Subjective Well-Being:
An Intersection between Economics and Psychology
By
Christopher J. Boyce
Thesis submitted in fulfilment of the requirements for the degree of
Doctor of Philosophy in Psychology
University of Warwick, Department of Psychology
September 2009
ii
TABLE OF CONTENTS
1
Subjective Well-Being Research: An Overview...................................................................... 1
1.1
The Development of Subjective Well-Being Research in Economics and Psychology .. 2
1.1.1
The Use of Subjective Well-Being Data as a Proxy for Utility ............................... 3
1.1.2
Subjective Well-Being – A Viable Tool for Economic Analysis ............................ 6
1.2
Overview of Key Research Areas .................................................................................... 8
1.2.1
Income and Well-Being ........................................................................................... 9
1.2.1.1 Evidence of a Relationship between Income and Well-Being............................. 9
1.2.1.1.1 Income and Well-Being over Time................................................................ 9
1.2.1.1.2 Income and Well-Being within a Country ................................................... 10
1.2.1.1.3 Income and Well-Being across Countries.................................................... 10
1.2.1.2 Explaining the Income and Well-Being Data – Relative Income Effects ......... 11
1.2.1.2.1 Relative Income Effects in Economics ........................................................ 12
1.2.1.2.2 Relative Judgment Models in Psychology ................................................... 13
1.2.1.2.3 Rank Income Effects .................................................................................... 14
1.2.1.3 Explaining the Income and Well-Being Data – Income is relatively
Unimportant for Well-Being .............................................................................................. 15
1.2.1.4 Explaining the Income and Well-Being Data – Personality .............................. 18
1.2.1.4.1 Controlling for Personality in Economics.................................................... 20
1.2.1.4.2 Personality Interacts with Demographic Characteristics ............................. 21
1.2.2
Employment Status and Well-Being...................................................................... 23
1.2.2.1 Unemployment ................................................................................................... 23
1.2.2.2 Occupational Status ........................................................................................... 24
1.3
Overview of the Thesis .................................................................................................. 25
2
Money and Happiness: Rank of Income, not Income, Affects Life Satisfaction .................. 27
2.1
Abstract .......................................................................................................................... 27
2.2
Introduction .................................................................................................................... 28
2.3
Method ........................................................................................................................... 30
2.4
Results ............................................................................................................................ 31
2.5
Discussion ...................................................................................................................... 36
3
Money or Mental Health: The Cost of Alleviating Psychological Distress with Monetary
Compensation versus Psychological Therapy ............................................................................... 38
3.1
Abstract .......................................................................................................................... 38
3.2
Introduction .................................................................................................................... 39
3.3
Money - A Common Metric for valuing Life Events and the Movement towards
Compensation ............................................................................................................................ 40
3.4
The Clinical and Cost Effectiveness of Psychological Therapy .................................... 41
3.5
A Cost Effectiveness Comparison between Psychological Therapy and Direct Financial
Compensation ............................................................................................................................ 42
3.6
Practical Implications of our Argument ......................................................................... 45
3.6.1
For Judges .............................................................................................................. 45
3.6.2
For Policy Makers and Society .............................................................................. 47
4
Understanding Fixed Effects in Human Well-Being ............................................................. 49
4.1
Abstract .......................................................................................................................... 49
4.2
Introduction .................................................................................................................... 50
iii
4.3
The Use of Personality Measures in Economics............................................................ 56
4.4
Methodology .................................................................................................................. 58
4.5
Data ................................................................................................................................ 63
4.6
Results ............................................................................................................................ 67
4.7
Conclusion ..................................................................................................................... 77
4.8
Appendix ........................................................................................................................ 80
4.8.1
Note to Tables ........................................................................................................ 80
4.8.2
Personality Variables in GSOEP ........................................................................... 81
4.8.2.1 Big Five Personality Inventory .......................................................................... 81
4.8.2.2 Positive and Negative Reciprocity ..................................................................... 82
4.8.2.3 Locus of Control ................................................................................................ 83
4.8.2.4 Pessimism .......................................................................................................... 83
5
Which Personality Types have the Highest Marginal Utilities of Income? ........................... 85
5.1
Abstract .......................................................................................................................... 85
5.2
Introduction .................................................................................................................... 86
5.3
Methodology .................................................................................................................. 90
5.4
Data ................................................................................................................................ 92
5.5
Results ............................................................................................................................ 95
5.5.1
Robustness Tests .................................................................................................. 100
5.6
Conclusion ................................................................................................................... 103
5.7
Appendix ...................................................................................................................... 106
5.7.1
Note to Tables ...................................................................................................... 106
5.7.2
Personality Variables in GSOEP ......................................................................... 106
5.7.2.1 Big Five Personality Inventory ........................................................................ 107
5.7.2.2 Individual Autonomy ....................................................................................... 107
5.7.2.3 Pessimism ........................................................................................................ 108
5.7.2.4 The Construction of Personality Measures ...................................................... 108
6
The Dark Side of Conscientiousness: Conscientious People Suffer more from
Unemployment ............................................................................................................................. 110
6.1
Abstract ........................................................................................................................ 110
6.2
Introduction .................................................................................................................. 111
6.3
Method ......................................................................................................................... 113
6.3.1
Participants and Procedure ................................................................................... 113
6.3.2
Measures .............................................................................................................. 114
6.4
Results .......................................................................................................................... 115
6.5
Discussion .................................................................................................................... 118
7
Do People Become Healthier after Being Promoted?.......................................................... 121
7.1
Abstract ........................................................................................................................ 121
7.2
Introduction .................................................................................................................. 122
7.3
Earlier Work................................................................................................................. 122
7.4
Methodology ................................................................................................................ 125
7.5
Data and Estimation Issues .......................................................................................... 126
7.6
Results .......................................................................................................................... 129
7.7
Objections and Counter Arguments ............................................................................. 136
7.7.1
Issue #1: Noise ..................................................................................................... 137
7.7.2
Issue #2: Endogeneity .......................................................................................... 137
iv
7.7.3
Issue #3: Poor Health as a Predictor .................................................................... 138
7.7.4
Issue #4: Sample Changes ................................................................................... 139
7.8
Conclusion ................................................................................................................... 141
7.9
Appendix ...................................................................................................................... 142
7.9.1
Notes to Tables .................................................................................................... 142
7.9.2
Sample Construction ............................................................................................ 143
7.9.2.1 Control Groups................................................................................................. 143
7.9.2.2 Treatment Groups ............................................................................................ 143
7.9.3
Definition of GHQ Mental Ill-health ................................................................... 144
8
Conclusion ........................................................................................................................... 145
8.1
Summary ...................................................................................................................... 145
8.2
Implications for Economic-Psychology Subjective Well-Being Research ................. 148
8.2.1
The Use of Large Data Sets in Psychology ......................................................... 148
8.2.2
Improved Understanding of Social Comparisons ................................................ 149
8.2.3
Rank Based Comparisons .................................................................................... 151
8.2.4
Subjective Well-Being Research and Policy ....................................................... 151
8.2.5
The Link between Health and Occupational Status ............................................. 152
8.2.6
Personality within Economics.............................................................................. 153
8.3
Conclusion ................................................................................................................... 154
9
References ............................................................................................................................ 155
v
LIST OF TABLES AND FIGURES
Table 2.1: Pooled OLS regression on life satisfaction comparing logarithm of absolute income
and income rank by sample.................................................................................................... 32
Table 2.2: Pooled OLS regressions on life satisfaction comparing logarithm of mean income and
income rank using various reference groups ......................................................................... 34
Table 4.1: Summary statistics across the 6 year panel used in analysis and a longer 12 year panel
(N = 93016/135486) – non-standardised ............................................................................... 66
Table 4.2: Fixed effect, REMT and pooled OLS life satisfaction regressions .............................. 68
Table 4.3: Predicting the fixed effects residual (from column 2 of Table 4.2) using the mean
levels of various objective characteristics and personality variables ..................................... 70
Table 4.4: Correlations between observable characteristics and the unobservable component of
the fixed effect residual errors ............................................................................................... 73
Table 4.5: Introducing personality into life satisfaction regressions using the fixed effect vector
decomposition technique (3rd stage) and the random effects model ..................................... 74
Table 5.1: Summary statistics (N = 93256) – non-standardized .................................................... 94
Table 5.2: Fixed effect and pooled OLS life satisfaction regressions ........................................... 96
Table 5.3: Fixed effects and pooled OLS analysis of income interactions with personality ......... 99
Table 5.4: Robustness of the personality-income interactions .................................................... 101
Table 6.1: Two hierarchical regression analyses predicting the life satisfaction of individuals in
the years following unemployment ...................................................................................... 116
Table 7.1: Pearson correlation coefficients for the three ill-health measures .............................. 129
Table 7.2: Ill-health over time within the whole sample ............................................................. 129
Table 7.3: Cross-section regression equations for subjective ill-health, visits to the doctor, and
mental strain ......................................................................................................................... 130
Table 7.4: Ill-health among the non-promoted non-supervisors and those promoted to manager
(at time T) ............................................................................................................................ 132
Table 7.5: Ill-health among the non-promoted and those promoted to any category (at time T) 135
Table 7.6: Difference-in-Difference ((T+3)-(T-1)) estimates (with controls) for individuals
working in the public sector and in the manufacturing industry, those individuals who stay
at the same address across all 5 years and those who stay in the promoted position up until
T+5 ....................................................................................................................................... 136
Table 7.7: Probit equations using health at T-1 as a predictor of promotion............................... 138
Table 7.8: Regressions showing health differences across promoted groups, and those who
subsequently left the workforce or changed role ................................................................. 140
Figure 6.1: The life satisfaction change following unemployment as moderated by
conscientiousness ................................................................................................................. 118
vi
ACKNOWLEDGMENTS
This thesis would not have been possible without the help, advice and encouragement
from so many people. I have been extremely fortunate.
First, I would like to thank Gordon Brown and Andrew Oswald for absolute first rate
supervision. Both have always made themselves available to give me great guidance in my
academic development. I have taken great pleasure from our many meetings and both have never
failed to inspire me. I am also in great debt to Alex Wood who has been pivotal in my
development as a researcher and who helped me develop so many of the ideas in this thesis.
I would also like to thank the department of psychology at the University of Warwick for
making the transition from economics to psychology surprisingly easy and providing an excellent
academic climate in which to pursue my research. Everyone in my office has been hugely
supportive and I’d like to thank them for lifting my spirits on many a grey day.
I am extremely grateful to Roxanne Rees-Channer whom without I would never have
thought it possible for me to do a PhD. Thank you for helping me to believe in myself. There are
also many close friends who have shared with me both the good and bad times during the last 3
years. Thank you for listening to my moans. I’d also like to thank everyone I’ve ever had a
conversation with about my “happiness” research and for helping me to stay passionate about
what I do.
Finally, I’d like to thank my family for their love.
vii
DECLARATION
The research reported in this thesis is my own work unless otherwise stated. No part of
this thesis has been submitted for a degree at another institution.
Chapter 2 was written in collaboration with Gordon Brown and Simon Moore. Chapters 3
and 5 were written in collaboration with Alex Wood. Chapter 6 was written in collaboration with
Alex Wood and Gordon Brown and Chapter 7 was written in collaboration with Andrew Oswald.
Chris Boyce
viii
NOTE ON INCLUSION OF PUBLISHED WORK
Certain chapters have been previously published during the period of the PhD registration
Copyright of these papers resides with the publishers, but under the terms of the copyright
agreements these papers are reproduced as chapters in this thesis. These papers are as follows.
Chapter 2: Boyce, C. J., Brown, G. D. A., Moore, S. C. (in press). Money and happiness: Rank
of income, not income, affects life satisfaction. Psychological Science.
Chapter 4: Boyce, C. J. (in press). Understanding fixed effects in human well-being. Journal of
Economic Psychology.
Chapters 3, 5 and 7 are currently under review.
ix
LIST OF ABBREVIATIONS
ALT: Adaptation Level Theory
BHPS: British Household Panel Survey
CBT: Cognitive Behaviour Therapy
FE: Fixed Effects
FEVD: Fixed Effect Vector Decomposition
GHQ: General Health Questionnaire
GP: General Practitioner
GSOEP: German Socio-Economic Panel
OLS: Ordinary Least Squares
R: Income Rank
RE: Random Effects
REMT: Random Effects with a Mundlak (1978) Transformation
RFT: Range-Frequency Theory
SES: Socio-Economic Status
SR: Subjective Income Rank
SWB: Subjective Well-Being
x
ABSTRACT
This thesis uses subjective well-being data to understand the impact that an individual’s
economic circumstances have on their well-being. Chapters 2, 3, 4 and 5 look specifically at the
role of income on well-being; whilst Chapters 6 and 7 focus on the effect of employment status.
This thesis draws heavily on psychological concepts and ideas; highlighting that an
interdisciplinary approach to subjective well-being data can have substantial benefits to the study
of well-being.
Chapter 2 seeks to understand how people compare their incomes with one another.
Relative judgment models from psychology are explored and the evidence suggests that
individuals may be concerned with their rank position rather than their absolute position or how
they compare relative to a mean level. Applying this idea to relative income studies it is shown
that an individual’s rank income provides a better explanation of life satisfaction than either
absolute income or their income relative to the mean income of those around them.
Chapter 3 highlights that although more money may reduce psychological distress it is a
relatively inefficient way to do so. This chapter provides medical evidence to suggest that
psychological therapy is a more efficient way to reduce psychological distress. Income growth
does not appear to increase national well-being in developed countries so this chapter suggests
that increasing access to mental health care could be a better way to raise national well-being.
Personality, although appropriately controlled for, is mostly ignored by economists
researching subjective well-being data. Chapters 4, 5 and 6 therefore explore the use of
personality measures in economic subjective well-being research. Chapter 4 proposes a new
methodological technique that incorporates personality measures. Chapters 5 and 6 then show
that personality interacts with important economic variables. These chapters show that
personality is an important aspect to be understood by economists.
Chapter 7 demonstrates the importance of using longitudinal data to understand causal
effects on well-being. Improvements to occupational status have been argued to lead directly to
improvements to health. This argument has been based solely on the cross-sectional association
that individuals with high occupational status tend to have better health. Chapter 7 shows that
improvements to occupational status actually tend to increase mental strain.
Taken altogether these studies suggest that subjective well-being data provides a useful
arena in which interdisciplinary research can be conducted.
1
CHAPTER 1
1
SUBJECTIVE WELL-BEING RESEARCH: AN OVERVIEW
This thesis takes a cross disciplinary approach to understanding human happiness. How
can we improve our lives? How can we be healthier, happier and more satisfied? Such questions
seem fundamental to the human condition and have been continually discussed and debated
throughout human history. Ancient Greek philosophers, for example, considered eudaimonia –
directly translated as “happiness”, but a more accurate meaning would be “human flourishing” –
to be the most important of human goals. In the United States’ Declaration of Independence of
1776, the pursuit of happiness, alongside life and liberty, is touted as one of the unalienable
rights of its citizens. In the present day some researchers have suggested that well-being indices
should be favoured over commonly used economic indices like Gross National Product (Diener,
2000; Easterlin, 1974; Kahneman, Krueger, Schkade, Schwarz, & Stone, 2004; Marks & Shah,
2005). Although the importance of and concern for happiness within an individual’s life is
unquestionable, what do we really know about what makes an individual happy?
When compiling research in the 1960s on the characteristics of the happy individual,
Wilson (1967) remarked that their had been relatively few advances in the theory of the happy
life since the ancient Greeks. Whether this statement was an accurate reflection of the times or
not, the research into what makes an individual happy has since thrived. Researchers now
routinely use self reported measures of well-being to understand human happiness. Such
subjective well-being research, however, includes more than just the study of the transitory
emotion of happiness. The term reflects both the cognitive and affective evaluations an
individual has about their existence, including aspects of an individual’s physical and mental
health. Such evaluations have been shown to be distinctly separate (Lucas, Diener, & Suh, 1996),
2
having their own set of correlates, which can be usefully analysed. The use of subjective wellbeing data, therefore, allows researchers to empirically test many of their well-being hypotheses.
It would perhaps be fair to now say that there have been considerable advances in the study of
human well-being since the ancient Greeks (Diener, Suh, Lucas, & Smith, 1999).
This thesis attempts to add to our knowledge of what improves human well-being. To do
this the thesis focuses entirely on an individual’s economic circumstances, for example their
income and employment status. Specifically this thesis considers the importance of income rank
for individual well-being, suggests an alternative way of increasing the well-being of our nations
besides economic growth, shows that personality measures can help us understand how economic
circumstances impact on well-being, and questions whether improvements to occupational status
will bring health benefits. The topics covered in this thesis, although wide-ranging, share one
common theme: They all draw heavily on concepts and ideas from psychology to answer
important economic questions.
1.1
The Development of Subjective Well-Being Research in Economics and Psychology
Subjective well-being research forms part of what is often referred to as the positive
psychology movement (Seligman & Csikszentmihalyi, 2000). Positive psychology is a focus on
the positive elements of the human experience. The movement is concerned with human
flourishing and akin to the Greek’s intended meaning of the word eudaimonia. More simply the
movement attempts to understand how individuals can live their lives in more fulfilling and
satisfying ways. This concern for understanding the positive aspects of the human experience
arose in reaction to a discipline that, post-World War 2, had largely become pre-occupied with
the negative. Psychologists had a relatively good understanding of mental illness and its
treatment, but could offer very little in the way of help to individuals who, although not
3
necessarily mentally ill, could neither be defined as mentally healthy. Although early humanistic
psychologists, such as Maslow (1954) and Rogers (1959), saw mental health as a continuum and
developed theories on how to improve individual functioning and mental health, it is only in
recent years that psychology has fully embraced the study of the positive. Subjective well-being
research forms an important part of this study of the positive, seeking to understand how
individuals can shine during relatively benign conditions (Seligman & Csikszentmihalyi, 2000).
There has been a great deal of progress in subjective well-being research in recent
decades, as the extensive reviews carried out by Wilson (1967), Diener (1984) and Diener et al.
(1999) show. Initially the concern was with determining the individual characteristics that were
most strongly correlated with high well-being. Wilson (1967) found – using only bivariate
associations – that happy individuals tended to be young, healthy, well-educated, well-paid,
extroverted, optimistic, worry-free, religious, married with high self esteem, have high job
morale and modest aspirations. Many of these early conclusions, however, which focused heavily
on demographic factors, have since been overturned and psychologists have increasingly focused
on psychological factors, such as personality, adaptation, goal striving and coping strategies.
More up to date evidence suggests that individuals with high subjective well-being are those that
have a positive temperament, do not ruminate excessively over bad events, live in economically
developed societies, have strong social relationships and possess adequate resources to progress
towards their goals (Diener et al., 1999).
1.1.1 The Use of Subjective Well-Being Data as a Proxy for Utility
Psychologists have naturally dominated the development of subjective well-being
research but in the last decade or so economists have begun to show considerable interest in the
area. Economists have recognised that subjective well-being research has the potential to unlock
4
answers to questions that have previously been thought of as unanswerable by acting as a proxy
for an individual’s utility. The concept of utility – the satisfaction derived from consuming goods
– permeates modern economic thought. Essentially all economic behaviour is assumed to reflect
an attempt to increase one’s utility and therefore any observation of an individual’s behaviour
should reveal something about their preferences. This revealed preference approach, pioneered
by economists such as Samuleson (1938), is central to modern economic thought and has been
for most of the twentieth century. If an individual chooses one bundle of goods over another then
theory states that it is simply because it is preferred and therefore must yield more utility.
Although an individual’s decision is based on the principle of utility maximisation this does not
necessarily mean the maximum utility will actually be realised. For this to realistically occur,
individuals would need to be perfectly rational, have perfect recall and foresight, and have access
to all information.
Researchers have pointed out that there is often a wide gap between an individual’s
“decision utility” and their “experienced utility”, and that as a result individuals may conceivably
make sub-optimal choices (Kahneman, Fredrickson, Schreiber, & Redelmeier, 1993; Kahneman,
Wakker, & Sarin, 1997). It has now been well documented, mostly by psychologists, that
individuals exhibit many cognitive biases and consistently make bad, or seemingly illogical,
economic decisions. For instance, it is well known that individuals favour the avoidance of losses
over the acquisition of gains and that they subjectively weight probabilities non-linearly
(Kahneman & Tversky, 1979). There are strong endowment effects, whereby individuals tend to
place higher values on the objects that they own than ones they do not (Kahneman, Knetsch, &
Thaler, 1990; Thaler, 1980), and an economic decision can be heavily dependent on the way it is
framed (Tversky & Kahneman, 1981). Some researchers argue that individuals have neither the
5
access to all the necessary information nor the time to make truly optimal decisions and therefore
rely mainly on heuristics – short cut answers (Gigerenzer & Selton, 2001; Gigerenzer & Todd,
1999).
Nevertheless, in order to understand an individual’s preferences and what brings the
greatest utility, economists have generally analysed an individual’s economic decisions.
However, utility, as it was originally conceptualised, was more akin to “experienced utility”. The
concept of utility was initially developed by utilitarianists, such as Bentham (1748-1832) and
Mill (1806-1873), who suggested using the term utility to represent the tendency for an object or
action to increase or decrease overall happiness (Read, 2007). Their proposition was that
individuals ought to desire the things that brought them the most utility. They suggested that, in
theory at least, it would be possible to obtain the total utility of some action by summing up the
total amount of pleasure that an action brought to an individual. They further argued for the
maximisation of social utility and that the morally right action was the one that produced the
greatest amount of pleasure for the greatest number of people. Although their theory of utility
suggested that individuals desired the greatest amount of utility, at no point did their formulation
imply that individuals would necessarily choose the option that would yield the most utility.
The inability to measure utility served as a major limitation to the utilitarianists’ idea of
utility. Utility as it was originally conceptualised fell afoul of economists, such as Vilfredo
Pareto (1848-1923) and Lionel Robbins (1898-1984) (Vaggi & Groenewegen, 2003), and as
discussed in Read (2007), the original conception of utility was eventually abandoned.
Economists soon came to favour the observation of an individual’s behaviour and decisions. At
the time economists had not figured out a way to measure utility, and hence actual experienced
utility was unobservable. The argument was that behaviour was observable and that decision
6
utility would in any case make a good approximation to experienced utility. The revealed
preference approach, which assumed that rational individuals chose the option according to a
stable set of preferences, soon became dominant. In this model utility became of no intrinsic
relation to happiness and could not be compared or aggregated across, or even within,
individuals. Utility simply became an abstraction that could be used to state that if an individual
had chosen x over y then x must necessarily have a higher utility than y. However, with the rise
in subjective well-being research the measurement of utility in the way utilitarianists originally
suggested finally seemed a possibility.
1.1.2 Subjective Well-Being – A Viable Tool for Economic Analysis
The use of subjective well-being data rests on the assumption that such data are a suitable
proxy for an individual’s utility. Researchers at Leyden University, such as Van Praag (1971)
and Kapteyn (1994), were the first to suggest the use of subjective data to measure welfare in
economics. The Leyden approach, as it became known, initially focused solely on economic
welfare and set out to determine what individuals considered to be, for example, very bad, bad,
sufficient, good or very good incomes. The values assigned were found to vary considerably
across individuals and were dependent on a number of economic and non-economic factors (Van
Praag & Frijiters, 1999). An interest in economic welfare soon developed into a concern for
general welfare and well-being. At the time the Leyden approach was particuarly novel and as
such came against substantial criticism from economists who generally consider utility as
immeasurable (e.g. Seidl, 1994).
Even in the present day economists are unconvinced of the use of subjective data to
answer economic questions or inform policy. For example, Johns and Ormerod (2007) comment
that national well-being levels, although they may not have risen in line with GNP, have also not
7
appeared to have risen in line with other aspects that one would firmly expect to contribute to
well-being, such as rising longevity, greater gender equality, and improved democratic
institutions. They take this as evidence that subjective well-being data is an insensitive measure
of welfare. Will Wilkinson (2007) argues that current well-being measures, due to very simple
and sometimes inconsistent questioning, are blunt instruments that cannot be realistically
interpreted or aggregated across individuals. Researchers have also warned of increased
paternalism (Johns & Ormerod, 2007; W. Wilkinson, 2007), believing that the individual is
always in the best position to make decisions that will promote their own well-being. However,
many of the findings from the use of subjective well-being data suggest that free-market models
might not always be the best way of promoting overall happiness (Layard, 2006a). In economics
subjective well-being research is still relatively young and many of the findings challenge
economic thought of the last 200 years. Although some scepticism within economics is only
natural the current findings are somewhat compelling and need considerable reflection.
There is much less scepticism in psychology where subjective well-being data has been in
use for much longer. Psychologists have spent considerable time ensuring that the self-reported
measures they use have both reliability and validity. That is, measures of well-being yield
consistent results across repeated tests and they truly represent the intended underlying construct
(Fordyce, 1988; Lepper, 1998; Lucas et al., 1996; Pavot & Diener, 1993a; Sandvik, Diener, &
Seidlitz, 1993). Further, there is strong evidence that self reported measures predict observable
behaviour. For instance, self report measures have been shown to be related to how often an
individual genuinely smiles – the Duchenne smile (Ekman, Friesen, & Davidson, 1990), the
length of an individual’s life (Palmore, 1969) and coronary heart disease (Sales & House, 1971).
Individuals that report themselves to be satisfied with life are much less likely to commit suicide
8
(Koivumaa-Honkanen et al., 2001) and positive well-being has been shown to directly relate to
health-relevant biological processes. It has additionally been shown that happy countries have
less hypertension (Blanchflower & Oswald, 2008).
Ultimately subjective responses from individuals about how happy or satisfied they are
with their lives cannot simply be ignored. Self report measures are of course far from perfect. For
example, there is still some concern over whether scores can be treated as cardinal or ordinal
(Ferrer-i-Carbonell & Frijters, 2004) and, due to it being unlikely that scores will be comparable
across individuals, self report measures cannot yet be aggregated across individuals in the way
originally suggested by utilitarianists (Read, 2007). Nevertheless, the use of subjective wellbeing data to proxy for utility has helped researchers understand the contribution that economic
circumstances have to an individual’s well-being. Economic subjective well-being research, or
“happiness economics”, as it has popularly and perhaps misleadingly become known, is a rapidly
growing area of research. The use of subjective well-being data in economics is seen by some
economists as a large step forward and its use is gradually, and rightfully, being acknowledged as
a useful counterpart to the revealed preference approach (Frey and Stutzer, 2002).
1.2
Overview of Key Research Areas
Subjective well-being is currently researched by economists and psychologists with equal
rigour. In spite of this there are often huge gaps in the way that economists and psychologists
approach the topic. In part, this arises due to fundamentally different research questions.
Additionally, however, subjective well-being researchers are not always aware of the progress
made outside of their immediate field. Much of the work in this thesis centres on the
relationships between income and well-being and an individual’s employment status and wellbeing. These relationships have received extensive cross-disciplinary coverage, as the next
9
section demonstrates, but there are key differences across the disciplines and therefore important
gaps. This thesis attempts to highlight this disciplinary divide and show how these gaps could be
bridged to benefit both disciplines and ultimately our understanding of human happiness.
1.2.1 Income and Well-Being
The importance of income for well-being is an area in which there has been a healthy
amount of interest from both psychologists and economists. This section first discusses the
evidence for a relationship between income and well-being. It then discusses possible
explanations and offers ways in which these explanations can be explored and refined.
1.2.1.1 Evidence of a Relationship between Income and Well-Being
There are several ways in which to determine whether income is positively related to
well-being – researchers can identify the correlation between income and well-being either; over
time, within a country or across countries. The evidence for each is discussed below.
1.2.1.1.1 Income and Well-Being over Time
One of the earliest and most influential studies on income and well-being came from the
economist, Richard Easterlin in 1974. Many years ahead of the explosion of subjective wellbeing research in economics, Easterlin (1974) questioned whether economic growth had
improved the human lot. Instead of using productivity data or standard of living indexes that
would give an objective answer to his question, Easterlin used the proportion of randomly
sampled individuals that stated they were “very happy” with their lives. He was able to
demonstrate using this subjective data that, in developed countries at least, economic growth had
not seemed to improve the human lot. It appeared that once countries achieved a certain level of
economic development, further development was associated with very little increases in average
10
national well-being. Although there are some that are sceptical of this finding (C. S. Fischer,
2008; Hagerty & Veenhoven, 2003; Stevenson & Wolfers, 2008) it has been replicated on
numerous occasions, by both economists and psychologists (Blanchflower & Oswald, 2004;
Diener & Oishi, 2000; Easterlin, 1995; Kenny, 1999). Such a finding could invite one to question
why economic growth remains a priority in developed countries.
1.2.1.1.2 Income and Well-Being within a Country
But does money buy happiness at the individual level? If we only observed people’s
behaviour then our conclusions would be that it surely must. Utility maximising individuals
choose to spend money on goods and services; implying that money, by allowing them to
increase the goods and services that they can buy, ought to bring them more utility. Early
research into the relationship between income and well-being within a country was mostly
carried out by psychologists (Diener, 1984; Diener & Biswas-Diener, 2002; Haring, Stock, &
Okun, 1984; Myers & Diener, 1995; W. Wilson, 1967) and it has consistently been found that
within a country individuals with higher incomes tended to also have higher well-being.
Economists have further shown using longitudinal data that this relationship may be truly causal
(Frijters, Haisken-DeNew, & Shields, 2004; Gardner & Oswald, 2007).
1.2.1.1.3 Income and Well-Being across Countries
Easterlin (1974) also asked whether richer countries were happier countries. The evidence
he presented suggested that richer countries were not happier. However, there were a number of
issues with his original research as pointed out in Veenhoven and Hagerty (2003). Easterlin has
since updated his work (Easterlin, 1995) and, as has been replicated on numerous occasions,
there is in fact a positive relationship between a countries income and their well-being (e.g.
11
Diener, Sandvik, Seidlitz, & Diener, 1993; Hagerty & Veenhoven, 2003). However, this positive
relationship appears to be concave and beyond $10,000, the average income level appears to
have very little effect on average well-being (Frey & Stutzer, 2002). The relationship between
income and well-being across countries is not entirely reliable. Not only are there cultural
limitations to making subjective well-being comparisons across countries but further it is likely
that there are other factors that accompany high income per capita, such as democracy (Inglehart,
Foa, Peterson, & Welzel, 2008), health or basic human rights (Frey & Stutzer, 2002), that may
have resulted in the higher happiness levels.
1.2.1.2 Explaining the Income and Well-Being Data – Relative Income Effects
The combination of a correlation between income and well-being within a country and
the fact that economic growth does not appear to increase national well-being leaves something
of a puzzle to economists researching subjective well-being (Clark, Frijters, & Shields, 2008).
