Economic Studies 156
Gabriella Chirico Willstedt
Demand, Competition and Redistribution in Swedish Dental Care
Gabriella Chirico Willstedt
Demand, Competition and Redistribution
in Swedish Dental Care
Department of Economics, Uppsala University
Visiting address: Kyrkogårdsgatan 10, Uppsala, Sweden
Postal address: Box 513, SE-751 20 Uppsala, Sweden
Telephone:
+46 18 471 00 00
Telefax:
+46 18 471 14 78
Internet:http://www.nek.uu.se/
_______________________________________________________
ECONOMICS AT UPPSALA UNIVERSITY
The Department of Economics at Uppsala University has a long history.
The first chair in Economics in the Nordic countries was instituted at
Uppsala University in 1741.
The main focus of research at the department has varied over the years
but has typically been oriented towards policy-relevant applied economics,
including both theoretical and empirical studies. The currently most active
areas of research can be grouped into six categories:
*
Labour economics
*
Public economics
*Macroeconomics
*Microeconometrics
*
Environmental economics
*
Housing and urban economics
_______________________________________________________
Additional information about research in progress and published reports is
given in our project catalogue. The catalogue can be ordered directly from
the Department of Economics.
Dissertation presented at Uppsala University to be publicly examined in Hörsal 2,
Ekonomikum, Kyrkogårdsgatan 10, Uppsala, Friday, 15 January 2016 at 10:15 for the degree
of Doctor of Philosophy. The examination will be conducted in English. Faculty examiner:
Professor Tor Iversen (University of Olso).
Abstract
Chirico Willstedt, G. 2015. Demand, Competition and Redistribution in Swedish
Dental Care. Economic studies 156. 119 pp. Uppsala: Department of Economics.
ISBN 978-91-85519-63-7.
Essay 1: Individuals with higher socioeconomic status (SES) also tend to enjoy better health.
Evidence from the economics literature suggests that a potential mechanism behind this “social
health gradient” is that human capabilities, that form SES, also facilitate health-promoting
behaviors. This essay empirically investigates the significance of socioeconomic differences in
health behaviors, using dental care consumption as an operationalization of health investments.
I focus on adults at an age where lifetime trajectories for SES can be taken as given and
use lifetime income to capture SES. I estimate the impact of lifetime income on dental care
consumption and find robust evidence that the social gradient in dental care consumption
steepens dramatically over the life-cycle. Considering that dental care consumption only reflects
a small part of individuals' health investments the results suggest that lifetime effects of SES on
health behaviors could be substantial in other dimensions.
Essay 2: This essay studies the effect of competition on prices on a health care market where
prices are market determined, namely the Swedish market for dental care. The empirical strategy
exploits that the effect of competition differs across services, depending on the characteristics
of the service. Price competition is theoretically more intense for services such as examinations
and diagnostics (first-stage services), compared to more complicated and unusual treatments
(follow-on services). By exploiting this difference, I identify a relative effect of competition on
prices. The results suggest small but statistically significant negative short-term effects on prices
for first-stage services relative to follow-on services. The results provide evidence that pricesetting among dental care clinics responds to changes in the market environment and substantial
effects of competition on prices over time cannot be ruled out.
Essay 3: The Swedish dental care insurance subsidizes dental care costs above a threshold
and becomes more generous as dental care consumption increases. On average, higher-income
individuals consume more dental care and have better oral health than low-income individuals.
Therefore, the redistributional effects of the Swedish dental care insurance are ambiguous a
priori. I find that the dental care insurance adds to the progressive redistribution taking place
through other parts of the Swedish social insurance (SI) for individuals aged 35-59 years
whereas it reduces the progressivity in the SI for those aged 60-89 years. While the result for the
oldest individuals is problematic from an equity point of view, the insurance seems to strengthen
the progressitivy of the Swedish social insurance for the vast majority of patients.
Keywords: Health, dental care, Grossman model, socioeconomic status, health disparities,
social health gradient, competition in health care, public health insurance, dental care
insurance, social insurance, redistribution.
Gabriella Chirico Willstedt, Department of Economics, Box 513, Uppsala University,
SE-75120 Uppsala, Sweden.
© Gabriella Chirico Willstedt 2015
ISSN 0283-7668
ISBN 978-91-85519-63-7
urn:nbn:se:uu:diva-267476 (http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-267476)
For Robert
Acknowledgments
As a four-year-old I wanted to be a bus driver so that I could drop of my future kids
at day care while working. It may seem clear that a kid with that kind of interest in
efficiency would end up an economist but the road hasn’t been straight and I wouldn’t
be where I am today without the help and encouragement of the people around me.
First of all, I am deeply grateful to my main supervisor Professor Per Johansson.
Per, thank you for taking me in as a summer intern at IFAU, for encouraging me to
apply to the PhD program and for all the support and patience along the way. Your
take on research and enthusiasm for understanding human behavior is inspiring and an
important reminder that what we do as economists is really exciting! I am so grateful
for how generous you are with your time and immense knowledge and I have very
much appreciated our many discussions over the years. Not only has your guidance
and support improved this thesis a great deal and taught me a lot about economics and
econometrics, it has also made me feel more confident about my work and myself as a
researcher. To my co-advisor Erik Grönqvist, I want to express my deepest gratitude
for your careful reading of all my unfinished drafts that has greatly enhanced the quality
of this thesis. I have always felt that you are genuinely interested in my research and
your constructive feedback has pushed me to really think about what we can take away
from my results and why. In addition, I would like to thank you for introducing me
into various health econ contexts, making me feel like a colleague.
I would like to express my deepest thanks to the discussant on my final seminar
covering the first and the third essay, Erik Lindqvist, for giving excellent feedback and
suggestions. Many thanks also to Per Molander for giving me the opportunity to spend
time at The Swedish Social Insurance Inspectorate throughout the completion of this
thesis. I would also like to thank Douglas Lundin for showing interest in my work and
for taking the time to discuss my projects.
Thank you Laura Hartman for welcoming me with open arms at IFAU back when I
knew nothing about economics research. You inspired me to apply to the program and
have been an important inspiration ever since.
During my years at the Department of Economics at Uppsala University I have
had the pleasure to meet a lot of intelligent and kind people. I would like to start
out by thanking Katarina, Ann-Sofie, Berit, Nina, Åke and Stina for doing such an
excellent job running this department and being so nice. I am also very grateful to
Mikael Lindahl for encouraging and contributing to the relaxed and open discussions
in the microeconometrics study group at the department. I would especially like to
thank Johan for being not only a great fellow PhD student but also my best friend
and coach. If it wasn’t for you I wouldn’t have laughed as much as I have through
these years, nor would I understand (as much as I think I do) about how everything
works. Most importantly, you helped me be strong throughout this journey, not only
by energetically encouraging me to lift heavier in the gym. Special thanks also to
Tove, my ol’ office pal, for all the profound talks on economics and the sometimes too
distant life outside the walls of Ekonomikum. I would also like to thank the rest of my
cohort Chris, Daniel, Glenn, Haishan and Jon for the moral support, the occasional
post-exam party and the nice company at countless dinners enjoyed in the fikarum.
Special thanks also to Lovisa. You are a true intellectual in all the right ways and I
have learned a lot from our discussions on both social science and American celebs.
Your support during the completion of this thesis means a great deal to me and I am
happy to have you as a friend. Thanks to Susanne for being such a great friend on
and outside the office (both old and new!). Your encouragement and acting as a career
coach has meant a lot to me and I am so excited about what the future holds for our
duo. Thanks to Erik, Arizo and Oscar for being both great colleagues and friends. I
would also like to thank my fellow health econ students Anna, Evelina and Mattias Ö
for inspiring discussions about health stocks, identification and lack thereof. Thanks
to Jonas, Fredrik, Linna, Selva, Irina, Mohammad, Linuz, Ylva, Kicki and Jenny with
whom I’ve shared countless chats, after works and office lunches. And thank you to
Adrian, Mattias N, Rita, Teodora, Alex, Georg and Micke L for being nice, supportive
and providing advice to a junior colleague.
I am so grateful for all my intelligent, strong and inspiring friends outside econ
academia. Elin, Eva, Theresa, Karin, Kristina, Tove, Malin and Gabriella; thank you
for your patience, for listening, for cheering me on and for all the good food, drinks
and lovely hang-outs along the way. I love you!
Last but far from least, I would like to thank my family. Lennart and Jeanette
for making me a part of the Willstedts, being very understanding of our way too
infrequent visits and for raising an extraordinary son. My parents, for raising me to be
independent, curious and confident. You are the best support system I could ever wish
for and I am so grateful that my brother and I grew up in such a loving home where
we always knew that you are proud of us. Thank you Oscar, for being the best brother
one could wish for. I especially want to thank you and Angelica for all your support
and encouragement during my first year in the program when I really needed it. My
extended family, the wonderful Winklers, thank you for being my second home, my
fan club and for giving me the privilege of having Kevin, Elvis and Otis in my life.
As always in big moments like this, I also send a thought to my late and much-loved
grandmother Eva.
Finally, I would like to thank my husband Robert, the love of my life. Even though
you have a lot of responsibilities of your own you have always put us first and stand
firmly by my side, helping and supporting me in all ways possible. Thank you for
loving me as much as I love you, for making me laugh all the time, for always believing
in me and for giving me freedom. You make my life so much bigger.
Stockholm, November 2015
Gabriella Chirico Willstedt
Contents
Introduction
........................................................................................................
13
Essay 1. Socioeconomic status and health investments over the life-cycle . . . . . . . . . . .
1.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2
Related literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3
Empirical setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4
Descriptive analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.5
Empirical modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.6
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.7
Concluding discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Appendix A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21
22
24
31
33
37
39
45
50
51
Essay 2. Price competition in Swedish dental care . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2
Competition in health care . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3
Theoretical framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4
Institutional setting and data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5
Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.6
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.7
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Appendix B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
59
60
62
64
67
72
73
79
83
84
Essay 3. Redistribution in the Swedish dental care insurance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.2
Related literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.3
Institutional setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.4
Methodological framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.5
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.6
Empirical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.7
Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Appendix C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
Introduction
“Life is better now than almost any time in history.”
– Angus Deaton, The Great Escape.
The quote is the opening line from Angus Deaton’s book “The Great Escape: Health,
Wealth and the Origins of Inequality” (2013a) telling the story of why our lives are
longer, healthier and wealthier than it used to be. Deaton also tells the story of those
who are left behind as “the tale of progress is also the tale of inequality” (Deaton,
2013a, p. xii). Throughout history, the “great escapes” from poverty and depravation
has made nations prosper but at the same time inequalities have increased, not least in
terms of health. As Deaton has noted elsewhere, “when health improvements come
through innovation and new knowledge, the first beneficiaries are likely to be those
with the understanding and wherewithal to adopt them, which will usually be the better
educated and better off” (Deaton, 2013b, p. 265).
Economists’ interest in health can somewhat simplified be understood from two
perspectives. The first is that health is an important part of our wellbeing, not just
because we enjoy being healthy but also because health is part of our human capital and
thus affects our capability of leading the kind of life we want to lead. Health is therefore
closely related to lifetime opportunities and consequently affects the distribution of
welfare. As economists generally are interested in human progress and prosperity (or
most definitely should be!) it is quite natural to also focus on the determinants of health.
And, “perhaps surprisingly”, as suggested by Glied and Smith (2011), economics as
a discipline has made important contributions to our understanding of what makes
people healthy. From Grossman’s (1972b) theoretical model of health production to
the fairly recent contributions on the long-term health effects of early life conditions
made by e.g. Almond and Currie1 and Cunha and Heckman2 to mention a few.
The second, and clearly interrelated, perspective is that the health sector makes
up a large part of developed economies today. In the Netherlands, France, Belgium,
Germany, Denmark and Austria health care spending exceeded 10 percent as a share of
GDP in 2012 (Eurostat, 2015). The corresponding figure for Sweden was 9.6 percent
(OECD, 2014). While average growth in health care spending is decreasing, the
demographic trend with more people living longer will most likely place a tremendous
strain on the systems for financing and providing care services in the future. This, in
turn, poses a challenge for governments to increase efficiency in the welfare systems
at large. In the light of these developments, many OECD countries have taken steps
towards a more market-oriented health service sector, which has been studied carefully
by industrial organization economists.
This thesis consists of three self-contained essays dealing with different aspects
of the economics of health from the perspectives discussed above. The essays are
1 Some of which are reviewed in Currie and Almond (2011).
2 Outlined briefly in Heckman (2007).
13
focused around the themes demand (essay 1), competition (essay 2) and redistribution
(essay 3) and the empirical application throughout the thesis is dental care in Sweden.
All three essays study behaviors, either on behalf of the individual as a consumer of
dental care and producer of health, or on behalf of the supplier of care services. Essay
1 investigates the relationship between income, as one dimension of socioeconomic
status, and investments in health in the form of dental care consumption. The essay
relates to the bigger question about how we can understand the well-documented
positive relationship between markers of socioeconomic status and health—the social
health gradient—resulting in the inequalities addressed by Deaton (2013a). Essay 3
also deals with the relationship between income and dental care consumption, but
from the perspective of what this relationship implies for redistribution within the
Swedish dental care insurance. The starting point is the theoretical concept of public
health insurance as a means of redistribution and thus a potential way of alleviating
the consequences of the social health gradient. Essay 2 turns to the supply side and
studies price competition on the Swedish dental care market as a way of increasing our
understanding of health service markets.
I find dental care and oral health to be an interesting application for investigating all
these behaviors. First, there is no reason to believe that disparities in oral health related
to socioeconomic status should differ systematically from disparities in general health.
Second, dental care is a suitable application for studying individual behavior in relation
to health as oral diseases—to a large extent—can be avoided by preventive measures
(Coulter et al., 1994). Dental care consumption therefore constitutes an important
part of maintaining a good oral health. Third, the Swedish dental care market is a
suitable setting for investigating the effects of competition in a health service market
as patients have a substantial cost share and prices are market determined. This
institutional setting differs from the US hospital market—the most active area in the
literature on competition in health care—where e.g. prices are set through hospitalinsurer contracts.3 Fourth, the Swedish dental care insurance is an interesting setting
for studying redistributional features within a public health insurance as the insurance
subsidizes care that is predominantly supplied through private clinics and patients, as
mentioned, make fairly large co-payments.
The social health gradient
The social health gradient, i.e. the positive relationship between health outcomes and
markers of socioeconomic status, is a salient feature of all developed countries. Cutler
et al. (2011) find that household income protects against 5-year mortality in all agegroups among US adults over the age of 25. In addition, Case and Deaton (2005) find
that both men and women in the bottom income quartile in the US are significantly
more likely to report being in poor health compared to their wealthier counterparts.
Case and Deaton (2005) also finds that self-reported health among those in the bottom
income quartile deteriorates faster compared to those in the top which results in a
widening of the income gap in health over the life-cycle. The patterns are similar in
Europe. A consortium on health inequalities in the EU (European Union, 2013) found
3 See e.g. Gaynor and Town (2011) for a survey of the literature on competition in health care
markets.
14
a steep social gradient between material deprivation and adverse health outcomes in
2010. Self-reported poor health or long-standing health problems become increasingly
more frequent when moving down in the distribution of socioeconomic status, whether
measured by education, income or material deprivation. Maskileyson (2014) reports
results for Sweden and finds a positive and statistically significant association between
family wealth and health status measured by a severity-weighted index of several
self-reported measures. The size of the association is not statistically different from
estimates for Germany, the Czech Republic, the UK or Israel, suggesting similar social
health gradients despite substantially different welfare-regimes.
A large body of literature has emerged in economics during the past decades, focusing on the evolution of human capabilities throughout life as a way of understanding
the origins of socioeconomic inequalities (outlined in e.g. Cunha et al., 2010). This
research agenda is motivated by evidence of early life conditions having a substantial
impact on health and economic outcomes later in life.4 One dimension of the pathway
from conditions in the beginning of life to adult health is the effect on subsequent biology; it is now widely recognized that e.g. coronary heart disease and type 2 diabetes,
in part, is determined by gene-environment interactions early in life (Gluckman and
Hanson, 2006). Another dimension is behavioral and builds on the notion that the
same processes that form human capabilities and ultimately socioeconomic status also
form individual’s health behaviors. Heckman (2007) reviews the evidence of early life
conditions affecting the formation of capabilities in childhood which, in turn, affects
skill-attainment later in life. The process is dynamic in the sense that skills and abilities
attained early in life serve as a “production technology” to attain further capabilities
later in life and therefore “skill beget skill”. In addition, Cunha and Heckman (2007)
suggests that individual’s stocks of skills “facilitate the accumulation of health capital
through self-regulation and choices” (Cunha and Heckman, 2007, p. 45). An important
insight from this literature is therefore that the emergence of the social health gradient
should be understood from a life-cycle perspective and that disadvantages, originating
in early life conditions, may be amplified by differences in behaviors. However, little
is known about the empirical significance of these differences as health behaviors are
generally not observed by researchers.
In Essay 1, I utilize the detailed Swedish register data on dental care consumption
and oral health to empirically investigate the significance of socioeconomic differences
in health behaviors. The concepts of the Grossman model (1972a, 1972b) serves as
a theoretical framework for the empirical analysis. The model introduced the notion
of health as a capital stock. In order to maintain or increase the size of the stock,
individuals can invest in their health by undertaking any health promoting behavior. I
use dental care consumption as an operationalization of health investments and focus
on adults aged 35-64 years, an age where most have completed their education and
settled into a career path. This suggests that lifetime profiles in terms of socioeconomic
status (SES) can be taken as given and be measured by lifetime incomes. The first part
of the empirical analysis estimates the relationship between lifetime income and oral
4 For reviews of the literature on the effects of early childhood and in utero environments
on human capital and skill formation see e.g. Currie (2009), Almond and Currie (2011),
Currie and Almond (2011) and Cunha et al. (2006). The idea of long-term effects of in utero
environments originates in epidemiology and is sometimes referred to as the “fetal origins
hypothesis”, or the “Barker hypothesis” after Professor David J. P. Barker.
15
health and dental care consumption respectively. The analysis is performed separately
by age-groups to take the age-related depreciation of health into account, and highlights
four salient patterns. First, in line with Grossman’s model, oral health decreases while
dental care consumption increases with age. The Grossman model (2000) assumes
that individuals’ stocks of health deteriorate faster as they grow older which is to say
that older individuals, on average, are less healthy than younger individuals, all else
equal. At the same time, because of the increasing depreciation of health, individuals
need to make relatively larger health investments to maintain the health stock. Age
therefore affects the demand for health and investments in opposite directions; the
demand for health decreases with age whereas the demand for health investments
increases. Second, oral health decreases faster for individuals in the bottom quartile of
the income distribution compared to those in the top. Third, individuals with higher
incomes consume more dental care at all ages. Finally, the income gap in both oral
health and dental care consumption widens substantially as individuals grow older.
In the second part of the analysis, I estimate the effect of lifetime incomes on
dental care consumption and condition the analysis on oral health. This is done
to account for the inherently dynamic process of accumulating oral health capital;
dental care consumption affects oral health which affects consumption and so on. The
results paint a clear picture of how differences in dental care consumption related to
socioeconomic status increases monotonically over the life-cycle. The lifetime patterns
are striking; from midlife to the years before the legal retirement age, the estimated
effect of lifetime income on dental care consumption increases by a factor of 4.8
for men and 2.7 for women. Considering that dental care consumption is only one
dimension of individual’s health behaviors, the results indicate that lifetime effects of
socioeconomic status can be even larger when considering other dimensions.
While I cannot determine the precise mechanisms giving rise to the documented
gradient in oral health, the results support the notion that differences in health behaviors
may account for part of it. Taken together with the general conclusion from the
literature on human capability formation that early life conditions have a significant
effect on both socioeconomic status and health, the results suggest that policies aimed
at addressing inequalities in lifetime opportunities in general are most likely to succeed.
This argument is reinforced by the literature5 that exploits wealth shocks in the form of
e.g. lottery winnings, and generally find no causal effect of income as such on health.
The market as a way of supplying care
Many OECD countries have undertaken market-oriented health care reforms over
the past decades (Gaynor, 2012). Some reforms have been motivated as a way of
increasing consumer choice. Others have been motivated as a means of cost control
since introducing market mechanisms into the provision of health care is assumed
to strengthen incentives for providers to become more efficient (Docteur and Oxley,
2003). At the same time, it is often argued that the special features of health care have
implications for the scope for competition. The development towards market-oriented
reforms therefore raises questions about when and how competition works on health
5 See e.g. Apouey and Clark (2014), van Kippersluis and Galama (2014) and Cesarini et al.
(2015).
16
care markets. In Essay 2, I study the effect of competition on prices on the Swedish
market for dental care, a setting where patients have a substantial cost share (on average
about 80%) and prices for health care services are market determined.
The empirical strategy exploits that competition—theoretically—has different effects across service types due to differences in consumers’ price sensitivity. More
precisely, competition has greater effects on informative services, such as examinations and diagnostics, compared to more complicated, therapeutic services. By
exploiting this difference, a relative effect of competition on prices is estimated. The
reduced form results show small but statistically significant effects of competition on
prices for first-stage services relative to follow-on services. Competition is measured
as the number of clinics within a fixed distance from each clinic. The main results
suggest that a 1% increase in the number of clinics is followed by a 0.024% decrease in
prices for basic examination and diagnostics relative to tooth extractions. The results
are robust across analyses of different kinds of services and model specifications. The
effects are small, but should be interpreted as short-term effects of increased competition. Consequently, substantial effects of competition on prices over time cannot be
ruled out.
A possible concern with the result is that clinics, in response to increased competition, lower prices for informative services while increasing prices for therapeutic
services. To investigate whether this is the case, the absolute effect of competition is
assessed through simulations, based on the estimates of the relative effect. The simulations suggest that the absolute effect of competition on prices is negative for both
informative and therapeutic services. Hence, the results suggest that there is room for
price decreases in the Swedish market for dental care and that increased competition
would imply a redistribution of welfare from sellers to buyers by lowering prices.
The conclusion is that increased competition is followed by an increased price
difference between informative and therapeutic services. The results therefore provide
evidence of strategic behavior of clinics in the sense that their price-setting indeed
responds to changes in the market environment. The effects are small, but substantial
effects of competition on prices over time cannot be ruled out. The assessment
of the absolute effects suggests that increased competition lowers prices for both
types of services. Furthermore, the simulations do not suggest that clinics fully
compensate price decreases for examinations and diagnostics with price increases for
more complicated and uncommon services. This implies that competition increases
welfare for consumers in the short run. However, the increase is moderate. All results
are statistically significant and robust for sensitivity analyses.
Public health insurance as a means of redistribution
Public health insurance systems are commonly justified on grounds of equity and
as a means of redistributing welfare from high-income individuals to low-income
individuals. The case for public health insurance as an instrument for redistribution
has been studied carefully theoretically. Blomqvist and Horn (1984), Rochet (1991),
Cremer and Pestieau (1996) find that health insurance, as a complement to income
taxation, can achieve redistribution more efficiently than distortionary income taxes
alone. This result is developed in models where income is assumed to be negatively
17
related to health risks. While this assumption is empirically sound, the argument
abstracts from a setting where the care that is covered by the insurance is more
than a “repair technology”, i.e. has value for the individual beyond its restoring
effect on health. In fact, dental care consumption increases with incomes whereas
individuals with higher incomes also tend to have better oral health, on average. Essay
3 investigates the redistributional features of the Swedish dental care insurance.
The insurance includes cost-sharing subsidies for dental care costs above a threshold,
at two different rates depending on the value of consumed care over a given period.
The size of the subsidy increases with dental care costs implying that the generosity
of the insurance increases with consumption. The insurance constitutes a small part
of the relatively large Swedish social insurance system and, therefore, redistribution
within the dental care insurance is defined as whether or not it adds to the progressive
redistribution taking place through other parts of the social insurance.
The empirical analysis is guided by Grossman’s framework (1972a, 1972b, 2000)
in which individuals are assumed to demand dental care because it serves as an input
in the production of oral health. Given that oral health is a normal good, the demand
for both oral health and dental care increases with incomes. The Grossman model
therefore provides a theoretical foundation for the documented positive relationship
between income and oral health as well as investments in oral health in the form of
dental care consumption.
The empirical strategy exploits that the insurance scheme consists of three different segments; one where individuals pay market prices and two where dental care
is subsidized. Consumption below the first threshold provides information on how
consumption varies with income under market prices. Consumption above this threshold also provides information on how consumption varies with income, but under
subsidized prices. The empirical analysis investigates if the consumption response
to the insurance, i.e. the lowered price, varies with income. The analysis covers all
individuals in Sweden aged 35-89 years with positive dental care consumption during
the period July 2008-December 2011.
I find that individuals with different incomes respond differently to the insurance.
For individuals aged 35-59 years, the dental care insurance adds to the progressive
redistribution of the social insurance system; as we move up in the income distribution,
individuals move from the consumption segment where dental care is subsidized to the
segment where they pay market prices. Higher-income individuals aged 35-59 year
are thus more likely to bear the full cost of their dental care consumption compared
to lower-income individuals. At the same time, the dental care insurance reduces
the progressivity in the social insurance system for those aged 60-89 years. This is
problematic from an equity point of view as the positive relationship between income
and oral health is stronger among those aged 65 years or above. Moreover, it is
reasonable to assume that the oral health stock deteriorates faster as an individual
age suggesting that older individuals will have a greater need for dental care. It is
however noteworthy that the dental care insurance makes redistribution through the
Swedish social insurance system more progressive for the vast majority of patients.
This finding may be surprising to some, as dental care in Sweden stands out in the
welfare sector by having pronounced market elements such as free price-setting and
fairly large private co-payments and such arrangements are often questioned on the
basis of equity concerns.
18
References
Almond, D. and Currie, J. (2011). Killing Me Softly: The Fetal Origins Hypothesis.
The Journal of Economic Perspectives, 25 (3), 153–172.
Apouey, B. and Clark, A. E. (2014). Winning big but feeling no better? The effect
of lottery prizes on physical and mental health. Health Economics.
Blomqvist, Å. and Horn, H. (1984). Public health insurance and optimal income
taxation. Journal of Public Economics, 24 (3), 353–371.
Case, A. and Deaton, A. S. (2005). Broken Down by Work and Sex: How Our
Health Declines. In Analyses in the Economics of Aging, University of Chicago
Press, pp. 185–212.
Cesarini, D., Lindqvist, E., Östling, R. and Wallace, B. (2015). Wealth, Health,
and Child Development: Evidence from Administrative Data on Swedish Lottery
Players. IFN Working Paper No. 1060.
Coulter, I. D., Marcus, M. and Atchison, K. A. (1994). Measuring oral health
status: Theoretical and methodological challenges. Social Science & Medicine,
38 (11), 1531–1541.
Cremer, H. and Pestieau, P. (1996). Redistributive taxation and social insurance.
International Tax and Public Finance, 3 (3), 281–295.
Cunha, F. and Heckman, J. J. (2007). The Technology of Skill Formation. The
American Economic Review, 97 (2), 31–47.
—, —, Lochner, L. and Masterov, D. V. (2006). Interpreting the evidence on life
cycle skill formation. Handbook of the Economics of Education, 1, 697–812.
—, — and Schennach, S. M. (2010). Estimating the technology of cognitive and
noncognitive skill formation. Econometrica, 78 (3), 883–931.
Currie, J. (2009). Healthy, wealthy, and wise: Is there a causal relationship between
child health and human capital development? Journal of Economic Literature,
47 (1), 87–122.
— and Almond, D. (2011). Chapter 15 - Human capital development before age five.
In D. C. a. O. Ashenfelter (ed.), Handbook of Labor Economics, vol. 4, Part B,
Elsevier, pp. 1315–1486.
Cutler, D. M., Lleras-Muney, A. and Vogl, T. (2011). Socioeconomic Status and
Health: Dimensions and Mechanisms. In S. Glied and P. C. Smith (eds.), The
Oxford Handbook of Health Economics, Oxford : New York: Oxford University
Press.
Deaton, A. (2013a). The Great Escape: Health, Wealth, and the Origins of
Inequality. Princeton: Princeton University Press.
— (2013b). What does the Empirical Evidence on SES and Health Tell Us About
Inequity and About Policy? In N. M. Eyal, S. A. Hurst, O. F. Norheim and
D. Wikler (eds.), Inequalities in health: concepts, measures, and ethics,
Population-level bioethics, Oxford: Oxford University Press.
Docteur, E. and Oxley, H. (2003). Health-Care Systems: Lessons from the Reform
Experience. OECD Health Working Papers No. 9, OECD.
European Union (2013). Health inequalities in the EU final report of a consortium Consortium lead: Sir Michael Marmot.
Eurostat (2015). Healthcare statistics.
http://ec.europa.eu/eurostat/statistics-explained/index.php/
Healthcare_statistics#Healthcare_expenditure.
19
Gaynor, M. (2012). Reform, Competition, and Policy in Hospital Markets. Working
Party No. 2 on Competition and Regulation, OECD.
— and Town, R. J. (2011). Competition in Health Care Markets. Handbook of Health
Economics, vol. 2, Elsevier, pp. 499 – 637.
Glied, S. and Smith, A. (2011). Introduction. In S. Glied and P. C. Smith (eds.), The
Oxford Handbook of Health Economics, Oxford : New York: Oxford University
Press.
Gluckman, P. and Hanson, M. (eds.) (2006). Developmental Origins of Health and
Disease. Cambridge: Cambridge University Press.
Grossman, M. (1972a). The Demand for Health: A Theoretical and Empirical
Investigation. New York: NBER, Occasional Paper 119, Columbia Press.
— (1972b). On the concept of health capital and the demand for health. The Journal
of Political Economy, 80 (2), 223.
— (2000). Chapter 7 The human capital model. In A. J. Culyer and J. P. Newhouse
(eds.), Handbook of Health Economics, vol. 1, Part A, Elsevier, pp. 347–408.
Heckman, J. J. (2007). The economics, technology, and neuroscience of human
capability formation. Proceedings of the National Academy of Sciences, 104 (33),
13250–13255.
Maskileyson, D. (2014). Healthcare system and the wealth–health gradient: A
comparative study of older populations in six countries. Social Science &
Medicine, 119, 18–26.
OECD (2014). OECD Health Statistics 2014. How does Sweden compare? http://
www.oecd.org/els/health-systems/Briefing-Note-SWEDEN-2014.pdf.
Rochet, J.-C. (1991). Incentives, redistribution and social insurance. The Geneva
Papers on Risk and Insurance Theory, 16 (2), 143–165.
van Kippersluis, H. and Galama, T. J. (2014). Wealth and health behavior: Testing
the concept of a health cost. European Economic Review, 72, 197–220.
20
Essay 1.
Socioeconomic status and health investments
over the life-cycle
Abstract Individuals with higher socioeconomic status (SES) also tend to enjoy
better health. Evidence from the economics literature suggests that a potential
mechanism behind this “social health gradient” is that human capabilities, that
form SES, also facilitate health-promoting behaviors. This essay empirically
investigates the significance of socioeconomic differences in health behaviors,
using dental care consumption as an operationalization of health investments. I
focus on adults at an age where lifetime trajectories for SES can be taken as given
and use lifetime income to capture SES. I estimate the impact of lifetime income
on dental care consumption and find robust evidence that the social gradient in
dental care consumption steepens dramatically over the life-cycle. Considering
that dental care consumption only reflects a small part of individuals’ health
investments the results suggest that lifetime effects of SES on health behaviors
could be substantial in other dimensions.
I would like to thank Per Johansson, Erik Grönqvist, Erik Lindqvist and Douglas Lundin for
careful reading and very valuable feedback. I would also like to thank participants at the 36th
Nordic Health Economists’ Study Group meeting (NHESG) as well as seminar participants at
the The Research Institute of Industrial Economics (IFN) for helpful comments.
21
1.1 Introduction
Individuals with higher socioeconomic status also tend to enjoy better health. This is
true across countries and across individuals within countries (Deaton, 2003, Currie,
2009, Cutler et al., 2011). While the positive correlation between health and markers
of socioeconomic status—the social health gradient— is documented in an extensive
body of literature, there is no general consensus about its determinants. Over the past
decade, a vivid strand of literature1 in economics has evolved, focusing on the evolution of human capabilities as a way of understanding the origins of socioeconomic
inequalities. This literature provides evidence that early life conditions have a substantial impact on health and economic outcomes in adulthood and that the negative effects
of early life disadvantages accumulate over the course of life.2 The long-term effects
are partly explained by the ways in which early life health insults affect subsequent
biology3 and partly by the effect of early life conditions on the dynamic formation of
human capabilities throughout life. This is because skill attainment in early childhood
promotes skill attainment later in life which forms the stock of skills and abilities
in adulthood and, ultimately, socioeconomic status. In addition, individual’s stocks
of “cognitive and non-cognitive skills facilitate the accumulation of health capital
through self-regulation and choices” (Cunha and Heckman, 2007, p. 45). This suggests that the same processes that form human capabilities and socioeconomic status
also form individual’s health behaviors. The social health gradient should therefore be
understood from a life-cycle perspective and, in part, as a result of the ways in which
capabilities shape behaviors. However, health behaviors are generally unobservable
for researchers and therefore the processes through which socioeconomic differences
in health are formed is largely an open question.
