voyage*Female
(0.924)
0.120
0.274
(0.039)
(0.043)
(0.002)
(<0.001)
-0.081
-0.002
(0.020)
(0.029)
(<0.001)
Post WWI
Post WWI*Female
(0.952)
-0.123
0.172
(0.039)
(0.039)
(0.001)
(<0.001)
0.157
0.151
(0.016)
(0.034)
(<0.001)
British ship
British ship*Female
Constant
Observations
H1–H8
(<0.001)
0.240
-0.132
(0.030)
(0.029)
(<0.001)
(<0.001)
-0.178
-0.105
(0.016)
(0.027)
(<0.001)
(<0.001)
0.216
0.217
0.436
0.205
0.326
0.086
-0.002
(0.033)
(0.033)
(0.035)
(0.033)
(0.020)
(0.023)
(0.048)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(0.973)
10,976
10,976
10,976
10,976
10,976
10,976
10,976
R-squared
0.270
0.270
0.271
0.271
0.275
0.276
0.279
Notes. Linear probability models. The dependent variable (Survival) is binary and equals one if the person survived the disaster and
zero if the person died. Robust standard errors and p-values in parentheses below the coefficients. All specifications include
controls for shipwreck specific fixed effects.
21
49
Table S11. Results from unweighted regressions on MS augmented with the
Lusitania
VARIABLES
Female
Crew
WCF order
WCF order*Female
H3
H4
H5
H6
H7
H8
(1)
(2)
(3)
(4)
(5)
(6)
(7)
-0.133
-0.142
-0.083
-0.073
-0.149
-0.070
-0.277
(0.009)
(0.011)
(0.009)
(0.018)
(0.011)
(0.011)
(0.044)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
0.157
0.154
0.154
0.152
0.151
0.151
0.160
(0.010)
(0.010)
(0.010)
(0.010)
(0.010)
(0.010)
(0.010)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
-0.132
-0.144
(0.039)
(0.023)
(0.001)
(<0.001)
0.133
0.239
(0.022)
(0.023)
(<0.001)
Quick
Quick*Female
(<0.001)
0.101
0.014
(0.038)
(0.023)
(0.008)
(0.558)
0.060
0.136
(0.016)
(0.027)
(<0.001)
Small share of women
Small share of women*Female
(<0.001)
0.224
0.239
(0.030)
(0.029)
(<0.001)
(<0.001)
-0.120
-0.005
(0.020)
(0.026)
(<0.001)
More than one day voyage
More than one day voyage*Female
(0.852)
-0.096
-0.055
(0.041)
(0.034)
(0.019)
(0.103)
-0.047
0.015
(0.020)
(0.025)
(0.018)
Post WWI
Post WWI*Female
(0.548)
0.089
0.061
(0.040)
(0.028)
(0.027)
(0.032)
0.088
0.173
(0.016)
(0.028)
(<0.001)
British ship
British ship*Female
Constant
Observations
H1–H8
(<0.001)
0.226
-0.015
(0.030)
(0.028)
(<0.001)
(0.591)
-0.087
-0.090
(0.016)
(0.024)
(<0.001)
(<0.001)
0.459
0.227
0.105
0.425
0.329
0.103
0.288
(0.034)
(0.033)
(0.022)
(0.036)
(0.020)
(0.023)
(0.033)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
12,934
12,934
12,934
12,934
12,934
12,934
12,934
R-squared
0.226
0.224
0.226
0.224
0.225
0.225
0.234
Notes. Linear probability models. The dependent variable (Survival) is binary and equals one if the person survived the disaster and
zero if the person died. Robust standard errors and p-values in parentheses below the coefficients. All specifications include
controls for shipwreck specific fixed effects.
22
50
Table S12. Results from unweighted regressions on MS augmented with the
Titanic
VARIABLES
Female
Crew
WCF order
WCF order*Female
H3
H4
H5
H6
H7
H8
(1)
(2)
(3)
(4)
(5)
(6)
(7)
-0.135
0.034
-0.115
-0.078
-0.004
-0.075
-0.393
(0.009)
(0.011)
(0.010)
(0.018)
(0.012)
(0.011)
(0.054)
(<0.001)
(0.003)
(<0.001)
(<0.001)
(0.718)
(<0.001)
(<0.001)
0.147
0.127
0.131
0.126
0.126
0.128
0.150
(0.010)
(0.010)
(0.010)
(0.010)
(0.010)
(0.010)
(0.010)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
0.094
-0.245
(0.038)
(0.022)
(0.013)
(<0.001)
0.562
0.738
(0.022)
(0.028)
(<0.001)
Quick
Quick*Female
(<0.001)
0.127
0.111
(0.039)
(0.030)
(0.001)
(<0.001)
-0.164
0.175
(0.017)
(0.031)
(<0.001)
Small share of women
Small share of women*Female
(<0.001)
0.206
0.232
(0.030)
(0.022)
(<0.001)
(<0.001)
0.235
0.070
(0.019)
(0.026)
(<0.001)
More than one day voyage
More than one day voyage*Female
(0.006)
-0.102
0.042
(0.041)
(0.023)
(0.014)
(0.073)
0.057
0.033
(0.020)
(0.028)
(0.004)
Post WWI
Post WWI*Female
(0.241)
0.096
0.054
(0.040)
(0.028)
(0.018)
(0.057)
-0.063
0.243
(0.017)
(0.034)
(<0.001)
British ship
British ship*Female
Constant
Observations
H1–H8
(<0.001)
-0.100
-0.024
(0.040)
(0.028)
(0.013)
(0.391)
0.092
-0.018
(0.017)
(0.027)
(<0.001)
(0.515)
0.228
0.204
0.122
0.432
0.330
0.430
0.203
(0.032)
(0.033)
(0.022)
(0.036)
(0.020)
(0.035)
(0.030)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
13,184
13,184
13,184
13,184
13,184
13,184
13,184
R-squared
0.246
0.211
0.216
0.206
0.206
0.207
0.253
Notes. Linear probability models. The dependent variable (Survival) is binary and equals one if the person survived the disaster and
zero if the person died. Robust standard errors and p-values in parentheses below the coefficients. All specifications include
controls for shipwreck specific fixed effects.
23
51
Table S13. Results from unweighted regressions on FS
VARIABLES
Female
Crew
WCF order
WCF order*Female
H3
H4
H5
H6
H7
H8
(1)
(2)
(3)
(4)
(5)
(6)
(7)
-0.139
0.029
-0.094
-0.080
-0.003
-0.078
-0.023
(0.009)
(0.011)
(0.009)
(0.018)
(0.011)
(0.011)
(0.048)
(<0.001)
(0.010)
(<0.001)
(<0.001)
(0.798)
(<0.001)
(0.637)
0.130
0.110
0.114
0.112
0.112
0.114
0.129
(0.009)
(0.009)
(0.009)
(0.009)
(0.009)
(0.009)
(0.009)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
0.093
-0.185
(0.038)
(0.023)
(0.014)
(<0.001)
0.387
0.492
(0.019)
(0.022)
(<0.001)
Quick
Quick*Female
(<0.001)
0.123
0.085
(0.039)
(0.024)
(0.001)
(<0.001)
-0.123
-0.045
(0.016)
(0.029)
(<0.001)
Small share of women
Small share of women*Female
(0.117)
-0.113
0.246
(0.040)
(0.030)
(0.005)
(<0.001)
0.208
0.124
(0.019)
(0.025)
(<0.001)
More than one day voyage
More than one day voyage*Female
(<0.001)
0.104
-0.002
(0.038)
(0.034)
(0.007)
(0.951)
0.063
-0.149
(0.020)
(0.026)
(0.002)
Post WWI
Post WWI*Female
(<0.001)
-0.105
0.103
(0.038)
(0.028)
(0.006)
(<0.001)
-0.068
0.050
(0.016)
(0.030)
(<0.001)
British ship
British ship*Female
Constant
Observations
H1–H8
(0.095)
-0.103
-0.027
(0.040)
(0.028)
(0.011)
(0.326)
0.092
-0.124
(0.016)
(0.025)
(<0.001)
(<0.001)
0.234
0.210
0.444
0.228
0.332
0.434
0.209
(0.032)
(0.033)
(0.035)
(0.033)
(0.020)
(0.035)
(0.033)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
15,142
15,142
15,142
15,142
15,142
15,142
15,142
R-squared
0.203
0.181
0.185
0.178
0.179
0.179
0.212
Notes. Linear probability models. The dependent variable (Survival) is binary and equals one if the person survived the disaster and zero
if the person died. Robust standard errors and p-values in parentheses below the coefficients. All specifications include controls for
shipwreck specific fixed effects.
24
52
Results from regressions with respect to H6.1
In this section we present regression results with respect to H6.1, i.e. when More than
one day voyage is replaced with Small ship (see Table S14). Like More than one day voyage,
the variable Small ship could be seen as a proxy for the degree of social proximity on board
the ship. Following the arguments surrounding the discussion of H6 in the main text, Small
ship is thus hypothesized to have a positive effect on the relative survival rate of women.
Regarding MS we note that the coefficient on Small ship*Female in column 1 is statistically
insignificant. A similar result is obtained from the joint test (column 2). Turning to the
unweighted regressions we see that the coefficient from the individual test (column 3) is
statistically significant (p<0.05). The estimated effect is positive suggesting that the survival
rate of women, relative to that of men, is lower (4.5 percentage points) in shipwrecks
involving ships with a comparably small complement. However, the joint test (column 4)
yields a statistically insignificant (p>0.10) coefficient estimate. Regarding the corresponding
results for FS (column 5−8) they reveal that the number of people on board the ship indeed
have an effect on the relative survival rate of women. The coefficients on Small ship*Female
are with no exceptions negative and statistically significant (p<0.001). Taken together the
results presented above are similar to the results with respect to More than one day voyage.
25
53
Table S14. Regression result from an alternative test of H6
MS
FS
Weighted
VARIABLES
Female
Crew
Unweighted
H1–H8
H6.1
H1–H8
H6.1
H1–H8
H6.1
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
-0.145
-0.140
-0.133
-0.237
-0.077
-0.182
-0.017
-0.198
(0.009)
(0.035)
(0.009)
(0.030)
(0.009)
(0.031)
(0.009)
(0.027)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(0.049)
(<0.001)
0.156
0.158
0.189
0.191
0.124
0.131
0.108
0.127
(0.014)
(0.014)
(0.011)
(0.011)
(0.012)
(0.012)
(0.009)
(0.009)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
Quick
Quick*Female
Small share of women
Small share of women*Female
-0.130
-0.096
-0.168
0.163
(0.027)
(0.027)
(0.026)
(0.024)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
0.123
0.148
0.379
0.483
(0.047)
(0.045)
(0.025)
(0.021)
(0.009)
(0.001)
(<0.001)
(<0.001)
0.069
0.042
0.097
0.060
(0.030)
(0.029)
(0.029)
(0.020)
(0.020)
(0.150)
(0.001)
(0.002)
0.014
0.120
0.029
0.020
(0.034)
(0.025)
(0.031)
(0.023)
(0.673)
(<0.001)
(0.350)
(0.378)
0.240
0.278
0.223
0.337
(0.022)
(0.022)
(0.022)
(0.025)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
-0.044
-0.003
0.067
0.158
(0.029)
(0.026)
(0.028)
(0.024)
(0.139)
(0.907)
(<0.001)
-0.120
0.076
-0.141
0.129
-0.076
0.134
-0.305
(0.024)
(0.022)
(0.025)
(0.022)
(0.024)
(0.021)
(0.025)
(0.021)
(<0.001)
(<0.001)
(0.002)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
-0.011
-0.039
0.045
0.037
-0.091
-0.179
-0.104
-0.172
(0.023)
(0.032)
(0.022)
(0.030)
(0.023)
(0.025)
(0.022)
(0.023)
(0.638)
(0.225)
(0.047)
(0.212)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
Post WWI*Female
British ship
British ship*Female
0.139
0.118
0.145
0.140
(0.025)
(0.025)
(0.025)
(0.025)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
0.072
0.147
0.134
0.129
(0.032)
(0.028)
(0.028)
(0.026)
(0.026)
(<0.001)
(<0.001)
(<0.001)
-0.077
-0.078
-0.082
-0.141
(0.025)
(0.025)
(0.025)
(0.022)
(0.002)
(0.002)
(0.001)
(<0.001)
-0.108
-0.106
-0.096
-0.112
(0.029)
(0.025)
(0.027)
(0.024)
(<0.001)
Observations
(0.017)
-0.405
Post WWI
Constant
H1–H8
(<0.001)
WCF order*Female
Small ship*Female
Unweighted
H6.1
WCF order
Small ship
Weighted
(<0.001)
(<0.001)
(<0.001)
0.733
0.347
0.248
0.318
0.204
0.336
0.200
0.213
(0.014)
(0.023)
(0.015)
(0.023)
(0.014)
(0.023)
(0.015)
(0.022)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
(<0.001)
10,976
10,976
10,976
10,976
15,142
15,142
15,142
15,142
R-squared
0.242
0.248
0.270
0.279
0.206
0.219
0.179
0.213
Notes. Linear probability models. The dependent variable (Survival) is binary and equals one if the person survived the disaster and zero if
the person died. Robust standard errors and p-values in parentheses below the coefficients. All specifications include controls for shipwreck
specific fixed effects. Observations in regressions in column 1−2 and 5−6 are weighted by the inverse of the number of individuals on
board the ship. The joint test of H1–H8 (in column 2, 4, 6, and 8) includes H6.1 instead of H6.
26
54
References
1.
2.
3.
4.
5.
6.
7.
Joyce CA ed (2007) The World Almanac and Book of Facts 2008 (World Almanac Books).
Frey BS, Savage DA, & Torgler B (2010) Interaction of natural survival instincts and internalized
social norms exploring the Titanic and Lusitania disasters. Proceedings of the National Academy of
Science U S A 107(11):4862-4865.
Frey BS, Savage DA, & Torgler B (2011) Behavior under Extreme Conditions: TheTitanicDisaster.
Journal of Economic Perspectives 25(1):209-222.
Hall W (1986) Social class and survival on the S.S. Titanic. Social Science & Medicine 22(6):687-690.
Mawson AR (2005) Understanding mass panic and other collective responses to threat and disaster.
Psychiatry 68(2):95-113.
Becker SW & Eagly AH (2004) The heroism of women and men. American Psychologist 59(3):163178.
Cameron CA & Trivedi PK (2005) Microeconometrics: Methods And Applications (Cambridge
University Press).
27
55
Essay 2: Estate division: Equal sharing as
choice, social norm, and legal requirement
♣
Co-authored with Henry Ohlsson
♣
We would like to thank Mikael Elinder, Mikael Lindahl, Katrina Nordblom, and seminar
participants at Uppsala for helpful comments and suggestions. Financial support from Handelsbankens Forskningsstiftelser is gratefully acknowledged.
57
1 Introduction
This paper is about decisions that deceased Swedish parents made before
they died.2 The objective is to study to what extent parents divide their
wealth unequally between their children when transferring the wealth. We
are, moreover, interested in what determines if children receives more than
their siblings.
Unequal division of parental transfers is, for example, a necessary condition for theories of altruistic (dynastic) behavior to hold (Becker, 1974,
Barro, 1974). Simple versions of altruistic models of intergenerational transfers predict that total transfers will be compensatory. Children with less economic resources (consumption possibilities) will receive more transfers than
siblings with more economic resources.
There are also other models of intergenerational transfers that predict unequal division. According to the exchange model, transfers from parents
reflect the payment of services and visits provided by children (Cox, 1987).
Children who provide more of these services will receive more transfers than
siblings who provide less.
It is crucial to understand the determinants of parental property transfers
for a wide range of economic issues. Some of these are the possible effects
of fiscal policy, the determinants of savings and wealth, the equality of opportunity, and the optimal design of tax systems. In macroeconomics, for
example, the Ricardian equivalence predictions about fiscal policy inefficiency, rest on the assumption of dynastic altruistic behavior.
Parents can transfer wealth while they are alive by providing inter vivos
gifts. They can also transfer wealth post mortem as bequests. It is also possible to transfer wealth using (life) insurance policies. Parents can directly
choose to divide gifts and insurance benefits unequally between their children while unequal bequests require writing a will.
The basic theoretical models of intergenerational transfers predict how the
total transfers from parents to their children are divided between the children. They have less to say about if and how the different types of transfers,
bequests and inter vivos gifts, are unequally divided.
Most empirical studies find that unequal division of bequests is not very
common. This has been viewed as a puzzle in light of the theories of intergenerational transfers. Another general finding is that bequests are typically
divided equally between children, regardless of their incomes.
The low incidence of unequal bequest division does not, however, necessarily mean that it is uncommon to write wills. Wills may very well deal
with other issues than estate division. A will might, for example, concern
who shall get which asset without implying unequal division of the trans2
The essay is related to Ohlsson (2007) which is a pilot study using a much smaller data set
with a limited number of variables.
58
ferred values. Another example is that wills might stipulate that a particular
transferred asset should be separate property, and not joint property, of the
recipient.
For inter vivos gifts other hand, the empirical findings are that these types
of transfers tend to be unequally divided. Most studies also find that inter
vivos gifts are compensatory.3 This is a second puzzle. Is it possible to give a
theoretical basis for why parents choose inter vivos gifts instead of bequests
to make unequal transfers? There are several different explanations suggested in the literature: psychological costs, social norms, whether information about transfers is public or private, parental affection, the transfers’ role
as insurance, etc.
We use a new administrative dataset based on the estate reports for almost
70,000 Swedish widows, widowers, and divorcees deceased in 2002–2004
with positive estates and two or more children. There are several advantages
with this dataset as compared to previously used datasets:
•
•
•
•
•
The dataset has many observations and many variables. It covers all
deceased in a country during several years.
The deceased’s share of the estate can be separated from the estate
share of a previously deceased spouse not previously transferred to
heirs.
Taxable gifts during the previous ten years and taxable (life) insurance benefits are also included in the dataset.
There is information on the family relationship between the donor
and donee for each transfer. It is, therefore, possible to calculate the
transfer to each family line.
There is information on the person identity numbers of the donors
and the donees. This makes it possible to merge the dataset with
other administrative registers that have information on other demographic and economic variables.
The bequests from the deceased are unequally divided between the children in 3 percent of the cases. Disregarding small variations from equal division reduces the frequency of unequal division by definition. Summing the
bequests to children, grandchildren, and great grandchildren in each family
line gives a frequency of unequal division between family lines of 5 percent.
These shares are low compared to what has been found for other countries.
Equal sharing of the estate between legal heirs is the legal default in Sweden if there is no written will. A will is, therefore, necessary for unequal
3
Hochguertel and Ohlsson (2009) show that inter vivos gifts are motivated by reasons other
than allowing children to overcome liquidity constraints and that lifetime poorer children
receive higher transfers than their lifetime richer siblings, although the gifts do not make up
the entire difference in lifetime incomes.
59
sharing of the estate. Only 17 percent of the deceased in our sample have
written wills. This share is considerably lower than what has been found for
the United States.
The bequests from the deceased are unequally shared between the children in 12 percent of the cases when there is a will. This shows that most
written wills deal with other issues than the division of the estate between
the children. Adding the bequests to grandchildren, and great grandchildren
in each family line to the bequests to each child gives a (will) conditional
frequency of unequal division of 16 percent.
As previously mentioned, there are, however, other ways of transferring
wealth and to do it unequally. Inter vivos gifts and (life) insurance policies
can be used for this. We have information of taxable gifts during the previous ten years and taxable insurance benefits.4 This information does not,
however, covers all transfers using gifts and insurance. Gifts made more than
ten years ago, non-taxable gifts (below the annual gift tax exemption level),
and non-taxable insurance are not included. Tax non-compliance might also
be important.
Still, we believe that the information we have on taxable gifts and taxable
insurance benefits is very useful. There are taxable gifts related to slightly
more than 2 percent of the estates. In almost half of the cases with taxable
gifts, the gifts are unequally shared between the children. This is consistent
with the findings of previous empirical studies of inter vivos gifts that gifts
are unequally shared. There are taxable insurance benefits related to
2.5 percent of the estates. The insurance benefits are unequally divided in
one out of five cases with taxable insurance benefits. As far as we are aware
this is the first study to study unequal sharing of insurances.
Adding taxable gifts and taxable insurance payments to the bequests increase share of unequal division between children from 3 percent to
4 percent. Unequal division between family lines increase from 5 percent to
6 percent. The corresponding will conditional shares increase from
12 percent to 14 percent and from 16 percent to 18 percent, respectively. The
conclusion is that although few cases are affected by taxable gifts and taxable insurance benefits, these cases contribute a lot to unequal division of the
transfers from parents to their children.
The dataset is used to estimate econometric models where we test if a
number of different economic and demographic variables significantly affect
the parents’ decisions. First, we estimate models for the likelihood of un-
4
There are probably considerable amounts transferred from decedents to heirs via different
insurance arrangements. Most of this wealth does not show up in the estate inventory reports.
This is particularly true for insurance policies with premia that have been paid for with money
that already has been taxed. Some insurance policies are, however, tax-deferred. When an heir
received the benefits from such a policy, the benefit amount was added to the inheritance
amount when the inheritance tax due was calculated.
60
equal sharing and for the likelihood of writing wills. Second, we also estimate models for the inherited amounts, controlling for fixed family effects.
The main results from the estimations are:
•
The probability of a written wills is increasing in the size of the estate and in the parent’s income. The average permanent income of
the children affects the probability of a will negatively while the
within-family income dispersion between siblings affects it positively.
•
The probability of unequal division of bequests, taxable gifts and
taxable (life) insurance benefits is increasing in the size of the estate.
The average permanent income of the children affects the unequal
division probability negatively while the within-family income dispersion between siblings affects it positively.
•
Almost all children inherit the same amount as their siblings.
For the few estates where the amounts differ, the permanent incomes
of the children do not significantly affect the differences in inherited
amounts. This suggests that parents do not use transfers at death to
equalize differences in consumption possibilities across children.
Women, children living in the same municipality as the parent, and
firstborns inherit more than siblings without these characteristics. To
the extent that geographical proximity, gender and birth order are
proxies for the child’s ability to provide services to the parent these
findings are in line with the predictions of the exchange model.
The paper is structured as follows: We present data and descriptive facts
about the deceased, the estates, and the heirs in Section 2. Section 3 presents
the descriptive evidence on unequal sharing. The results from estimating
probability models for writing wills and unequal sharing are presented in
Section 4. Section 5 presents the results from estimations of models with the
inherited amount as dependent variable. Section 6 concludes. An appendix
provides additional descriptive statistics.
61
2 Data and descriptive facts
2.1 Data
The number of inhabitants in Sweden was slightly more than 9 million at the
end of 2004 according to the Population Register. About 91,000 inhabitants
had died during that year. The corresponding numbers for the previous two
years were 93,000 deceased in 2003 and 95,000 deceased in 2002.
This paper is based on data from the Belinda databases.5 Statistics Sweden
was commissioned to organize data on intergenerational transfers (estates,
inheritances, taxable gifts during the previous ten years, and insurance payments) using the Inheritance Tax Register of the Swedish Tax Agency as a
starting point. Three data sets have been produced.6
We use the dataset with information on all estates 2002–2004.7 The Tax
Agency’s Inheritance Tax Register provides economic information for all
these estates.8 This gives a schematic view of the different aspects of intergenerational transfers. The items of the estate are valued at tax values and
not at market values. The information has enough detail, however, to study
estate division and the incidence of wills. There are about 90,000 observations per year and more than 80 variables in this dataset. The Swedish inheritance tax was repealed from 2005. There are, therefore, no similar data
available from 2005 on.
When creating our working sample we have proceeded in the sequence:
•
•
•
•
All deceased during the period 2002–2004.
There is no surviving spouse, the household is exiting (the deceased
was a widow, widower, divorced, or unmarried). The civil law protects surviving spouses which implies that there in most cases is no
or only partial estate division when a married person dies. We, therefore, condition the sample on the household exiting which means
that there is an estate division.
The estate is positive; otherwise there is nothing to transfer.
The deceased has two or more children; otherwise there is no division between children.
5
Henry Ohlsson is project leader for the BELINDA project. Access to the data has been
granted to the researchers at the Department of Economics at Uppsala University associated
with project Intergenerational transfers: causes and consequences. Due to its sensitive and
confidential nature, the data cannot be exported from the closed server environment at Statistics Sweden. Data are however available, subject to the usual standard secrecy examination,
for interested researchers through Statistics Sweden’s remote access system MONA.
6
The Swedish Research Council has funded the BELINDA project.
7
One of the other two dataset has data on all taxable gifts during the period 2002–2004. The
other dataset has detailed balance sheets at death in 2004 and 2005 for representative samples.
8
It has been compulsory to file estate reports since 1734. The Tax Agency is responsible for
keeping the register since 2001.
62
Table 1 summarizes the implications of the selection criteria.9 Two thirds
are exit households; there is a surviving spouse in a third of the cases. Four
out of five deceased leave a positive estate. About half of the deceased have
two or more children. This leaves us with a sample of slightly more than one
fourth of the total number of deceased. The remaining sample of almost
70,000 deceased is still of considerable size and much bigger than the samples used in previous empirical studies.
Table 1: Selection criteria
total number of
deceased
264,715
selection criterion:
264,715
each criterion:
share of
total, %
accumulated criteria:
share of
those kept
in previous step,
%
kept dropped
yes
no
exit households
65.87
174,368
90,347
65.87
174,368
90,347
positive estate
83.85
221,964
42,751
81.52
142,140
32,228
two or more children
55.36
146,546 118,169
48.57
69,039
73,101
will
22.26
58,926 205,789
17.37
11,991
57,048
Equal sharing of the estate between legal heirs is the legal default in Sweden if there is no written will. The civil law, moreover, stipulates that half
the estate should be equally shared between legal heirs even if there is a will.
The other half of the estate can be freely bequeathed.
The wills can be of any type. Some stipulate unequal sharing, others
stipulate that property received should be separate property. Some wills are
recent, others are old. Many written wills are mutual between spouses and
concern the property rights of a surviving spouse. Such wills are included in
the estate report file when the surviving spouse passes away (Ohlsson,
2007).
A will is, therefore, a necessary, but not sufficient, criterion for unequal
division of an estate. How common are wills? Slightly more than 17 percent
of the deceased in our sample of almost 70,000 estates had written wills.10
This contrasts the estimates of the incidence of wills in the United States.
9
One explanation for why the total number of decedents in the database does not match up
with the total number of deceased in the Population Register is that it, in some cases takes
several years before the estate inventory report becomes definite.
10
People without legal heirs are more likely to have written wills.
63
Approximately 40-50 percent of the population, and as many as two thirds of
those older than 70 years, have a will (Rossi and Rossi, 1990; Lee, 2000;
Goetting and Martin, 2001; Schwartz, 1993; McGranahan, 2006).
Very few bequests in Sweden go outside the family; to other people and
to charities (Ohlsson, 2007). There is strong support for the proverb that
blood is thicker than water!11
The dataset details the names, person identity numbers of the decedents
and the heirs, as well as their relationship.12 For each decedent there is also
information on citizenship, marital status, and date of death. Relevant demographic characteristics for the heirs that do not appear in the estate reports,
such as date of birth, sex, nationality, have been collected from the Swedish
Birth Register. We have retrieved information on the level of education of
the children and the parents from the Integrated Database for Labour Market Research. Information on marital status of the children is also collected
from this data source. Data on personal income and wealth are gathered from
the Income Registers provided by the Swedish Tax Agency. The Tax
Agency collects the information directly from the relevant sources, such as
personal tax files for incomes, and financial institutions and intermediaries
for wealth. Demographic and economic variables are available for each year
over the period 1999-2009.13 Because the Belinda database does not contain
information on relationships between heirs across generations we use the
Multi-Generation Register, which contains information on all parent-child
relations in Sweden, to link children with their offspring’s (i.e. the deceased’s grandchildren).
2.2 The parents and the estates
The average age of the deceased parents was 83.6 years. Figure 1 shows the
distribution of age at death.14 In appendix A, Table A1, Column1, we present
descriptive statistics with respect to demographics and economic characteristics of the decedents. More than two thirds of the decedents, 68 percent,
were women. Concerning marital status, 79 percent of the deceased parents
were widow or widower, while 19 percent were divorced, and about 2 percent was never married. The marital status variable does not inform us about
whether the deceased was cohabiting. We can however use information the
heirs’ relationship with the deceased to conclude this. It turns out that about
3 percent of the decedents had a cohabitating spouse.
11
There is a considerable theoretical and empirical literature on charitable bequests. We find,
however, that such bequests are much rarer in Sweden than in the United States.
12
Decedents and heirs are linked through the case number assigned to each estate inventory in
the Inheritance Tax Register.
13
Information on wealth is only available up to year 2007 because the wealth tax was repealed in that year.
14
We have estimated the distribution using the kdensity command in the Stata package.
64
Figure 1: The distribution of the parent’s age at death, years
The vast majority (99 percent) of the decedents were Swedish citizens.
We have information on level of education for 83 percent of the decedents.
The majority (57 percent) had only primary education. Slightly more than 19
percent had lower or secondary education and 6 had upper secondary or post
graduate education. Two fifths of the decedents lived in either one of the
three big city counties in Sweden (Stockholm, Skåne, and Västra Götaland).
The average number of children is 2.8. There are about equally many sons
and daughters.
We have information on deceased’s taxable employment income, including pensions, for each of the three years preceding death. The mean of annual employment income averaged of over the available years is
SEK 126,000.
Table 2 reports the basic facts about the estates. The average value of the
estates of the deceased was SEK 215,000.15 This is based on the tax values of
the different assets and debts. The tax values were lower than the market
values for some assets.16 The inheritance taxation integrated taxable gifts
during the previous ten years from the deceased to the heir and taxable insurance paid by the deceased with the heir as beneficiary. Taxable gifts and
taxable insurance benefits add almost SEK 12,000 to the average estate
amount.
15
This corresponds to EUR 51,000; GBP 35,000; or USD 64,000 using the 2004 exchange
rates of 9.13 SEK/EUR, 13.46 SEK/GBP, and 7.35 SEK/USD.
16
There were several exemptions from the principle of market prices. The most important
exception concerned real estates. The tax value of this asset was supposed to be 75 percent of
the market value. Any assets that were realized by the estate manager before the actual estate
division were valued at market prices.
65
Table 2: The estates and the inheritances, SEK
mean
p10
p50
p90
p99
Standard
deviation
The parents (n=68,090):
Estate of the deceased
215,008
14,328
100,247
487,457
1,713,470
482,778
Total transfer made by the
deceased: estate of the deceased, taxable gifts, and
taxable insurance
226,836
14,622
103,816
514,919
1,809,256
528,436
Inheritance from the deceased
73,533
3,894
31,376
171,101
641,171
166,108
Total transfer received from
the deceased: inheritance from
the deceased, taxable gifts,
and taxable insurance
77,211
3,954
32,155
180,192
672,171
178,437
The children (n=190,163):
The distributions of the different measures of the estates are skewed. The
medians are less than half the means. Figure 2 shows the distribution of the
logarithm of the estate of the deceased.
Figure 2: The distribution of the logarithm of the estate of the deceased.
66
2.3 The children and the inheritances
The deceased parents have in total 190,163 children.17 We use the term family to denote the parent-children entity. The average age of the children is 54
years. Figure 4 shows the distribution of the children’s age when inheriting.
Column 5 in Table A1, Appendix A, shows that there are about equally
many men and women among the children. The vast majority (99 percent) of
the decedents were Swedish citizens. Concerning marital status, almost 57
percent of the children are married, while 16 percent are divorced, 22 percent are unmarried, and 3 percent are widow or widower. The education
level is higher among the children than among the parents: 26 percent have
only primary education, slightly more than 44 percent secondary education
and as many as 26.7 percent has upper secondary or post graduate education.18 About 42 percent of the children lived in either one of the three big
city counties and about half of the children resided in the same municipality
as the parent. The predictions of transfer theories regarding the connection
between bequests and incomes are based on permanent income. Taking the
average of taxable employment income over the three years preceding the
parent’s death gives us a proxy for the child’s permanent income.19 The
mean of this variable is almost SEK 234,000. Table A1 also shows that the
children are wealthy: the mean value of net worth is SEK 636,000.
The lower panel in Table 2 reports the basic facts about the inheritances.
All amounts are before transfers taxes were paid. The average value of the
inheritance from the deceased is SEK 73,500. Taxable gifts and taxable insurance add slightly more than SEK 4,000 to this amount. Similarly to the
estates, the distributions of the different measures of the inheritances are
skewed, as indicated by the small medians relative to the means. Relating the
means in Table 2 to the economic variables in Table A1 shows that the transfers are small relative to permanent income and net worth.
17
This number refers to the number of surviving children.
Information on education level is missing for three percent of the children.
19
We do not include income in the year of the death as it is unclear whether this is observable
to all parents.
18
67
Figure 3: The distribution of the child’s age when inheriting, years.
3 Descriptive evidence on the frequency of unequal
sharing
It is possible to calculate several different measures of the frequency of unequal sharing using the present dataset. A first, fundamental, issue is how to
think about those who have not written wills. One approach (one extreme) is
to view the decision not to write a will as a decision to divide the bequests
equally between the children. We should then calculate the frequency of
unequal sharing using all 70,000 observations in our sample, hereafter denoted Total sample. Another approach (the other extreme) is to assume that
only those who have written wills have made conscious decisions whether or
not the divide the bequests equally. We should then calculate the frequency
of unequal sharing only using the subsample of 12,000 observations with
written wills, hereafter Will sample.
A second issue to decide is which transfers to include. There are two obvious alternatives: inheritances from the deceased and total transfers (inheritances, taxable gifts, and taxable insurance benefits) from the deceased, as
these are the amounts that the deceased had the right to decide about.
A third issue is to decide how much the shares may differ before the sharing is considered to be unequal. We define unequal sharing based on the
heirs receipt relative to the mean inheritance calculated across the children in
the family. The estate is considered exactly equally divided if the standard
deviation of the within-family mean inheritance is zero. We also use two
broader definitions previously used in the literature. The first classifies deviations larger than +/-2 percent of the mean as unequal division. This is the
68
definition used in Wilhelm (1996). The second definition follows Tomes
(1988) and considers a difference between the maximum and the minimum
inheritance exceeding 25 percent of the within-family mean as unequal division.
A fourth issue is which heirs to include. Are we interested in equal sharing between the children to the deceased? This is one possibility. But is also
possible to include bequests to grandchildren and great grandchildren and
study how the estate is divided between children including their offspring, or
in other words, how the estate is divided between family lines.
Table 3 and Table 4 show how the degree of unequal sharing differs depending on the three first choices, for children and for family lines, respectively. The upper panel is for the total sample whereas the lower panel is for
the sample limited to decedents with written wills. Suppose that we only
look at the children (as in Table 3), restrict the measure to the bequests from
the deceased, and allow for a variation up to ± 25 percent without considering the division to be unequal. Then sharing is unequal in only 2 percent of
the cases in the present example. But if we instead look at total transfers
from the deceased to children and include all cases with deviations from
exact equal sharing, then sharing is unequal in 4.4 percent of the cases.
It should be noted that the observed sharing patterns reported in Table 3
and Table 4 are not necessarily the most desirable from the perspective of
the decedents. This is because the legal system in Sweden puts a boundary
on the extent to which the parent can divide his or her estate unequally.
Children are always entitled to at least the statutory share of the estate,
which is fifty percent of their legal inheritance. Hence, for a parent with two
children the most unequal distribution possible is one which leaves one child
with three-fourths of the estate and the other child with the remaining onefourth. Similarly, for a parent with three children, any particular child could
not be given less than one sixth of the estate. Unfortunately, we cannot study
how the parent would have divided the estate had her or she been granted
full testamentary freedom. What we can do however is to study the incidence
of cases for which the law is most likely to be a constraint, i.e. those for
which at least one child receive the statutory share. Our calculations imply
that only 5-7 percent of the unequally divided estates are divided in accordance with the most unequal distribution rule. Given that this corresponds to
less than one percent of the total number of estates we can conclude that the
legal context has had little influence on the observed sharing patterns.