How can increases to income seemingly improve an individual’s utility but not total societal
utility? These two findings are commonly referred to as the Easterlin paradox and form the
cornerstone of much economic research into well-being. One of the most popular explanations
for the Easterlin paradox is that individuals are not concerned with absolute income but with
income relative to their peers. Income will improve an individual’s well-being but only if it rises
at a faster rate than the income of others. Hence, if everyone’s income increased by the same
amount then no one would actually be any better off because relative positions remain
unchanged.
12
1.2.1.2.1 Relative Income Effects in Economics
There has been some debate since using simple cross-sectional data by psychologists over
whether the effect of income is relative or absolute (Diener et al., 1993; Hagerty & Veenhoven,
2003; Veenhoven, 1991). Such evidence, however, is based solely on cross-sectional
observations and can only be taken as circumstantial. Simple correlations are not enough to
conclusively show that individuals care about relative income. Rather than relying on such
circumstantial evidence economists have attempted to show that an individual’s relative income
is an important predictor of well-being. For example, relative income variables have been shown
to significantly predict various measures of well-being (Clark & Oswald, 1996; McBride, 2001).
Additionally economists have also shown that there is causal link – with increases to an
individual’s relative income leading to increases in well-being (e.g. Ferrer-i-Carbonell, 2005;
Luttmer, 2005; Senik, 2004). It now seems fairly clear that individual’s care about their income
compared to those around them and will feel less satisfied in the presence of higher earning
others.
The presence of relative income effects should come as no surprise to economists. The
importance of relative concerns and social comparisons has received discussion throughout the
history of economics and can be traced back to the works of Adam Smith (1976) and Karl Marx
(1952). Veblen (1899) coined the phrase “conspicuous consumption” to refer to a type of
consumption that, although not necessary to our survival, signalled to everyone else something
about one’s standing or status in the community. Duesenberry (1949) discussed relative
concerns with respect to savings. He suggested that in order to maintain self esteem individuals
were likely to sacrifice their savings to consume goods that other people have. This “keeping up
with the Joneses” effect, or “luxury fever” as Frank (1999) terms it, may mean that collectively
13
individuals may be working too hard and buying things that they don’t really need. However, it
has always been difficult to determine whether behaviour is motivated by relative, rather than
absolute, effects. Neumark and Postlewaite (1998) demonstrate that the rise in women entering
the workforce can, in part, be explained by relative concerns by comparing the employment
decisions of sisters and sister-in-laws. However, as Luttmer (2005) points out, to really
differentiate between relative and absolute income effects a proxy for utility in the form of
subjective well-being data is essential. It is important that the concern for relative performance is
appropriately included in an individual’s utility function (Clark & Oswald, 1998).
1.2.1.2.2 Relative Judgment Models in Psychology
Psychologists have carried out extensive research to suggest that individual’s make
relative judgments and have therefore been ready to accept that an individual’s utility is
influenced by relative income. They have, for example, demonstrated under experimental
conditions that judgments are often based on some type of relative concern (Helson, 1964;
Parducci, 1965; Stewart, Brown, & Chater, 2005). Helson (1964) proposed Adaptation Level
Theory (ALT) to model how individuals subjectively assessed an objective stimulus within the
context of a set of other stimuli. For example, a typical experiment scenario might require an
individual to assess how subjectively heavy a weight feels in comparison to other weights, where
1 = “very light” and 10 = “very heavy”. The model is simple and proposes that an assessment is
made by intuitively making a comparison with a weighted mean of the background stimuli; in the
example above, an average weight in the context of the other weights might be given a 5 or a 6.
If the set of stimuli and mean were to increase so should the comparison level to which an
assessment would be made to. The previously average weight would perhaps now be given a 3 or
4. ALT predicts that the individual’s evaluation of a given stimulus will therefore adapt to the
14
context of comparison. This approach is very similar to that of economists researching relative
income effects. It is assumed that individuals compare with the average level of income of those
around them.
However within psychology the data better support an alternative model of how
individuals make subjective assessments: Range-Frequency Theory (RFT) (e.g. Parducci, 1965,
1995). RFT suggests that an assessment is given by a weighting of the stimulus’ rank (frequency)
and cardinal position relative to the highest and lowest values (range) within the set of stimuli.
One issue with ALT is that two differently distributed sets of stimuli can have identical means.
The distribution is not considered. An assessment under RFT, however, is modelled on a
uniformly distributed rank but anchored by distribution extremes. RFT has been useful in
modelling subjective assessments in an array of stimuli but importantly for economics it has been
found to help model assessments of both prices (Niedrich, Sharma, & Wedell, 2001; Qian &
Brown, 2007) and incomes (Brown, Gardner, Oswald, & Qian, 2008; Mellers, 1986). RFT has
also been shown to be applicable to social comparisons (R. H. Smith, Diener, & Wedell, 1989).
Outside of experimental conditions the range and skew of an income distribution, as predicted by
RFT, affect the average happiness level across communities (Hagerty, 2000).
1.2.1.2.3 Rank Income Effects
The evidence from psychology suggests that the relative income models used in
economics can be improved and provides the motivation for Chapter 2. This chapter uses
subjective well-being data to explore two relative income models – the reference income model,
where individuals compare to the average income of those around them, and a rank income
model, which suggests that it is the rank of their income within the comparison set that is
important. The reference income model is dominant within economists’ relative income studies
15
so we test which model has the greatest explanation of life satisfaction. Our evidence, consistent
with RFT, favours the rank income hypothesis. We also further consider that comparison is
mostly made with those that are better than oneself. There is already some evidence of upward
comparisons using the reference income approach (Blanchflower & Oswald, 2004; Ferrer-iCarbonell, 2005). However, we investigate upward comparison using the rank income
framework and we show that individuals compare up to twice as much with those above than
those below.
1.2.1.3 Explaining the Income and Well-Being Data – Income is relatively
Unimportant for Well-Being
A further, and perhaps less popular explanation, is that more income might simply not be
very important for either national or individual wellbeing. The perceived importance of income
for well-being depends heavily on how we choose to interpret statistical associations. Generally
economists and psychologists interpret the effects of income on well-being very differently. A
recent review, for example, puts the correlation between individual income and well-being within
countries at between 0.17 and 0.21 (Lucas & Dyrenforth, 2006). The correlation between income
and well-being varies considerably across countries and can often be much larger in developing
countries (Howell & Howell, 2008). However, the correlation can also be much smaller, for
example, Diener, Sandvik, Seidlitz & Diener (1993) obtained a correlation of 0.12 in the United
States. Income and well-being correlations are nearly always strongly significant, owing to the
large sample sizes, but correlations of this size might typically be referred to by psychologists as
small (Cohen, 1992). These correlations suggest that, at best, income explains 4% of an
individual’s well-being. A psychologist might therefore argue that this small correlation makes
16
income and well-being largely of little interest, particularly when other factors, such as
personality, seem to explain much more of an individual’s well-being.
Economists, on the other hand, do not use correlations to view the effect of income on
well-being. Simple correlation coefficients are much more useful for interpreting the nonmeaningful scales commonly used within psychology. Income is an objective characteristic that
can be meaningfully interpreted, enabling economists to make statements, such as, “£10,000 is
associated with a well-being rise of X” or “the well-being increase from some life event is
equivalent to having an income rise of X” (e.g. Blanchflower & Oswald, 2004; Di Tella,
MacCulloch, & Oswald, 2003; Ferrer-i-Carbonell, 2005). Some researchers have therefore
argued that simple correlation coefficients applied to income and well-being studies can actually
be misleading. Lucas and Schimmack (2009), in an attempt to convince psychologists sceptical
of income effects on well-being, demonstrate that small correlations can be translated into large
mean life satisfaction differences between the rich and the poor. For example, they show that the
life satisfaction for those earning over $200,000 a year is between 0.79 and 0.88 standard
deviations higher than those earning less than $10,000. This observation could lead one, in spite
of the low correlation in their data set of between 0.17 and 0.20, to conclude that income is in
fact important for well-being.
However, the figures provided by Lucas and Schimmack (2009) must be interpreted with
some caution. Firstly, correlation does not imply causality. Their analysis is based on simple
bivariate associations and the association between income and well-being is mostly driven by
third variables, such as an individual’s personality (see section 1.2.1.4). Secondly, one must take
into consideration the likelihood of an individual’s income increasing from below $10,000 to
$200,000. Unfortunately Lucas and Schimmack (2009) do not present the standard deviation of
17
household income but it is likely to be around $30,000. This suggests that to go from $10,000 to
$200,000 would require something like a six standard deviation increase in income. When we
consider that six standard deviations (3 standard deviations either side of the mean) should
contain around 99.7% of the observations in a normally distributed sample, we begin to see how
unrealistic a change in income of such a magnitude is likely to be. The correlation coefficient is
designed to describe the relationship between two variables in a standardised way. The
correlation coefficient reflects how much one variable will move as a result of a one standard
deviation change in the other. Hence, an alternative, and perhaps fairer interpretation, of Lucas
and Schimmack (2009) is that an individual who had an income one standard deviation higher
(around $30,000) than another individual would be approximately 0.17 to 0.2 standard deviations
higher in life satisfaction. If causal effects were considered then typically the association between
income and well-being shrinks by about a third (Ferrer-i-Carbonell & Frijters, 2004), suggesting
$30,000 would increase well-being by less than 0.1 standard deviations.
It is the intention of chapter 3 to illustrate how low the correlation between income and
well-being actually is relatively to other aspects of life. To do this, we bring together disjoint
areas of research from economics, psychology, law and medicine. This chapter centres its
discussion on the law courts and focuses on the use of monetary compensation as a means to help
individuals deal with traumatic life events. In recent work Oswald and Powdthavee (2008a,
2008b) use subjective well-being equations to suggest that current recommended compensation
payouts are too low. They suggest that to truly compensate someone financially then payouts
would actually need to be much higher. We argue that, although this would be a correct
interpretation of subjective well-being equations, the high values suggested actually reflect
money’s inefficiency at alleviating psychological distress. We make an alternative suggestion –
18
the use of psychological therapy. We then go on to illustrate, using medical evidence of
psychological therapy’s cost-effectiveness, that psychological therapy could be over 30 times
more cost effective than money at alleviating psychological distress.
We use this evidence to question the pursuit of income growth as a means to increase
well-being in developed countries (Blanchflower & Oswald, 2004; Diener & Oishi, 2000;
Easterlin, 1995; Kenny, 1999). If money is relatively unimportant to well-being then developed
societies might be better off pursuing objectives that are more likely to increase national wellbeing. Mental illness appears to be rising worldwide (Michaud, Murray, & Bloom, 2001) and we
suggest that there needs to be a greater focus on mental health and improved access to mental
health care, such as psychological therapy. In Chapter 3 we argue that good mental health is
undervalued in our societies and that as a result there could be enormous benefit if resources
were channelled into mental health care rather than solely focusing on economic growth.
1.2.1.4 Explaining the Income and Well-Being Data – Personality
A further explanation of the data, which shows that national income growth does not lead
to increases in well-being and that there is only a small correlation between an individual’s
income and their well-being, is that there is a third variable that causes both high income and
high well-being. For example, it is likely that aspects of an individual’s personality may drive
them to earn a higher income yet also mean they have higher levels of well-being over the course
of their life. Psychologists have shown that an individual’s well-being can mostly be explained
by either personality or genetic factors. For example, Lykken and Tellegen (1996) show that
between 44% and 52% of well-being is the result of individual differences. Once personality is
controlled for the effect of income on well-being decreases by up to a third (Ferrer-i-Carbonell &
Frijters, 2004). It has further been argued that demographic factors contribute substantially less to
19
well-being compared to other factors such as personality (Argyle, 1999). Psychologists have,
therefore, begun to focus less on demographic factors and are instead interested in understanding
the types of personality and the psychological processes that accompany high well-being (Diener
et al., 1999).
One theory of well-being that seeks to understand the role of personality and the
psychological processes is the set-point theory of subjective well-being. This is the idea that
individuals receive short term fluctuations in their well-being due to changes in their life
circumstances, but that given time, they revert back to a baseline level of well-being that is
dependent on an individual’s personality (Headey & Wearing, 1989). Lykken and Tellegen
(1996) estimate the stable component of well-being to be around 80%. Adaptation to traumatic
life events represents a practical application of set-point theory. An early demonstration of
adaptation came from Brickman, Coates and Janof-Bulman (1978), who seemed to show that
lottery winners were not significantly happier than a control group and that individuals with
spinal-cord injuries were not as unhappy as one might expect. However, these early results have
often been criticised (Lucas, 2007; Oswald & Powdthavee, 2008b) - not only are the results
based on cross-sectional differences using tiny samples of individuals, but the results have often
been misinterpreted to support complete adaptation.
Frederick and Loewenstein (1999) provide an in-depth discussion of adaptation and
suggest that whilst considerable adaptation seems to take place in some domains (e.g.
imprisonment, disability and income), it does not appear to in others (e.g. noise, cosmetic surgery
and food). Recent longitudinal evidence suggests that, although it can take a while, individuals
partially adapt to the loss of a loved one (Oswald & Powdthavee, 2008a). Researchers have also
observed that individuals appear to fully adapt to marriage, divorce, widowhood and the birth of
20
child (Clark, Diener, Georgellis, & Lucas, 2008) but not necessarily to unemployment (Clark,
Diener et al., 2008; Lucas, Clark, Georgellis, & Diener, 2004). The process of adaptation offers
an explanation as to why income growth may not provide as much of an increase to individual
and national well-being as one might expect. After the initial benefit of an income increase wears
off individuals drift back to a fixed long term level of well-being. Individuals get use to their new
wealthier environment (Easterlin, 2001, 2005). Much of the well-being work in psychology
attempts to understand what determines this important long term level of well-being.
1.2.1.4.1 Controlling for Personality in Economics
Economists would generally agree that personality explains a substantial component of
individual well-being (Ferrer-i-Carbonell & Frijters, 2004). However, in economics personality is
only of limited interest. Economists are much more interested in areas of life that can be changed
and influenced by policy, hence personality, due to its apparent non-changing nature (Costa &
McCrae, 1980, 1988), seems of little direct intrinsic interest. Economists are very aware of the
need to control for non-changing characteristics when trying to establish causal effects on wellbeing. Their main concern is, therefore, with how to control for personality simply and
effectively. Although economists have brought advanced statistical techniques to subjective wellbeing research to indirectly control for personality this seems to have precluded an awareness of
the huge literature on the development of personality measures. Psychologists routinely use
personality measures that have impressive levels of validity and reliability in subjective wellbeing studies and such measures could prove useful for economic research generally (Borghans,
Duckworth, Heckman, & ter Weel, 2008).
This thesis devotes several of its chapters to trying to illustrate that personality measures
could be hugely beneficial to economic research into subjective well-being. In turn the thesis also
21
attempts to demonstrate to psychologists that an understanding of demographic influences on
well-being can also be improved with the aid of personality (Gutierrez, Jimenez, Hernandez, &
Puente, 2005). Our work mainly relies on the German Socio-Economic Panel, which is a large
data set primarily used by and developed for economists. In a recent wave a number of
personality questions were included. Since psychologists do not routinely use these types of large
data sets it is therefore rare to find measures like personality included in them.
In Chapter 4 these personality measures are used to suggest an alternative to the dominant
statistical technique used by economists in subjective well-being research. Typically economists
use what is often referred to as a fixed effect model to control for unobservable personality
characteristics. Essentially the fixed effect model focuses on explaining the within-person
variation and controls for the aspects of an individual’s well-being that could be considered
fixed. This fixed component of well-being is similar to the long term level of well-being that was
addressed earlier and discussed in Lykken & Tellegen (1996). Chapter 4 first tries to determine
how much of the fixed effect can be explained by personality. The chapter then goes on to
suggest an alternative statistical technique to the standard fixed effect model that exploits
personality measures and then argues that the alternative technique produces more reliable casual
estimates on individual well-being for characteristics such as marital status, disability and
income.
1.2.1.4.2 Personality Interacts with Demographic Characteristics
Chapters 5 and 6 extend the argument for using personality measures in economic
research. These two chapters demonstrate that personality interacts with variables that
economists are typically more interested in, suggesting that personality cannot be simply
indirectly controlled for and largely ignored in the way that it has up to now.
22
Although part of this thesis demonstrates that income is relatively unimportant for wellbeing this does not mean the topic is of little interest. Economists will always naturally be more
interested in the effects of income on well-being simply because income is central to the
discipline. However, there are two further reasons that should make research into income and
well-being of interest to researchers across all disciplines. Firstly, income, unlike personality, can
be actively changed through policy and secondly, some understanding is needed as to why
income does not bring as much improvement to well-being as one might expect. Although the
data seem to suggest that income is relatively unimportant there remains a belief within our
societies that more money will bring substantially greater well-being. Individuals desire higher
incomes and the use of financial incentives, either through the tax system or through pay
bonuses, to change behaviour is endemic in many developed societies. This sets up the
motivation for Chapter 5. Most income and well-being studies only focus on average effects
across a large sample. We present evidence that the marginal utility of income varies greatly
according to an individual’s personality and suggest that policy could benefit from being
personality specific.
There has been some call for understanding the interaction between personality traits and
external circumstances (Diener et al., 1999) and the work in Chapter 5 answers this call. Chapter
6 offers a further demonstration of when personality can be important for the experience of a life
event. Chapter 6 explores the role of conscientiousness when becoming unemployed. The topic
of Chapter 6, however, will be discussed in relation to the employment status and well-being
literature, which is outlined in the next section.
23
1.2.2 Employment Status and Well-Being
Another area in which there has been a great deal of cross-disciplinary research with
respect to well-being concerns individuals’ employment status. In the standard economic utility
function work typically enters negatively, but there is substantial evidence to suggest that work
has many benefits for an individual’s well-being that goes beyond the simple acquisition of
income (Myers & Diener, 1995). The remainder of the thesis addresses two separate areas: the
psychological consequences from unemployment and understanding the positive association
between occupational status and physical and mental health. Researchers across the disciplines
have approached these areas in fairly similar ways. The largest difference appears to be mainly
methodological.
1.2.2.1 Unemployment
Typically individuals suffer deep psychological consequences from unemployment
(McKee-Ryan, Song, Wanberg, & Kinicki, 2005) and this goes beyond the simple loss of
income. Psychologists have generally been much more interested in the topic of unemployment
since, unlike income, it seems to have a substantial impact on an individual’s well-being. For
example, in their meta-analysis McKee-Ryan et al. (2005) review over one hundred, mainly
psychological, studies, and observe that there is a mental health difference between employed
and unemployed of 0.57 standard deviations. Although longitudinal studies are less common they
further observe that mental health drops on average by 0.38 standard deviations when an
individual becomes unemployed. Economists, although beginning their research much later, have
generally approached the topic of unemployment in a similar way (Frey & Stutzer, 2002).
However, economists have brought the use of large scale longitudinal data sets to subjective
well-being research and with the help of their advanced econometric techniques (see for example
24
work by Clark, Diener et al. (2008), Clark et al. (2001) and Winkelmann & Winkelmann (1998)),
they have advanced the understanding of the causal effects of unemployment on well-being.
One issue identified in unemployment studies, as with that of income studies, is that
estimates are generally based on the average effect across a sample. Researchers have established
specific circumstances under which unemployment is at its most psychologically damaging. For
example, the local unemployment rate (Clark, 2003; Powdthavee, 2007; Russell, 1987; Turner,
1995), the length of unemployment (Clark et al., 2001; Creed, Lehmann, & Hood, 2009;
Winefield & Tiggemann, 1989) and the degree of unemployment protection (Nordenmark,
Strandh, & Layte, 2006) all moderate the effect that unemployment has on well-being. However,
much less is known about which specific types of individuals suffer the most psychologically
following unemployment. This provides the motivation for Chapter 6, which presents evidence
that individuals with high levels of conscientiousness suffer the most during unemployment.
Typically conscientiousness has a positive relationship with well-being (DeNeve & Cooper,
1998; Hayes & Joseph, 2003) but our finding suggests that under certain circumstances this
relationship can be completely reversed.
1.2.2.2 Occupational Status
Individuals with high occupational status tend to have much better mental and physical
health (Marmot, Shipley, & Rose, 1984). The idea of status links in with the earlier discussion of
the importance of rank. Epidemiologists have been trying to understand the reasons for this
strong correlation and the theory that has dominated the literature is that there is a causal
relationship going from status to health (Marmot, 2004; R. Wilkinson, 2001). The argument
underlying their theory is convincing and suggests that individuals in low status positions have
low control over their life. A low control in life results in psychosocial stressors and it is these
25
stressors that can be detrimental to human health. It therefore follows that any improvement to
status will improve an individual’s level of control and psychosocial stressors will diminish. The
result should be a direct improvement to an individual’s health. Most of the evidence for a causal
link running from status to health, however, comes mainly from cross-sectional evidence.
The problem with relying solely on cross-sectional associations is that it is impossible to
determine the direction of causality. A cross-sectional association could be the result of one
variable causing a change in another or alternatively it could mean that the two variables move
together as a result of a change in a third variable. This means that whilst higher occupational
status may lead to health the cross-sectional association another explanation could be simply that
healthy individuals get promoted. Alternatively it could be that some other factor, such as
behavioural or genetic factors, causes movements in both. Generally economists are less satisfied
with simple cross-sectional associations and much of their subjective well-being research has
focused on establishing causal effects. Economists are more aware of the availability of large
longitudinal data sets that enable them to control for pre-existing levels of some variable.
Chapter 7 applies a longitudinal analysis to the question of whether occupational status
improves an individual’s health. As far as we are aware this is the first longitudinal analysis of its
kind. We illustrate that although individuals in high occupational positions have much better
health than those in lower occupational positions, a promotion between these positions does not
result in better health. We show that in some promotion categories mental strain may even
increase.
1.3
Overview of the Thesis
This thesis, although primarily concerned with economic decisions and how they
contribute to well-being, takes a cross disciplinary approach to the study of subjective well-being
26
data. This thesis uses concepts and ideas from psychology to enhance economic research into
subjective well-being. First, a rank income interpretation of well-being is presented (Chapter 2).
Second, the low importance of income for well-being is highlighted and it is further suggested
that developed societies could gain substantial well-being benefits by investing in mental health
care (Chapter 3). Third, the thesis demonstrates that personality measures can be of considerable
use to economists researching subjective well-being by showing that they allow alternative
statistical techniques to be used (Chapter 4) and that personality interacts with important
economic variables (Chapters 5 and 6). Finally, the thesis demonstrates the importance of
longitudinal data to draw causal conclusions by showing that, against common belief, improved
job status does not bring health benefits (Chapter 7).
27
CHAPTER 2
2
MONEY AND HAPPINESS: RANK OF INCOME, NOT INCOME, AFFECTS LIFE SATISFACTION
2.1
Abstract
Does money buy happiness, or does happiness come indirectly from the higher rank in society
that money brings? Here we test a rank hypothesis, according to which people gain utility from
the ranked position of their income within a comparison group. The rank hypothesis contrasts
with traditional reference income hypotheses, which suggest utility from income depends on
comparison to a social group reference norm. We find that the ranked position of an individual’s
income predicts general life satisfaction, while absolute income and reference income have no
effect. Furthermore, individuals weight upward comparisons more than downward comparisons.
According to the rank hypothesis, income and utility are not directly linked: Increasing an
individual’s income will only increase their utility if ranked position also increases and will
necessarily reduce the utility of others who will lose rank.
Previously published as: Boyce, C. J., Brown, G. D. A., Moore, S.C. (in press). Money and
Happiness: Rank of Income, not Income, Affects Life Satisfaction. Psychological Science
28
2.2
Introduction
Is there a true causal relation between money and happiness? According to conventional
economics, there is: Money can buy happiness because it can be exchanged for goods that will
increase an individual’s utility. Thus money and happiness are assumed to be causally linked,
and higher incomes should lead to greater happiness. In line with this absolute income hypothesis
richer people are happier than those less well off within the same society (Diener, 1984). The
correlation between money and happiness is often small, but effect sizes are larger in low-income
developing economies (Howell & Howell, 2008) and even small correlations can reflect
substantial real differences in happiness (Lucas & Schimmack, 2009). Such results, however, do
not necessarily reflect a simple causal relation between money and happiness. The idea that
absolute income leads to increased happiness is unable to account for the Easterlin paradox – that
income and happiness are positively associated within a country at a given time, but not (or less
well) correlated within a country over time (Easterlin, 1974).
Furthermore, being amongst people richer than oneself can be detrimental to wellbeing
variously measured (Blanchflower & Oswald, 2004; Clark, Frijters et al., 2008; Clark & Oswald,
1996; Ferrer-i-Carbonell, 2005; Luttmer, 2005), consistent with income comparison. Self-rated
happiness and satisfaction scores have been shown to act as valid and reliable proxies for utility
(e.g. Lepper, 1998; Sandvik et al., 1993). The data have therefore been taken to suggest that an
individual’s utility is influenced not by absolute level of income but instead by their income
relative to that of their peers.
The reference income hypothesis is the dominant model of income comparison and
suggests that individuals care about how their income compares to the norm, or reference
income, of a socially constructed comparison group. Again, a direct causal link is assumed:
29
Increased income will lead to increased utility for an individual if all else is held constant.
Individuals gain utility to the extent that their income exceeds the average or reference income of
people in their comparison set, and lose it to the extent that their own income falls below the
reference level. The average income of an assumed reference group typically negatively and
significantly predicts a number of variables related to well-being, consistent with the reference
income approach (e.g. Clark & Oswald, 1996).
Here we suggest instead that utility is based on an individual’s ranked position within a
comparison group – the rank income hypothesis. According to the rank-based model, people gain
utility from occupying a higher ranked position within an income distribution rather than from
either absolute income or their position relative to a reference wage (Brown et al., 2008; Clark,
Kristensen, & Westergard-Nielsen, 2009a; Clark, Masclet, & Villeval, in press; Hagerty, 2000;
R. H. Smith et al., 1989). For example, people might care about whether they are the second
most highly paid person, or the eighth most highly paid person, in their comparison set (which
might contain fellow workers of a similar age and qualification level, neighbours, friends from
college, etc). The ranked position of an income will be highly correlated with the position of that
income relative to a mean, so evidence previously taken to support reference income accounts
may be consistent with a rank income account. Not only do rank and reference based models
predict very different savings and consumption behaviour (Bilancini & Boncinelli, 2007) but
also, according to the rank income hypothesis, there is no simple causal relationship between
money and happiness: An increase in income need not increase ranked position and hence need
not increase happiness.
A rank based approach to judgment is independently motivated by the fact that judgments
about items within a context of other items are known to be influenced by the ranked position of
30
the item along the dimension of interest. This perspective originated within psychophysics in the
judgment of quantities like weight or pitch, but has since been extended to economic and social
phenomena (e.g. Mellers, 1986; Niedrich et al., 2001; Parducci, 1995; Stewart, Chater, & Brown,
2006). Subjective judgments of utility may be governed by context just like judgments of other
quantities (Parducci, 1995).
There is already some evidence that rank income rather than reference or absolute income
may be important, although previous large scale studies have looked only at satisfaction with
economic conditions and not overall life satisfaction. In a study of 16,000 British workers wage
satisfaction depended on the ordinal rank of an individual’s wage within a workplace (Brown et
al., 2008). Further, a study of 9,000 small neighbourhoods researchers found that satisfaction
with economic conditions increased with ranked position within a neighbourhood (Clark et al.,
2009a). Other studies have considered rank in the broader context of range-frequency theory
(Hagerty, 2000; R. H. Smith et al., 1989). However no large-scale study has examined the effect
of income rank on self reported general life satisfaction. Here we use data from 12,000 British
adults to examine this question. We also examine whether upward comparison (the number of
people earning more than oneself) has a greater influence on life satisfaction than downward
comparison (Duesenberry, 1949).
2.3
Method
We test a simple rank-based model according to which the individual compares themself
to a sample of other people in their reference group and assesses whether each sampled
individual earns more or less than themselves (Stewart et al., 2006). Those assigned “worse than”
(i-1) are compared to the total number within the reference group (n-1). The ratio gives the
individual a relative rank (Ri) normalized between 0 and 1:
31
(2.1) Ri 
i 1
n 1
We use Ri to predict life satisfaction in a multiple regression after the influence of other
relevant variables have been partialled out. Data are taken from seven years of the British
Household Panel Survey (BHPS), which is a representative longitudinal sample of British
households. All adults, from 1997 to 2004, who answered a life satisfaction question, are
included in the analysis1 (n= 86679). Life satisfaction is the respondent’s answer on a 1 to 7 scale
to the question: “how dissatisfied or satisfied are you with your life overall?” and is taken here to
proxy for an individual’s utility and standardized. Household incomes were adjusted for regional
living cost differences and number of individuals in the household: Total household income was
divided by 2004 regional living costs and weighted by household size (adults = 1 unit; each child
= 0.5 units). After such adjustment those with children, or those that may stay at home in the
presence of a big income earner, will have comparable spending powers. Demographic
characteristics were controlled for in all analyses.
We first report analyses comparing rank income and income in the overall sample, then
divide the sample into reference groups to test the rank income hypothesis against the reference
income hypothesis. Finally, we look for evidence of asymmetric (upward) comparison.
2.4
Results
First, the ranked position of each individual’s income within the entire sample in a given
year was compared to the individual’s absolute income as a predictor of life satisfaction. Table
1
The 2001 wave included no life satisfaction question and was therefore excluded.
32
Table 2.1: Pooled OLS regression on life satisfaction comparing logarithm of absolute income and income
rank by sample
Independent Variables:
Dependent Variable: Life Satisfaction (standardized)
1
2
3
Income Ranka
0.288
(21.46)**
Log(Household Incomeb)
R-Squared
Observations
0.302
(10.60)**
0.10
(18.66)**
-0.006
(0.53)
0.0826
86679
0.0838
86679
0.0838
86679
Absolute value of t-statistics in parentheses
* significant at 5% level; ** significant at 1% level
All analyses included demographic controls: age, gender, education, marital status, children, housing
ownership, labour force status and disabilities, and dummy variables identifying both region and wave.
In all cases, these variables accounted for significant variation in life satisfaction.
a. Based on the individual’s household income adjusted for household size and deflated by regional
living costs
b. Adjusted for household size and deflated by regional living costs
2.1 compares absolute income (logarithmically transformed2) and rank income variables. Each is
significant when entered as the only income-related predictor after controls (columns 1 and 2).
The coefficient from column 1 suggests that, once controlling for other factors, the life
satisfaction difference between the highest and lowest earners is 0.29 standard deviations.