In this essay, I utilize exceptionally detailed Swedish register data and take on the
behavioral dimension of the formation of health by estimating the effect of socioeconomic status (SES) on health investment decisions in adults. I restrict the analysis to
individuals aged 35-64 years, an age where most have completed their education and
settled into a career path.4 Individuals’ lifetime profiles in terms of SES can therefore
be taken as given, suggesting that the direction of causality is clear. This argument
is strengthened by using dental care consumption as the operationalization of health
investments; dental care consumption in adulthood arguably has negligible (if any)
effect on the formation of human capabilities. Individual’s SES is operationalized by
lifetime incomes.
The theoretical foundation underlying both this essay and the literature on the
formation of human capabilities is largely based on Grossman’s (1972a, 1972b) seminal
1
2
3
4
This literature is briefly outlined in (Heckman, 2007).
For reviews of the literature on the effects of early childhood and in utero environments
on human capital and skill formation see e.g. Currie (2009), Almond and Currie (2011),
Currie and Almond (2011) and Cunha et al. (2006). The idea of long-term effects of in utero
environments originates in epidemiology and is sometimes referred to as the “fetal origins
hypothesis”, or the “Barker hypothesis” after Professor David J. P. Barker.
This has been explored in other disciplines, predominantly epidemiology. It is now widely
recognized that e.g. coronary heart disease and type 2 diabetes, in part, is determined by geneenvironment interactions early in life (Gluckman and Hanson, 2006a). See also Gluckman
and Hanson (2006b) for an excellent review of this literature.
In 2007, the average graduation age for university students in Sweden was over 29 years
(Uusitalo, 2011).
22
health production model which introduced the notion of health as a capital stock that
can be built up by investments. The capability formation literature considers a broad
spectrum of investments in human capabilities whereas Grossman focuses on health
investment decisions in adults. Investments in Grossman’s model are produced by
any health-promoting behavior which includes the consumption of dental care and
other health care services. Current health is therefore, in part, determined by choices
regarding health behaviors throughout life.
The empirical analysis builds on individual-level register data on oral health, dental
care consumption and socioeconomic variables covering over 3,8 million adults in
Sweden. Dental care is a particularly suitable application for studying the consumption dimension of individual’s health investment for several reasons. First, utilizing
dental care regularly is an important part of maintaining a good oral health since oral
diseases—to a large extent—can be avoided by preventive measures (Coulter et al.,
1994). The consumption motive is therefore more apparent for dental care than it is
for e.g. health care in general. Second, there is a strong link between consumer choice
and utilization in dental care since out-of-pocket payments in dental care are generally
high while rationing is low.5 Finally, the oral health measure compares favorably to the
literature on disparities in general health which, to a large extent, employs mortality or
self-reported health status (SRHS). While death undeniably is an objective measure,
it is a blunt one and, as pointed out by Case and Deaton (2005), it does not allow investigating the dynamic process of accumulating health over the life cycle. SRHS are
arguably more nuanced but have the inherent risk of being endogenous as the subjective response, conditional on objective health, may be correlated with socioeconomic
status.
The Grossman model (2000) assumes that the health stock deteriorates faster as
individuals grow older implying that individuals’ health stocks decline with age. At
the same time, because of the increasing depreciation rate, relatively more health
investments will be required to maintain the health stock. Age therefore affects the
demand for health and investments in opposite directions; the demand for health
decreases with age whereas the demand for health investments increases. The empirical
analysis is performed separately by age-groups to account for the age effects on oral
health and health behaviors.
An introductory, descriptive analysis highlights four patterns. First, in line with
the Grossman model, oral health decreases while dental care consumption increases
with age. Second, oral health decreases faster for individuals in the bottom quartile
of the income distribution compared to those in the top. Third, individuals with
higher incomes consume more dental care at all ages. Finally, the income gap in
both oral health and dental care consumption widens substantially as individuals grow
older. In the second part of the analysis I estimate the effect of income on dental care
consumption by conditioning on oral health. This is done to account for the inherently
dynamic process of accumulating oral health capital; dental care consumption affects
oral health which affects consumption and so on. Failing to account for individual’s
5
There is larger variation in public financing schemes for dental care than for health care in
general. For example, adult dental care is not at all publicly financed in the Netherlands or
Switzerland. Residents in England pay above £200 for a full dental care service and in France,
co-insurance rates varies from 30-90% depending on the type of dental care services (van der
Wees et al., 2015).
23
oral health underestimates the effect of income on dental care consumption as it
neglects that individuals with higher lifetime incomes generally will have consumed
more dental care in the past, and thus have better oral health. This is reflected in the
empirical analysis; the estimated average effect of income on dental care consumption
doubles when conditioning on oral health. The argument that controlling for oral
health solves the omitted variable bias is strengthened by the finding that the effect of
income increases even more when only considering consumption of non-acute dental
care.
The empirical analysis paints a clear picture of how differences in dental care
consumption related to socioeconomic status increases monotonically over the lifecycle. The results are striking; from midlife to the years before the legal retirement age
the estimated effect increases by a factor of 4.8 for men and 2.7 for women. This overall
pattern is robust when controlling for oral health non-parametrically and when only
considering non-acute dental care. Considering that dental care consumption is only
one dimension of individual’s health behaviors, the results indicate that lifetime effects
of socioeconomic status on health behaviors could be substantial in other dimensions.
While the empirical analysis cannot identify the precise mechanism that forms the
documented social gradient in oral health, the results indicate that disparities may
be amplified by differences in health behaviors. In addition, the age-patterns of the
effect of income on dental care consumption mimics the widening of the social health
gradient over the life-cycle, documented in e.g. Smith (2007), Baum and Ruhm (2009)
and Islam et al. (2010).
The rest of the paper is organized as follows. In section 1.2, I review the literature
most closely related to the paper, starting out with the concepts of the Grossman
model and the evidence on the empirical importance of these concepts. I then discuss
previous findings on the relation between socioeconomic status and health and the
relation between income and consumption of health care. Section 1.3 outlines the
empirical setting and presents the data. Section 1.4 presents the descriptive analysis.
In section 1.5, I discuss the empirical modelling and section 1.6 presents the results. I
conclude with a discussion in section 1.7.
1.2 Related literature
1.2.1 The Grossman model as a conceptual framework
Grossman’s seminal work on the human capital model of health (first developed in
Grossman, 1972a,b) serves as a conceptual framework for understanding health investment decisions in adults. In the Grossman model health is viewed as a capital
stock which, together with individual’s abilities and skills, makes up an individual’s
human capital. The health stock depreciates over the life cycle and can be augmented
by investments. Drawing on Becker’s (1965) household production approach, individuals are considered as producing health, alongside the commodity “consumption”, by
inputs of market goods and own time. This is not to say that individuals have perfect
control over how their health evolves, but rather that these inputs are systematically
related to their health (Zweifel et al., 2009). Individuals are assumed to demand health
for its “consumption benefits” (they enjoy being healthy) and for its “investment benefits” (being healthy increases productivity). It is an innocuous assumption that general
24
health increases productivity which implies that an important motivation to invest in
general health is to reduce sick-time, i.e. time spent away from market and non-market
activities. However, the effect of oral health on productivity is arguable negligible6
suggesting that individuals consume dental care because of the consumption benefits
of oral health.
Grossman (1972a, 1972b, 2000) sets up the individual’s problem as an optimization
of lifetime utility which, in turn, is an increasing function in health and consumption.
The evolution of the health stock over time is determined by the process by which
investments are translated into health:
Ht+1 = It + (1 − δt )Ht ,
(1.1)
where δt is depreciation and It is gross investments at age t. Note that if It = δt ,
individuals are simply maintaining their health stock, which also implies that perfect
repair of the health stock is theoretically possible.7 Any health-promoting goods or
services can be considered as inputs in producing health investments. Considering
dental care as an input in producing oral health leads to the notion of dental care
demand as a derived demand; individuals demand good oral health, not dental care
per se.
Turning to resources, individuals are constrained by their lifetime budget, implying
that the present value of spending on health investments cannot exceed the present
value of their lifetime wealth, and time spent on investing in health cannot exceed the
total amount of time available. The equilibrium quantity of health, i.e. the optimal
health stock, is found by maximizing lifetime utility subject to (1.1) and the lifetime
budget constraint. One of the most central results of the Grossman model is simply that
individuals will invest in their health stock up to the point where marginal consumption
and production benefits of health equals the cost of an additional unit of health capital.
Marginal benefits of health are assumed to be diminishing, implying that healthy
individuals will place less value on an additional unit of health compared to less healthy
individuals ceteris paribus. The marginal cost is a function of the money price and
opportunity cost of investing in health, and depreciation. The rate of depreciation
is generally assumed to be exogenous and to increase with age, implying that the
health stock not only decreases over time but at an increasing rate (see e.g. Grossman,
2000). This leads to two predictions regarding age-effects. First, the demand for health
decreases over the life cycle since the marginal cost of holding an additional unit of
health stock increases with age ceteris paribus. Therefore, the optimal stock of health
falls with age, which is just to say that older individuals will have smaller stocks of
health. Second, the demand for health investments, in this case dental care, increases
with age. This is because it will take more dental care to compensate for the loss of
oral health as the depreciation rate increases.8 Naturally, the oral health stock will also
depreciate randomly due to illness.
6
7
8
In general, bad oral health does not impede the ability to work.
Individuals are assumed to have an inherited health stock, H0 , at t = 0.
An increase in δt lowers the supplied amount of health capital from a given gross investment.
If the decrease in supply is larger than the decrease in demand for health, individuals will
want to close the gap by consuming more health care. As discussed in (Grossman, 2000,
p. 367-370), this is true under plausible conditions.
25
1.2.2 Empirical tests of the Grossman model
There are some empirical tests of the Grossman model, most of which are based on
various cross sectional surveys on either the individual or the household level.9 The
results provide mixed support for the theoretical predictions which potentially can
be explained by the lack of adequate data with respect to the dynamic nature of the
model. Zweifel et al. (2009) argues that it is particularly problematic that e.g. Wagstaff
(1986) using the Danish Welfare Survey finds health, approximated by questions on
non-chronic health problems, to be negatively related to the demand for health care.
However, an important limitation of using cross-sectional survey data is that it cannot
address the problem of reverse causality, i.e. that individuals with poorer health
consume more health care. To appropriately capture the dynamic relationship between
health and health care consumption one therefore needs to observe individuals over
time. Wagstaff (1993) therefore argues that his and other early empirical tests of the
Grossman model should not necessarily be interpreted as contradicting the model’s
predictions. Moreover, there is quasi-experimental empirical evidence suggesting that
greater amounts of health care indeed improves health.10
Wagstaff (1993) also points out that the proper way to account for the age-related
depreciation rate would be to estimate separate models for different age-groups. However, to my knowledge this has so far only been done in Wagstaff (1993) where he fits
an empirical formulation of the Grossman model separately for two groups consisting
of adults under the age of 41 and adults aged 41 or above respectively.11 Contrary to
Grossman’s predictions, he finds that the depreciation of health is larger in the younger
age-group and that the depreciation rate decreases with age. Wagstaff (1993) argues
that his may be due to how the model is specified.12
Zweifel and Nocera (1998) test the Grossman model using panel data and find
evidence supporting that age reduces the demand for health, but that the negative effect
of age on health, approximated by subjective health status, can be counteracted with a
healthy lifestyle. They also find that age increases the demand for health care and that
health care and health are positively related.
1.2.3 Socioeconomic status and health
The Grossman framework can be used for understanding the process by which health
capital is accumulated over the life cycle, but it does not explicitly analyze socioeconomic disparities in health. Grossman (2000) does however highlight human capital,
9
See e.g. Grossman (1972a), Cropper (1981), Wagstaff (1986), Wagstaff (1993), Zweifel and
Nocera (1998) and Gerdtham et al. (1999).
10
Card et al. (2009) exploit that eligibility for Medicare at age 65 in the United States introduces
a discontinuity in the use of hospital services while the underlying severity of illness is similar
on both sides of the threshold. They find a 20 percent reduction in deaths for those aged 65
years compared to those just below the Medicare threshold. Doyle (2011) finds that individuals
visiting Florida, who fall ill in areas with high levels of health care spending have significantly
lower mortality rates compared to visitors falling ill in lower spending areas.
11
Wagstaff (1993) uses the Danish Health Study following a 1000 households during twelve
months starting 1982.
12
The depreciation rate enters both as a linear function and as a step-function of age. Wagstaff
(1993) argues that a more consistent treatment would be to specify the age effect as a step
function only.
26
defined as primarily being determined by years of formal schooling, as a factor that
makes individuals more efficient in their production of health.13 This can be related
to the literature on the formation of human capital and Cunha et al.’s (2006) model in
which investments in human capabilities early in life, e.g. education, raises skill attainment later in life. Cunha et al. (2006) use the term skills in a broad sense, including
non-cognitive skills or abilities such as self-discipline and time-preferences. Hence,
a “capability formation perspective” is that the stock of skills affect the accumulation
of health capital through their effect on choices (Cunha and Heckman, 2007). These
choices may include lifestyle factors, such as smoking as shown by Heckman et al.
(2006), and health investment decisions. Goldman and Smith (2005) finds differences
in self-management of Type 1 diabetes between individuals with high and low educational levels. Individual’s stock of cognitive and non-cognitive skills is also an
important determinant of e.g. scholastic achievement and earnings (Heckman, 2007,
Cunha et al., 2006) which form individual’s lifetime trajectories for socioeconomic
status. Combining this perspective with the Grossman framework suggests that one
of the potential mechanisms behind the social health gradient is that individuals with
higher socioeconomic status, as a function of acquired capabilities, invest more in
their health. The results in Goldman and Smith (2005) also suggest that one way that
socioeconomic status may influence health is that individuals with more human capital
are better able to treat illnesses by e.g. following a complex medical regimen.
The empirical analyzes in the literature that links the acquisition of human capital to improvements in adult health generally builds on data with health outcomes
recorded at one point in time in adulthood.14 This provides important insights on
the relationship between markers of socioeconomic status and health from a life-cycle
perspective. However, given that health is a stock that evolves throughout life in response to investments and age-related depreciation, it leaves the precise process by
which inequalities in health are formed as a “black box”. This is reasonably not due
to lack of interest but rather a result of restrictions in data; longitudinal data including
both health measures and socioeconomic variables are rare and, in addition, health
behaviors are generally unobservable for researchers. Little is therefore known about
the empirical significance of differences in health behaviors as a cause of the social
health gradient.
Campbell et al. (2014), being somewhat of an exception in the capability formation
literature, use biomedical data to estimate the long-term health effects of the Carolina
Abecedarian Project (ABC). ABC was an educational child-care program in the US
that has been shown to have boosted IQ and academic achievement among treatment
group members (Cunha et al., 2006). Campbell et al. (2014) find that individuals who
were randomly assigned to the treatment also had significantly better physical health in
their mid-30s and that treated males were more likely to be cared for at a hospital or a
doctor’s office when sick. While Campbell et al. (2014) cannot determine what caused
13
It follows from the Grossman model that individuals with more human capital also will be
healthier. However, as more educated individuals produce health more efficiently it does not
necessarily follow that they also demand more health care (Grossman, 2000).
14
Currie and Almond (2011) provide an extensive survey of the literature. Cunha et al. (2006)
review the empirical literature on skill formation and note that evaluations of early interventions
generally do not follow individuals into adulthood. The Perry Preschool Project was an
exception by following participants up to the age of 27. Most evaluations do not follow
participants past adolescence (see table 6 in Cunha et al., 2006).
27
the effects on health, the results indicate that differences in health behavior may account
for part of them. In a recent contribution on the relation between human capabilities
and health, Öhman (2015) use Swedish enlistment data and finds both cognitive and
non-cognitive abilities to be significantly associated with premature mortality in men
aged 44-59 years.15 In addition, Öhman (2015) finds that both types of abilities are
important predictors of mortality within income and educational groups, but more so
for those with low incomes and less than college education.
A closely related literature is that concerned with the broader question of whether
income affects health. Case et al. (2002) assess the origins of the social health gradient
by investigating the relationship between household income and children’s health. The
motivation for looking at children is to eliminate the direct channel that runs form
health to income in adults. Using cross-sectional US data they find that children from
lower-income households enter adulthood with lower socioeconomic status and worse
health compared to children from higher-income households. Part of this is explained
by the finding that children from low-income households are more likely to suffer
from chronic conditions and that they are in poorer health given a chronic condition
compared to children from wealthier households. In addition, the detrimental effect of
low income on health accumulates as children grow older. Case et al. (2002) cannot
identify the precise mechanism through which income protects against the adverse
effects of chronic conditions but are able to rule out insurance, health at birth and
genetics as key factors. Instead, they highlight that wealthier parents may be better
able to invest in their children’s health by e.g. consuming appropriate medical care
or follow treatment regimens. The suggestion of a behavioral dimension underlying
the relation between income and health is supported by the finding that children in
wealthier families are more likely to have a regular bedtime and wearing seat belts
(Case and Paxson, 2002) and that these proxies for health-related behaviors indeed are
significantly associated with better child health (Case et al., 2002).
Case and Deaton (2005) study the evolution of health with respect to age, sex
and socioeconomic status, also using cross-sectional US health survey data. In their
empirical analysis they find—in line with the Grossman model—that average selfreported health declines with age. Self-assessed health is on average worse for women,
but depreciates more slowly with age for women than for men. They also find that
family income has a protective effect on health, that this effect is larger for men than for
women and that the protective effect of income increases over time. The latter implies
that the health gradient widens over the life cycle, which has also been documented
elsewhere. Smith (2007) finds a clear age-pattern in the social health gradient where
income related disparities in self-reported health increases up to the age of 50.16 The
widening of the social health gradient is also documented in e.g. Baum and Ruhm
(2009) and Islam et al. (2010).
A subset of the literature on the effects of income on health exploits plausibly exogenous variation in lifetime incomes. Lindahl (2005), Gardner and Oswald (2007),
15
The enlistment data covers almost 700,000 men, enlisted during a period when military
enlistment was mandatory for all young men in Sweden. Cognitive abilities were tested with
a non-standard IQ test and non-cognitive abilities were assessed through interviews with a
psychologist.
16
Smith (2007) finds that the health gradient slowly begins to fade after the age of 50. However,
up to the end of life, moving down in the income distribution at any age is associated with
worse self-reported health.
28
Apouey and Clark (2014) and Cesarini et al. (2015) use lottery winnings, van Kippersluis and Galama (2014) use lottery winnings and inheritance and Erixson (2014) use
a Swedish inheritance tax reform. Lindahl (2005) stands out in the “lottery literature”
by finding that income increases cause statistically and economically significant health
improvements17 in Sweden. Apouey and Clark (2014) find no effects on self-assessed
health and health problems in Great Britain and Cesarini et al. (2015), using Swedish
data, find no effects on mortality or hospitalizations. Similarly, Erixson (2014) find
no effects of increased lifetime incomes on hospitalizations, sick-leave or mortality.
Cheng et al. (2015) exploit lottery winnings to estimate the causal effect of exogenous
income changes on health care demand in the United Kingdom. Lottery winnings have
little to no effect on health care utilization on the extensive margin whereas household
income is positively and statistically significantly associated with some types of medical examinations and dental care utilization. Gardner and Oswald (2007) and Apouey
and Clark (2014) find positive effects on mental health status, measured by health
questionnaires in Great Britain, and Cesarini et al. (2015) report a small reduction in
prescriptions of mental health drugs. These results are in line with theories suggesting
that the social health gradient, to some extent, can be explained by the psychosocial
stress caused by a perceived low position in the social hierarchy (see e.g. Brunner,
1997, Wilkinson, 1999, Marmot and Wilkinson, 2005). Cesarini et al. (2015) use a
large estimation sample from Swedish administrative records and the positive effects
they find are at most modest. The difference in their results compared to those in
Lindahl (2005) may therefore be due to lack of statistical power in Lindahl (2005);
the effects found in Cesarini et al. (2015) may be too small to be detected in smaller
samples.
The lack of positive health effects in the literature exploiting exogenous wealth
shocks suggests that financial resources per se does not determine health. While this
is an important insight it does not rule out that there is an effect of (endogenously determined) lifetime income on health, operating through differences in health behaviors
linked to e.g. non-cognitive skills. This is precisely because the wealth shocks make
differences in lifetime income exogenous with respect to individual characteristics, including acquired human capital. If we think that the long-run effect of lifetime income
on health operates through the ways in which skills and other capabilities affect choices
and health behaviors throughout life, it is clear that this effect may be very different
from the parameter that is identified from a random increase in unearned income.
1.2.4 Income and consumption of health care
There is an extensive literature on the relationship between income and health care
spending. This literature is not directly linked to that on the social health gradient.
It is focused on either investigating the determinants of medical care expenditures
or estimating income elasticities, although the vast majority are correlation studies.
In an early study, Newhouse (1977) estimated the income elasticity of medical care
to be between 1.15 and 1.31, suggesting that health care is a luxury good. Similar
estimates of the income elasticity have been presented in several other cross-sectional
17
Measured by the Standardized Index of Bad Health.
29
studies18 using aggregated data from OECD countries (see Gerdtham and Jönsson,
2000, for a review), whereas studies using more disaggregated data generally find
income elasticities to be significantly lower than one.19 In a review of studies using
household survey data on health care expenditure, Getzen (2000) report estimates
ranging from less than zero to 0.1. Manning et al. (1987) report non-experimental
estimates of income elasticities from the RAND study that are at most 0.2. Acemoglu
et al. (2013) address the causal nature of the question by using oil price shocks to
instrument for local area incomes in the South of the United States. Their main
estimate of the income elasticity of aggregate health care spending is 0.7.
There are some studies that explicitly focus on the income elasticity of demand for
dental care. All but two of these studies were conducted more than twenty-five years ago
and report elasticities ranging from 0.06 to over 2.20 In a more recent study, Grönqvist
(2012) use register data on dental care consumption from the Swedish Public Dental
Service linked to administrative register data on individual characteristics to estimate
the relationship between income and dental care consumption. The rich data is an
important improvement over the previous literature as it allows controlling for several
confounding covariates and follow individuals over time. Grönqvist (2012) finds that
an increase of SEK 10,000 in yearly income is associated with a 0.27 percentage
point and statistically significant increase in the probability of visiting the dentist.21
However, the estimate decreases with two thirds when adding further socioeconomic
individual controls. Grönqvist (2012) also reports an income elasticity of 0.017 for
preventive dental care and 0.019 for restorative dental care on the intensive margin. In
another recent study, Meyerhoefer et al. (2014) use a nationally representative survey on
American household’s medical expenditures and price data from the American Dental
Association to estimate probabilities of using preventive and restorative services. Their
estimates suggest income elasticities ranging from 0.04 for basic restorative services
to 0.09 for preventive care.
To summarize, the literature finds income to be positively associated with consumption of both health care and dental care. For dental care this relationship holds
for both the extensive and the intensive margin of consumption. Considering this
18
The validity of these estimates have been questioned in e.g. Parkin et al. (1987), Blomqvist
and Carter (1997) and Sen (2005). The main critique points out that income elasticities above
one could be a result of underspecified models and that inferences about individual behavior
are made based on aggregated data. Another concern is the appropriateness of pooling data
from several countries into one panel.
19
One way of understanding this discrepancy, that has been discussed in the literature, is that
individual health care spending is less restricted by income due to public or private insurance
schemes whereas health care spending on the country level must be limited by country income
(Blomqvist and Carter, 1997).
20
Upton and Silverman (1972) report an elasticity of 2.2 which essentially is an association
between incomes and the number of dental care treatments and is estimated using very limited
data. Holtmann and Olsen Jr (1976) use an American household questionnaire and report
income elasticities varying from 0.12 to 0.41, depending on functional form of the multivariate
regressions. Manning and Phelps (1979) use a cross sectional survey and estimate statistically
significant income elasticities ranging from 0.54 to 0.88 for all services except for extractions
where the elasticities are negative but not statistically significant. Conrad et al. (1987) estimate
income elasticities using a household mail survey covering families with different types of
dental insurance plans and find the income elasticity of any use to be in the range 0.06 to 0.09
for adults.
21
SEK 10,000 is about e1000 in November 2015.
30
consumption as health-promoting behaviors, the descriptive evidence suggests that
there is an income gradient in health investments. However, it is not clear how these
associations relate to socioeconomic status more broadly defined, as they are estimated
using cross-sectional variation in current incomes.
1.3 Empirical setting
Dental care in Sweden is provided through private clinics and the public National
Dental Service and is free for children and youth aged 19 or younger. Patients are
free to choose dental care provider and price setting is free.22 The majority (60-80 %)
of dental care for adults is supplied by private clinics. There is a universal and tax
financed dental care insurance run by the government including a dental care voucher,
that patients can use at all clinics that are subscribed with the dental care benefit system
(approximately 96 % of all clinics). The voucher amounts to about e15 (SEK 150)
each year for those aged 30-74 and e30 (SEK 300) for those aged 20-29 and 75 and
older.23 There is also a high-cost protection scheme including different coinsurance
rates. As long as the accumulated (reimbursable) expenditures are below e310 during
a year, patients pay the full dental care cost, i.e. the price set by the dentist less the
voucher. There is an administratively set reference price list that serves as a basis
for determining eligibility for high-cost protection and a patient’s accumulated cost is
made up of the reference price for each consumed service.24 It is therefore not possible
to become eligible for high-cost protection simply by choosing an expensive dentist.
1.3.1 Data and variables
The empirical analysis is based on administrative register data covering the period July
2008-December 2011. While the population consists of all individuals in Sweden aged
20 years and above the sample is restricted to one observation per individual aged 3564 years. This results in a sample consisting of 3,883,052 observations. Among these,
just above 80 percent, or 3,157,324 individuals, have positive dental care consumption
during the studied period. These individuals are referred to as the patient sample.25
The sample is restricted to individuals aged 35-64 years in order to capture individuals’ socioeconomic status as accurately as possible. Most individuals in this age range
have completed their education and settled into a career path, implying that individual’s
lifetime trajectories for income can be treated as already being determined.26 In other
words, the dynamic process by which individual’s form skills, abilities and behaviors,
as described by e.g. Heckman (2007), mainly takes place before the mid-30’s. This
22
The National Dental Service in each county has a legal responsibility for ensuring the supply
of dental care to its citizens (National Dental Service Act, Tandvårdslagen (1985:125)).
23
The voucher covers preventive dental care and dental care that reduce pain and enables the
patient to eat, chew and speak without impediment.
24
If the price charged by the dentist is lower than the reference price, the charged price is used
when determining eligibility.
25
About 66 percent of the population have some positive dental care consumption.
26
In 2007, the average graduation age for university students in Sweden was over 29 years
(Uusitalo, 2011) and the standard retirement age in Sweden is 65 years.
31
implies that incomes in this age range can be viewed as capturing lifetime incomes
and as being determined by previously acquired capabilities. In addition, Nybom and
Stuhler (2015) find that incomes at prime age are the most accurate measures of the
ranking of lifetime income profiles.
I use individual’s mean disposable income over three years as an operationalization
of socioeconomic status. As the level of disposable income will vary over the life
cycle, the empirical analysis is performed by age-group. The income measure comes
from administrative registers collected at Statistics Sweden (LISA-database) and is
given by households after-tax earnings and transfers weighted by each individual’s
consumption weight. After-tax earnings include e.g. wage income, income from own
businesses, interest income, share dividends and profits on sold shares in interest funds.
The weighting accounts for the number of adults and children in the household.27 The
register data from Statistics Sweden also contain annual information for 2008-2011 of
demographic variables such as educational level and attachment to the labor market.
The source for the dental care data is the Swedish Dental Care Register at the
Swedish Social Insurance Agency. It covers all dental care produced and purchased at
clinics that are subscribed with the dental care benefit system and contains information
about visits, diagnosis and co-payment rates. It covers over 30 million dental care
treatments in total during the studied period. The main outcome variable is the sum
of dental care consumption during July 2008- December 2011 for each individual.
Consumption is measured by the price charged by the dentist. It does not necessarily
correspond to individual’s out-of-pocket payments, due to the high-cost protection
scheme described above. The motivation for using charged prices rather than out-ofpocket payments is to capture individuals’ actual consumption choices, i.e. the amount
of dental care that the individual wishes to consume.28
The Dental Health Register at the The National Board of Health and Welfare
includes the number of remaining teeth and the number of intact teeth. The latter
refers to teeth without fillings, prosthetics or damages requiring treatment. These
variables are reported in connection to visits where the patient uses their dental care
voucher. As with general health, an individual’s oral health can be understood and
measured in several dimensions. Edentulousness, i.e. loss of all teeth, is a longstanding
measure of oral health in the odontological literature (Gift and Atchison, 1995) while
the number of intact teeth is a proxy for previous treatments. Since oral diseases—to
a large extent—can be avoided by preventive measures (Coulter et al., 1994), both
the number of remaining teeth and the number of intact teeth can be understood as
depending on previous dental care consumption. Oral health measured in this way
is therefore a suitable operationalization of the theoretical concept of an individual’s
health stock as described in the health production literature.
In the descriptive analysis, I use the number of intact teeth as a measure of oral
health in order to make the results readily interpretable. In the main empirical analysis
the number of remaining teeth and the number of intact teeth are reduced into an oral
health index using principal component analysis.29 The index is constructed such that
27
The variable is “DispInkPersF04" in the LISA-database.
As a sensitivity analysis I use out-of-pocket payments to define consumption and the results
are qualitatively similar to those presented in section 1.6.1.
29
“Oral health” is the principal component of the covariance matrix of the variables “Remaining
teeth” and “Intact teeth”. It has an eigenvalue of 69.7 and explains 89% of the total variance.
28
32
a higher value represents better oral health. Since the oral health stock is a function
of consumed dental care, the first available measure of each individual’s oral health
index is used.
Table 1.1. Summary statistics
All aged 35-64
Sample
Women
Men
209.29
(676.19)
218.87
(737.00)
189.65
(382.27)
249.83
(980.79)
50.66
(9.29)
51.12
(9.26)
51.08
(9.28)
51.16
(9.23)
Compulsory schooling
0.11
0.10
0.08
0.11
Secondary schooling
0.47
0.47
0.46
0.49
Post-secondary schooling
0.36
0.38
0.41
0.34
11.32
(49.19)
9.85
(45.96)
9.58
(45.15)
10.13
(46.80)
Employed
0.77
0.81
0.78
0.83
Born outside Nordic country
0.04
0.03
0.02
0.03
0.12
(0.32)
0.12
(0.32)
0.15
(0.35)
0.09
(0.28)
3,883,052
3,157,324
1,624,224
1,533,100
Disposable income
Age
Days in unemployment
No. sick-spells
Observations
Notes: The sample consists of those with some positive consumption during July 2008December 2011. Disposable income in SEK 100s. SEK100 ≈ e10.7 in September 2015.
Standard deviations in parentheses. Column 3 and 4 provide means for women and men in
the patient sample.
1.4 Descriptive analysis
Table 1.1 provides summary statistics for all individuals aged 35-64 years and the
patient sample respectively. Individuals in the patient sample are on average somewhat
older, have higher incomes, post-secondary schooling to a greater extent, and a stronger
attachment to the labor market.30 Women are on average more likely to consume dental
care which is in line with previous empirical findings suggesting that women tend to
have a more proactive health behavior than men (see e.g. Lee, 2010). At the same
time, women in the patient sample on average have about 24 percent lower disposable
incomes than men in the patient sample. Probit estimates suggest that being born
outside the Nordic countries and the number of sick-spells31 decreases the probability
of having positive consumption for both men and women. See further table A 1 in
Appendix A.
The Kaiser–Meyer–Olkin measure of sampling adequacy is 0.99 suggesting that the variables
are well suited for principal component analysis.
30
Attachment to the labor market is measured by employment status (reported to Statistics
Sweden in November each year) and the number of reported days in unemployment.
31
Defined as episodes with sickness benefits.