69
Table 3: Frequency of unequal sharing, children
Definition of equal sharing:
Exact
+/- 2 %
+/- 25 %
Estate of the deceased
3.20 (2,177)
2.36 (1,608)
2.02 (1,374)
Total transfer from the deceased:
estate of the deceased, taxable
gifts, and taxable insurance
4.40 (2,995)
3.37 (2,295)
2.49 (1,691)
Taxable gifts (n=1,525)
45.57 (695)
42.16 (643)
36.98 (564)
Taxable insurance (n=1,734)
21.28 (369)
16.09 (279)
13.26 (230)
Estate of the deceased
12.22 (1,441)
10.92 (1,287)
9.11 (1,074)
Total transfer from the deceased:
estate of the deceased, taxable
gifts, and taxable insurance
14.43 (1,701)
12.71 (1,498)
9.78 (1,153)
52.20 (285)
49.26 (269)
43.96 (240)
a
Total sample (N=68,090):
Will sampleb (N=11,790):
Taxable gifts (n=546)
Taxable insurance (n=590)
24.24 (143)
19.15 (113)
15.59 (92)
Note. Share, in percent, followed by the number of cases in parentheses. N denotes
the total number of estates in the respective sample. n denotes the number of estates
in which there are gifts or insurances. a Refers to the sample of exit households with
positive estate and for which there are two, or more children. b Refers to the sample of
exit households with positive estate and for which there are two, or more children,
and a will.
70
Table 4: Frequency of unequal sharing, family lines
Definition of equal sharing:
Exact
+/- 2 %
+/- 25 %
Estate of the deceased
4.37 (2,975)
3.33 (2,270)
2.54 (1,728)
Total transfer from the deceased:
estate of the deceased, taxable gifts,
and taxable insurance
5.61 (3,824)
4.38 (2,982)
3.02 (2,057)
Taxable gifts (n=1,553)
45.27 (703)
42.69 (663)
37.41 (581)
Taxable insurance (n=2,004)
17.11 (343)
14.27 (286)
11.73 (235)
Estate of the deceased
15.99 (1,886)
13.94 (1,644)
10.37 (1,221)
Total transfer from the deceased:
estate of the deceased, taxable gifts,
and taxable insurance
18.25 (2,152)
15.84 (1,868)
11.09 (1,308)
52.22 (294)
50.27 (283)
44.58 (251)
a
Totalsample (N=68,090):
Will sampleb (N=11,790):
Taxable gifts (n=563)
Taxable insurance (n=716)
19.97 (143)
17.18 (123)
14.25 (102)
Note. Share, in percent, followed by the number of cases in parentheses. N denotes
the total number of estates in the respective sample. n denotes the number of estates
in which there are gifts or insurances. a Refers to the sample of exit households with
positive estate and for which there are two, or more children. b Refers to the sample
of exit households with positive estate and for which there are two, or more children,
and a will.
Taxable gifts are unequally shared. There are taxable gifts reported in
connection to 1,500 estates. These gifts are unequally shared in 46 percent of
the cases. The higher frequency of unequal sharing of gifts, as compared to
bequest, is in line with the results in previous studies. It should be noted
however that the taxable gifts we study here are public information in the
same way as bequests are and therefore, that the results are not directly comparable to the results based on data on self-reported inter vivos gifts, which
may have taken place with only the donor’s and the recipients knowledge.
Moreover, there are taxable insurance benefits associated with 1,700 estates.
These payments are unequally shared in 21 percent of the cases.
Regarding the frequencies reported in the lower part of the panel, it can be
seen that unequal sharing is more common among decedents with wills. This
is expected given that a will is required for unequal sharing of bequest to
71
take place. Around 17 percent of the decedents in the Total sample had a
written will when they died. Considering the frequencies of unequal sharing
with respect to the bequest from the deceased they imply that between 9 and
12 percent of these wills prescribe that the deceased prescribed unequal division of their estates.
In Table 4 we report frequencies of unequal sharing across family lines.
Concerning the share of parents in the Total sample who divide their estate
unequally according to the exact definition it is about one percentage point
higher than the corresponding share in Table 3. This discrepancy in results is
evident also for the total transfer. Interestingly, gifts are unequally divided to
a similar extent across family lines as between children, whereas the frequency of unequal sharing of insurances is relatively lower for family lines.
Turning to the frequencies for the sample limited to decedents with wills,
in the lower part of Table 4, it can be seen that that these are around 3 percentage points higher compared to the corresponding frequencies in Table 3,
implying that the implied percentage differences in unequal sharing across
family lines and between children are similar in both samples.
Most empirical studies find that unequal sharing of bequests is not very
common. Still, the shares we find are lower than most of those previously
reported. Menchik (1980), Judge and Hrdy (1992), and Norton and Taylor Jr
(2005) all study estate reports from different parts of the United States. They
report frequencies of unequal sharing in the interval 17–46 percent. Tomes
(1981, 1988) are the exceptions finding unequal division in 51–79 percent of
the estates using a combination of estate reports and a survey. This was,
however, questioned by Menchik (1988) who only found unequal sharing in
12–16 percent of the estates reports from the same time and place.
Using French estate data, Arrondel et al. (1997) report that 8 percent of
the estates are unequally divided. Wilhelm (1996) uses US federal estate tax
data where the frequency of unequal sharing is 23–31 percent, while the
corresponding frequency in a US survey based on twin register data used by
Behrman and Rosenzweig (2004) is 8 percent.
An alternative source of information is survey data on the intended division of future bequests. Dunn and Phillips (1997), McGarry and Schoeni
(1997), McGarry (1999), and Light and McGarry (2004) all use US survey
data of this type. They report unequal sharing frequencies in the interval 8–
20 percent. Horioka (2009), using Japanese data, reports that 31 percent of
the respondents who plan to leave a bequests also plan to make it unequal.
Taken together, it can be noticed that the frequencies of unequal sharing
in the current paper are substantially lower than the frequencies reported in
previous studies and in particular, those from the United States.
72
4 Econometric evidence
4.1 The probability of writing wills
We study the characteristics of people who die with written wills in this subsection. This is unlike most previous studies that have primarily focused on
the determinants of will adoption among the living. We do not know when
the will was executed or its content. Thus, we cannot say anything about the
testators’ preferences regarding how they want specific assets to be distributed.20
We first need to decide which variables to include in the econometric
specifications as potential determinants of will writing. As there are no clear
theoretical predictions on the determinants of will writing; we let previous
empirical literature guide us.21
Wealth is perhaps the most obvious potential determinant. There is no
point of writing a will if the individual has no wealth. Having wealth, on the
other hand, means that the individual has something to decide about. Given
that all decedents in our sample have positive estates, we could test for
whether the probability of a written will is increasing in the size of the estate.
McGranahan (2006) studies will writing decisions in a sample of individuals who died in Ireland between 1901 and 1905. She finds that having a
written will at death is positively correlated with estate size. Marin and Goetting (2001) report similar results with respect to net worth for a sample of
elderly Americans. Implicit evidence of will writing being increasing in
wealth is reported by Su (2008) who finds that the probability of financial
end-of-life planning (as defined as having either a will, joint ownership
through which assets are transferred, and/or a trust) among the living is positively associated with the individual’s net worth.
Another possible determinant of will writing is income. High income individuals are likely to have greater access to legal and financial advice than
low income individuals. Palmer et al. (2006) find that (household) income is
positively associated with the probability of adopting a will whereas Goetting and Martin (2001) find no effect of income on the probability of holding
a will.
Age is another candidate likely to determine will writing. It lies close at
hand to conjecture that older individuals have had more time to think about
matters regarding distribution of assets and also that they know more about
end-of-life decisions than younger individuals. This might be a product of
their own life experiences as well as those of their spouses and age peers.
20
Light and McGarry (2003) use data on mothers’ planned division of estates among children
and conclude that variety of motives come into play when wills are established.
21
Table A1 in Appendix A compares the means of the sample characteristics discussed below, across parents and children of families with and without wills.
73
Another reason for why the probability of holding a written will at death
would be increasing in age is that older individuals, as compared to younger
individuals, are more likely to have experienced life events which may have
caused them to adopt a will, such as retirement, widowhood, and the onset of
disease (Palmer et al. 2006).
The empirical evidence on the relationship between age and will writing
is, however, less clear. Rossi and Rossi (1990) and McGranahan (2006) for
instance find that having a will is positively correlated with age whereas
Goetting and Martin (2001) do not find a relationship in their sample of elderly.
Gender may also be an important determinant of will writing. There are
studies reporting that men have better financial knowledge and are more able
to plan for retirement than women (Lusardi and Mitchell, 2011). Also, Su
(2008) finds that the incidence of financial end of life planning is lower
among women, compared to among men, suggesting that women would be
less likely to have a will. The empirical evidence on will writing, however,
suggests that men and women are equally likely to both adopt and have a
will (Palmer et al., 2006; Goetting and Martin, 2001).
Moreover, it is reasonable to assume that financial and legal knowledge is
related with level of education (Lusardi and Mitchel, 2007). Goetting and
Martin find that college graduates are 1.5 times more likely to have a will
than high school graduates. A similar finding is reported by Palmer et al.
(2006) with respect to will adoption and Su (2008) find, similarly, that the
probability of end of life financial planning is increasing in the level of education.
Previous work finds that becoming widow/widower increases the probability of adopting a will (Palmer at al., 2006). Marital status, as defined as
being married, is however not related with the probability of holding a will
(Goetting and Martin, 2001). Before the reform of the Marriage Act in 1988
married persons in Sweden had incentives to write wills to secure the financial situation of surviving spouse.22 We, therefore, expect widows and widowers to be more likely to have a will than never married individuals and
divorcees. Moreover, we expect a higher incidence of wills among cohabiting decedents as they had incentives to write a will to protect the cohabiting
spouse from unnecessary financial strains as a result of the estate division.
A written will may be seen by the individual as a tool to reduce conflicts
between children in the division of the estate. Support for this hypothesis is
found in Lee (2000) and Rossi and Rossi (1990) who document that the
presence of children increases the likelihood of will writing. Although the
same logic suggests that the likelihood of a will should increase in the number of children the authors find the opposite result.
22
See Brattström and Singer (2007)
74
The characteristics of the testator’s children may also influence the decision to write a will. Schwarts (1993) concludes that social influences and, in
particular, the influences of the family are the major determinant of the testators’ behavior. Models of transfer behavior also predict that characteristics of
the potential recipients of transfers are important determinants of the donor’s
motives regarding the distribution of assets. We would expect children characteristics proposed in this literature to affect the will writing decision to the
extent that the will reflects the deceased’s desire to divide the estate unequally. Economic and demographic characteristics enter the specifications
in the form of within-family (sibling) means and within-family (sibling)
coefficients of variation (for continuous variables).23
The unit of observation in the estimations is the parent. Each parent contributes one observation to the sample used for the estimation. The model
that we estimate is as follows:
(1)
,
where
is an indicator variable taking the value one if parent i has a will,
is a vector of parental characteristics assumed to afand zero otherwise,
is a vector containing exogenous characteristics of the
fect will writing,
children, and is a random disturbance term. We estimate the model on the
Total sample.
Table 6 shows that the probability of having a will is increasing in the
size of the estate. Permanent income is also significant and positively associated with will writing and so is age. Women are more likely to have a will
than men. This is opposite to what has been conjectured in the previous literature. As expected the likelihood of a will is lower among never married
and divorcees as compared to widows/widowers. Also, the indicator for cohabiting is positive and statistically significant.
The results with respect to education are in line with those in the previous
literature: individuals with more education have a higher probability of having a will than people with lower levels of education. Having three children,
as compared to having two, does not affect the likelihood of having a will
but having four or more children reduces the likelihood. These findings are
similar to those reported in Rossi and Rossi (1990) and Lee (2000). The
gender composition of the siblings is not significantly related to will writing.
Controlling for characteristics of the children does not affect the coefficient estimates on the parental variables significantly, see column 2. A
higher average permanent income of the children appears to reduce the like23
The coefficient of variation (cv) is obtained by dividing the standard deviation of the
within-family (sibling) mean with the within-family (sibling) mean. For cases where cv is
undefined, i.e. because the within-family (sibling) mean is zero, it has been replaced with
value zero.
75
lihood of a will, whereas higher inter-sibling dispersion increases it. The
latter finding may perhaps reflect the parent’s intentions of reallocating resources towards equalization of differences in consumption possibilities
across children.
Somewhat surprisingly perhaps, child wealth does not have an as strong
association with will incidence as permanent income. The older the children
are the more likely is the parent to have a will. Higher within-family age
dispersion is positively related with will holding. Moreover, a higher share
of daughters increases the likelihood of a will. The marital status of the children (as defined as the share of married children) is not associated with the
outcome.
The exchange model predicts that the transfer will increase in the amount
of services, e.g. visits, companionship, and home production, provided by
the child to the parent (Cox, 1987). We do not have information on services
provided by children; instead we use an indicator for whether the child lives
in the same municipality as the parent. The argument here is that services are
more easily delivered when parents and children live geographically close.
Distance, as a proxy for services, has been used in previous studies on transfer behavior, see e.g. Cox and Rank (1992) and Hochguertel and Ohlsson
(2009).
The coefficient estimates indicate that a higher share of children living in
the same municipality as the deceased reduces the likelihood of a will. The
result could be interpreted as if parents who have more children providing
services find it less meaningful to compensate any particular child. It may
also indicate the parent and the children have better relationship and that the
parent feels that it is unnecessary to write a will to minimize potential conflicts. Lastly, the results show that, parents with more children having upper
secondary education or post graduate education are more likely to have a
will.
76
Table 5: Linear probability models for the likelihood of having a written
will. Total sample. (page 1/2)
Variables
1
2
Parent:
Log of estate
0.0354***
0.0339***
(0.0010)
(0.0010)
Log of permanent income
0.0931***
0.0869***
(0.0043)
(0.0045)
Age
0.0027***
0.0009**
(0.0002)
(0.0004)
Woman
0.0299***
0.0224***
(0.0035)
(0.0037)
Marital status (reference:
widow/widower):
Never married
-0.1135***
-0.1111***
(0.0111)
(0.0115)
Divorced
-0.0726***
-0.0752***
(0.0037)
(0.0038)
Cohabiting
0.3591***
0.3641***
(0.0107)
(0.0111)
Education (reference: primary education):
Lower secondary
0.0206***
0.0172***
(0.0040)
(0.0041)
Upper secondary or post graduate
0.0459***
0.0383***
(0.0076)
(0.0080)
Missing
0.0040
0.0007
(0.0043)
(0.0044)
Big city county
0.0206***
0.0189***
(0.0029)
(0.0029)
Number of children (reference: 2 children)
3 children
-0.0018
-0.0019
(0.0034)
(0.0035)
4+ children
-0.0232***
-0.0225***
(0.0038)
(0.0040)
Children are of different sex
0.0032
0.0015
(0.0031)
(0.0032)
77
Table 5. Continued (page 2/2)
1
2
Children:
Permanent income, mean
Permanent income, cv
Wealth, mean
Wealth, cv
Age, mean
Age, cv
Woman, mean
Married, mean
Same municipality as parent, mean
Upper secondary or post graduate education, mean
-0.0070***
(0.0020)
0.0080***
(0.0019)
0.0005*
(0.0003)
0.0001
(0.0002)
0.0016***
(0.0004)
0.0015**
(0.0006)
0.0228***
(0.0047)
-0.0037
(0.0045)
-0.0240***
(0.0042)
0.0324***
(0.0050)
Dep. variable, mean
0.1731
0.1715
R2
0.0740
0.0775
No of observations
68,025
66,360
Notes. The estimation is conducted on the Total sample. Monetary variables
are reported in SEK100,000. Each specification includes controls for the
deceased’s year of death. Education refers to the highest achieved level.
Permanent income is calculated as the average of taxable labor income over
the three years preceding death. mean refers to within-family mean. cv
refers to the coefficient of variation. Robust standard errors in parentheses. *
significant at the 10 percent level, ** significant at the 5 percent level, ***
significant at the 1 percent level.
4.2 The probability of unequal sharing
In this subsection we report results from estimations of linear probability
models for unequal sharing.24 We consider two measures of transfers: the
bequest from the deceased and the total transfer from the deceased. For each
transfer we consider the case of unequal sharing between children as well as
the case of unequal sharing between family lines. The models we estimate
are similar to Model (1) but with the difference that, as dependent variable
we now use indicator variables for whether the transfer is unequally divided,
as defined by each child (or family line) receiving outside +/- 2 percent of
24
We have also considered a non-linear Probit model. This is to account for the possibility
that the estimated coefficients from the linear model can imply probabilities outside the unit
interval. The coefficients estimates from the Probit model are similar to the linear probability
estimates in terms of sign and statistical significance. Also, the implied marginal effects are
quantitatively similar to the estimates from the linear model.
78
the average transfer among children (or family lines) in the family. We report results for both the total sample (Table 6) and for the sample restricted
to decedents with wills (Table 7). As in the analyses in Subsection 4.1 we
use between-family variation and each decedent contributes one observation
to the estimation sample.
Starting by examining the results for the total sample, we see that the likelihood of unequal sharing of bequest between children is increasing in the
size of the estate. Judge and Hrdy (1992), Table 8, find the same, while the
wealth variables are not significant in the estimations reported by McGarry
(1999), Table 5. The coefficient estimate, which could be interpreted as
semi-elasticity as the estate value enters the model in logarithmic form, suggests that a one percent increase in the estate increases the likelihood of unequal division by around one percentage point, or 43 percent if compared to
the baseline probability (2.3 percent).
The finding that the decision to divide unequally is positively associated
with the estate holds true also when we consider unequal sharing between
family lines as well when we consider the total transfer from the deceased. It
lies close at hand to expect that parents find less of a point to divide unequally if the total estate is small. It can also be noted that unequal sharing is
increasing in the parent’s income. This is expected given the strong correlation between income and wealth.25
We also control for other characteristics of the parent that may be correlated with her taste or ability to divide unequally. The impact of these characteristics on the likelihood of unequal sharing can be summarized as follows: Older parents are more likely to divide unequally. This is perhaps expected given that they are likely to have had more time to think about the
distribution of their estates. Men and women are as likely to divide their
estates unequally but when including previous gifts and insurances we find
that women are more likely to divide their total transfers unequally. Marital
status seems to be rather unimportant in explaining unequal division of the
estate across children, whereas in the case of family lines it appears as if
divorcees are more likely to divide unequally than are widows and widowers.26 The estimate on the indicator for presence of a cohabiting spouse is
statistically significant only with respect to unequal bequest across children.
Moreover, level of education does not seem to be related with the decision to
divide unequally between children, but is positively related with unequal
25
To account for the correlation between income and the size of the estate we have estimated
the model without the former variable. The results, which are available on request, show that
the coefficient on log estate is largely similar to the corresponding coefficient in Table 6.
26
We have tested for whether distributive decision of widows/widowers has been influenced
by the previously deceased spouse by augmenting the econometrical specifications with an
indicator taking the value one if a positive bequest from a previous deceased spouse is transferred to the heirs, and zero otherwise. The coefficient on the covariates are largely unaffected
by the inclusion of this variable.
79
sharing across family lines. Parents living in any of the three most populated
counties in Sweden are more likely to divide their estates unequally than are
parents in other counties. Parents who have four or more children, relative to
those who have two children, are less likely to divide unequally. The fact
that the children are of different sex is positively associated with unequal
sharing of bequest across family lines and with the decision regarding total
transfer.
Both the altruistic model and the exchange model predict that characteristics of the children are important determinants of the parent’s transfer behavior. Children characteristics enter the specifications as the means calculated
among the siblings as in the empirical models for will writing. For continuous variables we also include the coefficient of variation.
We find that mean permanent income of the children is unrelated with unequal division of bequest but that a higher value is negatively associated
with unequal division of the total transfer between children as well as across
family lines.
A higher inter-sibling dispersion in permanent income is, however, associated with a higher likelihood of unequal sharing of both transfers. These
findings accord with McGarry (1999) who finds that a higher dispersion in
permanent income, as approximated by the inter-sibling difference in schooling reduces the probability of equal division.27 Although one may be keen to
interpret this as evidence that altruism play a role in transfer decisions, the
theory still requires that the parent gives more to the low-income children.
Also, the result is consistent with the exchange model, as long as the parent
gives more to the low-income children, for whom the price of time of providing time intensive services is low. In Section 5 we use within-family
variation in amounts to study whether low-income children receive more or
less than high-income siblings.
Regarding wealth, the mean is statistically significant and positive in
three of the four specifications and the coefficient of variation is significant
and positive in all four specifications.28
Moreover, we find that the likelihood of unequal sharing of total transfer
is increasing in the average age of the siblings. This may indicate that, the
older the children are the better information does the parent have on the
earning abilities of the children and hence, better bases to more effectively
distribute resources among children. Unequal division, of both transfers, is
27
We have tested for heterogeneous responses with respect to the within-family dispersion in
permanent income by separating the sample with respect to different values of the coefficient
of variation. The results, which are available on request, show that the frequency of unequal
sharing is, as expected, more common in families with relatively high income inequality
(coefficient of variation>sample median) than in families with relatively low inequality (coefficient of variation<sample median). The patterns of the coefficients on the control variable
are however largely similar across the different samples.
28
Estimating the model without controls for child wealth (income) does not change the coefficients on the child income (wealth) controls.
80
also more likely when siblings differ more in age, as indicate by the positive
estimate regarding the coefficient of variation. The resemblance of these
findings with those regarding permanent income suggests that age may act as
proxy for permanent income.
The results also indicate that the probability of unequal sharing is lower if
more children are married and if more children live in the same municipality
as the parent. The latter finding is in line with the exchange model.
Table 6: Linear probability models for the likelihood of unequal sharing. Bequests and total transfers.
Total sample. (Page 1/2)
Bequest from the deceased
Total transfer from the
deceased
Children
Family lines
Children
Family lines
1
2
3
4
Parent:
Log of estate
Log of income
Age
Woman
Marital status (reference:
widow/widower):
Never married
Divorced
Cohabiting
Education (reference: primary education):
Lower secondary
Upper secondary or post graduate
Missing
Big city county
Number of children (reference: 2
children)
3 children
4+ children
Children are of different sex
0.0100***
(0.0004)
0.0121***
(0.0019)
0.0013***
(0.0002)
0.0006
(0.0016)
0.0113***
(0.0005)
0.0156***
(0.0023)
0.0011***
(0.0002)
-0.0001
(0.0019)
0.0166***
(0.0005)
0.0234***
(0.0024)
0.0014***
(0.0002)
0.0046**
(0.0019)
0.0181***
(0.0006)
0.0272***
(0.0027)
0.0010***
(0.0002)
0.0035
(0.0021)
-0.0026
(0.0056)
0.0005
(0.0018)
0.0082*
(0.0043)
0.0085
(0.0071)
0.0121***
(0.0023)
0.0051
(0.0050)
0.0004
(0.0070)
-0.0027
(0.0021)
-0.0009
(0.0048)
0.0119
(0.0082)
0.0085***
(0.0025)
-0.0042
(0.0055)
0.0025
(0.0017)
0.0049
(0.0035)
-0.0067***
(0.0018)
0.0037***
(0.0012)
0.0054**
(0.0021)
0.0108***
(0.0042)
-0.0073***
(0.0021)
0.0050***
(0.0014)
0.0029
(0.0021)
0.0087**
(0.0044)
-0.0036*
(0.0021)
0.0045***
(0.0014)
0.0056**
(0.0023)
0.0124**
(0.0048)
-0.0040*
(0.0024)
0.0067***
(0.0016)
-0.0018
(0.0015)
-0.0079***
(0.0018)
0.0020
(0.0013)
-0.0033*
(0.0017)
-0.0060***
(0.0022)
0.0038***
(0.0015)
-0.0011
(0.0018)
-0.0082***
(0.0021)
0.0032**
(0.0015)
-0.0019
(0.0020)
-0.0065***
(0.0024)
0.0052***
(0.0017)
81
Table 6. Continued (page 2/2)
1
2
3
4
-0.0010
(0.0007)
0.0092***
(0.0019)
0.0001
(0.0001)
0.0083***
(0.0013)
-0.0001
(0.0002)
0.3002***
(0.0223)
-0.0029
(0.0019)
-0.0083***
(0.0018)
-0.0032**
(0.0016)
-0.0007
-0.0006
(0.0009)
0.0087***
(0.0022)
0.0002***
(0.0000)
0.0086***
(0.0015)
0.0004*
(0.0002)
0.3632***
(0.0236)
0.0012
(0.0022)
-0.0106***
(0.0022)
-0.0070***
(0.0020)
-0.0017
-0.0029***
(0.0009)
0.0177***
(0.0024)
0.0002*
(0.0001)
0.0067***
(0.0015)
-0.0003
(0.0002)
0.3370***
(0.0234)
-0.0044**
(0.0022)
-0.0109***
(0.0021)
-0.0012
(0.0019)
0.0011
-0.0028***
(0.0009)
0.0170***
(0.0025)
0.0003***
(0.0001)
0.0070***
(0.0017)
0.0004*
(0.0002)
0.4068***
(0.0245)
-0.0001
(0.0025)
-0.0130***
(0.0025)
-0.0046**
(0.0022)
0.0006
Children:
Permanent income, mean
Permanent income, cv
Wealth, mean
Wealth, cv
Age, mean
Age, cv
Woman, mean
Married, mean
Same municipality as parent, mean
Upper secondary or post grad. education, mean
(0.0020)
(0.0024)
(0.0024)
(0.0027)
Dep. variable, mean
0.0236
0.0333
0.0337
0.0438
R2
0.0281
0.0298
0.0425
0.0434
No of observations
67,684
67,684
67,684
67,684
Notes. Definition of unequal sharing is +/- 2 percent. Monetary variables are reported in
SEK100,000. Each specification includes controls for the deceased’s year of death. Education refers
to the highest achieved level. Permanent income is calculated as the average of taxable labor income
over the three years preceding death. mean refers to within-family mean. cv refers to the coefficient
of variation. Robust standard errors in parentheses. * significant at the 10 percent level, ** significant
at the 5 percent level, *** significant at the 1 percent level.
Moving to the results for the sample of parents who had a will (Table 7)
we see that there are some clear differences compared to what we found for
the Total sample. Regarding the parental characteristics it can be noted that
age, education, place of residence, and number of children, are not significant predictors of unequal sharing. Moreover, women are, relative to men,
less likely to divide unequally both between children and across family lines.
Cohabiting is, in contrast to previously, negatively associated with unequal
division. This may be a consequence of cohabiting parents demanding fewer
services from their children, as these are being provided by the cohabiting
spouse, and therefore have fewer reasons to divide unequally.
Regarding the characteristics of the children there are three noticeable differences compared to what we found for the Total sample. First, a higher
share of daughters (Woman) is negatively associated with the likelihood of
unequal sharing. Second, the share of children living in the same municipality as the parent is unrelated with the transfer decision. Third, a higher share
of children with upper secondary or post graduate education reduces the
likelihood of unequal division of bequest.
Taken together, the results in tables 6 and 7 show that: the estate of the
parent, the spread of wealth between siblings, and the spread in age among
82
siblings are statistically significant (p<0.01) positive related with the probability of unequal sharing across transfer measures and samples.
Table 7: Linear probability models for the likelihood of unequal sharing. Bequests and Total transfers, Will sample. (Page 1/2)
Bequest from the deceased
Total transfer from the
deceased
Children
Family lines
Children
Family lines
1
2
3
4
Parent:
Log of estate
Log of income
Age
Woman
Marital status (reference:
widow/widower):
Never married
Divorced
Cohabiting
Education (reference: primary education):
Lower secondary
Upper secondary or post graduate
Missing
Big city county
Number of children (reference: 2
children)
3 children
4+ children
Children are of different sex
0.0249***
(0.0020)
-0.0110
(0.0074)
0.0014
(0.0009)
-0.0275***
(0.0073)
0.0312***
(0.0023)
-0.0111
(0.0082)
0.0009
(0.0010)
-0.0327***
(0.0081)
0.0359***
(0.0021)
0.0019
(0.0081)
0.0016*
(0.0009)
-0.0196**
(0.0078)
0.0428***
(0.0024)
0.0038
(0.0088)
0.0007
(0.0009)
-0.0268***
(0.0084)
0.0573*
(0.0300)
0.0671***
(0.0112)
-0.1037***
(0.0109)
0.0770**
(0.0327)
0.0849***
(0.0122)
-0.1271***
(0.0120)
0.0475
(0.0312)
0.0623***
(0.0115)
-0.1109***
(0.0115)
0.0674**
(0.0336)
0.0795***
(0.0124)
-0.1340***
(0.0125)
0.0036
(0.0077)
-0.0007
(0.0117)
-0.0211**
(0.0091)
-0.0014
(0.0057)
0.0093
(0.0085)
0.0126
(0.0131)
-0.0156
(0.0102)
-0.0012
(0.0064)
-0.0004
(0.0081)
0.0076
(0.0130)
-0.0210**
(0.0097)
0.0005
(0.0061)
0.0025
(0.0088)
0.0135
(0.0141)
-0.0171
(0.0107)
0.0032
(0.0067)
0.0075
(0.0071)
-0.0053
(0.0092)
0.0089
(0.0061)
-0.0026
(0.0077)
-0.0111
(0.0101)
0.0106
(0.0068)
0.0105
(0.0075)
-0.0052
(0.0096)
0.0103
(0.0065)
0.0042
(0.0081)
-0.0121
(0.0104)
0.0117
(0.0071)
83
Table 7. Continued (page 2/2)
1
2
3
4
-0.0040
(0.0025)
0.0178**
(0.0080)
0.0002
(0.0001)
0.0333***
(0.0063)
0.0005
(0.0009)
0.6447***
(0.0743)
-0.0307***
(0.0092)
-0.0371***
(0.0089)
0.0060
(0.0076)
-0.0148*
-0.0044*
(0.0026)
0.0173*
(0.0089)
0.0002
(0.0001)
0.0301***
(0.0069)
0.0017
(0.0011)
0.6651***
(0.0807)
-0.0184*
(0.0102)
-0.0438***
(0.0101)
0.0027
(0.0087)
-0.0177*
-0.0064**
(0.0026)
0.0260***
(0.0087)
0.0003*
(0.0001)
0.0319***
(0.0066)
0.0003
(0.0010)
0.6682***
(0.0766)
-0.0325***
(0.0098)
-0.0374***
(0.0094)
0.0071
(0.0082)
-0.0135
-0.0078***
(0.0027)
0.0242**
(0.0094)
0.0003*
(0.0002)
0.0294***
(0.0071)
0.0022***
(0.0008)
0.7161***
(0.0684)
-0.0189*
(0.0107)
-0.0436***
(0.0104)
0.0068
(0.0092)
-0.0151
Children:
Permanent income, mean
Permanent income, cv
Wealth, mean
Wealth, cv
Age, mean
Age, cv
Woman, mean
Married, mean
Same municipality as parent, mean
Upper secondary or post grad. education, mean
(0.0090)
(0.0101)
(0.0097)
(0.0106)
Dep. variable, mean
0.1093
0.1398
0.1274
0.1590
R2
0.0497
0.0494
0.0598
0.0618
No of observations
11,706
11,706
11,706
11,706
Notes. Definition of unequal sharing is +/- 2 percent. Monetary variables are reported in
SEK100,000. Each specification includes controls for the deceased’s year of death. Education refers
to the highest achieved level. Permanent income is calculated as the average of taxable labor income
over the three years preceding death. mean refers to within-family mean. cv refers to the coefficient
of variation. Robust standard errors in parentheses. * significant at the 10 percent level, ** significant
at the 5 percent level, *** significant at the 1 percent level.
Table 3 and Table 4 showed that taxable gifts and insurances are unequally divided to a higher extent than are the estates. We have estimated
linear probability models for unequal sharing also of these transfers. The
analyses are based on the total sample, as a will is not a necessary condition
for unequal sharing of the transfers, but limited to cases in which a gift or an
insurance benefit has been given to at least one child, or one family line.
The results are reported Table 8. For taxable gifts (columns 1 and 2) it can
be noted that the likelihood of unequal sharing is, in contrast to what we
found for the bequests and the total transfers, unrelated to the size of the
estate. Likewise, the deceased’s income has no predictive power. Another
difference is that unequal sharing of gifts is decreasing in the parent’s age.
One possible explanation for this is that we only observe gifts during the last
ten years and that; older decedents may have given gifts at an earlier stage.
Women are, relative to men, less likely to divide gifts unequally, at least
between children, and so are parents who live in a big city county. Moreover, the number of children and their gender composition appears to be
important determinants of the decision to divide gifts unequally. In accordance with what we found for bequests, higher within-family dispersion in
84
permanent income and wealth is positively associated with the probability of
unequal sharing. Other child characteristics also display similar pattern to
what we found for bequests.
Columns 3 and 4 show how the probability of dividing insurance benefits
unequally is related to parent and child characteristics. It should be remembered that the insurance benefits are received at the time of death of the deceased and could therefore be considered more similar to a bequest than to
an inter vivos gift. However since the decision to assign a person as beneficiary to the insurance policy was made earlier, the motivations may be different than those governing the estate division. On the one hand, the parent
may have based the decision on the assumption that the current needs of the
beneficiary will remain also in the future. On the other hand, the parent may
have preferences for the beneficiary over the other children and may use the
insurance as a self-control device to assure that the money is not spent on
own consumption or is transferred to the other children as gifts.
We find that unequal sharing of insurance is positively related to the size
of the estate. Income is however only a significant predictor for unequal
sharing across family lines. Older decedents and women are more likely to
divide unequally than young decedents and men. The likelihood of unequal
sharing is lower if the deceased had a cohabiting spouse.
Having four or more children, as compared to having only two, is positively associated with unequal sharing of insurance. Moreover, the permanent income of the children is a significant predictor and the direction of the
relationship is in accordance with the findings regarding unequal sharing of
bequests and gifts. Unlike to what we found for gifts, however, the coefficients on the indicator for living in same municipality as the parent are statistically significant and positive.
85
Table 8: Linear probability models for the likelihood of unequal sharing, gifts and insurances, Total
sample. (Page 1/2)
Gift
Insurance
Children
Family lines
Children
Family lines
1
2
3
4
Parent:
Log of estate
Log of income
Age
Woman
Marital status (reference:
widow/widower):
Never married
Divorced
Cohabiting
Education (reference: primary education):
Lower secondary
Upper secondary or post graduate
Missing
Big city county
Number of children (reference: 2
children)
3 children
4+ children
Children are of different sex
0.0169
(0.0145)
0.0332
(0.0290)
-0.0076**
(0.0031)
-0.0520*
(0.0297)
0.0206
(0.0145)
0.0408
(0.0291)
-0.0073**
(0.0031)
-0.0441
(0.0297)
0.0183*
(0.0101)
0.0328
(0.0205)
0.0037*
(0.0019)
0.0363*
(0.0206)
0.0292***
(0.0095)
0.0509***
(0.0189)
0.0027
(0.0018)
0.0496***
(0.0183)
-0.0875
(0.2320)
0.0333
(0.0420)
0.0285
(0.0908)
-0.0629
(0.2238)
0.0316
(0.0417)
-0.0860
(0.0854)
0.1069
(0.0824)
0.0267
(0.0254)
-0.1078***
(0.0372)
0.0288
(0.0447)
0.0424*
(0.0236)
-0.2300***
(0.0249)
0.0420
(0.0315)
0.0314
(0.0455)
0.1077***
(0.0404)
-0.0686***
(0.0250)
0.0332
(0.0316)
0.0277
(0.0452)
0.1045***
(0.0401)
-0.0669***
(0.0250)
0.0143
(0.0212)
0.0050
(0.0301)
0.0939**
(0.0388)
0.0096
(0.0180)
0.0197
(0.0192)
0.0020
(0.0263)
0.0748**
(0.0355)
0.0107
(0.0161)
0.1135***
(0.0301)
0.1151***
(0.0407)
0.0572**
(0.0257)
0.1130***
(0.0298)
0.1044**
(0.0409)
0.0539**
(0.0256)
0.0300
(0.0222)
0.0800**
(0.0327)
0.0115
(0.0185)
0.0279
(0.0201)
0.0741**
(0.0304)
0.0099
(0.0166)
86
Table 8. Continued (page 2/2)
1
2
3
4
-0.0037
(0.0118)
0.1243***
(0.0332)
0.0001
(0.0004)
0.2174***
(0.0279)
0.0026
(0.0030)
0.8627***
(0.2337)
-0.0975**
(0.0383)
-0.1097***
(0.0368)
-0.0233
(0.0342)
0.0144
-0.0050
(0.0117)
0.1075***
(0.0332)
0.0001
(0.0004)
0.2107***
(0.0279)
0.0031
(0.0030)
0.8709***
(0.2380)
-0.1085***
(0.0381)
-0.1031***
(0.0370)
-0.0283
(0.0343)
0.0248
-0.0139
(0.0093)
0.1020***
(0.0252)
0.0000
(0.0001)
0.0013
(0.0190)
-0.0061***
(0.0018)
0.7556***
(0.1310)
-0.0076
(0.0287)
-0.0044
(0.0260)
0.0386*
(0.0232)
-0.0136
-0.0167***
(0.0062)
0.0804***
(0.0217)
0.0000
(0.0001)
-0.0020
(0.0164)
-0.0043**
(0.0018)
0.5941***
(0.1202)
-0.0055
(0.0251)
0.0002
(0.0242)
0.0395*
(0.0213)
-0.0207
Children:
Permanent income, mean
Permanent income, cv
Wealth, mean
Wealth, cv
Age, mean
Age, cv
Woman, mean
Married, mean
Same municipality as parent, mean
Upper secondary or post grad. education, mean
(0.0386)
(0.0387)
(0.0242)
(0.0219)
Dep. variable, mean
0.4224
0.4276
0.1638
0.1484
R2
0.1433
0.1316
0.1366
0.1325
No of observations
1,520
1,548
1,685
1,907
Note. Definition of unequal sharing is +/- 2 percent. Monetary variables are reported in SEK100,000.