Alternatively, the coefficient on the logarithm of household income shown in column 2 suggests
that on average an individual will be 0.1 standard deviations higher in life satisfaction than
someone earning about half as much. However, rank explains significantly more of the overall
variation (R2) in life satisfaction. Furthermore, when both income variables are entered
simultaneously, rank income dominates and absolute income accounts for no additional variance
2
The natural logarithm of income is the transformation typically used in income and happiness studies so provides a
useful benchmark against which to test rank income. Higher order polynomials in income against rank income were
also tested, but logarithm of income explained more of the variation in life satisfaction
33
(column 3) consistent with a role for ranked position of income, not income per se, in
determining life satisfaction.3
Next, we compared the rank and reference income hypotheses. To do this we constructed
various reference groups to explore the possibility that people compare their income to others in
the same geographical region (of which there were 19 in the BHPS), of the same gender and
education (three levels: graduate, college and neither) giving six groups in total, or of the same
age (we used 12 different age groupings: all less than 20 years old, 20-24, 25-29, 30-24, 35-39,
40-44, 45-49, 50-54, 55-59, 60-64, 65-69 and all older than 70). In each case we computed the
relative rank of each individual’s income within the reference group and also the mean income of
all individuals within the reference group. We then predicted each individual’s life satisfaction
from (a) their relative rank within the reference group, (b) their absolute income (logarithmically
transformed), and (c) mean reference group income (logarithmically transformed).
We were then able to test the rank income hypothesis against both absolute income and
reference income hypotheses. Results are shown in Table 2.2 and the t-statistics are adjusted for
clustering (Moulton, 1990). In all cases the rank position of an individual’s income within their
reference group dominated the explanation of life satisfaction. When geographically-defined
reference groups were assumed, rank income was significant whilst absolute income was not
(column 1). An R-squared comparison further reveals that rank income also explained more of
3
A fixed effect analysis, analyzing the within person variation, was also undertaken. The fixed effect analysis
controls for unobservable heterogeneous factors. Again rank dominates: when entered simultaneously the coefficient
on the rank variable is 0.06 and significant, whereas the coefficient on the absolute income variable is 0.02 and
insignificant.
34
Table 2.2: Pooled OLS regressions on life satisfaction comparing logarithm of mean income and income rank using various reference groups
Dependent Variable: Life Satisfaction (standardized)
Reference Group:
Independent Variables:
Log(Household Incomeb)
Income Ranka
Region
1
-0.004
(0.38)
0.101
(16.30)**
0.294
(9.36)**
Log(Mean Reference Group Incomeb)
Observations
R-Squared
2
86679
0.0838
Gender and Education
3
4
-0.004
(0.38)
-0.007
(0.50)
0.294
(9.46)**
0.289
(10.89)**
-0.050
(0.47)
0.011
(0.11)
86679
0.0826
86679
0.0838
86679
0.0839
Age
5
6
7
8
9
0.101
(7.43)**
-0.007
(0.50)
0.003
(0.20)
0.103
(9.43)**
0.013
(0.76)
0.289
(11.07)**
0.270
(4.95)**
-0.213
(0.79)
-0.130
(0.48)
86679
0.0826
86679
0.0839
86679
0.0838
0.244
(3.68)**
-0.365
(2.10)**
-0.263
(1.34)
86679
0.0831
86679
0.0840
Absolute value of t-statistics in parentheses (adjusted to account for clustering as a result of aggregated variables (see Moulton, 1990).
* significant at 5% level; ** significant at 1% level
All analyses included demographic controls: age, gender, education, marital status, children, housing ownership, labour force status and disabilities, and dummy variables
identifying both region and wave. In all cases, these variables accounted for significant variation in life satisfaction.
a. Based on the individual’s household income adjusted for household size and deflated by regional living costs
b. Adjusted for household size and deflated by regional living costs
35
the variation in life satisfaction than the reference group income model (column 2). Neither
reference income nor absolute income explained any additional variance over rank income
(column 3). Similar results were found when individuals were assumed to compare themselves to
others of the same education level and gender (columns 4, 5 and 6) or to others of similar age
(columns 7, 8 and 9).
The final analyses examined whether upwards comparisons were weighted more heavily.
It is commonly suggested that comparison is asymmetric, being made mostly to those above
oneself (Blanchflower & Oswald, 2004; Duesenberry, 1949; Ferrer-i-Carbonell, 2005). Does the
model improve when upward comparison is accommodated? The relative rank measure can be
adapted in a way such that higher ranked others have greater (or lesser) impact on the
individual’s assessment of their own income than those below (above). We refer to this as
subjective income rank (SR) (Brown et al., 2008):
(2.2) SRi  0.5 
(i  1)   (n  i)
2 (i  1)   (n  i)
Here, η captures the degree of upward comparison and increases the weight given to
those who earn more. If η = 1, equation 2.2 can be re-written as equation 2.1. When η > 1,
individuals earning more than i influence perception of the individual’s rank more than those
earning less. If η = 2, for example, the number of individuals that earn more than i matters twice
as much as those that earn less. Subjective rank, based on the whole sample for each wave
according to equation 2.2 with a given value of η, was compared to the simple relative rank
income variable (η = 1). With η set to 1.75 (the optimal value is based on explaining the highest
variation in life satisfaction) significant additional variance is accounted for [F(1, 86641) = 8.75;
p < 0.01]. The coefficient on the rank variable that incorporates this degree of upward
36
comparison is 0.394 and significant, whereas the coefficient on the absolute income variable is 0.03 and insignificant. This result supports Duesenberry’s (1949) claim that comparison is
primarily upwards and shows further that people compare to those above themselves one and a
three-quarter times more than those below.
2.5
Discussion
In analysis of more than 80,000 observations the relative rank of an individual’s income
predicts the individual’s general life satisfaction, and removes the effect of absolute income. In
analyses assuming that individuals compare themselves to smaller reference groups, relative rank
of income continues to dominate life satisfaction. Results suggest that individuals sample from a
reference group and compare their own income with sampled incomes ordinally – satisfaction is
gained from each “better than” comparison and lost for each “worse than” comparison. No
calculation of mean reference group income is required. We note that rank could be influencing
either an “underlying internal utility”, or an individual’s interpretation of their own utility. On the
latter interpretation, individuals will score themselves as more happy to the extent that they
perceive themselves as ranking higher in happiness than others. Although this possibility is
difficult to exclude, we note considerable evidence for relative effects in neuroscience (e.g.
Fliessbach et al., 2007) along with the observation that subjective wellbeing ratings correlate well
with observable behavioural measures (Ekman et al., 1990; Koivumaa-Honkanen et al., 2001).
We also note that income rank may well act as a proxy for more general social rank
(Powdthavee, in press), with the analyses then showing that social rank is key to wellbeing. The
rank hypothesis carries several implications. First, it assumes no direct causal relationship
between income and wellbeing. Unless the individual’s ranked position were perceived to
37
change, income could increase without increasing utility.4 Rank income also predicts a concave
utility function when comparison incomes are positively skewed, because an increasing income
at the lower end of the income distribution will increase rank faster (Brown et al., 2008;
Kornienko, 2004; Stewart et al., 2006). Finally, to the extent that there are effects only of rank,
income distribution cannot affect society’s income-derived utility. However, dissatisfaction could
still result from inequality per se (Alesina, Di Tella, & MacCulloch, 2004).
Our study underlines concerns regarding the pursuit of economic growth. There are fixed
amounts of rank in society – only one individual can be the highest earner. Thus pursuing
economic growth, although it remains a key political goal, might not make people any happier.
The rank hypothesis may explain why increasing the incomes of all may not raise the happiness
of all, while at the same time wealth and happiness are correlated within a society at a given
point in time.
4
We note the possibility that “previous self” may enter the comparison set (e.g. Vandestadt, Kapteyn, & Vandegeer,
1985), in which case any increase in income could lead to increased utility.
38
CHAPTER 3
3
MONEY OR MENTAL HEALTH: THE COST OF ALLEVIATING PSYCHOLOGICAL DISTRESS
WITH MONETARY COMPENSATION VERSUS PSYCHOLOGICAL THERAPY
3.1
Abstract
Money is the default way in which intangible losses, such as pain and suffering, are currently
valued and compensated in law courts. Economists have suggested that subjective well-being
regressions can be used to guide compensation payouts for psychological distress following
traumatic life events. We bring together studies from law, economic, psychology and medical
journals to show that alleviating psychological distress through psychological therapy could be at
least 32 times more cost effective than financial compensation. This result is not only important
for law courts but has important implications for public health. Mental health is deteriorating
across the world – improvements to mental health care might be a more efficient way to increase
the health and happiness of our nations than pure income growth.
Revise and resubmit for Health Economics, Policy and Law
39
3.2
Introduction
Putting a price tag on “pain and suffering” seems an impossible task but judges in law
courts are regularly expected to make such decisions. Equating money with an intangible loss
may seem peculiar but in tort law an individual who has suffered should, as nearly as possible, be
restored to the same position had they not sustained some wrong (Lunney & Oliphant, 2008). In
law courts monetary compensation is the expected remedy and there are established monetary
guidelines to compensate for the “pain and suffering” of various injuries (Mackay, Bruffell,
Cherry, Hughes, & Tillett, 2006). In the UK, the Fatal Accidents Act 1976 provides a one off
payment for the “bereavement” of family members. Economists have developed a method to
place monetary values on various life events (Blanchflower & Oswald, 2004; Ferrer-i-Carbonell
& van Praag, 2002; Powdthavee, 2008). It has further been suggested that such monetary values
could be offered as compensation to help overcome psychological distress after particularly
traumatic life events (Oswald & Powdthavee, 2008a, 2008b). The economists’ calculations
would suggest that court settlements would need to be much higher than present to fully
compensate an individual. The high monetary values reflect money’s ineffectiveness at
compensating someone for pain and suffering.
We assess the evidence across law, economic, psychology and medical journals and
suggest psychological therapy as an alternative. Psychological therapy would be substantially
more cost effective than financial compensation at alleviating psychological distress. We extend
our argument beyond the law courts and suggest that money’s low importance in achieving
mental health has important implications for public health. National happiness levels have
remained flat in developed countries in spite of large economic gains. Mental health, on the other
hand, appears to have been deteriorating across the world for some time and is estimated to
40
deteriorate still further (Michaud et al., 2001). The comprehensiveness and accessibility of
mental health services, in particular the provision of psychological therapies in publicly funded
services, have regularly been questioned. Increasing the investment in mental health and
generally broadening access therefore might be a more efficient way to increase the health and
happiness of our nations than pure income growth.
3.3
Money - A Common Metric for valuing Life Events and the Movement towards
Compensation
Monetary values have been calculated across numerous areas of life including; marriage
(Blanchflower & Oswald, 2004), social relationships (Powdthavee, 2008), the fear of crime
(Moore & Shepherd, 2006), noise (Van Praag & Baarsma, 2005), health (Ferrer-i-Carbonell &
van Praag, 2002) and disabilities (Oswald & Powdthavee, 2008b). The income equivalences
attached to such events are typically large. For example, the value of a marriage is estimated to
be equivalent to having an extra $100,000 (around £70,000 at today’s exchange rate) each year
(Blanchflower & Oswald, 2004). Such values have been calculated using subjective well-being
data and computation is relatively simple. The average impact of some life event on an
individual’s wellbeing can be determined statistically across a large sample. This impact is then
compared to the effect that income has on an individual’s well-being. Typically, a one standard
deviation rise in income would be expected to induce a rise in well-being of between 0.17 and
0.21 standard deviations (Lucas & Dyrenforth, 2006). By comparing the two effects researchers
can estimate how much extra income an individual would need on average to achieve an
equivalent level of well-being as the life event.
Compensation for injustices is an important aspect of society. In tort cases, particularly
those involving psychological distress, judges are commonly faced with the dilemma of awarding
41
compensation to restore an individual to the position they were before any injustice took place.
Such a decision is mostly subjective, arbitrary and normally takes the form of a one off monetary
payment (Mackay et al., 2006). Some researchers have suggested that the psychological impact
of particularly traumatic life events can be evaluated in purely financial terms (Clark & Oswald,
2002). With the aim of alleviating psychological distress recent studies have suggested that such
figures could be useful in a court of law to guide compensation payouts for individuals who have
lost family members (Oswald & Powdthavee, 2008a) or become disabled (Oswald &
Powdthavee, 2008b).
Such events can devastate the lives of individuals and the sums suggested to help
alleviate the psychological distress are large; much larger than those presently awarded by courts.
For example, were an individual to lose a partner, then it is suggested that a compensation
amount ranging from £114,000 to £206,000 per annum would be needed to overcome
psychological distress. For the loss of a child, individuals would require anything from £89,000
to £140,000 to compensate (Oswald & Powdthavee, 2008a). In the UK the Fatal Accidents Act
1976 recommends a substantially lower payout of just £10,000. If such a financial compensatory
argument were extended to unemployment, which is well known to have deep psychological
effects beyond the simple loss of income (Darity & Goldsmith, 1996), the income equivalence of
psychological distress alone would be in the region of £34,000 to £59,000 per annum (Oswald &
Powdthavee, 2008a).
3.4
The Clinical and Cost Effectiveness of Psychological Therapy
Researchers have assessed the clinical and cost effectiveness of various treatments for
depressed patients (Bower et al., 2000; Ward et al., 2000). The effectiveness of general
practitioner care with both cognitive-behaviour therapy (CBT) and non-directive counselling
42
were all compared. Over twelve months all treatments reduced average depression levels by at
least one and a half standard deviations. The average total cost, which even included indirect
costs such as work time lost, was less than £1,500. In fact, the improvement was achieved by
both CBT and non-directive counselling within the first four months at a total cost of less than
£800.
3.5
A Cost Effectiveness Comparison between Psychological Therapy and Direct Financial
Compensation
Comparing the cost effectiveness of psychotherapy and direct financial compensation in
alleviating individuals from severe psychological distress has not been carried out before. There
are several ways in which the costs between psychological therapy and direct financial
compensation can be compared using the studies already outlined. Oswald and Powdthavee
(2008a) estimate that the average psychological impact of losing a partner is about one standard
deviation. They suggest that the individual would need to be compensated with at least £114,000.
Pro-rata it would cost less than £600 to help the individual adjust to such a difficult life event
using psychological therapy. Similarly for unemployment, which has deep psychological
consequences, it would be more cost effective to provide individuals with psychological therapy
(around £100-£200 pro rata) to overcome their loss of purpose in life and help them back to work
rather than solely offering financial compensation.
To compare the costs in another way; psychological therapy alleviates psychological
distress by one and a half standard deviations at a cost of £800 over 4 months. To achieve a one
and a half standard deviation reduction in psychological distress using money alone would
require (based on estimates in Oswald and Powdthavee (2008a) and dependent on the statistical
technique) somewhere in the region of £179,000 to £292,000 of extra income every year. This
43
illustrates that the alleviation of severe psychological distress could be worth at least £179,000 of
extra income each year and suggests that financial compensation is an inefficient way of helping
individuals overcome distress.
The wide disparity between the effects of psychological therapy and income arises out of
the poor ability of income to improve mental health. However, income’s effect on psychological
distress in studies such as Oswald and Powdthavee (2008a) is likely to be under estimated. More
realistic income effects can be obtained by allocating income randomly to individuals and
observing the effects on individuals’ lives. Such an experiment is not possible but researchers
have analyzed longitudinally the effect of medium sized lottery wins on psychological distress.
An average lottery or pool win of £4,300 is found to bring approximately a quarter of a standard
deviation improvement to mental health two years after the win (Gardner & Oswald, 2007). This
suggests that even when the best income-psychological distress estimates are considered
(Gardner & Oswald, 2007), psychological therapy is still calculated to be at least 32 times more
cost effective than financial compensation. This figure is large and places huge questions on the
use of income as an effective compensation method.
Our argument is not without its limitations. There are of course inherent difficulties in
making inferences across studies and just like the proponents of the use of monetary
compensation it is necessary to draw some conclusions from one group of individuals to
another5. The evidence on the cost-effectiveness of psychological therapy is not based
exclusively on individuals experiencing loss as a result of a devastating life event. Some
psychological distress will undoubtedly follow from the loss of a loved one but one may question
5
The argument for using subjective well-being measures to calculate monetary compensation levels must
necessarily assume that the effect of income on psychological distress, based on estimates from the entire sample,
will be the same for the small sub-sample who have recently lost a family member
44
whether, what could be considered a fairly normal response, would benefit from psychological
therapy. This perhaps calls into question the purpose of psychological treatment more generally.
One view of psychological therapy is that mental health is a continuum and that everyone, no
matter their level of psychological distress, can benefit from psychological therapy. Under this
view any psychological distress would be viewed as mental disorder and therefore treatable.
Another viewpoint is that mental disorder is dichotomous, in that individuals either have mental
disorder or do not. This view suggests that mental disorder is the presence of distorted thought
processes that can result in protracted psychological distress. In this latter view the psychological
distress that arises as a normal reaction to loss might not be seen as directly treatable. However, it
is likely that, without adequate support, some unnecessary mental disorder will arise from the
loss. Therapy should aim to avert such mental disorder. Additionally, even if there were a
substantial cost difference in helping individuals with loss and those with mental disorders, the
cost-effectiveness comparison figure calculated earlier should be high enough to absorb any extra
expense.
A further issue that requires some discussion is the process of adaptation. People adapt to
many of life’s events (Brickman et al., 1978; Clark, Diener et al., 2008) and some individuals
will naturally return to their baseline level of psychological functioning given time. If individuals
adapt anyway then an argument could be made against the use of psychological therapy. The
same argument, however, could be made against the use of monetary compensation, the current
default method of addressing loss. In any case individuals do not always fully adapt and even if
they did resources should be used in the most efficient manner to relieve psychological distress in
the interim. Oswald and Powdthavee (2008b) address the issue of adaptation and as such suggest
that annual compensatory amounts should be reduced each year. Similarly, psychological therapy
45
could be proportionally reduced to help with the lower levels of distress in later years after an
event. However, it may turn out that psychological therapy, focusing directly on psychological
distress, speeds up the process of adaptation and may even mean therapy is unnecessary in later
years.
3.6
Practical Implications of our Argument
3.6.1 For Judges
In tort law an individual who has suffered should, as nearly as possible, be restored to the
same position had they not sustained some wrong. The loss of future earnings must undoubtedly
be compensated financially. The evidence suggests that following a devastating life event an
individual is likely to experience an adverse psychological reaction and our concern is with the
efficiency of compensating such an intangible and fairly normal psychological reaction
financially. Currently monetary compensation seems to be unquestionably taken in law courts as
the only way of helping an individual overcome psychological distress after a traumatic event.
The values currently offered as compensation are arbitrary (Mackay et al., 2006) and, according
to economists’ subjective well-being equations, should actually be much higher (Oswald &
Powdthavee, 2008a, 2008b). Rather than giving individuals more income to cope with distress it
seems sensible to consider other alternatives such as psychological therapy.
Can bestowing an individual with money really be expected to help an individual with
psychological distress after a traumatic life event? As a compensatory device, money, like most
things, can never truly fulfil such a role. The purpose of financial compensation, however, is to
alleviate an individual’s psychological distress by helping them to find enjoyment elsewhere.
Thus, if a tool is judged by its ability to alleviate an individual’s psychological distress then the
answer may be found in medical research. We have shown that psychological therapy could be
46
much more cost effective than financial compensation. We are not claiming that psychological
therapy will entirely prevent what could be considered as a normal adverse psychological
reaction but therapy may provide some short term relief. Moreover such devastating life events if
not adequately dealt with may lead to maladaptive thoughts and result in undue distress. It is
unlikely that financial compensation will protect against such maladaptive thinking.
On punitive grounds we are not necessarily suggesting that large court payouts are not
justified6. However, we are suggesting that the sums currently offered may not be the best way to
help the injured party overcome any psychological distress unless it is decreed, in the best
interests of the injured party, that the money gets spent on some form of psychological therapy.
There is currently no legal obligation for individuals to undertake steps to alleviate their
psychological distress. However, there is some legal support in New Zealand for an increased use
of psychological therapy to help individuals overcome psychological distress that occurs due to
the actions of a third party. The Injury Prevention, Rehabilitation and Compensation Act of 2001
requires that the cost of psychological therapy be borne by the state.
Financial compensation may not even have the causal effect on a traumatized individual
in the way that economists suggest. Additionally, individuals are likely to be unequally affected
by financial compensation, for example due to the diminishing marginal utility of income
wealthier individuals may require even more income to replenish lost well-being. Financial
compensation seems like a poor device for alleviating psychological distress. In contrast
psychological therapy does not attempt to act as compensation but instead focuses directly on
6
There are several theories concerning the purpose of tort, these include; deterrence, corrective justice, risk
allocation and distribution of loss. Some, depending on the legal system, would argue that retribution also plays an
important part
47
helping individuals to overcome their loss. Psychological therapy acknowledges that trauma is
person dependent and is therefore effective, inexpensive and compassionate.
3.6.2 For Policy Makers and Society
We believe our argument has important implications for public health. The high levels of
financial compensation that economists suggest help to highlight how inefficient money is at
alleviating psychological distress and generally improving individual well-being. Economists are
puzzled as to why developed societies are becoming no happier in spite of large income gains
(Easterlin, 1995). However, in 1999 unipolar major depression was the fifth leading cause of the
disease burden and by 2020 major depression is expected to rise to the second biggest burden
(Michaud et al., 2001). These conflicting findings, along with the evidence presented here,
indicate that although income growth may provide some benefit to well-being, the greatest
benefit to the health and happiness of our nations could come from improving access to mental
health care.
Our argument, although clearly comparing across studies and extending from the law
courts to wider society, adds a new perspective to public health debate on mental health.
Regarding the UK’s National Health Service, Richard Layard (2006b) has suggested that the cost
of helping individuals overcome mental health issues using psychological therapy more than
outweighs the money saved from reducing the individuals on benefits who are incapacitated by
their mental health. Our argument adds further support for increasing the availability of mental
health resources by instead suggesting that mental health in its own right is something to be
valued alongside economic progress. The importance of improving mental health for national
well-being needs to be further recognized and policy makers must consider improving mental
health care further. Individuals are also probably not fully aware of the powerful effects that
48
good psychological therapy can have on their mental health and general well-being. It needs to
be understood that aspiring to good mental health can often be more important than aspiring to
high income for well-being. Since individuals are unlikely to unilaterally invest in their own
mental health there is a strong case for public provision.
49
CHAPTER 4
4
UNDERSTANDING FIXED EFFECTS IN HUMAN WELL-BEING
4.1
Abstract
In studies of subjective well-being, economists and other researchers typically use a fixed or
random effect estimation to control for unobservable heterogeneity across individuals. Such
individual heterogeneity, although substantially reducing the estimated effect of many
characteristics, is little understood. This paper shows that personality measures can account for
20% of this heterogeneity and a further 13% can be accounted for by other observable betweenperson information. This paper then demonstrates that the use of personality measures, in a new
technique developed by Plumper and Troeger (2007), can help researchers obtain improved
estimates for important characteristics such as marital status, disability and income. The paper
argues that this has important practical implications.
Previously published as: Boyce, C. J. (in press). Understanding Fixed Effects in Human WellBeing. Journal of Economic Psychology
50
4.2
Introduction
Economists wishing to evaluate how economic circumstances benefit an individual’s life
are increasingly turning to subjective well-being data. There are now numerous studies by
economists that reveal the benefits of, for example, more income (Blanchflower & Oswald,
2004; Clark & Oswald, 1996; Ferrer-i-Carbonell, 2005; Ferrer-i-Carbonell & Frijters, 2004;
Frijters et al., 2004; Luttmer, 2005; Senik, 2004) and sustained employment (Clark & Oswald,
1994; Di Tella, MacCulloch, & Oswald, 2001; Winkelmann & Winkelmann, 1998). An
important concern in subjective well-being studies is how to deal with heterogeneity between
individuals that is largely considered to be unobservable. There is some uncertainty as to what
individual heterogeneity consists of but the term fundamentally represents fixed factors unique to
each individual that drive the individual to earn a higher income or remain in employment yet
also enable them to have higher levels of well-being over the course of their life. An example
would be the individual’s personality. Any researcher interested in the independent effects of
increasing an individual’s income must somehow control for these correlated but largely fixed
and unobservable heterogeneous factors.
A typical approach to overcome issues of heterogeneity is to exploit panel data and
perform either a random or fixed effect estimation. Since individual heterogeneity is generally
viewed as fixed across time, the observation of individuals at several time points allows
researchers to statistically control for the heterogeneous factors without having to directly
observe or quantify them. Inevitably, when important explanatory variables are unavailable
estimation by ordinary least squares (OLS) will result in biased estimates. Ferrer-i-Carbonell and
Frijters (2004) document this bias from not appropriately controlling for individual heterogeneity
and observe that the positive coefficient on income reduces by about 2/3 when moving from a
51
pooled OLS to a fixed effect estimator. The bias is large and illustrates the necessity of
controlling for important yet largely unobservable and unknown characteristics. Ferrer-iCarbonell and Frijters (2004) further suggest that individual heterogeneity, which appears to be
an important source of information as to why some individuals have higher levels of well-being
than others, needs to be better understood. This paper aims to elicit an understanding of
individual heterogeneity.
Aside from controlling for its presence there has been little work directed at
understanding what is truly included within the set of fixed heterogeneous factors. Within the
well-being literature there is no real consensus on what individual heterogeneity is and the term
is often used to reflect aspects of an individual’s character that researchers have difficulty
measuring. Some researchers have suggested that in part there is some bias in individual
responses to subjective well-being questions due to processes such as anchoring or whether the
individual has an optimistic or pessimistic view of life (Clark et al, 2005; Ferrer-i-Carbonell,
2005; Winkelmann & Winkelmann, 1998). It has also been argued that an individual’s health
(Winkelmann & Winkelmann, 1998), their capacity to deal with adversities (Ferrer-i-Carbonell,
2005) and their ability or family back ground (Di Tella & MacCulloch, 2006) are important
components of unobservable heterogeneity.
Perhaps the most widely cited and most important component of individual heterogeneity
is an individual’s personality. Many subjective well-being researchers have in fact made the
explicit assumption that the unobservable individual heterogeneity is mainly personality traits
(Booth & van Ours, 2008; Ferrer-i-Carbonell & Frijters, 2004; Frijters et al., 2004; Senik, 2004;
Vendrik & Woltjer, 2007). It is unclear what is truly contained within the all encompassing term
individual heterogeneity yet the assumption that personality is the main component has received
52
little scrutiny. Perhaps unfamiliar to many economists is the fact that personality can be reliably
measured and that personality research has a long history in psychology (Winter & Barenbaum,
1999). The use of such personality measures would enable researchers to improve their
understanding of individual heterogeneity and determine how important personality is compared
to other factors, such as the individual’s health or background, as a component of individual
heterogeneity. Further, an improvement to our understanding of individual heterogeneity may be
of benefit in the estimation of subjective well-being equations.
The estimation strategy that seems to be favoured in subjective well-being studies to
deal with individual heterogeneity is a fixed effect (FE) estimation. Based on the assumption that
the unobservable heterogeneity is correlated with explanatory variables the FE model focuses
solely on explaining the within-person variation. As a result the individual heterogeneity, which
is considered fixed across time, contains no within-person explanatory power. In fact, all of the
between-person information is not essential for estimation and is grouped together into what is
referred to as the individual fixed effects and mostly ignored. However, this focus on the withinperson variation is inefficient and the resultant loss of between-person information has been
described by Beck and Katz (2001) as like “throwing the baby out with the bath water”. It is
therefore not possible to obtain reliable estimates on characteristics that have zero or low withinperson variation using an FE estimation (Plumper & Troeger, 2007).
An alternative that circumvents the problem of obtaining reliable estimates on
characteristics with low within-person variation is a random effects (RE) estimation. Here, a
different assumption is made about individual heterogeneity, in that the heterogeneity is
uncorrelated with explanatory variables of interest. This is a strong assumption and, although the
RE model is more efficient than the FE model, since it uses both within and between-person
53
information, is likely to produce biased estimates if this core assumption is violated. Mundlak
(1978) proposed a solution to this problem by allowing for correlation between the unobservable
heterogeneity and some of the observable characteristics. The random effects model with a
Mundlak (1978) transformation (REMT) overcomes the inefficiency problems associated with
the FE model but still maintains many of the key FE assumptions. An example of the REMT’s
use in subjective well-being research can be seen in Ferrer-i-Carbonell (2005).
It is clear that the correlation between the unobservable heterogeneity and observable
characteristics is a key factor determining a subjective well-being researcher’s choice of
estimation strategy. However, if personality measures can add to our understanding of individual
heterogeneity then this correlation problem might not be so critical. The use of personality
measures in estimation may allow panel data estimation strategies to be applied to subjective
well-being research that are less restrictive than the FE or REMT. It has already been
demonstrated that personality measures are a useful substitute to control for individual
heterogeneity when panel data is unavailable (Anand et al., 2009). This paper also exploits
personality measures, but instead within a panel data framework, and suggests the fixed effect
vector decomposition (FEVD) model, developed by Plumper and Troeger (2007), as an
alternative to the FE and REMT models. The FEVD model is based on the standard FE model
but also recognises that some of the fixed individual heterogeneity is in fact observable. The
FEVD has three stages and, whilst giving the technique efficiency advantages over the FE model
by maximising the use of between-person variation, the technique still controls for the truly
unobservable component of individual heterogeneity. In the first of the three stages individual
fixed effect residuals are estimated using the FE model. Individual fixed effect residuals, which
absorb all between-person information, are then decomposed into an observable and
54
unobservable component. Here, available characteristics that have little or zero within-person
variation can be used to predict the fixed effect residual. The final stage then uses the error term
from stage two, representing the true unobservable heterogeneity, and all the observable
characteristics, as an explanatory variable in a pooled OLS estimation.
The FEVD allows the estimation of characteristics that have zero or only small withinperson variations. In studies of subjective well-being the application of FEVD, like the REMT,
seems immediately beneficial for obtaining estimates on various demographic characteristics that
mostly vary between individuals and not within, for example, age, gender and geographical
location, that are unobtainable using the FE model. There are, however, also other important
characteristics that are known to change very little over the individual’s life and the use of either
the FE or REMT models to estimate the effect of such characteristics can often result in
insignificant coefficients. An example would be the individual’s level of education. In some
studies, such as Luttmer (2005) and Ferrer-i-Carbonell (2005), the coefficient on education is
often found to be indeterminable and invites the conclusion that education does not improve the
individual’s life satisfaction.