33
The starting point for the descriptive analysis is to estimate the association between
incomes and dental care consumption for all individuals aged 35-64 years. This can be
understood as the average association between income and dental care consumption on
both the extensive and the intensive margin since the sample contains individuals with
no dental care consumption. Table 1.2 report the associations estimated including age
and sex and with the full set of controls respectively. When only controlling for age
and sex, a ten percent higher income is associated with a 1.3 percent increase in dental
care consumption among all aged 35-64 years. After including education, employment
status, a dummy for being born in a Nordic country and the number of sick-spells, the
estimate is cut by almost 18 percent, implying that a ten percent higher income is
associated with a 1 percent increase in dental care consumption.32 When estimated
separately for men and women, the associations are similar to those reported in table
1.2 but somewhat larger for men than for women, both with and without controls.33
Table 1.2. Income elasticities, extensive and intensive margin. All aged 35-64.
(1)
Log. disposable income
Controls
Number of observations
(2)
∗∗∗
0.1280
(0.0000)
0.1051∗∗∗
(0.0000)
Age & gender
Full set
3,883,052
3,883,052
Notes: Estimates of income elasticities for dental care consumption from a Poisson regression
model. The full set of controls include sex, age, education, employment status, a dummy for
being born in a Nordic country and the number of sick-spells. Standard errors in parenthesis.
∗
p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Table 1.3 reports the association between incomes and dental care consumption
estimated separately by age-group and for men and women. Comparing the first column
in table 1.2 with column (1) and (3) in table 1.3 gives that the average associations are
lower than the age-specific ones for men while the opposite is true for women. Including
covariates gives somewhat smaller, but not qualitatively different, results. Estimates
are consistently larger for men than for women, average dental care consumption is
however larger for women than for men within each age-group up until the oldest
group. Age-specific estimates are quantitatively similar across age-groups in table 1.3.
32
33
The associations are estimated with log-linear regression models.
Results reported in table A 3 in Appendix A.
34
Table 1.3. Income elasticities, extensive and intensive margin. All aged 35-64, by
age-group and gender.
Men
Age-group
(1)
Women
(2)
∗∗∗
(4)
35-39
0.1542
(0.0001)
0.1189
(0.0001)
0.0999
(0.0000)
0.1007∗∗∗
(0.0001)
40-44
0.1446∗∗∗
(0.0000)
0.1250∗∗∗
(0.0001)
0.1101∗∗∗
(0.0000)
0.1093∗∗∗
(0.0001)
45-49
0.1384∗∗∗
(0.0000)
0.1166∗∗∗
(0.0000)
0.1165∗∗∗
(0.0000)
0.1129∗∗∗
(0.0000)
50-54
0.1417∗∗∗
(0.0000)
0.1064∗∗∗
(0.0000)
0.1106∗∗∗
(0.0000)
0.0944∗∗∗
(0.0000)
55-59
0.1580∗∗∗
(0.0000)
0.1063∗∗∗
(0.0000)
0.1163∗∗∗
(0.0000)
0.0970∗∗∗
(0.0000)
60-64
0.1548∗∗∗
(0.0000)
0.1167∗∗∗
(0.0000)
0.0937∗∗∗
(0.0000)
0.0743∗∗∗
(0.0000)
No
Yes
No
Yes
Controls
∗∗∗
(3)
∗∗∗
Notes: Estimates of income elasticities for dental care consumption from a Poisson regression
model. The controls are sex, age, education, employment status, a dummy for being born in
a Nordic country and the number of sick-spells. Standard errors in parenthesis. ∗ p < 0.1,
∗∗
p < 0.05, ∗∗∗ p < 0.01
Turning to oral health and dental care consumption in the patient sample, figure 1.1
plots the average number of remaining teeth (left panel) and dental care consumption
(right panel) against age-groups for men and women in the top and bottom income
quartile respectively. For all groups, oral health, measured by the number of remaining
teeth, deteriorates with age, as suggested by the Grossman model. Men consistently
have better oral health than women within both the top and the bottom income quartile.
This is in line with Meisel et al. (2008) who found the risk of tooth loss to be
positively associated with child-bearing and estrogen-containing hormone therapy.
Similar explanations are discussed in a review in Russell et al. (2013).
Men and women in the top income quartile consistently have more remaining teeth
than their poorer counterparts and the gap between those in the top and those in the
bottom increases over the life cycle. This is in line with the Grossman framework and
previous findings suggesting that the protective effect of income on health increases
over the life cycle (Case and Deaton, 2005). The figure also suggests that men’s oral
health deteriorates faster than women’s oral health and that the income gap is larger
among men than women. Turning to the right panel of the figure, the income gap
in dental care consumption is also larger among men. Women, both in the top and
the bottom income quartile, consume more dental care than men up to the age of 45.
Men in the bottom income quartile consume less dental care than all other groups
and also experience a steep decline in oral health after the age of 50. The figure
thus provides graphical—and striking—evidence of the age-effects suggested by the
Grossman model; oral health depreciates with age while the demand for dental care
increases. In addition, it provides descriptive evidence of a widening in the income
gradient for both oral health and dental care consumption; the difference between
35
26
4000
No. remaining teeth
27
28
Dental care consumption, SEK
6000
8000
29
10000
30
Figure 1.1. Number of remaining teeth and dental care consumption, by age and
sex-specific income quartile.
Men, top quartile
Women, top quartile
2000
Men, bottom quartile
25
Women, bottom quartile
35−39
40−44
45−49
50−54
Age−cohort
55−59
60−64
35−39
40−44
45−49
50−54
55−59
60−64
Age−cohort
those in the top income quartile and those in the bottom with respect to the number
of remaining teeth and dental care consumption is larger in the oldest age-group
compared to the youngest. Figure 1.1 also illustrates the importance of accounting
for both differences between men and women and the increased depreciation of health
caused by age in order to get informative estimates of the impact of income dental care
consumption.
Figure 1.2 plots estimates from regressions of health and dental care consumption
respectively on disposable incomes.34 The models are estimated for the patient sample
and separately by age-groups and sex. The left panel plots estimates of the association
between income and the number of remaining teeth. The latter is measured prior to
the period when we observe individuals and their dental care consumption. Under the
assumption that the distribution of inherited oral health endowments are the same across
age-groups and that there has been no shift in the distribution of factors determining
socioeconomic status, the estimates plotted in figure 1.2 illustrate the evolution of the
number of teeth and dental care consumption with respect to socioeconomic status
over the life cycle. The plot in the left panel suggests that the social health gradient
is small, but increases monotonically with age. For the youngest age-group, a one
percent increase in income is associated with an increase of about 0.007 percent in the
number of remaining teeth. The corresponding figure for the oldest cohort is almost
ten times as high; a one percent increase in income is associated with about a 0.03
percent increase in the number of remaining teeth for both men and women. Hence,
the social gradient in oral health steepens sharply with age suggesting that the long-run
34
The association between the number of teeth and incomes are estimated with a log-linear
regression model whereas associations with dental care consumption are OLS estimates from
a log-log model.
36
Figure 1.2. Income elasticities with respect to the number of remaining teeth and
dental care consumption, by age-group and sex. Estimated for patient sample.
.06
Income elasticity
.02
.04
0
−.02
−.02
0
Income elasticity
.02
.04
.06
.08
Dental care consumption
.08
Remaining teeth
35−39
40−44
45−49
50−54
55−59
Age−group
Men
Women
60−64
35−39
40−44
45−49
50−54
55−59
60−64
Age−group
Men
Women
Notes: The left panel plots elasticities from a Poisson regression of no. remaining teeth on log.
disposable incomes. The right panel plots OLS estimates of log. dental care consumption on
log. disposable income. Models include age-specific fixed effects and estimates are plotted with
95 % CIs.
impact of income on oral health is substantial. An explanation for the observed pattern
is that the provision of free dental care for children and youth below the age of 20 in
Sweden mediates the effect of disparities in initial oral health endowments whereas the
association between incomes and oral health amplifies over the life cycle, potentially
operating through dental care consumption. This is supported by the right panel of
figure 1.2 suggesting that a one percent increase in income for the oldest cohort is
associated with an increase in dental care consumption of about 0.07 percent for men
and 0.06 percent for women while, in the youngest cohort, the association between
income and dental care consumption is substantially smaller for men and negative for
women. Taken together the results plotted in figure 1.2 are in line with the discussion
in the conceptual framework; income as a determinant of both oral health and dental
care consumption increases with age, suggesting that the social gradient widens over
the life cycle.
1.5 Empirical modeling
The descriptive analysis illustrates the social gradient for both oral health and dental care consumption. So far the analysis has focused on health and consumption
separately whereas the main interest lies in connecting them. Since dental care consumption affects oral health which affects dental care consumption and so on, dental
37
care consumption is determined by an autoregressive process where consumption today depends on consumption in the past. Given that dental care consumption today
depends on lifetime income, as a measure of socioeconomic status, this should also
hold for consumption in the past. Hence, lifetime income can be thought of as affecting
dental care consumption through two channels; directly and indirectly through past
dental care consumption. The right panel of figure 1.2 should therefore be interpreted
as plotting the “overall” social gradient for dental care consumption as it neglects that
individuals with higher incomes also have better oral health, on average. The analysis
now turns to identification of the direct effect by shutting down the indirect effect
operating through past consumption.
To fix ideas, the structural relations of interest can be described by the following
model:
(1.2)
ct = β0 + β1 Y + β2 Ht−1 + εt .
From equation 1.2, we have that dental care consumption depends on (time-invariant)
socioeconomic status measured by lifetime income, Y , and oral health in period t − 1.
The health stock is assumed to evolve according to the Grossman model;
Ht = γ0 + γ1 Ht−1 + γ2 ct−1 + ηt ,
(1.3)
where ct−1 is seen as the gross investment in oral health and ηt represents shocks
to oral health.35 Consequently, lifetime income affects oral health indirectly through
past dental care consumption (investments). Equation 1.3 also implies that, without
further assumptions, β1 in equation 1.2 does not identify the long-term effect of
lifetime income on dental care consumption since we cannot autonomously vary Y.
The identification problem can be illustrated by considering an experiment. If incomes
were exogenously determined across individuals through e.g. a lottery, this would shift
unearned income while Ht−1 would be held fixed as the lottery would not affect past
dental care consumption. The experiment would therefore allow us to obtain an
estimate of the direct effect of unearned income on dental care consumption from a
regression of ct on unearned income alone. However, this parameter may be widely
different from the effect of lifetime income on dental care consumption.
The effect of lifetime income on dental care consumption can be identified under
three assumptions; (1) individuals’ (future) dental care consumption (or investments)
is not determined by the same optimization that determines lifetime income36 , (2) an
individual’s position in the observed income distribution (i.e. disposable income) is
not affected by the dental care consumption and (3) oral health at t = 0 is independent
of Y .
I consider the first assumption to be innocuous. The second assumption can be
motivated by the fact that the analysis is performed within age-groups. It is simply
hard to see how dental care consumption could affect an individual’s position in the
income distribution within that individual’s age-group. Assumption (3) means that
baseline oral health, H0 , is seen as being determined by a random draw of “Nature”. In
The parameters γ1 and γ2 can be made time dependent without any loss of generality regarding
the empirical analysis.
36
It is likely that individuals’ choices of education, i.e. schooling investments, and occupation
is determined by counterfactual evaluations of lifetime incomes of different choices (see e.g.
Willis and Rosen, 1979). If these evaluations also consider health investments then observed
lifetime income is not exogenous to the chosen investment paths.
35
38
the Swedish context, this assumption can, in part, be motivated by that the provision of
free dental care before the age of 20 acts to weaken the link between family income and
early life oral health.37 Had H0 been the initial stock of general health, assumption (3)
would not hold considering the vast empirical evidence suggesting that poor health in
children hinders learning and hence the formation of capabilities (Currie and Almond,
2011). It is hard to see how the same mechanisms could be in place when considering
oral health.
Under assumption (3) we have that oral health stock at t is completely determined
by the consumption of dental care in t − 1 and the baseline oral health H0 , that is
Ht = f (ct−1 , H0 ).
The implication is that Ht−1 in equation 1.2 is exclusively determined by past dental
care consumption. Furthermore consumption in period t will be a function of lifetime
income, consumption in t − 2 and the baseline oral health H0 , that is
ct = f (Y, ct−2 , ct−3 ...c0 , H0 ).
Given that oral health has no direct effect on lifetime income, which follows from
assumptions 1 and 2, and that H0 is independent of lifetime income (assumption 3)
we can autonomously vary Y once we control for the oral health stock Ht−1 . This
is because controlling for the oral health stock “shuts” down the indirect effect from
earlier consumption streams which, in turn, are functions of Y . Under assumptions
1-3, the parameter β1 can thus be thought of as capturing the direct effect of lifetime
income on dental care consumption in any given period.
The descriptive analysis illustrated the importance of accounting for differences
between men and women and the age-related depreciation of the oral health stock.
Equation (1.4) gives the preferred specification where ci is individual i s observed
dental care consumption, Hi , is the first available measure of individual i s oral health
stock and μa is an age-specific fixed effect.
ln ci = β0 + β1 ln Yi + β2 ln Hi + μa + i ,
(1.4)
Equation (1.4) is estimated separately for age-groups (suggested by assumption 2) and
by sex.
1.6 Results
1.6.1 Main results
This section reports results from OLS-estimates of equation (1.4) with and without
including oral health. Both models include age-group fixed effects. Table 1.4 shows
the effect of income on dental care consumption, estimated for individuals in the patient
37
In a fully privately financed dental care system it is reasonable to assume that family income
would be an important determinant of children’s oral health. Together with the intergenerational transmission of human capital (see e.g. Black et al., 2005) this would imply that early
life oral health would not be independent of lifetime income.
39
sample with non-missing oral health measures.38 Without conditioning on oral health,
the estimated impact of income is 0.015 as given in column (1). After conditioning
on health, the estimated effect of income increases to 0.03 suggesting that failing to
take previous dental care consumption into account gives downward biased effects of
income as previously discussed.
Table 1.4. Estimated impact of income on dental care consumption. All individuals
in patient sample with non-missing oral health measures.
(1)
Log. disposable income
0.0151
(0.0009)
0.0300∗∗∗
(0.0009)
No
Yes
2,972,291
2,972,291
Oral health
Number of observations
(2)
∗∗∗
Notes: Results for individuals in the patient sample with non-missing health measures (94%
of patient sample). The outcome variable is log. dental care consumption. Models include
dummies for age-group. Standard errors in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗
p < 0.01
1.6.2 Results by age-group and gender
The descriptive analysis, along with the Grossman model, suggests that age-effects are
important for understanding health behaviors. Figure 1.3 therefore plots the impact of
income estimated separately for age-groups. The overall pattern is the same as in table
1.4; the effect of income is consistently larger when conditioning on oral health. In
addition, the effect of income increases strikingly over the life cycle. For the youngest
age-group the effect is negative and close to zero.39 From the age of 40 up to the years
before the legal retirement age 65, the effect of income increases by a factor of 9.3.
Hence, the social gradient in dental care consumption steepens considerably over the
life cycle. Considering that dental care is only one dimension of individual’s health
behaviors, the results suggest that the long-run effect of socioeconomic status on health
investments could be substantial.
Figure 1.4 plots the estimated impact of income on dental care consumption separately for age-groups by sex. The patterns are largely the same as for previously
reported results. The estimated impact of income increases when conditioning on oral
health and increases monotonically with age. For men, the effect of income is not
statistically significantly different from zero for the age group 40-44 and for women
only marginally so. From the mid-40’s up to the oldest age-group the effect of income
on dental care consumption increases by a factor of 4.8 for men and 2.7 for women.
In contrast to the descriptive results, income is a more important determinant of consumption for women than for men, except for the oldest age-group. This is more in line
with previous findings suggesting that women have a more pro-active health behavior
(Lee, 2010). Given that women on average are more cautious with their health and
38
39
94 % of the patient sample.
See table A 5 in Appendix A for further details.
40
that demand for health increases with socioeconomic status, we should expect larger
estimates for women. See table A 6 in Appendix A for details.
.04
.02
0
−.02
−.04
Impact of income on dental care consumption
.06
Figure 1.3. Estimated impact of income on dental care consumption, by age-group.
35−39
40−44
45−49
50−54
55−59
60−64
Age−group
Unconditional
Conditional on oral health
Notes: Results for individuals in the patient sample. The outcome variable is log. dental care
consumption. Models include age-specific fixed effects and estimates are plotted with 95 % CIs.
Overall, the results suggest that the effect of socioeconomic status on dental care
consumption increases as individuals grow older. This is in line with the literature
on human capital formation suggesting that the negative effects of disadvantages—
resulting in low SES—increases over the course of life due to the synergies in producing
health and other human capabilities (Heckman, 2012). It is also in line with findings of
an upward trend in income-related health disparities over the life-cycle (see e.g. Islam
et al., 2010). While I cannot determine the precise mechanism that causes the social
health gradient in oral health, the empirical results are in line with the notion that one
of these mechanisms is the effect of socioeconomic status on health behaviors.
41
Figure 1.4. Estimated impact of income on dental care consumption, by age-group
and sex.
Male
.06
.04
.02
0
−.04
−.02
Impact of income on dental care consumption
.04
.02
0
−.02
−.04
Impact of income on dental care consumption
.06
Female
35−39
40−44
45−49
50−54
55−59
60−64
35−39
40−44
Age−group
45−49
50−54
55−59
60−64
Age−group
Unconditional
Unconditional
Conditional on oral health
Conditional on oral health
Notes: Results for individuals in the patient sample. The outcome variable is log. dental care
consumption. Models include age-specific fixed effects and estimates are plotted with 95 % CIs.
1.6.3 Controlling for health non-parametrically
The effect of socioeconomic status on dental care consumption has economic significance in itself. In other words, an estimate of β1 in equation 1.4 identifies a parameter
that can be given a qualitative interpretation. We are however not interested in interpreting the parameter estimate of β2 in equation 1.4 and therefore do not need to
restrict oral health to enter in a linear fashion.
Consider the following model where oral health enters through some non-linear
function f (Hi ):
ci = θ0 + θ1 Yi + f (Hi ) + i ,
i = 1, ..., N.
(1.5)
Model 1.5 can be estimated using the approach suggested by Robinson (1988) which
includes applying conditional expectations on both sides of equation 1.5. We get:
E(ci | Hi ) = θ0 + θ1 E(Yi | Hi ) + f (Hi ),
i = 1, ..., N.
(1.6)
Subtracting 1.6 from 1.5 gives:
ci − E(ci | Hi ) = θ1 (Yi − E(Yi | Hi )) + i ,
i = 1, ..., N.
(1.7)
The conditional expectations are given by first sorting the data with respect to oral health
and then subtracting the observed ci and Yi for each individual from the antecedent
observation. The sorting is made by age-groups and sex. I can then estimate θ1 by
fitting 1.7 by ordinary least squares. Figure 1.5 plots the estimates of θ1 . The results
42
Figure 1.5. Estimated effect of lifetime income on dental care consumption by agegroup and sex using Robinson’s (1988) semi-parametric estimator.
.06
.04
Estimated effect of income
0
.02
.04
0
.02
Estimated effect of income
.06
.08
Female
.08
Male
35−39
40−44
45−49
50−54
55−59
60−64
35−39
40−44
Age−group
45−49
50−54
55−59
60−64
Age−group
Notes: Results for individuals in the patient sample.
are qualitatively similar to those previously reported. However, the point estimates are
consistently larger than those reported in figure 1.4and for men the estimated effect of
income decreases between age-group 45-49 to 50-54.
Model 1.5 has also been estimated with f (Hi ) defined by polynomials of Hi which
gives qualitatively similar results. See section A 2 and figures A 1-A 2 in Appendix A
for further details.
1.6.4 Alternative definition of consumption
So far the analysis has not separated between different types of dental care implying
that all dental care is treated as health investments. Doing so may understate the
effect of income on consumption since it is likely that low-income individuals face
a greater risk of needing acute care. As a sensitivity analysis I therefore define a
new measure of dental care consumption where dental care visits containing “acute
examinations” or treatment of pain and oral disease are excluded. This is done to create
a consumption measure that is closer related to the notion of health investments. The
estimated impact of income on consumption of non-acute dental care is reported in
table 1.5. Both estimates are larger than those reported for all dental care consumption.
The income effect estimated conditional on oral health (column 2) is 30 % larger than
the estimate for overall consumption. Figure 1.6 plots results for age-groups and by
sex.40 The overall patterns are the same as those reported previously but the effects of
40
See table A 10 in Appendix A for further details.
43
income estimated conditional on oral health are consistently larger (except for women
in the oldest age-group). For the youngest age-groups the estimated effect of income
is slightly positive for men and negative but statistically indistinguishable from zero
for women. From age 40-44 up to the oldest age-group, the estimated effect of income
increases by a factor of just over 3 for both men and women. These results indicate that
higher lifetime incomes indeed increases oral health investments since acute dental
care, that can either be viewed as pure consumption or random events, is excluded
from the analysis. In addition, the results strengthen the argument that controlling for
oral health solves the omitted variable bias that occurs when socioeconomic status is
positively related to healthy investment behaviors.
Table 1.5. Estimated impact of income on consumption of non-acute dental care.
(1)
Log. disposable income
Oral health
Number of observations
(2)
∗∗∗
0.02089
(0.00098)
0.03872∗∗∗
(0.00096)
No
Yes
2,942,444
2,942,444
Notes: Results for individuals in the patient sample with non-missing health measures (94%
of patient sample). The outcome variable is log. consumption of non-acute dental care.
Age-specific fixed-effects included. Standard errors in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05,
∗∗∗
p < 0.01
44
Figure 1.6. Estimated impact of income on consumption of non-acute dental care, by
age-group and sex.
Women
.04
.02
0
−.04
−.02
Impact of lifetime income on non−acute consumption
.04
.02
0
−.02
−.04
Impact of lifetime income on non−acute consumption
.06
.06
Men
35−39
40−44
45−49
50−54
55−59
60−64
Age−group
35−39
40−44
45−49
50−54
55−59
60−64
Age−group
Unconditional
Unconditional
Conditional on oral health
Conditional on oral health
Notes: Results for individuals in the patient sample. The outcome variable is log. consumption
of non-acute dental care. Models include age-dummies and estimates are plotted with 95% CI:s.
1.7 Concluding discussion
The socioeconomic health gradient is well documented but we still lack a full understanding of its causes. A growing empirical literature investigates the development
of human capabilities throughout life as a way of understanding inequalities in adult
health and other dimensions. This literature points out a positive relationship between
socioeconomic status, as a function of acquired abilities, as one possible mechanism
behind the gradient. The theoretical framework underlying this literature can be summarized by the notion that capabilities beget capabilities, implying that individuals
with greater e.g. cognitive and non-cognitive skills will not only be more successful
in terms of economic outcomes but also accumulate health capital more efficiently.
As both skills and health can be characterized as stocks that are built up through investments over the course of life, the social health gradient must be understood from a
life-cycle perspective.
While the idea of synergies in the production of health and other forms of human
capital as an important determinant of inequality is intuitively appealing and theoretically sound, empirical analyzes are constrained by limitations of data. In general,
health behaviors are unobserved for researchers, leaving the precise processes by which
socioeconomic status is translated into health an open question.
In this essay, I exploit uniquely rich Swedish register data to estimate the effect
of socioeconomic status on the consumption dimension of health behaviors in adults.
The empirical analysis is restricted to individuals aged 35-64 years and is performed
separately by age-groups. In addition, I use dental care consumption as an operational-
45
ization of health investments. This suggests that the direction of causality is clear
as the studied individuals are at an age where most have completed their education
and settled into a career path and the effect of health investments on the formation of
human capabilities is arguable negligible. In addition, the richness of data allows me
to control for individuals’ oral health and thus estimate the effect of socioeconomic
status on dental care consumption conditional on past oral health investments.
I use lifetime incomes as a measure of socioeconomic status and find that individuals
with higher incomes consume more dental care at all ages. I also find that oral health
deteriorates faster for those in the bottom of the income distribution compared to those
in the top and that the income gap in both oral health and dental care consumption
widens substantially over the life-cycle. The estimated impact of income on dental
care consumption increases when controlling for oral health. This is in line with
individuals with higher lifetime incomes having consumed more dental care in the
past. The income gradient steepens substantially as individuals grow older; from
midlife to the years before the legal retirement age the estimated effect increases by a
factor of 6.2 for men and 4.8 for women. Since dental care consumption is only one
dimension of individual’s health behaviors, the results indicate that lifetime effects of
socioeconomic status on health behaviors could be substantial in other dimensions.
Disparities in health across socioeconomic groups have not only spurred academic
interest but are also a common concern among policy-makers. This is illustrated, not
least, by Obama’s Affordable Care Act and the WHO’s global initiative of closing the
health gap in one generation (WHO, 2008). In order to design efficient policies to
address the gradient it is crucial to first have an understanding of what causes it. While
I cannot determine the precise mechanisms giving rise to the documented gradient
in oral health, the results support the notion that differences in health behaviors may
account for part of it. The empirical analysis paints a clear picture of how differences
in dental care consumption related to socioeconomic status increases monotonically
over the life-cycle. In addition, the literature exploiting exogenous wealth shocks does
not find support for the view that financial resources as such improve health. Taken
together with the general conclusion from the literature on human capability formation
that early life conditions have a significant effect on both socioeconomic status and
health, the results suggests that policies aimed at addressing inequalities in lifetime
opportunities in general are most likely to succeed.
46
References
Acemoglu, D., Finkelstein, A. and Notowidigdo, M. J. (2013). Income and health
spending: evidence from oil price shocks. Review of Economics and Statistics,
95 (4), 1079–1095.
Almond, D. and Currie, J. (2011). Killing Me Softly: The Fetal Origins Hypothesis.
The Journal of Economic Perspectives, 25 (3), 153–172.
Apouey, B. and Clark, A. E. (2014). Winning big but feeling no better? The effect
of lottery prizes on physical and mental health. Health Economics.
Baum, C. L. and Ruhm, C. J. (2009). Age, socioeconomic status and obesity growth.
Journal of Health Economics, 28 (3), 635–648.
Becker, G. S. (1965). A Theory of the Allocation of Time. The Economic Journal,
pp. 493–517.
Black, S. E., Devereux, P. J. and Salvanes, K. G. (2005). Why the Apple Doesn’t
Fall Far: Understanding Intergenerational Transmission of Human Capital. The
American Economic Review, 95 (1), 437–449.
Blomqvist, . G. and Carter, R. A. L. (1997). Is health care really a luxury? Journal
of Health Economics, 16 (2), 207–229.
Brunner, E. (1997). Socioeconomic Determinants of Health: Stress and the Biology
of Inequality. BMJ: British Medical Journal, 314 (7092), 1472–1476.
Campbell, F., Conti, G., Heckman, J. J., Moon, S. H., Pinto, R., Pungello, E. and
Pan, Y. (2014). Early Childhood Investments Substantially Boost Adult Health.
Science, 343 (6178), 1478–1485.
Card, D., Dobkin, C. and Maestas, N. (2009). Does Medicare Save Lives? The
Quarterly Journal of Economics, 124 (2), 597–636.
Case, A. and Deaton, A. S. (2005). Broken Down by Work and Sex: How Our
Health Declines. In Analyses in the Economics of Aging, University of Chicago
Press, pp. 185–212.
—, Lubotsky, D. and Paxson, C. (2002). Economic status and health in childhood:
The origins of the gradient. The American Economic Review, 92 (5), 1308–1334.
— and Paxson, C. (2002). Parental Behavior And Child Health. Health Affairs,
21 (2), 164–178.
Cesarini, D., Lindqvist, E., Östling, R. and Wallace, B. (2015). Wealth, Health,
and Child Development: Evidence from Administrative Data on Swedish Lottery
Players. IFN Working Paper No. 1060.
Cheng, T. C., Costa-Font, J. and Powdthavee, N. (2015). Do you have to win it to
fix it? A longitudinal study of lottery winners and their health care demand.
Conrad, D. A., Grembowski, D. and Milgrom, P. (1987). Dental care demand:
insurance effects and plan design. Health Services Research, 22 (3), 341.
Coulter, I. D., Marcus, M. and Atchison, K. A. (1994). Measuring oral health
status: Theoretical and methodological challenges. Social Science & Medicine,
38 (11), 1531–1541.
Cropper, M. L. (1981). Measuring the benefits from reduced morbidity. The
American Economic Review, pp. 235–240.
Cunha, F. and Heckman, J. J. (2007). The Technology of Skill Formation. The
American Economic Review, 97 (2), 31–47.
—, —, Lochner, L. and Masterov, D. V. (2006). Interpreting the evidence on life
cycle skill formation. Handbook of the Economics of Education, 1, 697–812.
47
Currie, J. (2009). Healthy, wealthy, and wise: Is there a causal relationship between
child health and human capital development? Journal of Economic Literature,
47 (1), 87–122.
— and Almond, D. (2011). Chapter 15 - Human capital development before age five.
In D. C. a. O. Ashenfelter (ed.), Handbook of Labor Economics, vol. 4, Part B,
Elsevier, pp. 1315–1486.
Cutler, D. M., Lleras-Muney, A. and Vogl, T. (2011). Socioeconomic Status and
Health: Dimensions and Mechanisms. In S. Glied and P. C. Smith (eds.), The
Oxford Handbook of Health Economics, Oxford : New York: Oxford University
Press.
Deaton, A. (2003). Health, Inequality, and Economic Development. Journal of
Economic Literature, 41 (1), 113–158.
Doyle, J. J., Jr. (2011). Returns to Local-Area Health Care Spending: Evidence
from Health Shocks to Patients Far From Home. American Economic Journal:
Applied Economics, 3 (3), 221–243.
Erixson, O. (2014). Health responses to a wealth shock: Evidence from a Swedish
tax reform. IFN Working Paper No. 1011.
Gardner, J. and Oswald, A. J. (2007). Money and mental wellbeing: A longitudinal
study of medium-sized lottery wins. Journal of Health Economics, 26 (1), 49–60.
Gerdtham, U.-G. and Jönsson, B. (2000). Chapter 1 International comparisons of
health expenditure: Theory, data and econometric analysis. In A. J. C. a. J. P.
Newhouse (ed.), Handbook of Health Economics, vol. 1, Part A, Elsevier, pp.
11–53.
—, Johannesson, M., Lundberg, L. and Isacson, D. (1999). The demand for
health: results from new measures of health capital. European Journal of Political
Economy, 15 (3), 501–521.
Getzen, T. E. (2000). Health care is an individual necessity and a national luxury:
applying multilevel decision models to the analysis of health care expenditures.
Journal of Health Economics, 19 (2), 259–270.
Gift, H. C. and Atchison, K. A. (1995). Oral health, health, and health-related
quality of life. Medical Care, pp. NS57–NS77.
Gluckman, P. D. and Hanson, M. A. (2006a). Chapter 1 - The developmental
origins of health and disease: an overview. In Developmental origins of health and
disease, Cambridge; New York: Cambridge University Press.
— and — (2006b). Developmental Origins of Health and Disease. Cambridge; New
York: Cambridge University Press.
Goldman, D. and Smith, J. P. (2005). Socioeconomic Differences in the Adoption of
New Medical Technologies. The American Economic Review, 95 (2), 234–237.
Grönqvist, E. (2012). The demand for dental care: Analyses of the importance of
price and income. Working Papers in Social Insurance 2012:1. Swedish Social
Insurance Agency.
Grossman, M. (1972a). The Demand for Health: A Theoretical and Empirical
Investigation. New York: NBER, Occasional Paper 119, Columbia Press.
— (1972b). On the concept of health capital and the demand for health. The Journal
of Political Economy, 80 (2), 223.
— (2000). Chapter 7 The human capital model. In A. J. Culyer and J. P. Newhouse
(eds.), Handbook of Health Economics, vol. 1, Part A, Elsevier, pp. 347–408.
48
Heckman, J. J. (2007). The economics, technology, and neuroscience of human
capability formation. Proceedings of the National Academy of Sciences, 104 (33),
13250–13255.
— (2012). The developmental origins of health. Health Economics, 21 (1), 24–29.
—, Stixrud, J. and Urzua, S. (2006). The Effects of Cognitive and Noncognitive
Abilities on Labor Market Outcomes and Social Behavior. Journal of Labor
Economics, 24 (3), 411–482.
Öhman, M. (2015). Be smart, live long: the relationship between cognitive and
non-cognitive abilities and mortality. Institute for Labour Market Policy
Evaluation (IFAU). Working paper 2015:21.
Holtmann, A. G. and Olsen Jr, E. O. (1976). The demand for dental care: a study
of consumption and household production. Journal of Human Resources, pp.