Each specification includes controls for the deceased’s year of death. Education refers to the highest
achieved level. Permanent income is calculated as the average of taxable labor income over the three
years preceding death. mean refers to within-family mean. cv refers to the coefficient of variation.
Robust standard errors in parentheses. * significant at the 10 percent level, ** significant at the 5
percent level, *** significant at the 1 percent level.
5 Inherited amounts and the characteristics of heirs
It is clear from the two previous sections that the overwhelming majority of
parents divide their estates equally between children and between family
lines. Still, it is interesting to study on what grounds the (few) parents who
make unequal transfers treat their children differently. The results in the
previous section showed, as predicted by transfer theories, that the parent’s
decision to divide the estate unequally was related to the economic circumstances of the children, and in particular with the inter-sibling dispersion in
income. However, the estimation results did not say anything about to what
degree parents use bequests to compensate for inter-sibling differences in
income.
The significant explanatory power of the coefficient of variation with respect to children’s income could, on the one hand, be seen as support for the
altruistic model which predicts that a parent who puts equal weight on each
87
child’s utility would transfer more to less well-off children.29 It could, on the
other hand, also imply that the parent has preferences for a particular child.
For example, a parent who has preferences for one child over the other(s)
may have invested heavily in that child’s education, which in turn has manifested into a higher relative income. If the parent favorites the child also with
respect to inheritance we would find, consistent with the results in the previous section, that the likelihood of unequal sharing is positively related with
the inter-sibling difference in income.
In this section we, therefore, take the analysis one step further and study
the connection between the economic circumstances of the children and the
transfer that they receive in more detail. We also consider other dimensions
along which parent may differentiate between children.
While the previous analyses used the parent as unit observation, the
analysis in this section is based on the children. Since we have information
the amounts received by all children within the family we can estimate the
impact of child’s characteristics using a within-family approach with controls for family fixed effects (see e.g. Wilhelm, 1996; McGarry, 1999).30 This
approach will account for unobserved heterogeneity across families and allow us to interpret the coefficients on the variables of interest as deviations
from the within-family mean.31 Using within-family variation rather than
between-family variation is also appealing as it is consistent with the predictions of the transfer theories. The models that we estimate take the following
form:
(2)
,
,
,
where , is the transfer, in SEK, received by child i of family f. is the
is a vector of exogenous child characchild’s permanent income, in SEK,
is a family fixed effect that varies across families, but is
teristics, and
common to all children within the same family. The fixed effect does not
only control for unobserved heterogeneity at the family level but also observable parent characteristics. The parameter of interest is . It measures
29
This is the so-called derivative condition, which implies that a child who loses an unit in
permanent income, while a sibling gains the same amount, should receive one unit more in
transfer relative to the sibling (see Cox, 1987). For tests of this condition with respect to
bequests see e.g. Wilhelm (1996), McGarry and Schoeni (1995), and Hochguertel and Ohlsson (2009) with respect to inter vivos gifts. Unlike the altruistic model the exchange model
makes no prediction about the correlation between transfer amounts and child income. It only
predicts that the probability of transfer is negatively related to child income, as a higher income implies a higher cost of the child’s time and thus a higher price of services.
30
Models using within-family variation (twins and siblings) have also been employed to study
the returns to education, see e.g. Aschenfelter and Krueger (1994) and Aschenfelter and
Zimmerman (1997).
31
In the case of two children the model reduces to a regression of the difference in incomes
between child i and his/her sibling j on the similar difference with respect to the transfer
amount.
88
how the transfer received by child i is related to her income, relative to the
within-family averages.32
The analysis is based on children whose parents chose unequal division.
That is because, if parents give to all children equally, there would be no
correlation between the transfer and the income; any deviation would be
random (McGarry, 1999). We study children of parents who have divided
the estates unequally, either between the children or between family lines,
separately.
Two outcomes are considered: the inheritance received from the deceased
and the total transfer received from the deceased. We use the “+/- 2 percent”
definition of equal sharing. Since the bequest from the deceased can only be
unequally divided if there is a will, the analysis with respect to this outcome
is based on a smaller sample of children than is the analysis with respect to
the total transfer.
The regression results are reported in Table 9. We start by reporting the
results from the analyses on children of parents who divided bequests unequally across children (Column 1). The coefficient estimate on the permanent income variable is negative but not statistically different from zero.
This is in accordance with the results in Wilhelm (1996). The magnitudes of
the estimates increase tenfold when total transfer received is used as dependent variable (Column 2). However, despite that the standard errors are
largely the same as for the previous outcome the coefficients remain statistically insignificant. These findings could be seen as evidence against the altruistic model’s prediction regarding perfect equalization which requires a
negative one-to-one relationship between the differences in incomes and
transfer amounts.33
One possible explanation for the absence of an effect is that the three year
average of income is a poor proxy for permanent income.34 The child’s net
worth may perhaps better capture her lifetime consumption possibilities.
Also, the assets which comprise net worth may be more observable to the
parent. The altruistic model predicts that parents should transfer more to
children who are less well off in terms of wealth, implying that we would
expect a negative coefficient if the theory holds up. Although the coefficient
32
We have also considered a version of Model (2) in which the transfer amount and permanent income enter in logarithmic form rather than in levels. The results in Table 8 are robust
to this change in functional form.
33
We have tested for heterogeneous responses across children from families with high and
low income inequality, similarly to what we did in Section 4.2. The hypothesis is that compensatory transfers are more common in families with relatively high income inequality than
in families with low income inequality. The results from this exercise, which are available on
request, show however that the relationship between transfer amount and permanent income is
statistically insignificant in both samples.
34
The model presented in McGarry (1999) predicts that the relationship between bequests and
current income is weak.
89
estimate is negatively signed it is statistically insignificant on conventional
levels, for both transfer measures.35
Regarding the effects of the other covariates they could provide us with
valuable information about, along which other dimensions parents may differentiate across their children. The results with respect to the bequest and
the total transfer may be summarized as follows. Older children receive
more than their siblings. This contrasts the findings in McGarry (1999).
Daughters receive more than sons. While this could be interpreted as if parents have preferences for daughters over sons (as opposed to the predictions
of Wedgewood, 1928 and Blinder, 1973) it could also be explained by the
possibility that daughters provide more services than sons and are compensated accordingly (Cox, 1987).36 An alternative explanation for why parent
might give more generously to a daughter than to a son is because the daughter’s offspring are certain to be genetic descendants (Cox, 2003).
Marital status does not have an impact on the transfer amount. Moreover,
we find that education has significant explanatory power: children with
lower secondary and upper secondary or post graduate education receive
more than siblings with only primary education. To the extent that education
is proxy for permanent income this finding supports the results with respect
to income and wealth, that inheritances are not compensatory.
In accordance with what the exchange model predicts, children living in
the same municipality as the parent receive more than their siblings. Moreover, being the oldest sibling is positively associated with the transfer
amount. This birth order effect suggests that parent’s decision confirms to
lineal geniture. The interaction between first born and female is, however,
negative implying that first born daughters receive less than first born sons,
possibly suggesting that within family inequality in heritance is explained by
primogeniture. Given the previous findings we would expect women living
in the same municipality as the parent to receive larger transfers than their
siblings. This is not what we find however. The coefficient on the interaction
term of these characteristics is statistically insignificant with respect to bequests and significant negative with respect to the total transfer.
We have also considered cases in which the transfer is divided unequally
between family lines (Columns 3 and 4). The samples over which we estimate Model (2) are larger because unequal sharing between family lines is
more common than unequal sharing across children. Nevertheless, the results
are largely similar to those appearing in Columns 1 and 2. The only apparent
35
We have tested for the independent effect of both permanent income and net worth on the
transfer amounts. The results, which are available on request, are largely similar to those
presented in Table 8.
36
Studies consistently report than women are more disproportionally involved in provision of
parental care (Coward and Dwyer, 1990; Stoller et al., 1992) and that this is probably a result
of their lower opportunity cost of time (see Ettner, 1996).
90
difference is that family lines with married children receive more than family
lines with unmarried children.
In sum, the analyses discussed previously indicate that inheritances from
parents are not compensatory in the sense that lifetime poorer children receive higher transfers than their lifetime richer siblings. While this finding
could be seen as rejection of the altruistic model it also cast some doubts on
the exchange model which predicts that the parent will purchase more services from a low-income child because the cost of that child’s time is relatively low. However, the fact that parent appear to use post mortem transfers
to compensate children who are likely to have provided them with services
may be considered as support for the exchange model.
Taken together, the results in Table 9 show that the coefficients on the indicators for Age, Woman, Same municipality as parent, and First born, are
positive and statistically significant at the one percent level in all reported
specifications.
91
Table 9: Family fixed-effects models of transfer amounts, inheritances and total transfers received.
Children
Family lines
Dependent variable:
Inheritance,
Total transInheriTotal transamounta
fer received,
tances,
fers reamountb
amounta
ceived,
amountb
1
2
3
4
Permanent income
-0.0003
-0.0185
-0.0004
-0.0156
(0.0255)
(0.0212)
(0.0188)
(0.0181)
Net worth
-0.0008
-0.0007
-0.0016
-0.0012
(0.0033)
(0.0025)
(0.0024)
(0.0019)
Age
0.0274***
0.0523***
0.0242***
0.0423***
(0.0063)
(0.0059)
(0.0054)
(0.0048)
Woman
0.2024***
0.2275***
0.1660***
0.1649***
(0.0750)
(0.0800)
(0.0587)
(0.0592)
Marital status (reference: never married):
Married
0.0409
0.0708
0.1026*
0.0925*
(0.0731)
(0.0666)
(0.0569)
(0.0485)
Divorced
0.0206
0.0316
0.0824
0.0691
(0.0822)
(0.0723)
(0.0636)
(0.0519)
Widow/widower
-0.0380
-0.1171
0.0266
-0.0283
(0.1516)
(0.1239)
(0.1118)
(0.0898)
Education (reference: primary education):
Lower secondary
0.1309**
0.1894***
0.1058**
0.1549***
(0.0572)
(0.0680)
(0.0465)
(0.0503)
Upper secondary or post graduate
0.2270***
0.1628
0.1319**
0.1201
(0.0848)
(0.1003)
(0.0664)
(0.0753)
Missing
-0.5740***
-0.3502*
-0.5222***
-0.3047**
(0.1613)
(0.1855)
(0.1455)
(0.1498)
Big city county
-0.0562
0.1704
-0.0285
0.1368
(0.0684)
(0.1236)
(0.0509)
(0.0866)
Same municipality as parent
0.2093***
0.3779***
0.1856***
0.3158***
(0.0746)
(0.0845)
(0.0609)
(0.0655)
First born
0.3593***
0.5297***
0.3037***
0.4044***
(0.1012)
(0.0843)
(0.0727)
(0.0637)
First born*Woman
-0.0672
-0.1921*
-0.0375
-0.1141
(0.1205)
(0.1058)
(0.0871)
(0.0770)
Same municipality as parent*Woman
-0.0551
-0.1812*
-0.0720
-0.1515**
(0.1021)
(0.0940)
(0.0808)
(0.0699)
No of children
3,476
6,167
4,367
8,105
No of families
1,285
2,289
1,642
2,973
2
R
0.8654
0.8906
0.9464
0.9215
Notes. aObservations are children in Will sample of families with unequally divided bequests. b Observations are children in Total sample of families with unequally divided total transfer. Definition of
unequal sharing is +/- 2 percent. Family fixed effects are included in each regression. Monetary variables are in SEK 100,000. Education refers to the highest achieved level. Permanent income is calculated as the average of taxable labor income over the three years preceding death. Robust standard
errors in parentheses. * significant at the 10 percent level, ** significant at the 5 percent level, ***
significant at the 1 percent level.
92
We have estimated Model (2) with taxable gift and taxable insurance
benefits as dependent variables. The analyses are based on children of parents who have given a gift or insurance to at least one child and who have
distributed the transfers unequally, either across children or across family
lines. The results are reported in Table 10.
Previous studies of gifts have documented that less well-off children
benefit disproportionately from transfers (McGarry and Schoeni, 1995;
McGarry, 1999; Hochguertel and Ohlsson 2009), results which are in line
with the predictions of both altruistic and exchange models.37 We do not,
however, find any indications that the gift amount is related to either the
child’s permanent income or to her wealth. This contrasts with the results
from the specifications using between-family variation (Table B1, Appendix
B) as well as with the results reported in previous work. However, since we
do not have information on when the gift was given we cannot disentangle
the effects of permanent and current income in the determination of the
transfers and therefore, test whether transfers are made in response to permanent differences in consumption or in response to liquidity constraints
(Cox, 1990; McGarry, 2000).
Concerning the additional covariates these show that children with higher
education receive more than their less educated siblings. Moreover, as for
inheritance, we find that living in the same municipality as the parent has a
positive impact on the gift amount. Likewise, firstborns receive more compared to younger siblings. The coefficients on the interactions between firstborn and woman and same municipality and woman are negative. Overall,
our results with respect to gifts display a pattern which is similar to that for
inheritances, suggesting that, as opposed to what has been demonstrated in
the previous literature, the parent’s decision regarding the distribution of
these two transfers are similar.
The results from family fixed effects models with respect to insurance
benefits are similar to the results from the corresponding specifications with
respect to bequest and gift amounts in that neither permanent income nor net
worth has statistically significant predictive power. However, unlike the
previous transfers none of the coefficients on the demographic covariates are
statistically significant. One explanation for the lack of relationship is that
there is too little within-family variation, as a consequence of the small sample of children and families, to estimate the coefficient with a sufficient degree of precision.
37
In their seminal work on inter vivos transfers, Cox (1987) and Cox and Rank (1992) found
evidence that parents tend to give more to better-off children, which in turn is inconsistent
with altruistic behavior.
93
Table 10: Family fixed-effects models of transfer amounts, gifts and insurances, total sample
Children
Family lines
Dependent variable:
Gifts,
Insurances,
Gifts,
Insurances,
amount
amount
amount
amount
1
2
3
4
Permanent income
-0.0361
-0.0910
-0.0360
-0.0816
(0.0501)
(0.0963)
(0.0490)
(0.1027)
Net worth
-0.0073
0.0018
-0.0072
0.0018
(0.0084)
(0.0041)
(0.0083)
(0.0040)
Age
0.0101
-0.0292
0.0087
-0.0269
(0.0107)
(0.0272)
(0.0106)
(0.0274)
Woman
0.1314
-0.2958
0.1315
-0.2748
(0.1404)
(0.4284)
(0.1360)
(0.4204)
Marital status (reference:
never married):
Married
-0.0951
-0.0108
-0.0934
-0.0125
(0.1005)
(0.4638)
(0.0994)
(0.4509)
Divorced
-0.0388
-0.2472
-0.0243
-0.3240
(0.1327)
(0.2964)
(0.1289)
(0.2859)
Widow/widower
-0.1073
-0.6382
-0.0922
-0.5880
(0.2558)
(0.6425)
(0.2452)
(0.7007)
Education (reference:
primary education):
Lower secondary
0.1953**
0.2796
0.2037**
0.3248
(0.0991)
(0.2256)
(0.0980)
(0.2678)
Upper secondary or
0.2613**
-0.4344
0.2668**
-0.3509
post graduate
(0.1295)
(0.4788)
(0.1270)
(0.3896)
Missing
-0.0964
-0.4228
-0.0833
-0.3589
(0.4509)
(0.4427)
(0.4509)
(0.4348)
Big city county
0.1911
1.3724
0.1784
1.3708
(0.1236)
(1.1082)
(0.1200)
(1.1106)
Same municipality as
0.4145***
0.7855
0.4030***
0.8251
parent
(0.1270)
(0.6409)
(0.1227)
(0.6792)
First born
0.3207**
-0.1675
0.2920**
-0.1513
(0.1350)
(0.3755)
(0.1313)
(0.3658)
First born*Woman
-0.3036*
-0.3761
-0.3055*
-0.3408
(0.1649)
(0.3844)
(0.1615)
(0.3579)
Same municipality as
-0.2423*
0.2781
-0.2259
0.2625
parent*Woman
(0.1435)
(0.2597)
(0.1385)
(0.2606)
No of children
1,669
646
1,717
663
No of families
643
276
663
283
0.8242
0.5365
0.8240
0.5485
R2
Notes. Observations are children in Total sample of families with a gift/insurance going to
at least one child, and who has divided the gifts/insurances unequally. Definition of unequal
sharing is +/- 2 percent. Family fixed effects are included in each regression. Monetary
variables are in SEK100,000. Education refers to the highest achieved level. Permanent
income is calculated as the average of taxable labor income over the three years preceding
death. Robust standard errors in parentheses. * significant at the 10 percent level, ** significant at the 5 percent level, *** significant at the 1 percent level.
94
6 Discussion and concluding remarks
The objective of this paper is to shed light on the motivations which are at
play when parents decide about issues regarding the distribution of their
assets when they have passed away. For this purpose we use a new administrative dataset on the estate reports for almost 70,000 Swedish widows, widowers, divorcees, and unmarried individuals deceased in 2002–2004 with
positive estates and two or more children, allowing us to study both to what
extent parents divide their estates unequally between their children and if
characteristics of the children determine whether they receive more than
their siblings.
We find that only 2-11 percent of the parents divide their estates unequally between their children. This finding could be seen as evidence
against the two most prominent models explaining bequest motives—the
altruistic model and the exchange model—which are both predicated on the
idea that parents make unequal bequests. Moreover, we do not find that the
transfer amount is correlated with the child’s economic circumstances,
measured either as permanent income or wealth. This could be seen as further evidence against the altruistic model which assumes that parents aim at
equalizing marginal utilities across children by giving larger bequests to
their least well-off children. Likewise, we do not find any relationship between the child’s economic position and taxable gifts received during the
past ten years. Although it should be noticed that parents might have compensated less advantage children earlier, with nontaxable amounts or unreported amounts, this finding strengthens our belief that the altruistic model
plays a minor role in explaining the observed transfer patterns. We do find,
however, that in families with unequally distributed estates, children who are
more likely to have provide services to the parent (either because they are
daughters or because they lived close to the parent) receive larger bequests
than their siblings and also, that these children benefit disproportionately
from taxable gifts. This could be interpreted as if, at least for some parents;
transfers are motivated by exchange.
Should the high degree of equal sharing be interpreted as if parents are
indifferent about the division of their assets, or that most bequests are “accidental”? Not necessarily. There are at least four possible reasons for why
equal sharing may reflect a deliberate choice.
First, the estate allocation is public information through the estate inventory report. This allows children to directly see how their shares compare
with their siblings’ and thereby, might interpret this as if they are loved more
or less than their siblings. Disfavored children may have reasons to consider
the parent as unjust. If the parent cares about her reputation after death, equal
treatment may be considered the most desirable outcome (Lundholm and
Ohlsson, 2000 and Bernheim and Severinov, 2003).
95
Second, parents may choose equal treatment because the alternative, unequal sharing, could lead to jealousy and conflicts among the children and
ultimately, a breakdown of the family as a social entity (Menchik, 1988 and
Wilhelm, 1996).
Third, parents might distribute their estates equally because it is less
costly and requires less effort and, therefore, may be more rational than other
distributive principles.38 The parent does not have to collect and compare
information on the financial status of the children and the parent does not
have to value the services provided by the children.
Fourth, parental bequests are received by people who are, on average, in
their fifties, a stage of life at which the most obvious financial hardships are
likely to have been solved. One possible explanation for the finding is therefore that few heirs are dependent on inheritance as a source of wealth to an
extent that parents would find it motivated to deviate from equal treatment.
Previous studies also document that equal sharing of bequests is the norm.
A noticeable difference, however, is that these studies report frequencies of
unequal sharing which are substantially higher than those reported in the
current paper, commonly around 20–40 percent. This is particularly true for
studies of the United States. Should this be interpreted as Swedes being more
equally minded than Americans? Not necessarily. We propose that differences between the countries in contextual factors, such as inheritance law,
tax treatment of transfers, income distribution, and welfare state, may explain the discrepancy in results.
Inheritance law. In Sweden, as in most other European countries, the default
of the succession law is that the estate should be divided equally between the
legal heirs. Thus, although the parent has the opportunity to follow her own
conceptions of distributive justice, by writing a will, she may feel obliged to
comply with the prevailing cultural (formal) norm, even if it contrasts with
her own self-interest (see Laitner, 1997). The legal system in the United
States on the other hand favors the testamentary freedom over the legal inheritance right. The deceased’s property rights and desires are the main focus and children inherit the parent only if there is no will which prescribes
differently, so-called intestate succession. In fact, the legislation allows parents to completely disinherit children. American parents may find it less
stigmatized to deviate from equality because of the legally sanctioned discretion to dispose wealth according to own wish.
Tax treatment of transfers. Differences in the taxation of estates, inheritances, and inter vivos gifts between the countries may also explain the differential findings. Under a progressive inheritance tax schedule, as in Swe38
Parents may economize on decision costs by following a mechanical decision rule as suggested by Elster (1989).
96
den during the period we study, parents have incentives to allocate the estate
equally as it minimizes the total tax burden, whereas under the case of estate
taxation, as in the United States, the total tax burden is invariant to the estate
division (Menchik, 1980; Laitner and Ohlsson, 2001).
Income distribution. The income distribution in Sweden is more compressed
than the income distribution in the United States. This suggests that income
inequalities within families are, or at least are perceived to be, smaller in
Sweden than in the United States and hence, that Swedish parents may find
it less required to divide unequally to achieve, what they consider to be, a
fair outcome.
Welfare state. A more extensive egalitarian welfare state in Sweden, as compared to in the United States, may lead children to having to engage less in
instrumental care of their parents and hence that, Swedish parents have
fewer reasons to discriminate between children on the grounds of equity and
reciprocity.
The Swedish society is often portrayed as equal in many aspects. In this
study, we have shown that this seems to apply even when it comes to the
distribution of wealth within families: equal division of parental bequests is
the rule rather than the exception. This rule also seems to be stronger in
Sweden than in the United States, which is often characterized as a more
unequal society. We propose some potential explanations why this is the
case, but we leave it to future studies to investigate which of them best explain the phenomenon. Our results nevertheless suggest that it is important to
account for contextual factors when comparing empirical estimates regarding intergenerational wealth transfers across countries.
97
Appendix
Appendix A: Descriptive statistics
98
68.7
99.0
1.7
20.1
78.1
2.1
58.8
18.8
5.0
17.3
42.2
2.8
1.7
19.3
78.9
3.3
57.4
19.6
5.8
17.2
43.2
2.8
(1.15)
65.4
126,218
(77,760)
215,008
(482,778)
278,784
(619,944)
235,476
181,312
65.5
121,029
2
83.7
1
83.6
(9.9)
68.0
99.0
485,591
375,911
64.4
150,992
2.7
50.7
23.2
9.5
16.6
48.3
1.6
15.4
83.0
9.2
64.9
99.2
3
83.5
Will sample
Parents
0.000
0.000
0.019
0.000
0.000
0.000
0.000
0.000
0.046
0.000
0.196
0.000
0.000
0.000
0.000
0.013
4
0.182
P-value (2-3)
233,993
(161,010)
231,596
27.0
44.9
25.4
3.0
40.7
50.4
56.5
22.4
15.3
3.3
56.6
21.9
15.6
3.2
25.9
44.4
26.7
3.0
41.8
49.6
49.4
99.6
6
54.0
245,996
20.8
41.6
33.7
4.0
47.1
45.3
56.9
19.6
16.8
3.1
51.0
99.7
7
53.8
Will sample
Children
No will
5
54.0
(10.5)
49.6
99.6
Total sample
636,132
587,816
880,074
(2,135,297)
No of obs.
68,090
56,300
11,790
190,163
158,502
31,661
Note. Indicator variables are reported in percent. Standard deviations follow from the means for indicator variables and are only reported for continuous variables in parentheses.
Net worth
Net worth at death (Total transfer of the deceased)
Net worth at death (Estate of deceased)
Children are of different sex
Permanent income
Woman
Swedish citizen
Marital status:
Married
Never married
Divorced
Widow/widower
Cohabiting
Education:
Primary
Lower secondary
Upper secondary or post graduate
Missing
Live in big city county
Live in same municipality as parent
No of children
Age
No will
Total sample
Table A1: Sample means, total sample and will sample, parents and children
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.250
0.000
0.000
0.259
0.000
0.013
8
0.010
P-value (6-7)
References
Arrondel, L., Masson, A., and Pestieau, P. (1997). “Bequest and inheritance: Empirical issues and France–U.S. comparison,” in Erreygers, G. and Vandevelde,
T. (eds.), Is Inheritance Legitimate? Ethical and Economic Aspects of Wealth
Transfers. Springer Verlag.
Arrondel, L., and A. Masson (2002). “Altruism, exchange or indirect reciprocity:
What do the data on family transfers show?” DELTA Working Paper 2002-18.
Ashenfelter, O., and A. Krueger (1994). “Estimates of the economic return to
schooling from a new sample of twins,” American Economic Review 84, 1157–
1174.
Ashenfelter, O., and D. J. Zimmerman (1997). “Estimates of the returns to schooling
from sibling data: Fathers, sons, and brothers,” The Review of Economics and
Statistics, 79, 1–9.
Behrman, J. R. and Rosenzweig, M. R. (2004). “Parental allocation to children: New
evidence on bequest differences among siblings,” Review of Economics and
Statistics, 86, 637–640.
Bernheim, B. D. and S. Severinov (2003). “Bequests as signals: An explanation for
the equal division puzzle,” Journal of Political Economy, 111, 733–764.
Blinder, A. S. (1973). “A model of inherited wealth,” Quarterly Journal of Economics, 87, 608–626.
Brattström, M. and A. Singer (2007). Rätt arv: Fördelning av kvarlåtenskap. Iustus
Förlag.
Coward, R. and Dwyer, S. (1990). “The association of gender, sibling network composition, and patterns of parent care by adult children,” Research on Aging 12,
158–181.
Cox, D. (1987). “Motives for private income transfers,” Journal of Political Economy, 95, 508–546.
Cox, D. (1990). “Intergenerational transfers and liquidity constraints,” The Quarterly
Journal of Economics, 105, 187–217.
Cox, D. (2003). “Fathers, mothers, sons, and daughters: how do people decide to
allocate transfers among family members?” in A. H. Munnell and A. Sundén
(eds.), Death and Dollars: the Role of Gifts and Bequests in America. Brookings
Institution Press.
Dunn, T.A. and Phillips, J.W. (1997). “The timing and division of parental transfers
to children,” Economics Letters, 54, 135–137.
Elster, J. (1989). “Social norms and economic theory,” Journal of Economic Perspectives, 3, 99–117.
Ettner, S. L. (1996). “The opportunity cost of elder care,” Journal of Human Resources, 31, 189–205.
Goetting, M.A., and Martin, P. (2001). “Characteristics of older adults with written
wills,” Journal of Family and Economic Issues, 22, 243-264.
Hochguertel, S., and Ohlsson, H. (2009). “Compensatory inter vivos gifts,” Journal
of Applied Econometrics, 24, 993-1023.
Horioka, C.Y. (2009). “Do bequests increase or decrease wealth inequalities?” Economics Letters, 103, 23-25.
Judge, D.S. and Hrdy, S.B. (1992). “Allocation of accumulated resources among
close kin: Inheritance in Sacramento, California, 1890–1984,” Ethology and Sociobiology, 13, 495–522.
Laitner, J. (1997). “Intergenerational and interhousehold economic links,” in
Rosenzweig, M. R. and Stark, O. (eds.), Handbook of Population and Family
Economics. North-Holland.
100
Laitner, J. and Ohlsson, H. (2001). “Bequest motives: a comparison of Sweden and
the United States,” Journal of Public Economics, 79, 205–236.
Lee, J. (2000). “Planned bequests: Decision to have a will and distributive preferences among the very old,” Dissertations Abstracts International, DAI-A 62/06,
p. 2194. (UMI No. 3018236).
Light, A. and McGarry, K. (2004). “Why parents play favorites: Explanations for
unequal bequests,” American Economic Review, 94, 1669–1681.
Lundholm, M. and Ohlsson, H. (2000). “Post mortem reputation, compensatory gifts
and equal bequests,” Economics Letters, 68, 165–171.
McGarry, K. (1999). “Inter vivos transfers and intended bequests,” Journal of Public
Economics 73, 321–351.
McGarry, K. (1995). “Transfer behavior within the family: Results from the Asset
and Health Dynamics Survey,” NBER Working Paper 5099.
McGarry, K. (2012). “Dynamic aspects of family transfers,” NBER Working Paper
18446.
McGarry, K. and Schoeni, R. F. (1995). “Transfer behavior in the Health and Retirement Study: Measurement and the redistribution of resources within the family,” Journal of Human Resources, 30, 184–226.
McGarry, K. and Schoeni, R. F. (1997). “Transfer behavior within the family: Results from the Asset and Health Dynamics Survey,” Journals of Gerontology,
52B, 82–92.
McGranahan, L. (2006). “Will writing and bequest motive: Early 20th century Irish
evidence,” Working Paper 2006-18, Federal Reserve Bank of Chicago.
Menchik, P. L. (1980). “Primogeniture, equal sharing and the U.S. distribution of
wealth,” Quarterly Journal of Economics, 94, 299–316.
Menchik, P. L. (1988). “Unequal estate division: Is it altruism, reverse bequests, or
simply noice?” in Kessler, D. and Masson, A. (eds.), Modeling the Accumulation and Distribution of Wealth. Clarendon Press.
Norton, E. C. and Taylor, Jr., D. H. (2005). “Equal division of estates and the exchange motive,” Journal of Aging and Social Policy, 17, 63–82.
Ohlsson, H. (2007). “The equal division puzzle - empirical evidence on intergenerational transfers in Sweden,” Working Paper No 2007:10, Department of Economics, Uppsala University.
Palmer, L. Bhargava, V., and G.-S. Hong (2006). “Will adoption and life events
among older adults,” Financial Services Review, 15: 281–295.
Schwartz, T. P. (1993). “Testamentary behavior: issues and evidence about individuality, altruism and social influences,” The Sociological Quarterly, 34, 337–
355.
Stoller, E. P., Foster, L. E. and Duniho, T. S. (1992). “Systems of parent care within
sibling networks,” Research on Aging 14, 28–49.
Su, Y.-T. (2008). “Looking beyond retirement: patterns and predictors of formal
end-of-life planning among retirement age individuals,” Journal of Family Economic Issues, 29, 654–673.
Tomes, N. (1981). “The family, inheritance, and the intergenerational transmission
of inequality,” Journal of Political Economy, 89, 928–958.
Tomes, N. (1988). “Inheritance and inequality within the family: Equal division
among unequals, or do the poor get more?” in Kessler, D. and Masson, A.
(eds.), Modeling the Accumulation and Distribution of Wealth. Clarendon Press.
Wedgewood, J. (1928). “The influence of inheritance on the distribution of wealth,”
Economic Journal, 38, 38–55.
Wilhelm, M. O. (1996). “Bequest behavior and the effect of heirs’ earnings: Testing
the altruistic model of bequests,” American Economic Review, 86, 874–892.
101
Essay 3: The impact of inheritances on heirs’
labor and capital income
♦
Co-authored with Mikael Elinder and Henry Ohlsson
♦
This essay is published as “The Impact of Inheritances on Heirs’ Labor and Capital Income,” The B.E. Journal of Economic Analysis & Policy: Vol. 12: Iss. 1 (Contributions),
Article 61 (2012). See www.reference-global.com. We would like to thank Adrian Adermon,
Niclas Berggren, Sven-Olov Daunfeldt, Susanne Ek, Henrik Jordahl, David Joulfaian, Verena
Kley, Wojciech Kopczuk, Ben Marx, Jukka Pittilä, Håkan Selin, Erik Spector, Daniel
Waldenström, and the members of the UCFS scientific advisory board as well as seminar
participants at Columbia University, Uppsala University, Ratio; Stockholm, IFN; Stockholm,
the 2010 IIPF Congress, the 2010 Swedish Economics Meeting, and the 2010 Öregrund
workshop in empirical economics for their valuable comments and suggestions. We are grateful to Johanna Westlin and Vilhelm Ax for excellent research assistance. Financial support
from the Jan Wallander and Tom Hedelius Foundation is gratefully acknowledged. Some of
the work was done when Erixson enjoyed the hospitality of Department of Economics, Columbia University, New York. And some of the work was done when Ohlsson enjoyed the
hospitality of School of Economics, UNSW, Sydney and Department of Economics, University of Melbourne during his sabbatical. Financial support from the Wenner-Gren Foundations
is gratefully acknowledged.
103
The B.E. Journal of Economic
Analysis & Policy
Contributions
Volume 12, Issue 1
2012
Article 61
The Impact of Inheritances on Heirs’ Labor
and Capital Income
Mikael Elinder∗
Oscar Erixson†
Henry Ohlsson‡
Uppsala University and IFN, Stockholm, [email protected]
Uppsala University, [email protected]
‡
Uppsala University, [email protected]
∗
†
Recommended Citation
Mikael Elinder, Oscar Erixson, and Henry Ohlsson (2012) “The Impact of Inheritances on Heirs’
Labor and Capital Income,” The B.E. Journal of Economic Analysis & Policy: Vol. 12: Iss. 1
(Contributions), Article 61.
DOI: 10.1515/1935-1682.3324
c
Copyright 2012
De Gruyter. All rights reserved.
104
The Impact of Inheritances on Heirs’ Labor
and Capital Income∗
Mikael Elinder, Oscar Erixson, and Henry Ohlsson
Abstract
The objective of this paper is to study when and how much labor and capital income of heirs
respond to inheritances. We estimate fixed effects models following direct heirs, inheriting in
2004, during the years 2000–2008 using Swedish panel data. Our first main result is that the more
the heir inherits, the lower her labor income becomes. This labor income effect appears in the
years after the heir had inherited and is stronger for old heirs than for young heirs. We also find
evidence of anticipation effects that occur before the actual transfer. Our second main result is that
the more the heir inherits, the higher her capital income becomes. This effect only appears in the
years after receiving the inheritance. It seems to be dissipating after a couple of years.
KEYWORDS: inheritances, windfall gains, labor income, capital income, anticipation
∗
We would like to thank Adrian Adermon, Niclas Berggren, Sven-Olov Daunfeldt, Susanne Ek,
Henrik Jordahl, David Joulfaian, Verena Kley, Wojciech Kopczuk, Ben Marx, Jukka Pirttilä,
Håkan Selin, Erik Spector, Daniel Waldenström, and the members of the UCFS scientific advisory
board as well as seminar participants at Columbia University, Uppsala University, Ratio; Stockholm, IFN; Stockholm, the 2010 IIPF Congress, the 2010 Swedish Economics Meeting, and the
2010 Öregrund workshop in empirical economics for their valuable comments and suggestions.