In the face of estimating variables with low or zero within-person variation it might be
sensible to opt for either an RE or REMT estimation. However, in addition to proposing the
FEVD as an alternative to the FE model for estimating characteristics with low within-person
variation, Plumper and Troeger (2007) also conduct Monte Carlo simulations to demonstrate the
conditions under which the FEVD provides superior estimates to the FE model. Such specific
conditions are not available for models like the RE or REMT. They show that, when a
characteristic of interest has a low enough within-person variation, estimation can be preferable
using FEVD. They also illustrate that what constitutes a low enough within-person variation is
55
heavily dependent upon how much that particular variable correlates with the unobservable fixed
heterogeneity. However, since the fixed heterogeneity is, by definition, unobservable, no
correlation can be obtained. To aid researchers with an application of their technique Plumper
and Troeger (2007) suggest that by being able to explain more of the fixed effect residual in stage
two reduces the likelihood of a high correlation. In the case of subjective well-being research this
would include measures of an individual’s personality.
Many characteristics that are typically estimated in subjective well-being equations have
fairly low within-person variations. This paper uses subjective well-being data from the German
Socio-Economic Panel, which included a number of validated and reliable personality questions
in a recent wave. First, it is shown that many of the conventional variables used in subjective
well-being research lack within-person variation. Second, it is shown using valid and reliable
personality measures that personality is the main component of individual heterogeneity and that
using personality measures to increase our understanding of individual heterogeneity
substantially reduces the correlation between the remaining unobservable heterogeneity and
important explanatory variables. Collectively these two results make estimation possible using a
FEVD. This paper argues that more reliable estimates are obtained on characteristics that have
only moderately low within-person variations and this includes the individual’s marital status,
disabilities and even income. The next section discusses the measurement of personality and its
use within economics. Section 4.4 gives greater technical detail regarding the FEVD and shows
how the technique can incorporate the use of personality measures and be applied to subjective
well-being research. Section 4.5 describes the German Socio-Economic Panel data used in the
analysis. The results are then given in section 4.6 and section 4.7 concludes.
56
4.3
The Use of Personality Measures in Economics
Personality research has a long history in psychology beginning with researchers such as
Allport (1937) and Cattell (1946) and has since developed into a systematic analysis of
individual differences. Models of personality are generally constructed around natural language
and the words used to describe people (for a full development of this lexical approach see John
and Srivastava (1999)). The personality literature is dominated by the Big Five Personality
Inventory, which suggests at the broadest level of abstraction there are five dimensions of
personality. Using factor analysis it has been shown that the large numbers of words used to
describe individuals load onto five main themes: openness to experience, conscientiousness,
extroversion, agreeableness and neuroticism. Although the model is atheoretical the personality
measures such as the Big Five have been shown to measure what they are supposed to (BenetMartinez & John, 1998; Borkenau & Ostendorf, 1990) making them distinct from indicators of
subjective well-being.
Models like the Big Five have been used to address some areas that are of economic
interest. For example, Mueller and Plug (2006), Nyhus and Pons (2005) and Groves (2005) have
all looked at the effect of personality on earnings. Mueller and Plug (2006) show that some
personality traits, such as openness and conscientiousness, are rewarded in the market place,
whereas other traits, such as agreeableness and neuroticism, are penalised. Nyhus and Pons
(2005) draw similar conclusions but also find that the degree of control an individual has is
important for earnings. They further find that the financial return to personality varies across
educational groups. Groves (2005) investigates the importance of psychological traits, such as
autonomy, social withdrawal and aggression in female earnings. Studies such as these may help
explain why even after controlling for many factors, which includes the improved cognitive
57
abilities that come through schooling, there are still large earning gaps. Bowles et al. (2001a)
have suggested that both school and family pass on many important behavioural traits that
enhance the individual’s earning success. However, the use of personality traits in the
determination of wages is very much in its infancy (Bowles, Gintis, & Osborne, 2001b).
More recently researchers have looked at personality’s relation to an individual’s
propensity to share knowledge with work colleagues (Matzler, Renzl, Muller, Herting, &
Mooradian, 2008). In fact, a recent discussion paper by Borghans el al. (2008) evaluates the
integration of personality into economic research more generally. From a theoretical perspective
they discuss ample evidence that suggests that personality should be given greater consideration
when discussing economic parameters and constraints. Borghans et al. (2008) further suggest that
there could be considerable benefit to understanding how economics incentives might influence
individuals with different personality traits.
In contrast to the approach by economists, a psychologist’s discussion of subjective wellbeing will often centre on personality. Personality enables a categorisation of people and their
behaviours and is therefore one of the strongest and most consistent factors predicting the
individual’s well-being (Diener & Lucas, 1999). Economists show greater interest in
demographic factors, like age, education, income and marital status, which although explaining
relatively less of the individual’s well-being (Argyle, 1999) can be influenced by policy. In
subjective well-being studies economists only require some way of controlling for personality.
By assuming personality is fixed, an assumption that is supported by work in psychology (Costa
& McCrae, 1980, 1988), economists are able to control for personality using the models like the
FE and REMT to obtain unbiased causal effects on various demographic factors. The use of
personality measures, however, to better understand the fixed effects will help merge subjective
58
well-being research of both psychologists and economists and may help uncover the true causes
of high subjective well-being (Gutierrez et al., 2005).
4.4
Methodology
At a given time point (t) the individual’s subjective well-being (SWB) is generated as
follows:
D
K
M
d 1
k 1
m 1
(4.1) SWBit     d  t    k xkit    m zmi  i   it
There are some characteristics (δ) which vary across time periods but not individuals, for
example, the economic or social conditions that affect everyone equally at a given time point.
Other characteristics (x) vary across individuals and time periods and may include aspects such
as the individual’s income and employment status. There are other observable characteristics (z)
that vary from individual to individual but not across time, such as gender. Finally, there are
other important factors (μ), considered unobservable, that do not change across time periods and
are referred to as individual heterogeneity.
The key assumption of the FE model is that the unobservable individual heterogeneity is
correlated with the observable characteristics. The FE model eliminates the need to worry about
any individual heterogeneity (μ) by focusing solely on how much individuals vary from their
time-means. This is known as the within-person variation and shown in equation 4.2.
K
M
k 1
m 1
(4.2) SWBit  SWBi  (   )  ( t   )   k  ( xkit  xki )   m  ( zmi zmi )  (i  i )  ( it   i )
59
where
SWBi 
1 T
1 T
1 T
1 T
1 T
,
,
,
,


x
z





SWB

z
x

 it i T 
 it i T 
  it
i
i
i
it
T t 1
T t 1
T t 1
t 1
t 1
By eliminating the individual heterogeneity (μ) unbiased estimates of the x characteristics can be
obtained. However, no estimates of the observable z characteristics are obtainable from such a
regression. Inadvertently they will be eliminated alongside the individual heterogeneity and
important information will be lost. In contrast the RE model makes better use of the information
available. The RE model assumes the unobservable heterogeneity is uncorrelated with observable
characteristics. This means that the time-invariant individual heterogeneity, μ, is subsumed by
the error term, ε. Reasonable estimates can be obtained as long as the error structure is
recognised using an estimator such as generalised least squares (GLS). However, the reliability
of these estimates depends on the strength of the assumption that observable characteristics are
not correlated with the individual heterogeneity. In studies of subjective well-being it would be
difficult to argue convincingly that this was the case. Some researchers have therefore proposed
the use of the REMT. The REMT circumvents this problem by allowing the error structure to
take account of the correlation, in a similar fashion to the FE model, by including the time-mean
values of the observable characteristics that are thought to be correlated with the unobservable
heterogeneity. An issue with such a technique, however, is that there is no clear cut way of
choosing which variables are correlated with the unobservable heterogeneity. There are some
individual characteristics that it would be difficult to argue weren’t correlated with the
unobservable heterogeneity but ultimately the decision is down to a researcher’s discretion.
The FEVD is an alternative estimation strategy proposed by Plumper and Troeger (2007)
which similarly attempts to overcome the loss of information that occurs using the FE model.
60
The advantage of this technique over the REMT is that Plumper and Troeger (2007) provide
clear conditions under which the FEVD estimation is superior to an FE model. Their technique
performs the FE model in its first stage in order to obtain an estimate of the fixed effect residual
( ̂i ). However, they note that the fixed effect residual ( ̂i ) obtained using the FE model is not
the same as the true unobservable heterogeneity (μ) outlined in equation 4.1. The fixed effect
residual also contains the eliminated information of characteristics contained in z as well as the
mean effects of the characteristics contained in x. An estimate of the fixed effect residual ( ̂i )
using the FE model effectively includes all observable and unobservable between-person
information.
K
(4.3) ˆi  SWBi   k xkit   i
k 1
Thus, in the second stage of the FEVD technique Plumper and Troeger (2007) suggest
decomposing the fixed effect residual into a part that is observable and a part that is not. It is this
stage in which greater understanding of the fixed effect can be obtained by using any available
between-person information, which would include personality variables. Here, it would possible
to determine what the main contributing factors to individual heterogeneity were. The
decomposition takes place using z characteristics to predict the fixed effect residual obtained
from stage one.
M
(4.4) ˆ i    m zmi  i
m 1
61
This leaves the true unobservable component of μ, captured in the error term from equation 4.4
and denoted here as η. Next, η is used in a third stage pooled OLS regression as an explanatory
variable.
D
K
M
d 1
k 1
m 1
(4.5) SWBit     d  t    k xkit    m zmi  i   it
Although z variables may have been correlated with μ, they are not correlated with η. Therefore,
by including the error term (η) from stage two the FEVD allows researchers to obtain reliable
estimates on z characteristics.
Plumper and Troeger (2007) discuss the conditions under which characteristics can be
classified as z characteristics and favourably estimated using the FEVD. There are some
variables that belong strictly in either the set of x or z variables; they change all the time or not at
all. However, there are other variables in which a strict categorisation is not possible. For
instance, there are some characteristics that simply do not change very much. Obvious examples
might include education or marital status. For a huge proportion of a population these types of
characteristics may never change, whilst in others they may be changing often. Hence, for part of
a given sample, a particular characteristic will behave like a z characteristic, whilst for others, the
characteristic will behave like an x characteristic. In the FE model the information from the part
of the sample that does not change cannot be used. Similarly, if the characteristic is treated as an
x variable in the FEVD then the information is also ignored. However, it is possible to treat some
characteristics as z variables if they have a low within-person variation. If the within-person
variation is sufficiently small enough then the trade-off between bias and efficiency favours the
efficient estimator.
62
In their Monte Carlo simulations Plumper and Troeger (2007) show using the root mean
squared error under what conditions the gain in information at the sacrifice to bias favours
estimation by FEVD. Plumper and Troeger (2007) pin-point the ratio of a particular variables
between-to-within person variation as a way of distinguishing whether that variable can be better
estimated using FEVD. This ratio, however, depends on how well the particular variable in
question is correlated with the unobservable heterogeneity (η). For example, when the correlation
is 0.5 the between-to-within person ratio must exceed 2.8 for the FEVD to be the superior
estimator. When the correlation drops to 0.3 the between-to-within person ratio only needs to be
1.7. These between-to-within person ratios are fairly low and may include many characteristics
that economists have so far only estimated using the FE or REMT models.
The correlation between unobservable heterogeneity and any potentially low withinperson characteristic is clearly unobservable. However, as Plumper and Troeger (2007) suggest,
including additional z variables in stage two of the FEVD and obtaining a better understanding of
the fixed effect is likely to decrease the size of the unobservable component (η). If the
unobservable component (η) of the individual heterogeneity (μ) is reduced then so too will the
likely correlation between any potentially low within-person variables and the true unobservable
component. Using the truly unobservable component of individual well-being, the error, η, from
stage two of FEVD, it is possible to determine empirically the approximate size of such a
correlation. The use of personality measures as additional z variables in stage two of the FEVD
increases our understanding of the fixed effect and therefore reduces this correlation allowing
many variables to be favourably estimated using the FEVD model. This reduction in correlation
through the use of personality variables may additionally make an RE estimation preferable to
REMT.
63
4.5
Data
The data used to aid the understanding of the fixed factors that contribute to well-being
comes from the German Socio-Economic Panel (GSOEP) survey, a representative longitudinal
sample of German households. The survey asks a number of questions about each individual’s
life. The list of questions includes a single item life satisfaction question:
How satisfied are you with your life, all things considered?
Possible responses range from 0, indicating complete dissatisfaction, to 10, indicating complete
satisfaction. The response to the life satisfaction question is assumed to be cardinal. Other
questions in the GSOEP uncover various objective circumstances of an individual’s life. The
variables used here include: demographics, education levels, household income, household size,
marital and employment status, the individual’s self-rated health, whether there are children in
the household and disabilities.7
In 2005, a series of questions designed to uncover aspects of an individual’s personality
were included in the GSOEP. Self-reported personality measures generally have high levels of
reliability and validity. Previous research has shown, for example, that self-reported personality
measures are highly stable over time (McCrae & Costa, 1990) and relate to peer ratings (McCrae
& Costa, 1987). Self report measures also predict both objective behaviour (Epstein, 1979) and
occupational success (Hogan, 2005), have biological correlates (Ryff et al., 2006) and relate to
changes in objective biological functioning (O'Cleirigh, Ironson, Weiss, & Costa, 2007). Of the
31 personality questions used here, 15 are a considerably shortened version of the standard Big
7
A description of all variables and how they were constructed is contained in the Notes to Tables in the Appendix to
this chapter
64
Five personality questionnaire, and a further 16 relate to an individual’s reciprocity, locus of
control and pessimism. Before being included in GSOEP the short item Big Five scale underwent
extensive pre-testing and has been shown to satisfactorily replicate the standard Big five
questionnaire (Gerlitz & Schupp, 2005). This scale has been used in studies such as Winkelmann
and Winkelmann (2008). There are 6 questions on an individual’s reciprocity. The questions on
reciprocity can be separated into positive and negative reciprocity and examples of their use can
be seen in Dohmen et al. (2008) and Fliessbach et al. (2007). There are 9 questions that indicate
an individual’s locus of control. This construct can be traced back to the work of Rotter (1966)
and the same set of questions has been used in Fliessbach et al. (2007). There is one question that
directly asks the degree to which an individual is pessimistic about the future.
Factor analysis confirmed that the 31 personality questions grouped meaningfully into the
personality traits outlined above.8 The sole purpose of using the personality measures in this
paper, however, is to maximize the explanation of individual heterogeneity. In the main analysis
all 31 measures are therefore included as separate predictors, rather than as the grouped
personality constructs. The personality questions asked in 2005 are assumed to be reliable
proxies for personality across all years of analysis. A key assumption of panel data models is that
individual heterogeneity is fixed across time so this seems like a reasonable assumption.
However, although innate personality may be relatively stable across time an individual will not
necessarily give the same response to a given question each year. It is therefore likely that
personality measures will be prone to some measurement error. This represents a limitation to
this study but more accurate personality measures are likely to only add to the explanatory power
of individual heterogeneity. It will be interesting to determine how much these available
8
The Appendix to this chapter contains a full list of the personality questions and a description of how the measures
combine into meaningful personality constructs
65
measures contribute to the explanation of heterogeneity when compared to other likely sources of
heterogeniety, including an individual’s health and background.
The panel constructed is unbalanced. All individuals are observed in 2005 and at least one
other time point. The period under analysis is 6 years from 2000 to 2005. This time-frame is
considerably shorter than the available data in GSOEP and means that the data set is likely to
have a lower within-person variation than a longer panel. Estimation by FEVD will therefore be
falsely superior according to the conditions set out by Plumper and Troeger (2007). However, a
short data set is needed to ensure that personality measures from 2005 are adequate proxies
across the entire period under analysis and can be adequately used to attempt an understanding of
the fixed effect. To counter the potential low-within person variation problem the descriptive
statistics across a 12 year panel form GSOEP are also shown alongside the 6 year panel in Table
4.1. Across both panels most characteristics have between variations that exceed the within
variation. Many important characteristics are also observed to have within-to-between variations
exceeding 2 and this suggests that a great deal of observable information will be discarded when
using the FE model. Depending on the correlation between the unobservable heterogeneity and
any characteristics of interests, as will be empirically approximated later, estimation using FEVD
might be the preferable estimation strategy.
Concentrating on the 6 year panel used in the main analysis there are 93016 individualyear observations coming from 17210 unique individuals. For ease of interpretation in the later
analyses all the variables with intrinsically non-meaningful scales, life satisfaction, self-rated
health and all of the personality measures, are standardised with a mean of zero and standard
deviation of one.
66
Table 4.1: Summary statistics across the 6 year panel used in analysis and a longer 12 year panel (N =
93016/135486) – non-standardised
Variable:
Life Satisfaction
Mean
6 Year Panel
Standard
BetweenDeviation
to-within
Mean
12 Year Panel
Standard
BetweenDeviation
to-within
Overall
Between
Within
7.00
1.74
1.38
1.07
1.29
6.98
1.73
1.32
1.14
1.16
Overall
Between
Within
2,662
1827.7
1779.9
821.4
2.17
2,511
1644.6
1749.5
801.7
2.18
Age
Overall
Between
Within
47.45
16.09
16.24
1.65
9.84
46.06
15.87
16.15
2.87
5.63
Female
Overall
Between
Within
0.52
0.50
0.50
n/a
n/a
0.52
0.50
0.50
n/a
n/a
Education (years)
Overall
Between
Within
12.07
2.64
2.66
0.30
8.87
11.91
2.61
2.61
0.68
3.84
Household Size
Overall
Between
Within
2.76
1.28
1.21
0.41
2.95
2.83
1.30
1.18
0.56
2.11
Married
Overall
Between
Within
0.65
0.48
0.46
0.14
3.29
0.66
0.48
0.45
0.18
2.5
Separated
Overall
Between
Within
0.02
0.13
0.10
0.09
1.11
0.02
0.13
0.09
0.10
0.90
Divorced
Overall
Between
Within
0.07
0.26
0.24
0.09
2.67
0.07
0.26
0.23
0.11
2.09
Widowed
Overall
Between
Within
0.06
0.23
0.22
0.06
3.67
0.05
0.22
0.21
0.07
3
Self-Rated Health
Overall
Between
Within
2.59
0.93
0.77
0.53
1.45
2.58
0.92
0.74
0.57
1.30
Unemployed
Overall
Between
Within
0.06
0.24
0.17
0.17
1.00
0.07
0.25
0.16
0.20
0.80
Retired
Overall
Between
Within
0.22
0.42
0.38
0.14
2.71
0.20
0.40
0.36
0.17
2.18
Disabled
Overall
Between
Within
0.10
0.30
0.27
0.13
2.08
0.10
0.30
0.25
0.14
1.79
Child dummy
Overall
Between
Within
0.32
0.47
0.43
0.18
2.39
0.34
0.47
0.42
0.24
1.75
Monthly Household Income
(Euros)
67
4.6
Results
Analysis begins by showing the importance of controlling for individual heterogeneity. In
Table 4.2 standardised life satisfaction equations are estimated using pooled OLS, FE and REMT
models. All models offer some interesting insights into what makes an individual satisfied with
their life. However, there are important differences between the models. Both the FE and REMT
appropriately control for individual heterogeneity and the FE model has often been interpreted as
producing causal effects. The pooled OLS represents a mere association. For example,
concentrating on the effect size on the log of household income, the FE model has a coefficient
of 0.17, whereas the pooled OLS estimator has a coefficient that is nearly twice as large at 0.33.
The coefficient on the FE model suggests that if the individual’s household income were doubled
then their life satisfaction could be expected to increase by 0.17 standard deviations. The pooled
OLS model coefficient would suggest that, ceteris paribus, if individual x were observed to have
a household income twice the size of an individual y, then individual x would be on average
more satisfied with their life by 0.33 standard deviations. The prediction from the pooled model
is after having controlled for all other observable characteristics and presents the main drawback
of the pooled OLS model – there are important factors correlated with both income and life
satisfaction that are unobservable. Not controlling for these unobservable factors results in biased
coefficients. At the sacrifice of efficiency it is sensible to opt for an unbiased estimator such as
the FE model.
There are, however, other important differences across the pooled OLS and FE models.
Some variables do not have enough within-person variation to enable reliable estimation. In
Table 4.2 variables with zero within-person variation, such as gender, cannot be included in the
fixed effects model. Controlling for age can also be problematic. Age changes within all
68
Table 4.2: Fixed effect, REMT and pooled OLS life satisfaction regressions
(1)
Dependent Variable:
Estimation type
Independent Variables:
Year Dummies
Regional Dummies
(2)
Life Satisfaction (Standardised)
Pooled OLS
Fixed Effect
(3)
REMT
Yes
Yes
Yes
Yes
Yes
Yes
Log of Monthly Household Income (Euros)
0.331
(47.40)**
0.170
(15.94)**
0.165
(15.91)**
Age
-0.021
(16.00)**
0.272
(20.90)**
0.055
(9.62)**
0.007
(5.51)**
-0.197
(19.96)**
0.135
(12.47)**
-0.255
(10.78)**
-0.046
(3.29)**
-0.031
(1.83)
-0.414
(129.18)**
-0.423
(34.72)**
0.061
(5.29)**
-0.087
(8.52)**
0.043
(4.90)**
-0.003
(0.44)
-0.081
(4.81)**
0.041
(1.65)
-0.123
(3.46)**
0.101
(2.96)**
-0.356
(7.86)**
-0.233
(61.62)**
-0.297
(23.13)**
0.016
(1.00)
-0.087
(5.06)**
0.057
(4.32)**
-0.012
(6.31)**
0.207
(10.83)**
0.066
(6.97)**
-0.001
(0.27)
-0.083
(5.54)**
0.065
(2.75)**
-0.137
(4.01)**
0.077
(2.36)*
-0.395
(9.15)**
-0.240
(63.95)**
-0.291
(23.03)**
0.057
(4.38)**
-0.037
(2.88)**
0.052
(5.03)**
Age squared/1000
Female
Education (years)
Log of Household Size
Married
Separated
Divorced
Widowed
Self-Rated Health (Standardised)
Unemployed
Retired
Disabled
Child dummy
Mean(Log of Monthly Household Income)
0.192
(12.44)**
-0.051
(7.71)**
0.034
(1.17)
-0.206
(3.11)**
-0.138
(3.39)**
0.423
(8.01)**
-0.289
(36.62)**
-0.324
(9.72)**
Mean(Log of Household Size)
Mean(Married)
Mean(Separated)
Mean(Divorced)
Mean(Widowed)
Mean(Self-Rated Health)
Mean(Unemployed)
Constant
Observations
Number of Never Changing Person ID
R-Squared (within)
R-Squared (between)
R-Squared (overall)
-2.006
(32.46)**
93016
0.27
Absolute value of t-statistics in parenthesis * significant at 5%; ** significant at 1%
-1.054
(6.17)**
93016
17210
0.08
0.27
0.19
-2.304
(22.15)**
93016
17210
0.08
0.41
0.29
69
individuals in the same way and when included in the FE model is only interpretable as a linear
time trend. As a result when time dummies are included there is little reason to include age as an
explanatory variable in the FE model. Contrastingly, both age and gender can be included in the
pooled OLS and the REMT models. These models suggest that women are more satisfied with
life and that there is a u-shape relationship between age and life satisfaction (life satisfaction
minimises at around 39 in the pooled OLS model). More importantly a closer observation across
all the models highlights the difficulty of obtaining reliable estimates on variables that could be
termed as slow changing. Table 4.1 showed that characteristics like education and marital status
had very low within-person variations. In both the REMT and FE models in Table 4.2 the
coefficient on education is indeterminable. The coefficients on marital status, on the other hand,
vary considerably and in conflicting directions from the estimates given in the pooled OLS
model. This leaves some concern over the reliability of the coefficients using both FE and REMT
estimations on variables with low within-person variations.
The FE model in Table 4.2 is used to estimate a fixed effect residual for each individual.
Observable characteristics can then be used to decompose the fixed effect residual in Table 4.3.
Column 1 begins by including only observable demographic characteristics; these variables
explain 7% of the variation. Column 2 extends the model by further including what could be
considered as very slow moving variables; marital status, education and whether the individual is
retired. Adding these variables increases the explanation of the fixed effect residual to 10%.
Column 3 indicates how much personality contributes to an explanation of the fixed effect
residual. When grouped into their 9 meaningful constructs the personality measures collectively
explain 18% of the fixed effect residual, with most of this explanation coming from an
individual’s level of pessimism, locus of control and neuroticism. The explanation of the fixed
70
Table 4.3: Predicting the fixed effects residual (from column 2 of Table 4.2) using the mean levels of various
objective characteristics and personality variables
(1)
Dependent Variable:
Independent Variables (mean levels):
Year Dummies
Regional dummies
(2)
(3)
(4)
(5)
Fixed Effect Residual (from column 2 of table 1)
No
Yes
No
Yes
No
No
No
Yes
No
Yes
-0.027
(11.16)**
0.326
(12.68)**
0.043
(4.28)**
0.035
(17.40)**
-0.015
(2.82)**
0.072
(3.71)**
-0.291
(4.85)**
-0.213
(7.91)**
0.338
(10.63)**
-0.016
(7.42)**
0.234
(10.12)**
0.065
(6.60)**
0.016
(8.29)**
-0.006
(1.38)
0.060
(3.45)**
-0.263
(4.92)**
-0.229
(9.49)**
0.307
(10.80)**
0.144
(12.18)**
-0.010
(4.79)**
0.183
(8.05)**
0.072
(7.51)**
0.001
(0.30)
-0.022
(3.73)**
0.069
(3.87)**
-0.199
(3.78)**
-0.178
(7.35)**
0.360
(12.53)**
-0.195
(27.08)**
-0.029
(1.09)
0.015
(0.63)
0.101
(4.23)**
0.106
(5.36)**
-0.051
(3.09)**
-0.003
(0.57)
0.014
(2.80)**
0.002
(0.38)
-0.003
(0.49)
0.012
(2.33)*
0.004
(0.75)
0.005
(0.99)
0.008
(1.49)
0.021
(3.69)**
0.002
(0.39)
0.011
(2.27)*
0.016
(2.94)**
0.011
(1.84)
0.026
(4.40)**
0.024
(4.75)**
0.008
(1.36)
0.021
(3.72)**
0.019
(3.89)**
Log of Monthly Household Income (Euros)
Age
Age squared/1000
Female
Education (years)
Log of Household Size
Married
Separated
Divorced
Widowed
-0.023
(13.21)**
0.309
(16.91)**
0.041
(4.03)**
Self-Rated Health (Standardised)
Unemployed
Retired
Disabled
Child dummy
Standardised Personality Variables:
Constructs of the “Big Five”
Openness
0.019
(3.59)**
Original
Values artistic experiences
Active imagination
Conscientiousness
-0.002
(0.45)
Thorough worker
Lazy
Effective and efficient
Extrovert
Communicative
Sociable
Reserved
Table 4.3 continues on the next page
0.021
(3.75)**
71
Table 4.3 continued
Agreeable
0.045
(8.20)**
Rude to others
0.002
(0.38)
0.028
(5.62)**
0.015
(2.74)**
0.002
(0.34)
0.027
(5.50)**
0.019
(3.52)**
-0.078
(14.99)**
0.012
(2.31)*
0.034
(6.31)**
-0.050
(9.65)**
0.012
(2.26)*
0.017
(3.28)**
0.077
(15.22)**
0.007
(1.42)
0.028
(6.09)**
-0.054
(10.37)**
-0.020
(4.05)**
-0.017
(3.18)**
-0.016
(3.27)**
0.006
(1.34)
-0.044
(8.20)**
0.069
(14.00)**
0.011
(2.25)*
0.023
(5.20)**
-0.048
(9.56)**
-0.019
(3.93)**
-0.018
(3.52)**
-0.012
(2.52)*
0.008
(1.68)
-0.036
(6.77)**
-0.149
(28.74)**
-0.150
(29.82)**
-0.127
(25.62)**
0.004
(0.79)
17210
0.18
-0.071
(0.96)
17210
0.29
-1.136
(11.13)**
17210
0.33
Forgiving nature
Considerate
-0.053
(10.24)**
Neuroticism
Worries a lot
Nervous
Deals well with stress
0.140
(25.25)**
Locus of control
Control over life
Belief in luck
Influencing social conditions
Others control their life
Success comes from hard work
Doubts own abilities
Opportunities depend on social conditions
Ability is more important than effort
Little control in life
Pessimism
Constant
Observations
R-squared
0.444
(6.03)**
17210
0.07
0.144
(1.76)
17210
0.10
Absolute value of t-statistics in parenthesis * significant at 5%; ** significant at 1%
72
effect residual rises to 20% when personality is included as 31 separate scores. Maximising the
explanation of the fixed effect residual is most important here so all 31 measures are used in the
subsequent analysis. In column 4 these personality variables are appended to the explanatory
variables considered to be slow moving. The overall explanation of the fixed effect residual rises
to 29%. Finally column 5 includes the between-person information about the household’s
income, whether there are children in the house and an individual’s health and disabilities. The
explanation rises to 33%.
Table 4.3 illustrates that personality provides the greatest explanation of individual
heterogeneity when compared to other observable characteristics. Pessimism, locus of control,
agreeableness and neuroticism are particularly important components of individual
heterogeneity. The individual’s health is also observed to be an important component. This
suggests that individual heterogeneity is mostly, although not exclusively, personality. Since a
large proportion of individual heterogeneity is in fact observable it may be sensible to consider
alternative estimation strategies. As has been discussed, many of the favoured models are based
on an assumed correlation between unobservable heterogeneity and the observable
characteristics.
The fixed effect residual can also be used to empirically approximate the likely
correlation between individual heterogeneity and other observable characteristics. The first
column in Table 4.4 shows that there is only low to moderate correlation between observable
characteristics and the fixed effect residual. All variables are below 0.2. The second column
controls for demographic characteristics with the correlation rising across many characteristics.
However, as personality and other characteristics with very low within-person variation are
included in Columns 3 and 4 respectively the correlations substantial reduce. As Plumper and
73
Troeger (2007) show when an observable characteristic is correlated with the unobservable
component of individual heterogeneity by just 0.3 the between-to-within person ratio only needs
to be 1.7 to make estimation by FEVD superior to the FE model. Focusing specifically on
income, the correlation is just 0.09 and has a between-to-within ratio of 2.17, suggesting that
estimation may be preferable using FEVD.