546–560.
Islam, M. K., Gerdtham, U.-G., Clarke, P. and Burström, K. (2010). Does
income-related health inequality change as the population ages? Evidence from
Swedish panel data. Health economics, 19 (3), 334–349.
Lee, C. (2010). Gender, health, and health behaviors. In Handbook of gender
research in psychology, Springer, pp. 471–493.
Lindahl, M. (2005). Estimating the effect of income on health and mortality using
lottery prizes as an exogenous source of variation in income. Journal of Human
Resources, 40 (1), 144–168.
Manning, W. G., Newhouse, J. P., Duan, N., Keeler, E. B. and Leibowitz, A.
(1987). Health insurance and the demand for medical care: evidence from a
randomized experiment. The American Economic Review, pp. 251–277.
— and Phelps, C. E. (1979). The demand for dental care. The Bell Journal of
Economics, pp. 503–525.
Marmot, M. and Wilkinson, R. (eds.) (2005). Social Determinants of Health.
Oxford University Press.
Meisel, P., Reifenberger, J., Haase, R., Nauck, M., Bandt, C. and Kocher, T.
(2008). Women are periodontally healthier than men, but why don’t they have more
teeth than men?:. Menopause, 15 (2), 270–275.
Meyerhoefer, C. D., Zuvekas, S. H. and Manski, R. (2014). The Demand for
Preventive and Restorative Dental Services. Health Economics, 23 (1), 14–32.
Newhouse, J. P. (1977). Medical-care expenditure: a cross-national survey. Journal
of Human Resources, pp. 115–125.
Nybom, M. and Stuhler, J. (2015). Biases in standard measures of
intergenerational income dependence. Institute for Labour Market Policy
Evaluation (IFAU). Working paper 2015:13.
Parkin, D., McGuire, A. and Yule, B. (1987). Aggregate health care expenditures
and national income: is health care a luxury good? Journal of Health Economics,
6 (2), 109–127.
Robinson, P. M. (1988). Root-N-consistent semiparametric regression.
Econometrica: Journal of the Econometric Society, pp. 931–954.
Russell, S. L., Gordon, S., Lukacs, J. R. and Kaste, L. M. (2013). Sex/Gender
Differences in Tooth Loss and Edentulism: Historical Perspectives, Biological
Factors, and Sociologic Reasons. Dental Clinics of North America, 57 (2),
317–337.
49
Sen, A. (2005). Is health care a luxury? New evidence from OECD data.
International Journal of Health Care Finance and Economics, 5 (2), 147–164.
Smith, J. P. (2007). The Impact of Socioeconomic Status on Health over the
Life-Course. The Journal of Human Resources, 42 (4), 739–764.
Upton, C. and Silverman, W. (1972). The demand for dental services. Journal of
Human Resources, pp. 250–261.
Uusitalo, R. (2011). Factors affecting the late graduation of Swedish university
students. In Långtidsutredningen 2011, Bilaga 3 till LU2011.
van der Wees, P. J., Wammes, J. J., Westert, G. P. and Jeurissen, P. (2015). The
Relationship Between the Scope of Essential Health Benefits and Statutory
Financing: An International Comparison Across Eight European Countries.
International Journal of Health Policy and Management.
van Kippersluis, H. and Galama, T. J. (2014). Wealth and health behavior: Testing
the concept of a health cost. European Economic Review, 72, 197–220.
Wagstaff, A. (1986). The demand for health: some new empirical evidence. Journal
of Health Economics, 5 (3), 195–233.
— (1993). The demand for health: An empirical reformulation of the Grossman
model. Health Economics, 2 (2), 189–198.
WHO (2008). Closing the gap in a generation: health equity through action on the
social determinants of health: Commission on Social Determinants of Health final
report. Geneva, Switzerland: World Health Organization, Commission on Social
Determinants of Health.
Wilkinson, R. G. (1999). Health, hierarchy, and social anxiety. Annals of the New
York Academy of Sciences, 896 (1), 48–63.
Willis, R. J. and Rosen, S. (1979). Education and Self-Selection. Journal of
Political Economy, 87 (5), S7–S36.
Zweifel, P., Breyer, F. and Kifmann, M. (2009). Health economics. Dordrecht ;
New York: Springer, 2nd edn.
— and Nocera, S. (1998). The demand for health: An empirical test of the
Grossman model using panel data. In Health, the Medical Profession, and
Regulation, Boston, MA: Springer US.
50
Appendix A
A 1 Additional tables and results
Table A 1. Full sample probit estimates of marginal effects on the probability of any
consumption.
Prob. of any consumption
Full sample
Men
Women
Log. disposable income
0.2455∗∗∗
(0.0014)
0.2669∗∗∗
(0.0018)
0.2148∗∗∗
(0.0021)
Female
0.3347∗∗∗
(0.0016)
Age
0.0190∗∗∗
(0.0001)
0.0203∗∗∗
(0.0001)
0.0177∗∗∗
(0.0001)
Compulsory schooling
0.1176∗∗∗
(0.0035)
0.1137∗∗∗
(0.0047)
0.1238∗∗∗
(0.0054)
Secondary schooling
0.2912∗∗∗
(0.0030)
0.2692∗∗∗
(0.0041)
0.3283∗∗∗
(0.0045)
Post-secondary schooling
0.4278∗∗∗
(0.0032)
0.3966∗∗∗
(0.0043)
0.4734∗∗∗
(0.0047)
No. of days in unemployment
0.0004∗∗∗
(0.0000)
0.0003∗∗∗
(0.0000)
0.0004∗∗∗
(0.0000)
Employed
0.4247∗∗∗
(0.0020)
0.4427∗∗∗
(0.0028)
0.3991∗∗∗
(0.0029)
Born outside Nordic countries
-0.5878∗∗∗
(0.0034)
-0.5434∗∗∗
(0.0046)
-0.6444∗∗∗
(0.0051)
No. sick-spells
-0.0202∗∗∗
(0.0025)
-0.0365∗∗∗
(0.0037)
-0.0097∗∗∗
(0.0033)
-2.0395
-2.2050
-1.5016
3,883,052
1,968,226
1,914,826
Constant
Number of observations
Notes: Estimates from probit model of having positive consumption during July 2008–
December 2011. Sick-spells defined as episodes with sickness benefits. Standard errors in
parentheses.
51
Table A 2. Income elasticities, extensive and intensive margin. All aged 35-64. Full
table.
(1)
Log. disposable income
0.1051∗∗∗
(0.0000)
Female
0.0857∗∗∗
(0.0000)
Age
0.0403∗∗∗
(0.0000)
Level of education
-0.0106∗∗∗
(0.0000)
Employed
0.1011∗∗∗
(0.0000)
No. of days in unemployment
0.0003∗∗∗
(0.0000)
Born outside Nordic countries
-0.1536∗∗∗
(0.0000)
No. sick-spells
0.0628∗∗∗
(0.0000)
Constant
5.9908∗∗∗
(0.0001)
Number of observations
3,883,052
Notes: Estimates of income elasticities for dental care consumption from a log-linear Poisson
regression. Standard errors in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Table A 3. Income elasticities, extensive and intensive margin. All aged 35-64, by
gender.
Men
(1)
Women
(2)
∗∗∗
(3)
∗∗∗
(4)
∗∗∗
Log. lifetime income
0.2600
(0.0000)
0.1151
(0.0000)
0.2142
(0.0000)
0.0918∗∗∗
(0.0000)
Number of observations
1,968,226
1,968,226
1,914,826
1,914,826
Notes: Estimates of income elasticities for dental care consumption from a log-linear Poisson
regression. Standard errors in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
52
Table A 4. Income elasticities for oral health and dental care consumption. By
age-group and gender.
Age-group
Men
(1)
(2)
Remaining teeth
Consumption
Women
(3)
(4)
Remaining teeth
Consumption
35-39
0.0073∗∗∗
(0.0007)
0.0131∗∗∗
(0.0033)
0.0078∗∗∗
(0.0007)
-0.0119∗∗∗
(0.0033)
40-44
0.0111∗∗∗
(0.0007)
0.0256∗∗∗
(0.0034)
0.0114∗∗∗
(0.0007)
0.0152∗∗∗
(0.0034)
45-49
0.0169∗∗∗
(0.0007)
0.0345∗∗∗
(0.0033)
0.0150∗∗∗
(0.0007)
0.0352∗∗∗
(0.0034)
50-54
0.0247∗∗∗
(0.0007)
0.0439∗∗∗
(0.0035)
0.0220∗∗∗
(0.0008)
0.0470∗∗∗
(0.0037)
55-59
0.0310∗∗∗
(0.0007)
0.0605∗∗∗
(0.0034)
0.0284∗∗∗
(0.0008)
0.0563∗∗∗
(0.0035)
60-64
0.0356∗∗∗
(0.0005)
0.0741∗∗∗
(0.0025)
0.0322∗∗∗
(0.0005)
0.0590∗∗∗
(0.0025)
Notes: Results for the patient sample. Column (1) and (3) report estimates from a log-linear
Poisson regression model of no. remaining teeth on log. lifetime incomes. Column (2)
and (4) report OLS estimates from regressing log. dental care consumption on log. lifetime
income. All models are estimated separately age-groups and gender. Age-specific fixed
effects included. Standard errors in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Table A 5. Estimated impact of income on dental care consumption, by age-groups.
All individuals in patient sample with non-missing health measures
Age-group
(1)
(2)
35-39
∗∗∗
-0.0273
(0.0024)
-0.0070∗∗∗
(0.0023)
40-44
-0.0128∗∗∗
(0.0024)
0.0066∗∗∗
(0.0023)
45-49
-0.0013
(0.0024)
0.0160∗∗∗
(0.0023)
50-54
0.0072∗∗∗
(0.0025)
0.0225∗∗∗
(0.0024)
55-59
0.0291∗∗∗
(0.0024)
0.0445∗∗∗
(0.0024)
60-64
0.0533∗∗∗
(0.0017)
0.0639∗∗∗
(0.0017)
Health
No
Yes
Notes: Results for individuals in the patient sample with non-missing health measures (94%
of patient sample). The outcome variable is log. dental care consumption. Age-specific
fixed-effects included. Standard errors in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
53
Table A 6. Estimated impact of income on dental care consumption by age-group and
gender.
Men
Age-group
(1)
Women
(2)
∗∗∗
(3)
∗
(4)
∗∗∗
35-39
-0.0200
(0.0035)
-0.0057
(0.0034)
-0.0393
(0.0034)
-0.0177∗∗∗
(0.0033)
40-44
-0.0104∗∗∗
(0.0035)
0.0042
(0.0034)
-0.0130∗∗∗
(0.0034)
0.0057∗
(0.0033)
45-49
-0.0025
(0.0034)
0.0114∗∗∗
(0.0033)
0.0060∗
(0.0035)
0.0198∗∗∗
(0.0034)
50-54
0.0005
(0.0035)
0.0124∗∗∗
(0.0034)
0.0185∗∗∗
(0.0037)
0.0316∗∗∗
(0.0036)
55-59
0.0238∗∗∗
(0.0035)
0.0352∗∗∗
(0.0034)
0.0324∗∗∗
(0.0035)
0.0464∗∗∗
(0.0035)
60-64
0.0468∗∗∗
(0.0025)
0.0551∗∗∗
(0.0025)
0.0443∗∗∗
(0.0025)
0.0533∗∗∗
(0.0024)
Health
No
Yes
No
Yes
Notes: Results for individuals in the patient sample with non-missing health measures (94%
of patient sample). The outcome variable is log. dental care consumption. Age-specific
fixed-effects included. Models are estimated separately by age-group and gender. Standard
errors in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
54
A 2 Controlling for health non-parametrically using polynomials
Figure A 1 and A 2 plot OLS estimates of θ1 from fitting the model
ci = θ0 + θ1 Yi + f (Hi ) + i ,
(1.8)
where f (Hi ) is given by polynomials of Hi . I estimate model (1.8) using second
through fourth degree polynomials of oral health. The results are qualitatively similar
to the main results presented in section 1.6.1. The estimates are also presented in table
A 7 for all patients with non-missing oral health measures and for men and women
separately in table A 8 and A 9 respectively.
Impact of income on dental care consumption
0
.02
.04
.06
.08
Figure A 1. Estimated impact of income on dental care consumption, by age-group.
Models with up to 4th order polynomials of oral health.
−.02
Polynomial for health:
2nd order
3rd order
4th order
35−39
40−44
45−49
50−54
55−59
60−64
Age−group
Notes: Results for individuals in the patient sample. The outcome variable is log. dental care
consumption. Models include up to 4th order polynomial of oral health, Hi , age-specific fixed
effects and estimates are plotted with 95 % CIs.
55
Figure A 2. Estimated impact of income on dental care consumption, by age-group
and sex. Models with up to 4th order polynomials of oral health.
Male
.04
.02
0
Impact of income on dental care consumption
.04
.02
0
Impact of income on dental care consumption
.06
.06
Female
Polynomial for health:
2nd order
4th order
−.02
−.02
3rd order
35−39
40−44
45−49
50−54
55−59
60−64
Age−group
35−39
40−44
45−49
50−54
55−59
60−64
Age−group
Notes: Results for individuals in the patient sample. The outcome variable is log. dental care
consumption. Models include up to 4th order polynomial of oral health, Hi , age-specific fixed
effects and estimates are plotted with 95 % CIs.
Table A 7. OLS estimates of the effect of income on dental care consumption, by
age-group. Model with polynomials of oral health.
Health enters model with polynomial up to:
2nd order
3rd order
4th order
Log. disposable income
0.03088∗∗∗
(0.00121)
0.03266∗∗∗
(0.00121)
0.03307∗∗∗
(0.00121)
Number of observations
3,001,589
3,001,589
3,001,589
Notes: Results for individuals in the patient sample with non-missing health measures (94%
of patient sample). The outcome variable is log. dental care consumption. Age-specific
fixed-effects included. Models are estimated separately by age-group and gender. Standard
errors in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
56
Table A 8. OLS estimates of the effect of income on dental care consumption for men,
by age-group. Model with polynomials of oral health.
Health enters model with polynomial up to:
3rd order
4th order
Age-group
2nd order
35-39
-0.00002
(0.00511)
0.00596
(0.00510)
0.00621
(0.00510)
40-44
0.01488∗∗∗
(0.00486)
0.02078∗∗∗
(0.00485)
0.02086∗∗∗
(0.00485)
45-49
0.02382∗∗∗
(0.00462)
0.02669∗∗∗
(0.00461)
0.02682∗∗∗
(0.00461)
50-54
0.02826∗∗∗
(0.00470)
0.02814∗∗∗
(0.00469)
0.02818∗∗∗
(0.00469)
55-59
0.04367∗∗∗
(0.00446)
0.04274∗∗∗
(0.00446)
0.04243∗∗∗
(0.00446)
60-64
0.05910∗∗∗
(0.00305)
0.05820∗∗∗
(0.00305)
0.05776∗∗∗
(0.00305)
Notes: Results for individuals in the patient sample with non-missing health measures (94%
of patient sample). The outcome variable is log. dental care consumption. Age-specific
fixed-effects included. Models are estimated separately by age-group and gender. Standard
errors in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Table A 9. OLS estimates of the effect of income on dental care consumption for
women, by age-group. Model with polynomials of oral health.
Health enters model with polynomial up to:
3rd order
4th order
Age-group
2nd order
35-39
-0.00365
(0.00496)
0.00460
(0.00495)
0.00463
(0.00495)
40-44
0.01705∗∗∗
(0.00478)
0.02229∗∗∗
(0.00477)
0.02233∗∗∗
(0.00477)
45-49
0.04141∗∗∗
(0.00465)
0.04485∗∗∗
(0.00464)
0.04485∗∗∗
(0.00464)
50-54
0.04382∗∗∗
(0.00489)
0.04387∗∗∗
(0.00488)
0.04386∗∗∗
(0.00488)
55-59
0.04907∗∗∗
(0.00451)
0.04878∗∗∗
(0.00450)
0.04882∗∗∗
(0.00450)
60-64
0.05784∗∗∗
(0.00296)
0.05738∗∗∗
(0.00296)
0.05762∗∗∗
(0.00295)
Notes: Results for individuals in the patient sample with non-missing health measures (94%
of patient sample). The outcome variable is log. dental care consumption. Age-specific
fixed-effects included. Models are estimated separately by age-group and gender. Standard
errors in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
57
A 3 Results with alternative definition of consumption
Table A 10. Estimated impact of income on consumption of non-acute dental care by
age-group and gender.
Men
Age-group
Women
(1)
(2)
(3)
(4)
35-39
-0.0054
(0.0036)
0.0094∗∗∗
(0.0035)
-0.0272∗∗∗
(0.0035)
-0.0049
(0.0034)
40-44
0.0020
(0.0035)
0.0174∗∗∗
(0.0034)
-0.0026
(0.0035)
0.0166∗∗∗
(0.0034)
45-49
0.0103∗∗∗
(0.0034)
0.0250∗∗∗
(0.0033)
0.0152∗∗∗
(0.0035)
0.0299∗∗∗
(0.0034)
50-54
0.0115∗∗∗
(0.0036)
0.0244∗∗∗
(0.0035)
0.0248∗∗∗
(0.0037)
0.0386∗∗∗
(0.0036)
55-59
0.0320∗∗∗
(0.0035)
0.0442∗∗∗
(0.0034)
0.0356∗∗∗
(0.0036)
0.0503∗∗∗
(0.0035)
60-64
0.0490∗∗∗
(0.0026)
0.0580∗∗∗
(0.0026)
0.0421∗∗∗
(0.0025)
0.0517∗∗∗
(0.0025)
Health
No
Yes
No
Yes
Notes: Results for individuals in the patient sample with non-missing health measures (94%
of patient sample). The outcome variable is log. consumption of non-acute dental care. Agespecific fixed-effects included. Models are estimated separately by age-group and gender.
Standard errors in parenthesis. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
58
Essay 2.
Price competition in Swedish dental care
Abstract This essay studies the effect of competition on prices on a health
care market where prices are market determined, namely the Swedish market
for dental care. The empirical strategy exploits that the effect of competition
differs across services, depending on the characteristics of the service. Price
competition is theoretically more intense for services such as examinations and
diagnostics (first-stage services), compared to more complicated and unusual
treatments (follow-on services). By exploiting this difference, I identify a
relative effect of competition on prices. The results suggest small but statistically
significant negative short-term effects on prices for first-stage services relative to
follow-on services. The results provide evidence that price-setting among dental
care clinics responds to changes in the market environment and substantial effects
of competition on prices over time cannot be ruled out.
I would like to thank Per Johansson and Erik Grönqvist for very valuable feedback. I would
also like to thank seminar participants at EARIE 2013, the Department of Economics at
Uppsala University, the Swedish Competition Authority, the CINCH Academy 2013 and the
ERSA Summer School 2012.
59
2.1 Introduction
Many OECD countries have undertaken market-oriented health care reforms over
the past decades (Gaynor, 2012). Some reforms have been motivated as a way of
increasing consumer choice. Others have been motivated as a means of cost control
since introducing market mechanisms into the provision of health care is assumed
to strengthen incentives for providers to become more efficient (Docteur and Oxley,
2003). At the same time, it is often argued that the special features of health care have
implications for the scope for competition.1 The development towards market-oriented
reforms therefore raises questions about when and how competition works on health
care markets. In this essay, I study the effect of competition on prices on the Swedish
market for dental care, a setting where patients have a substantial cost share (on average
about 80%) and prices are market determined.
One of the main challenges of the empirical literature on competition in health
care is to identify the effect of competition on prices (Gaynor and Town, 2011). The
challenge stems from the endogeneity of market structure, i.e. that the market structure
depends on factors that also affects prices. If not treated properly, this will yield biased
estimates of the effects of competition on prices. This essay adds to the existing
empirical literature by suggesting a strategy for identifying the effect of competition
on prices, which is grounded in the theoretical literature on competition in health care.
Identification is achieved by exploiting that competition theoretically has different
effects on prices for different types of services due to differences in consumers’ price
sensitivity.
The theoretical starting point for the identification strategy is recognizing health care
as a multiproduct industry. Following Pauly (1978), the products can be categorized
roughly as being either “informative” or “active-therapeutic”, where consumption of
the former precedes and determines demand for the latter. Consumption can thus be
viewed as taking place in two stages. The first—informative—stage consists of examinations and diagnostics (first-stage services) and the second—therapeutic—stage
consists of any subsequent treatments (follow-on services). In general, the same seller
provides both kinds of services. Consumers are on average better informed about
services in the first stage, as they are consumed with some frequency and are uncomplicated. Consequently, demand is more price sensitive for first-stage services compared
to follow-on services (Dranove and Satterthwaite, 1992, Pauly, 1978). Moreover, consumers who want to switch to a different provider in the second stage will face costs,
e.g. transaction costs for transferring medical records and search costs for finding a
new provider. This makes demand relatively less price sensitive in the second stage
compared to the first stage. In addition, switching costs make new consumers valuable
from the perspective of the seller, because of their future purchases (Klemperer, 1995,
Farrell and Klemperer, 2007). These arguments all suggests that sellers on health
care markets where prices are market determined will compete in prices for first-stage
services, whereas price competition is less fierce for follow-on services. This difference is exploited in the empirical strategy to identify the relative effect of competition
on prices for first-stage services relative to follow-on services. The identification can
1
Arrow (1963) argues that the medical industry differs sharply from the standard competitive
model and that medical care cannot be understood as other “economics textbook commodities”.
This view has been differentiated by e.g. Pauly (1978, 1988) and Shleifer (1998). The special
features of health care markets are also discussed in Dranove and Satterthwaite (2000).
60
be understood analogously to a difference-in-differences framework. Prices for both
kinds of services are effected by competition, but the magnitude of the effects should,
theoretically, differ across services. Focusing on the difference between the effects
across services allows identification as the differencing controls for the unobserved
factors that are correlated with both competition and prices.
Another important contribution to the literature is that the data used in this essay
comes from administrative registers covering all dental care consumed by adults in
Sweden during the years 2008–2011. This is a major improvement over the existing literature that mainly relies on imperfect price measures and markets defined by
geopolitical boarders that do not necessarily delineate the actual market (Dranove and
Satterthwaite, 2000). In Sweden, prices for dental care are set freely by providers as
opposed to negotiated by insurance plans (as in the US setting) or being regulated. I
use the actual price charged by the dentist for each service rather than constructed price
measures commonly used in the existing literature. Furthermore, the data contains the
coordinates of each clinic’s location. Competition is defined as the number of clinics
located within a fixed distance from each clinic.
Finally, by exploiting auxiliary data on the pricing of services, I perform policy
relevant simulations to assess the absolute effects of competition on prices. The
simulations build on the assumption that prices for follow-on services can be expressed
as a function of prices for first-stage prices. The assumption reflects that sellers are
relatively more constrained by competition in their price setting of first-stage services
and therefore may recoup foregone profits with high prices for the follow-on services.
The reduced form results show small but statistically significant effects of competition on prices for first-stage services relative to follow-on services. Competition is
measured as the number of clinics within a fixed distance from each clinic. The main
results suggest that a 1% increase in the number of clinics is followed by a 0.024%
decrease in prices for basic examination and diagnostics relative to tooth extractions.
In terms of dental clinics the estimate implies than an additional clinic within 1 kilometer reduces the price of first-stage services relative to follow-on services with about
0.1%. The results are robust across analyses of different kinds of services and types
of markets. The effects are small, but should be interpreted as short-term effects of
increased competition. Consequently, substantial effects of competition on prices over
time cannot be ruled out.
The simulation results suggest that a 1% increase in competition is followed by
a decrease in the price for first-stage services of about 0.05%. Relating this to the
number of times the first-stage services are performed on average during a year, the
effects corresponds to a redistribution from providers to patients in the range 1.6-3.9
million SEK. The price decrease for follow-on services fall in the range 0.02% to 0.05%,
corresponding to a redistribution from providers to patients in the range 700,000 SEK
to 1 million SEK. This implies that prices on the Swedish dental care market are set
above the competitive level and that increased competition would increase consumer
welfare by lowering prices. Moreover, the results suggest that price-setting among
dental care clinics indeed responds to changes in the surrounding market environment.
The remainder of the paper is structured as follows: section 2.2 gives a background
to the empirical literature on competition in health care. Section 2.3 gives a theoretical
background to the empirical strategy. Section 2.4 gives an overview of dental care
in Sweden and the data, section 2.5 describes the empirical strategy and, section 2.6
61
presents the results, a set of robustness tests and policy simulations. In section 2.7, I
conclude.
2.2 Competition in health care
There has been a rapid and steady growth in health care spending during the past
century throughout all OECD countries.2 In the 1960s and 1970s, government efforts
to control costs were aimed at the macro level, i.e. by rationing through regulation of
prices and volumes. However, during the 1980’s the focus of cost containment policies
started to shift towards the micro level, i.e. by encouraging more efficient provision
rather than cuts in entitlements (Docteur and Oxley, 2003). Markets therefore play an
important role today in the provision of health care services.3 Prices, qualities4 and
quantities are thus set through interactions in the market place between providers and
payers (Gaynor and Town, 2011). Consequently, there is an increasing literature on
the industrial organization of health care.
2.2.1 Empirical literature on price competition
The most active area in the literature on competition in health care is focused on
hospital competition in the United States, more precisely the relationship between
hospital concentration and prices. The recent empirical literature is surveyed in Gaynor
and Town (2011) and general finding is that prices on average are higher on hospital
markets with little competition.5 However, as discussed by Capps and Dranove (2004)
and Gaynor and Town (2011), there are caveats with the existing literature and the
results should therefore be interpreted with some caution.
Firstly, the price variables are often constructed, aggregated measures and not actual
transaction prices.6 This may be especially problematic in the US setting, where prices
are set in complicated hospital-insurer contracts (Gaynor and Town, 2011) and public
and private payers pay different prices for the same services (Dranove et al., 1993).
Without the possibility to control for these differences, the results may be misleading.
Secondly, a common approach has been to use zip-codes to create measures of concentration or delineating markets by geopolitical boarders (Dranove and Satterthwaite,
2000). Failing to take the actual distance between providers into account may lead to
either defining the market as too narrow or too wide.7
2
3
4
5
6
7
Several explanations to the health care spending growth has been put forth in the literature,
the most common being technological change combined with increased coverage (Chernew
and Newhouse, 2011).
The US has the longest experience of competition in health care. Market-oriented reforms
has been implemented or are considered in the U.K., Sweden, France, Germany and the
Netherlands (Gaynor, 2012).
Quality is here referred to as any non-price attributes.
Similar results are found in earlier reduced form studies (Noether, 1988, Dranove et al., 1993).
See for example Capps and Dranove (2004); Dranove et al. (1993); Noether (1988); Thompson
(2011).
Two providers can be in different zip-codes but still be very close to each other and thus
compete for the same patients. On the other hand all care givers in one geographical area, say
state, are not necessarily competing with each other.
62
Thirdly, the endogeneity of market structure needs to be addressed explicitly in order
to estimate a causal effect of competition on prices.8 Concerns about the endogeneity
are common to all empirical approaches to estimate the effect of concentration on prices
(Gaynor and Town, 2011). Attempts have been made to mitigate these concerns.
However, they have been focused on improving concentration measures by adding
structure9 rather than finding reduced-form approaches to identify the effect under
weaker assumptions..10
The market for dental care is arguably more similar to the market for physician
services than the hospital market. A patient consulting a physician is generally looking
to resolve uncertainty (Dranove and Satterthwaite, 2000) rather than undergoing a
certain treatment. This also applies to dental care. Empirical work on competition and
prices on physician service markets is however sparse11 , mainly due to lack of data.
Instead, most part of the literature on physician service markets has focused on issues
concerning asymmetric information and agency in the relationship between the patient
and the physician (Gaynor and Town, 2011).
2.2.2 Economic studies on dental care
Very little economic research has been done on dental care and even less on the effects
of competition. The focus of previous studies has mainly been on utilization and
estimating demand elasticities. Estimates of income and price elasticities generally
has the expected signs, but vary in size across studies.12
Grytten and Sørensen (2000) studies the short-term effect of a deregulation of the
fee system for dental services in Norway. They find that the mean expenditure per
consultation is not related to dentists’ subjective perception of the price elasticity
directed towards their own practice. This is interpreted as evidence against dentists’
exploiting their perceived market power. It is noteworthy that dentists’ subjective
estimates of price elasticities are lower on average than the estimated mean price
elasticity. The validity of both estimates are however questionable. First, it is unclear
what the dentists’ perceptions are based on. Second, it is unclear how the estimate
that Grytten and Sørensen (2000) call the “actual price” elasticity is identified. The
results should therefore be interpreted with caution.Grytten and Sørensen (2000) also
find that a 1% increase in dentist density is associated with a fall in mean expenditures
and fees in the range 0.12% to 0.31%. This is interpreted as competition having a
8
As discussed by Capps and Dranove (2004), some of the previous studies use cross-sectional
data (e.g. Staten et al. (1988) and Melnick et al. (1992)). Estimates based on cross-sectional
data cannot, in general, be given a causal interpretation because of unobservable characteristics
that may affect both concentration and prices.
9
Kessler and McClellan (2000) construct a hospital-specific concentration measure based on
patients’ choice of hospital as a function of travel distance. A similar approach is used in
Gowrisankaran and Town (2003).
10
It is noteworthy that the literature on the industrial organization of health care has moved
away from reduced form towards more structural approaches. This is also true for the broader
industrial organization literature. See Nevo and Whinston (2010) for further discussion.
11
There are examples of more structural approaches to investigate market structure and pricing
conduct of physician services. See for example Wong (1996) and Gunning and Sickles (2012).
12
See Sintonen and Linnosmaa (2000) for a review.
63
weak effect on dentists’ price setting. A concern with these results is that it is unclear
what drives changes in the density.
Eriksson (2004) is the only previous study of price setting in Swedish dental care.
The focus of Eriksson (2004) is price leadership rather than price competition per se.13
2.3 Theoretical framework
This section gives the theoretical rationale for the empirical strategy to identify the
effect of competition on prices. I start out with a discussion on assumptions about
the nature of competition and a description of the strategy outline. I then describe the
theoretical background in more detail.
The nature of competition is implicitly assumed to be monopolistic in the sense
that there is both an element of market power and an element of competition. Monopolistic competition has become a workhorse model for describing markets for health
care; while services are differentiated, consumers cannot perfectly observe the service
attributes. Providers therefore face a downward sloping demand, implying that they
can raise prices without losing all their patients to competing providers, selling similar
but not identical services (Dranove and Satterthwaite, 2000). However, an important
aspect is that consumers information regarding quality and price differs across types
of services (Pauly, 1978). Another important assumption for the empirical strategy is
that the number of competitors surrounding each clinic can be used as a measure of
competition. This is because an increase in the number of clinics in a given area will
provide consumers with greater choice, making the demand facing each clinic more
elastic. At the same time, it is reasonable to assume that the price elasticity of demand
differs across types of services (Pauly, 1978). More precisely, that demand is more
price sensitive for services that are consumed with some frequency, such as diagnostics
and examinations, compared to more uncommon therapeutic services. Competition
therefore has heterogeneous effects on prices. By exploiting both the variation in competition across clinics and the difference in consumers’ price sensitivity across services,
I identify a relative effect of competition on prices for diagnostic services compared
to therapeutic services. The identified effect is given by the difference between the
average effect of competition on prices for diagnostic services and therapeutic services
respectively. The purpose of the identification strategy is to address the endogeneity of
market structure, i.e. that market structure is correlated with unobserved factors that
affect prices. The endogeneity is driven by the fact that providers are not randomly
allocated across the country, but rather choose to establish in areas where demand and
prices are high. An observed correlation between prices and competition can therefore
not be given a causal interpretation. By focusing on the difference in the effect between
services across levels of competition, the unobserved endogeneity is differences away.
13
To test for price leadership, Eriksson (2004) estimates the effect of Public Dental Service
prices on prices of the private clinics for different kinds of treatments and finds that the
effect is positive and both statistically and economically significant. He finds support for the
hypothesis that the Public Dental Services acts as a price leader. The results also suggest
that the private dentists follow the public prices treatment by treatment, as opposed to across
treatments.
64
2.3.1 Consumer demand and price competition
The identification strategy can be understood analogously to a difference-in-differences
framework. Prices for both kinds of services are effected by competition, but theoretically the magnitude of the effects should differ across services. The rationale for the
strategy is best understood by considering how health care services can be characterized. An important feature of health care markets is that consumers buy information
in the form of diagnostic services and other assessments of their health status. This
information is then used as a basis for decision-making about future consumption
(Pauly, 1978). The consumption can thus be divided into two stages.