We are grateful to Johanna Westlin and Vilhelm Ax for excellent research assistance. Financial support from the Jan Wallander and Tom Hedelius Foundation is gratefully acknowledged.
Some of the work was done when Erixson enjoyed the hospitality of Department of Economics,
Columbia University, New York. And some of the work was done when Ohlsson enjoyed the
hospitality of School of Economics, UNSW, Sydney and Department of Economics, University
of Melbourne during his sabbatical. Financial support for the sabbatical from the Wenner-Gren
Foundations is gratefully acknowledged.
105
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
1
Introduction
Decedents in Europe and the United States have become wealthier and wealthier,
leaving larger bequests to their heirs.1 Knowledge about behavioral effects of inheritances is essential for a wide range of economic questions. The distribution
effects of intra-family wealth transfers and effects of inheritance and estate taxation, for instance, depend on how inheritances affect labor supply, consumption and
savings decisions.2 An inheritance tax reduces the amount received by the heir.
If inheritances reduce labor supply, then a higher inheritance tax may yield higher
income tax revenues. The idea that inheritances depress work effort and encourage
spendthrift behavior has been labeled “the Carnegie conjecture” in the economic
literature (Holtz-Eakin, Joulfaian, and Rosen, 1993). There are, however, few studies on behavioral responses to inheritances because micro data on inheritances are
not easily available. The existing studies have with rare exceptions focused on the
United States and used survey data or register-based data covering only the very
wealthy.3
Sweden differs from the United States in at least two important respects.
First, decedents cannot disinherit their children. Second, labor income taxes are
higher. These institutional differences are likely to affect the responses to inherited
wealth.
We contribute to this literature by using Swedish register-based panel data to
study how inheritances affect labor and capital income. Special emphasis is placed
on the timing of these responses. A standard life-cycle model of consumption predicts that the timing and magnitude of behavioral responses to inheritances critically
depend on whether the inheritance is anticipated or not. Responses in consumption
and labor supply will be smoothed over the entire lifetime if inheritances are anticipated. Responses will, on the other hand, take place after receiving the inheritance
if the heir does not anticipate the inheritance.
Our data cover individuals who received inheritances in 2004 and include
information on the inherited amount, family characteristics, and the heirs’ labor and
capital income for the period 2000–2008. The average inheritance in our sample
1 For an excellent description of the evolution of inheritances in France, see Piketty (2011), for
estate tax revenue in the United States, see Joulfaian (2011), and for inheritance tax revenue in
Sweden, see Ohlsson (2011).
2 Kopczuk (2009) discusses the implications of transfer taxes. Horioka (2009) discusses inequality aspects of inheritances and bequests.
3 Holtz-Eakin et al. (1993), Joulfaian and Wilhelm (1994), Weil (1994), Joulfaian (2006), Brown,
Coile, and Weisbenner (2010) are studies from the United States on the effects of inheritances on
labor supply, consumption and savings. Faria and Wu (2012) study the effect of inheritance on labor
supply of entrepreneurs in the United Kingdom.
Published by De Gruyter, 2012
1
106
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
amounts to about SEK 300,000.4 This is significantly smaller than the inheritances
examined in the earlier studies.
The panel structure of our data allows us to estimate fixed effect models.
We can in this way account for unobservable individual characteristics, such as
taste for leisure and risk aversion, which may correlate with both the inheritance
and the outcome variables. The length of our panel also allows us to consider that
the timing of responses to inherited wealth may be different from responses to other
wealth shocks. For instance, the deceased’s health has usually deteriorated before
the death. Important behavioral responses could, therefore, also occur before the
demise if heirs spend time caring for their parents.5 Mourning may also depress
work effort. Thoughts about how to incorporate inheritances optimally into the
allocation of assets may be put aside for similar reasons.6 This discussion also
suggests that it would be inappropriate to compare responses in heirs’ labor and
capital income to those of individuals who have not lost a relative. We therefore
limit the analysis to individuals who have received inheritances and use variation in
the inherited amount to estimate responses in labor and capital income.
We present estimates of the marginal propensity to earn labor income out
of wealth, also known as the marginal propensity to consume leisure.7 It may take
time to adjust labor supply in response to a wealth shock. Responses may also occur
before receiving the inheritance if the heirs anticipate it. Therefore, it is important to
trace the dynamics of the responses from both a theoretical and a policy perspective.
We present estimates for the four years before inheriting, the year when the
inheritance is received, and for the four years after inheriting. We find substantial
negative impacts of inheritances on labor income in each of the four years after
inheriting. On the other hand, there are no effects in the year when the inheritance
4 The exchange rate has fluctuated around 7 SEK/USD and 9 SEK/EUR during the studied period.
A USD bought more SEK in the beginning of our sample period and fewer in the end. The SEK/EUR
rate was comparatively stable.
5 Studies have documented a negative relation between the provision of informal care to elderly
parents and labor market outcomes of the children. The outcomes are labor force attachment, earnings, hours of work, etc., see Ettner (1995), Bolin, Lindgren, and Lundborg (2008), Fevang, Kverndokk, and Røed (2012).
6 Studies have found that the death of a parent is associated with severe grief among adult children - presumably with worse performance at work (Umberson, Wortman, and Kessler, 1992, Umberson and Chen, 1994, Kessler, 1997, Bennedsen, Pérez-González, and Wolfenzon, 2010). Schulz,
Mendelsohn, Haley, Mahoney, Allen, Zhang, Thompson, and Belle (2003) find that individuals who
provided informal care to their dying parents are likely to show depressive symptoms. They also
increase their consumption of antidepressant medicines immediately after the bereavement. A year
after the demise the symptoms had, however, returned to levels lower than those prior to the death.
7 Effects of wealth shocks on labor income also complement the literature on labor income elasticities that focuses on changes in income taxes, Feldstein (see 1995, 1999), Gruber and Saez (see
2002), Kopczuk (see 2005), Blomquist and Selin (see 2010), Saez, Slemrod, and Giertz (see 2012).
2
107
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
is received or the four years before. Our estimates of the marginal propensity to
earn labor income after receiving an inheritance range from -0.04 to -0.09. This
suggests that labor income decreases by an amount corresponding to 4–9 percent of
the wealth increase. The labor income effect is stronger for old heirs than for young
heirs.
Our estimates are higher than what has been reported in previous studies
of labor income responses to wealth shocks. The magnitudes in these studies suggest that labor income decreases by a corresponding 1 to 2 percent of the wealth
increase.8 We want to provide a more complete picture of how heirs respond to inheritances. We therefore follow Joulfaian (2006) and also present estimates of the
marginal propensity to earn capital income.
The theoretical model suggests that if heirs increase their capital income
less than what the return on the inheritance would yield, then consumption has
increased. In reality, however, there is another possibility. Positive responses in
capital income may reflect that capital gains on inherited assets have been realized.
Realized capital gains can be used for consumption or be invested in new assets.
Both, however, are important behavioral responses that have not been sufficiently
studied. 9
We do not find any responses in capital income before the inheritance is
received. We do, however, find large increases in capital income in the years immediately after inheriting. Effects this large suggest that capital gains on inherited
assets have been realized.
Our results imply that the short-run increase in capital income tend to outweigh the decrease in labor income. The heirs make themselves better off in terms
of leisure as well as consumption possibilities.
Our empirical approach is limited in one respect. The estimates only capture
responses to inheritances that were not anticipated prior to 2000, which is the start
of our sample period. The individual fixed effects capture the responses taking
place before the start of our sample period. We can, however, correlate the estimated
individual fixed effects with the inherited amounts in a second step. This is to assess
the importance of anticipation effects occurring before our sample period. This way
we can separate the responses before the sample period from the responses during
the sample period but before the inheritance is received.
8 Cheng and French (2000) and Poterba (2000) review the literature on the marginal propensity
to earn and consume out of wealth. The wealth effects on labor supply and consumption have also
been studied using lottery winnings (Imbens, Rubin, and Sacerdote, 2001), stock market returns
(Coronado and Perozek, 2003, Juster, Lupton, Smith, and Stafford, 2006, Coile and Levine, 2006),
and housing capital gains (Engelhardt, 1996).
9 Engelhardt (1996) and Juster et al. (2006) discuss the difference between active and passive
saving.
Published by De Gruyter, 2012
3
108
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
The results suggest that the heirs might have reduced their labor incomes
already before the start of our sample period. We interpret this as indicating that
the inheritances were at least partly anticipated. There is, on the other hand, no
evidence of any responses in capital income during the years before the beginning
of the sample period.
2
Theoretical framework
Our starting point is a life-cycle model of consumption and labor supply (MaCurdy,
1981) to illustrate how inheritances affect heirs’ optimal labor supply, consumption,
and savings decisions (Weil, 1996, Joulfaian, 2006). Heirs derive utility from consumption and leisure, which are both assumed to be normal goods. We assume
that labor supply is endogenous and that there is a bequest motive.10 Our focus is
on how anticipation, labor income taxation, and liquidity constraints might affect
behavioral responses when inheriting.11
Suppose that an heir receives an inheritance at the time of the donor’s death.
The heir, however, knows the size and the time of the transfer already from young
age. Such anticipated inheritances will not affect either labor supply (and labor income) or consumption when inheriting. However, the responses to an anticipated
inheritance will affect the heir’s optimal consumption and labor supply paths already from the beginning of the life-cycle. Consumption will be smoothed over the
life-cycle. Similarly, the labor supply response will occur from the beginning of the
life-cycle. There will be no change in labor supply at the time of the inheritance.
The inheritance will cause a one-to-one change in wealth when received as there
will be no changes in consumption or hours worked.
On the other hand, if the inheritance is not anticipated, it will affect labor supply, consumption, and savings. The heir will act as if the inheritance is a
windfall gain and there will be discontinuities in the consumption and labor supply
paths: The inheritance will increase consumption in each period during the heir’s
remaining life-time. The labor supply path will follow the same logic with decreasing hours of work. The larger the unanticipated inheritance, the larger is the
corresponding response. Consequently, the inheritance will cause a smaller than
10 The
bequest motives might be consistent with joy-of-giving (egoistic) models (Blinder, 1976,
Abel and Warshawsky, 1988), altruistic models (Becker, 1974, Barro, 1974), or strategic models
(Cox, 1987, Bernheim, Shleifer, and Summers, 1985).
11 A formal version of the model is provided in the first version of this paper; see Elinder, Erixson,
and Ohlsson (2010).
4
109
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
a one-to-one wealth increase when inheriting. An inheritance that is partly anticipated and partly unanticipated can be viewed as two separate inheritances: one that
is perfectly anticipated and one that is unanticipated.
Suppose there is a proportional tax on labor income. How will changes in
labor income tax rates affect labor supply and savings responses to inheritances?
An increase in the tax rate further increases the heir’s optimal leisure time. In other
words, the negative labor supply response will be larger when the tax rate is higher.
People may not always be able to borrow against a future inheritance. One
should keep this in mind when using this framework to study responses to inheritances. There will be no consumption and labor supply responses before the inheritance is received if the heir faces binding liquidity constraints. This result holds
even if the inheritance is perfectly anticipated. It is, therefore, not possible to use
responses after the inheritances are received to infer if they are anticipated or unanticipated in this situation. The outcome will be similar if heirs who anticipate the
inheritance are risk averse or prudent (Kimball, 1990, Weil, 1996).
3
Institutional context
It is likely that the extent to which an inheritance is anticipated by the heir depends
on the institutional context.12 Succession rules in many European countries, more
or less, follow the Roman tradition with restricted testamentary freedom. Parents
are prohibited from completely disinheriting a child. Germany, France, and Sweden, are examples of this.13 Anglo-Saxon countries, e.g., the United States and the
United Kingdom, on the other hand, often grant full testamentary freedom. Therefore, it is likely that a larger fraction of received inheritances are anticipated in
countries with restricted testamentary freedom. Children are certain to inherit at
least a share of the estate if it is positive.14 This implies that heirs’ labor income
responses after inheriting may be smaller in Sweden than in the United States.
Sweden and many other European countries, however, have higher marginal
tax rates on labor income than the United States. Theoretically, the higher tax rate,
the larger the increase in leisure in response to inheritance.
12 There is a more extensive presentation of Swedish succession rules in the first version of this
paper, see Elinder et al. (2010).
13 Pestieau (2003) is an excellent review of differences in institutions governing inheritances and
transfer taxes across countries.
14 The elderly also face lower risks of large out-of-pocket health care expenses in the final stages
of life in countries with generous public insurance systems. This makes a parent’s bequeathable
wealth more predictable.
Published by De Gruyter, 2012
5
110
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
Taken together, it is therefore not clear if responses in labor and capital
income after receiving an inheritance can be expected to be larger in Sweden than in
the United States. Restricted testamentary freedom and high marginal tax rates also
give the heirs incentives to reduce their labor income long before the inheritance is
received.
This section presents the main rules governing the deceased and heirs in
Sweden. The default succession scheme in Swedish civil law implies that closer
relatives to the deceased inherit before more distant relatives. The deceased’s descendants, i.e., children, grandchildren, etc., are the first in line to inherit.
A surviving spouse does not inherit the deceased’s estate if the deceased
has children. These children will inherit. A surviving spouse, however, has the
right to dispose the estate freely for the remainder of her life if the deceased and
the surviving spouse have common children.15 Common children are referred to as
direct heirs with a postponed right to inherit. They have to wait until their second
parent dies to receive their inheritances.
More distant relatives will inherit the estate if there are no direct heirs. The
estate will go to a public fund, The Swedish Inheritance Fund, if there are no legal
heirs and if there is no surviving spouse.
The default succession scheme can be set aside by a will. This is a legally
binding document declaring the deceased’s last wish on how the estate should be
divided. The testator is, however, only allowed to bequeath up to half the estate.
The remaining part is divided among legal heirs according to the default succession
rules.16
The inheritance data we use are collected from estate inventory reports. The
estate inventory report provides information about the deceased’s complete balance
sheet at the time of her death.17 It served as a basis for the inheritance tax until the
tax repeal from 2005.18
The estate was reported by those in charge of the estate, usually a surviving
spouse or a child for smaller estates, or banks or law firms for large estates. The
reported values should be supported by documentation from banks, financial insti15 This has been the rule since the reform of the Marriage Act in 1988 (Brattström and Singer,
2007). Free disposal means that the surviving spouse can spend, but not bequeath, the money.
16 See Angelini (2009), Table 2, for corresponding rules in other European countries.
17 The estate inventory report should be prepared within three months after the time of death. It is
to be filed with the Swedish Tax Authority within a month after its completion.
18 The inheritance tax was to be repealed from January 1, 2005. Parliament, however, later
changed the repeal date to December 17, 2004. The reason was that many Swedes died in the Asian
Tsunami on December 26, 2004. The gift tax was repealed at the same time. Surviving spouses
were exempted from tax already from January 1, 2004. The inheritance tax was progressive in two
dimensions; first, large inheritances were taxed at higher rates than small. Second, direct heirs faced
lower tax rates than more distant heirs.
6
111
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
tutions, real estate agents, etc. Heirs had incentives to underreport the estate value,
to lower their tax payments, until the repeal of the inheritance tax. The deceased
may also have engaged in tax planning (or evasion) both during life, and shortly
before death, (see Bernheim, Lemke, and Scholz, 2004, Joulfaian, 2004, Nordblom
and Ohlsson, 2006, Kopczuk, 2007, Eliason and Ohlsson, 2010).
Most assets and debts were to be valued at market prices. For example,
financial assets were to be declared according to their market values as of the date
of death.
There are, however, reasons to believe that the reported inheritance values
until and including 2004 understated the market values of the transfer. There were
several exemptions from the principle of market prices. The most important exception concerned real estates. The tax value of this asset was supposed to be
75 percent of the market value. Any assets that were realized by the estate manager
before the actual estate division were valued at market prices. Heirs, therefore, had
tax incentives to postpone realization of capital gains on real estate until the estate
was divided.
It is an option, not an obligation, to inherit according to Swedish law. Heirs
can never be forced to pay the debts of estates in deficit. In many situations, inter
vivos gifts are regarded as inheritances received in advance. The law defines when
and how such transfers should be taken into account when dividing an estate. The
objective is to ensure that succession rules are not circumvented by inter vivos gifts.
4
Data and empirical strategy
It is difficult to obtain inheritance data of high quality. Inheritance data from surveys
are likely to be influenced by errors such as recall biases (Brown et al., 2010) and
underreporting (Kurz, 1984, Juster, Smith, and Stafford, 1999). Administrative tax
records have more precisely measured data. The disadvantage is that the value
of an inheritance (or an estate) needs to be above a certain threshold to be taxed
(Behrman and Rosenzweig, 2004). It is therefore not possible to draw conclusions
for individuals who are not affected by the taxation. The threshold for taxation
was very low in Sweden. Administrative data on inheritances in Sweden, therefore,
cover a much larger part of the population of heirs than in the United States.
We use a dataset of decedents and heirs that was originally collected from
the Swedish Tax Authority’s Inheritance Tax Register. The objective was to study
the incidence and the determinants of unequal sharing of bequests between heirs
(Ohlsson, 2007). To make data collection feasible, the sample was limited to dece-
Published by De Gruyter, 2012
7
112
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
dents registered in the City of Stockholm who passed away in 2004 and their heirs.19
The sample was also limited to deceased who had a will, more than one child, and
a positive estate, as these are necessary conditions for unequal sharing. In addition,
the sample only includes decedents who were not married at the time of their death.
This was done to avoid the uncertainties in estate division that might appear when
there is a surviving spouse.20
While this is not a perfectly representative sample of decedents and heirs,
we still believe that it is appropriate for this study. The decedents and the heirs are
relatively wealthy. It is therefore less likely that the heirs are liquidity constrained.
Furthermore, most heirs are in their prime age making it likely that they have good
access to credit markets. This is an advantage. It should be noted, however, that
being relatively wealthy in Sweden is quite different from being very wealthy in the
United States. The average estate in our sample is about twice as large as the average net worth in the Swedish population (Berg, 2006). Kopczuk and Saez (2004)
reports that the average net worth of the two percent of the decedents who file estate
tax returns in the United States is more than 13 times higher than average wealth.
The complete sample contains 232 decedents and 820 heirs. The estates
were divided up in inheritances transferred to 573 children, 176 grandchildren,
8 partners, 45 relatives, and 18 other individuals and charities. Few lots go outside the family. This suggests that testators tend to follow the principles of the
default succession rules. The dataset has information on the net worth of the estate,
the value of each inheritance received by the heirs, and data on possible taxable
inter vivos gifts made by the decedent to each heir during the last ten years.
We have added data on annual labor income, capital income, self-employment income, taxable wealth, and real estate wealth for decedents and for heirs.
These data come from the Tax Authority’s Register of Final Tax on Income. This
concerns the nine years 2000–2008 for the heirs. Taxable wealth is measured on a
household basis, whereas the other variables are measured at the individual level.
Demographic characteristics, such as sex, marital status, year of birth, number of
children, place of residence etc., have been collected from the Tax Authority’s Total
Register of the Population.
We study responses to inheritances in a sub-sample of heirs who were between 21 and 59 years old in 2004. There are three reasons for this. First, we want
to separate labor income responses from normal retirement and education decisions.
Second, we do not want to include retired heirs. People tend to spend down their
wealth after retirement and this might result in negative saving. Third, we do not
19 The
data were collected manually from estate reports at the Uppsala Tax Office.
can expect that individuals who no longer have any parents alive do not anticipate any large
inheritances in the future. This is an advantage for this study.
20 We
8
113
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
want to include heirs who are minors during the sample period.
We only include direct heirs (children) in our sample. The responses of
other heirs could be quite different. It becomes more clear to which population our
results generalize when the sample is limited to direct heirs.21 We have complete
observations for every year for about 96 percent of the heirs. We lack income data
for at least one year for the remaining 4 percent. This leaves us with an unbalanced
panel of 374 direct heirs and 3,310 observations during 9 years.
The reported inherited amount is our main explanatory variable. It is calculated as the amount inherited by the heir excluding inter vivos gifts.22 The amount
is also net after taxes.
Ideally we would have preferred to have data on hours worked, effort, and
wage rates. This is not possible, so we use taxable labor income instead. Taxable
labor income includes salaries, social insurance benefits (sickness, parental, and
unemployment), and pension payments. This, in some sense, captures aspects of
hours worked, effort, and wages. Changes in taxable labor income are likely to
reflect conscious decisions. It is desirable from a tax revenue perspective to use
taxable labor income, since the effects on tax revenues come via taxable income.
Capital income includes interest received on financial assets, dividends, and
realized capital gains minus interest paid on loans and realized capital losses. Capital income is the result of past and present savings and investment decisions. It is
taxed independently of how long a particular asset has been held. Capital gains are
only taxed when realized. The capital income tax rate has been 30 percent throughout the sample period.
One can think of using capital income to calculate an approximate measure
of wealth.23 The analysis could then be done using wealth rather than capital income. We would, however, need to assume rates of return on assets. This is difficult
since we do not have information about the heirs’ portfolios composition. Returns
on different assets have also varied substantially during the sample period.
21 The
original data contained one decedent with a very considerable estate. The heirs of this
estate have been excluded from the analysis.
22 Gifts are excluded because only gifts that have been declared for taxes are reported in our data.
Furthermore, we do not know when they were received.
23 The correlation between taxable wealth in period t −1 (for those with taxable wealth) and capital
income in period t is 0.21. This is statistically significant at all conventional levels.
Published by De Gruyter, 2012
9
114
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
4.1
The heirs
Table 1 provides descriptive statistics for the heirs.24 Inheritances vary between
zero and SEK 2,6 million with an average value of SEK 300,000.25 The P90:P10
ratio shows that those in the top of the distribution inherit around 12 times more
than those at the bottom. The heirs have an annual labor income of on average
SEK 310,000 in 2003. Inheritances are, on average, almost as large as average
annual labor income. This suggests that inheritances may considerably influence
economic behavior.
The heirs in our sample are on average 50 years old when inheriting.26
Women are in slight majority and about half of the heirs are married. Heirs have, on
average, 1.45 siblings and 1.75 children. Almost all the direct heirs in the sample,
96 percent, are children of the decedents.27 Furthermore, 75 percent of the heirs
live in Stockholm County.
Our sample of heirs is not completely representative for the entire population of Swedes. These heirs have more siblings than Swedes on average.28 This is
a consequence of the sampling criteria. They also earn more. The corresponding
mean in labor income for Swedes aged from 21 to 64 was 219,800 in 2003.
There is one clear advantage of having a sample with wealthier decedents
and higher income heirs than Swedes on average. It is less likely that our heirs
face binding liquidity constraints. Otherwise, such constraints might confound the
interpretation of our empirical results.
4.2
Empirical challenges and estimation framework
We guide our estimation strategy by visually inspecting how labor income and capital income evolve over the sample period. We observe labor and capital income for
24 Descriptive
statistics for the decedents are presented in Appendix A.
are reported in the 2004 price level. All amounts reported in the paper are, therefore,
deflated to the 2004 price level. Inflation was low during the studied period. The CPI increase from
2000 to 2008 was 14.6 percent. The year by year CPI increases were: 2.4 percent (2001); 2.2 percent
(2002); 1.9 percent (2003); 0.4 percent (2004); 0.5 percent (2005); 1.4 percent (2006); 2.2 percent
(2007); 3.5 percent (2008).
26 This is considerably older than in previous studies. The heirs in Joulfaian and Wilhelm (1994)
are 42 years old on average. Holtz-Eakin et al. (1993) and Joulfaian (2006) report average ages of
39.
27 The remaining 4 percent are grandchildren. We have only included grandchildren when they
are direct heirs. Grandchildren become direct heirs when a child of a decedent is already deceased.
Omitting grandchildren does not affect the empirical results.
28 Blomquist (1979) report that the inherited amount is inversely related to the number of children
of the deceased.
25 Estates
10
115
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
Table 1: Descriptive statistics for the heirs.
Mean
S.d.
Inheritance, 2004
299.3
313.4
Age, 2004
49.97
7.88
50 years and older in 2004, percent
55
Male, percent
47
Married, percent
52
Number of siblings
1.45
0.69
Number of children
1.75
1.10
Direct heir child of the deceased, percent
96
Living in Stockholm County, percent
75
Labor income, 2003
311.3
231.5
Capital income, 2003
-1.3
64.5
Self-employment income, 2003
7.7
61.8
Share with taxable wealth in 2003, percent
20
Taxable wealth, 2003a
2,468
1,974
Share with taxable real estate in 2003, percent
68
Taxable real estate, 2003b
972
1,341
Share with taxable gift(s), percent
8
Value of gift(s), 2004c
241.9
274.3
Number of observations
374
Notes. Amounts are measured in SEK thousands, price level 2004.
a the value is conditional on having taxable wealth,
b the value is conditional on having taxable real estate,
c the value is conditional on having received a gift.
Published by De Gruyter, 2012
Median
192.3
50
1
2
269.7
-4.3
0
2,114
701
155
11
116
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
each heir in the sample during four years before the decedent passed away (2000–
2003), the year when the decedent died (2004), and the following four years (2005–
2008).
We classify heirs into two groups: The high inheritance group consists of
heirs with inheritance higher than the sample mean (SEK 299,000). Those with
inheritances below the sample mean are classified as the low-inheritance group. We
then compute sub-sample means for inheritances in the two groups.29 The question
is: Do the two groups differ in labor income and capital income responses?
The two lines in Figure 1 show how labor income evolves for the heirs in
the two inheritance groups. We first observe that those who inherit less have higher
labor income on average than those who inherit more. This result perhaps surprises
some. It might be expected that inheritance amounts would be positively correlated
with the heir’s earnings potential. However, those who inherit more might also
have anticipated a larger inheritance. They therefore have had stronger incentives
to reduce their labor supply long before the actual inheritance is received.
Figure 1 also shows that the two groups have similar labor income trajectories in the pre-inheritance period, although the levels differ. Labor income in the
high inheritance group declines gradually for all years after the inheritance is received. On the other hand, labor income in the low-inheritance group increased
dramatically up to 2007. This pattern is consistent with a situation in which those
in the high inheritance group inherit more than anticipated. It is also consistent with
the low inheritance group inheriting less than what they anticipated. We emphasize,
however, that Figure 1 shows the unconditional means. The differences between the
groups do not necessarily reflect causal effects of inheritances.
Figure 2 shows how capital income evolves during the sample period. The
capital income trajectories of the two groups are similar during the pre-inheritance
years. Capital income increases from zero to almost SEK 48,000 in 2004, when the
inheritances are received. The inherited amount in itself may explain the dramatic
surge. Some heirs may decide to realize inherited capital gains, for instance, to
re-optimize their asset composition.
The financial upturn starting in 2004 may also explain the increase in capital
incomes. The Riksbank’s decrease of the repo rate from 2.75 in 2003 to 2 percent
in 2004 could also contribute to higher capital incomes for the heirs who had debt.
Capital income in the low-inheritance group decreases in 2005. On the other hand,
the high-inheritance group’s capital income continues to rise. Capital income increases for both groups in the subsequent years followed by a dramatic fall in 2008.
The difference between the two groups almost disappears as a consequence.
29 We
exclude heirs with incomes outside 1.96 standard deviations from the sample mean.
12
117
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
Mean Value of Labor Income SEK
240,000 260,000 280,000 300,000 320,000 340,000
Figure 1: Annual labor income 2000–2008.
Note. The solid vertical line indicates the year when inheritance is received.
2000
2001
2002
2003
2004
2005
2006
2007
2008
Year
High Inheritance
Mean Value of Net Capital Income SEK
0
50,000
100,000
Low Inheritance
2000
2001
2002
2003
2004
Year
Low Inheritance
2005
2006
2007
2008
High Inheritance
Figure 2: Annual capital income 2000–2008.
Note. The solid vertical line indicates the year when inheritance is received.
Published by De Gruyter, 2012
13
118
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
There are at least two plausible explanations for why capital incomes converge. First, the stock market share of wealth is increasing in wealth (Guiso,
Haliassos, and Jappelli, 2002). The financial crisis in 2008 might therefore have
hit wealthier heirs harder. Second, those who inherited more may also have had a
larger share of their wealth invested in assets that yield little or no cash flow, e.g.,
residential real estate.
It is clear that the two groups follow the same trends in labor and capital
income in the years before the inheritance was received. This is, however, not
a formal test. The observed differences during the post-inheritance period may
depend on differences in inherited amounts. But the observed differences may also
depend on differences in the heirs’ characteristics, characteristics that are correlated
with the inherited amount. We need a more thorough econometric analysis to test
this. We will, therefore, now turn to the empirical model.
4.3
Empirical model
It is reasonable to believe that heirs differ in unobservable characteristics, such
as taste for leisure, risk aversion, early family upbringing, ability, etc. This may
affect their behavior and, therefore, also their labor and capital income. Cross
section analysis is likely to yield upward-biased estimates of the labor and capital income responses if, for example, inheritances are positively correlated with
unobserved income potential. We deal with this type of omitted variable bias by
estimating models with individual fixed effects. This approach is similar in spirit to
a difference-in-differences approach. Individuals inheriting different amounts serve
as counterfactuals to each other in our approach.
Suppose that one heir inherits more than another heir. It is crucial for our
approach that these two heirs would have had the same labor or capital income had
they inherited the same amount. This is necessary for obtaining unbiased estimates,
at least conditional on a fixed effect. We cannot test this assumption. Figure 1 and
Figure 2, however, indicate that this assumption is reasonable. The labor income
trajectories of the high- and low-inheritance groups follow the same trends before
the year when the inheritances are received. The same is also true for the capital
income trajectories.
We start by running different versions of the following regression to explore
the effects of inherited amounts on our dependent variables:
yit =
2008
∑
t=2000
δt Ii2004 + β Xit + θt + ai + uit ,
(1)
14
119
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
where yit is individual i’s labor or capital income in year t. The sample period is t =
2000 through 2008. These variables are both measured in SEK. The main variable
of interest is the inherited amount, Ii2004 . It is also measured in SEK. We interact
the inherited amount with year indicators to estimate the annual responses of our
dependent variables. The vector Xit includes controls for a third order polynomial
in age,30 and in some specifications, also individual specific trends. We want to
avoid having differences in age-related factors (e.g., human capital, labor market
experience, and job tenure) bias our estimates.
We include a full set of year dummies, θt , to capture time effects such as
macroeconomic changes etc. Moreover, ai captures an individual fixed effect. This
includes factors that are assumed to be constant over time and correlated with the
inherited amount and the dependent variables. We also report results from specifications where we control for individual-specific linear time trends.
The idiosyncratic error uit is assumed to be uncorrelated with Ii2004 , Xit , and
ai . We cluster the standard errors on the family level to allow for within-family correlation of the error term. The F and t statistics are valid under these assumptions.
We can consistently estimate δt as the annual marginal causal effect of an additional
SEK inherited.
Our theoretical model suggests that the individual fixed effects capture responses to the anticipations of inherited amounts formed before the studied period.
This implies that δt , t = 2000, . . . , 2003, capture responses to updated anticipations
of inherited amounts.
On the other hand, δt , t = 2004, . . . , 2008, capture responses to inherited
amounts that were not anticipated at the time when the inheritance was received.
The annual effects δt , in other words, underestimate the total causal effects if inheriting was anticipated before the studied period. Therefore, we also test for early
anticipation effects by correlating the estimated individual fixed effects from the
empirical model (1) with the inherited amounts.
5 Results
Subsection 5.1 presents the main regression results for labor income and several robustness tests. Corresponding results for capital income are in Subsection 5.2. The
estimated coefficients of the inheritance-year interaction variables are the parameters of prime interest. We choose 2003, the year before the deceased passed away,
as the reference year. We present the results from the tests for pre-sample period
anticipation effects in Subsection 5.3.
30 The
first order polynomial term is omitted since it is colinear with time.
Published by De Gruyter, 2012
15
120
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
5.1
Labor income
Table 2, column 1, reports small and statistically insignificant effects of the inherited amount on labor income in the years before the inheritance was received. This
suggests that anticipation effects are either small or that the heir has adjusted her
labor income already before the start of our sample period. We also do not find
any statistically significant effects the year when the inheritance is received. We
cannot, however, observe the exact dates when the estates are divided and the inheritances are transferred to the heirs. A possible explanation for the insignificant
effect is that many inheritances were received late in 2004 (or even early in 2005).
Therefore, these inheritances may have had relatively small effects on the labor income in 2004. The estimated coefficients for the subsequent years 2005–2008 are
economically and statistically significant. Heirs use some of their newly obtained
wealth to increase their consumption of leisure as expected.
The estimated effect of the interaction term for 2005 is that labor income
decreases by SEK 0.048 for each additional SEK received. The estimated responses
are almost twice as large the following two years. The results suggest that it takes
time for the responses to materialize in labor income.
Assuming that the response in 2008 reflects a new optimum, we can approximate the effect on the life-time income of the heir. On average, the heirs are
50 years old and the inheritances are SEK 299,000. Therefore, the coefficient estimate for 2008 implies that average annual after tax labor income decreases by
approximately SEK 14,400.31
Suppose that we also assume that heirs retire at age 65 and that there is no
time discounting. A back-of-the-envelope calculation then suggests that the impact
on life-time labor income corresponds to about 72 percent of the inherited amount.
The impact of the inherited amounts on labor income is large. We also see that the
coefficients of the age variables are imprecisely estimated. A Wald-test, however,
tells us that the age variables are jointly significant at the ten percent level.
There is a potential drawback with our empirical model (1). It does not
allow for heirs to have different trends in labor income growth. This was also mentioned when discussing Figure 1. Although average wage increases in Sweden were
moderate during our sample period, income growth may still differ substantially between socioeconomic groups.
Therefore, we also include individual specific linear trends in our empirical
model. Column 2 in Table 2 presents the results. Although the coefficient estimates are somewhat smaller in absolute magnitude than the baseline estimates in
31 We assume that the average heir pays 30 percent in labor income tax. This is a plausible assumption as most heirs in our sample were in this labor income tax bracket in 2004.
16
121
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
Table 2: Labor income.
1
Inherited amount*Year
2000
2001
2002
2003, reference
2004
2005
2006
2007
2008
Age2 /100
2
3
young heirs,
21−49
4
old heirs,
50−59
-0.028
(0.029)
-0.038
(0.029)
-0.012
(0.019)
-0.038
(0.027)
-0.046*
(0.027)
-0.016
(0.018)
0.016
(0.024)
0.008
(0.019)
-0.001
(0.019)
-0.064
(0.043)
-0.074*
(0.043)
-0.025
(0.023)
-0.026
(0.018)
-0.048**
(0.021)
-0.092***
(0.029)
-0.085***
(0.024)
-0.069***
(0.021)
-0.022
(0.018)
-0.040*
(0.022)
-0.080***
(0.030)
-0.070**
(0.027)
-0.049*
(0.028)
-0.002
(0.019)
-0.017
(0.027)
-0.100
(0.066)
-0.038
(0.033)
-0.033
(0.034)
-0.042
(0.028)
-0.069***
(0.022)
-0.091***
(0.023)
-0.114***
(0.031)
-0.091***
(0.025)
-13.9
(79.1)
-2.90
(56.68)
692
(1,240)
Yes
-11.2
(77.9)
-4.91
(55.82)
3,658
(3,554)
Yes
-13.3
(142.9)
-8.26
(120.05)
571
(1,593)
Yes
Year dummies
Yes
Yes
Yes
Yes
Individual specific
time trends
Number of heirs
No
Yes
No
No
374
374
150
224
Age3 /10,000
Constant
Individual fixed effects
Number of obs
3,310
3,310
1,326
Notes. Amounts are measured in SEK thousands, price level 2004.
Standard errors are clustered on family.
* significant at the 10 percent level, ** significant at the 5 percent level,
*** significant at the 1 percent level
Published by De Gruyter, 2012
-1,715
(1,456)
1,006
(881)
34,662
(28,698)
Yes
1,984
17
122
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
column 1, this exercise largely confirms our previous findings.