Table 4.4: Correlations between observable characteristics and the unobservable component of the fixed
effect residual errors
Fixed Effect Residual
controlling for:
Error from predicting the fixed
effect residual in:
Log of Monthly Household
Income (Euros)
Age
Female
Education (years)
Log of Household Size
Married
Separated
Divorced
Widowed
Self-Rated Health (Standardised)
Unemployed
Retired
Disabled
Child dummy
(1)
No
controls
(2)
Demographics
(3)
Demographics
and personality
(4)
Demographics, personality and
characteristics with low withinperson variation
Table 4.3
Column 2
0.10**
-0.01**
0.00
0.05**
0.02**
0.05**
-0.05**
-0.12**
0.07**
0.10**
-0.06**
-0.00
-0.04**
-0.01
0.09**
-0.01
0.00
0.00
0.03**
0.03**
-0.03**
-0.02**
-0.01
0.09**
-0.05**
-0.00
-0.04**
-0.01
Table 4.3
Column 1
0.13**
0.14**
0.03**
0.08**
-0.07**
0.04**
-0.05**
-0.11**
0.15**
0.20**
-0.13**
0.13**
-0.03**
-0.08**
0.16**
-0.01*
-0.00
0.12**
0.01**
0.04**
-0.05**
-0.10**
0.07**
0.26**
-0.10**
-0.01**
-0.08**
-0.01**
* significant at 5%; ** significant at 1%
The estimations carried out in Table 4.3 represent the second stage of the FEVD
technique. Table 4.5 therefore displays the third stage FEVD results by including the error terms
from Table 4.3 as explanatory variables in a pooled OLS estimation. In column 1 the error term
from the second column in Table 4.3, where only demographic and slow moving variables were
used to predict the fixed effect residual in the second stage, is used in estimation. Although
efficiency of estimation has been increased it is important to note that the coefficients on
variables not used in stage two are similar to those using the standard FE model seen in Table
74
Table 4.5: Introducing personality into life satisfaction regressions using the fixed effect vector decomposition
technique (3rd stage) and the random effects model
(1)
Dependent variable:
Independent variables:
Year dummies
Regional dummies
31 Personality Variables
Log of Monthly Household Income (Euros)
Age
Age squared/1000
Female
Education (years)
Log of Household Size
Married
Separated
Divorced
Widowed
Self-Rated Health (Standardised)
Unemployed
Retired
Disabled
Child dummy
Error from predicting the
fixed effect residual in:
Constant
(2)
(3)
Life Satisfaction (Standardised)
(4)
Yes
Yes
No
Yes
Yes
Yes
Yes
Yes
No
Yes
Yes
Yes
0.178
(36.53)**
-0.025
(28.17)**
0.305
(33.78)**
0.044
(11.09)**
0.030
(34.98)**
-0.109
(15.94)**
0.100
(13.30)**
-0.270
(16.46)**
-0.089
(9.15)**
-0.069
(5.95)**
-0.235
(102.62)**
-0.296
(35.00)**
-0.001
(0.19)
-0.091
(12.87)**
0.058
(9.62)**
Table 4.3 Column
2
0.999
(316.84)**
-0.968
(22.53)**
93016
0.288
(57.76)**
-0.015
(16.06)**
0.221
(24.04)**
0.067
(15.31)**
0.001
(1.65)
-0.129
(18.61)**
0.101
(13.24)**
-0.223
(13.43)**
-0.064
(6.49)**
-0.060
(5.07)**
-0.332
(141.91)**
-0.295
(34.42)**
0.059
(7.38)**
-0.070
(9.69)**
0.019
(3.13)**
Table 4.3 Column
5
1.001
(283.39)**
-2.220
(42.41)**
93016
0.276
(33.41)**
-0.021
(11.41)**
0.274
(14.59)**
0.051
(5.40)**
0.016
(8.66)**
-0.148
(12.13)**
0.098
(6.55)**
-0.198
(7.63)**
-0.029
(1.49)
-0.110
(4.50)**
-0.308
(92.25)**
-0.345
(29.12)**
0.026
(1.99)*
-0.134
(10.51)**
0.043
(4.13)**
0.238
(29.45)**
-0.015
(8.33)**
0.221
(12.42)**
0.066
(6.89)**
0.006
(3.14)**
-0.109
(9.24)**
0.094
(6.65)**
-0.203
(8.06)**
-0.045
(2.39)*
-0.118
(5.10)**
-0.279
(83.99)**
-0.326
(27.99)**
0.031
(2.44)*
-0.102
(8.29)**
0.032
(3.20)**
-1.681
(20.03)**
93016
0.08
0.40
0.26
-1.852
(19.40)**
93016
0.08
0.48
0.33
Observations
R-Squared (within)
R-Squared (between)
R-Squared (overall)
0.65
0.64
Absolute value of t-statistics in parenthesis * significant at 5%; ** significant at 1%
75
4.2. On the other characteristics used in stage 2, however, some interesting changes occur. The
estimation in the first column of Table 4.5 suggests that education has a positive effect on life
satisfaction, contrasting the negative and insignificant coefficient seen in the FE and REMT
models in Table 4.2.
Here, it is useful to comment on the coefficients on marital status in the third stage of the
FEVD when compared with the FE model in Table 4.2. The change in the coefficients across the
models is not consistent. Married has a coefficient that is nearly three times larger in the FEVD
model. In contrast, the coefficient on widowhood is at least three times smaller using FEVD, and
the variable divorced reverts from a positive to a negative coefficient. These changes across
marital status are inconsistent and this highlights an important issue with the FE model. When
comparing the FE and FEVD models it is important to consider the difference between a change
in circumstance and maintaining a permanent situation. In marital status, for example, only the
changes to the individual’s marital status are important in the FE model. Thus, if there is no
change to the individual’s marital status then the individual will yield no useful information for
the FE estimation. However, that same individual may have greater life satisfaction simply from
sustaining a particular marital situation for a considerable period of time. For example, it is likely
that getting married will increase the individual’s life satisfaction but additionally there are also
likely to be well-being benefits from sustaining a healthy marriage for an extended period of
time. Similarly, getting divorced may bring immediate life satisfaction benefits as seems to be
supported in work by Gardner and Oswald (2006). However, if the individual were to remain in a
divorced state for a sustained period then it would seem plausible that there could eventually be
adverse consequences for their well-being. The coefficients on widowed invite a similar
argument – becoming a widow initially has a large negative impact on the individual’s life. Over
76
time, however, the impact reduces as the individual adapts to their loss, supporting conclusions in
both Clark et al (2008) and Gardner and Oswald (2006). The issue associated with marital status
generalises to other variables and highlights a practical concern with the FE model.
Column 2 of Table 4.5 shows a further estimation using FEVD using the error term from
column 5 of Table 4.3. Since the use of personality measures substantially reduce the correlation
between unobservable heterogeneity and important characteristics it enables many of the
observable variables to be classified as slow moving enough to be preferably estimated using
FEVD. The coefficients are different to the pooled OLS, FE and REMT models seen in Table 4.2
and, given the core model assumptions, possibly reflect more accurate estimates. Another way of
using personality variables, given that the correlation with unobservable heterogeneity and
observable characteristics is substantially reduced, could be by using a standard RE model.
Columns 3 and 4 show the results from an RE estimation both without and with personality
variables respectively. The coefficients are fairly similar to the FEVD coefficients with the
personality variables attenuating many of the coefficients in comparison to the pooled OLS
model. The FEVD, however, is the preferred model since it satisfies the specific conditions set
out by Plumper and Troeger (2007).
The pooled OLS model, by not appropriately dealing with unobservable correlated
factors, has a tendency to produce biased coefficients. The FE model, on the other hand, discards
all between-person information and without a true understanding of individual heterogeneity
underestimates the effect on life satisfaction of various individual characteristics. The FEVD by
offering an alternative way to deal with individual heterogeneity combines elements of both
techniques to enable efficient yet unbiased estimates. The results from column 2 in Table 4.5
suggest that doubling the individual’s household income will actually increase their life
77
satisfaction by 0.29 standard deviations. The discrepancy with the FE model arises due to the
well-being benefits that come about from having a permanently high level of household income
as well as increases to household income. Additionally the FEVD provides more reliable
coefficients on the effect of age, education, marital status, disabilities and having children.
4.7
Conclusion
This paper attempts to understand individual heterogeneity, which has been shown to
substantially attenuate estimates of effects when moving from a pooled OLS to an FE model in
subjective well-being studies. Here, personality measures are used to increase the understanding
of individual heterogeneity and help confirm that personality is one of the main components of
individual heterogeneity. Health and other demographic characteristics also provide some
explanation of heterogeneity. A greater understanding of individual heterogeneity reduces the
correlation between the remaining unobservable heterogeneity. Reducing this correlation is key,
since it allows alternative techniques to be explored, and enables more reliable estimates on
variables that have low within-person variations, such as income, education, marital status,
disabilities and having children.
The use of a FEVD model with personality variables has a tendency to produce estimates
that lie someway between estimates on the pooled OLS and FE estimations. For example, using
the FEVD the individual’s household income is estimated to be more than 1.6 times more
beneficial for the individual’s well-being than is suggested by the FE and REMT models, but
around 0.85 that of a pooled OLS model. One potential explanation is that there are still other
important unobservable components not controlled for using the FEVD. However, another reason
could be that the FE and REMT models, by making strong assumptions about the correlation
with the unobservable heterogeneity, are simply too restrictive. Specifically, the FE model leaves
78
no room for uncovering improvements to the individual’s subjective well-being that may simply
arise, for example, from having a permanently high income or being in a permanently healthy
relationship. Only focusing on changes detracts from the benefits to well-being that may accrue
from sustaining a high level or state.
The observable between-person information is shown to predict 13% of the fixed effect
residual. This decomposition of the fixed effect residual suggests that the fixed effect residual
should not be completely termed as unobservable individual heterogeneity and simply
disregarded. It contains valuable observable information. The fixed effect residual is potentially
an untapped source in providing answers as to why some individuals have higher subjective wellbeing than others. The personality measures used here alone explain 20% of the fixed effect
residual. Compared to the explanation given by other characteristics, such as health and other
demographic characteristics, this contribution is large. However, there still remains a substantial
unexplained component. There are three possible explanations. The unexplained component
could be due individual heterogeneity that is still largely unknown, for example an individual’s
ability. Alternatively, the measures used, particularly the one-item life satisfaction scale, are
imperfect and are likely to be measured with some error. Lastly, the FE model, focusing on only
changes as discussed earlier, may underestimate the importance of permanent state effects. It is
likely that there is some combination of the three but simply terming the individual heterogeneity
simply as personality traits and discarding the information seems inappropriate.
An important consideration for the future is the availability of personality measures in the
large data sets commonly used by economists. Currently many representative national surveys
like the GSOEP do not include questions on an individual’s personality. Such unavailability may
prove problematic for the approach outlined in this paper. Economists are relatively unfamiliar
79
with the idea that personality can be measured and this has no doubt influenced the demand for
inclusion of such measures in their data sets – this needs to change. Personality has already been
shown to be an important determinant of wages (Bowles et al., 2001a) but further work is needed
around this area. More generally Borghans et al. (2008) have convincingly argued that economic
research has much to gain from using reliable and valid personality measures that are used
extensively by psychologists. Personality appears to be one of the biggest and most consistent
predictors of well-being (Diener & Lucas, 1999) and as shown here is important for
understanding both fixed and variable components of well-being. This paper adds to the support
for the increased use of personality measures in subjective well-being research (Anand et al.,
2009). A wider inclusion of personality measures in data sets like GSOEP would be of enormous
benefit to both personality and economic research.
This paper has gone some way in understanding the important fixed effect and shown the
importance of exploiting between person information. The use of personality in this context is
novel and may allow researchers to relax the statistical technique used to estimate subjective
well-being equations. The use of personality measures combined with the FEVD technique may
therefore provide an important methodological advance for subjective well-being research.
80
4.8
Appendix
4.8.1 Note to Tables
Variable
Description
Life Satisfaction
A self reported measure of how satisfied the individual is with their
life, all things considered, where 0=completely dissatisfied and
10=completely satisfied
Monthly Household Income
The household’s income in which the individual resides
(Euros)
Age
Individual’s age
Female
Individual is female (excluded dummy: male)
Education (years)
Number of years of education
Household Size
The number of members in the individual’s household
Married, Separated, Divorced,
Individual is married, separated, divorced or widowed (excluded
Widowed
dummy: single)
Self-Rated Health
Individuals are asked to give a self rating of their current health
where 1=Very Good, 2=Good, 3=Satisfactory, 4=Poor and 5=Bad
Unemployed
Individual is unemployed (excluded dummies: any other responses
to occupation position except retired)
Retired
Individual is retired (excluded dummies: any other responses to
occupation position except unemployed)
Disabled
Disability status of the individual
Child dummy
Whether there is at least one child in the household (excluded
dummy: no children in the household)
Personality variables
31 personality variables that measured 8 underlying constructs;
openness-to-experience, conscientiousness, extroversion,
agreeableness, neuroticism, individual autonomy, social
responsibility and pessimism. See section 4.8.2 for a full description
of the personality questions and how questions grouped into
constructs.
81
4.8.2 Personality Variables in GSOEP
In the questionnaire section entitled “What kind of personality do you have?” individuals
are asked 30 questions. 15 of these relate to the “Big five” personality inventory, whilst a further
15 cover aspects of the individual’s reciprocity and control in life. A further question, on
pessimism, comes from the “Attitudes and opinions” section. Using factor analysis the measures
were found to load onto 9 different personality constructs: openness to experience,
conscientiousness, extroversion, agreeableness, neuroticism, positive and negative reciprocity,
locus of control and pessimism.
4.8.2.1 Big Five Personality Inventory
Individuals are asked whether they see themselves as someone who…
1. …does a thorough job
2. …is communicative, talkative
3. …is sometimes somewhat rude to others
4. …is original, comes up with new ideas
5. …worries a lot
6. …has a forgiving nature
7. …tends to be lazy
8. …is outgoing, sociable
9. …values artistic experiences
10. …gets nervous easily
11. …does things effectively and efficiently
12. …is reserved
13. …is considerate and kind to others
82
14. …has an active imagination
15. …is relaxed, handles stress well
Individuals are asked whether the statement applies to them on a 1 to 7 scale, with 1 meaning the
statement does not apply to them at all and 7 that it applies perfectly. These 15 variables load
onto five personality dimensions: Openness to experience, conscientiousness, extroversion,
agreeableness and neuroticism. Questions 4, 9 and 14 relate to an individuals openness to
experience; questions 1, 7 & 11 relate to conscientiousness, questions 2, 8 & 12 relate to
extroversion; questions 3, 6 & 13 relate to agreeableness; questions 5, 10 & 15 relate to
neuroticism. These groups of questions can be reverse coded (as appropriate) and combined to
give an underlying score of the personality dimension.
4.8.2.2 Positive and Negative Reciprocity
Individuals are asked to what extent the following apply to them
16. If someone does me a favour, I am prepared to return it
17. If I suffer a serious wrong, I will take revenge as soon as possible, no matter what the
cost
18. If somebody puts me in a difficult position, I will do the same to him/her
19. I go out of my way to help somebody who has been kind to me before
20. If somebody offends me, I will offend him/her back
21. I am ready to undergo personal costs to help somebody who helped me before
Individuals are asked whether the statement applies to them on a 1 to 7 scale, with 1 meaning the
statement does not apply to them at all and 7 that it applies perfectly. Questions 16, 19 & 21 load
83
onto a construct termed positive reciprocity and questions 17, 18 & 20 load onto the individual’s
negative reciprocity.
4.8.2.3 Locus of Control
Individuals are asked their attitudes towards their life and future.
22. How my life goes depends on me
23. What a person achieves in life is above all a question of fate or luck
24. If a person is socially or politically active, he/she can have an effect on social conditions
25. I frequently have the experience that other people have a controlling influence over my
life
26. One has to work hard in order to succeed
27. If I run up against difficulties in life, I often doubt my own abilities
28. The opportunities that I have in life are determined by the social conditions
29. Inborn abilities are more important than any efforts one can make
30. I have little control over the things that happen in my life
Individuals are asked whether they agree with the statements on a 1 to 7 scale, with 1
representing complete disagreement and 7 that they completely agree. These measures reflect an
individual’s locus of control and factor analysis show that questions 22, 25, 27 & 30 can be
grouped together to give an indication of this trait.
4.8.2.4 Pessimism
This is a one item scale. Individuals are asked
31. When you think about the future, are you…optimistic, more optimistic than pessimistic,
more pessimistic than optimistic, pessimistic?
84
This variable is treated as cardinal.
85
CHAPTER 5
5
WHICH PERSONALITY TYPES HAVE THE HIGHEST MARGINAL UTILITIES OF INCOME?
5.1
Abstract
Economics implicitly assumes that the marginal utility of income is independent of an
individual’s personality. We show that this is wrong. This is the first demonstration that there are
strong personality-income interactions. Individuals who are conscientious, have pessimistic
tendencies or feel they have low control over their life get substantially higher marginal utility
from their income. Our findings are highly robust and have important implications for the use of
financial incentives to influence behaviour. In the future, public policy may benefit from being
personality-specific.
Under review at Journal of Economic Behavior and Organization
86
5.2
Introduction
Will more money improve an individual’s satisfaction with life, and if so, by how much?
The use of subjective well-being data has helped researchers evaluate the role of income in an
individual’s life. For example, it has been shown that there are large well-being differences
between low and high income earners (Lucas & Schimmack, 2009) and that an exogenous
increase to an individual’s income can raise their well-being (Frijters et al., 2004; Gardner &
Oswald, 2007). Researchers have also revealed that individuals are mainly concerned with their
income relative to others (Ferrer-i-Carbonell, 2005; Luttmer, 2005) and that this relative income
effect is thought to explain why economic growth in developed countries has not always
increased national well-being (Blanchflower & Oswald, 2004; Easterlin, 1995).
The literature on income and well-being is extensive (Clark, Frijters et al., 2008; Howell
& Howell, 2008) but the relationship is far from fully understood. Current research into income
and well-being almost always focuses on average effects across a sample (for example, Layard et
al. (2008) estimate the average elasticity of income across various samples). Researchers have
shown, however, that the benefit from income can vary according to an individual’s health
(Finkelstein, Luttmer, & Notowidigdo, 2008; D. M. Smith, Langa, Kabeto, & Ubel, 2005).
Individuals are also likely to have heterogeneous preferences (Barsky, Juster, Kimball, &
Shapiro, 1997), yet very little is known about how the marginal utility of income might vary
across a population. How an individual spends their money can be important for well-being and
recent research has shown, for example, that engaging in pro-social spending has a strong
positive effect on well-being (Dunn, Aknin, & Norton, 2008). Such a finding could indicate that
individuals with particular types of preferences could extract more utility from a given increase
to their income. Some researchers have suggested that the role of emotions are hugely
87
understated in economic theory, even though emotions are likely to influence an individual’s
enjoyment of particular economic activities (Elster, 1998; Loewenstein, 2000). The experience of
emotions habitually is closely linked to an individual’s personality (Revelle & Scherer, 2008),
hence it is likely that an individual’s marginal utility of income could be dependent upon their
personality.
Personality measures are used extensively in psychology (Pervin & John, 1999) and selfreported personality judgments have impressive levels of reliability and validity. For example,
self-reported traits are highly stable over time (McCrae & Costa, 1990), are related to peer
ratings (McCrae & Costa, 1987), predict objective behaviour (Epstein, 1979) and occupational
success (Hogan, 2005), have biological correlates (Ryff et al., 2006), and prospectively predict
changes in objective biological functioning over time (O'Cleirigh et al., 2007). Such findings
have led to personality psychology being studied and applied in many contexts, including health,
clinical, psychiatric, educational, and occupational settings.
Measures of personality enable a categorization of people and their behaviours but,
mostly due to a lack of familiarity, such measures have not yet been fully integrated into
economic research (Borghans et al., 2008). In relation to well-being research, it is fairly clear that
personality is one of the biggest and most consistent predictors of well-being (Diener & Lucas,
1999). Some authors estimate that between 44% and 52% of the variation in well-being is
attributable to individual differences (Lykken & Tellegen, 1996). Economists will also be
familiar with the importance of controlling for individual heterogeneity when trying to determine
the causal effects of income on well-being (Ferrer-i-Carbonell & Frijters, 2004). We are
concerned, however, that aspects of individual heterogeneity may interact with an individual’s
income. For example, the relationship between a change in income and well-being may be
88
dependent on an individual’s personality type. To test this hypothesis we use a well-known
longitudinal data set that recently included standard psychological measures of personality to
determine whether there are any systematic personality differences between those that gain more
utility from their income than others and those that get less.
Theoretically the case for the use of personality measures in economics seems strong.
Borghans et al. (2008) have argued that personality should be given greater consideration when
discussing economic parameters and constraints. They suggest that there could be considerable
benefit to understanding how economic incentives might influence individuals with different
personality traits. From a psychologist’s perspective personality research has a long history (see
Winter & Barenbaum, 1999) and has developed into a systematic understanding of individual
differences. Nevertheless, it is relatively uncommon to find empirical studies that use personality
measures within economics. This is beginning to change; with a number of recent studies
investigating an area of economic importance – the determination of an individual’s wages.
Mueller & Plug (2006), Nyhus & Pons (2005), Groves (2005) and Semykina & Linz (2007) have
all used personality measures to predict an individual’s wages. Mueller & Plug (2006) show that
some personality traits, such as openness and conscientiousness, are rewarded in the market
place, whereas other traits, such as agreeableness and neuroticism, are penalized. Nyhus & Pons
(2005) draw similar conclusions but also find that the degree of autonomy an individual has is
also important. They further find that the financial return to personality varies across educational
groups. Groves (2005) investigates the importance of psychological traits, such as autonomy,
social withdrawal and aggression in female earnings. Semykina & Linz (2007) find that
personality traits explain as much as 8% of the gender wage gap.
89
These types of empirical study may help explain why, after controlling for many factors,
including the improved cognitive abilities that come through schooling, there are still large
earning gaps. Although the use of personality traits in the determination of wages is very much in
its infancy (Bowles et al., 2001b), the findings indicate that personality is an important
determinant. Bowles et al. (2001a) have suggested that both school and family pass on many
important behavioural traits that enhance the individual’s earning success. Other very recent
empirical contributions have assessed personality’s relation to performance in ultimatum games
(Schmitt, Shupp, Swope, & Mayer, 2008; Swope, Cadigan, Schmitt, & Shupp, 2008), the
propensity for an individual to share knowledge with work colleagues (Matzler et al., 2008) and
job matching (Winkelmann & Winkelmann, 2008). Researchers have also shown the importance
of conscientiousness and self control in the individual’s accumulation of wealth (Ameriks,
Caplin, & Leahy, 2003; Ameriks, Caplin, Leahy, & Tyler, 2007).
In income and well-being research personality measures have rarely been used. Due to
important policy consequences researchers are concerned with determining causal effects of
income on well-being. Hence, researchers are mostly concerned with controlling for personality
– not its independent effect. It is argued that personality is most convincingly controlled for by
using panel data and trying to explain the within-person variation in subjective well-being
(Ferrer-i-Carbonell & Frijters, 2004). Personality, being largely thought of as fixed (Costa &
McCrae, 1980, 1988; Srivastava, John, Gosling, & Potter, 2003), is considered to offer no
explanation to the within-person variation in subjective well-being. Within this statistical
framework personality measures are, therefore, not directly needed. However, if personality were
thought to interact with income then personality measures would aid an investigation. Here, we
use personality measures to show that there are substantial income-personality interaction effects.
90
Individuals with high levels of conscientiousness, pessimism or lack of control seem to gain
more utility from income than others. Our results stand up to a number of alternative
explanations. Such a finding poses new questions on the links between income and well-being
and may have important implications for the use of financial incentives to influence behaviour.
In the future, public policy may benefit from being personality-specific in a similar way as has
been suggested for gender (Alesina & Ichino, 2007).
This paper is structured as follows. Section 2 details the methodology, section 3 describes
the data, section 4 discusses the results, including robustness tests, and section 5 concludes.
5.3
Methodology
The standard approach within economics to determine causal effects of income on
subjective well-being (SWB) is the fixed effects estimator. A fixed effect analysis is easily
performed by observing multiple individuals across several time-points.
(5.1) SWBit    Dit   log yit  k it  i   it
The subjective well-being of a given individual, i, at a given time period, t, is dependent
upon a number of factors other than income; specific regional and time period factors, D, a series
of observable time varying characteristics, X, and individual heterogeneity that, although varying
across individuals, does not vary across time, μ. A causal effect of income can only be obtained
provided all these correlated factors are controlled for. Heterogeneous factors, although often
unavailable, immeasurable or simply unknown, are captured by the parameter μ. Assuming that
the factors contained within μ have zero within-person variation then any changes to an
individual’s SWB must have arisen from changes to the individual’s circumstances.
91
It is fairly common for researchers to assume that individual heterogeneity is mostly
personality (for explicit illustrations of this assumption see Booth & van Ours (2008), Ferrer-iCarbonell & Frijters (2004), Frijters, Haisken-DeNew & Shields (2004), Senik (2004) and
Vendrik & Woltjer (2007)). However, although personality measures may be available, the fixed
effect estimator may still be the best way to control for individual heterogeneity, which may
include much more than simply personality. Our estimation strategy is therefore based on the
premise that the measures of personality, P, are a subset of μ. A fixed effect estimator is therefore
used on equation 5.2 to determine whether the well-being effects from a change to an
individual’s income is dependent upon a vector of personality characteristics, P.
(5.2) SWBit    Dit   log yit   Pi  log yit  k it  i   it
Such an estimation strategy must assume that the vector of personality measures, P, is
appropriately controlled for using the fixed effect estimator which eliminates μ. Given the
widespread use of the fixed effect analysis to control for personality factors, and individual
heterogeniety more generally, this assumption seems appropriate. It is possible that this
assumption is too strong, but later we present an alternative estimation strategy that relaxes this
assumption. Our interest in the main analysis therefore lies simply with whether the personality
measures P, a subset of μ and therefore already controlled for, interact with income. A well
determined coefficient on any of the personality-income interaction terms would signify that the
degree to which an individual benefits from income is dependent upon personality.
92
5.4
Data
It is relatively unusual to find a representative longitudinal data set typically used in
economic analysis that contains reliable personality measures frequently used by psychologists.
The lack of availability of such measures has probably not helped personality’s integration into
economic research. However, in a recent wave of the German Socio-Economic Panel (GSOEP) a
set of personality questions were asked. These included questions that related to the Big Five
model of personality. The Big Five model suggests that there are five overarching dimensions to
personality, that of an individual’s openness-to-experience, conscientiousness, extroversion,
agreeableness and neuroticism. The hierarchical organization of the Big Five model suggests that
there are also lower order personality facets (Wood, Joseph, & Maltby, 2008). Such facets can
include an individual’s level of autonomy over their life, and pessimism, which were also
measured in the GSOEP. Factor analysis confirmed that autonomy and pessimism had unique
variation from the Big Five9. Such self-reported personality judgments have impressive levels of
reliability and validity. For example, self-reported traits are highly stable over time (McCrae &
Costa, 1990), are related to peer ratings (McCrae & Costa, 1987), predict objective behaviour
(Epstein, 1979) and occupational success (Hogan, 2005), have biological correlates (Ryff et al.,
2006), and prospectively predict changes in objective biological functioning over time
(O'Cleirigh et al., 2007). The personality variables contained in the GSOEP have also been used
by Winkelmann and Winkelmann (2008) to investigate job matching. Here, to aid an
interpretation of the results the personality scores used are standardized.
The personality variables were asked only in 2005. Personality is generally regarded as
fixed across time (Costa & McCrae, 1980, 1988, Srivastava et al., 2003) so we assume that these
9
A full description of the personality questions and how the personality dimensions were constructed using factor
analysis can be found in the Appendix to this chapter
93
personality measures can be used as an acceptable proxy for personality across all years of
analysis. Although an individual’s innate personality may be fixed across time it is possible that
an individual would not have answered the same every single year. For example, an individual’s
response to a given personality question could be highly dependent on an individual’s
circumstances at the time of questioning. Answers across years would be expected to correlate
but this potential problem may lead to the conclusion that there are interaction effects when there
are none. A robustness test is offered later to counter this possibility.
In all other respects the GSOEP is a representative longitudinal sample of German
households. The survey has been used in a number of important subjective well-being studies
(for example Clark, Diener et al., 2008; Ferrer-i-Carbonell, 2005; Ferrer-i-Carbonell & Frijters,
2004; Frijters et al., 2004) and alongside the standard objective characteristics10 contains a single
item life satisfaction question:
How satisfied are you with your life, all things considered?
Individuals are asked to respond to this question on an 11-point scale, where 0 indicates complete
dissatisfaction and 10 indicates complete satisfaction. Since it has been shown that there is little
difference between estimating effects using cardinal or ordinal models (Ferrer-i-Carbonell &
Frijters, 2004) the life satisfaction measure is treated as cardinal. We use the household income
per month as the income variable but we include within our standard set of controls the size of
the individual’s household.
10
A full description of these variables is contained in the Notes to Tables in the Appendix to this chapter
94
The panel used to determine whether there is an income-personality interaction effect on
well-being is unbalanced and limited to just six years, from 2000 to 2005. The only requirement
Table 5.1: Summary statistics (N = 93256) – non-standardized
Variable:
Mean
Standard
Deviation
Life Satisfaction (non-standardized)
Monthly Household Income (Euros)
Age
Female
Education (years)
Household Size
Married
Separated
Divorced
Widowed
Unemployed
Retired
Disabled
Child dummy
7.00
2,662
47.45
0.52
12.07
2.76
0.65
0.02
0.07
0.06
0.06
0.22
0.10
0.32
1.74
1827.4
16.08
0.50
2.64
1.28
0.48
0.13
0.26
0.23
0.24
0.42
0.30
0.47
we make is that individuals answer the personality questions in 2005 and also completed the
questionnaire in at least one other time point. The chosen time frame is considerably shorter than
the available data in the German panel but this is necessary to ensure that the personality
measures answered in 2005 are realistic proxies across the entire period under analysis. Although
there is no reason to expect innate personality to change the use of a large time lag has the
potential to be problematic and is not necessary for the analysis. The 6 year unbalanced panel
used in the main analysis contains 17241 individuals, producing 93256 individual time-point
observations with the descriptive statistics shown in Table 5.1. Life satisfaction is presented in its
raw form but for the main analysis life satisfaction scores have been standardized to give a
meaningful interpretation.
95
5.5
Results
The analysis begins in Table 5.2 by estimating the average effect of income on a
standardized life satisfaction variable. The pooled OLS and fixed effect models show that income
has a positive effect on individual well-being. As is typical the coefficients attached to the pooled
model in column 1 are much larger than those of the fixed effect model in column 2. The
difference reflects the importance of controlling for heterogeneous factors between individuals.