Following Pauly’s (1978) classification, services in the first stage can be characterized as being primarily “informative”, whereas services in the second stage can be
characterized as “active-therapeutic”. Hence, the first stage consists of examinations
and diagnostics (first-stage services) and the second stage consists of somewhat more
extensive treatments (follow-on services) if found necessary in the first stage. As
mentioned above, providers are assumed to have some market power, which is mainly
driven by consumers’ imperfect information. However, as first-stage services are consumed relatively frequent, consumers can become informed about these services at
a low cost (Pauly, 1978). On the other hand, most consumers purchase follow-on
services infrequently and becoming informed about these services are costly. Hence,
consumers are on average more able to evaluate prices for first-stage services compared
to follow-on services. In addition, evaluating prices for follow-on services is costly
and thus it is reasonable to assume that the price elasticity of demand is greater for
first-stage services than for follow-on services. Moreover, first-stage services is the
kind a service a patient would get when visiting a provider for the first time. These
services are therefore more likely to be subject to price comparisons among patients.
The difference in the price elasticity of demand across stages implies that price
competition will be more intense for first-stage services compared to follow-on services.
This argument is reinforced by considering that consumers will face a cost for switching
provider between the first and the second stage. This cost is over and above the search
cost for finding and evaluating a new provider in the second stage. The switching
cost may be either perceived or real, but have in common that consumers will find it
cheaper to buy all services from the same provider. Consumers would therefore become
“locked-in” after purchasing first-stage services from a given provider, which makes
demand for follow-on services less elastic. In addition, switching costs make new
customers valuable from the perspective of sellers, because of their future purchases
(Farrell and Klemperer, 2007, Klemperer, 1995). A standard results from models
of switching costs is therefore that sellers compete fiercely to attract new customers
and exploit locked-in customers, by charging higher prices (Padilla, 1991). These
arguments all suggest that price competition will be more intense for first-stage services
compared to follow-on services.
Consumers on health care markets face switching costs for several reasons. A
patient who wants to switch provider after being examined will face transaction costs,
because the patient will need either to be examined again or somehow transfer their
medical records. Switching costs in health care can thus be viewed as start-up costs for
establishing a new patient-provider relationship. Switching costs are also caused by
uncertainty of the quality (and price) of untested providers. Medical care and dental
care may be defined as experience goods (Nelson, 1970), in the sense that consumers
65
learns about quality only after consuming the service. Consumers can therefore be
viewed as facing a switching cost that is equal to what they at most would be willing
to pay to be guaranteed that the services offered by the new provider has the same
value to them as their current provider (Klemperer, 1995). Finally, there may be what
Klemperer (1995) calls psychological costs of switching, or non-economic brandloyalty. These kinds of switching costs relates to what Samuelson and Zeckhauser
(1988) calls ”status quo bias” in decision making, referring to the inclination of
sticking to a previous decision, i.e. the status quo. Samuelson and Zeckhauser (1988)
and Strombom et al. (2002) find evidence of status quo inertia when studying individual
health care plan choices among employees at Harvard University and University of
California respectively.
A standard framework in the theoretical literature on the implications of consumer switching costs is a two-period model (Gehrig and Stenbacka, 2002, Klemperer,
1987b,a, 1995, Padilla, 1992). Consumers enter the market and make their purchase
in the first period. Once consumers have chosen a provider, they face switching costs
and hence become locked-in (or at least attached). The main result from these models
is that firms compete fiercely in the first period to attract new customers and exploit
locked-in customers in the second period, by charging higher prices (Padilla, 1991).
The differing degree of competition across periods gives rise to a pricing schedule
that follows a pattern of introductory offers, sometimes also referred to as a bargainthen-ripoffs structure.14 The pattern is clearest when sellers can distinguish between
new and old customers (Farrell and Klemperer, 2007). In the case of health care markets, sellers can clearly distinguish between customers in different stages, depending
on what services they are purchasing. In a two-period model of switching costs, a
consumer purchasing informative services would be a ”new” customer. On the other
hand, an individual that has already been examined and is about to undergo some
therapeutic service, is ”locked-in”. Another result from the core two-period model is
that market shares become valuable and a determinant of future profits, because of
locked-in customers’ repeated purchases.
The firm’s problem is to set prices in the first period such that total discounted
profits are maximized.15 Hence, the firm takes both current-period profits and the
effect of current-period markets shares on future profits into account when setting
first-period prices.16 Since firms take overall profits into account, they are even willing
to price below costs in the first period, as foregone profits can be recouped in the
second period (Farrell and Klemperer, 2007). Beggs and Klemperer (1992) model
pricing with switching costs on a growing market in a multi-period setting, where the
number of consumers increases for every time period. With new customers entering
the market, the proportion of locked-in customers is reduced and competition for new
customers is intensified. The intuition is that a steady growth in market size makes
the future relatively more important, because the amount of new customers that can
14
It is shown in Klemperer (1987b) that the effect of switching costs on competition depends
on consumers’ expectations about prices. If consumers have rational expectations, they will
recognize that low prices today will be followed by high prices tomorrow. Foreseeing this will
thus make customers less sensitive to price cuts or introductory offers. Note that the overall
effect of switching costs on competition is ambiguous, as the tough first-period competition
may be offset by the relaxed competition in the second-period.
15
See Farrell and Klemperer (2007) for a review of the literature on pricing with switching costs.
16
See Klemperer (1995) for a thorough discussion of the model.
66
be locked-in (and exploited) is increasing. The market size in terms of the number of
competing firms also has implications for penetration pricing. This can be illustrated
with an example of a monopolist; if consumers have nowhere else to go there is no
role for penetration pricing or introductory offers (Farrell and Shapiro, 1988). In sum,
the effect of switching costs on current-period prices is intensified by the number of
competing firms. The difference in the pricing schedule between first-stage services
and follow-on services therefore increases as the number of competing firms increases.
2.3.2 Summary
The theoretical framework outlined above offers two interrelated reasons for why price
competition is more intense for diagnostic services compared to therapeutic services.
First, consumers have more information about first-stage services compared to followon services as the former is purchased with some frequency. Consumers’ demand for
first-stage services is therefore relatively more price sensitive. Second, consumers face
costs for switching provider between the first and second stage and thus find it cheaper to
buy all services from the same provider. Consequently, demand for follow-on services
is relatively less price sensitive. New customers are valuable from the perspective of
the provider, because of their follow-on purchases. Naturally, therapeutic services will
on average be more expensive than diagnostic services, because they are generally more
complicated and time consuming. However, it follows from the reasoning in 3.1 that
the price difference between services will increase with the intensity of competition.
This is because i) competition will have greater effects on services for which demand is
sensitive to price and ii) it becomes more important for providers to lock in consumers
as the number of competitors increases. Thus, changes in competition would be
reflected by changes in the price difference across services.
2.4 Institutional setting and data
Dental care in Sweden is provided through both private clinics and the Swedish Public
Dental Service, which is the county councils’17 dental care organization. The majority
(60–80%) of dental care for individuals aged 20 and above18 is supplied by private
clinics. Patients can freely choose either public or private clinics as providers and price
setting is free. The Swedish dental care system was reformed in 2008. The system now
contains a general dental care subsidy, which applies to all citizens aged 20 years and
above, and a high-cost protection plan. The size of the subsidy covers preventive dental
care and dental care that reduce pain and enables the patient to eat, chew and speak
without impediment. The subsidy is 300 SEK/year for individuals aged 20–29 years
or above 74 years and 150 SEK/year for individuals aged 30–74 years. Patients are
dental care costs are subsidized at 50% through the high-cost protection plan between
17
Sweden is divided into 21 counties at the regional level. The county councils’ main responsibility is health and medical service (about 80% of total expenditures). The Swedish Public
Dental Service has a legal responsibility for ensuring the supply of dental care to the citizens
in the county (National Dental Service Act, tandvårdslagen (1985:125)).
18
Dental care is free for children and young people aged 19 or under.
67
SEK 3 001 and SEK 15 000 during (at most) a twelve-month period. For costs above
SEK 15 000 the subsidy is 85%.
The Dental and Pharmaceutical Benefits Agency (TLV) decides which procedures
are subsidized and determines a reference price list, which serves as a basis for calculating the size of the subsidy and reimbursements within the high-cost protection plan.
The reference price list is revised every year based on general cost trends in dentistry
with regard to technological developments such as new treatments and changes in the
use of materials. The reference prices are supposed to reflect a “normal price” for
each treatment, that reflects actual costs (SOU 2007:19, 2007). The dental care system
is administered by the Swedish Social Insurance Agency (SSIA), who also hold the
Dental Care Register. Since the dental care subsidy and the high-cost protection plan
applies to all dental care, both publicly and privately provided dental care is in the
register.19 Subsidies in the high-cost protection plan are calculated over the consumption during a twelve-month period. Therefore, all dental care is registered, even if an
individual does not reach the first threshold in the high-cost protection. Hence, the
register also covers dental care that is fully paid for by the patient.20
19
All public clinics and about 96% of the private clinics are connected to the dental care system.
Little is known about the 4% of the private clinics that operate entirely outside the national
dental care system.
20
Apart from purely aesthetic dental care, which is never subsidized.
68
Table 2.1. Summary statistics for dental care in Sweden June 2008-December 2010.
Patients aged 20 years and above.
Number of clinics
All
Private
Public
Number of patients by type of clinic
All
Private
Public
Charged price per service
Mean
Charged price per visit
Mean
Patient’s price per visit
Mean
Number of patients per day per clinic
Mean
Number of patients per year per clinic
Mean
2008
Year
2009
2010
3,472
2,726
746
3,762
2,932
830
3,656
2,877
779
2,370,182
1,417,200
952,982
3,976,166
2,360,211
1,615,955
4,091,337
2,419,277
1,672,060
750.1
(993.6)
832.6
(1,419.2)
844.8
(1,375.0)
1,327.1
(2,121.2)
1,573.5
(3,123.8)
1,583.4
(3,085.3)
1,023.4
(1,079.6)
1,036.2
(1,276.4)
1,035.3
(1,265.0)
All
Clinic type
Private
Public
26.0
(145.4)
21.3
(143.6)
46.3
(151.2)
3,697.7
(22,570.4)
2,806.7
(22,877.1)
7,596.3
(20,740.4)
Notes: Some clinics are categorized as neither private nor public in the register. These
clinics are excluded. Standard deviation in parenthesis.
2.4.1 Data and descriptive statistics
The data comes from the Dental Care Register at The Swedish Social Insurance Agency
linked to geographical variables collected at Statistics Sweden. The Dental Care
Register covers all dental care produced at clinics that are subscribed with the dental
care benefit system (≈96% of all clinics in Sweden). The Dental Care Register contains
information about diagnosis, treatment items and prices among other variables. The
price variables include the price charged by the dentist for each service, the gross price
for all treatment items in one visit and the total price paid by the patient, i.e. the
price net of allowance and any reimbursement. The Dental Care Register covers over
47 million treatments during the studied period. The geographic variables contain
information about the clinics’ location, given by the midpoint coordinate of a 100 x
100 meter square around the clinic. The coordinates are defined within the national
69
horizontal reference system RT 90. All data covers the period July 2008–December
2010.
Table 2.1 provides some summary statistics for dental care in Sweden during the
studied period June 2008–December 2010. There are roughly four times as many
private clinics as public clinics and consequently, the majority of patients visit private
clinics. It is noteworthy that the Swedish Dental Service, i.e. the public clinics, is
responsible for children’s dental care. However, the data only covers patients aged 20
years and above. The total number of patients for public clinics is therefore understated.
2.4.2 Operationalization of variables
The measure of competition is constructed by counting all clinics21 within a certain
distance from the clinic. In the baseline model, I use 1 km as a distance, but for
robustness checks I increase the distance to 5 km. Table 2.2 provides descriptive
statistics for the competition measures. The means are the number of clinics within
the specified distance. There is large variation and the clinics with most competitors
within 1 kilometer are located in big cities.
Table 2.2. Competition measure, means.
Variable
Mean
SD
Max
Competition measure, all clinics
Clinics within 1 km
Clinics within 5 km
Clinics within 10 km
17.4
73.3
127.2
33.4
133.7
214.6
218
520
731
Clinics within 1 km, by geographical area
Big city
Mid city
Non-metropolitan area
53.8
14.1
3.4
57.2
11.8
3.6
218
49
27
Notes: SD = Standard deviation. Definitions of geographical areas are given in the appendix.
Table 2.3 summarize mean prices and occurrences for services included in the
analysis. First-stage services are defined as the kind of service a patient generally
would get during a routine check-up or when visiting the dentist for the first time.
The first-stage services are therefore defined as examinations performed by a dentist
or a dental hygienist. The follow-on services are defined as a somewhat complicated
treatment for which the patient has a diagnosed need, e.g. a cavity or a bacterial
infection. All prices are in SEK.22 As expected, the first-stage services are on average
cheaper than the follow-on services. Only services that account for at least 0.5% of the
total amount of services in the register are included.23 One common service that has
21
Clinics are defined as the unique combination of an organizational number and an establishment
number (CFAR) at a given coordinate. The identity numbers for clinics (mottagningsnummer)
from SSIA are not used in this definition. This is done to minimize the risk of incorrectly
defining clinics with more than one identification number as different clinics.
22
10 SEK ≈ EUR 1.1 in November 2015.
23
There are over 40 million services registered during the period July 2008–December 2010.
70
not been included in the analysis are fillings. The reason is that it is not straightforward
whether filling a cavity should be viewed as a follow-on service or not. Even though
it clearly is a treatment preceded by an examination, an uncomplicated filling may be
performed in connection with an examination performed by a dentist.24
Table 2.3. Prices and occurrences for first-stage and follow-on service.
Type of service
Mean
SD
Min
Max
Occurrences
First-stage services
Examination by dentist
Examination by dental hygienist
671.5
613.6
65.0
65.1
578
390
885
777
6,862,693
2,194,725
815.9
2,440.0
1,534.4
76.8
298.7
128.9
690
1,894
1,100
1,114
3,232
2,025
886,858
270,963
232,525
Follow-on services
Tooth extraction, one tooth
Root canal, one filling
Clinic-made crown
Notes: SD = Standard deviation. All prices in SEK. 10 SEK ≈ 1.2 EUR in April 2013.
In the main analysis, I use basic examination and tooth extractions. The reason
is that these services are the most common first-stage service and follow-on service
respectively. The other services listed in table 2.3 are used to perform sensitivity
analyses.
Basic examination performed by a dentist is the single most common service in the
register and accounts for about 17% of all services. It contains basic diagnostics and
minor treatments such as removal of dental calculus and fluoride rinsing. Full examination performed by dental hygienist is a somewhat more comprehensive examination,
as it also contains an assessment of the patients’ general health status and habits. A
patient may be called back for an examination by a dentist to further evaluate findings
made by the dental hygienist. However, the most common procedure is that patients
with e.g. indications of caries, are called back to see a dentist for treatment rather than
re-examination.
Tooth extraction simply refers to the removal of a tooth. Teeth may be removed with
different techniques and for several reasons, e.g. tooth decay, fractions or infections.
Non-surgical tooth extraction, which is the most common type of extraction in the
register, is used in the analysis. Crown therapy is used to restore teeth. A crown
encircles or caps the tooth and is used when the tooth is severely damaged. Clinicmade crowns are made in plastic material.25 Root canal therapy refers to treatments of
the pulp or the tissue surrounding the root. It is used to treat bacterial infections that
may occur due to e.g. deep cavities or fractures.
24
Other services among the top 0.5% that have been excluded are acute examinations, x-rays
(apart from pictures taken during examinations), information and instruction for patients at risk
for e.g. caries, professional tooth cleaning in connection with examination and non-surgical
periodontal therapy (somewhat more complicated dental plaque removal).
25
Another kind of crown therapy is “Laboratory-made crown” (service no. 801), which is a
crown made by using a methodology called CAD/CAM (computer-aided design/computeraided manufacturing). Laboratory-made crowns may in some cases be replaced by implant
therapy, which is in general more expensive than 801. This has led to misreporting in the
price variable for 801; the service type is registered as 801 whereas the prices are for the
replacement services.
71
2.5 Empirical strategy
In the following, I discuss the empirical strategy and how the effect of competition on
prices is identified.
Suppose that the relationship between prices and competition could be described
by the following simple model:
log ptjl = α + β log ctj + utjl
(2.1)
where log ptjl is the log price charged by clinic j, in area l,in time period t and log ctj
is log competition. The parameter of interest is β, capturing the effect of competition
on prices. Simply estimating β in model (2.1) by running a regression is not sufficient
to establish a causal relationship. The identification problems can be illustrated by
decomposing the error term into three parts; utjl = ηl + atj + εtjl . The term ηl
captures overall demand in area l. It is correlated with ctj if there, for example,
are more clinics in areas with higher incomes.26 The term atj captures demand for
both types of services facing clinic j, and changes in demand over time. Given that
high demand is reflected by high prices, atj is correlated with ctj if clinics choose to
establish where prices historically have been high. The potential relationship can be
expressed as atj = f (p(t−1,j) ). The implication is that ctj is not strictly exogeneous.
The last part of the error term, εtjl , is an idiosyncratic error. Estimating (2.1) with
OLS will yield biased estimates of β.27 Since ηl and atj are unobservable parts of the
error, they cannot fully be "controlled away". However, I can use that β in (2.1) differs
across services as discussed in section 2.3.
Consider two kinds of services, k = 1, 2. Where k = 1 represents the first-stage
service and k = 2 represents the follow-on service. We can write:
ptjl1 = β1 + β11 cjt + utjl1
ptjl2 = β2 + β12 cjt + utjl2
(2.2)
(2.3)
where the outcome variable, log ptjlk , is the mean log price of treatment k at clinic j,
in are l at time period t. βk captures the average price for services of type k. With a
log-log specification, β1k captures the average elasticity of prices for services of type
k with respect to competition. Because of difference in consumers’ price sensitivity
across service types it is assumed β11 > β12 , reflecting that the average effect of
competition on prices is larger for the first-stage services compared to the follow-on
services. The error term, utjlk can still be thought of as consisting of three parts.
The idiosyncratic error term εtjlk , is service-specific. However, overall demand ηl
and changes in overall demand, atj , are common to both the first-stage service and the
follow-on service. Subtracting (2.2) from (2.3) gives:
log ptjl1 − log ptjl2 = δ1 + δ2 log cjt + εjt1 − εjt2
(2.4)
where δ1 = β1 − β2 and δ2 = β11 − β12 . δ2 is the relative elasticity of prices with
respect to competition, for first-stage services versus follow-on services. By taking
26
This is the case if the income elasticity is positive. Tentative results suggests a positive income
elasticity for dental care (Grönqvist, 2012, Manning and Phelps, 1979, Holtmann and Olsen Jr,
1976).
cov(utjl ,ctj )
27
This is because cov(utj , ctj ) = 0 and we get: p lim β̂ − β = var(u
>0
tjl )
72
the difference over service types for given levels of competition, ηl and atj have been
differenced away. Hence, the identifying assumption is that the unobserved factorsηl
and atj are only associated with the level of competition. This assumption is consistent
with clinics establishing in areas where demand for first-stage services are high. Note
that the competition measure is general to all services as it captures the number of
clinics within a certain distance.
When ηl and atj have been differenced away, we now only have the idiosyncratic
error terms,εtjlk . I can therefore estimate (2.4) with OLS, adding a dummy d for the
first-stage service:
log ptjkl = α0 + α1 d + α2 log cjt + α3 d log cjt + γj + εtjlk
(2.5)
Model (2.5) exploits variation in competition across clinics. α1 captures the average
association between k = 1 and price and α2 captures the average association between
competition and price. The parameter of interest is α3 capturing the difference across
services in the average effect of competition on prices (β11 − β12 ) interacted with
competition. Following the reasoning in the theoretical background, α3 is hypothesized
to be negative, reflecting that competition has a relatively larger effect on first-stage
services compared to follow-on services.
Apart from the identification issues discussed above, there may also be a clinic
specific component of the error, γj . By further adding the year specific effect γt , we
get:
log ptjkl = α0 + α1 d + α2 log cjt + α3 d log cjt + γj + γt + εtjlk
(2.6)
The variation used in the estimation of model (2.6) is the within-clinic variation in
competition. The identifying assumption is that the evolution of the price difference
between services on the clinic level would have been unchanged in the absence of a
change in competition. Given that the unobserved part of the error term is indeed
associated with the level of competition rather than clinic specific characteristics or
some year specific shock common to all clinics, models (2.5) and (2.6) should yield
quantitatively similar results. This is because the differencing across services solves the
fundamental market endogeneity problem and therefore the cross-sectional variation
in competition should be sufficient to identify α3 .
2.6 Results
2.6.1 Main results
This section presents the results from the baseline estimation. The sample used in
the analysis contains clinics that have performed both the first-stage service and the
follow-on service during the same year. In addition the sample is restricted to be
balanced over time.28 The estimation sample differs depending on what services are
included in the analysis.
Table 2.4 gives the main results. The first column gives the pooled OLS estimates
of the baseline model (2.5) where the only included control variable is a public clinic
28
Including all clinics does not change the results qualitatively.
73
dummy. The results from the pooled OLS-estimate (column 1) suggest that a 1%
increase in competition is followed by a 0.024% decrease in the price for “Basic
examination & diagnostics” relative to “Tooth extraction”. This estimate can be related
to the mean of clinics within 1 km, which is 22 for the estimation sample. An increase
in competition with one new clinic corresponds to a 4.5% increase in the competition
measure. Thus, one extra clinic within 1 km is followed by an increase in the price
difference between basic examinations and tooth extractions with 0.11% (≈ 4.5 ×
0.024).
The point estimates does not change when adding clinic fixed effects (column 2) and
a time-dummy (column 3) capturing year specific shocks, but precision is increased.
The estimated effect of competition on prices from the pooled OLS is thus not driven
by unobserved heterogeneity across clinics or year specific factors. However, this is not
surprising, as the differencing across service types, described in section 2.5, removes
clinic specific factors that are common to all services.
Table 2.4. Tooth extraction vs. Basic examination & diagnostics, performed by dentist.
(1)
k×Competition 1 km
(2)
∗∗∗
(3)
-0.02318
(0.00144)
-0.02318
(0.00436)
-0.02318∗∗∗
(0.00436)
No
No
Yes
No
Yes
Yes
12,350
12,350
12,350
Clinic FE
Year dummy
Number of observations
∗∗∗
Notes: Standard errors in parenthesis. The pooled model in column (1) is estimated with a
public clinic dummy. Standard errors in column (2) and (3) clustered at the clinic level. All
models weighted with the number of patients per year at each clinic. ∗ p < 0.1, ∗∗ p < 0.05,
∗∗∗
p < 0.01
Table 2.5. Tooth extraction vs. Basic examination & diagnostics, performed by dentist.
k×Competition 5 km
(1)
(2)
(3)
-0.01037∗∗∗
(0.00111)
-0.01037∗∗∗
(0.00346)
-0.01037∗∗∗
(0.00346)
No
No
Yes
No
Yes
Yes
13,700
13,700
13,700
Clinic FE
Year dummy
Number of observations
Notes: Standard errors in parenthesis. The pooled model in column (1) is estimated with a
public clinic dummy. Standard errors in column (2) and (3) clustered at the clinic level. All
models weighted with the number of patients per year at each clinic. ∗ p < 0.1, ∗∗ p < 0.05,
∗∗∗
p < 0.01
2.6.2 Sensitivity analysis
To test the sensitivity of the results to functional form the same models have been
estimated in levels instead of logs for both “Basic examination & diagnostics” relative
74
to “Tooth extraction” and “Full examination, performed by dental hygienist“ relative
to “Tooth extraction”. The estimates are negative and given in table B1 and B2 in
Appendix B. As a further sensitivity analysis, I re-estimate the models using clinics
within five kilometers as a measure of competition. The point estimates decreases, but
the results do not change qualitatively. The results are given in table 2.5.
The models have also been estimated separately for different types of geographical
areas. The elasticites are -0.04% for “Metropolitan municipalities” and -0.017% for
“Large cities”. These areas are defined as municipalities with a population of over 200
000 inhabitants and a population of 50 000–200 000 respectively. The point estimate
for “Forest counties” is -0.026%, defined as sparsely populated counties (consisting of
several municipalities) with a large proportion of forest land. Municipalities that does
not fit into any of the aforementioned categories forms the group “All other areas”.
The elasticity for this group, consisting of about half of all municipalities, is -0.019%.
Definitions of the geographical areas and tables are given in Appendix B.
Table 2.6 presents results where the first-stage service is defined as “Full examination, performed by dental hygienist“. A 1% increase in competition is followed by
a 0.03% decrease in the price for “Full examination” realtive to “Tooth extraction”.
The mean of clinics within 1 km is 17 for the sample used in the analysis. Relating
the point estimate to the mean number of clinics gives that one extra clinic within 1
km is followed by an increase in the price difference between full examinations and
tooth extractions with 0.17% (≈ 5.8x0.03). The sample is about half of that used in
the baseline model. This can be explained by the fact that auxiliary personnel such as
dental hygienists are much more common in public clinics. Therefore, many private
clinics are excluded from the estimation sample.
Table 2.6. Tooth extraction vs. Examination by dental hygienist.
(1)
k×Competition 1 km
Clinic FE
Year dummy
Number of observations
(2)
∗∗∗
(3)
∗∗∗
-0.02842
(0.00237)
-0.02842
(0.00495)
-0.02842∗∗∗
(0.00495)
No
No
Yes
No
Yes
Yes
5,384
5,384
5,384
Notes: Standard errors in parenthesis. The pooled model in column (1) is estimated with a
public clinic dummy. Standard errors in column (2) and (3) clustered at the clinic level. All
models weighted with the number of patients per year at each clinic. ∗ p < 0.1, ∗∗ p < 0.05,
∗∗∗
p < 0.01
The results from the analysis using root canal therapy and clinic-made crowns are
given in Appendix B table B7 and B8. The results are qualitatively similar to the main
results in table 2.4 and shows that an increase in competition with 1% is followed by a
decrease in the price for first-stage services relative to follow-on of -0.014% and -0.012
% respectively.
75
2.6.3 Policy simulations
The policy relevant effect is (mainly) the absolute effect of competition on prices,
rather than the relative effect. In this section, I asses the absolute effect by performing
policy simulations.
Recall that the relative effect, α3 is derived from the difference β11 − β12 , where
β1k is the average effect of competition on prices for services of type k, where k =
1 represents the first-stage service. As discussed above, firms compete to attract
customers who are in the first stage. When customers face switching costs, sellers
may try to recoup foregone profits from low prices for first-stage services with high
prices for the follow-on services. An interpretation of β11 and β12 is therefore that β11
depends on competition, whereas β12 is a choice variable for the firm. In other words,
the price for first-stage services is determined by competition with other firms, whereas
the price for the follow-on services is set such that overall profits are maximized. The
parameter β12 can thus be thought of as a function of β11 . This can be expressed in
the following simple way:
β12 = ρβ11 .
(2.7)
If ρ < 0, the negative effect of competition on prices for first-stage services is
offset by a positive price effect for follow-on services. In other words, firms increase
prices for follow-on services as competition increases. This would require substantial
switching costs, since firms otherwise would lose follow-on customers to competitors.
If ρ > 0, competition has a negative effect on prices for both kinds of services.
However, up to the point where ρ = 1, the negative effect is larger for first-stage
services. Note that if ρ = 1 the average effect of competition on prices β1k , does not
vary across services and hence α3 is not identified. From (2.7), we get α3 = β11 (1−ρ)
and by rearranging, we get:
β11 =
α3
.
(1 − ρ)
(2.8)
By plugging in values for ρ we can back out the implied β11 (and β12 ). Figure 2.1 plots
the simulation results for β11 and β12 from the estimate of α3 for “Tooth extraction”
versus “Basic Examination & Diagnostics”. Figure 2.2 plots the results for “Tooth
extraction” versus “Full examination, performed by dental hygienist”. The figures
shows the implied values for β11 and β12 , for −1 < ρ < 1. It is clear that the welfare
implications for consumers differ depending on where we are on the abscissa, i.e. the
value of ρ. If ρ is positive, competition has negative effects on prices for both types of
services. Therefore, ρ can be thought of as a measure of the overall competitiveness
of the market.
In order to say something about welfare implications, I need to assess the value
of ρ. An estimate of ρ should reflect overall competitiveness and from (2.7) we have
that ρ captures how β11 and β12 are related. A natural measure of competitiveness is
firms’ market power, defined as their ability to set prices above costs (Corts, 1999). I
therefore define ρ as the average markup for first-stage services relative to the average
markup for follow-on services. Mark-ups are defined by relating treatment prices to
their respective reference price (described in section 2.4). Using reference prices as a
76
proxy for marginal costs, I get an estimate of the mark-up in the following way:
Mkjt =
Pkjt − Ref.Pkt
.
Pkjt
(2.9)
Pkjt is the average price for treatment k at clinic j in year t and Ref.Pkt is the
average reference price for treatment k in year t. On average, the mark-up is higher for
tooth extractions compared to examinations, implying a lower mark-up for the latter.
Finally, I define ρ in the following way:
ρjt =
M2jt − M1jt
.
M2jt
(2.10)
Where M1jt is the average markup for examinations and M2jt is the average markup
for tooth extractions. Mkjt can be smaller than zero and larger than one, as clinics
may set prices below and well above the reference price.
Consequently, ρjt is not restricted to be in the range -1 to 1. As a means of
normalization I therefore use the median of ρjt as an estimate of ρjt , rather than the
mean.
The estimated value of ρ is 0.56 (mean 1.44) for “Tooth extractions vs. Basic
Examination & Diagnostics”. The backed out average response in prices to a 1%
increase in competition is -0.052% for first-stage prices and -0.029% for the follow-on
prices. Relating this to an increase in the number of clinics gives that one extra clinic
within 1 km, would decrease prices for “Basic examinations and diagnostics” with
0.24%, corresponding to a price decrease with about 1.6 SEK. Considering that “Basic
examinations and diagnostics” are on average performed 2.4 million times per year,
the price decrease corresponds to a redistribution from providers to patients of about
3.9 million SEK on average per year.
The estimated value of ρ is 0.81 (mean 1.40) for “Tooth extractions vs. Full
Examination, performed by dental hygienist”. This corresponds to a decrease in prices
of -0.056% for the first-stage service and -0.045% for the follow-on service. An
increase of one extra clinic within 1 km, would decrease prices for “Full Examination,
performed by dental hygienist” with -0.34%. This corresponds to a price decrease with
about 2 SEK and a redistribution from sellers to buyers of just under 1.6 million SEK
on average per year.
The policy simulations thus suggest that the absolute effects of competition are
negative for prices for both types of services. This in turn implies that competition is
welfare enhancing for consumers by lowering prices. In addition, the results suggest
that prices are set above the efficient level as there is room for price decreases.
77
Figure 2.1. Policy simulation: Tooth extraction vs. Basic Examination & Diagnostics,
performed by dentist.
Assigned value of ρ
Implied effect on prices
0
−.05
−.1
−.15
−.2
−.25
−1
−.8
−.6
−.4
−.2
0
.2
β11
.4
.6
.8
1
β12
Figure 2.2. Policy simulation: Tooth extraction vs. Full examination, performed by
dental hygienist.
Assigned value of ρ
Implied effect on prices
0
−.05
−.1
−.15
−.2
−.25
−1
−.8
−.6
−.4
−.2
0
β11
78
.2
.4
β12
.6
.8
1
2.7 Conclusions
Markets today play an important role in the provision of health care services throughout the developed world. There is therefore a growing literature on the industrial
organization of health care markets. A large part of the literature is focused on the
effects of competition on prices. However, much of the existing empirical literature
fails to credibly identify the effect. This essay suggests a framework to identify the
effect of competition on prices in a setting where price setting is free and competition
is patient-driven, namely the Swedish market for dental care.
The rationale for the identification strategy comes from the theoretical literature on
competition in health care and switching costs. The strategy exploits that the effect of
competition differs across services; price competition is more intense for services such
as examinations and diagnostics (first-stage services), compared to more complicated
and unusual services (follow-on services). This is because patients are relatively better
informed about first-stage services and face costs for switching once they have chosen a
provider. Consequently, consumers’ demand is relatively less price sensitive to followon services and therefore competition has heterogeneous effects across services. By
exploiting this difference in the effect of competition on prices, I identify a relative
effect of competition.
The results show small, but statistically significant negative effects of competition on
the price difference between first-stage services and follow-on services. Competition
is measured as the number of clinics within a fixed distance from each clinic. A one
percent increase in competition is followed by an increase in the price difference in
the range -0.01% to -0.02%. This implies that one more clinic within a radius of one
kilometer lowers prices for first-stage services relative to follow-on services with up to
0.1%. The results are robust across analyses of different kinds of services and market
areas.