Our theoretical approach suggests that the magnitude of responses to inheritances depends negatively on the length of the heirs’ remaining lifetime. Fewer
remaining years to live lead to fewer years during which the inheritance can be consumed. We would, therefore, expect labor income responses to be larger for old
heirs than for young. We have estimated separate models for young and old heirs
to test for this. The threshold age is 50.
The results, presented in Table 2, suggest that old heirs (column 4) reduce
their labor income more than young heirs (column 3). This result is in accordance
with our hypothesis. Imbens et al. (2001) report similar results for lottery prize
winners.
In Appendix B, Table 6, we present results from a set of specifications similar to those presented in Table 2. The difference, however, is that we let the entire
pre-inheritance period (2000–2003) be the reference period. Regarding the results,
we note first of all that the coefficient estimates become smaller in size. Moreover, we see that the responses in 2005 become statistically insignificant as a result
of this change in reference period. However, the main finding, that inheritances
reduce labor income, remains.
A related literature studies if receiving an inheritance enables the heir to
become an entrepreneur by relaxing liquidity constraints. The results are clear-cut:
Inheriting increases the probability of starting a business (Holtz-Eakin, Joulfaian,
and Rosen, 1994a,b, Lindh and Ohlsson, 1996, Blanchflower and Oswald, 1998,
Hurst and Lusardi, 2004). It also improves the performance of existing businesses.
Our labor income measure does not include income from small businesses
and sole proprietorships. This is a limitation. We have information about the heirs’
self-employment income; however, only about 12 percent of the heirs in our sample
report this type of income. Therefore, it would not be fruitful to perform a separate
analysis of self-employment income.
Instead, we have calculated an extended labor income variable by adding
self-employment income to labor income. We have estimated the empirical model
with extended labor income as dependent variable; see the Appendix B, Table 7,
column 1. The results are both quantitatively and qualitatively similar to those for
labor income.
The labor income distribution is positively skewed. We have accounted for
this by estimating the empirical model with the logarithm of labor income as dependent variable. The results are similar to the main results in Table 2. Table 7,
column 2, in Appendix B reports the results.
The results are also robust to specifications where we limit the sample to
heirs with labor income within 1.96 standard deviations from the sample mean. We
report the details in Table 7, column 3, in Appendix B.
18
123
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
We pointed out in Section 3 that real estate wealth is declared according to
its tax value in the estate inventory report. This implies that the reported values of
inheritances of real estate in our sample are below the market value. Suppose that
we adjust the inheritances to reflect market values rather than the reported values.
How will our previous results change? We will now come closer to the true effect. The previously understated estate value, leading to upward biased coefficient
estimates, is corrected.
We divide the tax value of real estate with the factor 0.75 to get the market value. The difference between the market value and the reported value is then
distributed equally between the legal heirs of each donor. We calculate the average inherited amount at market value to SEK 330,000. Estimating the empirical
model (1) using the inherited amount at market value, however, yields results that
are akin to those presented in Table 2. Table 7, column 4, in Appendix B presents
the detailed results.
We have also estimated the empirical models without any age restrictions;
see Table 7, column 5, in Appendix B. It is reassuring that these estimations give
virtually the same results as the baseline specification. (The statistically insignificant coefficient estimate for 2005 is the exception.)
The responses to inheritances may, as discussed earlier, be affected by liquidity constraints. We estimate separate models for heirs with positive and negative
capital income in 2000 to study the potential impact of liquidity constraints. The
idea is that heirs with positive capital income are less likely to be liquidity constrained. We use capital income as early as in 2000 to reduce the risk of capital income being endogenous with respect to the inheritance. Larger responses for heirs
with negative capital income than for heirs with positive capital income suggest that
liquidity constraints are important. However, we find slightly smaller responses for
heirs with negative capital income. We conclude that our results are not likely to be
driven by liquidity constraints. The results are presented in Appendix B, Table 7,
columns 6 and 7.
We conclude that the inherited amount affects the heir’s labor income negatively during all the four years following the transfer. In other words, we find that
the negative impact of the inherited amount lasts a long time. This is contrary to the
findings of Joulfaian and Wilhelm (1994) but similar to those with respect to lottery
winnings in Imbens et al. (2001). Our findings suggest that the inherited amount
is at least partly not anticipated. The succession rules guarantee each direct heir a
share of the estate. Inheriting per se is, therefore, expected. However, our results are
not consistent with the inherited amounts being perfectly anticipated. The relatively
large responses are consistent with the theoretical prediction that high marginal tax
rates create disincentive effects.
Published by De Gruyter, 2012
19
124
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
5.2
Capital income
We proceed by estimating responses in capital income. Column 1 in Table 3 shows
that there are no statistically significant effects during the pre-inheritance years or
the year when the inheritance was received. It is possible that the insignificant
response in 2004 has to do with insufficient adjustment time for labor income. We
find that the response in 2005 is positive and statistically significant at the 1 percent
level. This is consistent with our predictions.
The coefficient estimate suggests that capital income in 2005 increases by
27.5 percent of the inherited amount. We consider this response non-trivial. A large
immediate response is, however, reasonable if the heirs decide to realize capital
gains to increase consumption or to re-optimize the portfolio. The undervaluation
of the inherited amount may also partly explain the sizeable effect.
Let us turn to the years 2006–2008. The estimated coefficients are positive and statistically significant on at least the 10 percent level. Capital income
responses first decline gradually and then drop sharply in 2008, the last year of the
sample period. However, without detailed information on the heirs’ asset holdings,
it is difficult to say how important the financial crisis is for this drop. We also note
that the coefficients on the age polynomials are both individually and jointly insignificant. The stock market return was unusually high during the post-inheritance
years, except for 2008. We may, therefore, find estimates that are higher than otherwise if heirs’ held stock.
Our estimates, nevertheless, suggest that the heirs’ capital income increases
substantially for up to three years after the transfers. Suppose that capital income
in 2007 reflects a new steady state. The average inherited amount will lead to a
SEK 38,000 increase in annual capital income net of tax.
We also extend our empirical specification to include individual specific
time-trends. The second column in Table 3 presents the results. Allowing for individual specific trends gives results that differ somewhat from to those in column 1.
The estimated responses reported are marginally lower. Also, the coefficient estimates for 2007 and 2008 become statistically insignificant. These results, nevertheless, reinforce the conclusion that the responses in capital income are temporary.
We then estimate separate models of capital income response for young and
old (see columns 3 and 4). The prediction follows the same logic as for labor
income. Old heirs, compared to young heirs, will consume inheritances at a higher
rate, and as a consequence have a higher short term capital income, as capital gains
are realized earlier and to a larger extent. We observe that some of the responses
are imprecisely estimated. This is probably because the sample sizes are small.
Nevertheless, the results suggest that increases in capital income are larger for old
heirs than for young.
20
125
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
1
Inherited amount*Year
2000
2001
2002
2003, reference
2004
2005
2006
2007
2008
Age2 /100
Table 3: Capital income.
2
3
young heirs,
21−49
4
old heirs,
50−59
0.195
(0.154)
-0.102
(0.111)
-0.019
(0.024)
0.200
(0.164)
-0.098
(0.109)
-0.017
(0.026)
0.271
(0.304)
0.010
(0.036)
-0.060
(0.051)
0.141
(0.172)
-0.168
(0.177)
0.011
(0.019)
-0.013
(0.029)
0.275***
(0.095)
0.213***
(0.065)
0.182*
(0.098)
0.048*
(0.026)
-0.015
(0.029)
0.270***
(0.089)
0.207***
(0.060)
0.173
(0.106)
0.037
(0.042)
-0.078
(0.053)
0.187***
(0.064)
0.151
(0.092)
0.152**
(0.073)
0.051
(0.060)
0.034
(0.042)
0.343***
(0.128)
0.256***
(0.080)
0.218
(0.156)
0.054***
(0.019)
7.59
(76.29)
-10.29
(66.55)
-58
(1,046)
Yes
6.04
(74.93)
-9.18
(65.39)
-1,764
(6,203)
Yes
-57.77
(96.88)
59.98
(85.03)
553
(1,048)
Yes
Year dummies
Yes
Yes
Yes
Yes
Individual specific
time trends
Number of heirs
No
Yes
No
No
374
374
150
224
Age3 /10,000
Constant
Individual fixed effects
Number of obs
3,310
3,310
1,326
Notes. Amounts are measured in SEK thousands, price level 2004.
Standard errors are clustered on family.
* significant at the 10 percent level, ** significant at the 5 percent level,
*** significant at the 1 percent level
Published by De Gruyter, 2012
-4,600**
(2,102)
2,770**
(1,268)
90,904**
(41,493)
Yes
1,984
21
126
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
We have also estimated the models in Table 3 with the pre-inheritance years
(2000–2003) as reference period. Table 8 in Appendix B reports the results. These
results are similar to those in Table 3. The estimates are slightly lower, however,
and not statistically significant for 2007 and 2008.
The overall finding from the previous exercises is that there are large effects
during the first two years after inheriting, then the effects decline. It is, however,
difficult to separate temporary effects from effects from the financial crisis as the
2007 coefficient still is quite large.
It is possible that the rate of return on the inherited amount affects the estimates in Table 3. We follow Joulfaian (2006) and adjust capital income to reflect
the change in wealth, less the inherited amount, to account for this. We first assume
a uniform 4.67 percent rate of return on inherited assets.32 Our dependent variable
becomes capital income excluding the return in the inherited amount. We then rerun
the empirical model (1). Table 9, column 1, in Appendix B reports the results from
this regression. These results are almost identical to those presented in Table 3.
The distribution of capital income is less skewed than that of labor income.
This is clear from the descriptive statistics in Table 1. We still want to test if our
results are robust accounting for the influence of outliers. The estimation results
using the logarithm of capital income as dependent variable are both economically
and significantly akin to those in Table 3. Table 9, column 2, in Appendix B presents
the results.
We obtain a similar result if we omit heirs with capital income deviating
more than 1.96 standard deviations of the sample mean; see Table 9, column 3, in
Appendix B. Nevertheless, it is reassuring that results except for 2008 reasonable
variations in specifications and sample definitions do not alter our main.
We have also varied the sample’s age restrictions. The main difference is
that the response in 2008 is positive but not statistically significant. Table 9, column 5, in Appendix B report the detailed results.
Following the same reasoning as we did in the analysis with respect to labor income, we have also tested for confounding effects of liquidity constraints.
Here too, we use information on capital income in 2000 to distinguish between potentially liquidity constrained heirs and heirs not being liquidity constrained. See
Table 9, columns 6 and 7 in Appendix B. We find no statistically significant responses in the pre-inheritance years, or in the year of the receipt, for either of the
two groups. Likewise, the coefficient estimates for the first two years following the
receipt are similar with respect to statistical significance, but smaller in magnitude
for heirs with negative capital income in 2000. The heirs do not seem to be liquidity
32 This is the average of the official long-term central government borrowing rate during the studied period.
22
127
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
constrained. We conclude that the large short-run responses in capital income are
most likely due to the realization of capital gains.
5.3
Anticipations formed before the sample period
The results presented so far do not indicate any behavioral responses during the four
years preceding the transfer. Still, it is reasonable to believe that some heirs during
a long time may have had a fairly good idea of how much they will inherit and
roughly when. They may, therefore, have adjusted their labor supply and savings
behavior before the start of our sample period. We would potentially risk overlook
important behavioral responses if this is the case.
The empirical model (1)’s fixed effects capture the impact of unobserved
factors on the income level. The unobserved factors that are constant over time
include the inheritance anticipations formed before the sample period.
Therefore, we can predict the pre-sample period anticipations by correlating
the estimated fixed effects and the inherited amount. This procedure would, however, lead to biased estimates if the inherited amount correlates positively with the
(unobserved) earnings potential. It is difficult to completely deal with this omitted
variable bias by controlling for the heirs’ observable characteristics. Our approach,
however, yields conservative estimates of pre-sample period anticipation effects.
The reason is that the expected bias goes in the opposite direction to the expected
income responses.
We regress the estimated individual fixed effects from the regressions presented in Table 2 and Table 3 on the inherited amount. Column 1 in Table 4 reports
a negative and statistically significant relationship between the inheritance amounts
and the labor income fixed effects. It is a good idea to be cautious in interpreting
the magnitude of this relationship. Nevertheless, it indicates that the more an heir
has inherited, the lower is the level of the annual gross labor income already from
the start of the sample period.
The second column provides the estimates of anticipations formed before
the sample period for capital income. The correlation between the estimated fixed
effects from the capital income model and the inherited amount is close to zero. It
is statistically insignificant at all conventional levels. The level effects in capital
income are, in other words, unrelated to the inherited amount. This contrasts with
the findings for labor income. It is difficult to tell if there is no effect for capital
income or that the estimates suffer from upward bias.
We obtain reasonable values for the estimated coefficients of the control
variables in both regressions. This makes us more confident in the results. However,
it remains to find more definite answers.
Published by De Gruyter, 2012
23
128
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
Table 4: Anticipations formed before the sample period.
1
2
Dependent variable: Estimated individual fixed effect, SEK thousands
Labor income Capital income
-0.0735**
-0.0001
(0.0353)
(0.0400)
Age
105.5
9.5
(66.4)
(29.2)
Age2 /100
-229.6
-46.0
(164.3)
(72.4)
188.4
53.4
Age3 /10,000
(130.4)
(57.5)
Female
-71.2***
4.7
(23.7)
(20.6)
Married
8.5
4.4
(25.2)
(17.2)
Children
25.4
1.1
(26.3)
(12.4)
Stockholm County
97.5***
1.8
(25.7)
(21.8)
Taxable real estate
108.2***
39.6***
(23.3)
(12.1)
Constant
-2,023**
-227
(849)
(381)
Number of observations
374
374
Notes. Standard errors, in parentheses, are clustered on family.
Independent variables are measured in 2004, except
Taxable real estate, which is measured in 2003.
Female, Married, Children, Stockholm County, and Taxable real estate
are binary variables taking the value one if category indicated
by name is satisfied, and zero otherwise.
** significant at the 5 percent level,
*** significant at the 1 percent level
Inherited amount, SEK thousands
24
129
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
6
Concluding remarks
This paper presents new evidence on the impact of inheritances on heirs’ labor
income and capital income. We use data from administrative records for a sample
of Swedish decedents and their heirs. Our focus is on how the heirs’ marginal
propensities to earn labor and capital income out of wealth evolve during the years
before and after receiving an inheritance.
Labor and capital income responses will, according to theory, critically depend on whether the inheritance is anticipated or not. The behavioral responses to
inheritances are likely to take place already before receiving the inheritance if it
is anticipated. On the other hand, inheritances that the heir did not anticipate will
generate responses after the inheritance is received.
In contrast to previous studies, we find that inheritances have persistent effects on labor income. The effect is negative and considerable in each of the four
years following the transfer. It is stronger for old heirs than for young heirs. The
corresponding effects on life-time labor income, calculated using some simplifying
assumptions, are large relative to the inherited amount. We also show results supporting that heirs have reduced their labor income already before the start of our
sample period.
There are several possible explanations for why we find larger and longer
lasting responses in labor income than those reported by, e.g., Joulfaian and Wilhelm (1994). First, labor income taxes in Sweden are higher than those levied in
the United States. This suggests that the opportunity cost of leisure is higher in the
United States than in Sweden. Second, the responses may be larger because the
heirs in our sample are significantly older than those in the studies from the United
States. Both the theoretical model and previous empirical findings (Imbens et al.,
2001) suggest that the magnitude of responses to wealth shocks should increase in
age.
Our results show that even relatively small inherited amounts affect economic behavior. This contrasts with the belief that wealth shocks need to be significant to overcome frictions that may intrude on labor supply decisions (see, e.g.,
Pencavel, 1986, Card, 1994, Blundell and MaCurdy, 1999).
We also find large positive responses in capital income during the three years
following the transfer. There is a sharp decline in the response, however, in 2008.
It is difficult to say to what extent this is a consequence of the financial crisis that
started in 2008.
The temporary increase in capital income is sufficiently large to outweigh
the corresponding loss in labor income. The heirs make themselves better off both
in terms of leisure and consumption possibilities. We conjecture that the large capital income responses partly arise because previously unrealized capital gains were
130
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
realized. The high returns on the stock market during this period may also have contributed. It is, however, necessary to have more detailed data on the assets inherited
and the associated rates of return to draw more refined conclusions.
The results in our paper contribute to the literature in several ways. First, the
results provide detailed information about the dynamic effects of inheritances. This
can be useful for policy makers who want to account for behavioral responses when
designing optimal estate or inheritance tax schedules. One important implication is
that inheritance taxes are likely to also increase revenue from labor income taxes.
On the other hand, revenue from capital taxes might decrease. Second, it is not
sufficient to look only at labor income responses when studying the welfare effects
of inheritances. Third, it is important to study the effects of inheritances in different
institutional contexts to better understand their impact on economic behavior.
26
131
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
Appendix
Appendix A
Table 5 presents summary statistics for the decedents. The average age at the time
of death in the sample is 85 years. This is 7 years older than the average age at
death in 2004. This suggests that the decedents were healthier than the overall
Swedish population. There are fewer men than women among the decedents. This
is expected since women live longer than men. We have sampled households where
the deceased was a widow(er), divorced, or unmarried.
The number of children of the deceased varies between 2 and 5. We also
note that about 20 percent of the decedents in the sample paid wealth taxes in
2003. The corresponding share for the total population of Swedes age 60 years and
older was 6.3 percent in 2004. Furthermore, we see that the average estate in 2004
amounts to SEK 960,000 with a median value of SEK 600,000. The average value
of labor income in 2003 was rather low. This reflects that the majority of deceased
were retired in the last year of their lives. Moreover, we have information on the
share in the sample with self-employment income. Self-employment income can
be a good proxy for whether the estate included a small business. Table 5 reveals,
however, that the share of decedents with self-employment income is negligible.
It is clear that the sampling strategy has resulted in a sample of deceased
who were both healthier and wealthier than the deceased Swedes in 2004 in general.
Table 5: Descriptive statistics for the decedents.
Mean
S.d.
Age, 2004
85.3
8.8
Male, percent
31
Widow(er), percent
81
Number of children
2.45
0.69
Labor income, 2003
216.8
139.5
Capital income, 2003
27.7
100.6
Share with taxable wealth in 2003, percent
22
Taxable wealth, 2003a
2,608
1,300
Share with taxable real estate in 2003, percent
26
Taxable real estate, 2003b
858.3
763.7
Share with self-employment income in 2003, percent
0.5
Estate, 2004
959.0
978.0
Number of decedents
194
Notes. Amounts are measured in SEK thousands, price level 2004.
a the value is conditional on having taxable wealth
b the value is conditional on having taxable real estate
Published by De Gruyter, 2012
Median
86.5
2
188.2
5.3
2,117
669.8
600.0
27
132
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
Appendix B
Table 6: Labor income, alternative reference years.
1
2
3
young heirs,
21−49
4
old heirs,
50−59
Inherited amount*Year
2000–2003, reference
2004
2005
2006
2007
2008
Age2 /100
Age3 /10,000
Constant
Individual fixed effects
-0.005
(0.021)
-0.027
(0.020)
-0.071**
(0.030)
-0.064***
(0.024)
-0.047**
(0.021)
-134
(790)
-0.331
(5.666)
638
(1,130)
Yes
0.001
(0.019)
-0.019
(0.018)
-0.060**
(0.028)
-0.050**
(0.024)
-0.030
(0.026)
-116
(781)
-0.468
(5.589)
2,564
(3,593)
Yes
-0.008
(0.021)
-0.023
(0.025)
-0.107
(0.066)
-0.044
(0.028)
-0.040
(0.028)
-147
(1,425)
-0.707
(12.02)
538
(1,419)
Yes
0.000
(0.031)
-0.026
(0.028)
-0.047
(0.029)
-0.070**
(0.035)
-0.045
(0.030)
-16,724
(14,306)
98
(87)
30,981
(25,862)
Yes
Year dummies
Yes
Yes
Yes
Yes
Individual specific
time trends
Number of heirs
No
Yes
No
No
374
374
150
224
Number of obs
3,310
3,310
1,326
Notes. Amounts are measured in SEK thousands, price level 2004.
Standard errors are clustered on family.
* significant at the 10 percent level, ** significant at the 5 percent level,
*** significant at the 1 percent level
1,984
28
133
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
Table 7: Sensitivity analyses, Labor income.
Inherited amount * Year:
2000
2001
2002
2003, reference
2004
2005
2006
2007
2008
1
extended
Labor
incomea
2
log
Labor
incomeb
3
Labor income
± 1.96 sd
from meanc
4
inheritance
at market
valuesd
no age
restrictions
6
heirs with
Capital income
<0 in 2000
7
heirs with
Capital income
>0 in 2000
-0.0414
(0.0294)
-0.0493*
(0.0288)
-0.0160
(0.0182)
-0.000066
(0.000088)
-0.000103
(0.000074)
-0.000007
(0.000065)
-0.00438
(0.0197)
-0.0126
(0.0179)
-0.00867
(0.0183)
-0.0248
(0.0276)
-0.0339
(0.0280)
-0.0112
(0.0162)
-0.0318
(0.0276)
-0.0449
(0.0282)
-0.00556
(0.0159)
-0.011
(0.031)
-0.009
(0.031)
-0.010
(0.032)
-0.043
(0.049)
-0.066
(0.051)
-0.013
(0.021)
0.0625
(0.0848)
-0.0548**
(0.0218)
-0.101***
(0.0289)
-0.0805***
(0.0244)
-0.0655***
(0.0206)
-0.000083
(0.000107)
-0.000209*
(0.000108)
-0.000248**
(0.000103)
-0.000352**
(0.000140)
-0.000438**
(0.000180)
-0.0147
(0.0166)
-0.0422*
(0.0218)
-0.0715***
(0.0254)
-0.0652***
(0.0212)
-0.0632***
(0.0211)
-0.0263
(0.0175)
-0.0410*
(0.0208)
-0.0865***
(0.0286)
-0.0815***
(0.0235)
-0.0641***
(0.0202)
0.00826
(0.0313)
-0.0152
(0.0418)
-0.0834***
(0.0244)
-0.0803***
(0.0222)
-0.0827***
(0.0227)
-0.012
(0.029)
-0.041
(0.028)
-0.062**
(0.028)
-0.072**
(0.032)
-0.070**
(0.031)
-0.036
(0.026)
-0.053*
(0.028)
-0.123**
(0.050)
-0.096***
(0.035)
-0.067**
(0.029)
14.0
(79.4)
-23.9
(56.4)
264
(1,246)
Yes
Yes
197
1,765
-32.4
(101.0)
11.8
(74.4)
966
(1,567)
Yes
Yes
177
1,545
Age2 /100
5
-51.7
-0.148
-35.0
-9.7
-24.7
(72.4)
(0.641)
(75.0)
(79.8)
(46.2)
24.7
0.092
17.1
-5.9
6.7
(52.5)
(0.435)
(52.3)
(57.1)
(29.3)
Constant
1,279
8.047
927
625
941
(1,129)
(10.345)
(1,187)
(1,250)
(873)
Individual fixed effects
Yes
Yes
Yes
Yes
Yes
Year indicators
Yes
Yes
Yes
Yes
Yes
Number of heirs
374
374
359
374
552
Number of observations
3,310
3,310
3,183
3,310
4,891
Notes. Amounts are measured in SEK thousands, price level 2004. Standard errors are clustered on family.
a the sum of Labor and Self-employment incomes. b Labor income in natural logarithms.
c heirs with incomes within 1.96 s.d. from the sample means.
d inherited amount is market value adjusted with respect to inherited taxable real estate wealth.
* significant at the 10 percent level, ** significant at the 5 percent level,
*** significant at the 1 percent level
Age3 /10,000
134
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
Table 8: Capital income, alternative reference years.
1
2
3
young heirs,
21−49
4
old heirs,
50−59
Inherited amount*Year
2000–2003, reference
2004
2005
2006
2007
2008
Age2 /100
Age3 /10,000
Constant
Individual fixed effects
-0.034
(0.061)
0.254***
(0.071)
0.192***
(0.062)
0.160
(0.114)
0.026
(0.057)
55
(776)
-0.858
(6.758)
-47
(986)
Yes
-0.138
(0.098)
0.127
(0.084)
0.091
(0.120)
0.090
(0.113)
-0.013
(0.100)
801
(1,109)
7.874
(9.625)
777
(1,122)
Yes
0.037
(0.095)
0.347***
(0.087)
0.260***
(0.062)
0.221
(0.182)
0.058
(0.077)
-46,632**
(20,892)
281**
(126)
84,556
(37,810)
Yes
-0.026
(0.058)
0.264***
(0.064)
0.205***
(0.054)
0.177
(0.118)
0.046
(0.059)
77
(746)
-1.019
(6.518)
2,342
(5,811)
Yes
Year dummies
Yes
Yes
Yes
Yes
Individual specific
time trends
Number of heirs
No
Yes
No
No
374
374
150
224
Number of obs
3,310
3,310
1,326
Notes. Amounts are measured in SEK thousands, price level 2004.
Standard errors are clustered on family.
* significant at the 10 percent level, ** significant at the 5 percent level,
*** significant at the 1 percent level
1,984
30
135
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
Table 9: Sensitivity analyses, Capital income.
Inherited amount * Year:
2000
2001
2002
2003, reference
2004
2005
2006
2007
2008
1
adjusted
Capital
incomea
2
log
Capital
incomeb
3
Capital income
± 1.96 sd
from meanc
4
inheritance
at market
valuesd
no age
restrictions
6
heirs with
Capital income
<0 in 2000
7
heirs with
Capital income
>0 in 2000
0.192
(0.154)
-0.104
(0.111)
-0.0196
(0.0240)
0.00108*
(0.00056)
0.00028
(0.00036)
0.00021
(0.00036)
0.021
(0.036)
0.000322
(0.0203)
-0.0265
(0.0281)
0.163
(0.139)
-0.102
(0.110)
-0.0155
(0.0213)
0.140
(0.135)
-0.111
(0.0825)
-0.0317
(0.0242)
-0.019
(0.018)
0.018
(0.030)
-0.061
(0.042)
0.343
(0.236)
-0.226
(0.232)
0.018
(0.020)
-0.0127
(0.0290)
0.275***
(0.0952)
0.214***
(0.0650)
0.184*
(0.0980)
0.0519*
(0.0264)
-0.00011
(0.00046)
0.00179***
(0.00042)
0.00166***
(0.00045)
0.00149***
(0.00049)
0.00090
(0.00056)
-0.0133
(0.0348)
0.205***
(0.0558)
0.224***
(0.0731)
0.114**
(0.0572)
0.0482
(0.0309)
-0.0179
(0.0277)
0.241***
(0.0915)
0.190***
(0.0646)
0.162*
(0.0826)
0.0414*
(0.0242)
-0.0236
(0.0301)
0.236***
(0.0854)
0.198***
(0.0581)
0.155*
(0.0846)
0.0375
(0.0262)
-0.048
(0.041)
0.148**
(0.058)
0.152**
(0.061)
0.167*
(0.086)
0.049
(0.048)
0.015
(0.038)
0.371***
(0.123)
0.257***
(0.087)
0.184
(0.163)
0.041
(0.031)
31.5
(62.1)
-20.8
(45.3)
-527
(960)
Yes
Yes
197
1,765
-65.1
(131.2)
40.3
(114.0)
1,112
(1,804)
Yes
Yes
177
1,545
Age2 /100
5
7.6
2.12
14.7
4.2
-23.5
(76.3)
(1.75)
(35.6)
(76.7)
(45.3)
-10.3
-1.86
-7.8
-7.3
16.8
(66.5)
(1.22)
(28.5)
(66.8)
(30.2)
Constant
-72
-27.2
-265
-13
415
(1,046)
(28.2)
(520)
(1,052)
(834)
Individual fixed effects
Yes
Yes
Yes
Yes
Yes
Year indicators
Yes
Yes
Yes
Yes
Yes
Number of heirs
374
308
365
371
552
Number of observations
3,310
1,502
3,237
3,283
4,891
Notes. Amounts are measured in SEK thousands, price level 2004. Standard errors are clustered on family.
a Capital income is net of the return on the inherited amount.
The annual rate of return on inherited capital (2000–2008) is 4.67 percent.
b Capital income in natural logarithms.
c The sample is limited to heirs with incomes within 1.96 s.d. from the sample means.
d Inherited amount is market value adjusted with respect to inherited taxable real estate wealth.
* significant at the 10 percent level, ** significant at the 5 percent level,
*** significant at the 1 percent level
Age3 /10,000
Published by De Gruyter, 2012
31
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
References
Abel, A. and M. Warshawsky (1988): “Specification of the joy of giving: Insights
from altruism,” Review of Economics and Statistics, 70, 145–149.
Angelini, V. (2009): “The strategic bequest motive: Evidence from SHARE,”
Working paper, University of Padua, draft.
Barro, R. J. (1974): “Are government bonds net wealth?” Journal of Political Economy, 82, 1095–1117.
Becker, G. S. (1974): “A theory of social interactions,” Journal of Political Economy, 82, 1063–1093.
Behrman, J. R. and M. R. Rosenzweig (2004): “Parental allocation to children:
New evidence on bequest differences among siblings,” Review of Economics and
Statistics, 86, 637–640.
Bennedsen, M., F. Pérez-González, and D. Wolfenzon (2010): “Do CEOs matter?”
Working paper.
Berg, L. (2006): “Förmögenhetskällan: De svenska hushållens förmögenhetsutveckling,” Rapport, Nordea, Stockholm.
Bernheim, B. D., R. J. Lemke, and J. K. Scholz (2004): “Do estate and gift taxes
affect the timing of private transfers?” Journal of Public Economics, 88, 2617–
2634.
Bernheim, B. D., A. Shleifer, and L. H. Summers (1985): “The strategic bequest
motive,” Journal of Political Economy, 93, 1045–1076.
Blanchflower, D. G. and A. J. Oswald (1998): “What makes an entrepreneur?”
Journal of Labor Economics, 16, 26–60.
Blinder, A. (1976): “Intergenerational transfers and life cycle consumption,” American Economic Review, 66, 87–93.
Blomquist, S. (1979): “The inheritance function,” Journal of Public Economics, 12,
41–60.
Blomquist, S. and H. Selin (2010): “Hourly wage rate and taxable labor income
responsiveness to changes in marginal tax rates,” Journal of Public Economics,
94, 878–889.
Blundell, R. and T. MaCurdy (1999): “Labor supply: a review of alternative approaches,” in O. Ashenfelter and D. Card, eds., Handbook of Labor Economics,
volume 3A, Amsterdam: North-Holland, 1559–1695.
Bolin, K., B. Lindgren, and P. Lundborg (2008): “Your next of kin or your own
career? Caring and working among the 50+ of Europe,” Journal of Health Economics, 27, 718–738.
Brattström, M. and A. Singer (2007): Rätt arv: Fördelning av kvarlåtenskap, Uppsala: Iustus Förlag, 2 edition.
32
137
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
Brown, J. R., C. C. Coile, and S. J. Weisbenner (2010): “The effect of inheritance
receipt on retirement,” Review of Economics and Statistics, 92, 425–434.
Card, D. (1994): “Intertemporal labour supply: An assessment,” in C. A. Sims,
ed., Advances in Econometrics: Sixth World Congress, volume II, Cambridge
University Press, chapter 2, 49–80.
Cheng, I.-H. and E. French (2000): “The effect of the run-up in the stock market on
labor supply,” Economic Perspectives, 24, 48–65.
Coile, C. C. and P. B. Levine (2006): “Bulls, bears, and retirement behavior,” Industrial & Labor Relations Review, 59, 408–429.
Coronado, J. L. and M. G. Perozek (2003): “Wealth effects and the consumption of
leisure: Retirement decisions during the stock market boom of the 1990s,” FEDS
Papers 2003-20, Board of Governors of the Federal Reserve System, Washington,
D.C.
Cox, D. (1987): “Motives for private income transfers,” Journal of Political Economy, 95, 508–546.
Eliason, M. and H. Ohlsson (2010): “Timing of death and the repeal of the Swedish
inheritance tax,” Working Paper 2010:2, Uppsala Center for Fiscal Studies, Department of Economics, Uppsala University.
Elinder, M., O. Erixson, and H. Ohlsson (2010): “The effect of inheritance receipt
on labor and capital income: Evidence from Swedish panel data,” Working Paper
2010:3, Uppsala Center for Fiscal Studies, Department of Economics, Uppsala
University.
Engelhardt, G. V. (1996): “House prices and home owner saving behavior,” Regional Science and Urban Economics, 26, 313–336.
Ettner, S. (1995): “The impact of “parent care” on female labor supply decisions,”
Demography, 32, 63–80.
Faria, J. R. and Z. Wu (2012): “From unemployed to entrepreneur: The role of the
absolute bequest motive,” Economics Letters, 114, 120–123.
Feldstein, M. (1995): “The effect of marginal tax rates on taxable income: A panel
study of the 1986 tax reform act,” Journal of Political Economy, 103, 551–572.
Feldstein, M. (1999): “Tax avoidance and the deadweight loss of the income tax,”
Review of Economics and Statistics, 81, 674–680.
Fevang, E., S. Kverndokk, and K. Røed (2012): “Labor supply in the terminal
stages of lone parents’ lives,” Journal of Population Economics, 25, 1399–1422.
Gruber, J. and E. Saez (2002): “The elasticity of taxable income: Evidence and
implications,” Journal of Public Economics, 84, 1–32.
Guiso, L., M. Haliassos, and T. Jappelli (2002): Household portfolios, Cambridge,
MA: MIT Press.
Holtz-Eakin, D., D. Joulfaian, and H. S. Rosen (1993): “The Carnegie conjecture:
Some empirical evidence,” Quarterly Journal of Economics, 108, 413–435.
Published by De Gruyter, 2012
33
138
The B.E. Journal of Economic Analysis & Policy, Vol. 12 [2012], Iss. 1 (Contributions), Art. 61
Holtz-Eakin, D., D. Joulfaian, and H. S. Rosen (1994a): “Entrepreneurial decisions
and liquidity constraints,” RAND Journal of Economics, 25, 334–347.
Holtz-Eakin, D., D. Joulfaian, and H. S. Rosen (1994b): “Sticking it out: Entrepreneurial survival and liquidity constraints,” Journal of Political Economy,
102, 53–75.
Horioka, C. Y. (2009): “Do bequests increase or decrease wealth inequalities?”
Economics Letters, 103, 23–25.
Hurst, E. and A. Lusardi (2004): “Liquidity constraints, household wealth, and
entrepreneurship,” Journal of Political Economy, 112, 319–347.
Imbens, G., D. Rubin, and B. Sacerdote (2001): “Estimating the effect of unearned
income on labor earnings, savings, and consumption: Evidence from a survey of
lottery players,” American Economic Review, 91, 778–794.
Joulfaian, D. (2004): “Gift taxes and lifetime transfers: time series evidence,” Journal of Public Economics, 88, 1917–1929.
Joulfaian, D. (2006): “Inheritance and saving,” NBER Working Paper No. 12569,
National Bureau of Economic Research, Cambridge, MA.
Joulfaian, D. (2011): “The federal estate tax: History, law, and economics,” Working paper, Office of Tax Analysis, US Department of the Treasury, Washington,
DC, draft.
Joulfaian, D. and M. O. Wilhelm (1994): “Inheritance and labor supply,” Journal
of Human Resources, 29, 1205–1234.
Juster, F. T., J. P. Lupton, J. P. Smith, and F. Stafford (2006): “The decline in
household saving and the wealth effect,” Review of Economics and Statistics, 88,
20–27.
Juster, F. T., J. P. Smith, and F. Stafford (1999): “The measurement and structure of
household wealth,” Labour Economics, 6, 253–275.
Kessler, R. (1997): “The effects of stressful life events on depression,” Annual
Review of Psychology, 48, 191–214.
Kimball, M. (1990): “Precautionary saving in the small and in the large,” Econometrica, 58, 53–73.
Kopczuk, W. (2005): “Tax bases, tax rates and the elasticity of reported income,”
Journal of Public Economics, 89, 2093–2119.
Kopczuk, W. (2007): “Bequest and tax planning: Evidence from estate tax returns,”
Quarterly Journal of Economics, 122, 1801–1854.