In the pooled OLS model there are factors that cannot be controlled for that drive an individual to
be both more satisfied and earn higher levels of income. The fixed effect model, on the other
hand, by focusing only on the changes that occur within individuals, successfully controls for
such factors. Each individual, once controlling for all other changes to their life, will have a
unique slope that represents how changes to their income across the panel related to changes in
their life satisfaction. The fixed effect estimates represent the average of all these individual
slopes and could be interpreted as the average causal effect on individual well-being. From a
practical perspective there would naturally be more interest in the results from the fixed effect
model, helping the understanding of how an increase to an individual’s income might raise wellbeing. There would be far less concern for the cross-sectional association reflected in the pooled
OLS model. The pooled model does not control for the fixed individual heterogeneity that drives
a large proportion of the association between income and well-being.
In column 3 of Table 5.2 personality variables are introduced into the pooled OLS model.
Like the fixed effects estimates in column 2, although not nearly as much, the coefficients are
attenuated downwards compared with the pooled OLS model in column 1. In the final column of
Table 5.2 the individual fixed effect residuals from the fixed effect regression are predicted using
personality. It is observed that at least 20% of the individual heterogeneity can be explained
96
Table 5.2: Fixed effect and pooled OLS life satisfaction regressions
Dependent Variable:
Independent Variables:
Log of Monthly Household Income (Euros)
Age
Age squared/1000
Female
Education (years)
Log of Household Size
Married
Separated
Divorced
Widowed
Unemployed
Retired
Disabled
Child dummy
(1)
(2)
(3)
Life Satisfaction (Standardized)
Pooled OLS
0.410
(54.57)**
-0.034
(24.35)**
0.333
(23.69)**
0.020
(3.25)**
0.018
(13.45)**
-0.237
(22.28)**
0.119
(10.18)**
-0.281
(11.00)**
-0.050
(3.32)**
-0.049
(2.69)**
-0.490
(37.31)**
0.033
(2.65)**
-0.442
(41.53)**
0.060
(6.37)**
Fixed Effect
0.178
(16.34)**
-0.002
(0.22)
-0.085
(4.99)**
0.042
(1.67)
-0.122
(3.37)**
0.106
(3.05)**
-0.360
(7.81)**
-0.309
(23.59)**
0.031
(1.92)
-0.148
(8.48)**
0.057
(4.24)**
Standardized Personality Variables:
Pessimism
Individual Autonomy
Openness-to-Experience
Conscientiousness
Extroversion
Agreeableness
Neuroticism
Year Dummies
Regional Dummies
Constant
Yes
Yes
-2.102
(33.59)**
93256
Yes
Yes
-1.080
(6.20)**
Observations
93256
Number of Never Changing Person ID
17241
R-squared
0.14
0.04
Absolute value of t-statistics in parenthesis * significant at 5%; ** significant at 1%
Pooled OLS
0.324
(45.37)**
-0.023
(17.56)**
0.250
(18.83)**
0.051
(8.28)**
0.008
(6.29)**
-0.153
(15.27)**
0.095
(8.64)**
-0.285
(11.85)**
-0.092
(6.49)**
-0.097
(5.69)**
-0.409
(33.07)**
0.043
(3.67)**
-0.347
(34.45)**
0.027
(3.09)**
(4)
Individual’s Residual
(estimated from column 2)
Pooled OLS
-0.179
(56.08)**
0.136
(41.00)**
0.020
(5.94)**
0.017
(5.17)**
0.030
(9.03)**
0.039
(12.52)**
-0.086
(26.76)**
Yes
Yes
0.709
(16.79)**
93256
-0.191
(34.86)**
0.156
(26.80)**
0.030
(5.28)**
0.006
(1.09)
0.028
(4.73)**
0.048
(8.90)**
-0.092
(16.60)**
No
No
0.010
(2.06)*
17241
0.24
0.23
97
using the personality measures. This shows that personality measures explain much of the
variation between individuals and confirms that personality is one of the single biggest predictors
of well-being. However, for those interested in causality the use of personality measures at this
stage may be unconvincing. To uncover causal effects of income on well-being heterogeneous
factors must be fully controlled for in the most convincing way. It is unlikely that the available
personality measures will completely control for individual heterogeneity. Individual
heterogeneity is most convincingly controlled for using the fixed effect model. Although
personality measures are time invariant and cannot be used directly in the standard fixed effect
model, it is still possible to use them to interact with an individual’s income that does vary across
time.
Table 5.3 begins the analysis of whether individuals with different personality traits have
different well-being reactions to changes in their income. For example, do individuals that score
high on neurotic indicators enjoy changes to their income more or less than individuals that score
low on neuroticism? Are extroverted individuals more adversely affected by a decrease in their
income? To test such hypotheses the personality measures, openness-to-experience,
conscientiousness, extroversion, agreeableness, neuroticism, degree of autonomy and pessimism
are interacted with the income variable. After controlling for all other changes to an individual’s
circumstances Table 5.3 displays the results of the income-personality interactions. The
coefficients on the demographic controls, since they are very similar to those in Table 5.2, are not
reported. For completeness the results of both the pooled OLS and fixed effects models are
included. In columns 1 and 3 we test for simple linear interactions whilst columns 2 and 4
explore quadratic interactions. There is strong evidence that individuals with high levels of
conscientiousness or pessimism are likely to gain more satisfaction with their lives as a result of
98
an increase to their income. The same is true for individuals that have low levels of control over
their lives. These results are consistent across estimation types.
Neuroticism, however, is found to have a negative and significant association with
subjective well-being in the pooled model but a positive coefficient on the edge of significance in
the fixed effect model. However, the models measure fundamentally different things; one model
concerns the estimation of an association and the other the variation within individuals. Although
there may be larger differences in well-being across the income distribution for neurotic
individuals this finding is not incompatible with the result that neurotic individuals benefit less
from an increase to their income.
The effects presented in Table 5.3 are substantial. For example, concentrating on the
results from the fixed effect model in column 3; if the income of a typical individual were
doubled, then life satisfaction is estimated to increase by around 0.19 standard deviations. Since
the personality variables have been standardized the estimates suggest that the life satisfaction of
an individual with moderately high levels of conscientious (i.e. one standard deviation above the
mean) would increase by around 0.23 standard deviations if their income were doubled.
Alternatively, moderately conscientious individuals could be viewed as benefiting from income
by around 20% more than those with average conscientiousness levels. Therefore, holding
everything else constant, a typical individual would need to receive 20% more income than
someone with moderate levels of conscientiousness to reach the same level of satisfaction with
life. The effect is even more dramatic were we to consider doubling the income of individuals at
the extremes. An individual with very high levels of conscientiousness (two standard deviations
above the mean) would receive a life satisfaction rise of around 0.28 standard deviations. In
contrast, an individual with extremely low levels of conscientiousness (two standard deviations
99
Table 5.3: Fixed effects and pooled OLS analysis of income interactions with personality
(1)
Dependent Variable:
Independent Variables:
Log of Monthly Household
Income (Euros)
(2)
(3)
(4)
Life Satisfaction (Standardized)
Pooled OLS
Fixed Effect
0.323
(45.08)**
0.309
(28.75)**
0.185
(16.82)**
0.139
(7.33)**
0.034
(5.98)**
0.023
(3.94)**
-0.001
(0.22)
-0.009
(1.40)
-0.002
(0.48)
0.000
(0.05)
-0.003
(0.83)
0.035
(5.13)**
0.014
(3.70)**
0.003
(0.43)
0.011
(2.55)*
0.003
(0.59)
-0.005
(1.05)
0.016
(2.80)**
-0.004
(0.92)
0.054
(5.25)**
0.051
(4.92)**
0.009
(1.09)
-0.041
(3.62)**
0.001
(0.19)
-0.009
(0.84)
0.002
(0.31)
0.062
(4.85)**
0.017
(2.43)*
0.010
(0.85)
-0.003
(0.34)
0.010
(0.91)
0.011
(1.40)
-0.021
(1.94)
0.007
(0.90)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
0.278
(8.59)**
93256
0.329
(10.04)**
93256
0.274
(1.72)
93256
17241
0.274
(1.72)
93256
17241
0.24
0.04
0.04
Personality Interactions:
Pessimism
(Pessimism)2
Individual Autonomy
-0.012
(2.04)*
(Individual Autonomy)2
Openness-to-Experience
0.002
(0.32)
(Openness-to-Experience)2
Conscientiousness
0.016
(3.03)**
(Conscientiousness)2
Extroversion
-0.002
(0.36)
(Extroversion)2
Agreeableness
0.006
(1.16)
(Agreeableness)2
Neuroticism
0.014
(2.50)*
(Neuroticism)2
Year Dummies
Regional Dummies
Demographic Variables
Constant
Observations
Number of Never
Changing Person ID
R-squared
0.24
Absolute value of t-statistics in parenthesis
* significant at 5%; ** significant at 1%
-0.040
(3.67)**
-0.008
(0.78)
0.045
(4.43)**
0.013
(1.19)
0.008
(0.77)
-0.020
(1.88)
100
below the mean) is predicted to gain just 0.10 standard deviations in life satisfaction. The
differences between individuals at the extremes are huge and similar effect sizes are exhibited
across individual autonomy and pessimism. Individuals with combinations of these personality
traits are also predicted to have marginal utilities of income that differ greatly to a typical
individual. The evidence for any quadratic interaction effects seems limited.
5.5.1 Robustness Tests
Overall the estimates from Table 5.3 clearly indicate that aspects of an individual’s
personality interact with income. It is possible, however, that the results could be explained in a
number of alternative ways. The alternative explanations are explored here.
It is possible that how an individual answers a given personality question could be
heavily biased by their current circumstances. Factors such as pessimism and individual
autonomy may be particularly affected if, for example, an individual had just received a large pay
rise in that year. The interaction effects could be driven primarily by the wave in which
personality variables were answered. This possibility is tested by excluding from a fixed effect
analysis the wave in which personality was measured and also the year previous to that. This
minimizes any bias in personality responses that may come from an individual’s present situation
and essentially becomes a test of whether there is still an income-personality interaction with
personality measures that individuals will answer in the future. The results are shown in the first
column of Table 5.4. Other than neuroticism now being significant there are no substantive
differences.
A further potential issue with the main analysis is that an investigation into interactions
will typically include at least three effects related to the interaction; the independent level effects
of both characteristics and an interaction effect (Aiken & West, 1991). Since personality does not
101
Table 5.4: Robustness of the personality-income interactions
(1)
Dependent Variable:
Independent Variables:
Log of Monthly Household
Income (Euros)
(2)
(3)
Life Satisfaction (Standardized)
FE (excl. 2004
Fixed Effect Vector
Pooled
& 2005)
Decomposition
OLS
(4)
(5)
Fixed
Effect
Fixed
Effect
0.169
0.177
0.339
0.244
0.172
(11.22)**
(34.91)**
(18.36)**
(10.09)**
(4.33)**
-0.000
(0.09)
-0.000
-0.000
(2.19)*
0.000
(1.34)
0.000
(0.74)
-0.000
(1.14)
(0.50)
Monthly Household Income
Monthly Household
Income-Squared
Monthly Household
Income-Cubed
Interactions with Aspects of
Openness:
Values artistic experiences
-0.013
(2.20)*
0.014
Original/Comes up with
new ideas
(1.75)
Personality Interactions:
Pessimism
Individual Autonomy
Openness-to-Experience
Conscientiousness
Extroversion
Agreeableness
Neuroticism
Year Dummies
Regional Dummies
Demographic Variables
Constant
Observations
Number of Never Changing
Person ID
R-squared
0.056
(3.84)**
-0.040
(2.59)**
0.022
(1.48)
0.053
(3.50)**
0.007
(0.45)
-0.008
(0.57)
-0.030
(2.03)*
0.009
(2.29)*
-0.007
(1.69)
-0.002
(0.46)
0.008
(2.04)*
0.002
(0.58)
0.002
(0.43)
-0.003
(0.79)
0.033
(5.79)**
-0.009
(1.54)
0.004
(0.74)
0.017
(3.06)**
-0.002
(0.41)
0.004
(0.72)
0.014
(2.49)*
0.053
(5.08)**
-0.038
(3.42)**
-0.006
(0.56)
0.046
(4.52)**
0.013
(1.17)
0.006
(0.57)
-0.023
(2.14)*
0.056
(5.37)**
-0.042
(3.80)**
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
0.321
(1.59)
59418
16790
0.879
(30.00)**
93256
0.290
(7.72)**
93256
0.363
(2.23)*
93256
17241
0.275
(1.73)
93256
17241
0.03
0.63
0.24
0.04
0.04
Absolute value of t-statistics in parenthesis * significant at 5%; ** significant at 1%
0.043
(4.17)**
0.008
(0.75)
0.012
(1.12)
-0.019
(1.75)
102
change across time the independent effect of a particular personality characteristic cannot be
determined within the standard fixed effect framework. An assumption in the main analysis is
that the fixed effect model fully controls for the level of personality. Although this is a
reasonable assumption and shouldn’t cause too much concern, personality effects not fully
eliminated using the fixed effect technique may surface in an investigation of interactions. This
assumption is relaxed by exploring interactions within the recently developed fixed effect vector
decomposition (FEVD) technique (Plumper & Troeger, 2007). This technique allows
independent effects of the personality variables to be obtained and has three stages. In the first
stage a standard fixed effect regression is performed as seen earlier in Table 2 with no
personality interactions. Each individual’s fixed effect residual is then predicted and saved. In the
second stage all variables with zero within person variation, such as age, gender and personality
are used to estimate this saved fixed effect residual.
(5.3) ˆi   Pi  1agei  2 genderi  ... i
The error from equation 5.3, η, which represents the true unobservable error, is then used in a
final stage pooled OLS regression with all explanatory variables, including the incomepersonality interaction terms. As demonstrated in the previous chapter the FEVD model has been
shown to allow independent effects of variables with zero or low within person variation to be
obtained (Plumper & Troeger, 2007). The results using the FEVD model are shown in the second
column of Table 5.4. Although the coefficients are not as large, they suggest that, even using a
technique that effectively controls for personality twice, there is still significant evidence that
103
individuals who are conscientious, lack control in their life and are pessimistic gain more utility
from a change to their income.
The final robustness test checks that the strong interaction effects are not being driven by
non-linearities in income. In columns 3 and 4 of Table 5.4 we show that introducing the level of
income up to its cubed root does not alter the interaction effects. In the final column we also
show that there are components of openness-to-experience that also interact with income.
Whether an individual values artistic experiences has a negative and significant coefficient as
might be expected intuitively; yet there is a positive coefficient, although only on the border of
significance, on whether an individual sees themselves as original or good at coming up with
new ideas. Due to the counteractive effects this explains why there is no overall significance on
the openness-to-experience measure. The results in column 5 add support for our general
personality-income interactions model; confirming an intuitive result that valuing artistic
experiences means money has less of an impact on an individual’s life and that being original
may mean the individual is good at spending their money in new and innovative ways to produce
greater well-being.
5.6
Conclusion
The overall conclusion is that the extra utility gained from an increase in income is
heavily dependent upon an individual’s personality. Economists normally concentrate on the
average effect of an increase in income across an entire population. We show that individuals
who are conscientious, have low control in their life or are pessimistic actually get more utility
from a given change in income. For example, individuals that are moderately un-conscientious
are predicted to need around a 20% greater increase to their income than those moderately
conscientious to achieve the same increase in well-being. Our results are robust to a number of
104
alternative explanations. This could be an important finding for policy makers in two respects.
Firstly, with regards to increasing national welfare and which individuals might benefit the most
from rises to their income and secondly, in understanding how individuals might react to
economic incentives. We provide some insights into the complex relationship between income
and well-being showing that individuals have heterogeneous preferences. Not everyone appears
to benefit from changes to their income in the same way and it could be problematic to assume
that they do. If the marginal utility from income is different across individuals then individuals
will behave differently towards a given financial incentive. This is an important policy concern
and suggests tailoring policy according to an individual’s personality may make policy more
effective. A similar argument has been made with regards to gender-specific taxation (Alesina &
Ichino, 2007).
Our results generate some more important questions. There may be strong evidence that
an individual who is either conscientious, has low control in life or pessimistic has a higher
marginal utility of income than others but this leaves the question as to why? On this we can only
speculate. Perhaps individuals with certain personalities have a habitual spending pattern that
increases their well-being more than others. If so then it is important to establish what these
spending patterns are. It would seem plausible that conscientious individuals might be better
planners enabling them to make wiser purchases with their income. The positive well-being link
between conscientiousness and income may relate to recent work by Ameriks et al. (2007). They
found that individuals with high levels of conscientiousness accumulate more wealth. However,
is this because they simply enjoy their income more and are therefore driven to accumulate
wealth or do non-conscientious individuals simply make bad decisions? Perhaps income is not as
important to everybody. Some people may gain more utility from non-monetary areas of life,
105
such as social relationships, cultural or physical activity, linking in with our result that someone
who values artistic experiences is less responsive to extra income.
These are many important questions that still need answering but we have shown how
exploiting standard psychological measures of personality can help to do this. The findings
presented here perhaps produce many more questions than are solved but could shed new light on
future directions in which income and well-being research might take.
106
5.7
Appendix
5.7.1 Note to Tables
Variable
Description
Life Satisfaction
A self reported measure of how satisfied the individual is with their
life, all things considered, where 0=completely dissatisfied and
10=completely satisfied
Monthly Household Income
The household’s income in which the individual resides
(Euros)
Age
Individual’s age
Female
Individual is female (excluded dummy: male)
Education (years)
Number of years of education
Household Size
The number of members in the individual’s household
Married, Separated, Divorced,
Individual is married, separated, divorced or widowed (excluded
Widowed
dummy: single)
Unemployed
Individual is unemployed (excluded dummies: any other responses
to occupation position except retired)
Retired
Individual is retired (excluded dummies: any other responses to
occupation position except unemployed)
Disabled
Disability status of the individual
Child dummy
Whether there is at least one child in the household (excluded
dummy: no children in the household)
Personality variables
See section 5.7.2 for full description of the personality variables
5.7.2 Personality Variables in GSOEP
There were a number of questions asked in the 2005 wave of the GSOEP that attempt to quantify
aspects of an individual’s personality. In the questionnaire section entitled “What kind of
personality do you have?” there were 15 questions related to the “Big five” personality inventory,
and a further 9 that covered aspects of an individual’s autonomy. A further question, on
pessimism, comes from the “Attitudes and opinions” section. These questions were as follows:
107
5.7.2.1 Big Five Personality Inventory
Individuals are asked whether they see themselves as someone who…
1. …does a thorough job
2. …is communicative, talkative
3. …is sometimes somewhat rude to others
4. …is original, comes up with new ideas
5. …worries a lot
6. …has a forgiving nature
7. …tends to be lazy
8. …is outgoing, sociable
9. …values artistic experiences
10. …gets nervous easily
11. …does things effectively and efficiently
12. …is reserved
13. …is considerate and kind to others
14. …has an active imagination
15. …is relaxed, handles stress well
Individuals are asked whether the statement applies to them on a 1 to 7 scale, with 1 meaning the
statement does not apply to them at all and 7 that it applies perfectly.
5.7.2.2 Individual Autonomy
Individuals are asked their attitudes towards their life and future.
108
16. How my life goes depends on me
17. What a person achieves in life is above all a question of fate or luck
18. If a person is socially or politically active, he/she can have an effect on social conditions
19. I frequently have the experience that other people have a controlling influence over my
life
20. One has to work hard in order to succeed
21. If I run up against difficulties in life, I often doubt my own abilities
22. The opportunities that I have in life are determined by the social conditions
23. Inborn abilities are more important than any efforts one can make
24. I have little control over the things that happen in my life
Individuals are asked whether they agree with the statements on a 1 to 7 scale, with 1
representing complete disagreement and 7 that they completely agree.
5.7.2.3 Pessimism
This is a one item scale. Individuals are asked
25. When you think about the future, are you…optimistic, more optimistic than pessimistic,
more pessimistic than optimistic, pessimistic?
This variable is treated as cardinal.
5.7.2.4 The Construction of Personality Measures
Factor analysis was used to check whether these variables were measuring the same underlying
trait and suggested that the questions related to Big Five personality model of openness to
experience, conscientiousness, extroversion, agreeableness and neuroticism, with individual
autonomy and pessimism explaining additional variance.
109
Not all 25 personality questions were used to construct the 7 personality measures, as some were
poorly defined and appeared to measure no underlying trait. The remaining questions were used
to construct the personality measures in the following ways:
Openness to experience used questions 4, 9 and 14.
Conscientiousness used questions 1, 7 and 11 (question 7 was reverse coded)
Extroversion used questions 2 and 8 (question 12 was not used as factor analysis showed it did
not measure the same thing as the other two questions)
Agreeableness used questions 3, 6 and 13 (question 3 was reverse coded)
Neuroticism used questions 5, 10 and 15 (question 15 was reverse coded)
Individual Autonomy used questions 16, 19, 21 and 24 (questions 19, 21 and 24 were reverse
coded)
Pessimism used question 25 and factor analysis showed it was unique
110
CHAPTER 6
6
THE DARK SIDE OF CONSCIENTIOUSNESS: CONSCIENTIOUS PEOPLE SUFFER MORE FROM
UNEMPLOYMENT
6.1
Abstract
Conscientious individuals tend to achieve more and have higher well-being. This has led to a
view that conscientiousness is always positive for well-being. We hypothesize that
conscientiousness may actually be detrimental to well-being when failure is experienced, such as
when individuals become unemployed. In a longitudinal study of 9,530 individuals we show that
the drop in an individual’s life satisfaction following unemployment is significantly moderated
by their conscientiousness. An individual with a high level of conscientiousness (i.e. a person one
standard deviation above the mean) experiences a 200% higher drop in life satisfaction than
someone at low levels. The difference is not temporary and is sustained through to the second
year of unemployment. Thus the positive relationship typically seen between conscientiousness
and well-being is completely reversed: Conscientiousness is therefore not always good for wellbeing.
111
6.2
Introduction
Conscientiousness has a strong positive effect on an individual’s well-being (DeNeve &
Cooper, 1998; Hayes & Joseph, 2003). Conscientious individuals are orientated towards life
situations that are beneficial for well-being (Mccrae & Costa, 1991), they set themselves higher
goals (Barrick, Mount, & Strauss, 1993; DeNeve & Cooper, 1998), and have high levels of
motivation (Judge & Ilies, 2002). Conscientious individuals are therefore more likely to achieve
highly (McGregor & Little, 1998) and obtain higher levels of well-being (DeNeve & Cooper,
1998). Overall, this body of literature has lead conscientiousness to be conceptualized as a
positive, adaptive personality trait that is important for well-being, employment, and personal
functioning (DeNeve & Cooper, 1998).
Despite the evidence that conscientiousness is generally positively related to well-being
and functioning, it is also possible that there are situations where this pattern is reversed, and
where being more conscientiousness forms a risk factor for increased suffering and less
productivity. These situations have not previously been studied, leading to a perhaps erroneous
view that being more conscientious is better all of the time. Given the strong links between
conscientiousness and goal setting, motivation, and achievement, we hypothesize that under
conditions of failure conscientious people may suffer more, having sharper decreases in wellbeing. We use a nationally representative dataset of around 10,000 people to investigate the
causal role of conscientiousness on well-being following a life event that represents a severe and
chronic failure. Specifically, we examine how prospectively measured conscientiousness may
interact with unemployment to affect well-being.
Unemployment is an ever present aspect of our societies. As of July 2009 there were 14.5
million unemployed individuals in the United States representing an unemployment rate of 9.4%;
112
a rate not seen since 198311. Many individuals face the prospect of unemployment at some point
in their lives and the experience can be devastating. The loss of work generally represents a
failure in life and can be extremely harmful to well-being (e.g. Frey & Stutzer, 2002; Oswald,
1997). In addition to the loss of earnings, unemployment represents a loss of purpose and can
erode an individual’s identity and sense of self-worth (Ashforth, 2001; Turner, 1995). It is not the
case that less happier people are selected in unemployment (Diener et al., 1999), and a metaanalysis of longitudinal studies shows that unemployment has an average causal effect size of .38
on mental health (McKee-Ryan et al., 2005). Additionally it can be difficult to fully recover
psychologically from unemployment (Clark et al., 2001; Lucas et al., 2004).
Although conscientious people may potentially suffer more following failure, there are
three additional reasons to suggest conscientious people may suffer more from unemployment.
First, conscientious people tend to accumulate more wealth (Ameriks et al., 2003; Ameriks et al.,
2007) and obtain larger well-being increases when their incomes rise (as demonstrated in Chapter
5). To the extent that accumulating wealth is a goal of conscientious people, unemployment will
represent a chronic blocking of a goal, which is known to lead to decreased well-being (Emmons,
1992). Second, employment may be more important to conscientious people, as it offers
opportunities for conscientious people to pursue goals and use their particular strengths (c.f.
Barrick et al., 1993; DeNeve & Cooper, 1998). Both the increased importance of employment
and the use of strengths have been related to well-being (McKee-Ryan et al., 2005; Seligman,
Steen, Park, & Peterson, 2005), and becoming unemployed would remove the opportunities for
conscientious people to gain emotional benefits in this way. Third, being conscientious may lead
to different appraisals of the reasons for unemployment. Specifically, un-conscientious people
11
Figures from the Current Population Survey, July 2009 -http://www.bls.gov/web/cpseea1.pdf
113
may be able to attribute unemployment to a lack of effort whilst working in the previous job (a
temporary and specific cause for failure). In contrast, conscientious people who worked to their
ability would not be able interpret the situation in this way, and would be more likely to attribute
their failure to their own lack of ability (a stable and general cause of failure). This attribution
style has been related to both clinical depression (Alloy, Abramson, Whitehouse, & Hogan,
2006; Mongrain & Blackburn, 2005), anxiety (Ralph & Mineka, 1998), and negative affect
(Sanjuan, Perez, Rueda, & Ruiz, 2008).
As conscientious people are theoretically more likely to (a) suffer from failure, (b) have
accumulating wealth as a goal, (c) value their workplace more, and (d) appraise unemployment
differently, we suggest that conscientious people would suffer more from unemployment. It is
not the purpose of this study to examine which of these mechanisms is responsible for the effect,
but rather to demonstrate that the usually observed positive relationship between
conscientiousness and well-being can be in some situations reversed. In doing so, we aim to
encourage a broader study of conscientiousness, which considers both the situations in which
conscientiousness is adaptive, and when it becomes a risk factor. Additionally, this will provide
the first study to suggest that the effects of unemployment on well-being depend on any
personality characteristic. This observation may have applied implications for the support given
to people post-employment.
6.3
Method
6.3.1 Participants and Procedure
The sample comprised a nationally representative sample of 9530 people (4538 males,
4992 females), all of whom were in employment, and who completed measures at three time
points, each one year apart. Ages ranged from 17 to 83 years (M = 41.75, SD = 11.67). Income
114
varied from €200 to €30,000 each month (M = 3112.44, SD = 1789.78, Mdn = 2800.00).
Participation was part of the German Socio-Economic Panel Study (GSOEP), a longitudinal
sample of German households, with questions relevant for this analysis only included during the
2005, 2006 and 2007 waves. All members of the household were invited to participate with
questionnaires being administered through yearly face-to-face interviews. Further data on
sampling is available in Haisken-De New and Frick (1998).
All participants completed the measures of conscientiousness and well being in 2005, and
were employed at this time. For analysis purposes, participants were separated into two groups
according to their employment status in 2005, 2006 and 2007. The first group included those that
were employed in 2005, 2006 and 2007 (n = 9361) and the second group included those that
were employed in 2005 but became unemployed in 2006 and were still unemployed in 2007 (n =
169). The GSOEP also contained participants who were not employed or were retired in 2005;
these people were not included in this sample.
6.3.2 Measures
Life Satisfaction: Life satisfaction was measured using a one-item scale. Participants
were asked “how satisfied are you with your life, all things considered?” and responded to this
question on an 11-point scale, from 0 (complete dissatisfaction) to 10 (complete satisfaction).
The life satisfaction measure was standardized. The single item scale, although fairly typical in
large data sets, is a limitation of the study and could result in an underestimation of the true effect
size.
Conscientiousness: A 3 item scale was used to uncover participants’ pre-unemployment
levels of conscientiousness. The questionnaire asked individuals to rate each of the three
statements, which concerned whether they saw themselves as someone who “does a thorough
115
job”, “tends to be lazy” or “does things effectively and efficiently”, on 7-point scales, from 1
(does not apply to them) to 7 (applies perfectly to them). The individual’s answer to “tends to be
lazy” was reverse coded and then all three scores were aggregated to obtain the
conscientiousness scale. The scale was then standardized across individuals. This short scale was
developed specifically for the GSOEP to enable individual conscientious levels to be determined
where limited space for questions was available. Gerlitz and Schupp (2005) document the
extensive pre-testing that took place to ensure the 3-item scale replicated established longer
conscientiousness scales.
Demographic controls: Each participant’s age, gender, education level (years), marital
status, disability status, household income, household size and whether they had children in their
household were used as control factors. Regional dummy variables were also included.
6.4
Results
Two hierarchical multiple regressions were performed to predict life satisfaction at one
(T2) and two years (T3) after unemployment. All baseline measures were taken at T1 prior to
unemployment. For both regressions, in the first step, T1 conscientiousness, T1 life satisfaction,
and T1 demographic controls were entered. In the second step, a dummy variable was entered,
representing whether the person had become unemployed between T1 and T2 (coded 0 =
Individual was employed in 2005, 2006 and 2007, 1 = Individual is employed in 2005 but
unemployed in 2006 and 2007). This tested the unique impact of unemployment on individuals’
well-being. In the third step, an interaction was entered between T1 conscientiousness and the
unemployment dummy variable. This tested whether baseline levels of conscientiousness
changed the impact of unemployment on well-being. The analysis followed Aiken and West’s
(1991) recommendations for moderation analysis; conscientiousness was standardized prior to
116
Table 6.1: Two hierarchical regression analyses predicting the life satisfaction of individuals in the years following unemployment
Regression 1:
Regression 2:
Predicting life satisfaction at T2
Predicting life satisfaction at T3
b
SE
β
b
SE
β
Life satisfaction at T1
0.54
0.01
.54***
0.51
0.01
.05***
Conscientiousness at T1
0.04
0.01
.04***
0.03
0.01
.03***
Life satisfaction at T1
0.54
0.01
.54***
0.50
0.01
.50***
Conscientiousness at T1
0.04
0.01
.04***
0.03
0.01
.03***
Employed at T1 but unemployed in T2 and T3
-0.34
0.06
-.05***
-0.54
0.07
-.07***
Life satisfaction at T1
0.54
0.01
.54***
0.50
0.01
.50***
Conscientiousness at T1
0.04
0.01
.04***
0.03
0.01
.03***
Employed at T1 but unemployed in T2 and T3
-0.34
0.06
-.05***
-0.54
0.07
-.07***
-0.17
0.05
-.03**
-0.19
0.06
-.03***
Variables
Step 1
Step 2
Step 3
Conscientiousness at T1 x employed at T1 but unemployed in T2 and T3
2
2
Notes: Year 1 - For Step 1, F(29, 9500) = 195.37, R = 0.3736 (p < .001); for Step 2, F(30, 9499) = 190.43; R = 0.3756 (p < .001); for Step 3, F(31, 9498) = 184.79;
R2 = 0.3762 (p < .001); Year 2 - For Step 1, F(29, 9500) = 157.49; R2 = 0.3247 (p < .001); for Step 2, F(30, 9499) = 155.71; R2 = 0.3297 (p < .001); for Step 3, F(31,
9498) = 151.22; R2 = 0.3305 (p < .001)
**p < .01 ***p < .001
All regressions include demographic controls; age, gender, education level (years), marital status, disability status, household income, household size, a child in their
household dummy and regional dummies.