In order to assess the policy relevant absolute effect of competition on prices I add
the assumption that prices for follow-on services can be expressed as a function of the
average effect on first-stage prices. This reflects that sellers recoup foregone profits
from low prices for first-stage services with high prices for follow-on services. By
exploiting auxiliary data on the pricing of services, the model assumptions allow me
to perform simulations to assess the absolute effects.
The simulation results suggest that the absolute effect of competition on prices is
negative for both kinds of services. This implies that even though the magnitude of
the effect on prices indeed differs across services, competition increases welfare for
consumers. A one percent increase in competition is followed by a decrease in the
price for first-stage services around -0.05 %. Relating this to the number of times the
first-stage services are performed on average during a year, the effects corresponds to a
redistribution from providers to patients in the range of 1.6-3.9 million SEK. The price
decrease for follow-on services fall in the range 0.03% to 0.05%, corresponding to a
redistribution from providers to patients in the range 700,000 SEK to 1 million SEK.
Furthermore the policy simulations suggest that there is room for price decreases. This
implies that prices on the Swedish dental care market are set above the competitive
level and that increased competition would increase consumer welfare by lowering
prices. While all simulation results point in the same direction, it is noteworthy that
the economic significance of the welfare improvement varies depending on which
reduced form estimate the simulation is based on.
79
In sum, the reduced form results, suggests that competition has an effect on the
difference between first-stage services and follow-on services. These results are robust
across different definitions of services and specifications. Combined with the policy
simulations, the conclusion is that i) the absolute effect of competition on prices is
negative for both kind of services and ii) competition on average has greater negative
effects for first-stage services compared to follow-on services.
All effects should be interpreted as short-term effects of competition as the identifying variation is changes in competition from one year to the next. It is therefore
noteworthy that providers show evidence of strategic behavior in their price setting.
Furthermore, substantial effects of competition on prices over time cannot be ruled
out.
An important result in this essay is that competition indeed has different effects on
prices for different types of services. This is a feature with competition on health care
markets that was pointed out several decades ago by Pauly (1978), but has been widely
overlooked in the empirical literature.
80
References
Arrow, K. (1963). Uncertainty and the welfare economics of medical care. The
American Economic Review, 53 (5), 941–973.
Beggs, A. and Klemperer, P. (1992). Multi-Period Competition with Switching
Costs. Econometrica, 60 (3), 651–666.
Capps, C. and Dranove, D. (2004). Hospital consolidation and negotiated PPO
prices. Health Affairs, 23 (2), 175–181.
Chernew, M. E. and Newhouse, J. P. (2011). Chapter One - Health Care Spending
Growth. In T. G. M. a. P. P. B. Mark V. Pauly (ed.), Handbook of Health
Economics, Handbook of Health Economics, vol. 2, Elsevier, pp. 1–43.
Docteur, E. and Oxley, H. (2003). Health-Care Systems: Lessons from the Reform
Experience. OECD Health Working Papers No. 9, OECD.
Dranove, D. and Satterthwaite, M. (1992). Monopolistic competition when price
and quality are imperfectly observable. The RAND Journal of Economics, pp.
518–534.
— and Satterthwaite, M. A. (2000). The industrial organization of health care
markets. Handbook of Health Economics, 1, 1093–1139.
—, Shanley, M. and White, W. D. (1993). Price and Concentration in Hospital
Markets: The Switch from Patient-Driven to Payer-Driven Competition. Journal of
Law and Economics, 36 (1), 179–204.
Eriksson, R. (2004). Testing for price leadership and for reputation goods effects:
Swedish dental services. Working Paper 5/2004, Swedish Institute for Social
Research (SOFI), Stockholm University.
Farrell, J. and Klemperer, P. (2007). Coordination and lock-in: Competition with
switching costs and network effects. Handbook of Industrial Organization, 3,
1967–2072.
— and Shapiro, C. (1988). Dynamic competition with switching costs. The RAND
Journal of Economics, pp. 123–137.
Gaynor, M. (2012). Reform, Competition, and Policy in Hospital Markets. Working
Party No. 2 on Competition and Regulation, OECD.
— and Town, R. J. (2011). Competition in Health Care Markets. In T. G. M. Mark
V. Pauly and P. P. Barros (eds.), Handbook of Health Economics, Handbook of
Health Economics, vol. 2, Elsevier, pp. 499 – 637.
Gehrig, T. and Stenbacka, R. (2002). Introductory Offers in a Model of Strategic
Competition. Published: No. 3189, CEPR Discussion Paper Series.
Gowrisankaran, G. and Town, R. J. (2003). Competition, Payers, and Hospital
Quality. Health Services Research, 38 (61), 1403–1422.
Grönqvist, E. (2012). The demand for dental care: Analyses of the importance of
price and income. Working Papers in Social Insurance 2012:1. Swedish Social
Insurance Agency.
Grytten, J. and Sørensen, R. (2000). Competition and dental services. Health
Economics, 9 (5), 447–461.
Gunning, T. S. and Sickles, R. C. (2012). Competition and market power in
physician private practices. Empirical Economics, pp. 1–25.
Holtmann, A. and Olsen Jr, E. (1976). The demand for dental care: a study of
consumption and household production. Journal of Human Resources, pp.
546–560.
81
Kessler, D. P. and McClellan, M. B. (2000). Is Hospital Competition Socially
Wasteful? Quarterly Journal of Economics, 115 (2), 577–615.
Klemperer, P. (1987a). The competitiveness of markets with switching costs. The
RAND Journal of Economics, pp. 138–150.
— (1987b). Markets with Consumer Switching Costs. The Quarterly Journal of
Economics, 102 (2), 375 –394.
— (1995). Competition when consumers have switching costs: An overview with
applications to industrial organization, macroeconomics, and international trade.
The Review of Economic Studies, 62 (4), 515–539.
Manning, W. and Phelps, C. (1979). The demand for dental care. The bell journal of
economics, pp. 503–525.
Melnick, G. A., Zwanziger, J., Bamezai, A. and Pattison, R. (1992). The effects of
market structure and bargaining position on hospital prices. Journal of Health
Economics, 11 (3), 217–233.
Nelson, P. (1970). Information and consumer behavior. The Journal of Political
Economy, 78 (2), 311–329.
Nevo, A. and Whinston, M. D. (2010). Taking the dogma out of econometrics:
Structural modeling and credible inference. The Journal of Economic Perspectives,
24 (2), 69–81.
Noether, M. (1988). Competition among hospitals. Journal of Health Economics,
7 (3), 259–284.
Padilla, A. (1991). Consumer Switching Costs: A Survey. Investigaciones
Económicas (Segunda época), 15 (3), 485–504.
— (1992). Mixed pricing in oligopoly with consumer switching costs. International
Journal of Industrial Organization, 10 (3), 393–411.
Pauly, M. (1978). Is medical care different. In Competition in the Health Care
Sector: Past, Present, and Future: Proceedings of a Conference Sponsored by the
Bureau of Economics, Federal Trade Commission, March 1978.
Pauly, M. V. (1988). Is medical care different? Old questions, new answers. Journal
of Health Politics, Policy and Law, 13 (2), 227–237.
Samuelson, W. and Zeckhauser, R. (1988). Status quo bias in decision making.
Journal of risk and uncertainty, 1 (1), 7–59.
Shleifer, A. (1998). State versus Private Ownership. Journal of Economic
Perspectives, 12 (4), 133–150.
Sintonen, H. and Linnosmaa, I. (2000). Economics of dental services. Handbook of
Health Economics, 1, 1251–1296.
SOU 2007:19 (2007). Friskare tänder - till rimliga kostnader. Slutbetänkande av
Utredningen om ett nytt tandvårdsstöd för vuxna.
Staten, M., Umbeck, J. and Dunkelberg, W. (1988). Market share/market power
revisited: A new test for an old theory. Journal of Health Economics, 7 (1), 73–83.
Strombom, B. A., Buchmueller, T. C. and Feldstein, P. J. (2002). Switching costs,
price sensitivity and health plan choice. Journal of Health Economics, 21 (1), 89.
Thompson, A. (2011). The effect of hospital mergers on inpatient prices: a case study
of the New Hanover–Cape Fear transaction. International Journal of the
Economics of Business, 18 (1), 91–101.
Wong, H. S. (1996). Market structure and the role of consumer information in the
physician services industry: An empirical test. Journal of Health Economics,
82
15 (2), 139–160.
83
Appendix B
To test the sensitivity of the results to functional form, the models 2.4-2.6 are estimated
in levels instead of logs. The results and corresponding elasticities are given in table
B1and B2.
Table B1. Tooth extractions vs. Basic examination & diagnostics, performed by
dentist. Levels of price and competition.
(1)
k×Competition 1 km
Clinic FE
Year dummy
Number of observations
(2)
∗∗∗
(3)
∗∗∗
-0.74169
(0.09893)
-0.74169
(0.08379)
-0.74169∗∗∗
(0.08379)
No
No
Yes
No
Yes
Yes
14,712
14,712
14,712
Notes: Standard errors in parenthesis. The pooled model in column (1) is estimated with a
public clinic dummy. Standard errors in column (2) and (3) clustered at the clinic level. All
models weighted with the number of patients per year at each clinic. ∗ p < 0.1, ∗∗ p < 0.05,
∗∗∗
p < 0.01
Table B2. Tooth extractions vs. Full examination, performed by dental hygienist.
Levels of price and competition.
(1)
k×Competition 1 km
Clinic FE
Year dummy
Number of observations
(2)
∗∗∗
(3)
∗∗∗
-1.01882
(0.07222)
-1.01882
(0.17134)
-1.01882∗∗∗
(0.17137)
No
No
Yes
No
Yes
Yes
6,876
6,876
6,876
Notes: Standard errors in parenthesis. The pooled model in column (1) is estimated with a
public clinic dummy. Standard errors in column (2) and (3) clustered at the clinic level. All
models weighted with the number of patients per year at each clinic. ∗ p < 0.1, ∗∗ p < 0.05,
∗∗∗
p < 0.01
Table B3-B6 present results from when models models 2.4-2.6 are estimated separately for different geographical regions. The definitions of geographical areas are
taken from the Swedish Association of Local Authorities and Regions (SKL). “Metropolitan municipalities” are defined as municipalities with a population of over 200,000
inhabitants, i.e. Stockholm, Gothenburg and Malmo. “Large cities” (31 municipalities) are defined as municipalities with 50,000-200,000 inhabitants with more than
70% of the population living in urban areas. “Forest counties” are sparsely populated
counties (consisting of several municipalities) with a large proportion of forest land.
These counties are Värmland, Dalarna, Gävleborg, Jämtland, Västernorrland, Västerbotten and Norrland excluding the municipalities in the Large city category. All other
municipalities are in the category “All other areas”.
84
Table B3. Tooth extraction vs. Basic examination & diagnostics, performed by dentist.
Metropolitan municipalities.
(1)
k×Competition 1 km
Clinic FE
Year dummy
Number of observations
(2)
∗∗∗
(3)
∗∗∗
-0.04071
(0.00302)
-0.04080
(0.00804)
-0.04077∗∗∗
(0.00804)
No
No
Yes
No
Yes
Yes
4,702
4,702
4,702
Notes: Standard errors in parenthesis. The pooled model in column (1) is estimated with a
public clinic dummy. Standard errors in column (2) and (3) clustered at the clinic level. All
models weighted with the number of patients per year at each clinic. ∗ p < 0.1, ∗∗ p < 0.05,
∗∗∗
p < 0.01
Table B4. Tooth extraction vs. Basic examination & diagnostics, performed by dentist.
Large cities.
(1)
k×Competition 1 km
Clinic FE
Year dummy
Number of observations
(2)
∗∗∗
(3)
∗∗∗
-0.01730
(0.00299)
-0.01721
(0.00488)
-0.01721∗∗∗
(0.00488)
No
No
Yes
No
Yes
Yes
4,303
4,303
4,303
Notes: Standard errors in parenthesis. The pooled model in column (1) is estimated with a
public clinic dummy. Standard errors in column (2) and (3) clustered at the clinic level. All
models weighted with the number of patients per year at each clinic. ∗ p < 0.1, ∗∗ p < 0.05,
∗∗∗
p < 0.01
Table B5. Tooth extraction vs. Basic examination & diagnostics, performed by dentist.
Forest counties.
(1)
k×Competition 1 km
Clinic FE
Year dummy
Number of observations
(2)
∗∗∗
(3)
∗∗∗
-0.02599
(0.00746)
-0.02599
(0.00928)
-0.02599∗∗∗
(0.00929)
No
No
Yes
No
Yes
Yes
1,094
1,094
1,094
Notes: Standard errors in parenthesis. The pooled model in column (1) is estimated with a
public clinic dummy. Standard errors in column (2) and (3) clustered at the clinic level. All
models weighted with the number of patients per year at each clinic. ∗ p < 0.1, ∗∗ p < 0.05,
∗∗∗
p < 0.01
85
Table B6. Tooth extraction vs. Basic examination & diagnostics, performed by dentist.
All other areas.
(1)
k×Competition 1 km
Clinic FE
Year dummy
Number of observations
(2)
∗∗∗
(3)
∗∗∗
-0.01893
(0.00309)
-0.01855
(0.00682)
-0.01855∗∗∗
(0.00683)
No
No
Yes
No
Yes
Yes
6,389
6,389
6,389
Notes: Standard errors in parenthesis. The pooled model in column (1) is estimated with a
public clinic dummy. Standard errors in column (2) and (3) clustered at the clinic level. All
models weighted with the number of patients per year at each clinic. ∗ p < 0.1, ∗∗ p < 0.05,
∗∗∗
p < 0.01
Table B7. Root canal, one filling vs. Basic examination & diagnostics, performed by
dentist.
(1)
k×Competition 1 km
Clinic FE
Year dummy
Number of observations
(2)
∗∗∗
(3)
∗∗∗
-0.01390
(0.00151)
-0.01390
(0.00454)
-0.01390∗∗∗
(0.00454)
No
No
Yes
No
Yes
Yes
11,794
11,794
11,794
Notes: Standard errors in parenthesis. The pooled model in column (1) is estimated with a
public clinic dummy. Standard errors in column (2) and (3) clustered at the clinic level. All
models weighted with the number of patients per year at each clinic. ∗ p < 0.1, ∗∗ p < 0.05,
∗∗∗
p < 0.01
Table B8. Clinic-made crown, plastic material vs. Basic examination & diagnostics,
performed by dentist.
(1)
k×Competition 1 km
Clinic FE
Year dummy
Number of observations
(2)
∗∗∗
(3)
∗∗∗
-0.01202
(0.00163)
-0.01202
(0.00301)
-0.01202∗∗∗
(0.00301)
No
No
Yes
No
Yes
Yes
10,294
10,294
10,294
Notes: Standard errors in parenthesis. The pooled model in column (1) is estimated with a
public clinic dummy. Standard errors in column (2) and (3) clustered at the clinic level. All
models weighted with the number of patients per year at each clinic. ∗ p < 0.1, ∗∗ p < 0.05,
∗∗∗
p < 0.01
86
Essay 3.
Redistribution in the Swedish dental care
insurance
Abstract The Swedish dental care insurance subsidizes dental care costs above
a threshold and becomes more generous as dental care consumption increases.
On average, higher-income individuals consume more dental care and have
better oral health than low-income individuals. Therefore, the redistributional
effects of the Swedish dental care insurance are ambiguous a priori. I find
that the dental care insurance adds to the progressive redistribution taking place
through other parts of the Swedish social insurance (SI) for individuals aged
35-59 years whereas it reduces the progressivity in the SI for those aged 60-89
years. While the result for the oldest individuals is problematic from an equity
point of view, the insurance seems to strengthen the progressitivy of the Swedish
social insurance for the vast majority of patients.
I would like to thank Per Johansson, Erik Grönqvist and Erik Lindqvist for very valuable
feedback. I would also like to thank participants at the Swedish Health Economics Association
(SHEA) conference 2015 as well as seminar participants at the Department of Economics at
Örebro university school of business and the Department of Economics at Uppsala University
for helpful comments.
87
3.1 Introduction
Health care is publicly financed to some extent in all OECD countries (OECD, 2013).
Although the generosity of the systems varies, all types of public health insurance
schemes include elements of redistribution. As shown by Blomqvist and Horn (1984),
Rochet (1991) and Cremer and Pestieau (1996), a public health insurance can be an
efficient complement to income taxation as a means of redistribution from high-income
individuals to low-income individuals given that income is negatively related to health
risks. This is because the insurance—by covering part of individuals’ health care
costs—redistributes welfare between individuals in different health states rather than
between individuals of different productivity. Therefore the insurance can alleviate
the welfare losses associated with a distortionary income tax. This theoretical result
is developed in models where health care consumption is treated only as a way to
restore health after falling sick and where there is no motive for healthy individuals
to consume health care. It is however hard to reconcile the notion of health care as
solely a repair technology with the substantial empirical evidence suggesting a positive
relationship between income and health care consumption (see Blazquez-Fernandez
et al., 2014, for a review) given that individuals with higher incomes, on average, have
better health. Therefore, in a setting where the care that is covered by the insurance
has value for the individual beyond its restoring effect on health, redistribution will
be determined by how consumption depends on income. This essay investigates the
redistributional features of a public health insurance system in such a setting, namely
the Swedish dental care insurance.
Sweden has historically had a high level of social protection. In 2012, Sweden was
among the top ten countries in the EU in terms of expenditures on social protection
per inhabitant (Eurostat, 2015) and among the top 15 countries in the OECD in terms
of public spending on health as a share of GDP (OECD, 2014). The Swedish social
insurance system is universal, financed through taxes and contains several redistributive
elements. The redistributional features of the dental care insurance is therefore defined
in this essay by whether or not the dental care insurance adds to the progressive
redistribution taking place through other parts of the social insurance.1 The dental
care insurance includes cost-sharing subsidies at two different rates depending on the
value of consumed care over a given period.2 Dental care costs below EUR 320 are paid
out of pocket by the patient. Above this threshold, the coinsurance rate is 50 percent
which falls to 15 percent once accumulated dental care costs are above EUR 1,600. The
generosity of the insurance therefore increases with consumption. It is reasonable to
assume that dental care consumption on average increases with income and decreases
with oral health. Under the assumption of a negative relationship between income
and the need for care—commonly made in the theoretical literature—this implies that
income has a positive direct effect on dental care consumption and a negative indirect
1
2
The other parts of the Swedish social insurance system are not connected to consumption
but rather an insurance for loss of income due to disabilities, illnesses or parental leave. In
addition, benefits generally correspond to a fraction of individuals’ salaries up to a cap. It is
therefore more straightforward to characterize redistribution through these parts of the social
insurance system as being progressive.
The value of consumed care is determined by administratively set reference prices. The part
of the dental care cost that is covered by the insurance is paid directly by the Swedish Social
Insurance Agency to the provider. The rest is paid out-of-pocket by patients. The period is at
most twelve months.
88
effect, operating through better oral health. If the total effect of income on dental
care consumption is negative, insurance coverage will to a larger extent accrue to
low-income individuals than high-income individuals. This implies that the dental
care insurance adds to the progressive redistribution through other parts of the social
insurance. If the total effect of income instead is positive, the dental care insurance
will redistribute regressively in the sense that it will to a greater extent subsidize dental
care consumption among high-income individuals than low-income individuals. The
direction of the total effect of income is clearly an empirical question.
The empirical analysis is guided by Grossman’s framework (Grossman, 1972b,a,
2000) implying that individuals are assumed to demand dental care because it serves
as an input in the production of oral health. Given that oral health is a normal good, the
demand for both oral health and dental care increases with incomes. The Grossman
model therefore provides a theoretical foundation for a positive relationship between
income and health as well as investments in health in the form of any health promoting
activities. In addition, the Grossman model assumes that the marginal benefits of
health are diminishing, implying that the demand for dental care is decreasing in oral
health.
The analysis builds on detailed register data on dental care consumption, income
and other individual characteristics on the entire Swedish population aged 35 years
and above for the period July 2008- December 2011. The empirical strategy exploits
that the insurance scheme consists of three different segments; one where individuals
pay market prices and two where dental care is subsidized. Consumption in the
lowest segment provides information on how consumption varies with income under
market prices. Consumption in the higher segments also provides information on
how consumption varies with income, but under subsidized prices. By estimating
a multinomial logit model for reaching one of the segments with subsidized prices
using the no-subsidy segment as the reference category, I can test if the consumption
response to the insurance is greater among low-income individuals than high-income
individuals. Since incomes and oral health varies over the life-cycle, the model is
estimated separately for age-groups.
The paper contributes to the empirical literature on public health insurance by combining the theoretical literature on public health insurance with the literature on the
relationship between income and health care consumption as a way on understanding
distributional effects. The theoretical literature on both moral hazard and redistribution is substantial while the empirical work is often limited by availability of data.
Using register data is an improvement compared to the existing literature which often
relies on insurance claims data with the limitation that it does not contain important
socioeconomic variables (Breyer et al., 2011). Claims data also has the problem that it
only covers those who have chosen to enroll in an insurance plan whereas the data used
in this essay covers the universe of individuals aged 35 and above in Sweden during
2008-2011. In addition, the Swedish dental care insurance is an interesting application
to study as it covers care that is provided through private clinics to a large extent,
includes a relatively large co-payment from patients and accounts for only about 8 %
of public expenditures on health care, medical care and social services in Sweden.
I find that the response to the insurance differs across incomes and that the redistributional features of the dental care insurance differ across age-groups. The insurance
adds to the progressive redistribution taking place through other parts of the Swedish
89
social insurance system for individuals aged 35-59 years whereas it reduces the progressivity in the social insurance for those aged 60 years and above. The result for the
older individuals is problematic from an equity point of view as the income gradient
in oral health is more pronounced among older individuals compared to the younger
age-groups. However, given the market elements in Swedish dental care and the political debate surrounding e.g. patient choice and private co-payments in health care, it
is noteworthy that the insurance seems to strengthen the progressiveness of the social
insurance system for the vast majority of patients. The results should however be
interpreted as a “lower bound” due to the institutional setting; dental care in Sweden is
free up to the age of 19. This plausibly increases oral health in the Swedish population
and decreases the social oral health gradient. It is likely that disparities in adult oral
health related to income would be more pronounced if dental care for children and
youth were privately financed suggesting a larger scope for a dental care insurance to
redistribute progressively in such a setting.
3.2 Related literature
Health care, along with education, is the largest government expenditure item in most
OECD countries. Government intervention in health care is primarily motivated by
equity concerns.3 More precisely, that access to care should not be conditioned
on income (Poterba, 1996), and concerns about the distribution of welfare between
individuals resulting from differential risks and incidence of illness (Blomqvist and
Horn, 1984). The case for public health insurance as an instrument for redistribution
has been studied carefully theoretically. The starting point in this literature is models
where individuals differ in their productivity and health. The government is imperfectly
informed about individuals’ health risks and therefore cannot differentiate income
taxes accordingly. If they could, there would be no role for public insurance based
on equity concerns (Blomqvist and Horn, 1984). The main result in these models is
that health insurance, as a complement to income taxation, can achieve redistribution
more efficiently than distortionary income taxes alone, given that income is negatively
related to health risks (Blomqvist and Horn, 1984, Rochet, 1991, Cremer and Pestieau,
1996). The intuition for this result is that the insurance and the tax redistribute welfare
along different margins. While the income tax redistributes between individuals of
different productivity, the insurance redistributes between individuals in different states
of “nature”, i.e. good or bad health. In order for the insurance to alleviate the welfare
losses associated with the distortionary tax there must be a higher proportion of lowincome individuals falling ill and a possibility to compensate for any losses caused
by illness by lump-sum transfers (Boadway et al., 2006). Moreover, as pointed out
by Boadway et al. (2003) the models assume that there are no behavioral changes in
response to the insurance.
There is a rather extensive literature on consumer incentives in health care detailing
behavioral changes caused by insurance, the so called moral hazard effect (Arrow, 1963,
Pauly, 1968). Distortions of individuals’ decisions are inherent in all health insurance
since the insurance lowers the cost associated with illness (see e.g. Pauly, 2011, for
a review) while both the insurer and health care provider are imperfectly informed
3
Other motives include concerns about the inefficiency of private health markets.
90
about individuals’ behaviors. Ex ante moral hazard refers to reductions in preventive
activities caused by insurance, e.g. if individuals stop flossing because they have dental
care insurance. The empirical evidence of ex ante moral hazard is limited (Dave and
Kaestner, 2009) which can potentially be explained by the fact that the loss of health
itself generally implies an uncompensated utility loss for the individual (Cutler and
Zeckhauser, 2000). Another potential explanation is lack of data. Ex post moral hazard
is the substitution effect of individuals demanding more care due to the price decrease
caused by the insurance. Hence, ex post moral hazard refers to actions to affect the size
of a loss caused by e.g. falling ill. This theoretical prediction is supported by empirical
evidence suggesting positive and economically significant effects of insurance on
consumption. Recent evidence from the Oregon Health Insurance Experiment4 suggest
substantial and statistically significant positive effects of insurance on health care
utilization (Finkelstein et al., 2012). Furthermore, Aron-Dine et al. (2013) reexamine
the data from the Rand Health Insurance Experiment5 using techniques developed over
the three decades of advances in empirical methods. They find elasticities of medical
spending with respect to out-of-pocket prices in the range -0.04 and -0.6 suggesting
that medical spending indeed responds to price changes induced by health insurance.
Grönqvist (2006) use a dental insurance quasi-experiment in Sweden and utilize several
empirical strategies to identify the impact of full-coverage insurance on dental care
consumption. He finds large increases in utilization as a result of full-coverage and
evidence of moral hazard for expensive treatments.
Boadway et al. (2003) reconciles the theoretical literature on redistributive public
insurance with the empirical evidence, by introducing moral hazard.6 The resulting
conclusion is that there is a case for public health insurance even with behavioral
responses but that the sign and the size of the optimal rate of insurance will depend
on an efficiency term and an equity term. The efficiency term is how the insurance
affects expenditures on health care. This is expected to be positive due to ex post moral
hazard. The second term captures the covariance between the marginal social utility
of income7 , πr zir , and expected health care spending. The sign of the covariance
is theoretically ambiguous. It is fairly reasonable to assume that redistribution from
low-risk households to high-risk households is socially desirable both from an equity
and an efficiency perspective. This implies a positive covariance between marginal
social utility of income, bir , and risk probabilities, πr . However, expected health care
spending is given by the risk probability multiplied with care expenditures following
a realized negative health risk, zir . In other words, expected health care spending is
a function of both health and how households respond to illness in terms of health
4
5
6
7
An experiment in Oregon conducted in 2008 where low-income adults were selected by lottery
to be given a chance to apply for Medicaid.
A large scale randomized experiment conducted in the U.S. 1974-1981. In the experiment,
households where randomized into insurance plans with different levels of cost-sharing. See
Newhouse et al. (1996) for a detailed description.
Boadway et al. (2003) sets up a Mirrlees-type taxation model (first introduced in Mirrlees,
1971) developed by Rochet (1991) and Cremer and Pestieau (1996), where incomes are
uncertain due to the risk of falling ill. Boadway et al. (2006) also considers the case for public
health insurance with distortionary taxes and moral hazard.
The governments object is to maximize a utilitarian social welfare function. The marginal
social utility of income is the derivative of the social welfare function with respect to a
lump-sum transfer to households.
91
care spending. The sign of cov(bir , πr zir ) therefore depends on the relation between
income and zir . Signing the covariance is thus an empirical question (Boadway et al.,
2003).
Turning to the relationship between income and the use of health care, there is
empirical evidence suggesting that health care spending increases with incomes on
average. Despite that most OECD countries provide at least near-universal coverage for
core health care services8 , there is evidence of disparities in utilization across income
groups (Devaux, 2013). In addition, Bredenkamp et al. (2014) find evidence of a prorich bias in the incidence of government health expenditures on some types of health
care.9 Van Doorslaer et al. (2000) find that high-income individuals in the U.S. and
several European countries10 report significantly more physician contacts compared
to low-income individuals even after controlling for need, defined as individual’s
expected health care utilization given their age, sex and self-reported health status.
Similarly, Van Doorslaer et al. (2004) use the European Community Household Panel
and find that high income individuals are substantially more likely to visit medical
specialists despite their low need compared to less wealthy individuals. Morris et al.
(2005) use register data from the English National Health Service (NHS) to find that
all types of hospital care increases with income when controlling for a wide range of
health indicators.11 All these studies describe a pattern for health care consumption
suggesting that income matters for the utilization decision. However, these studies
do not investigate how the relationship between income and utilization is associated
with insurance coverage, i.e. if differences in consumption across income-groups are
related to the ways in which health care spending is financed.
A positive relationship between income and health care consumption can be understood from Grossman’s seminal health production model (Grossman, 1972b,a, 2000)
which treats health as a capital stock, considered as being a part of an individual’s
human capital. In the model, health promoting activities are viewed as investments in
individual’s health stocks. A central proposition in Grossman’s model is therefore that
the level of health should not be considered as exogenously determined, but rather as
a function of the resources invested in it. Given that health in itself is a normal good,
the demand for it will increase with incomes and consequently individuals with higher
incomes will invest more in their health. This broadens the perspective on health care
consumption to more than solely a repair technology and provides a rationale for why
healthy individuals also consume health care. In addition it provides a theoretical
foundation for a positive relationship between income and health.
In conclusion, there is an extensive literature on both health insurance and the
relationship between income and health care consumption. While the theoretical
literature makes a case for public health insurance as a means of redistribution the
implications of a positive relationship between income and health care consumption is
left aside. At the same time, the literature on income and health care spending generally
8
In Europe, Bulgaria, Greece and Cyprus are exceptions (OECD and European Union, 2014).
Bredenkamp et al. (2014) study 69 countries in different stages of development and finds that
government health expenditures (GHE) on contracted private facilities and all outpatient care
is pro-rich. When looking at GHE on all types of care together, they find that it is neither
pro-rich nor pro-poor in a majority of the studied countries.
10
Sweden is one of them.
11
Morris et al. (2005) also reports that low-income individuals are more likely to visit a GP, but
this estimate is not statistically significant.
9
92
abstracts from the effects of health insurance. Viewing health care consumption
through the lens of the Grossman model suggests that the consumption should be
seen as having an investment dimension rather than just being a repair technology.
Taken together, this may have implications for redistribution within a public health
insurance. If the insurance covers part of individuals’ health care costs and highincome individuals consume more than low-income individuals, the insurance will
redistribute regressively. However, low-income individuals will on average have worse
health. The redistributional features of such an insurance system will therefore be
determined by the total effect of income on consumption which is something that has
not been addressed explicitly in the existing literature.
3.3 Institutional setting
Dental care in Sweden is provided through both private clinics and the public National
Dental Service.12 Dental care is free for children and young people aged 19 or under.
The majority (60-80 %) of dental care for individuals aged 20 years and above is
supplied by private clinics. Patients can freely choose either public or private clinics
and price setting is free. The Swedish dental care insurance is administered by the
Swedish Social Insurance Agency (SSIA) and contains a general dental care allowance,
which applies to all citizens aged 20 years and above, and a high-cost protection (HCP)
plan. The allowance covers preventive dental care and dental care that reduce pain
and enables the patient to eat, chew and speak without impediment. The size of the
allowance varies with age; 300 SEK/year for individuals aged 20-29 years or above 74
years and 150 SEK/year for individuals aged 30-74 years (approximately 30 EUR and
15 EUR respectively).
The high-cost protection scheme includes coinsurance at two different rates which
is essentially a subsidy for dental care costs over a given threshold. The insurance
scheme is depicted in figure 3.1. Eligibility for high-cost protection, i.e. subsidized
care, is determined over a period of at most twelve month starting the first time the
individual visits the dentist.13 This is referred to as a HCP-period. Patients pay the
full price directly to the dentist for dental care costs below 3,001 SEK within each
HCP-period. The coinsurance rate, i.e. the fraction of the cost borne by patient, is 50
percent for dental care costs between SEK 3,001 and SEK 15,000 (approximately 320
EUR and 1600 EUR respectively). Once accumulated costs are pushed above SEK
15,000 within a HCP-period, the coinsurance rate falls to 15 percent. The remaining
costs are paid directly to the dentist by the Swedish Social Insurance Agency.