Kopczuk, W. (2009): “Economics of estate taxation: A brief review of theory and
evidence,” Tax Law Review, 63, 139–157.
Kopczuk, W. and E. Saez (2004): “Top wealth shares in the United States, 1916–
2000: Evidence from estate tax returns,” National Tax Journal, 57, 445–488.
Kurz, M. (1984): “Capital accumulation and the characteristics of private intergenerational transfers,” Economica, 51, 1–22.
34
139
Elinder et al.: The Impact of Inheritances on Heirs' Labor and Capital Income
Lindh, T. and H. Ohlsson (1996): “Self-employment and windfall gains: Evidence
from the Swedish lottery,” Economic Journal, 106, 1515–1526.
MaCurdy, T. E. (1981): “An empirical model of labor supply in a life-cycle setting,”
Journal of Political Economy, 89, 1059–1085.
Nordblom, K. and H. Ohlsson (2006): “Tax avoidance and intra-family transfers,”
Journal of Public Economics, 90, 1669–1680.
Ohlsson, H. (2007): “The equal division puzzle – empirical evidence on intergenerational transfers in Sweden,” Working Paper 2007:10, Department of Economics,
Uppsala University.
Ohlsson, H. (2011): “The legacy of the Swedish gift and inheritance tax,” European
Review of Economic History, 15, 539–569.
Pencavel, J. (1986): “Labor supply of men: A survey,” in O. Ashenfelter and R. Layard, eds., Handbook of Labor Economics, volume 1, Amsterdam: North Holland, chapter 1, 3–102.
Pestieau, P. (2003): “The role of gift and estate transfers in the United States and in
Europe,” in A. H. Munnell and A. Sundén, eds., Death and Dollars. The Role of
Gifts and Bequests in America, Washington, DC: Brookings, chapter 3, 64–85.
Piketty, T. (2011): “On the long-run evolution of inheritance: France 1820–2050,”
Quarterly Journal of Economics, 126, 1071–1131.
Poterba, J. M. (2000): “Stock market wealth and consumption,” Journal of Economic Perspectives, 14, 99–118.
Saez, E., J. B. Slemrod, and S. H. Giertz (2012): “The elasticity of taxable income
with respect to marginal tax rates: A critical review,” Journal of Economic Literature, 50, 3–50.
Schulz, R., A. Mendelsohn, W. Haley, D. Mahoney, R. Allen, M. S. Zhang,
L. Thompson, and B. Belle (2003): “End-of-life care and the effects of bereavement on family caregivers of persons with dementia,” New England Journal of
Medicine, 349, 1936–1942.
Umberson, D. and M. Chen (1994): “Effects of a parent’s death on adult children:
Relationship salience and reaction to loss,” American Sociological Review, 59,
152–168.
Umberson, D., C. B. Wortman, and R. C. Kessler (1992): “Widowhood and depression: Explaining long-term gender differences in vulnerability,” Journal of
Health and Social Behavior, 33, 10–24.
Weil, D. N. (1994): “The saving of the elderly in micro and macro data,” Quarterly
Journal of Economics, 109, 55–81.
Weil, D. N. (1996): “Intergenerational transfers, aging, and uncertainty,” in D. A.
Wise, ed., Advances in the Economics of Aging, Chicago: University of Chicago
Press, chapter 10, 321–342.
Published by De Gruyter, 2012
35
140
Essay 4: Health responses to a wealth shock:
Evidence from a Swedish tax reform
♠
♠
Valuable comments and suggestions from Adrian Adermon, Mikael Elinder, Mikael Lindahl, Eva Mörk, Katarina Nordblom, Mattias Nordin, Henry Ohlsson, Håkan Selin and Erik
Spector, and seminar participants at IFN Stockholm and the Department of Economics at
Uppsala University are gratefully acknowledged. Sebastian Escobar provided excellent research assistance. Some of the work was done when I enjoyed the hospitality of the Department of Economics, Columbia University. Financial support from the Jan Wallander and Tom
Hedelius Foundation is gratefully acknowledged.
141
1 Introduction
It has long been recognized that there exists a positive relationship between
many measures of economic wealth and a variety of health outcomes.39
This ‘gradient’ has become a significant concern for politicians and public
health officials as it implies that inequalities between rich and poor do not
only appear as differences in consumption and material well-being, but also
in life-expectancy and quality of life. Unfortunately, any policy intervention
targeted at reducing these inequalities, or promoting public health in general,
suffers from the fact that we still know little about if and how economic
wealth affects health.
Answering these questions is further complicated by the possibility that
causation may go in the opposite direction, from health to wealth.40 It could
also be that unobserved factors, such as genetic endowment, early childhood
exposures or time preferences, influence wealth and health in the same direction without a causal link.41
Given the practical constraints on randomizing people to receive different
amounts of wealth, researchers have tried to solve these methodological
challenges with quasi-experimental designs, in particular by exploiting exogenous variation generated from individual wealth or income shocks. Important examples include lottery winnings (Lindahl, 2005; Gardner and
Oswald, 2007), stock market fluctuations (Schwandt, 2012), inheritances
(Meer et al. 2003; Kim and Ruhm 2012; Carman 2013) and unanticipated
policy changes (Jensen and Richter, 2003; Case, 2004; Frijters et al., 2005;
Snyder and Evans, 2006).42 The general finding is that wealth and income
have limited impacts on adult health in the short to medium run.
Previous studies are limited by the fact that they are based almost entirely
on survey data on subjective general health status. Although it has been argued that general health status is a good predictor of future morbidity and
mortality (Idler and Benjamini, 1997; van Doorslaer and Gerdtham, 2003),
there are reasons to question general health status as a dependent variable in
this context. Subjective health status is, for example, likely to be influenced
by factors such as social norms regarding health, use of health care as well as
understanding of the survey questions, which are themselves systematically
related to wealth and income in such a way that the coefficient estimates are
biased towards zero (see for example Murray and Chen, 1992 and Bago
39
See Marmot (1999), Smith (1999), Deaton (2003), and Cutler et al. (2011) for reviews of
the literature.
40
For examples of studies investigating the impact of health shocks on labor market outcomes, see Lundborg et al. (2011), and on wealth, see Wu (2003).
41
For studies discussing these issues, see for example Barker (1997), Almond and Currie
(2013), Fuchs (1982) and Barsky et al. (1997).
42
Other quasi-experimental designs in this context include IV estimators (see for instance
Ettner, 1996) and Granger causality testing (see for instance Adams et al., 2003 and Michaud
and van Soest, 2008).
142
d’Uva et al., 2008). Moreover, subjective general health status does not separate between different aspects of health. For instance, it has been shown that
improved wealth leads to, on the one hand, harmful behaviors like smoking
and drinking and, on the other hand, to reduced obesity, lower stress and
enhanced mental well-being, suggesting that important health effects may go
unnoticed (Lindahl, 2005; Apouye and Clark, 2013; Kim and Ruhm, 2012).
This paper tackles causality by exploiting a previously untapped and policy-relevant source of exogenous variation in wealth; namely the repeal of
the Swedish inheritance tax on December 17, 2004.43 Heirs who received
inheritance above the tax threshold from parents who passed away after the
reform are defined as being treated as they experienced a favorable wealth
shock equal to what their tax payments would have been had the decedent
died before the reform. Calculations indicate that the wealth shock amounted
to, on average, SEK 70,000 (about USD 9,500 in 2004 values), or 7 percent
of initial wealth. The empirical strategy is to estimate the causal effect of the
wealth shock on health by approximating the counterfactual outcome with
the health experiences of heirs who received inheritance above the tax
threshold before the reform date. The relevant sample is collected from an
administrative database covering the entire population of heirs of deceased
Swedes over the period 2003–2005. Results from several tests show that the
treated and the controls are comparable in predetermined characteristics,
including health, implying that any difference in health between the two
groups following the inheritance could reasonably be attributed to the wealth
shock. I also conduct placebo experiments which tests for responses in a
sample of heirs who received parental bequests below the tax threshold and
hence, for whom the reform should have no impact. The results from these
tests support the validity of the empirical strategy.
The health outcomes are collected from medical records, death certificates
and the Swedish sickness insurance register and share the feature that they
are based on the medically qualified opinions of physicians. As far as I am
aware, this is the first study to use objective health outcomes from administrative registers, other than mortality, to investigate the effects of increased
economic resources on health.
The main health outcome is an indicator of whether the individual has
been hospitalized for any cause in a given year. Comparing the incidences of
hospitalization between the treated and controls over time, ten years before
and six year after the inheritance, show that the wealth shock increases the
probability of hospitalization by five percent. This is equal to the impact of
being four years older. Tests for heterogeneous responses suggest that the
effect is primarily driven by the relatively old, women and those with a low
level of education.
43
Eliason and Ohlsson (2013) use the repeal of the inheritance tax to study behavioral responses to taxation among individuals leaving inheritances.
143
At a first blush, the positive effect on hospitalization may be interpreted
as if the wealth shock has detrimental consequences for health, especially
since health care in Sweden is universal and basically free of charge.44 Tests
for heterogeneous responses across diagnoses reported in connection with
the hospital admissions show, however, that the wealth effect is evident in
only two diagnose categories: ‘symptoms of disease’ (e.g. shortness of
breath, fever, general feeling of illness, etc.) and ‘cancer’. Regarding cancer,
previous studies document that improved wealth leads to more smoking and
drinking, behaviors which are positively related with the disease. That the
current wealth effect is operating through these channels seems unlikely,
however, given the relatively limited time period over which it is estimated.
If the wealth shock leads to more smoking and drinking I should rather see
responses in diagnoses which are more immediately related with these risk
factors (e.g. injuries, mental problems, respiratory diseases, etc.). Likewise,
if the shock leads to reduced obesity or improved mental well-being (which
has also been indicated by previous studies) I should be more likely to find a
reduction in cancer incidence rather than an increase. A more realistic explanation is therefore that cancer has been detected during health care visits
for minor health contingencies (i.e. symptoms of disease). That the wealth
shock leads to more health care visits, although health care in Sweden is
free, could potentially be explained by people demanding good health to
fully benefit from their improved future consumption prospects.
To get a better understanding of how the wealth shock affects different
aspects of health, I test for responses in (publicly insured) sick leave amounting to more than two weeks and in all-cause mortality, as these two health
outcomes are likely to capture health events which are both less and more
severe than those resulting in hospital admissions. The results show that the
wealth shock does not have any statistically significant effects on either of
the two outcomes. Although the insignificant wealth effect on sick leave
may be attributed to the fact that the analysis is based on the working-age
population (for whom the wealth shock has no detectable effect on hospitalization), the finding lends additional support for the conclusion that the
wealth shock has negligible consequences for health. The insignificant effect
on mortality is expected given the insignificant effect on the prevalence of
diseases other than cancer (for which the impact is apparently too small to
translate into mortality, at least over a period of six years).
In sum, the results show that more wealth has limited short to medium run
consequences for objective adult health. This is line with what has been
found regarding subjective health. It appears however as if the wealth shock
leads to preventive behaviors, which may have long-term beneficial consequences for health.
44
See Glengård et al. (2005) for an excellent description of the Swedish health care system.
144
The outline of this paper is as follows. In Section 2, I discuss the theoretical predictions regarding the effect of wealth on health together with an
overview of the previous empirical literature. Section 3 describes the inheritance tax, with a particular focus on the unexpected repeal. In section 4, I
discuss the data used in the empirical analysis. Section 5 presents the empirical strategy and in section 6, I present evidence that the wealth shock is exogenous. Section 7 provides the results and section 8, finally, is a concluding
discussion. Each section begins with a short summary of its main points.
2 Review of related literature
This section starts with a discussion on the theoretical arguments for why
increased wealth may affect health. The second sub-section gives a review of
the previous empirical literature. The general finding is that wealth shocks
have a limited impact on self-assessed general health status and longevity. It
appears, however, as if improved financial resources, on the one hand, leads
people to engage more in health behaviors and lifestyles which are possibly
detrimental in the long run (e.g. smoking and drinking) and, on the other
hand, have beneficial consequences in form of reduced obesity, lower stress
and improved mental well-being.
2.1 Theoretical arguments for causal effects of wealth on health
The common hypothesis in the literature is that improved economic resources lead to better health. Although it is largely motivated by stylized
facts regarding the positive correlation between wealth and health, theoretical support for the hypothesis can be found in Grossman’s model of health
capital (Grossman, 1972, 2000).45 According to this model, people demand
health for the consumption benefits (health provides utility), in addition to
the production benefits (more healthy time available for activities such as
work, consumption and health investments). Healthy time available for market and non-market activities depends on the stock of health capital, which
depreciates over the lifecycle to a critical threshold where death occurs. The
individual, however, may counteract the natural deterioration process by
investing in her health. In accordance with Becker’s model of household
production (Becker, 1976), health is produced by combining market goods
and time. More wealth will make health investments subjectively cheaper,
lead to increased demand for health and, eventually, improved health.
In recent years, additions to the health-capital model have been made to
account for the possibility that the individual derives utility not only from
45
See Muurinen (1982) and Ehrlich and Chuma (1990) for extensions of the Grossman
framework.
145
health enhancing consumption (e.g. healthy foods and exercise), but also
from consumption which is negatively correlated with health (e.g. drinking
and smoking), see for example Galama and Van Kippersluis (2010) and Van
Kippersluis and Galama (2013).46 According to these models, improved economic resources will relax the individual’s budget constraint allowing a
higher level of both types of consumption. Nevertheless, as unhealthy consumption is associated with a cost in the form of reduced health and shorter
lifespan, the rise in healthy consumption will be relatively larger.
2.2 Findings in the previous literature
Three previous studies have used inheritances to identify the effects of
wealth on health. Meer et al., (2003) use data from the Panel Study of Income Dynamics to analyze the impact of wealth on self-reported health
status. The authors use receiving an inheritance as an instrument for changes
in wealth and find what they interpret as “a quantitatively small effect” and
conclude that the wealth-health connection is not driven by short-term
changes in wealth. There are two concerns regarding the identification strategy employed by Meer et al. First, inheritances need not randomly distributed, but correlated with unobserved determinants of health. Second, the
interpretation of the effect is complicated by the possibility that inheritances
are anticipated. If the heir has adjusted her health behavior or lifestyle in
anticipation of the inheritance, the estimate will then understate the true effect. In a related study, Kim and Ruhm, (2011) compare health consequences
of people in the Health and Retirement Study (HRS) who have received inheritances in excess of $10,000 with people who have inherited small
amounts (<$10,000), which are assumed to not affect health. The authors
attempt to account for unobserved individual heterogeneity by estimating
models with large sets of observable characteristics, including lagged health,
and they exploit data on the individual’s subjective probability of receiving
an inheritance in order to address the issue of possible anticipatory effects.
The results show that the wealth shock has no effect on self-reported health
status, but that it seems to lead to an increase in the prevalence and intensity
of social drinking, in addition to a reduction in obesity. In a recent study,
Carman (2013) contributes to the two previous studies by comparing the
results from models with and without individual fixed effects to test for the
influence of unobserved heterogeneity across individuals who receive and
not receive inheritance in the PSID. Her first main result is that the inherited
amount does not have any effect on self-reported health status, independ46
These extensions are largely motivated by epidemiological research which documents that a
large fraction of the socioeconomic disparities in adult health in developed countries can be
accounted for by disparities in lifestyles and consumption (McGinnis and Foege, 1993; Mokdad et al., 2004; Contoyannis and Jones, 2004; Cutler et al., 2011).
146
ently of model specification. Her second main result is that the effect of receiving inheritance (irrespectively of amount) is positive and significant in
the no-fixed-effects specification, but not in the fixed-effects specification.
This suggests that individuals who receive inheritances have better health
than those who do not receive inheritances, but that there is no change in
health following the receipt.
Another source of plausibly exogenous variation in economic resources is
lottery winnings. Using data on lottery winners from the Swedish Level of
Living Surveys, Lindahl (2005) finds that increased income is associated
with improved health, measured by an index of self-reported illnesses and
symptoms, as well as increased life expectancy. The income effect on health
appears to be strongest for the oldest individuals. Moreover, Lindahl (2005)
finds evidence of decreased obesity as a result of higher lottery winnings,
suggesting that wealth may affect health through health-related consumption,
such as exercise and healthy food. Unfortunately, however, the sample is
limited to winners and contains no information on the frequency of lottery
playing. In a related study, Gardner and Oswald (2007) focus solely on lottery winners in the British Household Panel Survey and identify causation
with variation in the size of the prize. By doing so, they implicitly assume
that winners of small and large prizes have similar unobserved characteristics, which is not obvious. Their results show that winning a large prize,
compared to a small, enhance subjective mental well-being two years after
winning. Apouye and Clark (2013) use the same dataset and identification
strategy as Gardner and Oswald to test for responses, not only in mental
well-being, but also in self-reported measures of physical and general health.
Their results show that the wealth shock has no detectable effect on general
health but that it produces better mental health. The authors explain the lack
of effect in the former variable by showing that winning the lottery leads to
more smoking and drinking, behaviors with plausibly detrimental effects on
general health. The main objection against lottery winnings is that they are
randomly assigned and only conditional on participation in the lottery and,
thus, that the results may be confounded by selection bias (Van Kippersluis
and Galama, 2013). More specifically, because lottery players tend to have
lower incomes and less education than non-players, the empirical estimates
are likely to generalize only to the lower segments of the socioeconomic
distribution.
Stock-market fluctuations constitute another source of variation in wealth
which is unlikely to be induced by health (Smith, 1999). Schwandt (2012)
exploits the wealth gains and losses generated in the US stock market during
a time-period of 18 years. Using data on a sample of retirees from the HRS,
he finds that a ten percent wealth increase over two years leads to a significant improvement in an index constructed of different survey measures of
physical and mental health, as well as reduced mortality. It appears as if the
wealth shock reduces the incidence of diseases of the heart, hypertension and
147
psychiatric problems, suggesting that psychological factors may be the
mechanism through which the wealth effect operates. As with lottery winnings, however, stock market swings are experienced by a specific subset of
the population, which in this case tend to be relatively wealthy (Mankiw and
Zeldes, 1991; Poterba and Samwick, 2003; Smith, 2004).
A second branch of studies in the field have exploited variation in income
and wealth generated by changes in government policies. One advantage
with policy changes is that they usually affect a larger segment of the population. Therefore, they may be more relevant from a policy perspective than
individual shocks. Using cross-sectional data on self-reported health status
of Black South Africans who had their income doubled due to a change in
the pension system, Case (2004) finds evidence of improvements in general
health. These, interestingly, not only manifest themselves for the recipient,
but for all household members. Moreover, Case shows that the effect is
likely to stem from improved sanitation, housing, health care as well as reduced stress. It is, however, unclear whether these results are applicable to a
Western population. Jensen and Richter (2003) study a pension crisis in Russia during which many retirees did not receive their pensions for an extended
period of time. The average decrease in income for this group was 24 percent. Examining the longitudinal effects of this adverse shock, the authors
find evidence of reduced nutritional intake and utilization of health care in
the short run. They also find that the likelihood to die in the two years following the crisis increased by five percent. Similarly, Snyder and Evans
(2006) use a legislative change in the US Social Security system which unexpectedly lowered the benefits for people born after January 1, 1917 - the
so called “Notch” generation. A comparison of five-year mortality rates after
age 65 for males born in the first quarter of 1917 and the last quarter of 1916
show that the Notch had slightly lower five-year mortality rates than the
previous cohort. The authors suggest that this countervailing finding is partly
due to the fact that the people in the Notch cohort increased their postretirement labor supply, which in turn had beneficial health effects through
reduced social isolation. Fritjers et al. (2005) take advantage of the fact that
the German reunification in 1990 resulted in large income transfers to the
East German population but not to West Germans. As the collapse of East
Germany was unanticipated, the authors could attribute differences in health
consequences between the two groups to the resulting increase in real income. The results show a significant, but small, positive effect of the income
shock on health satisfaction.
148
3 The Swedish inheritance tax and how it was
unexpectedly repealed
This section begins with a short description of taxation of inheritance prior
to the repeal. This is to get an understanding of the source of variation I use
to identify the causal effect of wealth on health. After that, I discuss the way
in which the tax reform was proposed, passed and implemented. The main
point is that the decision to repeal the tax was largely unexpected and that
the reform was enacted in a rapid way. This would imply that the affected
population had limited incentives or abilities to react vis-à-vis the reform
before it was implemented.
3.1 Taxation on inheritances before the reform
Prior to December 2004, legal heirs and beneficiaries of wills in Sweden
were subject to inheritance taxation according to the laws stipulated in the
Inheritance and Gift Tax Ordinance.47 The inheritance tax, similarly, depended on the succession scheme of the relationship between the deceased
and the heir.48 For the deceased’s descendants (i.e. the deceased’s children
and their descendants), amounts exceeding a basic deductible exemption of
SEK 70,000 were taxed according to a progressive tax schedule consisting of
three marginal tax brackets of: 10 percent, 20 percent and 30 percent. Table
1 reports the tax schedule for the deceased’s descendants.
Table 1: Tax rates on inheritances for the deceased’s descendants.
Taxable inheritance
0-70
Tax rate
0
70-370
370-670
670-
10%
30+20% within bracket
90+30% within bracket
Note. All monetary values are in 1,000SEK.
3.2 The unexpected reform
Concerned with the growing criticism against the inheritance tax, the Social
Democratic Government announced, in the Budget on September 20, 2004,
47
See Ohlsson (2011) and Du Rietz et al. (2012) for excellent historical reviews of the inheritance tax.
48
The law defined three classes of taxpayers. Class 1 contained the children and their descendants, and, before 2003, spouses and cohabiters. Class 2 constituted all other legal heirs, and
Class 3 legal entities such as public institutions, charities and foundations.
149
that the Inheritance and Gift Tax Ordinance was to be repealed starting
January 1, 2005.49
The legislation had been criticized for complicating distributions of estates, especially those involving transfers of family firms. Escalating tax
values on real estate in the early 2000’s had also led to public criticism of the
inheritance tax, as many heirs, especially widows, had difficulties affording
the increasingly large tax payments. Although the general impression was
that the legislation was in need of a reform, the Government’s decision to
completely remove the tax came as a surprise (Silfverberg, 2005). The tax on
bequests to spouses had been removed in January 2004, but at that time there
had been no indication of a removal of the tax for other heirs (SOU 2003:3).
As late as in June 2004, The Property Tax Committee had presented its final
report Reform of inheritance and gift taxes (SOU 2004:66). This report did
not propose a complete removal of the tax, but rather a series of adjustments
to the existing rules.50 However, none of these were considered appropriate
to implement at the time.
Unfortunately, there has been no systematic research undertaken on what
factors contributed to the repeal of the inheritance tax (Du Rietz et al., 2012).
According to Silfverberg (2005), the Government’s “radical” decision to
abolish the inheritance tax was probably a consequence of The Property Tax
Committee’s inability to review all rules in the Inheritance and Gift Tax
Ordinance and work out a new modern legislation in time for the Budget.
That the decision fell on the inheritance tax and not on the wealth tax, which
had also been heavily debated and evaluated by The Property Tax Committee, was, according to Lodin (2009), a result of a horse trade between the
Social Democrats and the Left Party.51
After the announcement of the repeal, things happened very rapidly. The
Ministry of Finance worked out a memorandum bill, the Tax Agency and the
Appeal Court in Stockholm gave their comments, and on December 16, only
three months after the initial announcement, the bill was passed in the Parliament. The Council of Legislation was critical of the quick manner in
which the reform had been enacted and, in particular, of the limited preparation work that had preceded the bill. According to Silfverberg (2005), the
49
The main motivation was that it would be impossible to tackle the criticism of the tax with
other legislative changes. It was also emphasized that the inheritance tax generated low revenues relative to its costly administration.
50
The report had been preceded by several governmental investigations of the Swedish tax
system; none of which had proposed a complete abolition of the inheritance tax, but rather
reductions of the tax rates and reforms of the valuation rules (see for example SOU 2002:52).
51
According to Lodin (2009), Prime Minister Göran Person invited the Left Party leader Lars
Ohly to a private discussion, during which he demanded that Ohly agree on removing the
inheritance tax and the wealth tax. Ohly refused to abolish both taxes, but after Person issued
an ultimatum—one of the taxes would in any case be removed—Ohly agreed to remove the
inheritance tax.
150
swiftness of the legislative process was a contributing factor as to why the
bill caused almost no political debate.52
The Parliamentary decision on December 16 was that the Inheritance and
Gift Tax Ordinance would expire at the end of 2004. However, of concern of
the bereaved relatives of the many Swedes who died in the Asian Tsunami
on December 26, the Parliament passed a law in April 2005 on inheritance
tax exemption for the period December 17–31, 2004, implying that the tax
was affectively abolished on December 17.
A direct consequence of the repeal of the Inheritance and Gift Tax Ordinance is that inheritances from decedents who die after December 17, 2004
are exempted from taxation. Tax exemption also applies to inheritances
which are received after December 17, but originates from a previously deceased parent who died prior to the reform (so-called postponed inheritances). However, if the tax liability occurred prior to December 17, the old
law applies.
4 Data
In this section, the dataset is presented.53 In the first subsection, I describe the
construction of the working sample. I also describe how I separate between
individuals who were affected and unaffected by the tax reform and particularly, how I approximate the heir’s tax status using data on the deceased
parent’s net worth. The last subsection details the health outcomes used in
the empirical analysis. These include: hospitalization, and the resulting diagnoses, insured sick leave and mortality.
4.1 The sample and approximation of tax status
Information on individuals who received inheritances before and after the
repeal of the inheritance tax is collected from the Belinda database. This
database consists of estate report data covering information on the entire
population of heirs and beneficiaries of deceased Swedes over the period
2003-2005. The database contains around 1,120,000 individuals, but for the
empirical analysis I restrict my attention to heirs who have received inheri52
The limited debate which occurred focused mainly on the proposed date of repeal. The
opposition parties argued that the tax should be abolished retroactively from 20 September
2004, i.e. from the day when the government announced the proposal in the Budget, as it
would otherwise lead to an “inhuman situation” for heirs of decedents who would die in the
last quarter of 2004. In its response, the Government argued that this would result in an unfair
outcome because many (irreversible) cedes had already been made.
53
Access to the data has been granted to the researchers at the Department of Economics at
Uppsala University associated with project Intergenerationella överföringar: orsaker och
konsekvenser. Due to its sensitive and confidential nature, the data cannot be exported from
the closed server environment at Statistics Sweden.
151
tance before (136,920) and after (76,992) the tax reform from parents who
were widowed, divorced, or unmarried when they died and whose deaths
resulted in an estate inventory report.54 These sample restrictions more or
less follow from the succession scheme default rules and yield a sample
which is representative of the population of heirs in Sweden who receive
parental bequests.
The main focus of the empirical analyses is on the heirs who were affected by the tax repeal, or, put differently, those with inheritances large
enough to have rendered liability to pay the inheritance tax had the tax remained in effect. Unfortunately, the Belinda database only contains information on economic variables like the value of estate and the inherited amount
for heirs who inherited before the tax reform, implying that I cannot directly
observe which heirs who received inheritance exceeding the tax threshold
after the reform. My solution to this problem is to approximate the heir’s
inheritance using data on the deceased parent’s net worth from the Swedish
Wealth Register. A novel feature of the wealth register is that the valuation
principles are similar to those that apply to estates, i.e. assets and debts are
valued at market values. This implies that heirs, for whom the product of the
parent’s net worth times the inheritance share exceeds the tax threshold,
could be categorized as affected by the reform.55
I measure net worth three years before death for decedents who died both
before as well as after the reform. This is to avoid that differential incentives
for tax planning (or evasion) has resulted in systematic differences in characteristics between heirs inheriting before and after the reform.56 To account
for the possibility that economic conditions have affected the net worth for
decedents dying on each side of the reform date differently, I adjust it with
the annual official long-term central government borrowing rate.57 Moreover,
because the inheritance law stipulates that heirs can never be forced to pay
the debts of estates in deficit, negative net worth is replaced with the value
zero.
54
Swedish citizens not residing in Sweden and with no assets in Sweden are exempted from
this rule. Exemption from the rule is also given to the deceased’s whose assets are only sufficient to cover funeral expenses and do not comprise real estate. In the latter case, a so-called
estate notification should be established.
55
Given that the sample is restricted to offspring, the inheritance share is calculated as one
divided by the number of offspring appearing in the estate report, information which is available both before and after the reform.
56
Recent studies show that people engage in estate tax planning (or evasion), both during life
and shortly before death, and that this behavior tends to be positively correlated with wealth
(Joulfaian, 2004; Nordblom and Ohlsson, 2006; Kopczuk, 2007; Eliason and Ohlsson, 2013).
57
The estate three years before death is calculated as Estatet-3=Net wortht-3*(1+it-2)*(1+it1)*(1+ it), were i is the yearly official long-term central government borrowing rate and t
denotes the year of death. The i:s during the considered years were: 5.34 percent (2000); 4.98
percent (2001); 5.15 percent (2002); 4.39 percent (2003); 4.30 percent (2004); 3.24 percent
(2005).
152
For each heir, I calculate the (gross) inheritances, referred to as imputed
inheritance, as well as the corresponding tax payment (imputed tax payment)
using the tax rates that applied before the reform, see Table 1. For deceased
widows/widowers, the net worth may in some instances contain the inheritance of the previously deceased spouse, implying that the heirs of widowed
decedents effectively receive two inheritances. To account for the fact both
inheritances were subject to the deductible exemption I divide the net worth
of widow/widowers into two equally sized parts, which I then distribute
evenly between their children. This is in accordance with the schematic distribution applied by the Tax Agency. I then subtract SEK 70,000 from each
of the two inheritances received by the heir before calculating the total tax
payment.58
To test how well the imputed tax payment corresponds to actual tax payment, I calculate the correlation between the two measures for heirs inheriting before the repeal of the tax. (i.e. in 2003 and 2004). The raw correlation
is 0.842 (p<0.01), suggesting that the imputed measure is a valid proxy for
actual tax payment. I have data on inheritances for a representative sample of
three percent of heirs of decedents who died in 2005. The correlation between the two tax measures in this sample is almost identical to that for heirs
inheriting before the tax repeal (0.837, p<0.01). Moreover, the share of heirs
with positive tax payments is very similar across the years. In sum, these
calculations suggest that the imputed measure is valid both within and across
the inheritance cohorts and that it can effectively be used to decide the heirs’
tax status.
In total, 79,777 heirs received inheritances above the tax threshold. They
are the main focus of the empirical analysis, hereafter referred to as Main
sample. Heirs who received inheritance below the tax threshold (133,920),
however, are not omitted completely from the analysis. They are used in
placebo experiments and in the estimation of wealth effects on mortality,
hereafter referred to as Placebo sample.
4.2 Health outcomes59
The health outcomes in this paper are collected from three administrative
registers: the Swedish National Patient Register, which contains detailed
58
Because the distribution depends on the deceased’s marital status, I restrict the sample to
heirs whose decedents had the same marital status (i.e. widow, unmarried, or divorced) three
years before death and at death.
59
Relevant demographic and socioeconomic variables like year of birth, sex, nationality,
marital status, and education, are collected from the Birth Register and the LISA database,
whereas data on incomes and wealth are gathered from population registers provided by the
Tax Agency. The tax agency collects the information directly from relevant sources, such as
personal tax files for incomes, and financial institutions and intermediaries for wealth. The
variables are available for each year over the period 1999–2009 (except wealth which is
available up to 2007).
153
data on all hospital admissions (inpatient care), including data on diagnoses,
concerning Swedish citizens, the Integrated Database for Labour Market
Research (LISA), which contains information on sick spells covered by the
national sickness insurance60 exceeding fourteen days, and the Cause of
Death Register, which contains data on the date and cause of death for all
Swedes who die. Below, I describe the health outcomes which are obtained
from these data sources.
•
Hospitalization is an indicator variable which takes value one if the
individual has been hospitalized, for any cause, at least once during
the year, and otherwise zero. The variable is available for each year
over the period 1993–2011, for all individuals. It should be noticed
that Hospitalization captures health conditions severe enough to require the medical and technical expertise of hospitals.61
•
Diagnose is represented by a set of indicator variables representing
each of the 21 chapters in the WHO’s International Statistical Classification of Diseases and Related Health Problems (ICD), see Table
A1, Appendix A. More specifically, the indicator variables take
value one if the individual, in the given year, has been hospitalized
for any diagnosis appearing in the specific chapter, and otherwise
zero.62 The reason for using this categorization is twofold. First,
there is not enough variation to provide reliable estimates with respect to specific diagnoses. Second, it solves the problem of tractability of diagnoses before and after the reform of the ICD system in
1997, which replaced the previous ICD-9 system with the new ICD10. The diagnose variables are available for each year over the period 1993–2011, for all individuals, and are used to investigate the
reasons for the hospital admissions. The focus is on the ten variables
with the highest pre-inheritance period incidences, see Table 2 (and
variables in bold in Table A1). The remaining variables are grouped
into one variable called Others.
60
See Larsson (2002) and Hesselius et al. (2008) for informative reviews of the Swedish
sickness insurance.
61
Treatment of less severe conditions, medical check-ups and other forms of preventive care
is a matter for the primary (outpatient) care. Since 2001, The Swedish Board of Health and
Welfare keeps a register on outpatient care admissions. Unfortunately, these data are still of
low quality and not recommended to be used for research purposes.
62
The physician is required to report the diagnosis (mapped into ICD code) for the disease or
symptom that the patient was treated for.
154
•
Sick leave is a indicator variable which takes value one if the individual has received sickness benefits for more than two weeks during the year, and otherwise zero. Sick leave could be considered an
objectives measure of health since, in order to receive sickness benefits, the individual has to send in a doctor’s certificate to the Swedish
Social Insurance Agency verifying that the reduced working capacity is due to illness. The variable is available for each year over the
period 1993–2009 for the working aged population (16–65) and
functions as a complement to Hospitalization as it also captures minor health conditions, which are not severe enough to result in hospital admissions. A regression of Hospitalization on Sick leave
yields a coefficient estimate of 0.51 (p<0.001) implying that the outcomes are partly correlated. This is in accordance with previous
studies reporting that medically certified sick leave is a good predictor of clinically defined ill health (Marmot et la., 1995; Kivimäki et
al., 2003).
•
Mortality is represented by six indicator variables (Mortality1,…,
Mortality6) which take the value one if the individual dies from any
cause, within one up to, within six years after the inheritance, respectively and otherwise zero. The variables are available for all individuals. Mortality, similarly to Sick leave, functions as a complement to Hospitalization, but it captures the most severe state of ill
health, namely death.
I have standardized Hospitalization, Diagnose, and Sick leave so that they
are measured for the same number of years before (ten) and after (Hospitalization, Diagnose: six, Sick leave: four) the inheritance receipt for heirs inheriting in 2003, 2004 and 2005. Table 2 reports the annual incidences of the
variables for the pre-inheritance years, as well as the share of heirs who die
in any year over the six years following the inheritance (Mortality6).
To establish that the empirical estimates in this paper are not artifacts of
the current dataset I estimate the cross-sectional relationship between wealth
and health prior to the inheritance. The results, which are reported in Appendix B, show that the there is a statistically significant wealth gradient in
Hospitalization as well as in Sick leave, implying that wealth is protective
against ill health. This holds true both for the Main sample and the Placebo
sample.
155
Table 2: Health outcomes, incidences, in percent.
Health outcome
Hospitalization
Incidence
a
6.65
a
Diagnose :
Neoplasms
0.55
Mental
0.57
Nervous
0.26
Circulatory
0.77
Respiratory
0.31
Digestive
0.78
Musculoskeletal
0.52
Genitourinary
0.53
Symptoms
0.84
Injury
0.73
Others
0.79
Sick leave
a1
13.3
Mortality6
3.51
a
Notes. Incidence calculated as annual average over the
ten years before the inheritance. 1The incidence is calculated for the working-age population (16-65).
5 Empirical strategies
In this section, I present the empirical strategies to identify the causal effect
of the wealth shock on the health outcomes discussed in the previous section.
A direct consequence of the repeal of the Inheritance and Gift Tax Ordinance is that offspring who received inheritances, amounting to more than
the basic deductible exemption, from parents who died after December 17,
2004 experienced beneficial shocks to their inheritances equal in size to what
their tax payments would have been had the parents died before that date.