117
analysis, and the interaction term was a product of the standardized conscientiousness variable
and the unemployed dummy.
Table 6.1 shows the results from both multiple regressions. Controlling for baseline
variables, the effect of becoming unemployed on life satisfaction was d = -.34 at one year post
unemployment, which rose to d = - .54 at two years. However, this effect was significantly
moderated by conscientiousness. Adding the interaction term in Step 3 significantly improved the
prediction for both regressions12. The moderation is plotted in Figure 6.1. The moderation was
substantial: At the first year of unemployment, people high in conscientiousness (defined as 1 SD
above the mean) had life satisfaction decreases of d = .51 from their pre-unemployment levels.
In contrast, people low in conscientiousness (defined as 1 SD below the mean) only had
decreases of d = .17. These differences persisted into the second year of unemployment, where
highly conscientious people had decreases in life satisfaction of d = .73 from pre-unemployment
levels, compared to individuals with low conscientiousness who only had decreases of d = .35. In
relative terms these results suggest that highly conscientious people suffer from unemployment
approximately three times as much as people low in conscientious people in the first year of
unemployment, and more than twice as much in the second year.
As a robustness check, we further tested whether our result was driven by conscientious
individuals being more or less likely to face a period of unemployment. We conducted a
logistical regression to predict the probability of an individual in our sample being employed at T
but unemployed in T+1 and T+2. Conscientiousness did not predict unemployment (β = 0.01, p >
0.9).
12
Note that the increment in R-squared is not readily interpretable, as given that 98.2% of the sample did not
become unemployed, the moderation would not explain a large amount of variance in life satisfaction for the whole
sample. A more representative indication of effect sizes is provided by Figure 1.
118
Figure 6.1: The life satisfaction change following unemployment as moderated by conscientiousness
- 1 SD Conscientiousness
Mean Conscientiousness
+ 1 SD Conscientiousness
0.9
0.8
Change in life satisfaction
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1
2
Years following unemployment
Error bars denote standard errors calculated according to Aiken and West (1991)
at the appropriate levels of conscientiousness.
6.5
Discussion
We show that the personality trait conscientiousness is not always beneficial for
individual well-being. Whilst individuals high on levels of conscientiousness may achieve more
throughout their lives (Barrick et al., 1993), leading to higher levels of well-being, we show that
during times of failure being conscientious can actually be detrimental. In longitudinal study of
around 10,000 individuals we first show that those entering unemployment suffer severe
psychological consequences. We then show, using an individual’s level of conscientious before
they became unemployed, that the psychological consequences are significantly greater for those
that are conscientious. This difference continues into the second year of unemployment. Thus,
119
the normal positive relationship between conscientiousness and well-being is completely
reversed.
We propose a number of possible explanations as to why this might take place. Firstly,
unemployment represents a failure to achieve. Conscientious individuals care more about
achieving their goals and so any failure could be more detrimental to their well-being. Secondly,
conscientious individuals tend to value wealth accumulation as demonstrated in Chapter 5
(Ameriks et al., 2003; Ameriks et al., 2007) and unemployment prevents them from achieving
this goal. Third, the work environment allows conscientious individuals to work to their strengths
and they are more likely to see work as a central part of their identity. The loss of a job may
therefore erode a conscientious individual’s core sense of meaning and purpose in life to a much
greater extent than someone with low levels of conscientiousness. It is also possible that a
conscientious individual will attribute their job loss to their lack of ability as opposed to a lack of
effort. Lastly, conscientious individuals may carry out a more efficient job search and, although
they may find re-employment quicker, there is some evidence to suggest that individuals that are
more motivated to find work also have higher levels of depressive affect (Feather & Davenport,
1981). In accordance with this it had has been demonstrated that job search effort during
unemployment is negatively related to well-being (McKee-Ryan et al., 2005).
Our analysis can not unpick the extent to which these mechanisms drive conscientious
individuals to suffer more during unemployment. However, we provide strong evidence to
suggest that conscientiousness is not always beneficial for well-being. Whilst conscientious
individuals may on the whole have higher well-being, there are some circumstances that cause
them to have lower well-being. More research is needed around this area.
120
The psychological consequences of unemployment have been researched extensively.
However, previous research into unemployment has not been looked at in relation to individual
differences. Our research provides further evidence that personality traits should be considered
when trying to understand economic behaviour (Ameriks et al., 2003; Ameriks et al., 2007;
Borghans et al., 2008; Bowles et al., 2001b). Our study also has important practical implications.
Conscientious individuals are a risk group psychologically during unemployment and these
individuals may benefit the most from extra support during unemployment.
121
CHAPTER 7
7
DO PEOPLE BECOME HEALTHIER AFTER BEING PROMOTED?
7.1
Abstract
This paper explores the hypothesis that greater job status makes a person healthier. It first
replicates the well-known cross-section gradient in health across different levels of job seniority.
Then -- following a large sample of randomly selected individuals through time -- it turns to
longitudinal patterns. When it does so, the paper can find little evidence that promotees exhibit a
health improvement. In the private sector, promotion worsens people’s psychological health (on
a standard GHQ mental-strain measure). The data suggest that it is people who start with good
health who are promoted.
Revise and resubmit for Health Economics
122
7.2
Introduction
Human beings with high occupational status have good health and low rates of premature
mortality. Cross-section evidence for this correlation has been found many times (Johnson,
Sorlie, & Backlund, 1999; Macleod, Smith, Metcalfe, & Hart, 2005; Marmot et al., 1984). The
difficulty, however, is to know how to interpret the association. Is it causal in the sense that job
status somehow leads to a later boost in a person’s health?
Surprisingly, there appears to be no published truly longitudinal test of this hypothesis –
one in which the investigator is able to observe individuals’ health both before and after they are
promoted. This paper attempts to design such a test. The focus is on individuals’ job rank and
thus their degree of control within the workplace. We draw upon a panel data set, collected
annually between 1991 and 2006, with information on approximately 1000 individual
promotions. We follow what happens to the health of those who gain seniority when compared
to the health of those who do not.
With one exception, our longitudinal study does not find compelling evidence in favour
of a status-causes-health theory. Moreover, job promotion in many instances brings about a
worsening of mental health. Nevertheless, after being promoted to the position of manager,
people do go on to reduce the number of times that they visit a doctor (by up to 20%).
7.3
Earlier Work
Researchers such as Marmot (2004) and Wilkinson (2001) have argued that there may be
a cause-and-effect connection between status and health. According to this account, high status
can itself boost health: psychosocial stressors are detrimental to the human condition, especially
to cardiovascular health and the auto-immune system, and they can explain much of the social
123
health gradient. Griffin et al. (2002) and others have suggested that greater control at work
improves mental health.
Our paper also relates to a stream of work on the connections between mental well-being,
health and economics, such as Van Praag et al (2003) and Graham et al (2004). Here one
possibility is that greater psychological well-being, which in principle might be thought to stem
from job seniority, can lead to economic success.
Whilst there may be a causal chain running from status to health, one may also operate in
the opposite direction, with the healthiest individuals going on to obtain the highest status
(Deaton, 2003; J. P. Smith, 1999; West, 1991). Alternatively, a third unobservable influence,
such as genetic factors, could cause both good health and job success (Adams, Hurd, McFadden,
Merrill, & Ribeiro, 2003; Cutler, Deaton, & Lleras-Muney, 2006).
Using various socio-economic status (SES) indicators, attempts have been made to
address the problem of causality. For example, Adams et al (2003) for the US, and Adda,
Chandola and Marmot (2003) for Sweden and the UK, use longitudinal data and control for
initial health. No clear causal effects from SES to health are found. Similarly, Gardner and
Oswald (2004) control for initial health at T in an annual panel on individuals and find that
income does not influence survival probability at T+10. Whilst they adjust for pre-existing
health conditions, these studies cannot discount the possibility that individuals’ early health led
to their SES. Using instrumental variables, however, Ettner (1996) argues that more income
appears to result in significantly better physical and mental health. Lorgelly and Lindley (2008)
also find, using fixed effects regressions, that although absolute income influences health there
are no independent effects of either relative income or income inequality. Wilkinson (1986)
124
examines the link between changes to both occupational mortality and occupational incomes
over a twenty year period.
Sapolsky (2004) experimentally documents the fact that health consequences emerge
relatively quickly after rank is established across groups of animals. Sapolsky (2004) further
suggests that this pattern extends to humans. However, the social context has been shown to
differ across species. For example, subordinate animals that embark in cooperative breeding
(Abbott, Saltzman, Schultz-Darken, & Tannenbaum, 1998) generally do not suffer from elevated
release of glucocorticoids, a classic negative stress response documented in Sapolsky, Romero
and Munck (2000). Similarly, this stress response within species is dependent on whether the
subordinate animals are subjected to high levels of harassment by dominant individuals and
whether they have social support networks (Abbott et al., 2003).
Exogenous manipulation of a human individual’s status is not possible, but nearexperiments potentially provide a way in which one might try to uncover causal effects. Rablen
and Oswald (2008) offer support for a causal effect, running from social status to health, among
Nobel Prize winners. Their results are similar to, though use different statistical methods than,
that on Academy Award winners carried out by Redelmeier and Singh (2001). Snyder and Evans
(2006) focus on a quasi-experiment in the realm of income. They find, counter-intuitively, that
those with higher incomes as a result of changes to social security payments also have greater
mortality rates. This result is somewhat consistent with the finding by Ruhm (2000) that
temporary upturns in the economy are bad for people’s health.
The simple correlation between health and income is strong (recent econometric evidence
includes Cantarero and Pascual (2005), Duleep (1986), Frijters, Haisken-DeNew and Shileds
(2005), McDonough et al. (1997), Menchik (1993) and Wolfson et al. (1993)), and similarly so
125
for education (Feldman, Makuc, Kleinman, & Cornonihuntley, 1989; Lahelma & Valkonen,
1990). Income, education and occupation all give fairly good indications of an individual’s SES.
Duncan et al (2002) argue that economic measures, such as pay, are preferable over other
measures of SES. Yet income correlates well with psychosocial aspects and therefore with
health. It is therefore necessary to isolate independent SES effects; exclusion of correlated
variables will bias the estimates (Fuchs, 2004). There are several studies that centre around
occupation as an indicator of SES. Ala-Mursula et al (2005) conclude that women with less
work-time control have an increased risk of health problems. Fischer and Sousa-Poza (2009)
show that increased job satisfaction improves the individual’s health. Anderson and Marmot
(2007) try to exploit differences in promotion rates across departments in the British civil service
as an instrument for individual promotion.
7.4
Methodology
Consider an individual who is promoted at time T. If causality runs solely from
occupational status to health, then, after controlling for other factors correlated with health and
promotion (such as age, education and gender), there should be no significant differences, at T-1,
in the health of those who are promoted and those who are not promoted. At T+1 there should
begin to be a difference. If there is only reverse causality -- that is, causality running from health
to occupational status -- then promoted individuals should exhibit significantly better health to
the same degree at both T-1 and after T+1. Were two-way causality to exist, a promoted group
would exhibit a combination of these two effects.
Using longitudinal data, on a large sample of British workers, here cross-sectional and
difference-in-difference methods are used to explore these three hypotheses. Our promoted
126
group includes those who improve their occupational status internally and those who gain extra
seniority after a move to a different employer.
7.5
Data and Estimation Issues
Seniority and job status come in myriad forms. An empirical inquiry has to make some
taxonomic assumption. In this study, an individual’s role in the workplace is assumed to be
captured by whether they report in the British Household Panel Survey that
: their job is one of… manager, supervisor, or neither of these.
In the data set, these are uniquely different classifications13, which are similar to those used by
Macleod et al. (2005). While this approach necessarily aggregates across sectors in a way that
may produce some measurement error, it offers an indication of the seniority and hence the
degree of control each individual can be expected to have in the job. This taxonomy of seniority
assumes away complex role overlaps, and assumes too that an individual is employed, which
means that any association between unemployment and poor health will be largely ignored in our
main analysis. We return later to this issue.
Data come from the British Household Panel Survey (BHPS), a representative
longitudinal sample of British households. Running from 1991-2006, the Survey tracks over
10,000 adults in each of 16 years. Our analysis concentrates on a particular proportion of this
sample, namely, those who worked for at least five consecutive years, from, in our notation, T-1
up to T+3. We observe who is promoted14 at T. This gives us approximately 1000 individual
promotions. There is some loss to this research design, because we are unable to say whether
those who left work entirely, or subsequently changed role again, went on as a result to have
13
Those indicating neither of these are termed here as non-supervisors.
14
See the Appendix to this chapter for sample construction notes.
127
better or worse health15. But it allows a simple focus upon longitudinal health within an
individual’s work setting.
Health typically declines as people age. For our test, simple within-promoted group
comparisons are therefore likely to be insufficient; we cannot merely measure the same
individual’s health across periods T-1 to T+3. It would be difficult to discern whether declining
health across time is due to extraneous factors, or, perhaps more plausibly, to the natural process
of ageing. We overcome this by comparing particular individuals’ health levels with those
among a control group. The sample is separated into treatment and control groups -- those
promoted at T and those never promoted -- and comparisons made between them.
Our study examines three possible types of promotions: workers promoted from
(1) non-supervisor to supervisor,
(2) supervisor to manager and
(3) those going directly from non-supervisor to manager.
The final promotion type represents the largest gain in occupational status. Each
promotion case has individuals’ health and changes to health contrasted to that of an appropriate
control group, namely, individuals who remain as non-supervisors for promotion types 1 and 3
and supervisors for the 2nd promotion category. The comparison, in the analysis, takes place
across the entire 5-year period. This gives us a total sample size of up to 18,000 individual fiveyear observations.
The BHPS contains several indicators of an individual’s health. Here, we make use of
three:
(i)
15
subjective ill-health,
Section 7.7 attempts to deal with this issue.
128
(ii)
number of visits to the doctor,
(iii)
mental strain.
We do so to allow a degree of corroboration of the regression results on any single health
variable. These three variables are coded such that a higher value indicates worsening health.
The paper therefore estimates ill-health16 regression equations (we use cardinal methods but
ordered estimators give the same results).
Subjective ill health is a self rating of one’s health on a cardinal 5-point scale, where
5=very poor through to 1=excellent. The number of visits to the doctor -- available in BHPS data
as a grouped variable -- is another simple measure of how healthy an individual might be. The
final dependent variable is that of psychological ill-health. It is captured here using a General
Health Questionnaire (GHQ) measure of mental strain, on a 0 to 36 scale. The same variable -defined more fully in the Appendix -- has been used in a large medical and psychiatric literature
such as Cardozo et al (2000) and Pevalin and Ermisch (2004), and in health-economics research
by, for example, Shields and Wheatley Price (2005) and Gardner and Oswald (2004). All three
variables have positive skew; most individuals mark themselves in surveys as relatively healthy.
Individuals’ mean rating of their subjective ill-health is 2.02. They visit their doctor (i.e., their
General Practitioner, or GP, in British jargon) on average 1.78 times each year. This is on a
numerical 0-10 scale. They have mean mental strain of 10.76 on a 0-36 scale. A simple
correlation matrix is shown in Table 7.1. As might be expected, people who are less healthy on
one criterion are more likely to be recorded as less healthy on the other two.
The health measures across the entire sample from T-1 to T+3 are shown in Table 7.2,
which shows that subjective ill health, visits to the GP, and mental strain all deteriorate over a
16
A fuller description of the variables is given in the section Notes to Tables at the end of this chapter.
129
five-year period of aging. Subjective ill-health worsens in Table 2 by 0.06 points; visits to the
doctor by 0.10 points; mental health by 0.37 points. The observed health deterioration highlights
the importance of not relying merely on a within-promoted group comparison.
Table 7.1: Pearson correlation coefficients for the three ill-health measures
Subjective Ill-health
Visits to the Doctor
Mental Strain
Subjective Ill Health
Visits to the Doctor
Mental Strain
1
0.40 (n=18218)
0.27 (n=17169)
1
0.20 (n=17126)
1
All coefficients are statistically significantly different from zero at the 0.01 level
Table 7.2: Ill-health over time within the whole sample
Subjective Ill Health
Standard
N
Mean
Deviation
T-1
T
T+1
T+2
T+3
18282
18282
18282
18282
18282
2.01
2.02
2.05
2.06
2.07
0.80
0.80
0.81
0.81
0.82
Visits to the Doctor
Standard
N
Mean Deviation
18233
18233
18233
18233
18233
1.78
1.78
1.81
1.82
1.88
7.6
2.27
2.30
2.34
2.36
2.44
Mental Strain (GHQ)
Standard
N
Mean Deviation
17184
17184
17184
17184
17184
10.66
10.76
10.85
10.92
11.03
4.77
4.78
4.85
4.90
4.97
Results
We begin by depicting the cross-sectional differences in health across levels of seniority.
This is demonstrated, with gradual inclusion of a set of control variables, in Table 7.3.
Table 7.3’s evidence reveals the positive association commonly seen in empirical studies
of socio-economic status and health. The strongest correlation with occupational grade in the
table is observed for subjective ill-health. Managers in column 1 of Table 7.3 report themselves
0.185 points healthier than non-supervisors; supervisors are 0.047 points healthier than nonsupervisors. Here, managers’ health remains significantly different from both of the other
occupational grades even when other socio-economic variables, such as income and education,
130
Table 7.3: Cross-section regression equations for subjective ill-health, visits to the doctor, and mental strain
1
Explanatory Variables
Manager
Supervisor
Year Dummies
Age
Age-squared
Female
Married
Smoker
Education Level
College
Graduate
Logarithm of Income (at
2005 living costs)
Logarithm of work hours
Constant
-0.185
(6.24)**
-0.047
(2.22)*
2
Subjective Ill Health
3
4
-0.164
(5.59)**
-0.040
(1.92)
-0.130
(4.32)**
-0.031
(1.50)
-0.288
(3.39)**
-0.064
(1.06)
Jointly
Significant
0.006
(1.79)
-0.000
(1.17)
0.039
(3.32)**
-0.024
(1.79)
0.163
(12.34)**
Jointly
Significant
0.012
(3.39)**
-0.000
(2.73)**
0.020
(1.46)
-0.028
(2.04)*
0.155
(11.65)**
0.004
(0.27)
-0.055
(2.47)*
-0.030
(5.04)**
0.033
(1.99)*
5
Visits to the Doctor
6
7
-0.155
(1.84)
0.075
(1.25)
-0.139
(1.61)
0.056
(0.93)
-0.761
(4.21)**
-0.264
(2.06)*
Jointly
Significant
-0.053
(5.58)**
0.001
(5.25)**
0.874
(25.81)**
0.036
(0.93)
0.066
(1.76)
Jointly
Significant
-0.058
(5.73)**
0.001
(5.40)**
0.904
(22.86)**
0.036
(0.91)
0.060
(1.56)
8
Mental Strain (GHQ)
-0.648
(3.62)**
-0.164
(1.29)
-0.697
(3.80)**
-0.152
(1.18)
Jointly
Significant
0.258
(12.57)**
-0.003
(11.92)**
1.328
(18.27)**
-0.499
(5.91)**
0.079
(0.97)
Jointly
Significant
0.254
(11.76)**
-0.003
(11.22)**
1.277
(15.02)**
-0.504
(5.94)**
0.112
(1.36)
0.042
(1.04)
-0.201
(3.13)**
0.032
(1.85)
-0.020
(0.41)
2.035
1.766
1.693
1.799
2.319
2.317
10.815
(320.17)**
(26.86)**
(20.21)**
(98.60)**
(12.33)**
(9.66)**
(276.67)**
Observations
18282
18282
18282
18233
18233
18233
17184
R-squared
0.0023
0.03
0.04
0.0012
0.04
0.04
0.0012
Absolute value of t-statistics in parentheses; * significant at 5% level; ** significant at 1% level
This is a full-sample regression that combines all promoted control and treatment group samples. Subsequent analysis separates these.
9
0.045
(0.53)
0.271
(1.98)*
0.035
(0.96)
-0.207
(2.04)*
5.186
(12.86)**
17184
0.03
5.806
(11.29)**
17184
0.03
131
are added to the regression equation. For the number-of-visits-to-the-GP variable, the evidence
in Table 7.3 is not as clear. The coefficients, although beginning with some significance for
managers, only border on the 10% level of significance once controls are added. Mental strain
follows a similar, and slightly stronger, pattern. Managers have lower levels of mental strain and
the coefficients are well-defined.
In Table 7.3, smoking appears to have the negative consequences that might be expected.
Ceteris paribus, women rate their own health worse. Moreover, they go to their doctor more
often, and have higher mental strain. Education appears to contribute an important effect across
all measures of an individual’s health. Income is only significant for subjective ill-health.
Interestingly, the least-educated and the married individuals have significantly lower levels of
mental strain. Concentrating on subjective health, we can evaluate the importance of a position
of control in the workplace in light of other variables. Being a manager, for instance, appears to
have a similar health impact to smoking and has over twice the benefit of being educated to
degree level. Even once we control for the individual’s access to resources and his or her
education level, large benefits from job seniority still remain. Although consistent with decades
of previous evidence on the positive association between health and socio-economic status, Table
7.3 should not be viewed as proof of causality.
Later tables move to longitudinal patterns. They draw upon samples of up to 18,000
person-year observations and approximately 1000 promotions. The sub-tables in Table 7.4, for
example, report both the raw means and the differences between groups - both with and without
controls17 - for subjective ill-health, visits to the doctor, and GHQ mental strain. The data here
run from T-1 to T+3. In other words, the job promotion itself occurs at time period T, and data
17
The regressions with controls are available upon request
132
Table 7.4: Ill-health among the non-promoted non-supervisors and those promoted to manager (at time T)
Subjective Ill Health
Non-Promoted Group
Promoted Group
Time Period
T-1
T
T+1
T+2
T+3
Change Over Time
(T)-(T-1)
(T+3)-(T-1)
(T+1)-(T)
(T+3)-(T)
N
15911
15911
15911
15911
15911
Mean
2.02
2.03
2.06
2.07
2.08
Standard
Deviation
0.80
0.80
0.81
0.81
0.83
N
331
331
331
331
331
Mean
1.86
1.81
1.82
1.87
1.84
Standard
Deviation
0.81
0.74
0.72
0.80
0.76
15911
15911
15911
15911
0.01
0.06
0.02
0.05
0.82
0.90
0.82
0.89
331
331
331
331
-0.05
-0.02
0.01
0.03
0.82
0.87
0.72
0.81
Difference in
Mean across
Groups
-0.16**
-0.23**
-0.24**
-0.20**
-0.24**
Difference in
Mean across
Groups
(with controlsa)
-0.13**
-0.16**
-0.17**
-0.13**
-0.15**
-0.06
-0.08
-0.01
-0.01
-0.04
-0.05
-0.02
-0.02
Difference in
Mean across
Groups
-0.10
-0.32*
-0.43**
-0.37**
-0.44**
Difference in
Mean across
Groups
(with controlsa)
0.08
-0.15
-0.25^
-0.16
-0.16
-0.22^
-0.34*
-0.11
-0.12
-0.22^
-0.30*
-0.10
-0.08
Difference in
Mean across
Groups
-0.42
-1.07**
-0.19
0.05
0.10
Difference in
Mean across
Groups
(with controlsa)
-0.33
-0.96**
-0.10
0.15
0.19
-0.64*
0.52^
0.88**
1.16**
-0.66*
0.42
0.86**
1.08**
Visits to the Doctor
Non-Promoted Group
Promoted Group
Time Period
T-1
T
T+1
T+2
T+3
Change Over Time
(T)-(T-1)
(T+3)-(T-1)
(T+1)-(T)
(T+3)-(T)
N
15869
15869
15869
15869
15869
Mean
1.79
1.80
1.83
1.84
1.89
Standard
Deviation
2.27
2.30
2.36
2.36
2.44
N
332
332
332
332
332
Mean
1.69
1.48
1.40
1.47
1.45
Standard
Deviation
2.21
1.80
1.66
2.00
2.00
15869
15869
15869
15869
0.01
0.10
0.03
0.09
2.33
2.70
2.36
2.66
332
332
332
332
-0.21
-0.24
-0.08
-0.03
2.03
2.51
2.03
2.17
Mental Strain (GHQ)
Non-Promoted Group
Promoted Group
Time Period
T-1
T
T+1
T+2
T+3
Change Over Time
(T)-(T-1)
(T+3)-(T-1)
(T+1)-(T)
(T+3)-(T)
Numbers subject to rounding
N
14925
14925
14925
14925
14925
Mean
10.70
10.81
10.88
10.95
11.04
Standard
Deviation
4.79
4.83
4.88
4.91
4.99
N
321
321
321
321
321
Mean
10.27
9.75
10.69
11.00
11.14
Standard
Deviation
5.59
4.54
4.80
5.58
5.16
14925
14925
14925
14925
0.12
0.35
0.07
0.23
4.99
5.57
5.00
5.47
321
321
321
321
-0.52
0.87
0.95
1.39
5.59
6.96
4.37
5.92
^ significant at 10% level; * significant at 5% level; ** significant at 1% level
a. For the time-period regressions, time dummies, age, gender, smoking and marital status, education, income and hours of work are used as
controls at the appropriate time point. For the change over time regressions, controls are time dummies, age, gender, smoking and marital status,
education at T and also the appropriate changes that took place in income and hours of work.
133
are also given on the person the year before that promotion, and for each of three years after that.
18
Table 7.4 studies ‘large’ promotions. These are for people who become managers. The
case is particularly interesting because these people initially begin in a non-supervisory role: this
group of individuals, it might be said, are given the greatest boost to their status.
Importantly, in Table 7.4 there is evidence that the (future) promotees begin with much
better health. Subjective ill-health is significantly better -- compare the mean of 2.02 with the
mean of 1.86 -- to begin with than among those who will not be promoted. At period T, the
promoted group even visit the doctor significantly fewer times and have lower mental strain.
These health differences persist but do not significantly improve for the subjective ill-health
variable up until T+3. However, recently promoted managers go on to visit their doctor less
often in both the short and medium term. The effect is large. As the mean of Visits in the
combined sample is 1.78, the estimates imply approximately a 20% fall in visits to the doctor
after promotion to manager. Moreover, there is consistency in this evidence for an improvement
from time T. This appears to be more encouraging for the claim that taking a promotion
improves health, although a critic might potentially raise an alternative explanation, namely, that
managers simply become short of time.
Contrastingly, mental stress actually worsens after promotion. This can be seen in the
bottom panel of Table 7.4. The null of zero on the key difference-in-difference can be rejected.
There appear to be some benefits to psychological health in the run up to promotion at T.
The promoted group have significantly better psychological health at T and the value -0.64* from
18
It is possible to start and end the analysis at different time points, but that greatly reduces the sample size without
affecting our principal findings. The results of extending the analysis up until T+5, for example, is shown in table
7.6
134
T to T-1 indicates that this improvement is significant. However, as soon as the promoted group
reach T+1, any health improvements in the lead up to promotion have dissipated. By T+3 they
have the same mental strain levels as those not promoted and when compared to T there is strong
evidence that the promoted group suffer a relative worsening in their mental strain. See, for
instance, the difference in Table 7.4 of 1.16 points on a GHQ mental strain score by period T+3.
This is a substantial deterioration compared to the non-promoted controls who remain at their
original level of seniority.
This result runs counter to the hypothesis that promotion improves health: those who
obtain the largest boost to status here show the clearest deterioration in mental health.
Other factors are associated with health and promotion. The second-last column in Table
7.4 shows the differences once the time period, age, education, gender, marital status, income,
hours worked and smoking status at T are held constant.
Similar results are found -- though sometimes less sharply as would be expected -- in the
full sample of all job-promotions (many of which are ‘smaller’ promotions, such as from nonsupervisor to supervisor19). These results are depicted in Table 7.5. Here psychological health
among promotees has worsened in T+3 by 0.62 GHQ points.