The Dental and Pharmaceutical Benefits Agency (TLV) decides which procedures
are covered by the insurance and determines a reference price list, which serves
as a basis for determining eligibility for high-cost protection. Hence, a patient’s
12
The National Dental Service has a legal responsibility for ensuring the supply of dental care
to the citizens in the county (National Dental Service Act, Tandvårdslagen (1985:125)).
13
A patients’ HCP period can be closed prematurely. This is done if the patient e.g. is in the
tenth month of a HCP-period and will have to undergo costly treatments in the near future.
The dentist may then start a new period in order for the patient to become eligible for coverage
through the HCP.
93
accumulated cost is made up of the reference price for each consumed service.14 It is
therefore not possible to become eligible for high-cost protection simply by choosing
an expensive dentist. The Swedish Social Insurance Agency (SSIA) hold the Dental
Care Register. Since the public dental care insurance applies to all dental care, the
register covers both publicly and privately provided dental care.15 Subsidies through
the high-cost protection plan are calculated over the consumption during a HCP-period.
Therefore, all dental care is registered, even if an individual does not reach the first
threshold in the high-cost protection. Consequently, the register also covers dental care
that is fully paid for by the patient.16
Figure 3.1. High cost protection scheme
Cost of care
6
Total payment
6
Insurer payment
(
(((
((((
(
Individual payment
Coinsurance 50%
Coinsurance 15%
3000
15000
Patient’s accumulated cost, SEK
3.4 Methodological framework
The Grossman model (1972b, 1972a, 2000) serves as a framework for understanding
individuals demand for dental care. In the original model, individuals are assumed to
demand health because they enjoy being healthy and because being healthy increases
14
If the price charged by the dentist is lower than the reference price, the charged price is used
when determining eligibility.
15
All public clinics and about 96 % of the private clinics are connected to the dental care system.
Little is known about the 4 % of the private clinics that operate entirely outside the national
dental care system.
16
Purely aesthetic care is not registered.
94
productivity. In this application, the production benefits are arguable negligible17
suggesting that individuals consume dental care because of the consumption benefits of
oral health. Viewing oral health as a capital stock allows thinking about health behavior
as other investment decisions considered in the economics literature. Individuals will
invest in their health stock up to the point where marginal benefits equals the marginal
costs of holding an additional unit of health. In this context, the benefits are the flow
of services provided by the stock, i.e. enjoying having a good oral health whereas the
marginal cost is simply the cost faced by individuals for holding an additional unit of
health capital. The costs are measured not only by e.g. the price for dental care but
also by the opportunity cost of own effort, time etc. From the law of the downward
sloping demand curve it follows that the demand for health investments decreases with
costs (Grossman, 1972a). Grossman also assumes that marginal benefits of health are
diminishing, implying that healthy individuals will place less value on an additional
unit of health compared to less healthy individuals. Therefore the demand for dental
care is assumed to be decreasing in oral health.
Age-effects are an important component of the Grossman model. The health stock
is assumed to depreciate over time at an increasing rate, implying that the health stock
deteriorates faster as we grow older.18 This suggests that individuals have to invest
continuously over the life-cycle and that the cost of maintaining a given health stock
is greater for older individuals. At the same time, due to the higher depreciation rate,
it is plausible that the demand for health investments increases with age as a way of
making up for the greater health loss.19 This illustrates the importance of taking age
into account in order to understand any health-related behavior.
Given that oral health is a normal good, the demand for both oral health and dental
care increases with income. However, income varies over the life-cycle, in part due
to variations in health. The most accurate way of capturing the relationship between
dental care consumption and income is therefore to use lifetime incomes. Using dental
care consumption as an operationalization of investments in the oral health stock,
optimal dental care consumption at age t can be expressed as:
c∗t = c(ht , y, p),
(3.1)
where ht is the oral health stock, y is lifetime income and p is the price of dental care.
From the discussion above, we have that:
∂c∗
∂c∗
∂c∗t
< 0, t < 0 and t > 0.
∂p
∂ht
∂y
Note that equation (3.1) abstracts from any insurance schemes and therefore c∗t should
be understood as what individuals would have consumed had they paid the full price,
or as a notion of individuals’ valuation of dental care without any subsidy.
17
In general, bad oral health does not impede the ability to work.
Depreciation does not necessarily start at birth but rather after some stage in the life cycle
(Grossman, 1972b). As pointed out by Sintonen and Linnosmaa (2000), a difference between
the stock of oral health and the stock of general health considered in the Grossman model, is
that death occurs when the latter falls under a minimum level. This is not true for oral health.
19
Health investments increase the health stock and hence increases “the supply of health” for the
individual. If the change in health stock from a given investment is larger than the decrease in
demand for oral health, individuals will want to close the gap by consuming more dental care.
As discussed in Grossman (2000, p. 367-370), this is true under plausible conditions.
18
95
Interest lies in whether the response to the insurance differs across incomes. To fix
ideas, consider the model:
c∗t = a0 + a1 ht (y) + a2 y + a3 p + a4 (y × p) + a5 (ht (y) × p) + εt ,
(3.2)
where εt is an error term. ht (y) is health at age t and is allowed to vary with lifetime
income. Furthermore, we allow for differences in responses to prices across both
incomes and health, captured by parameters a4 and a5 respectively. Following the
Grossman model (1972b, 1972a, 2000), a1 is assumed to be negative. a3 is also
negative as optimal consumption decreases with price and given that oral health is
a normal good and dental care increases with lifetime income, a2 is assumed to be
positive.
If individuals face a lower price ps (i.e. ps < p), optimal consumption should
increase. In the current application, this implies that the subsidy in the high-cost
protection scheme causes observed dental care consumption, i.e. optimal consumption
given the subsidy, to increase. That is, given the model (3.2):
ct − c∗t = a3 (ps − p) + a4 (y × (ps − p)) + a5 (ht (y) × (ps − p)) > 0.
The interaction term a4 can be thought of as an “income-driven” response to a price
change whereas a5 can be though of as the indirect “health-driven” response, reflecting
that health increases with income. a4 and a5 cannot be signed a priori as we do not
know if the consumption response to a price change differs across incomes. Given
model (3.2) the empirical question simplifies to test if:
dc∗
dc∗t
|ps − t |p = a4 (ps − p) + a5 ht (y)(ps − p) < 0.
dy
dy
(3.3)
Since ps < p we have that the price difference, ps − p, is known. This allows
for testing (3.3) using a reduced form model which is discussed below and in detail
in Appendix C2. Failing to reject (3.3), suggests that the response to the insurance
decreases with lifetime incomes which implies that the dental care insurance adds to
the progressive redistribution taking place through other parts of the Swedish social
insurance.
The empirical strategy to test the hypothesis in equation (3.3) can be illustrated by
figure 3.2. The horizontal axis depicts optimal dental care consumption, c∗ , and the
vertical axis depicts observed consumption, c, i.e. dental care consumption with the
insurance in place. Maintaining the assumption of oral health and hence optimal dental
care consumption varying with age, the subscript t is dropped for ease of exposition.
Recall that the two thresholds for receiving subsidies are SEK 3,000 and SEK 15,000
respectively. As long as an individual’s accumulated dental care costs (AC) is below
the first threshold within a HCP-period, i.e consumption is in the left-most bracket in
figure 3.2, the marginal price equals the price charged by the dentist and c∗ equals
c. However, as soon as AC is pushed above the first threshold, the marginal price
decreases. Once AC is above SEK 3,000, the individual only pays 50 percent of
the cost of dental care consumed up to the point where AC is pushed above SEK
15,000, after which the individual only pays 15 percent. Observed consumption in the
two upper consumption brackets in figure 3.2 is therefore higher than what optimal
consumption would have been without the subsidy.
96
Figure 3.2. Observed versus optimal consumption
c
6
15000
3000
3000
9000
- c∗
The lowest consumption bracket provides information on how consumption varies
with income under market prices whereas consumption above the first threshold provides information on how consumption varies with income under subsidized prices.
The insurance adds to the progressive redistribution in the Swedish social insurance
system if the response to the insurance decreases with income, i.e. if lower-income
individuals are more likely the get their dental care costs subsidized through the insurance compared to higher-income individuals. A way of testing the hypothesis in
equation (3.3) is therefore to estimate the marginal effect of income on the probability
of being in one of the subsidized consumption bracket relative to being in the lowest
consumption bracket and paying market prices. Assuming that individuals decide
which consumption bracket to be in given their optimal consumption and taking the
subsidy into account, the hypothesis is tested by estimating a multinomial logit model
for individual’s choice of consumption bracket using the no-subsidy segment as the
reference category. The value added from the empirical strategy is that it exploits
the fact that some individuals choose to consume below the first threshold and paying
market prices for dental care, and that it takes the difference in the subsidy into account.
Evaluating how the size of the subsidy varies with income would underestimate dental
care consumption on the extensive margin as all individuals consuming below the first
threshold would be dropped from the analysis.
The three different consumption segments are represented by Si where i is 0, 1 or
2 and given by the highest consumption bracket the individual reaches during each
97
HCP-period. The reduced form model is given by:
S=0
if (β00t + β10t y) ≤ 3, 000 − εt
S=1
S=2
if 3, 000 − εt < (β01t + β11t y) ≤ 15, 000 − εt
if (β02t + β12t y) > 15, 000.
where β1jt represents consumption responses to lifetime incomes within each agegroup t evaluated at prices p = 1 i.e. the full price paid under the first threshold, and
p = 0.5 and p = 0.15 in the subsidized consumption brackets. See Appendix C2 for
the derivation of the reduced form.
Under the assumption that εit is i.i.d., i = 1, ..., nt , the parameters can be estimated
using a multinomial logit model, where nt is a sample of individuals in age-group t.
This means using a maximum likelihood estimator where the probability for individual
i to be in one of the three possible consumption brackets given lifetime income yi is
equal to:
eβ0jt +β1jt yi
.
Pr(Si = j|yi , t) = 2
β0jt +β1jt yi
j=0 e
t = 35−40, . . . , 60−64
and
j = 0, 1, 2.
The model is estimated with year-specific fixed effects and standard errors are clustered
at the individual level since individuals can have several HCP-periods during the
studied period. As interest lies in potential differences in consumption responses to the
insurance across the income distribution, the hypothesis (3.3) is tested by normalizing
the income effect to zero under market prices and test whether β11t = 0 and/or β12t =
0. That is, to test if the probability of being in one of the higher consumption brackets
relative to the being in the lowest consumption bracket changes as we move up in the
income distribution. If β11t > 0 and/or β12t > 0, the dental care insurance adds to
the progressive redistribution taking place through other parts of the Swedish social
insurance.
3.5 Data
Dental care data comes from the Dental Care Register at the Swedish Social Insurance
Agency and The Dental Health Register at the The National Board of Health and
Welfare. The dental care data is for the period July 2008-December 201120 and covers
all dental care produced and purchased at clinics that are subscribed with the dental
care insurance, which is approximately 96 % of all clinics in Sweden. The Dental
Care Register contains information about visits, diagnosis and insurance coverage and
covers over 40 million treatments during the studied period.
The key variables from the Dental Care Register is expenditures on dental care and
which consumption bracket the individual is in. The register contains information on
the charged price, out-of-pocket payment and the administratively set reference price
for each treatment item. The outcome of interest is if an individual’s dental care costs
are subsidized within the high-cost protection or not. It is captured by an indicator
20
Both registers were created when the Swedish dental care system was reformed and the new
dental care insurance was introduced in 2008.
98
variable that takes on values 0, 1 or 2 reflecting the three different segments of the
insurance. The indicator is constructed from the same variable as the one used by the
Swedish Social Insurance Agency to determine eligibility for the subsidy. It captures
the patient’s accumulated consumption during each HCP-period, measured by the sum
of the reference price for all treatment items consumed. Hence, the outcome variable
takes the value one for an individual with accumulated costs between SEK 3,001
and SEK 15,000 within a HCP-period, two if costs are above SEK 15,000 and zero
otherwise.
The Dental Health Register covers all individuals in the Dental Care Register and
includes the number of remaining teeth and the number of intact teeth. The dental
care data is linked to administrative registers collected at Statistics Sweden (LISAdatabase), covering the universe of individuals aged 35 and above. However, the main
sample is restricted to one observation per year and individual for those aged 35-64
years, resulting in a sample of 8,737,193 observations. In the empirical analysis I
use individuals with positive dental care consumption during the studied period and
use one observation per HCP-period and individual, referred to as the patient sample
(n=6,213,867). I also perform the analysis on non-working individuals aged 65 years
and above, referred to as the retired sample.
The registers contain annual information for 2008-2011 on a wide variety of demographic variables such as income, education, employment status and number of days in
unemployment during each year. To capture individuals’ lifetime incomes, I use mean
disposable income over three years. In a study using long series of Swedish income
data, Nybom and Stuhler (2015) find that incomes at prime age are the most accurate
measures of the ranking of lifetime income profiles. Since individuals with higher
lifetime incomes tend to be in higher education in their late twenties and consequently
have low incomes, the analysis focuses on individuals from their mid-30’s and onward.
Disposable incomes are given by households after-tax earnings and transfers weighted
with each individuals’ consumption weight. After-tax earnings include e.g. wage
income, income from own business, interest income, share dividends and profits on
sold shares in interest funds. The weighting accounts for the number of adults and
children in the household.21
21
The variable is “DispInkPersF04" in the LISA-database.
99
Table 3.1. Summary statistics, all individuals.
All aged 35-64
Non-patients
Patients
198.08
(765.48)
161.19
(234.77)
213.07
(894.84)
49.54
(8.71)
47.69
(8.55)
50.29
(8.66)
Female
0.49
0.40
0.52
Born outside Nordic countries
0.04
0.10
0.02
Disposable income, SEK1000s
Age
Compulsory schooling
0.11
0.15
0.10
Secondary schooling
0.48
0.48
0.47
Post-secondary schooling
0.35
0.27
0.38
Employed
0.79
0.65
0.84
No. of days in unemployment
11.63
(49.52)
17.49
(60.26)
9.25
(44.20)
No. sick-spells
0.12
(0.34)
0.12
(0.33)
0.13
(0.34)
8,737,193
2,523,326
6,213,867
Number of observations
Notes: One observation per year & unique individual aged 35-64 years during the period July 2008-December 2011. Non-patients are never patients during the studied period
whereas patients have some positive consumption during the period. Standard deviations in
parentheses.
3.5.1 Descriptive statistics
Table 3.1 provides means for disposable incomes and demographic variables for all
aged 35-64 years, non-patients and patients respectively. Summary statistics for the
retired sample are given in table C1 in the Appendix. Non-patients are defined as
individuals who do not show up in the dental care data during the studied period.
Disposable incomes are on average lower among non-patients compared to patients.
However, the latter group is older on average. Patients are also women to a somewhat
higher extent, more likely to be born in one of the Nordic countries, more likely to
have post-secondary schooling and to be employed. It is also noteworthy that patients
on average have a slightly higher number of sick-spells22 compared to non-patients.
22
Defined as episodes with sickness benefits.
100
Table 3.2. Elasticities of dental care consumption with respect to income (full sample)
and oral health (patient sample).
(1)
Log. disposable income
(2)
∗∗∗
0.1108
(0.0014)
-0.6781∗∗∗
(0.0030)
Log. oral health index
Number of observations
8,737,193
5,357,067
Notes: One observation per year & unique individual during the period July 2008-December
2011. Poisson regression estimates of elasticities of dental care with respect to income and
oral health. Column 2 reports estimates for patients with non-missing oral health measures.
Models include a dummy for sex and fixed effects for year and age-group. Standard errors
in parenthesis, clustered on individual level. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Table 3.3. Oral health and income.
(1)
Patient sample
(2)
Retired sample
Log. disposable income
0.0251∗∗∗
(0.0003)
0.0651∗∗∗
(0.0007)
Number of observations
5,357,067
2,134,370
Notes: One observation per year & unique individual during the period July 2008-December
2011. Regression estimates of the log. oral health index on log. disposable incomes for
patients with non-missing oral health measures. Models include a dummy for sex and fixed
effects for year and age-group. Standard errors in parenthesis, clustered on individual level.
∗
p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Table 3.2 reports estimates of elasticities of dental care consumption with respect to
income and oral health.23 The oral health index is constructed by reducing the variables
“Remaining teeth” and “Intact teeth” into one principal component. In line with the
Grossman model and the assumptions made in the methodological framework, dental
care consumption increases with income and decreases with oral health on average.
The results suggests that a one percent increase in disposable income is associated with
an increase in dental care consumption of 0.11 percent whereas a one percent increase
in the oral health index is associated with a 0.68 decrease in dental care consumption.
Table 3.3 reports elasticities of oral health with respect to income for individuals aged
35-64 years (patient sample) and non-working individuals age 65-89 (retired sample)
respectively. The estimates confirm that oral health increases with incomes on average.
Table 3.4 turns to the extensive and intensive margins of dental care consumption.24
Column (1) provides probit estimates of the marginal effects on the probability of
any consumption during the period July 2008-December 2011. Column (2) gives
OLS estimates of the association between individual characteristics and the value of
consumed care, defined by the sum of the reference prices for all treatment items
23
24
See table C2 in Appendix for corresponding results for the retired sample.
See table C3 in Appendix for corresponding results for the retired sample.
101
consumed each year, i.e. the variable that determines subsidizes though the highcost protection. Age is controlled for using fixed-effects for age-groups defined over
five years. Disposable incomes are associated with a higher probability of any visits
during the period but a lower value of consumed care. Comparing the estimates across
columns in table 3.4 gives that women on average are more likely than men to have
any visits, while men on average consume more care. This is in line with previous
empirical evidence suggesting that women have a more pro-active health behavior
than men (see e.g. Lee, 2010). Being employed, secondary and post-secondary
schooling is associated with a higher probability of any consumption but a lower level
of consumption. The opposite is true for the number of sick-spells whereas the number
of days in unemployment slightly increases consumption on both the extensive and
intensive margin. For the retired sample, dental care consumption increases with
income on both the extensive and the intensive margin.
Table 3.4. Extensive and intensive margin of consumption.
(1)
Prob. of any consumption
(2)
Log. value of consumption
Log. disposable income
0.2159∗∗∗
(0.0013)
-0.0161∗∗∗
(0.0008)
Female
0.3651∗∗∗
(0.0018)
-0.0465∗∗∗
(0.0010)
Born outside Nordic countries
-0.5712∗∗∗
(0.0036)
0.2962∗∗∗
(0.0036)
Compulsory schooling
0.1629∗∗∗
(0.0039)
0.0235∗∗∗
(0.0029)
Secondary schooling
0.3484∗∗∗
(0.0034)
-0.0539∗∗∗
(0.0026)
Post-secondary schooling
0.5186∗∗∗
(0.0035)
-0.1301∗∗∗
(0.0026)
Employed
0.5646∗∗∗
(0.0020)
-0.1075∗∗∗
(0.0015)
No. of days in unemployment
0.0001∗∗∗
(0.0000)
0.0002∗∗∗
(0.0000)
No. sick-spells
-0.0503∗∗∗
(0.0022)
0.0630∗∗∗
(0.0015)
-2.0303
7.3893
8,737,193
6,101,740
Constant
Number of observations
Notes: One observation per year & unique individual aged 35-64 years during the period
July 2008-December 2011. Models include fixed effects for year and age-group. Standard
errors in parenthesis, clustered on individual level. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Table 3.5 provide means by patient group for the same set of individual characteristics as in table 3.1. Just above 47 percent of the patients have 50 percent coinsurance
at some point whereas about 7 percent also have 15 percent coinsurance at some point.
Individuals in the mid consumption segment have the highest disposable incomes on
102
average across the three categories of patients, followed by individuals in the highest
consumption segment. Those in the latter group are also older on average and have
fewer remaining and intact teeth compared to the other two patient categories. Another
way of illustrating the relation between income and the different segments of the insurance are given in figure 3.3 where the probabilities of moving upward in the high-cost
protection scheme is plotted over income groups. Taken at face value, the left panel
of figure 3.3 suggests that the dental care insurance redistributes regressively as the
probability of becoming eligible for 50 % coinsurance increases strikingly as we move
upwards in the income distribution. It is however important to note that the plotted
probabilities does not account for age and can therefore be entirely driven by older
individuals having both higher incomes and worse oral health. The corresponding
table for the retired sample (C4 in Appendix) illustrates a somewhat different pattern;
average disposable incomes are highest among those that have 15 percent coinsurance
at some point and lowest for those that never reaches the high-cost protection.
Table 3.5. Summary statistics by patient group.
Never in HCP
Coinsurance 50%
Coinsurance 15%
207.05
(623.42)
219.75
(1121.55)
212.68
(263.58)
48.34
(8.48)
52.46
(8.34)
54.83
(7.67)
Female
0.53
0.51
0.52
Born outside Nordic countries
0.02
0.03
0.04
Disposable income, SEK1000s
Age
Compulsory schooling
0.09
0.11
0.13
Secondary schooling
0.47
0.48
0.48
Post-secondary schooling
0.41
0.35
0.29
Employed
0.87
0.81
0.74
No. of days in unemployment
8.56
(42.05)
10.02
(46.45)
12.31
(52.01)
No. sick-spells
0.12
(0.32)
0.14
(0.36)
0.15
(0.38)
No. remaining teeth
28.33
(3.07)
27.14
(3.84)
24.69
(5.15)
No. intact teeth
17.20
(7.03)
12.98
(7.03)
10.28
(7.03)
No. visits year
2.30
(1.22)
4.92
(3.62)
7.73
(5.92)
3,268,261
2,945,605
434,730
Number of observations
Notes: One observation per unique patient & year during the period July 2008-December
2011. Column 2 and 3 refers to patients whose dental care costs are ever reimbursable in
that bracket of the high-cost protection plan. Standard deviations in parentheses.
In sum, the descriptive evidence is mixed. The analysis so far suggests that there
is a social gradient in consumption on the extensive margin but this is not as clear
103
when it comes to the intensive margin. However, individuals whose consumption
ever becomes subject to the subsidy have higher disposable incomes on average while
they also have worse oral health as measured by the number of intact and remaining
teeth. The latter, again, points to the importance of taking age into account in order to
understand the redistributional effects of the insurance.
Figure 3.3. Probabilities of moving to one of the higher consumption brackets, by
income-groups.
.2
.15
0
.05
.1
Pr.(Move from HCP1 to HCP2)
.15
.1
0
.05
Pr.(Move from 0 to HCP1)
.2
.25
Probability of having 15% coinsurance
.25
Probability of having 50% coinsurance
1 2 3 4 5 6 7 8 9 1011121314151617181920
1 2 3 4 5 6 7 8 9 1011121314151617181920
Notes: The left panel plots the probability of moving from the left-most segment of the insurance
to 50 % coinsurance. The right panel plots the probability of moving from 50 % to 15 %
coinsurance. Both probabilities are plotted over twenty equally sized income groups.
3.6 Empirical results
3.6.1 Main results
Figure 3.4 plots estimates from multinomial logit models estimated separately for
age-groups defined over five years for all individuals in the patient sample. The
outcome variable is an indicator for the highest consumption bracket the individual
belongs to within each HCP-period, with the lowest bracket as the reference category.
The estimates in the upper panel of figure 3.4 plots results for the mid consumption
segment, i.e. having 50 % coinsurance rather than paying the full cost of dental care.
The lower panel plots the corresponding estimates for 15 % coinsurance. All estimates
plotted in figure 3.4 are statistically significant on the 1 % level. The results are also
presented in table C5 in Appendix.
Larger disposable incomes consistently decreases the probability for getting dental
care costs covered within both the first and second segment of the insurance compared
to paying market prices for dental care. Put differently, as incomes increases individuals
104
move from the higher consumption brackets to the left-most segment, where dental
care costs are not subsidized. This suggests that we can reject the hypothesis that
consumption responses to the insurance does not differ across the income distribution.
It also suggests that the dental care insurance redistributes progressively in the sense that
it subsidizes dental care for lower-income individuals to a larger extent than for higherincome individuals. However, under the assumption that the distribution of inherited
oral health endowments are the same across age-groups so that the age-patterns reflect
lifetime patterns, the results suggest that the insurance is less progressive for older
individuals than for young individuals. For the oldest age-group, individuals with
higher incomes are more likely to be in the mid consumption segment compared to
paying market prices which is reflected by a positive estimate.
Figure 3.4. Multinomial logit estimates of consumption bracket choice for patient
sample, by age-group. Plotted with 95 % confidence interval.
50 % Coinsurance
Age−groups
35−39
40−44
45−49
50−54
55−59
15 % Coinsurance
60−64
35−39
40−44
45−49
50−54
55−59
60−64
−.5
−.4
−.3
−.2
Income
−.1
0
Notes: Estimates from a multinomial logit model with the highest reached consumption bracket
within each HCP-period as the outcome variable and the no-subsidy bracket as the reference
category. Results for the patient sample, estimated separately by age-group. Models include
year-specific fixed effects.
Figure 3.5 plots corresponding estimates for the retired sample, consisting of nonworking individuals aged 65-89 years.25 All estimates are statistically significant on
the 1 % level. The age-pattern is the same as for the patient sample but the estimates are
consistently positive implying that the dental care insurance subsidizes dental care costs
for individuals with higher incomes to a larger extent than for lower-income individuals.
Hence, the results suggest that the dental care insurance reduces the progressivity of the
Swedish social insurance for non-working elderly individuals. Moreover, the gradient
of the estimates in figure 3.5 is steeper in the lower panel compared to the upper
25
See table C6 in Appendix C for further details.
105
panel implying that the progressivity reduces faster as individuals grow older. This is
problematic from an equity point of view as the oral health stock also decreases faster
as individuals age.
Figure 3.5. Multinomial logit estimates of consumption bracket choice for patient
sample, by age-group for retired sample. Plotted with 95 % confidence intervals.
15 % Coinsurance
50 % Coinsurance
Age−groups
65−69
70−74
75−79
80−84
85−89
65−69
70−74
75−79
80−84
85−89
0
.1
.2
.3
.4
.5
Income
Notes: Estimates from a multinomial logit model with the highest reached consumption bracket
within each HCP-period as the outcome variable and the no-subsidy bracket as the reference
category. Results for the retired sample, estimated separately by age-group. Models include
year-specific fixed effects.
The positive estimates plotted in figure 3.5 could potentially be explained—in part—
by the lower opportunity cost of consuming dental care faced by retired individuals
compared to working individuals. As the opportunity cost is also assumed to be
increasing in incomes this would cause wealthier individuals in the retired sample to
be more likely to be in the upper consumption brackets and get subsidized dental care.
3.6.2 Results by gender
The descriptive analysis suggests that dental care consumption patterns differ across
men and women. The analysis has therefore been performed separately by men
and women and the results are plotted in figure 3.6. All estimates in figure 3.6 are
statistically significant on the 1 % level. The overall pattern is the same as for the main
results. However, only the estimate for women is positive for the oldest age-group
in the patient sample. In addition, progressivity decreases monotonically with age.
The sex pattern is somewhat different for the higher consumption bracket where the
gradient of the estimates is steeper for women than for men.
106
Figure 3.6. Multinomial logit estimates of consumption bracket choice for patient
sample, by age-group and gender. Plotted with 95 % confidence intervals.
50 % Coinsurance
Age−groups
35−39
40−44
45−49
50−54
55−59
60−64
Men
−.5
Women
−.4
−.3
−.2
−.1
0
−.2
−.1
0
Income
15 % Coinsurance
Age−groups
35−39
40−44
45−49
50−54
55−59
60−64
−.5
−.4
−.3
Income
Notes: Estimates from a multinomial logit model with the highest reached consumption bracket
within each HCP-period as the outcome variable and the no-subsidy bracket as the reference
category. Results for the patient sample, estimated separately by age-group and gender. Models
include year-specific fixed effects.
107
Figure 3.7. Multinomial logit estimates of consumption bracket choice for retired
sample, by age-group and gender. Plotted with 95 % confidence intervals.
50 % Coinsurance
Age−groups
65−69
70−74
75−79
80−84
Men
85−89
−.2
Women
0
.2
Income
.4
.6
.4
.6
15 % Coinsurance
Age−groups
65−69
70−74
75−79
80−84
85−89
−.2
0
.2
Income
Notes: Estimates from a multinomial logit model with the highest reached consumption bracket
within each HCP-period as the outcome variable and the no-subsidy bracket as the reference
category. Results for the retired sample, estimated separately by age-group and gender. Models
include year-specific fixed effects.
Figure 3.7 plots estimates by gender for the retired sample. All estimates are
statistically significant at the 1 % level. As with the results plotted in figure 3.5,
the gradient is steeper in the lower panel compared to the upper panel. The results
suggests that the dental care insurance reduces the progressivity of the social insurance
system for all age-groups in the retired sample, except for men aged 64-69 years when
considering the highest consumption bracket. In addition, income responses to the
subsidy are consistently larger for women except for in the oldest age-group.
108
3.7 Concluding remarks
Public health insurance systems are commonly justified on grounds of equity and
as a means of redistributing welfare from high-income individuals to low-income
individuals. This abstracts from a setting where the costs that are covered by the
insurance increase with incomes. This essay investigates the redistributional features
of a public health insurance system in such a setting, namely the Swedish dental care
insurance. The insurance includes cost-sharing subsidies above a threshold of 3,000
SEK (EUR 320), at two different rates depending on the value of consumed care over
a given period. The size of the subsidy increases with dental care costs implying that
the generosity of the insurance increases with consumption. The insurance constitutes
a small part of the relatively large Swedish social insurance system and, therefore,
redistribution within the dental care insurance is defined as whether or not it adds to
the progressive redistribution taking place through other parts of the social insurance.
On average, individuals with higher incomes consume more dental care and have
better oral health compared to individuals with lower incomes. The positive income
elasticity for dental care can be understood from the perspective of the Grossman
model which treats health as a capital stock in which individuals invest through any
health promoting behaviors. Viewing dental care consumption as an investment in oral
health and assuming that oral health in itself is a normal good provides a theoretical
justification for why dental care consumption should increase with incomes. Together
with the inherent moral hazard problem in all health insurance implying that care
consumption is not exclusively driven by need, the redistributional features of the dental
care insurance are ambiguous. While individuals with higher incomes consume more
dental care on average they also have better oral health, suggesting a lower consumption.
Redistribution within the dental care insurance will therefore be determined by the total
effect of income on the response to the insurance.
I find that individuals with different incomes respond differently to the insurance.
For individuals aged 35-59 years, the dental care insurance adds to the progressive
redistribution of the social insurance system; as we move up in the income distribution,
individuals move from the consumption segment where dental care is subsidized to
the lowest consumption bracket, where there is no subsidy. However, the dental care
insurance reduces the progressivity in the social insurance system for individuals aged
60-89 years. This is problematic from an equity point of view as the income gradient
in oral health is more pronounced for the retired sample than in younger age-groups.
Moreover, it is reasonable to assume that the oral health stock deteriorates faster as
individuals age suggesting that older individuals will have a greater need for dental care.
It is however noteworthy that the dental care insurance makes redistribution through
the Swedish social insurance system more progressive for the vast majority of patients.
Especially considering that dental care in Sweden stands out in the welfare sector by
having pronounced market elements such as free price-setting and fairly large private
co-payments and the equity concerns commonly surrounding such arrangements.
109
References
Aron-Dine, A., Einav, L. and Finkelstein, A. (2013). The RAND Health Insurance
Experiment, Three Decades Later. Journal of Economic Perspectives, 27 (1),
197–222.
Arrow, K. (1963). Uncertainty and the welfare economics of medical care. The
American Economic Review, 53 (5), 941–973.
Blazquez-Fernandez, C., Cantarero, D. and Perez, P. (2014). Disentangling the
heterogeneous income elasticity and dynamics of health expenditure. Applied
Economics, 46 (16), 1839–1854.
Blomqvist, Å. and Horn, H. (1984). Public health insurance and optimal income
taxation. Journal of Public Economics, 24 (3), 353–371.
Boadway, R., Leite-Monteiro, M., Marchand, M. and Pestieau, P. (2003). Social
insurance and redistribution. Public Finance and Public Policy in the New
Century, MIT Press, Cambridge, MA, 333, 358.
—, —, — and — (2006). Social Insurance and Redistribution with Moral Hazard and
Adverse Selection. The Scandinavian Journal of Economics, 108 (2), 279–298.
Bredenkamp, C., Wagstaff, A., Bilger, M. and Buisman, L. R. (2014). Who
benefits from government health spending and why? A global assessment. World
Bank. Policy Research Working Paper 7044.
Breyer, F., Bundorf, M. K. and Pauly, M. V. (2011). Chapter eleven- Health Care
Spending Risk, Health Insurance, and Payment to Health Plans. Handbook of
Health Economics, vol. 2, Elsevier, pp. 691–762.