The core of the empirical strategy is to estimate the causal effect of this
wealth shock on health by approximating the counterfactual outcome (i.e.
health in the absence of the wealth shock) with the health experiences of
heirs who received inheritance above the tax threshold from parents who
died before the reform date.
Due to the fact that it is essentially a random process determining whether
an individual dies today or tomorrow, the ideal would be to compare the
health of individuals whose parents died in the days surrounding the reform.
This approach would be similar in spirit to a regression discontinuity design
framework, where the forcing variable would be the parent’s date of death.
However, because only about 300 individuals die in Sweden each day, and
156
even fewer with taxable estates, I would end up with a sample too small to
provide enough power for statistical analysis in the close vicinity of the reform.
To have any hope in precisely detecting differences in health between the
two groups, I define heirs receiving inheritances above the tax threshold
(Main Sample) after December 17, 2004 and in 2005 as being treated, and
heirs receiving inheritances above the tax threshold in 2004, before December 17, and in 2003 as controls. Heirs receiving inheritances below the tax
threshold (Placebo sample) over these periods are referred to as “treated”
and “controls”.
Table 3: Sample means with respect to inheritances, wealth shocks and Hospitalization (by time
period), for Main sample and Placebo sample
Hospitalization, by period:3
Inheritance1
Wealth shock2
Pre
Post
Post-Pre
N
Main sample
Treated
548,189
70,817
6.6
8.7
2.2
28,827
Controls
565,417
0
6.7
8.6
2.0
50,950
Placebo sample
”Treated”
32,923
0
7.6
10.1
2.4
48,165
”Controls”
34,671
0
7.8
10.1
2.3
85,967
Notes. Dummy variables are reported in percent. 1Refers to imputed inheritance, see Section 4.
2
Approximated by imputed tax payment, see Section 4. 3The means have been calculated as
yearly average over the given period.
Table 3 illustrates the variation in inherited wealth generated by the repeal
of the inheritance tax by reporting descriptive statistics on inheritances and
the corresponding wealth shocks for the treated (“treated”) and the controls
(“controls”). The upper panel displays the statistics for the Main sample
whereas the bottom panel displays the statistics for the Placebo sample.
It can be noted that the difference in inheritance between the treated and
the controls is small. This is reassuring, as it suggests that the wealth shock
is exogenous.63 A similar finding is noted for the Placebo sample. Regarding
the wealth shock (approximated by the imputed tax payment, see Section 4)
it is, by definition, zero for the controls and positive for the treated subjects
in the Main sample and zero for both groups in the Placebo sample. The
mean of the shock for treated subjects in the Main sample is SEK 70,817.64
For health outcomes which are observable over time, before and after the
inheritance receipt (i.e. Hospitalization, Diagnose, and Sick leave), I will
estimate the effect of the wealth shock by comparing the difference in inci63
In Section 7, I confirm this further by showing that the treated and the controls are balanced
in predetermined characteristics, including health.
64
See Table C1 in Appendix C for the sample distribution of the wealth shock
157
dences before and after the inheritance for the treated subjects with the similar difference for the controls. The last three columns in Table 3 report descriptive statistics necessary to calculate these difference-in-differences
(DID) with respect to Hospitalization (i.e. the incidences in the pre- and
post-periods, as well as the change in incidence over time (Post-Pre) for each
group). It can be noticed that the pre-period incidences are similar across
treated and controls. This indicates that the counterfactual identifying assumption of parallel trends in the absence of the shock is satisfied.65 A comparison of the change in Hospitalization (Post-Pre) between the treated and
the controls suggests that the wealth shock has a positive, but small, impact
on the incidence, around 0.2 percentage points. The question is, however,
whether or not we could interpret this impact as a causal effect?
To place this issue in perspective, one can compare the change in Hospitalization over time across the “treated” and the “controls” in the Placebo
sample. In contrast with what we should expect to see given that both these
groups were unaffected by the tax reform, the implied DID is positive and
indicates that the reform leads to a 0.1 percentage points increase in the outcome.
One possible explanation for this finding is that the DID:s obtained from
Table 3 only accounts for biases from common trends in the outcome, such a
health responses surrounding the death of the parent or an increasing trend in
health over time, and not for the fact that the time periods over which the
differences are calculated correspond to different calendar years for heirs
inheriting before and after the tax reform. This may be an issue, due to the
fact that recent studies show that health tends to respond to temporary fluctuations in the economy (Ruhm, 2000; Adda et al., 2009; Gerdtham and Johannesson, 2005). The impact and severity of aggregate seasonal health
shocks, such as the flu or the winter vomiting disease, may also differ between years. Although the influence of year-specific events is partly mitigated by using the average incidences for the pre- and post-periods, one may
still be concerned by the possibility that the response in the outcome is the
result of an adverse event taking place in the years surrounding the reform or
events in a year in the beginning or in the end of the sample period, rather
than the wealth shock. For instance, if something adversely impacts the
health of the treatment group in the last (calendar) year of the sample period,
we may wrongly conclude that a difference in health across the two groups is
the consequence of the wealth shock. Likewise, an adverse event in 2004
would be picked up as a pre-period effect for the treatment group and as a
post-period effect for the controls, implying that we may overestimate (underestimate) a positive (negative) effect of the wealth shock.
65
In Section 7, I present graphical evidence showing that the trajectories of Hospitalization
for the treated and the controls evolve similarly in the pre-inheritance period.
158
My strategy to account for this source of bias is to estimate panel data
models with cohort, time and year effects of the following form:
(1)
1
, , ,
2005:
0
, , ,
,
where , , , is outcome of individual i, of inheritance cohort j (j=2003,
2004, 2005) at time t, in year z.66 ,
and
are cohort, time and year
is an indicator variable which takes the value
fixed effects, respectively.
one (=1) from the year of the inheritance (t=0) and onwards for individuals
whose parents died after the tax reform (j=2005), and zero (=0) in all years
for individuals whose parents died in the years before the reform (j=2003,
2004), and , , , is an idiosyncratic error. The coefficient is the DID estimator which captures the average effect of the wealth shock over the years
following the inheritance.
The fact that the heir has to be alive at the time of the inheritance to be included in the sample means that Model 1 cannot be employed to estimate the
effect of the wealth shock on Mortality. Instead, I estimate the wealth effect
by comparing the difference in the likelihood of mortality between treated
and controls in the Main sample with the similar difference for heirs in the
Placebo sample. This alternative difference-in-differences strategy will account for biases from time-invariant differences between the treated and the
controls under the assumption that environmental conditions (i.e. aggregate
health shocks) during life, before the inheritance, have similar impacts on
mortality rates for offspring receiving inheritance above and below the tax
threshold.67 Likewise, it will account for differential annual trends in mortality under the assumption that external exposures over the period after the
inheritance have similar influences on mortality for heirs receiving inheritance above and below the tax threshold.
6 Exogeneity of the wealth shock and test of identifying
assumptions
In this section, I present two informal tests of the identifying assumptions
underlying the empirical strategy. The first test, which looks for differences
in predetermined characteristics between the treated and the controls, suggests that the wealth shock is exogenous. The second test compares the dy66
Here, cohort j=2005 includes the offspring who inherit over the period December 17-30,
2004.
67
This difference-in-difference strategy is similar in spirit to that used by Snyder and Evans
(2006), who estimate the effect of income on mortality by comparing mortality rates for men
born in the first quarter of 1917 (the Notch generation) with mortality rates for men born in
the fourth quarter of 1916, using women from the same two birth quarters as controls.
159
namics of Hospitalization over the sample period between the treated and
controls. Reassuringly, the trajectories evolve similarly in the pre-inheritance
period, suggesting that the parallel trends assumption is satisfied. Taken
together, the tests imply that any difference in health following the inheritance could reasonably be attributed to the wealth shock.
6.1 Test for differences in pre-determined characteristics
between treated and controls
Table 4 compares the sample means across the treated and the controls along
a number of different predetermined demographic and economic characteristics which are likely to be related with health. The first two columns report
the sample means for the treated and the controls, respectively, and the last
column (3) reports the p-values from t-tests of the difference in means between the groups.
As indicated in Section 5, the treated and the controls, by construction,
inherit in different calendar years (2005 vs. 2003 and 2004). A direct consequence of this sample design is that the treatment group contains heirs of
younger birth-cohorts than the control group, as indicated by difference in
mean birth year between the two groups. What consequences do this have
for other observable characteristics? It can be seen from Column 3 that there
are no statistically significant differences (p>0.10) in the fraction women,
fraction Swedish citizen, fraction with children in the household, fraction
with lower secondary education, earned income, or net worth across the
treated and the controls.68 The differences in observed characteristics that do
exist are in age, fraction married, fraction with primary education, and fraction with upper secondary or post graduate education. Although these differences are statistically significant (p<0.10) they are quantitatively small and
can easily be explained by the disparity in birth-year between the two
groups. It is generally acknowledged that younger cohorts tend to have
higher education, be married to a lower degree, and receive inheritance later
in life, than older cohorts. The econometrical model presented in Section 5
includes inheritance-cohort fixed effects, which should account for any unobserved heterogeneity related with birth-cohort across the groups.
In Table D1, Appendix D, I present similar descriptive statistics for the
Placebo sample. The differences in sample means between heirs inheriting
before and after the reform are comparable to the corresponding differences
for the Main sample, again suggesting that unobservable (inheritance) cohort
68
The means with respect to Earned income and Net worth have been calculated on the annual averages for the available pre-inheritance years to limit the influence of differential
macroeconomic exposures. Moreover, Earned income is adjusted for nominal wage growth,
in the Government sector (base year 2004) and Net worth is adjusted for inflation using CPI
(base year 2004).
160
specific factors have not manifested into persistent differences in observable
characteristics related with health.
Table 4: Comparison of sample means, predetermined demographic and socioeconomic characteristics, treated and controls, Main sample.
Treated
Controls
p-value 1-2
1
2
3
Birth-year
1951.4
1950.1
0.000
Age when inheriting
53.5
53.4
0.054
Woman
49.3
49.8
0.246
Swedish citizen
99.6
99.6
0.575
Married
55.9
57.3
0.000
Children in household1
38.3
38.6
0.346
Level of education2:
Primary
18.1
19.1
0.001
Lower secondary
42.6
42.6
0.939
Upper secondary or post
35.6
34.9
0.031
graduate
Earned income3
274,891
274,062
0.577
Net worth4
905,871
899,235
0.884
Number of obs.
28,827
50,950
Notes. Characteristics other than Birth-year, Age, Earned income and Net worth
are measured three years before the inheritance receipt. Indicator variables are
reported in percent. 1Refers to children younger than 18. 2Highest achieved level
of education. 3The means are calculated on annual incomes (adjusted for the
growth in nominal income, base year 2004) averaged over the available preinheritance years.4The means are calculated on annual net worth (adjusted to 2004
price level using CPI) averaged over the available pre-inheritance years.
6.2 Test for parallel trends in health
Figure 1 displays the dynamics of Hospitalization over the sample period for
the treated and the controls. Regarding the controls, I have separated between the heirs with respect to year of inheritance (2003 and 2004). The
reason behind this division, rather than representing the dynamics for the
controls with only one trajectory, is that it conforms better to Model 1, which
includes controls for inheritance-cohort. It should, however, be emphasized
that the graphs display the unconditional sample means by time period and
hence, do not account for the fact that the periods corresponds to different
calendar years for the treated and the controls.
The general pattern indicates that the incidence of Hospitalization is
rather stable in the beginning of the sample period, increases sharply around
two years before the inheritance and continues to do so thereafter. The increasing trend is expected given that the heirs become older. The sharp rise
surrounding the parent’s death (vertical line) may reflect increased illness
related to mourning and psychological distress (Scharlach, 1991; Umberson
and Chen, 1994; Kessler, 1997; Marks et al., 2007; Rostila and Saarela,
2001). Regarding the trajectories of the treated subjects and the two control
cohorts, these display similar trends in the pre-inheritance period, suggesting
that cohort specific influences have not manifested in persistent differences
161
in health. The small differences in incidence between the groups that do exist
could partly be explained by the fact that the years reported on the horizontal
axis correspond to different calendar years for the groups, and these will be
accounted for by the year controls in Model 1. The results nevertheless suggest that the parallel trends assumption is satisfied. It is, however, difficult to
get an indication of whether or not the wealth shock has any effect on Hospitalization by comparing the trajectories over the post-inheritance years. If
anything, the trajectory of the treated subjects appears to increase somewhat
more sharply than those of the two control cohorts, but, as previously noted,
one should be careful when interpreting this as causal effect since differential year trends are unaccounted for.
In Figure D1 in Appendix D, I report similar graphs for the “treated” and
the “controls” in the Placebo sample. The parallel trends assumption appears
to be satisfied for this sample as well. Moreover, a comparison of Figure 1
with Figure D1 suggests that the heirs receiving inheritances above and below the tax threshold experience similar health dynamics, although the incidences differ somewhat in levels. To the extent that trends in mortality are
similar to trends in Hospitalization, this finding could be seen as supporting
the identifying assumption underlying the estimation of wealth effects on
mortality.
Figure 1: The annual incidence of Hospitalization for treated and controls, Main sample.
Note. The vertical line indicates the point in time when inheritance is received. Controls 2004 does
not include offspring receiving inheritance from a parent over the period December 17-31.
162
7 Results
In this section, the empirical results are presented. The first sub-section details the results with respect to the effect of the wealth shock on Hospitalization. It shows that the wealth shock increases the likelihood of hospitalization for any cause by five percent. The causal interpretation of the estimate is
confirmed using a placebo test. The effect is more pronounced for women,
the elderly and individuals with low education. In Section 7.1.1, I show that
a non-trivial share of the effect in Hospitalization could be attributed to
higher incidences of signs and symptoms of disease and cancer. Section 7.2
reports that the wealth shock does not have any effect on Sick leave or on
Mortality.
7.1 The effect of the wealth shock on Hospitalization
The DID estimates in this section have been obtained from versions of
Model 1 estimated using OLS. Given that Hospitalization is binary, the estimates should be interpreted as the percentage point difference in the probability of the outcome between the treated and the controls. In connection to
the regression estimates, I report the mean of the dependent variable, in percent, for the post-inheritance period for the relevant control group (in brackets). Dividing the DID estimate by this statistic gives the percentage difference in incidence between the treated and the controls. In each specification,
the standard errors have been clustered at the individual level to account for
correlation within the individual over time.
Column 1 in Table 5 reports the DID estimate obtained from Model 1
without year controls. This is comparable to the naïve DID estimate implied
by the statistics in Table 3. As expected, the estimate implies that the treated
subjects have 0.2 percentage point higher probability of being hospitalized in
the pre-inheritance period relative to the controls. Column 2 reports the DID
estimate from Model 1 with year controls. The estimate, similarly to the
estimate in Column1, is positive and statistically significant (p<0.05) but,
notably, almost twice as large. The discrepancy suggests that the treated and
the controls experience differential year trends and that year controls indeed
are essential. Regarding the size of the effect, it suggests that the wealth
shock leads to a five percent increase in the probability of Hospitalization.69
69
Since it was decided retroactively that inheritances received during the period December
17-31, 2004 would be exempted from taxation, it may be a source of concern that the anticipatory effects of heirs inheriting during this period are different to those of other heirs inheriting after the reform. Reassuringly, however, the DID estimate is unchanged when I estimate
the model on a sample without these individuals. Moreover, recent studies have documented
that people may postpone their death to save taxes, see Eliason and Ohlsson (2013) and
Kopczuk and Slemrod (2005). Regarding the current reform, Eliason and Ohlsson show that
deceased with taxable estates were more likely to have passed away on January 1, 2005, from
when the tax was (supposed to be) repealed, rather than on December 31, 2004, compared to
163
Table 5: Difference-in-differences (DID) estimates, impact of wealth shock on Hospitalization (in percent), Main sample and Placebo sample.
Main sample
DID estimate :
Year FE
N
N*T
Placebo sample
1
2
3
4
0. 222**
0.432**
0.201**
-0.144
(0. 111)
(0.218)
(0.091)
(0.179)
[8.63]
[8.63]
[10.09]
[10.09]
No
Yes
No
Yes
79,802
79,802
134,172
134,172
1,356,634
1,356,634
2,280,924
2,280,924
Notes. Coefficient estimates are reported in percent. Standard errors (in percent) clustered
at individual, in parentheses. Mean of dependent variable (in percent), post-inheritance
period for control group, in brackets. * significant at the 10 percent level, ** significant at
the 5 percent level, *** significant at the 1 percent level.
Is this a large or a small effect? To gain perspective on this issue, I compute the cross-sectional relationship between age (in years) and Hospitalization, as it is well-known that age has a large impact on health. It turns out
that the effect of the wealth shock equals the impact of being about four
years older, suggesting that the wealth effect is non-trivial. However, when I
relate the wealth effect to the impact of education, another factor related with
health status (Lleras-Muney, 2005; Cutler and Lleras-Muney, 2010), I find
that having primary or lower secondary education, as compared to upper
secondary or post graduate education (i.e. the impact of having lower education) increases the probability of Hospitalization by 18 percent. This would
suggest that the effect of a seven percent increase in wealth should be considered relatively limited.70
To establish that the estimate in Column 2 represents a causal relationship, and not just a spurious correlation, I estimate Model 1 on the Placebo
sample (see columns 3 and 4 for results). An insignificant response in the
outcome, or at least a DID estimate which is smaller in magnitude than the
corresponding estimate for the Main sample, should be considered a validation of the casual interpretation of the main estimate. In accordance with
what the statistics in Table 3 suggested, the DID estimate from Model 1
without year controls is similar to the corresponding estimate for the Main
sample. However, in contrast to the corresponding main estimate, the estideceased without taxable estates. To account for the possibility that individuals whose parents
died during the days surrounding the reform are systematically different from other heirs, I
have redone the main analyses omitting heirs of parents who died over a period of up to two
weeks following the reform. Reassuringly, the results from this exercise are similar to the
main results.
70
One explanation for this is that education has been obtained early in life, and hence that its
effect has had more time to accumulate into health.
164
mate from the model with year controls (Column 4) is negative and statistically insignificant. This finding could be seen as lending additional support
for the full version of Model 1. Taking the difference between the estimates
in columns 2 and 4, as in a triple-difference estimator, suggests that, if anything, the wealth effect is underestimated.
I continue by testing for how the wealth effect varies with demographic
characteristics. The first dimension I consider is age. The results, displayed
in Table E1 in Appendix E, show that the effect is markedly higher for old
heirs (above mean age) than for young heirs (below mean age). This finding
corresponds with previous studies (e.g. Lindahl 2005). I also test for responses for the working age population (16-65) over a period of four years
following the inheritance. This is to obtain results which are comparable to
those with respect to Sick leave. The DID estimate from this exercise (Column 3) is of the same order of magnitude as that for young heirs, and here
too, statistically insignificant. Moreover, I find that the effect is primarily
driven by women and not by men (see columns 3 and 4 in Table E1).71 The
results in columns 5 and 6 in Table E1 show that the DID estimate is positive
and statistically significant for heirs with low education (primary or lower
secondary education) and negative and statistically insignificant for heirs
with high education (upper secondary or post graduate education). This
could possibly suggest that highly educated individuals have more knowledge about, and are better at avoiding and managing, harmful health effects
than their peers with lower levels of education (Goldman and Smith 2002).
Models of health production, as well as previous empirical results, suggest
that the wealth effect should be increasing relative to the size of the shock.
To explore this in more detail, I estimate Model 1 separately for heirs receiving inheritance within the first, second, third and fourth quartile of the sample distribution. The results from this exercise are reported in Table E2 in
Appendix E and suggest that the effect is quantitatively similar across the
subsamples. The coefficient estimates, however, are imprecisely measured,
which is probably a consequence of small sample sizes.
The results in Table 5 give us no sense of the dynamics of the wealth effect - whether it accelerates or stabilizes. To explore these dynamics, I estimate Model 1 with leads and lags of treatment. More specifically, I include
interactions between the treatment indicator and time dummies for each of
the ten years before the inheritance, the year of the receipt and for each of
the six subsequent years. The results, which are reported in Column 1, Table
E3 in Appendix E, show that the coefficient estimates on the lead indicators
are statistically insignificant. This is comforting, as it suggests that the parallel trends assumption indeed is satisfied. As for the pattern of the lag struc71
My result shows that women have a higher probability, relative to men, of being hospitalized in the pre- and post-inheritance periods. This is consistent with previous research on
gender differences in health (see for example Case and Paxson, 2005).
165
ture, it shows that the difference in probability of Hospitalization between
the treated and the controls increases sharply at the time of the inheritance
receipt. This should be taken as additional support for the causal interpretation of the wealth effect. It should be noted, however, that the implied effect
varies across the years and that it is only statistically significant for the second and fifth years after the inheritance. The lead and lag estimates obtained
from the Placebo sample (Column 2) are quantitatively similar (implying
that the tax reform has no impact), do in general have the opposite sign to
those for the Main sample and are, except for one lead indicator, statistically
insignificant.
Taken together, it is evident that the wealth shock causes an increase in
Hospitalization. At a first blush this finding suggests that increased wealth
has detrimental effects on health. It should, however, be remembered that
Hospitalization does not inform us about the reasons for the hospital admission. To place this issue in perspective, I therefore continue and test for heterogeneous response across the diagnoses reported in connection with the
hospital admissions.
7.1.1 Explaining the wealth effect on Hospitalization
In this section, I report regression results for the effect of the wealth shock
on the diagnose indicators detailed in Section 4.
Column 1 in Table 6 displays the DID estimates obtained from Model 1
(with year effects) estimated on the Main sample. It is noticeable that there
are only two outcomes for which the DID estimate is statistically significant:
Neoplasms and Symptoms and signs. The estimate with respect to Neoplasms
implies that the wealth shock causes a 12 percent increase in the probability
of the outcome, whereas for Symptoms and signs, the coefficient implies an
increase of 11 percent. Taken together, they explain around 60 percent of the
effect in Hospitalization. The fact that there is no significant response in any
other variable (neither in the single diagnose variables nor in the variable
Others) suggests that the wealth effect on Hospitalization is operating solely
through Symptoms and signs and Neoplasms. Moreover, the corresponding
estimates for the Placebo sample (see Column 2) are statistically insignificant, suggesting that the main estimates are causal.
What do the responses in these variables tell us about the mechanisms
through which wealth affects Hospitalization?
The variable Symptoms and signs, as the name indicates, captures symptoms and signs of disease (e.g. irregular heart rate, shortness of breath, fever,
senility, general feeling of illness, etc.) as well as unusual findings during
medical examinations (e.g. blood and urine samples).72 Given that the condition has resulted in a hospital admission, the response in the variable may, on
72
The results are similar when I exclude diagnoses due to abnormal clinical and laboratory
findings (ICD-10: R70-R99 and ICD-9: 790-799).
166
the one hand, imply that the wealth shock leads to worse health, and potentially more so had we investigated the effects over a longer period of time.73
On the other hand, the response could be interpreted as if the shock has made
people more prone to seek care for health irregularities, possibly to reduce
the likelihood of more severe conditions in the future. This is in line with
previous studies which document that economic circumstances are positively
associated with disease prevention (see for example Cawley and Ruhm,
2012)
Regarding Neoplasm, it contains diagnoses of cancers at different stages
of development (i.e. benign, potentially malignant, and malignant tumors).74
It is difficult to give an analytical explanation for why the wealth shock
causes an increase in the likelihood of cancer, especially since it is commonly considered an equal opportunity disease (Smith, 2004). Although
lifestyle factors such as smoking and drinking, which are reported to be positively related with improved wealth (Apouye and Clark, 2013; Kim and
Ruhm, 2012), are linked to many types of cancers (e.g. lung, head and neck,
pancreatic, liver, colon, gastric, etc., see for example Kushi et al., 2012) it
seems unlikely that an increase in these factors would manifest into higher
cancer incidence within a period of only six years. If the wealth shock has
caused people to smoke and drink more we would rather expect to find responses in diagnoses which are more immediately related to these behaviors,
such as injuries (e.g. alcohol poisoning), mental and behavioral disorders,
diseases in the digestive system (e.g. liver cirrhosis), respiratory diseases
(e.g. chronic obstructive lung disease) and circulatory diseases (e.g. coronary
heart disease and stroke) (WHO 2002). Moreover, previous studies report
that improved wealth leads to reduced obesity (Lindahl, 2005; Kim and
Ruhm, 2012) and improved mental well-being (Gardner and Oswald, 2007;
Apouye and Clark, 2013). But, if the wealth shock exploited here has led to
reduced obesity or improved mental well-being we should expect to find, if
anything, a reduction in cancer incidence, not an increase (Kushi et al.,
2012; Chida and Steptoe, 2008).
A more realistic explanation for the positive response in Neoplasm is
therefore that the wealth shock has led to more health care visits in general,
as indicated by the results with respect to Symptoms and sings, and that cancer, which otherwise would have been diagnosed later, is detected and possibly treated earlier.
In sum, the results in this section suggest that the higher incidence of
Hospitalization does not necessarily mean that the wealth shock has detri73
Minor medical problems generally heighten the odds of experiencing more severe health
problems. This is the so-called progressive nature of disease (Smith, 2005).
74
I have analyzed the effect on the wealth shock on cancerous tumors (malignant) and other
tumors (benign and potentially malignant) separately, but the estimates are imprecisely measured, probably as a result of not enough variation (i.e. not enough non-zero observations) in
the outcomes.
167
mental effects on health, but rather that it leads to more preventative actions
against future morbidity.
Table 6: Difference-in-differences (DID) estimates, impact of
wealth shock on diagnose categories (in percent), Main sample and
Placebo sample.
Main sample
Placebo sample
1
2
Outcome:
Neoplasms
0. 13*
-0.04
(0.08)
(0.06)
[1.13]
[1.26]
Mental
-0.05
-0.07
(0.07)
(0.06)
[0.64]
[0.75]
Nervous
-0. 03
-0.001
(0.05)
(0.04)
[0.40]
[0.47]
Circulatory
0.03
-0.09
(0.09)
(0.08)
[1.55]
[1.94]
Respiratory
0.04
-0.02
(0.05)
(0.04)
[0.48]
[0.62]
Digestive
0.02
-0.001
(0.07)
(0.06)
[0.98]
[1.18]
Musculoskeletal
0.001
-0.08
(0.07)
(0.05)
[0.88]
[1.12]
Genitourinary
0.05
0.02
(0.06)
(0.04)
[0.61]
[0.72]
Symptoms and signs
0.14*
0.08
(0.07)
(0.06)
[1.09]
[1.49]
Injury
0.06
0.03
(0.07)
(0.05)
[0.10]
[1.15]
Others
0.05
0.08
(0.08)
(0.07)
[1.46]
[1.62]
Year FE
Yes
Yes
N
79,802
134,172
N*T
1,356,634
2,280,924
Notes. Coefficient estimates are reported in percent. Standard
errors (in percent) clustered at individual, in parentheses. Mean of
dependent variable (in percent), post-inheritance period for control
group, in brackets. * significant at the 10 percent level.
168
7.2 The effect of the wealth shock on Sick leave and Mortality
In this section, I complement the previous analyses by investigating responses in outcomes capturing health events that are both less and more severe than those resulting in hospital admissions. More specifically, I estimate
the causal effect of the wealth shock on Sick leave (less severe) and Mortality (more severe).
Table F1 in Appendix F reports the DID estimates with respect to Sick
leave. These have been obtained from Model 1, estimated on the working
aged population over a period of ten years before and four years after the
inheritance receipt. A comparison of the estimates from the model with and
without year controls indicates that the treated and the controls experience
differential year trends in the variable. This is in line with what I found for
Hospitalization. Here, however, the DID estimate from the preferred specification of Model 1 (Column 2) is statistically insignificant, implying that the
wealth shock does not have any evident effect on the likelihood of sick
leave. The DID estimates for the Placebo sample (columns 3 and 4) are
similar in terms of sign and statistical significance to the corresponding estimates for the Main sample, but the implied responses are quantitatively
smaller. It should be noticed that I cannot rule out the possibility that the
wealth shock has consequences for health events captured by Sick leave for
heirs who are younger than 16 and older than 65. However, the fact that the
wealth effect with respect to both Sick leave and Hospitalization are statistically insignificant for the working-age population lends additional support to
the conclusion that the wealth shock generated by the tax repeal has no detectable consequences for health.
The causal effect of the wealth shock on mortality is estimated by comparing the difference in the probability of dying over the post-inheritance
period between treated and controls in the Main sample with the similar difference for the Placebo sample.75 The regression results with respect to each
of the six mortality indicators (i.e. Mortality1,..., Mortality6) are presented in
Table F2, Appendix F.76 Neither the differences estimates (for any of the two
samples) nor any of the DID estimates (which accounts for biases from timeinvariant differences and year trends), are statistically significant at conventional levels. These results suggest that the wealth shock has no detectable
effect on mortality within any year over a period of six years after the inheritance receipt. The results in Section 7.1.1 suggested that cancer is detected
75
The analysis is based on heirs inheriting in the year before (2004) and in the year after the
reform (2005). This is to limit the potential influence of confounding secular trends in mortality. I have, however, redone the analysis on samples including offspring inheriting in 2003
and obtained largely similar results.
76
The differences and the difference-in-differences are estimated with linear probability
models. The models include controls for age, age2, gender, marital status, presence of children, level of education, earned income and net worth, measured three years before the inheritance and aimed at accounting for any remaining unobserved heterogeneity.
169
earlier as a consequence of the wealth shock. It is evident however that this
potentially preventative effect is not sufficient to have any effect on all-cause
mortality, at least not over a period of six years. I have also tested explicitly
for the impact of the wealth shock on the likelihood of cancer mortality
within the six year period but the estimate of the wealth effect is imprecisely
measured.
8 Concluding discussion
In this paper I exploit the exogenous variation in wealth induced by the unexpected repeal of the Swedish inheritance tax to test for the impact of increased wealth on health outcomes commonly found in administrative registers.
The empirical analysis shows that the favorable wealth shock resulting
from the tax reform has limited consequences for objective health over a
period of six years following the shock. This is in line with what has been
documented previously regarding subjective health outcomes. If anything, it
appears as if the wealth shock leads to more health care visits for minor
health contingencies, which in turn result in that cancer is detected and possibly treated earlier. One possible explanation for this preventive response is
that people feel that their future consumption prospects have improved and
that good health is necessary for enjoying these benefits. Even if these findings suggest that increased wealth does not have any direct consequences for
health they should be of interest to policy makers, since prevention, and especially early detection of chronic diseases like cancer, has been brought
forward as one of the most valuable instruments to reduce health care costs
(see for example Kenkel, 2000). Preferably, one would like to complement
the analysis with data on outpatient care to say more about the wealth effect
on total number of health care visits and also, to pinpoint when in time cancer is initially discovered. Data on outpatient care of sufficient quality for the
sample period is, unfortunately, not available however.
Although the wealth shock exploited in this paper is received by people
who have suffered the loss of a parent – and therefore may be unhealthier
than the general population – the results generalize to people who are in their
fifties, as the death of a parent commonly occurs at this stage of life. The fact
that I can replicate the stylized facts concerning the cross-sectional relationship between wealth and health also suggests that the empirical results are
not specific for the current sample.
It should be noted, however, that I cannot rule out the possibility that potential effects of the wealth shock take more than six years to manifest into
health. From a policy perspective the results nevertheless seem particularly
relevant, suggesting that wealth changes that might be expected from tax
reforms of similar magnitudes as the repeal of the Swedish inheritance tax,
170
affecting this age-group, are unlikely to have any short or medium run consequences for health. The results, moreover, suggest that policies targeted at
reducing socioeconomic inequalities in health are likely to be more usefully
channeled toward interventions that directly improve health.
171
Appendix A: Description of diagnose variables
172
Fatcors
External
Injury
Symptoms
Congenital
Endocrine
Mental
Nervous
Eye
Ear
Circulatory
Respiratory
Digestive
Skin
Musculoskeletal
Genitourinary
Pregnancy
Perinatal
Blood
Infections
Neoplasms
Variable
XXI. Factors influencing health status and contact with health services
I. Certain infectious and parasitic diseases
II. Neoplasms
III Diseases of the blood and blood-forming organs and certain disorders
involving the immune mechanism
IV. Endocrine, nutritional and metabolic diseases
V. Mental and behavioral disorders
VI. Diseases of the nervous system
VII. Diseases of the eye and adnexa
VIII. Diseases of the ear and mastoid process
IX. Diseases of the circulatory system
X. Diseases of the respiratory system
XI. Diseases of the digestive system
XII. Diseases of the skin and subcutaneous tissue
XIII. Diseases of the musculoskeletal system and connective tissue
XIV. Diseases of the genitourinary system
XV. Pregnancy, childbirth and the puerperium
XVI. Certain conditions originating in the perinatal period
XVII. Congenital malformations, deformations and chromosomal abnormalities
XVIII. Symptoms, signs and abnormal clinical and laboratory findings, not elsewhere classified
XIX. Injury, poisoning and certain other consequences of external
causes
XX. External causes of morbidity
ICD chapter
Table A1: Diagnose variables, corresponding ICD chapters, and ICD codes (by version).
V01-V82
E01-E99
V00-Y99
Z00-Z99
800-999
780-799
740-759
240-279
290-319
320-389
360-379
380-389
390-459
460-519
520-579
680-709
710-739
580-629
630-676
760-779
280-289
001-139
140-239
ICD-9
S00-T88
R00-R99
Q00-Q99
E00-E89
F01-F99
G00-G99
H00-H59
H60-H95
I00-I99
J00-J99
K00-K94
L00-L99
M00-M99
N00-N99
O00-O99
P00-P96
D50-D89
A00-B99
C00-D49
ICD-10
Appendix B: Cross-sectional evidence of the wealthhealth gradient
This appendix shows that the dataset can reproduce the positive crosssectional association between wealth and health documented in the previous
literature.
Table B1 presents estimates from linear probability models with Hospitalization and Sick leave as dependent variables and wealth as explanatory
variable. To account for the fact that the relationship between wealth and
(good) health is documented to be concave (see for example Ettner 1996,
Smith 1999, Benzeval and Judge 2001), I apply the inverse hyperbolic sine
transformation to wealth (Burbidge et al., 1988). This transformation is preferred because, unlike the log transformation, it accommodates zeros values
in the variable. To account for the fact that negative values are not accommodated by the inverse hyperbolic sine however, I use the individual’s gross
wealth (i.e. the sum of real and financial assets, at market prices) instead of
net worth. The coefficient estimate on wealth can nevertheless be interpreted
as a as a semi-elasticity. The models also include controls for: a second order
polynomial in age, sex, marital status, presence of children, level of education, as these have been used in the previous literature, as well as controls for
year of inheritance. The outcomes as well as the covariates are measured
three years before the inheritance receipt to assure that they are exogenous
with respect to the tax reform. Columns 1 and 2 report the result with respect
to Hospitalization and Sick leave for the Main sample, whereas columns 3
and 4 report the corresponding results for the Placebo sample. Regarding the
Main sample the coefficient estimate on wealth is statistically significant at
the one percent level, indicating that higher wealth reduces the likelihood of
hospital admission. Divided by the sample mean, the estimate implies that a
one percent increase in wealth, all else equal, reduces the likelihood of Hospitalization by four percent. Similarly, the coefficient estimate on wealth
from the specification with Sick leave as dependent variable is statistically
significant (p<0.01). The estimate implies that a one percent increase in
wealth reduces the probability of the outcome with two percent. The results
for the Placebo sample display a similar pattern as those for the Main sample: the coefficient estimates on wealth are negative and statistically significant on conventional levels for both outcomes.
174
Table B1: Linear probability estimates (in percent) of the cross-sectional relationship
between wealth and Hospitalization and wealth and Sick leave, Main sample and Placebo
sample.
Main sample
Placebo sample
Outcome:
Hospitalization
Sick leavea
Hospitalization
Sick leavea
1
2
3
4
sine-1 Wealth
-0.31***
-0.40***
-0.33***
-0.32***
(0.02)
(0.03)
(0.01)
(0.02)
Mean of dep. var,
6.93
15.67
8.23
18.36
in percent
N
76,949
69,936
129,921
111,754
Notes. Robust standard errors (in percent), in parentheses. * significant at the 10 percent
level, ** significant at the 5 percent level, *** significant at the 1 percent level. The
specifications include control variables, measured three years before the inheritance.