Cremer, H. and Pestieau, P. (1996). Redistributive taxation and social insurance.
International Tax and Public Finance, 3 (3), 281–295.
Cutler, D. M. and Zeckhauser, R. J. (2000). Chapter 11 The anatomy of health
insurance. In A. J. Culyer and J. P. Newhouse (eds.), Handbook of Health
Economics, vol. 1, Part A, Elsevier, pp. 563–643.
Dave, D. and Kaestner, R. (2009). Health insurance and ex ante moral hazard:
evidence from Medicare. International journal of health care finance and
economics, 9 (4), 367–390.
Devaux, M. (2013). Income-related inequalities and inequities in health care services
utilisation in 18 selected OECD countries. The European Journal of Health
Economics, 16 (1), 21–33.
Eurostat (2015). Social protection statistics. http://ec.europa.eu/eurostat/
statistics-explained/index.php/Social_protection_statistics.
Finkelstein, A., Taubman, S., Wright, B., Bernstein, M., Gruber, J., Newhouse,
J. P., Allen, H. and Baicker, K. (2012). The Oregon Health Insurance
Experiment: Evidence from the First Year. The Quarterly Journal of Economics,
127 (3), 1057–1106.
Grönqvist, E. (2006). (M)oral hazard? SSE/EFI Working Paper Series in
Economics and Finance, No. 642.
Grossman, M. (1972a). The Demand for Health: A Theoretical and Empirical
Investigation. New York: NBER, Occasional Paper 119, Columbia Press.
— (1972b). On the concept of health capital and the demand for health. The Journal
of Political Economy, 80 (2), 223.
— (2000). Chapter 7 The human capital model. In A. J. Culyer and J. P. Newhouse
(eds.), Handbook of Health Economics, vol. 1, Part A, Elsevier, pp. 347–408.
110
Mirrlees, J. A. (1971). An exploration in the theory of optimum income taxation.
The Review of Economic Studies, pp. 175–208.
Morris, S., Sutton, M. and Gravelle, H. (2005). Inequity and inequality in the use
of health care in England: an empirical investigation. Social Science & Medicine,
60 (6), 1251–1266.
Newhouse, J. P., Rand Corporation and Insurance Experiment Group (1996).
Free for all?: Lessons from the Rand Health Insurance Experiment. Cambridge,
Mass.: Harvard University Press.
Nybom, M. and Stuhler, J. (2015). Heterogeneous Income Profiles and Life-Cycle
Bias in Intergenerational Mobility Estimation. Forthcoming. Journal of Human
Resources.
OECD (2013). Health at a glance 2013: OECD indicators. Paris: OECD.
OECD (2014). Social Expenditure Update - Social spending is falling in some
countries, but in many others it remains at historically high levels. Paris: OECD.
OECD and European Union (2014). Health at a Glance: Europe 2014. Paris:
OECD.
Pauly, M. V. (1968). The economics of moral hazard: Comment. The American
Economic Review, 58 (3), 531–537.
— (2011). Insurance and the Demand for Medical Care. In S. Glied and P. C. Smith
(eds.), The Oxford Handbook of Health Economics, Oxford : New York: Oxford
University Press.
Poterba, J. M. (1996). Government intervention in the markets for education and
health care: how and why? In Individual and social responsibility: Child care,
education, medical care, and long-term care in America, University of Chicago
Press, pp. 277–308.
Rochet, J.-C. (1991). Incentives, redistribution and social insurance. The Geneva
Papers on Risk and Insurance Theory, 16 (2), 143–165.
Sintonen, H. and Linnosmaa, I. (2000). Chapter 24 Economics of dental services.
In A. J. Culyer and J. P. Newhouse (eds.), Handbook of Health Economics, vol. 1,
Part B, Elsevier, pp. 1251–1296.
Van Doorslaer, E., Koolman, X. and Jones, A. M. (2004). Explaining
income-related inequalities in doctor utilisation in Europe. Health Economics,
13 (7), 629–647.
—, Wagstaff, A., Van der Burg, H., Christiansen, T., De Graeve, D.,
Duchesne, I., Gerdtham, U.-G., Gerfin, M., Geurts, J. and Gross, L. (2000).
Equity in the delivery of health care in Europe and the US. Journal of Health
Economics, 19 (5), 553–583.
111
Appendix C
C1 Descriptive statistics, retired sample.
Table C1. Summary statistics for retired sample.
Non-working, aged 65-
Non-patients
Patients
170.42
(190.34)
142.10
(120.93)
184.78
(215.81)
76.06
(7.90)
79.19
(8.60)
74.48
(7.02)
Female
0.57
0.58
0.56
Born outside Nordic countries
0.01
0.02
0.01
Disposable income, SEK1000s
Age
Compulsory schooling
0.07
0.06
0.07
Secondary schooling
0.34
0.27
0.38
Post-secondary schooling
0.16
0.07
0.21
Number of observations
4,328,259
1,456,408
2,871,851
Notes: One observation per year & unique non-working individual aged 65 years or above
during the period July 2008-December 2011. Non-patients are never patients during the
studied period whereas patients have some positive consumption during the period. Standard
deviations in parentheses.
Table C2. Elasticities of dental care consumption with respect to income and oral
health. Retired sample.
(1)
Log. disposable income
(2)
∗∗∗
0.2603
(0.0023)
-0.1878∗∗∗
(0.0032)
Log. oral health index
Number of observations
4,328,255
2,134,370
Notes: One observation per year & unique non-working individual aged 65 years or above
during the period July 2008-December 2011. Poisson regression estimates of elasticities of
dental care with respect to income and oral health. Column 2 reports estimates for patients
with non-missing oral health measures. Models include a dummy for sex and fixed effects
for year and age-group. Standard errors in parenthesis, clustered on individual level. ∗
p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
112
Table C3. Extensive and intensive margin of consumption. Retired sample.
(1)
Prob. of any consumption
(2)
Log. value of consumption
Log. disposable income
0.3716∗∗∗
(0.0026)
0.0711∗∗∗
(0.0016)
Female
0.1662∗∗∗
(0.0025)
-0.0754∗∗∗
(0.0016)
Compulsory schooling
0.2708∗∗∗
(0.0048)
0.1138∗∗∗
(0.0032)
Secondary schooling
0.3569∗∗∗
(0.0027)
0.0890∗∗∗
(0.0019)
Post-secondary schooling
0.6743∗∗∗
(0.0039)
0.1495∗∗∗
(0.0022)
Born outside Nordic countries
-0.7300∗∗∗
(0.0096)
0.1038∗∗∗
(0.0099)
Constant
Number of observations
-1.9242
7.2267
4,328,255
2,863,481
Notes: One observation per year & unique non-working individual aged 65 or above during
the period July 2008-December 2011. Age is controlled for with linear, square and cubic
term. Models include year dummies. Standard errors in parenthesis, clustered on individual
level. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
113
Table C4. Summary statistics by patient group. Retired sample.
Never in HCP
Coinsurance 50%
Coinsurance 15%
174.68
(202.02)
192.51
(225.49)
194.61
(232.34)
Female
0.59
(0.49)
0.54
(0.50)
0.54
(0.50)
Age
75.34
(7.20)
73.82
(6.80)
73.57
(6.72)
Level of education
2.52
(1.46)
2.76
(1.49)
2.73
(1.47)
No. of days in unemployment
0.24
(6.91)
0.34
(8.13)
0.39
(8.88)
Born outside Nordic countries
0.01
(0.08)
0.01
(0.08)
0.01
(0.10)
No. sick-spells
0.01
(0.07)
0.01
(0.08)
0.01
(0.09)
No. remaining teeth
22.91
(5.84)
23.34
(5.32)
21.70
(5.61)
No. intact teeth
8.26
(6.50)
7.61
(6.03)
6.94
(6.18)
1,244,348
1,627,503
322,463
Disposable income, SEK1000s
Number of observations
Notes: One observation per year & unique non-working patient aged 65 or above during the
period July 2008 - December 2011. Column 2 & 3 refers to patients whose consumption
is ever covered by the subsidy in that bracket. No. visits and no. treatments refers to the
number within each subsidy period. Standard deviations in parentheses.
114
C2 Further notes on the empirical strategy
Let S represent the three different consumption segments where S = 0 is the segment
where individuals bear the full cost of their consumption. The underlying optimal
dental care consumption can be mapped to the different segments of the insurance:
S=0
S=1
if c∗ ≤ 3, 000
if 3, 000 < c∗ ≤ 15, 000
S=2
if c∗ > 15, 000.
By normalizing the price p to 1 and using model (3.2) we get:
S=0
S=1
if (α00 + α10 ht (y) + α20 y) ≤ 3, 000 − εt
if 3, 000 − εt < (a01 + α11 ht (y) + α21 y) ≤ 15, 000 − εt
S=2
if (a02 + α12 ht (y) + a22 y) > 15, 000.
Where α00 = a0 + a3 , α10 = (a1 + a5 ), α20 = (a2 + a4 ), α01 = a0 + 0.5a3 , α11 =
a1 + 0.5a5 , α21 = a2 + 0.5a4 , α02 = a0 + 0.15a3 , α12 = a1 + 0.15a5 , α22 =
a2 + 0.15a4 . Note that:
α21 − α20 = a4 (0.5 − 1) and α22 − α20 = a4 (0.15 − 1)
and
α11 − α10 = a5 (0.5 − 1) and α12 − α10 = a5 (0.15 − 1).
This means that the hypothesis given in equation (3.3) can be tested by obtaining an
estimate of the average marginal effect from y on dental care consumption from a price
change. This effect is obtained directly from the reduced form which includes income
only. The reduced form model is:
S=0
S=1
if (β00t + β10t y) ≤ 3, 000 − εt
if 3, 000 − εt < (β01t + β11t y) ≤ 15, 000 − εt
S=2
if (β02t + β12t y) > 15, 000.
where:
β10t =
dc∗t
dc∗
dc∗
dc∗
dc∗
|p=1 , β11t = t |p − t |ps =0.5 and β12t = t |p − t |ps =0.15 .
dy
dy
dy
dy
dy
Note that the parameters are indexed by age to account for that the oral health stock
varies over the life-cycle and is likely a function of y. Dental care consumption is
also likely to vary with age due to the age-related depreciation and the parameters of
interest are therefore estimated separately for age-groups defined over five years for all
individuals aged 35-64.
115
C3 Result tables
Table C5. Multinomial logit estimates of consumption bracket choice, by age-group.
Age-group 35-39
Log. disposable income
Number of observations
Age-group 40-44
Log. disposable income
Number of observations
Age-group 45-49
Log. disposable income
Number of observations
Age-group 50-54
Log. disposable income
Number of observations
Age-group 55-59
Log. disposable income
Number of observations
Age-group 60-64
Log. disposable income
Number of observations
50 % coinsurance
15 % coinsurance
-0.1852∗∗∗
(0.0048)
-0.4552∗∗∗
(0.0129)
910,426
910,426
-0.1437∗∗∗
(0.0046)
-0.4142∗∗∗
(0.0118)
1,020,711
1,020,711
-0.1052∗∗∗
(0.0043)
-0.3728∗∗∗
(0.0100)
1,063,462
1,063,462
-0.0702∗∗∗
(0.0040)
-0.3271∗∗∗
(0.0085)
1,062,764
1,062,764
-0.0288∗∗∗
(0.0036)
-0.2465∗∗∗
(0.0072)
1,139,313
1,139,313
0.0250∗∗∗
(0.0033)
-0.1301∗∗∗
(0.0063)
1,291,229
1,291,229
Notes: Estimates from a mulitnomial logit model estimated separately for age-groups in
the patient sample. The model includes year fixed-effects. Standard errors in parenthesis,
clustered on individual level. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
116
Table C6. Multinomial logit estimates of consumption bracket choice for retired
sample, by age-group.
Age-group 65-69
Log. disposable income
Number of observations
Age-group 70-74
Log. disposable income
Number of observations
Age-group 75-79
Log. disposable income
Number of observations
Age-group 80-84
Log. disposable income
Number of observations
Age-group 85-89
Log. disposable income
Number of observations
50 % coinsurance
15 % coinsurance
0.1201∗∗∗
(0.0043)
0.0075
(0.0086)
932,601
932,601
0.1770∗∗∗
(0.0052)
0.1356∗∗∗
(0.0108)
774,095
774,095
0.2070∗∗∗
(0.0065)
0.2094∗∗∗
(0.0131)
599,475
599,475
0.2305∗∗∗
(0.0083)
0.2959∗∗∗
(0.0165)
431,244
431,244
0.2717∗∗∗
(0.0118)
0.4345∗∗∗
(0.0226)
236,636
236,636
Notes: Estimates from a mulitnomial logit model estimated separately for each age-group
in the retired sample. The model includes year fixed-effects. Standard errors in parenthesis,
clustered on individual level. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
117
Table C7. Multinomial logit estimates of consumption bracket choice, by age-group
and sex.
Men
Age-group 35-39
Log. disposable income
Number of observations
Age-group 40-44
Log. disposable income
Number of observations
Age-group 45-49
Log. disposable income
Number of observations
Age-group 50-54
Log. disposable income
Number of observations
Age-group 55-59
Log. disposable income
Number of observations
Age-group 60-64
Log. disposable income
Number of observations
Women
50 %
coinsurance
15 %
coinsurance
50 %
coinsurance
15 %
coinsurance
-0.1895∗∗∗
(0.0071)
-0.4776∗∗∗
(0.0176)
-0.2120∗∗∗
(0.0069)
-0.4864∗∗∗
(0.0188)
429,661
-0.1651∗∗∗
(0.0067)
480,765
-0.4580∗∗∗
(0.0162)
488,167
-0.1320∗∗∗
(0.0062)
-0.4374∗∗∗
(0.0134)
-0.4031∗∗∗
(0.0115)
621,706
-0.3328∗∗∗
(0.0152)
-0.0701∗∗∗
(0.0058)
-0.2745∗∗∗
(0.0127)
551,521
-0.3151∗∗∗
(0.0099)
548,089
-0.0152∗∗∗
(0.0049)
-0.0911∗∗∗
(0.0061)
551,925
511,243
-0.0616∗∗∗
(0.0052)
-0.4013∗∗∗
(0.0174)
532,544
511,537
-0.0916∗∗∗
(0.0058)
-0.1418∗∗∗
(0.0066)
-0.0420∗∗∗
(0.0053)
-0.2199∗∗∗
(0.0108)
591,224
-0.1996∗∗∗
(0.0090)
0.0039
(0.0048)
-0.1226∗∗∗
(0.0093)
669,523
Notes: Estimates from a mulitnomial logit model estimated separately for age-groups in
the patient sample. The model includes year fixed-effects. Standard errors in parenthesis,
clustered on individual level. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
118
Table C8. Multinomial logit estimates of consumption bracket choice for retired
sample, by age-group and sex.
Men
Age-group 65-69
Log. disposable income
Number of observations
Age-group 70-74
Log. disposable income
Number of observations
Age-group 75-79
Log. disposable income
Number of observations
Age-group 80-84
Log. disposable income
Number of observations
Age-group 85-89
Log. disposable income
Number of observations
Women
50 %
5 coinsurance
15 %
coinsurance
50 %
coinsurance
15 %
coinsurance
0.0659∗∗∗
(0.0066)
-0.1010∗∗∗
(0.0130)
0.0976∗∗∗
(0.0061)
0.0241∗∗
(0.0122)
422,214
0.1325∗∗∗
(0.0082)
510,387
0.0628∗∗∗
(0.0171)
351,784
0.1615∗∗∗
(0.0102)
0.1684∗∗∗
(0.0212)
91,299
0.1743∗∗∗
(0.0092)
0.1989∗∗∗
(0.0181)
336,896
0.2341∗∗∗
(0.0271)
177,877
0.2369∗∗∗
(0.0186)
0.1326∗∗∗
(0.0153)
422,311
262,579
0.1807∗∗∗
(0.0129)
0.1443∗∗∗
(0.0075)
0.2098∗∗∗
(0.0115)
0.2971∗∗∗
(0.0220)
253,367
0.4075∗∗∗
(0.0348)
0.2430∗∗∗
(0.0161)
0.4023∗∗∗
(0.0316)
145,337
Notes: Estimates from a mulitnomial logit model estimated separately for age-groups in
the patient sample. The model includes year fixed-effects. Standard errors in parenthesis,
clustered on individual level. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
119
Economic Studies
____________________________________________________________________
1987:1 1989:1 1989:2 1989:3 1989:4 1990:1 1990:2 1991:1 1992:1 1992:2 1992:3 12 13 14 15 16 17 Haraldson, Marty: To Care and To Cure. A linear programming approach to national health planning in developing countries. 98 pp. Chryssanthou, Nikos: The Portfolio Demand for the ECU: A Transaction Cost Approach. 42 pp. Hansson, Bengt: Construction of Swedish Capital Stocks, 1963-­‐‑87: An Application of the Hulten-­‐‑Wykoff Studies. 37 pp. Choe, Byung-­‐‑Tae: Some Notes on Utility Functions Demand and Aggregation. 39 pp. Skedinger, Per: Studies of Wage and Employment Determination in the Swedish Wood Industry. 89 pp. Gustafson, Claes-­‐‑Håkan: Inventory Investment in Manufacturing Firms. Theory and Evidence. 98 pp. Bantekas, Apostolos: The Demand for Male and Female Workers in Swedish Manufacturing. 56 pp. Lundholm, Michael: Compulsory Social Insurance. A Critical Review. 109 pp. Sundberg, Gun: The Demand for Health and Medical Care in Sweden. 58 pp. Gustavsson, Thomas: No Arbitrage Pricing and the Term Structure of Interest Rates. 47 pp. Elvander, Nils: Labour Market Relations in Sweden and Great Britain. A Com-­‐‑
parative 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. Banks, Ferdinand E.: A Modern Introduction to International Money, Banking and Finance. 1993. 303 pp. Mellander, Erik: Measuring Productivity and Inefficiency Without Quantitative Output Data. 1993. 140 pp. Ackum Agell: Susanne: Essays on Work and Pay. 1993. 116 pp. Eriksson, Claes: Essays on Growth and Distribution. 1994. 129 pp. Banks, Ferdinand E.: A Modern Introduction to International Money, Banking and Finance. 2nd version, 1994. 313 pp. 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 Apel, Mikael: Essays on Taxation and Economic Behavior. 1994. 144 pp. Dillén, Hans: Asset Prices in Open Monetary Economies. A Contingent Claims Approach. 1994. 100 pp. Jansson, Per: Essays on Empirical Macroeconomics. 1994. 146 pp. Banks, Ferdinand E.: A Modern Introduction to International Money, Banking, and Finance. 3rd version, 1995. 313 pp. Dufwenberg, Martin: On Rationality and Belief Formation in Games. 1995. 93 pp. Lindén, Johan: Job Search and Wage Bargaining. 1995. 127 pp. Shahnazarian, Hovick: Three Essays on Corporate Taxation. 1996. 112 pp. Svensson, Roger: Foreign Activities of Swedish Multinational Corporations. 1996. 166 pp. Sundberg, Gun: Essays on Health Economics. 1996. 174 pp. Sacklén, Hans: Essays on Empirical Models of Labor Supply. 1996. 168 pp. Fredriksson, Peter: Education, Migration and Active Labor Market Policy. 1997. 106 pp. Ekman, Erik: Household and Corporate Behaviour under Uncertainty. 1997. 160 pp. Stoltz, Bo: Essays on Portfolio Behavior and Asset Pricing. 1997. 122 pp. Dahlberg, Matz: Essays on Estimation Methods and Local Public Economics. 1997. 179 pp. Kolm, Ann-­‐‑Sofie: Taxation, Wage Formation, Unemployment and Welfare. 1997. 162 pp. Boije, Robert: Capitalisation, Efficiency and the Demand for Local Public Services. 1997. 148 pp. Hort, Katinka: On Price Formation and Quantity Adjustment in Swedish Housing Markets. 1997. 185 pp. Lindström, Thomas: Studies in Empirical Macroeconomics. 1998. 113 pp. Hemström, Maria: Salary Determination in Professional Labour Markets. 1998. 127 pp. Forsling, Gunnar: Utilization of Tax Allowances and Corporate Borrowing. 1998. 96 pp. Nydahl, Stefan: Essays on Stock Prices and Exchange Rates. 1998. 133 pp. Bergström, Pål: Essays on Labour Economics and Econometrics. 1998. 163 pp. 40 41 42 43 44
45
46
47
48
49
50
51
52
53
54
55 56 57 58 59 60 61 Heiborn, Marie: Essays on Demographic Factors and Housing Markets. 1998. 138 pp. Åsberg, Per: Four Essays in Housing Economics. 1998. 166 pp. Hokkanen, Jyry: Interpreting Budget Deficits and Productivity Fluctuations. 1998. 146 pp. Lunander, Anders: Bids and Values. 1999. 127 pp. Eklöf, Matias: Studies in Empirical Microeconomics. 1999. 213 pp. Johansson, Eva: Essays on Local Public Finance and Intergovernmental Grants. 1999. 156 pp. Lundin, Douglas: Studies in Empirical Public Economics. 1999. 97 pp. Hansen, Sten: Essays on Finance, Taxation and Corporate Investment. 1999. 140 pp. Widmalm, Frida: Studies in Growth and Household Allocation. 2000. 100 pp. Arslanogullari, Sebastian: Household Adjustment to Unemployment. 2000. 153 pp. Lindberg, Sara: Studies in Credit Constraints and Economic Behavior. 2000. 135 pp. Nordblom, Katarina: Essays on Fiscal Policy, Growth, and the Importance of Family Altruism. 2000. 105 pp. Andersson, Björn: Growth, Saving, and Demography. 2000. 99 pp. Åslund, Olof: Health, Immigration, and Settlement Policies. 2000. 224 pp. Bali Swain, Ranjula: Demand, Segmentation and Rationing in the Rural Credit Markets of Puri. 2001. 160 pp. Löfqvist, Richard: Tax Avoidance, Dividend Signaling and Shareholder Taxation in an Open Economy. 2001. 145 pp. Vejsiu, Altin: Essays on Labor Market Dynamics. 2001. 209 pp. Zetterström, Erik: Residential Mobility and Tenure Choice in the Swedish Housing Market. 2001. 125 pp. Grahn, Sofia: Topics in Cooperative Game Theory. 2001. 106 pp. Laséen, Stefan: Macroeconomic Fluctuations and Microeconomic Adjustments: Wages, Capital, and Labor Market Policy. 2001. 142 pp. Arnek, Magnus: Empirical Essays on Procurement and Regulation. 2002. 155 pp. Jordahl, Henrik: Essays on Voting Behavior, Labor Market Policy, and Taxation. 2002. 172 pp. 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 Lindhe, Tobias: Corporate Tax Integration and the Cost of Capital. 2002. 102 pp. Hallberg, Daniel: Essays on Household Behavior and Time-­‐‑Use. 2002. 170 pp. Larsson, Laura: Evaluating Social Programs: Active Labor Market Policies and Social Insurance. 2002. 126 pp. Bergvall, Anders: Essays on Exchange Rates and Macroeconomic Stability. 2002. 122 pp. Nordström Skans, Oskar: Labour Market Effects of Working Time Reductions and Demographic Changes. 2002. 118 pp. Jansson, Joakim: Empirical Studies in Corporate Finance, Taxation and Investment. 2002. 132 pp. Carlsson, Mikael: Macroeconomic Fluctuations and Firm Dynamics: Technology, Production and Capital Formation. 2002. 149 pp. Eriksson, Stefan: The Persistence of Unemployment: Does Competition between Employed and Unemployed Job Applicants Matter? 2002. 154 pp. Huitfeldt, Henrik: Labour Market Behaviour in a Transition Economy: The Czech Experience. 2003. 110 pp. Johnsson, Richard: Transport Tax Policy Simulations and Satellite Accounting within a CGE Framework. 2003. 84 pp. Öberg, Ann: Essays on Capital Income Taxation in the Corporate and Housing Sectors. 2003. 183 pp. Andersson, Fredrik: Causes and Labor Market Consequences of Producer Heterogeneity. 2003. 197 pp. Engström, Per: Optimal Taxation in Search Equilibrium. 2003. 127 pp. Lundin, Magnus: The Dynamic Behavior of Prices and Investment: Financial Constraints and Customer Markets. 2003. 125 pp. Ekström, Erika: Essays on Inequality and Education. 2003. 166 pp. Barot, Bharat: Empirical Studies in Consumption, House Prices and the Accuracy of European Growth and Inflation Forecasts. 2003. 137 pp. Österholm, Pär: Time Series and Macroeconomics: Studies in Demography and Monetary Policy. 2004. 116 pp. Bruér, Mattias: Empirical Studies in Demography and Macroeconomics. 2004. 113 pp. Gustavsson, Magnus: Empirical Essays on Earnings Inequality. 2004. 154 pp. 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 Toll, Stefan: Studies in Mortgage Pricing and Finance Theory. 2004. 100 pp. Hesselius, Patrik: Sickness Absence and Labour Market Outcomes. 2004. 109 pp. Häkkinen, Iida: Essays on School Resources, Academic Achievement and Student Employment. 2004. 123 pp. Armelius, Hanna: Distributional Side Effects of Tax Policies: An Analysis of Tax Avoidance and Congestion Tolls. 2004. 96 pp. Ahlin, Åsa: Compulsory Schooling in a Decentralized Setting: Studies of the Swedish Case. 2004. 148 pp. Heldt, Tobias: Sustainable Nature Tourism and the Nature of Tourists'ʹ Cooperative Behavior: Recreation Conflicts, Conditional Cooperation and the Public Good Problem. 2005. 148 pp. Holmberg, Pär: Modelling Bidding Behaviour in Electricity Auctions: Supply Function Equilibria with Uncertain Demand and Capacity Constraints. 2005. 43 pp. Welz, Peter: Quantitative new Keynesian macroeconomics and monetary policy 2005. 128 pp. Ågren, Hanna: Essays on Political Representation, Electoral Accountability and Strategic Interactions. 2005. 147 pp. Budh, Erika: Essays on environmental economics. 2005. 115 pp. Chen, Jie: Empirical Essays on Housing Allowances, Housing Wealth and Aggregate Consumption. 2005. 192 pp. Angelov, Nikolay: Essays on Unit-­‐‑Root Testing and on Discrete-­‐‑Response Modelling of Firm Mergers. 2006. 127 pp. Savvidou, Eleni: Technology, Human Capital and Labor Demand. 2006. 151 pp. Lindvall, Lars: Public Expenditures and Youth Crime. 2006. 112 pp. Söderström, Martin: Evaluating Institutional Changes in Education and Wage Policy. 2006. 131 pp. Lagerström, Jonas: Discrimination, Sickness Absence, and Labor Market Policy. 2006. 105 pp. Johansson, Kerstin: Empirical essays on labor-­‐‑force participation, matching, and trade. 2006. 168 pp. Ågren, Martin: Essays on Prospect Theory and the Statistical Modeling of Financial Returns. 2006. 105 pp. 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 Nahum, Ruth-­‐‑Aïda: Studies on the Determinants and Effects of Health, Inequality and Labour Supply: Micro and Macro Evidence. 2006. 153 pp. Žamac, Jovan: Education, Pensions, and Demography. 2007. 105 pp. Post, Erik: Macroeconomic Uncertainty and Exchange Rate Policy. 2007. 129 pp. Nordberg, Mikael: Allies Yet Rivals: Input Joint Ventures and Their Competitive Effects. 2007. 122 pp. Johansson, Fredrik: Essays on Measurement Error and Nonresponse. 2007. 130 pp. Haraldsson, Mattias: Essays on Transport Economics. 2007. 104 pp. Edmark, Karin: Strategic Interactions among Swedish Local Governments. 2007. 141 pp. Oreland, Carl: Family Control in Swedish Public Companies. Implications for Firm Performance, Dividends and CEO Cash Compensation. 2007. 121 pp. Andersson, Christian: Teachers and Student Outcomes: Evidence using Swedish Data. 2007. 154 pp. Kjellberg, David: Expectations, Uncertainty, and Monetary Policy. 2007. 132 pp. Nykvist, Jenny: Self-­‐‑employment Entry and Survival -­‐‑ Evidence from Sweden. 2008. 94 pp. Selin, Håkan: Four Empirical Essays on Responses to Income Taxation. 2008. 133 pp. Lindahl, Erica: Empirical studies of public policies within the primary school and the sickness insurance. 2008. 143 pp. Liang, Che-­‐‑Yuan: Essays in Political Economics and Public Finance. 2008. 125 pp. Elinder, Mikael: Essays on Economic Voting, Cognitive Dissonance, and Trust. 2008. 120 pp. Grönqvist, Hans: Essays in Labor and Demographic Economics. 2009. 120 pp. Bengtsson, Niklas: Essays in Development and Labor Economics. 2009. 93 pp. Vikström, Johan: Incentives and Norms in Social Insurance: Applications, Identification and Inference. 2009. 205 pp. 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. 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 Nilsson, Peter: Essays on Social Interactions and the Long-­‐‑term Effects of Early-­‐‑life Conditions. 2009. 180 pp. Johansson, Elly-­‐‑Ann: Essays on schooling, gender, and parental leave. 2010. 131 pp. Hall, Caroline: Empirical Essays on Education and Social Insurance Policies. 2010. 147 pp. Enström-­‐‑Öst, Cecilia: Housing policy and family formation. 2010. 98 pp. Winstrand, Jakob: Essays on Valuation of Environmental Attributes. 2010. 96 pp. Söderberg, Johan: Price Setting, Inflation Dynamics, and Monetary Policy. 2010. 102 pp. Rickne, Johanna: Essays in Development, Institutions and Gender. 2011. 138 pp. Hensvik, Lena: The effects of markets, managers and peers on worker outcomes. 2011. 179 pp. Lundqvist, Heléne: Empirical Essays in Political and Public. 2011. 157 pp. Bastani, Spencer: Essays on the Economics of Income Taxation. 2012. 257 pp. Corbo, Vesna: Monetary Policy, Trade Dynamics, and Labor Markets in Open Economies. 2012. 262 pp. Nordin, Mattias: Information, Voting Behavior and Electoral Accountability. 2012. 187 pp. Vikman, Ulrika: Benefits or Work? Social Programs and Labor Supply. 2013. 161 pp. Ek, Susanne: Essays on unemployment insurance design. 2013. 136 pp. Österholm, Göran: Essays on Managerial Compensation. 2013. 143 pp. Adermon, Adrian: Essays on the transmission of human capital and the impact of technological change. 2013. 138 pp. Kolsrud, Jonas: Insuring Against Unemployment 2013. 140 pp. Hanspers, Kajsa: Essays on Welfare Dependency and the Privatization of Welfare Services. 2013. 208 pp. Persson, Anna: Activation Programs, Benefit Take-­‐‑Up, and Labor Market Attachment. 2013. 164 pp. Engdahl, Mattias: International Mobility and the Labor Market. 2013. 216 pp. Krzysztof Karbownik. Essays in education and family economics. 2013. 182 pp. 141 142 143 144 145 146 147 148 Oscar Erixson. Economic Decisions and Social Norms in Life and Death Situations. 2013. 183 pp. Pia Fromlet. Essays on Inflation Targeting and Export Price Dynamics. 2013. 145 pp. Daniel Avdic. Microeconometric Analyses of Individual Behavior in Public Welfare Systems. Applications in Health and Education Economics. 2014. 176 pp. Arizo Karimi. Impacts of Policies, Peers and Parenthood on Labor Market Outcomes. 2014. 221 pp. Karolina Stadin. Employment Dynamics. 2014. 134 pp. Haishan Yu. Essays on Environmental and Energy Economics. 132 pp. Martin Nilsson. Essays on Health Shocks and Social Insurance. 139 pp. Tove Eliasson. Empirical Essays on Wage Setting and Immigrant Labor Market Opportunities. 2014. 144 pp. 149 Erik Spector. Financial Frictions and Firm Dynamics. 2014. 129 pp. 150 151 Michihito Ando. Essays on the Evaluation of Public Policies. 2015. 193 pp. 152 153 154 155 156 Selva Bahar Baziki. Firms, International Competition, and the Labor Market. 2015. 183 pp. Fredrik Sävje. What would have happened? Four essays investigating causality. 2015. 229 pp. Ina Blind. Essays on Urban Economics. 2015. 197 pp. Jonas Poulsen. Essays on Development and Politics in Sub-­‐‑Saharan Africa. 2015. 240 pp. Lovisa Persson. Essays on Politics, Fiscal Institutions, and Public Finance. 2015. 137 pp. Gabriella Chirico Willstedt. Demand, Competition and Redistribution in Swedish Dental Care. 2015. 119 pp.