These are: age, age2, gender, marital status, presence of children, and level of education.
The specifications also include controls for year of inheritance. a The specification has
been estimated on the working aged population (16-65).
Appendix C: Sample distribution of wealth shock,
treated subjects, Main sample
Table C1: Distribution of wealth shock, treated subjects, Main sample
Mean
p5
p10
p25
p50
p75
p90
p99
Sd
Count
70,817
1,176
2,533 7,323 20,046 50,930
150,676
769,817
358,073
28,827
Notes. Wealth shock is approximated by imputed inheritance tax payment, see Section 4 for description.
175
Appendix D: Sample characteristics, Placebo sample.
Table D1: Comparison of sample means, predetermined demographic and socioeconomic characteristics, “treated” and “controls”, Placebo sample.
“Treated”
“Controls”
p-value 1-2
1
2
IV
Birth-year
1950.2
1948.8
0.000
Age when inheriting
54.8
54.7
0.234
Woman
49.9
49.7
0.423
Swedish citizen
99.4
99.4
0.330
Married
56.2
58.1
0.000
Children in household1
32.8
33.5
0.018
Level of education2:
Primary
27.8
30.0
0.001
Lower secondary
45.9
45.2
0.008
Upper secondary or post
22.8
21.6
0.031
graduate
Earned income3
242,410
240,805
0.062
Net worth4
488,363
470,437
0.024
Number of obs.
48,165
85,970
Notes. Characteristics other than Birth-year, Age, Earned income and Net worth are
measured three years before the inheritance receipt. Indicator variables are reported in
percent. 1Refers to children younger than 18. 2Highest achieved level of education.
3
The means are calculated on annual incomes (adjusted for the growth in nominal
income, base year 2004) averaged over the available pre-inheritance years.4The means
are calculated on annual net worth (adjusted to 2004 price level using CPI) averaged
over the available pre-inheritance years.
Figure D1: The annual incidence of Hospitalization for “treated” and “controls”, Placebo sample.
Note. The vertical line indicates the point in time when inheritance is received. Controls 2004 does not
include offspring receiving inheritance from a parent over the period December 17-31.
176
Appendix E: DID estimates, heterogeneous effects,
Hospitalization
Table E1: Difference-in-difference estimates, impact of wealth shock on Hospitalization (in
percent), heterogeneous effects with respect to demographic characteristics, Main sample.
Age
Sex
Education
5
< Upper
secondary
or post
graduate
6
Upper
secondary
or post
graduate
7
0.621**
0.236
0.751***
-0.168
(0.255)
(0.312)
(0.305)
(0.279)
(0.345)
[7.58]
[8.66]
[8.59]
[9.07]
[7.79]
Yes
Yes
Yes
Yes
Yes
Yes
34,465
45,336
62,514
39,577
40,224
51,758
28,043
585,905
770,712
937,710
672,809
683,808
879,886
476,731
Young,
< mean
age
Old,
> mean
age
16-65
Women
Men
1
2
3
4
0.234
0.668**
0.275
(0.309)
(0.301)
[6.68]
[10.12]
Yes
N
N*T
DID estimate:
Year FE
Notes. Coefficient estimates are reported in percent. Standard errors (in percent) clustered at
individual, in parentheses. Mean of dependent variable (in percent), post-inheritance period for
control group, in brackets. ** significant at the 5 percent level, *** significant at the 1 percent
level.
Table E2: Difference-in-difference estimates, impact of wealth shock
on Hospitalization (in percent), heterogeneous effects with respect to
wealth shock, per quartile of the distribution, Main sample.
Wealth shock, by quartile of the distribution:
1st
1
2nd
2
3rd
3
4th
4
0.333
0.608
0.559
0.251
(0.431)
(0.459)
(0.438)
(0.416)
[9.14]
[8.81]
[8.48]
[8.07]
Yes
Yes
Yes
Yes
N
19,949
19,950
19,951
19,951
N*T
339,133
339,150
339,167
339,167
DID estimate:
Year FE
Notes. Coefficient estimates are reported in percent. Standard errors (in
percent) clustered at individual, in parentheses. Mean of dependent
variable (in percent), post-inheritance period for control group, in
brackets.
177
Table E3: Difference-in-difference estimates, impact of wealth shock on
Hospitalization (in percent), dynamics of responses, Main sample and
Placebo sample.
Main sample
Placebo sample
1
2
DID estimate by year since inheritance:
-8
0.0167
0.275
(0.314)
(0.255)
-7
-0.001
-0.302
(0.346)
(0.286)
-6
-0.001
-0.098
(0.352)
(0.292)
-5
-0.0136
-0.334
(0.355)
(0.293)
-4
-0.115
-0.303
(0.355)
(0.296)
-3
0.144
-0.245
(0.357)
(0.299)
-2
0.154
-0.420
(0.359)
(0.305)
-1
0.205
-0.687**
(0.373)
(0.308)
0
-0.199
-0.455
(0. 386)
(0. 316)
1
0.535
-0.200
(0.393)
(0.322)
2
0.891**
0.258
(0.396)
(0.324)
3
0.600
-0.380
(0.398)
(0.327)
4
0.424
-0.632
(0.404)
(0.670)
5
0.865**
-0.312
(0.409)
(0.338)
6
0.320
-0.636
(0.884)
(0.708)
Year FE
Yes
Yes
N
79,802
134,172
N*T
1,356,634
2,280,924
Notes. Coefficient estimates are reported in percent. Standard errors
(in percent) clustered at individual, in parentheses. Mean of dependent
variable (in percent), post-inheritance period for control group, in
brackets. ** significant at the 5 percent level.
178
Appendix F: DID estimates of the effect of the wealth
shock on Sick leave and Mortality
Table F1: Difference-in-differences (DID) estimates, impact of wealth shock on Sick leave
(in percent), Main sample and Placebo sample.
Main sample
Placebo sample
1
2
3
4
-1.12***
0.311
-1.44***
0. 211
(0. 201)
(0. 386)
(0. 174)
(0. 327)
[13.96]
[13.96]
[15.62]
[15.62]
No
Yes
No
Yes
N
61,584
61,584
93,961
93,961
N*T
911,750
911,750
1,394,978
1,394,978
DID estimate :
Year FE
Notes. Coefficient estimates are reported in percent. Standard errors (in percent) clustered
at individual, in parentheses. Mean of dependent variable (in percent), post-inheritance
period for control group, in brackets. *** significant at the 1 percent level.
Table F2: Differences estimates and Difference-in-differences (DID) estimates, impact of wealth shock on mortality (in percent).
Differences estimates:
DID estimates:
Main sample
Placebo sample
1-2
1
2
3
Outcome:
Mortality1
-0.06
-0.01
-0.06
(0.08)
(0.06)
(0.10)
[0.78]
[0.92]
Mortality2
-0.05
0.06
-0.11
(0.10)
(0.08)
(0.13)
[1.33]
[1.50]
Mortality3
-0.06
0.10
-0.16
(0.12)
(0.10)
(0.15)
[1.85]
[2.16]
Mortality4
-0.06
0.08
-0.14
(0.13)
(0.11)
(0.17)
[2.39]
[2.84]
Mortality5
-0.10
0.13
-0.23
(0.15)
(0.13)
(0.20)
[3.04]
[3.62]
Mortality6
-0.06
0.04
-0.10
(0.16)
(0.14)
(0.21)
[3.66]
[4.49]
N
51,835
86,733
138,568
Notes. Coefficient estimates are reported in percent. Robust standard errors
(in percent) in parentheses. Mean of dependent variable (in percent), for
control group in brackets. The estimates have been obtained from models
with controls for: age, age2, gender, marital status, presence of children, level
of education (highest achieved), earned income and net worth, measured
three years before the inheritance.
179
10 References
Adams, P., M.D. Hurd, D. Mcfadden, A. Merrill, and T. Ribeiro. (2003). “Healthy,
wealthy, and wise? Tests for direct causal paths between health and
socioeconomic status,” Journal of Econometrics, 112, 3–56.
Adda, J., J. Banks, and H.-M. von Gaudecker (2009). “The impact of income shocks
on health: evidence from cohort data,” Journal of the European Economic
Association, 7, 1361–1399.
Almond, D. and J. Currie (2013). “Killing me softly: the fetal origins hypothesis,”
Journal of Economic Perspectives, 25, 153–172
Apouey, B. and A. E. Clark (2013). “Winning big but feeling no better? The effect
of lottery prizes on physical and mental health,” CEP Discussion Paper No.
1228.
Bago d’Uva, T., E. van Doorslaer, M. Lindeboom, O. O’Donnell and S. Chatterji
(2008). “Does reporting heterogeneity bias the measurement of health disparities?” Health Economics, 17, 351–375.
Baker, M., M. Stabile and C. Deri (2004). “What do self-reported objective measures of health measure?” Journal of Human Resources, 39, 1067–1093.
Barker, D. J. (1997). “Maternal nutrition, fetal nutrition and diseases later in life,”
Nutrition, 13, 807–813.
Barsky, R. B., F. T. Juster, M.S., Kimball, and M.D. Shapiro (1997). “Preference
parameters and behavioral heterogeneity: An experimental approach in the
health and retirement study,” Quarterly Journal of Economics, 112, 537–579.
Becker, G. S. (1976). The Economic Approach to Human Behavior. University of
Chicago Press.
Benzeval, M. and K. Judge (2001). “Income and health: the time dimension,” Social
Science & Medicine, 52, 1371–1390.
Burbidge, J. B., Magee L., and A. L. Robb (1988). “Alternative transformations to
handle extreme values of the dependent variable,” Journal of the American Statistical Association, 83, 123–127.
Carman, K. G. (2013). “Inheritances, intergenerational transfers, and the
accumulation of health,” American Economic Review: Papers & Proceedings
2013, 103, 451–455.
Case, A. (2004). “Does money protect health status? Evidence from South African
pensions,” in D. Wise (ed.). Perspectives on the Economics of Aging.
University of Chicago Press.
Case, A. and C. Paxson (2005). “Sex differences in morbidity and mortality,”
Demography, 42, 189–212.
Cawley, J. and C. J. Ruhm (2011). “The economics of risky health behaviors,” in T.
G. M. Mark, V. Pauly and P. B. Pedro (eds.), Handbook of Health Economics.
Elsevier.
Chida, Y., and A. Steptoe (2008). “Positive psychological wellbeing and mortality: a
quantitative review of prospective observational studies,” Psychosomatic Medicine, 70, 741-756.
Contoyannis, P. and A. M. Jones (2004). “Socio-economic status, health and lifestyle,” Journal of Health Economics, 23, 965–995
Cutler, D. M. and A. Lleras-Muney (2010). “Understanding differences in health
behaviors by education,” Journal of Health Economics, 29, 1–28.
Cutler, D. M., A. Deaton, and A. Lleras-Muney (2011). “Socioeconomic status and
health: dimensions and mechanisms,” in Glied, S. and P. C. Smith (eds.), The
Oxford Handbook of Health Economics. Oxford University Press.
180
Deaton, A. (2003). “Health, inequality, and economic development,” Journal of
Economic Literature, 41, 113–158.
Deaton, A. S. and C. H. Paxson (1998). “Aging and inequality in income and
health,” American Economic Review, 88, 248–253.
Du Rietz, G., M. Henrekson, and D. Waldenström. (2012). “The Swedish inheritance and gift taxation, 1885–2004,” IFN Working Paper No. 936.
Eliason, M. and H. Ohlsson (2013). “Timing of death and the repeal of the Swedish
inheritance tax,” Journal of Socio-Economics, 45, 113–123.
Ehrlich, I. and H. Chuma (1990). “A model of the demand for longevity and the
value of life extension,” Journal of Political Economy, 98, 761-782.
Ettner, S. L. (1996). “New evidence on the relationship between income and health,”
Journal of Health Economics, 15, 67–85.
Frijters, P., J. P. Haisken-DeNew, and M. A. Shields (2005). “The causal effect of
income on health: Evidence from German reunification. Journal of Health
Economics, 24, 997–1017.
Fuchs, V. R. (1982). Time preferences and health: an exploratory study on economic
aspects of health. University of Chicago Press.
Galama, T. and H. Van Kippersluis (2010). “A theory of socioeconomic disparities
in health over the life cycle,” Rand Working Paper, WR-773.
Gardner, J. and A. J. Oswald (2007). “Money and mental wellbeing: a longitudinal
study of medium-sized lottery wins,” Journal of Health Economics, 26, 49–60.
Gerdtham, U.-G. and M. Johannesson (2005). “Business cycles and mortality:
results from Swedish microdata,” Social Science & Medicine, 60, 205–218.
Glenngård A.H., F. Hjalte, M. Svensson, A. Anell, and V. Bankauskaite (2005).
”Health Systems in Transition: Sweden,” WHO Regional Office for Europe on
behalf of the European Observatory on Health Systems and Policies, Copenhagen.
Goldman, D. P. and J. P. Smith (2002). “Can patient self-management help explain
the SES health gradient?” Proceedings of the National Academy of Science of
the United States of America, 99, 10929–10934.
Grossman, M. (1972). “On the concept of health capital and the demand for health,”
Journal of Political Economy, 80, 223–255.
Grossman, M. (2000). “The human capital model,” in A.J. Culyer and J.P.
Newhouse (eds.), Handbook of Health Economics. Elsevier.
Hesselius, P., P. Johansson, and L. Larsson (2008). “Monitoring and norms in sickness insurance: empirical evidence from a natural experience,” IFAU Working
Paper 2008:8.
Idler, E. L. and Y. Benyamini (1997). “Self-rated health and mortality: a review of
twenty-seven community studies,” Journal of Health and Social Behavior, 38,
21–37.
Jensen, R. T. and K. Richter (2004). “The health implications of social security
failure: evidence from the Russian pension crisis,” Journal of Public Economics,
88, 209–236.
Joulfaian, D. (2004). “Gift taxes and lifetime transfers: time series evidence,”
Journal of Public Economics, 88, 1917–1929.
Kenkel, D. S. (2000). “Prevention,” in A.J. Culyer and J.P. Newhouse (eds.), Handbook of Health Economics. Elsevier.
Kim, B. and C. J. Ruhm (2012). “Inheritances, health and death,” Health
Economics, 21, 127–144.
Kivimäki, M., Head. J., Ferrie, J.E., Shipley, M., Vahtera, J., and Marmot, M.G.
(2003). “Sickness absence as a global measure of health: evidence from mortality in the Whitehall II prospective cohort study,” BMJ, 327, 364–368.
181
Kopczuk, W. (2007). “Bequest and tax planning: Evidence from estate tax returns,”
Quarterly Journal of Economics, 122, 1801–1854.
Kopczuk, W. and J. Slemrod. (2005). “Denial of death and economic behavior,”
Advances in Theoretical Economics, 5, Article 1.
Kushi L.H., Doyle C., McCullough M., Rock, C. L., Demark-Wahnefried, W., Bandera, E.V. Gapstur, S., Patel, A.V., Andrews, K., and T. Gansler
(2012). “American Cancer Society Guidelines on nutrition and physical activity
for cancer prevention: reducing the risk of cancer with healthy food choices and
physical activity,” CA: Cancer Journal for Clinicians, 62, 30–67.
Larsson, L. (2006). “Sick of being unemployed? Interactions between
unemployment and sickness insurance,” Scandinavian Journal of Economics,
108, 97–113.
Lleras-Muney, A. (2005). “The relationship between education and adult mortality
in the United States,” Review of Economic Studies, 72, 189–221.
Lodin, S.-O. (2009). Professorn som blev näringslivstorped: min tid i skattepolitiken. Ekerlids Förlag.
Lundborg P., M. Nilsson and J. Vikström (2011). ”Socioeconomic heterogeneity in
the effect of health shocks on earnings: evidence from population-wide data on
Swedish workers,” IFAU Working Paper 2011:11.
Mankiw, N.G. and S.P. Zeldes (1991). “The consumption of stockholders and
nonstockholders,” Journal of Financial Economics, 29, 97–112.
Marks, N. F., H. Jun, and J. Song (2007). “Death of parents and adult psychological
and physical well-being: a prospective U.S. national study,” Journal of Family
Issues, 28, 1611–1638.
Marmot, M., A. Feeney, M. Shipley, F. North, and S. L. Syme (1995). “Sickness
absence as a measure of health status and functioning: from the UK Whitehall II
study,” Journal of Epidemiology and Community Health, 49, 124–130.
Marmot, M. (1999). “Multi-Level Approaches to Understanding Social Determinants,” in Berkman, L. and I. Kawachi, (eds.), Social Epidemiology. Oxford
University Press.
McGinnis, J. M. and W.H. Foege (1993). “Actual causes of death in the United
States,” JAMA, 270, 2207–2212.
Meer, J., D. L. Miller, and H. Rosen (2003). “Exploring the health-wealth nexus,”
Journal of Health Economics, 22, 713–730.
Michaud, P.C. and A. van Soest (2008). “Health and wealth of elderly couples:
causality tests using dynamic panel data models,” Journal of Health Economics,
27, 1312–1325.
Mokdad, A. H., J. S. Marks, D. F. Stroup and J. L. Gerberding (2004). “Actual
causes of death in the United States, 2000,” JAMA, 291, 1238–1245.
Murray, C. and L. Chen (1992). “Understanding morbidity change,” Population and
Development Review, 18, 481–503.
Muurinen, J.-M. (1982). “Demand for health: A generalized Grossman model,”
Journal of Health Economics, 1, 5–28.
Nordblom, K. and H. Ohlsson (2006). “Tax avoidance and intra-family transfers,”
Journal of Public Economics, 90, 1669–1680.
Ohlsson, H. (2011). “The legacy of the Swedish gift and inheritance tax, 18842004,” European Review of Economic History, 15, 539–569.
Poterba, J. M. and A.A. Samwick (2003). “Taxation and household portfolio
composition: US evidence from the 1980s and 1990s,” Journal of Public
Economics, 87, 5–38.
Rostila, M. and J. M. Saarela (2011). “Time does not heal all wounds: mortality
following the death of a parent,” Journal of Marriage and Family, 73, 236–249.
182
Ruhm, C. J. (2000). “Are recessions good for your health?” The Quarterly Journal of
Economics, 115, 617–650.
Schwandt, H. (2012). “Is wealth causing health? Evidence from stock market induced wealth shocks,” Job market paper Universitat Pompeu Fabra.
Scharlach, A. E. (1991). “Factors associated with filial grief following the death of
an elderly parent,” American Journal of Orthopsychiatry, 61, 307–313.
Silfverberg, C. (2005). ”Slopad arvs- och gåvoskatt,” Skattenytt 2005.
Smith, J.P. (1999). “Healthy bodies and thick wallets: the dual relation between
health and economic status,” Journal of Economic Perspectives, 13, 145–166.
Smith, J.P. (2004). “Unraveling the SES health connection,” Population and
Development Review, 30, 108–132.
Smith, J.P. (2005). “Consequences and predictors of new health events,” in D. Wise
(ed.), Analyses in the Economics of Aging. University of Chicago Press.
Snyder, S. E. and W. N. Evans (2006). “The effect of income on mortality: evidence
from the Social Security Notch,” The Review of Economics and Statistics, 88,
482–495.
SOU 2002:52. Beskattning av småföretagare, del II. Generationsskiften. Betänkande
av 3:12-utredningen. Fritzes.
SOU 2003:3. Egendomsskatter Dämpningsregel för fastighetsskatten och sänkt
arvsskatt. Fritzes.
SOU:2004:66. Egendomsskatter. Reform av arvs- och gåvoskatter. Slutbetänkande
av Egen-domsskattekommittén. Fritzes.
Umberson, D. and M. D. Chen (1994). “Effects of a parent’s death on adult children:
relationship salience and reaction to loss,” American Sociological Review, 59,
152–168.
Van Doorslaer, E. and U.-G. Gertham (2003). “Does inequality in self-assessed
health predict inequality in survival by income? Evidence from Swedish data,”
Social Science and Medicine, 57, 1621–1629.
Van Kippersluis, H. and T. Galama (2013). “Why the rich drink more but smoke
less: The impact of wealth on health behaviors,” Rand Labor and Population
Working Paper, WR-988.
Wu, S. (2003). “The effects of health events on the economic status of married
couples,” Journal of Human Resources, 38, 219–230.
World Health Organization (WHO) (2002). World health report 2002: reducing
risks, promoting healthy life. Geneva, Switzerland.
183
Economic Studies
____________________________________________________________________
1987:1
Haraldson, Marty: To Care and To Cure. A linear programming approach to
national health planning in developing countries. 98 pp.
1989:1
Chryssanthou, Nikos: The Portfolio Demand for the ECU:
Approach. 42 pp.
1989:2
Hansson, Bengt: Construction of Swedish Capital Stocks, 1963-87: An Application of
the Hulten-Wykoff Studies. 37 pp.
1989:3
Choe, Byung-Tae: Some Notes on Utility Functions Demand and Aggregation. 39
pp.
1989:4
Skedinger, Per: Studies of Wage and Employment Determination in the Swedish
Wood Industry. 89 pp.
1990:1
Gustafson, Claes-Håkan: Inventory Investment in Manufacturing Firms. Theory and
Evidence. 98 pp.
1990:2
Bantekas, Apostolos: The Demand for Male and Female Workers in Swedish
Manufacturing. 56 pp.
1991:1
Lundholm, Michael: Compulsory Social Insurance. A Critical Review. 109 pp.
1992:1
Sundberg, Gun: The Demand for Health and Medical Care in Sweden. 58 pp.
1992:2
Gustavsson, Thomas: No Arbitrage Pricing and the Term Structure of Interest Rates.
47 pp.
1992:3
Elvander, Nils: Labour Market Relations in Sweden and Great Britain. A Comparative Study of Local Wage Formation in the Private Sector during the 1980s. 43
pp.
Dillén, Mats: Studies in Optimal Taxation, Stabilization, and Imperfect Competition.
1993. 143 pp.
12
A Transaction Cost
13
Banks, Ferdinand E.: A Modern Introduction to International Money, Banking and
Finance. 1993. 303 pp.
14
Mellander, Erik: Measuring Productivity and Inefficiency Without Quantitative
Output Data. 1993. 140 pp.
15
Ackum Agell: Susanne: Essays on Work and Pay. 1993. 116 pp.
16
Eriksson, Claes: Essays on Growth and Distribution. 1994. 129 pp.
17
Banks, Ferdinand E.: A Modern Introduction to International Money, Banking and
Finance. 2nd version, 1994. 313 pp.
18
Apel, Mikael: Essays on Taxation and Economic Behavior. 1994. 144 pp.
19
Dillén, Hans: Asset Prices in Open Monetary Economies. A Contingent Claims
Approach. 1994. 100 pp.
20
Jansson, Per: Essays on Empirical Macroeconomics. 1994. 146 pp.
21
Banks, Ferdinand E.: A Modern Introduction to International Money, Banking, and
Finance. 3rd version, 1995. 313 pp.
22
Dufwenberg, Martin: On Rationality and Belief Formation in Games. 1995. 93 pp.
23
Lindén, Johan: Job Search and Wage Bargaining. 1995. 127 pp.
24
Shahnazarian, Hovick: Three Essays on Corporate Taxation. 1996. 112 pp.
25
Svensson, Roger: Foreign Activities of Swedish Multinational Corporations. 1996.
166 pp.
26
Sundberg, Gun: Essays on Health Economics. 1996. 174 pp.
27
Sacklén, Hans: Essays on Empirical Models of Labor Supply. 1996. 168 pp.
28
Fredriksson, Peter: Education, Migration and Active Labor Market Policy. 1997. 106 pp.
29
Ekman, Erik: Household and Corporate Behaviour under Uncertainty. 1997. 160 pp.
30
Stoltz, Bo: Essays on Portfolio Behavior and Asset Pricing. 1997. 122 pp.
31
Dahlberg, Matz: Essays on Estimation Methods and Local Public Economics. 1997. 179
pp.
32
Kolm, Ann-Sofie: Taxation, Wage Formation, Unemployment and Welfare. 1997. 162
pp.
33
Boije, Robert: Capitalisation, Efficiency and the Demand for Local Public Services. 1997.
148 pp.
34
Hort, Katinka: On Price Formation and Quantity Adjustment in Swedish Housing
Markets. 1997. 185 pp.
35
Lindström, Thomas: Studies in Empirical Macroeconomics. 1998. 113 pp.
36
Hemström, Maria: Salary Determination in Professional Labour Markets. 1998. 127 pp.
37
Forsling, Gunnar: Utilization of Tax Allowances and Corporate Borrowing. 1998. 96
pp.
38
Nydahl, Stefan: Essays on Stock Prices and Exchange Rates. 1998. 133 pp.
39
Bergström, Pål: Essays on Labour Economics and Econometrics. 1998. 163 pp.
40
Heiborn, Marie: Essays on Demographic Factors and Housing Markets. 1998. 138 pp.
41
Åsberg, Per: Four Essays in Housing Economics. 1998. 166 pp.
42
Hokkanen, Jyry: Interpreting Budget Deficits and Productivity Fluctuations. 1998. 146
pp.
43
Lunander, Anders: Bids and Values. 1999. 127 pp.
44
Eklöf, Matias: Studies in Empirical Microeconomics. 1999. 213 pp.
45
Johansson, Eva: Essays on Local Public Finance and Intergovernmental Grants. 1999.
156 pp.
46
Lundin, Douglas: Studies in Empirical Public Economics. 1999. 97 pp.
47
Hansen, Sten: Essays on Finance, Taxation and Corporate Investment. 1999. 140 pp.
48
Widmalm, Frida: Studies in Growth and Household Allocation. 2000. 100 pp.
49
Arslanogullari, Sebastian: Household Adjustment to Unemployment. 2000. 153 pp.
50
Lindberg, Sara: Studies in Credit Constraints and Economic Behavior. 2000. 135 pp.
51
Nordblom, Katarina: Essays on Fiscal Policy, Growth, and the Importance of Family
Altruism. 2000. 105 pp.
52
Andersson, Björn: Growth, Saving, and Demography. 2000. 99 pp.
53
Åslund, Olof: Health, Immigration, and Settlement Policies. 2000. 224 pp.
54
Bali Swain, Ranjula: Demand, Segmentation and Rationing in the Rural Credit Markets
of Puri. 2001. 160 pp.
55
Löfqvist, Richard: Tax Avoidance, Dividend Signaling and Shareholder Taxation in an
Open Economy. 2001. 145 pp.
56
Vejsiu, Altin: Essays on Labor Market Dynamics. 2001. 209 pp.
57
Zetterström, Erik: Residential Mobility and Tenure Choice in the Swedish Housing
Market. 2001. 125 pp.
58
Grahn, Sofia: Topics in Cooperative Game Theory. 2001. 106 pp.
59
Laséen, Stefan: Macroeconomic Fluctuations and Microeconomic Adjustments: Wages,
Capital, and Labor Market Policy. 2001. 142 pp.
60
Arnek, Magnus: Empirical Essays on Procurement and Regulation. 2002. 155 pp.
61
Jordahl, Henrik: Essays on Voting Behavior, Labor Market Policy, and Taxation. 2002.
172 pp.
62
Lindhe, Tobias: Corporate Tax Integration and the Cost of Capital. 2002. 102 pp.
63
Hallberg, Daniel: Essays on Household Behavior and Time-Use. 2002. 170 pp.
64
Larsson, Laura: Evaluating Social Programs: Active Labor Market Policies and Social
Insurance. 2002. 126 pp.
65
Bergvall, Anders: Essays on Exchange Rates and Macroeconomic Stability. 2002.
122 pp.
66
Nordström Skans, Oskar: Labour Market Effects of Working Time Reductions and
Demographic Changes. 2002. 118 pp.
67
Jansson, Joakim: Empirical Studies in Corporate Finance, Taxation and Investment.
2002. 132 pp.
68
Carlsson, Mikael: Macroeconomic Fluctuations and Firm Dynamics: Technology,
Production and Capital Formation. 2002. 149 pp.
69
Eriksson, Stefan: The Persistence of Unemployment: Does Competition between
Employed and Unemployed Job Applicants Matter? 2002. 154 pp.
70
Huitfeldt, Henrik: Labour Market Behaviour in a Transition Economy: The Czech
Experience. 2003. 110 pp.
71
Johnsson, Richard: Transport Tax Policy Simulations and Satellite Accounting within a
CGE Framework. 2003. 84 pp.
72
Öberg, Ann: Essays on Capital Income Taxation in the Corporate and Housing Sectors.
2003. 183 pp.
73
Andersson, Fredrik: Causes and Labor Market Consequences of Producer
Heterogeneity. 2003. 197 pp.
74
Engström, Per: Optimal Taxation in Search Equilibrium. 2003. 127 pp.
75
Lundin, Magnus: The Dynamic Behavior of Prices and Investment: Financial
Constraints and Customer Markets. 2003. 125 pp.
76
Ekström, Erika: Essays on Inequality and Education. 2003. 166 pp.
77
Barot, Bharat: Empirical Studies in Consumption, House Prices and the Accuracy of
European Growth and Inflation Forecasts. 2003. 137 pp.
78
Österholm, Pär: Time Series and Macroeconomics: Studies in Demography and
Monetary Policy. 2004. 116 pp.
79
Bruér, Mattias: Empirical Studies in Demography and Macroeconomics. 2004. 113 pp.
80
Gustavsson, Magnus: Empirical Essays on Earnings Inequality. 2004. 154 pp.
81
Toll, Stefan: Studies in Mortgage Pricing and Finance Theory. 2004. 100 pp.
82
Hesselius, Patrik: Sickness Absence and Labour Market Outcomes. 2004. 109 pp.
83
Häkkinen, Iida: Essays on School Resources, Academic Achievement and Student
Employment. 2004. 123 pp.
84
Armelius, Hanna: Distributional Side Effects of Tax Policies: An Analysis of Tax
Avoidance and Congestion Tolls. 2004. 96 pp.
85
Ahlin, Åsa: Compulsory Schooling in a Decentralized Setting: Studies of the Swedish
Case. 2004. 148 pp.
86
Heldt, Tobias: Sustainable Nature Tourism and the Nature of Tourists' Cooperative
Behavior: Recreation Conflicts, Conditional Cooperation and the Public Good Problem.
2005. 148 pp.
87
Holmberg, Pär: Modelling Bidding Behaviour in Electricity Auctions: Supply Function
Equilibria with Uncertain Demand and Capacity Constraints. 2005. 43 pp.
88
Welz, Peter: Quantitative new Keynesian macroeconomics and monetary policy
2005. 128 pp.
89
Ågren, Hanna: Essays on Political Representation, Electoral Accountability and Strategic
Interactions. 2005. 147 pp.
90
Budh, Erika: Essays on environmental economics. 2005. 115 pp.
91
Chen, Jie: Empirical Essays on Housing Allowances, Housing Wealth and Aggregate
Consumption. 2005. 192 pp.
92
Angelov, Nikolay: Essays on Unit-Root Testing and on Discrete-Response Modelling of
Firm Mergers. 2006. 127 pp.
93
Savvidou, Eleni: Technology, Human Capital and Labor Demand. 2006. 151 pp.
94
Lindvall, Lars: Public Expenditures and Youth Crime. 2006. 112 pp.
95
Söderström, Martin: Evaluating Institutional Changes in Education and Wage Policy.
2006. 131 pp.
96
Lagerström, Jonas: Discrimination, Sickness Absence, and Labor Market Policy. 2006.
105 pp.
97
Johansson, Kerstin: Empirical essays on labor-force participation, matching, and trade.
2006. 168 pp.
98
Ågren, Martin: Essays on Prospect Theory and the Statistical Modeling of Financial
Returns. 2006. 105 pp.
99
Nahum, Ruth-Aïda: Studies on the Determinants and Effects of Health, Inequality and
Labour Supply: Micro and Macro Evidence. 2006. 153 pp.
100
Žamac, Jovan: Education, Pensions, and Demography. 2007. 105 pp.
101
Post, Erik: Macroeconomic Uncertainty and Exchange Rate Policy. 2007. 129 pp.
102
Nordberg, Mikael: Allies Yet Rivals: Input Joint Ventures and Their Competitive Effects.
2007. 122 pp.
103
Johansson, Fredrik: Essays on Measurement Error and Nonresponse. 2007. 130 pp.
104
Haraldsson, Mattias: Essays on Transport Economics. 2007. 104 pp.
105
Edmark, Karin: Strategic Interactions among Swedish Local Governments. 2007. 141 pp.
106
Oreland, Carl: Family Control in Swedish Public Companies. Implications for Firm
Performance, Dividends and CEO Cash Compensation. 2007. 121 pp.
107
Andersson, Christian: Teachers and Student Outcomes: Evidence using Swedish Data.
2007. 154 pp.
108
Kjellberg, David: Expectations, Uncertainty, and Monetary Policy. 2007. 132 pp.
109
Nykvist, Jenny: Self-employment Entry and Survival - Evidence from Sweden. 2008.
94 pp.
110
Selin, Håkan: Four Empirical Essays on Responses to Income Taxation. 2008. 133 pp.
111
Lindahl, Erica: Empirical studies of public policies within the primary school and the
sickness insurance. 2008. 143 pp.
112
Liang, Che-Yuan: Essays in Political Economics and Public Finance. 2008. 125 pp.
113
Elinder, Mikael: Essays on Economic Voting, Cognitive Dissonance, and Trust. 2008.
120 pp.
114
Grönqvist, Hans: Essays in Labor and Demographic Economics. 2009. 120 pp.
115
Bengtsson, Niklas: Essays in Development and Labor Economics. 2009. 93 pp.
116
Vikström, Johan: Incentives and Norms in Social Insurance: Applications, Identification
and Inference. 2009. 205 pp.
117
Liu, Qian: Essays on Labor Economics: Education, Employment, and Gender. 2009. 133
pp.
118
Glans, Erik: Pension reforms and retirement behaviour. 2009. 126 pp.
119
Douhan, Robin: Development, Education and Entrepreneurship. 2009.
120
Nilsson, Peter: Essays on Social Interactions and the Long-term Effects of Early-life
Conditions. 2009. 180 pp.
121
Johansson, Elly-Ann: Essays on schooling, gender, and parental leave. 2010. 131 pp.
122
Hall, Caroline: Empirical Essays on Education and Social Insurance Policies. 2010.
147 pp.
123
Enström-Öst, Cecilia: Housing policy and family formation. 2010. 98 pp.
124
Winstrand, Jakob: Essays on Valuation of Environmental Attributes. 2010. 96 pp.
125
Söderberg, Johan: Price Setting, Inflation Dynamics, and Monetary Policy. 2010. 102 pp.
126
Rickne, Johanna: Essays in Development, Institutions and Gender. 2011. 138 pp.
127
Hensvik, Lena: The effects of markets, managers and peers on worker outcomes. 2011.
179 pp.
128
Lundqvist, Heléne: Empirical Essays in Political and Public. 2011. 157 pp.
129
Bastani, Spencer: Essays on the Economics of Income Taxation. 2012. 257 pp.
130
Corbo, Vesna: Monetary Policy, Trade Dynamics, and Labor Markets in Open
Economies. 2012. 262 pp.
131
Nordin, Mattias: Information, Voting Behavior and Electoral Accountability. 2012.
187 pp.
132
Vikman, Ulrika: Benefits or Work? Social Programs and Labor Supply. 2013. 161 pp.
133
Ek, Susanne: Essays on unemployment insurance design. 2013. 136 pp.
134
Österholm, Göran: Essays on Managerial Compensation. 2013. 143 pp.
135
Adermon, Adrian: Essays on the transmission of human capital and the impact of
technological change. 2013. 138 pp.
136
Kolsrud, Jonas: Insuring Against Unemployment 2013. 140 pp.
137
Hanspers, Kajsa: Essays on Welfare Dependency and the Privatization of Welfare
Services. 2013. 208 pp.
138
Persson, Anna: Activation Programs, Benefit Take-Up, and Labor Market Attachment.
2013. 164 pp.
139
Engdahl, Mattias: International Mobility and the Labor Market. 2013. 216 pp.
140
Krzysztof Karbownik. Essays in education and family economics. 2013. 182 pp.