A possible conclusion from these results is that causality does not run from status to
health. In part, it seems that the healthiest individuals get promoted, but this result alone does
not fully explain the social health gradient initially observed in Table 7.3. Arguably the crosssectional association is driven by a third unobservable factor, such as behavioural or genetic
factors. If there is a large benefit from being promoted, as potentially suggested by the Whitehall
studies, then it is undetectable across our observed time frame. Good health, at least in the long
19
When the smaller promotion groups are analysed separately there is no support for the status-causes-health theory
135
Table 7.5: Ill-health among the non-promoted and those promoted to any category (at time T)
Subjective Ill Health
Non-Promoted Group
Promoted Group
Time Period
T-1
T
T+1
T+2
T+3
Change
Over Time
(T)-(T-1)
(T+3)-(T-1)
(T+1)-(T)
(T+3)-(T)
N
17169
17169
17169
17169
17169
Mean
2.02
2.03
2.06
2.07
2.08
Standard
Deviation
0.80
0.80
0.82
0.81
0.83
17169
17169
17169
17169
0.01
0.06
0.02
0.05
0.82
0.90
0.82
0.88
N
1113
1113
1113
1113
1113
1113
1113
1113
1113
Mean
1.90
1.88
1.90
1.96
1.95
Standard
Deviation
0.81
0.77
0.75
0.80
0.79
Difference in
Mean across
Groups
-0.12**
-0.15**
-0.15**
-0.11**
-0.13**
Difference in Mean
across Groups
(with controlsa)
-0.09**
-0.11**
-0.11**
-0.06**
-0.07**
Instrumental
Variables
Estimation
-0.43**
-0.28^
0.12
0.05
0.02
-0.02
0.05
0.02
0.07
0.82
0.91
0.76
0.86
-0.03
-0.01
0.00
0.02
-0.02
-0.00
-0.00
0.02
0.06
-0.00
0.20
-0.07
Difference in
Mean across
Groups
-0.13^
-0.28**
-0.32**
-0.16*
-0.23**
Difference in Mean
across Groups
(with controlsa)
-0.01
-0.16*
-0.17*
-0.01
-0.03
Instrumental
Variables
Estimation
-0.65^
-0.11
0.05
0.48
1.18**
-0.15*
-0.10
-0.04
0.06
-0.15*
-0.07
-0.02
0.08
0.16
0.68^
-0.03
0.30
Difference in
Mean across
Groups
-0.33*
-0.68**
-0.17
-0.06
-0.05
Difference in Mean
across Groups
(with controlsa)
-0.22
-0.57**
-0.08
0.03
0.03
Instrumental
Variables
Estimation
-2.35**
0.03
1.03
1.07
1.38^
-0.34*
0.17
0.27
0.28
0.51**
0.62**
-0.37*
0.11
0.19
0.19
0.48**
0.55**
1.14
1.85*
1.76*
2.09*
0.60
0.80
Visits to the Doctor
Non-Promoted Group
Promoted Group
Time Period
T-1
T
T+1
T+2
T+3
Change Over
Time
(T)-(T-1)
(T+3)-(T-1)
(T+1)-(T)
(T+3)-(T)
N
17120
17120
17120
17120
17120
Mean
1.78
1.80
1.83
1.83
1.89
Standard
Deviation
2.28
2.32
2.36
2.37
2.45
17120
17120
17120
17120
0.01
0.11
0.03
0.09
2.32
2.70
2.35
2.66
N
1113
1113
1113
1113
1113
Mean
1.66
1.52
1.51
1.67
1.66
Standard
Deviation
2.13
2.00
1.98
2.16
2.18
1113
1113
1113
1113
-0.14
0.01
-0.01
0.15
2.13
2.58
2.14
2.38
Mental Strain (GHQ)
Non-Promoted Group
Promoted Group
Time Period
T-1
T
T+1
T+2
T+3
Change Over
Time
(T)-(T-1)
(T+1)-(T-1)
(T+2)-(T-1)
(T+3)-(T-1)
(T+1)-(T)
(T+3)-(T)
N
16127
16127
16127
16127
16127
Mean
10.68
10.80
10.86
10.92
11.03
Standard
Deviation
4.75
4.80
4.86
4.89
4.97
16127
16127
16127
16127
16127
16127
0.12
0.19
0.25
0.35
0.06
0.23
4.95
5.20
5.40
5.54
4.97
5.44
Numbers subject to rounding
N
1057
1057
1057
1057
1057
Mean
10.34
10.12
10.69
10.86
10.98
Standard
Deviation
5.06
4.40
4.68
5.10
5.02
1057
1057
1057
1057
1057
1057
-0.22
0.35
0.52
0.63
0.57
0.85
5.17
5.42
5.91
6.11
4.55
5.49
^ significant at 10% level; * significant at 5% level; ** significant at 1% level
a. For the time-period regressions, time dummies, age, gender, smoking and marital status, education, income and hours of work are used as
controls at the appropriate time point. For the change over time regressions, controls are time dummies, age, gender, smoking and marital status,
education at T and also the appropriate changes that took place in income and hours of work.
136
Table 7.6: Difference-in-Difference ((T+3)-(T-1)) estimates (with controls) for individuals working in the
public sector and in the manufacturing industry, those individuals who stay at the same address across all 5
years and those who stay in the promoted position up until T+5
Promoted Group
Public Sector1
Health Measure
Manufacturing
Industry
Difference-in-
a
Same address
across all 5 years
Remain in promoted
a
position until T+5a
((T+3)-(T-1))
((T+3)-(T-1))
((T+3)-(T-1))
((T+5)-(T-1))
Difference
Promoted to
Subjective Ill-Health
0.29 (17/839)
-0.31^ (36/2379)
-0.06 (165/10420)
-0.03 (150/8622)
Manager (from
Visits to the Doctor
-1.06 (17/855)
-0.49 (36/2378)
-0.48* (166/10397)
-0.33 (150/8584)
Non-supervisor)
Mental Strain
1.91 (17/812)
0.26 (35/2246)
0.11 (160/9710)
-0.10 (142/8048)
Any Promotion
Subjective Ill-Health
0.08 (69/1040)
-0.09 (158/2931)
-0.02 (636/12268)
-0.04 (499/8972)
at T
Visits to the Doctor
-0.39 (65/995)
-0.12 (154/2807)
-0.10 (635/12218)
-0.14 (498/8928)
Mental Strain
-0.41 (64/949)
-0.19 (144/2653)
0.17 (598/11437)
-0.18 (468/8380)
^ significant at 10% level; * significant at 5% level; ** significant at 1% level
a. The numbers in brackets refer to the numbers in treatment/sample group
As an aid to reading this table, the top left number of 0.29 (17/839) means that in the public sector a promotion to manager
increases subjective ill-health by 0.29 points, and there are 17 people in this category, with 839 in the whole sample.
term, apparently does not follow from job promotion. The decline in visits to the doctor of the
promoted group in the third column of Table 7.6 is the closest to evidence for the contrary.
7.7
Objections and Counter Arguments
There is inevitably some noise in the data. Hence (Issue #1) the findings might in
principle be the result of a Type II error. Moreover, promotion may be non-random in influential
ways (Issue #2). These include the possibility that (Issue #3) the promoted groups endured
substantial health deterioration relative to the control group in the years leading up to the
promotion, with promotion merely restoring it. Alternatively (Issue #4) the individuals who
really improve in health might somehow be missed from our sub-sample. This could occur if an
individual promoted at T went on to then get demoted or promoted within the three years, or
even left the work-force altogether.
We probe these possible explanations.
137
7.7.1 Issue #1: Noise
A simple check on the possibility that our negative conclusions stem from sheer noise and
Type II errors is temporarily to ignore the standard errors and to focus on coefficient signs. But,
when this is done, even the coefficient signs do not support a status-causes-health theory.
Our data set necessarily aggregates across different kinds of work and different sectors.
Therefore a further argument could be made against the occupational status variable. Whilst we
expect the individual’s answer to the status variable to have a large degree of internal
consistency, there may be variation across industries. We test this possibility by carrying out the
same analysis on individuals who work and remain in the manufacturing industry and again
separately for the public sector. As shown in Table 7.6 the treatment group declines substantially
in size, which make it difficult to pick up significant differences. However, if again we focus
only on the coefficient signs for this relatively homogenous set of individuals, there are,
consistently with the null of randomness, 9 instances out of 18 in which there is a negative value.
Table 7.6 further shows the difference-in-difference estimates for those that stayed at the
same address for the full period and separately for those that stayed in the promoted position
until T+5. We examine these groups of individuals since those that stay at the same address are
more likely to have gained a promotion within the same company than those who moved and the
period of analysis in our main analysis may have been too small to observe health benefits. The
sample sizes remain reasonably large but there are no significant differences, other than
individuals visiting the doctor less, in either of these tests.
7.7.2 Issue #2: Endogeneity
An important issue is that promotion is potentially non-random and endogenous. We try
an overcome this issue by using the variation in promotion rates in each year across industries as
138
an instrument for promotion. It is assumed that promotion varies across industries but that ability
of the individuals within the industry does not. A similar approach is taken by Anderson and
Marmot (2007). The results of using instrumental variables are shown for the any promotion
group in the final column of Table 7.5. Generally the coefficients are larger than the previous
estimates and are consistent with the paper’s previous findings; healthy individuals get promoted
and promotion brings on substantial mental strain.
It is perhaps also worth noting that the published literature on the cross-section
association, which argues promotion has a causal effect, ignores possible endogeneity bias.
7.7.3 Issue #3: Poor Health as a Predictor
To deal with the second objection, it is necessary to determine whether, prior to
promotion, poor health predicts promotion at T. Table 7.7’s evidence suggests not. The reverse
holds.
Table 7.7: Probit equations using health at T-1 as a predictor of promotion
Dependent Variable:
Explanatory Variables at T
Ill-Health at T-1
Age
Female
Married
Smoker
Education Level
College
Graduate
1
Any Promotion at T
Subjective Ill
Health
2
3
Visits to the GP
Mental Strain
-0.070
(3.64)**
-0.001
(0.08)
-0.003
(0.98)
-0.006
(4.25)**
-0.216
(7.12)**
-0.007
(4.34)**
-0.216
(6.96)**
-0.006
(3.87)**
-0.214
(6.80)**
0.106
(3.08)**
0.032
(0.89)
0.109
(3.17)**
0.021
(0.61)
0.104
(2.94)**
0.023
(0.64)
0.308
(8.83)**
0.788
(17.96)**
0.312
(8.96)**
0.799
(18.22)**
0.301
(8.42)**
0.777
(17.26)**
-1.325
-1.459
-1.436
(18.58)**
(23.09)**
(20.50)**
18282
18233
17184
Observations
Absolute value of t-statistics in parentheses; * significant at 5% level; ** significant at 1% level
Constant
139
7.7.4 Issue #4: Sample Changes
Promoted individuals are lost from the sample on three accounts: they get further
promoted or demoted within the three years; they leave the workforce; or they exit the BHPS
completely. On the last point, not a great deal can be done. However, it is hard to see,
intuitively, why the particularly healthy people should exhibit high attrition from the panel. The
first two can be tracked with comparisons against control and treatment groups. Table 7.8 mirrors
the previous estimation of health changes from T-1 to T+3. A separate comparison is made for
those who stay in employment, and those who leave the workforce. There is no evidence that
those who subsequently change roles become healthier. The only clear outcomes arise for those
who leave the workforce completely, but the coefficients indicate a worsening of health. Similar
effects are found in the changes in health across other time periods.
140
Table 7.8: Regressions showing health differences across promoted groups, and those who subsequently left
the workforce or changed role
1
Dependent
Variable:
Explanatory
Variables at T
Promoted at T
2
3
Promoted to Manager at T
(from Non-Supervisor)
Subjective
Visits to
Mental
Ill Health
the GP
Strain
4
5
6
Any Promotion at T
Subjective Ill
Visits to
Mental
Health
the GP
Strain
-0.069
(1.37)
-0.057
(0.06)
-0.295
(1.95)
2.732
(1.01)
0.407
(1.27)
8.863
(1.58)
-0.006
(0.21)
0.451
(1.58)
-0.073
(0.86)
1.540
(1.79)
0.201
(1.12)
3.415
(1.92)
Promoted at T but
subsequently
changed role
-0.035
(0.90)
-0.123
(1.08)
0.073
(0.30)
-0.006
(0.34)
-0.090
(1.68)
0.076
(0.67)
Age
0.001
(1.87)
0.005
(0.32)
0.009
(4.49)**
0.039
(0.91)
-0.024
(5.38)**
-0.061
(0.68)
0.001
(1.96)*
0.006
(0.47)
0.008
(4.23)**
0.036
(0.94)
-0.021
(5.37)**
-0.063
(0.78)
0.028
(1.80)
0.018
(1.13)
-0.100
(2.12)*
0.063
(1.33)
0.553
(5.44)**
-0.015
(0.15)
0.022
(1.59)
0.023
(1.62)
-0.070
(1.66)
0.108
(2.53)*
0.525
(5.86)**
-0.057
(0.63)
0.012
(0.69)
-0.012
(0.48)
0.001
(0.03)
-0.034
(0.44)
-0.053
(0.50)
0.019
(0.12)
0.013
(0.86)
-0.014
(0.64)
-0.014
(0.33)
-0.012
(0.18)
-0.004
(0.04)
0.018
(0.13)
-0.021
-0.194
-0.161
-0.017
-0.191
-0.155
(1.10)
-0.003
(0.51)
(3.46)**
0.039
(2.19)*
(1.34)
0.141
(3.72)**
(0.97)
-0.006
(1.10)
(3.65)**
0.027
(1.66)
(1.39)
0.133
(3.82)**
-0.014
-0.244
0.909
-0.008
-0.210
(0.46)
(2.73)**
(4.75)**
(0.30)
(2.58)**
16831
16789
15801
21457
21401
Observations
0.0012
0.0026
0.0043
0.0011
0.0023
Pseudo R-squared
Absolute value of t-statistics in parentheses; * significant at 5% level; ** significant at 1% level
0.840
(4.87)**
20181
0.0038
Promoted at T but
left workforce
Female
Married
Smoker
Education Level
College
Graduate
Logarithm of
work hours
Logarithm of
Income (at 2005
living costs)
Constant
141
7.8
Conclusion
This paper is one of the first fully longitudinal inquiries into the hypothesis that status
makes people healthy. It draws upon data from a nationally representative sample of employees.
The paper finds little evidence that promotion improves a person’s health.20 In fact, after they are
promoted, the GHQ mental health of managers typically deteriorates, and in a way that goes
beyond a short-term change. This result is a new one in the literature. Workers promoted in time
period T alter in one other way. From that point, they visit their doctor approximately 20% less
frequently, although this may simply be because new managers have less time for everything. 21
Further research will be needed before we have a complete understanding of the links
between human status and human health.
20
Our negative findings have one interpretive advantage: the likely bias goes in the other direction. If promotion
really improves people’s health, then to make sense of our results using a status-causes-health theory it would be
necessary to believe, against common intuition, that individuals with a high probability of deteriorating health are
the ones most likely to gain an increase in workplace seniority.
21
We are not sure how to reconcile these results with the more supportive ones that have been found, using data on
Oscar and Nobel Prize winners and nominees, in the work of Redelmeier and Singh (2001) and Rablen and Oswald
(2007). One conjecture might be that it takes a major change in status to make a difference to physical and mental
health; perhaps health does not respond in a linear dose-response way, but rather is a strongly convex function of
status.
142
7.9
Appendix
7.9.1 Notes to Tables
Regressions include observations across all years separated by promoted to manager/supervisor
groups where specified.
Description of Variables
Subjective Ill-health
Visits to the doctor
Mental strain
The individual’s rating of their health status over the last twelve months,
where 1=excellent, 2=good, 3=fair, 4=poor, 5=very poor
The number of visits to their General Practitioner in the past year. This
is a categorical variable. None, one or two, three to five, six to ten, or ten
plus are the available options. These are recoded to be the minimum
value in each category (0, 1, 3, 6 and 10). This is an underestimation, but
ensures consistency across categories and individuals.
This variable is a 1 to 36 scale of the level of mental strain obtained
from the General Health Questionnaire (GHQ). There are 12, zero to 3
point questions that include; among others, the individuals ability to
overcome problems, their decision capabilities, sleep, concentration and
general feelings of depression.
Non-supervisor
Individual’s managerial duties are neither manager or supervisory, yet
they are still in employment
Supervisor
Individual’s managerial duties are that of a supervisor or foreman
Manager
Individual’s managerial duties are that of a manager
Promoted at T
Individuals promoted at T (to supervisor or manager) and remained until
T+3
Promoted at T but left workforce
Individuals promoted at T (to supervisor or manager) but did not remain
until T+3 as they left the workforce at some point
Promoted at T but subsequently changed role
Individuals promoted at T (to supervisor or manager) but did not remain
until T+3 as they subsequently changed role through further promotion
or demotion
Age
Individual’s age
Female
Individual is female (excluded dummy: male)
Married
Individual is married (excluded dummy: all non married individuals
including single, widow, divorced and separated)
Smoker
Individual is a smoker (excluded dummy: all non-smoking individuals)
Education level
Specifies the individual’s highest level of education obtained.(College
and Graduate dummies used with excluded dummy for those who either
left after high school or dropped out)
Logarithm of work hours
The logarithm of the number of hours an individual works in a typical
week, including overtime
Logarithm of income (at 2005 living costs)
The logarithm of an individual’s own annual income, with all years,
deflated to 2005 living costs
143
7.9.2 Sample Construction
All three promoted groups are of interest in the analysis: those promoted from nonsupervisor to both supervisor and manager, and those promoted from supervisor to manager.
Individuals in employment and indicating their position at T were sourced from every wave of
the BHPS. Each observation at T was tracked from T-1 through to T+3, and where available, the
health measures taken. Occupational position changes were then analyzed and two groups, both
control and treatment, created.
7.9.2.1 Control Groups
A control group of those not promoted is required. Those who maintained the same
position (non-supervisor or supervisor) for the full five years were a control group for the
relevant promoted group.
7.9.2.2 Treatment Groups
Those who were initially in the control group at T-1 but promoted at T, and maintained
this until at least T+3, made up the treatment group. Inevitably, since requiring a full five years
of data, waves 1, 13, 14 and 15 could not be included. This makes an overall sample size of
approximately 18,000 observations. It is a balanced panel: individuals give answers to the health
question for each of the five years. Thus the sample size varies depending on the health variable
under analysis.
Those promoted at T but who did not remain in the promoted position for all three years
were separately coded -- depending on whether observed as still working, changing roles, or
leaving work entirely. These groups are used in the analysis in Section 7.7.
144
By the nature of our sample construction, some individuals appear as multiple
observations. This occurs in two circumstances. First, an individual may maintain a role for
longer than 5 years. Second, a single individual may enter on a number of occasions if they
experience a break in employment of which at least five years of employment exist either side. In
both scenarios it is difficult to know which observation should be included as all spells contain
valuable information. Both are kept as observations in the analysis.
7.9.3 Definition of GHQ Mental Ill-health
A GHQ score, defined to lie between zero and 36, is a psychiatric screening instrument
that is as an amalgamation of answers to the questions: Have you recently:
1. Been able to concentrate on whatever you are doing?
2. Lost much sleep over worry?
3. Felt that you are playing a useful part in things?
4. Felt capable of making decisions about things?
5. Felt constantly under strain?
6. Felt you could not overcome your difficulties?
7. Been able to enjoy your normal day-to-day activities?
8. Been able to face up to your problems?
9. Been feeling unhappy and depressed?
10. Been losing confidence in yourself?
11. Been thinking of yourself as a worthless person?
12. Been feeling reasonably happy all things considered?
145
CHAPTER 8
8
CONCLUSION
This thesis has argued that, although there is still much to learn about human happiness,
our understanding can be enhanced by taking a multi-disciplinary approach. This thesis merged
psychological concepts with economic methodology to provide answers to questions that have
economic and social interest. Specifically, (a) an individual’s rank income was shown to be a
stronger predictor of life satisfaction than either absolute income or an individual’s income
relative to those around them; (b) it was argued that income is an inefficient way of helping
individuals overcome psychological distress and that we should consider channelling more
resources into achieving mental health; (c) personality was shown to be a useful tool for
understanding the influence of demographic circumstances on subjective well-being; and (d)
improvements to an individual’s occupational status do not feed through to better health. This
chapter summarises these findings, addresses the limitations to the research and highlights areas
in which subjective well-being research can develop.
8.1
Summary
One of the dominant questions in economic subjective well-being research is: Does
money buy happiness? Money is of course central to economic thinking so this question is
essential. However, the evidence seems to suggest that, whilst money may bring some extra wellbeing, this extra well-being seems to be quite small relative to that produced by other potentially
more important factors such as social relationships, mental health and personality. Although
money seems to be less important than one typically might expect it is nevertheless a central
topic of subjective well-being research. It is important that research seeks to understand why
there is such a huge gap between how much happiness money is believed to bring and how much
146
it actually does. Such a puzzle can be in part put down to relative effects and the process of
adaptation (Clark, Frijters et al., 2008). However, the nature of such comparisons, to others and
past selves, need to be better understood. Chapter 2 presented a rank-based model of social
comparisons. In this model, individuals, instead of comparing to the mean income of those
around them, make simple binary comparisons, distinguishing whether another individual is in
either a better or worse position than themselves. This simple, and more cognitively realistic,
model of comparison (Stewart et al., 2006) results in a rank based income model. This relative
judgment model has support from psychology (Parducci, 1995) and we showed that rank income
significantly predicts satisfaction with life and explained more variation than either absolute or
mean reference income accounts.
Chapter 3 highlighted how relatively unimportant and inefficient money is at helping
individuals overcome psychological distress. This chapter began by pointing out that monetary
equivalences can be given to various life events using subjective well-being equations. The
chapter then showed that the use of such valuations have been suggested to be appropriate for
compensating individuals after traumatic life events. However, the amount of money needed to
compensate an individual would be high and Chapter 3 argued that this reflects the relative
unimportance and inefficiency of money at restoring lost well-being. The chapter proposed a
more efficient alternative, psychological therapy, referring to evidence from medicine to support
this claim. In the face of rising mental illness (Michaud et al., 2001) and stagnating well-being
levels in developed countries (Easterlin, 1995), the argument was extended to suggest that
societies may be better off by channelling resources into mental health as opposed to solely
focusing on income growth.
147
Individual differences and personality are treated very differently in psychology and
economics and much of the work in this thesis sought to highlight this divide. Psychologists have
spent considerable resources on developing reliable and valid measures of individual personality.
In spite of personality being one of the largest and most consistent predictors of well-being
(Diener & Lucas, 1999) personality measures have not typically been embraced by economists in
their subjective well-being research. Chapters 4, 5 and 6 sought, first and foremost, to highlight
to economic subjective well-being researchers the fact that personality can be reliably measured.
These chapters then illustrated useful ways in which personality measures can be used to answer
questions that are of interest to economists.
Chapter 4 specifically suggested an estimation technique that incorporated fixed
personality measures as an alternative to the current dominant estimation technique that seeks to
explain only the within-person variation in well-being. The benefit of the within-person approach
is that such an approach allows researchers to indirectly control for fixed and assumed
unobservable personality characteristics. The alternative estimation technique proposed in
Chapter 4 is argued to produce superior estimates on the effect on well-being of variables that
have low within-person variations, and this includes characteristics of economic interest, such as
income, marital status and retirement. Chapters 5 and 6 further highlighted the usefulness of
personality measures by presenting evidence that personality interacted with important economic
variables. Chapter 5 showed how different personality types get different marginal utilities out of
a given income rise and Chapter 6 showed that being conscientious can be a serious risk factor
during unemployment. Psychologists have generally been sceptical about researching the effect
of demographic factors on well-being since they often explain only a very small proportion of
well-being (Argyle, 1999). However, the work presented here illustrates to psychologists that
148
understanding the causes and correlates of high well-being might be better understood by
combining demographic and personality characteristics (Gutierrez et al., 2005).
Finally, Chapter 7 applied methodological rigour to a topic that, despite the wide claims
of causality, has so far only relied on cross-sectional data. It is often argued that improvements to
job status lead directly to improvements in health, explaining why people with high job status
generally have better health (Marmot, 2004; R. Wilkinson, 2001). This chapter showed, using a
difference-in-difference technique on longitudinal data, that the strong cross-sectional association
typically seen between occupational status and health does not appear to have a causal
explanation running from status to health. This chapter highlighted the importance of analysing
longitudinal data and showed that an improvement to occupational status – a promotion – may in
fact increase mental strain and allow less time to visit the doctor. This suggested promotion may
in fact be detrimental to health and not, as conventional wisdom might suggest, unequivocally
good.
8.2
Implications for Economic-Psychology Subjective Well-Being Research
8.2.1 The Use of Large Data Sets in Psychology
Economists routinely make use of publicly available datasets that survey large portions of
the population. Such data sets are normally representative and are carried out over a number of
years. Such data sets have not only allowed economists to answer important economic questions,
but have also enabled the routine use of more advanced statistical techniques that help uncover
causal links to higher well-being. However, the use of these data sets also represents one of the
biggest limitations to the research carried out in this thesis. All the chapters, baring Chapter 3,
relied on nationally representative longitudinal data sets. Such data sets are used by a wide range
of researchers across very broad topics and therefore have extremely limited space available for
149
questions. The thesis was therefore limited by the availability of questions. For example, singleitem scales had to be relied on and such scales are known to sometimes lead to an
underestimation of true effect sizes. Sometimes the most valid and reliable scales are not always
chosen by the compilers of nationally representative data sets and this is an important topic for
the future.
Psychologists, on the other hand, are often unaware that such data sets even exist. The use
of such data sets by psychologists has the potential to drastically improve their research.
Psychologists also have greater concern for using the best possible measures so it is likely that if
psychologists become familiar users of these large data sets then better quality measures will be
included. For example, the well validated satisfaction with life scale (Pavot & Diener, 1993b)
would be a useful addition. It is also likely that more psychological questions would then be
included and this has the potential to then benefit economic research.
8.2.2 Improved Understanding of Social Comparisons
It seems clear that individuals compare with one another but the empirical evidence in
economic subjective well-being research seems to suggest that this is a mainly negative process.
On the whole having higher earning others around is detrimental for individual well-being (e.g.
Luttmer, 2005). However, it seems unlikely that this is a one way process. There is some
evidence to suggest that being surrounded by individuals that earn higher incomes can actually
be positive for well-being (Clark, Kristensen, & Westergard-Nielsen, 2009b; Senik, 2004). It is
likely that being around higher earning others can be in some ways positive; whilst at the same
time can also be negative. The coefficient on the average income of those in one’s comparison
group, therefore, reflects the process that is most dominant. We need a much clearer
150
understanding of social comparison and to do this it would be worth drawing on ideas from the
social comparison literature in psychology.
The social comparison literature seeks to understand how and why individuals socially
compare to one another. For example, individuals may compare with one another in order to gain
information about how well one is doing in life (Festinger, 1954), or to learn how one’s
performance can be improved (S. R. Wilson & Benner, 1971). Comparisons undertaken for either
reason can be both beneficial and detrimental to the individual’s life (Buunk et al., 1990).
However, it now seems clear that, with the addition of evidence presented in Chapter 2,
individuals generally engage in upward comparison. But why would individuals engage in
upward comparison if it were detrimental to well-being? Such a comparison is also likely to
bring benefits by demonstrating ways in which the individual could improve their life (Buunk et
al., 1990). It has been suggested, however, that upward comparison can take place to indicate
how to improve life without necessarily damaging the individual’s well-being (Taylor & Lobel,
1989). It seems important to understand the reasons behind upward comparison, which current
relative income studies can not quite yet offer. The idea of positive upward comparison has been
interestingly modelled in the economic literature by Falk and Knell (2004) and has been looked
at theoretically in an evolutionary context (L. Samuelson, 2004). The thesis was unable to
explore these issues but further exploration is needed. A greater understanding of the social
comparison literature would help advance the relative income studies. It may be, for example,
that individuals compare negatively with some groups but positively with others. It would be
useful to determine who these groups are and how we can encourage more positive upward
comparisons.
151
Social comparison may also benefit by incorporating the range aspect of RFT. The range
aspect of RFT is the cardinal position of an individual’s income relative to the highest and lowest
incomes of those in their comparison group. Since no data on comparison groups is available it is
not possible to know what low and high earners an individual holds in their mind and deriving
such figures based on the data set was not practical. Chapter 2 was therefore unable to address
this issue but future research may benefit if comparison groups are explicitly determined.
8.2.3 Rank Based Comparisons
Income rank was found to dominate an explanation of life satisfaction. Rank offers a very
simple way of comparing to others and it is likely that there could be other areas in which a rank
based explanation dominates. It has already been shown theoretically that rank can account for
much economic behaviour (Stewart et al., 2006) so rank based comparisons need greater
exploration within economics.
8.2.4 Subjective Well-Being Research and Policy
Subjective well-being research has been developing at a rapid rate and uses the scientific
method to understand what causes humans to be happy. One of the key findings of subjective
well-being research is that average well-being levels have not increased in the developed world
in spite of large increases in GDP. This should be an enormous concern for policy. Affluence has
brought a new set of problems, for instance obesity, materialistic attitudes and impatience (Offer,
2006), which need addressing. Subjective well-being research has the potential to inform the
policy debate and help the public understand which actions are more likely to bring the greatest
improvements to well-being. As argued in this thesis the use of psychological therapy could be
one such approach. Subjective well-being researchers need to explore other ways in which health
152
and happiness can be improved. It has been argued that a society targeting well-being indices
would be radically different politically (Marks & Shah, 2005).
The utilitarianists’ original conception of utility argued that the morally correct action
was the one that brought the greatest amount of pleasure. What does this say about countries that
chase economic growth when it appears to bring no increases to well-being? The income rank
explanation to well-being as addressed in Chapter 2 explains why growth is still pursued in spite
of no gains to happiness. Such an explanation suggests that although it is rational for an
individual to try to do better than others, ultimately obtaining status is a zero-sum game.
Improvements to one individual’s status necessarily come at the expense of some else’s status. If
this were so then societies might be better off not engaging in the race for status. The problems of
pursuing status have been extensively discussed (Frank, 1999; Layard, 2006a) but as yet these
powerful ideas have not yet been embraced by policymakers. However, a solution doesn’t
necessarily need to be imposed by an outside body and informing individuals about subjective
well-being research may prevent people from becoming locked in the race for status and enable
them to take action to unilaterally improve the their lives. Currently, however, many remain
sceptical of subjective well-being research (Johns & Ormerod, 2007; W. Wilkinson, 2007) and
this needs to be countered by spending more time validating well-being scales in economics,
encouraging more research with direct applications to policy and fostering public debate around
some of the central issues.
8.2.5 The Link between Health and Occupational Status
Chapter 7 is one of the first studies to show that improved occupational status does not
result in better health. However, in part, the results rely on null effects and this gives rise to the
possibility of a Type 2 error. It could be that our sample sizes were not large enough to uncover
153
an effect, our occupational status measure was measured with considerable error or that the
health measures used were poor. The research was again limited by the data but even if future
researchers do find better measures they must ensure they use longitudinal data.
8.2.6 Personality within Economics
As has been highlighted personality is an area that has gone unexplored within
economics. Personality will almost undoubtedly influence our economic decisions and could
explain and help in our understanding of why there is heterogeneity in individual preferences and
behaviour. It seems difficult to ignore personality research. However, economists are relatively
unfamiliar with the idea that personality is a measurable construct. Part of this thesis’ intention is
to draw economists’ attention to the reliable and valid personality measures that are available.
Future research needs to explore the use of personality measures and further highlight ways in
which personality can answer important economic questions. As has been suggested in Chapter 5
individuals with different personalities may be differently motivated by money. This finding
brings up many questions that need answering. For instance, why do some people value money
more than others? How can this be used to influence behaviour of individuals with different
personality types? Further, it also seems to be the case that conscientiousness is important for an
individual’s economic behaviour and greater understanding of exactly how and why is needed.
The exploration of personality was hampered by the quality of personality measures
available. Not only were they only 3-item scales but they have so far been measured at just one
time point. Improved measures would only add to the quality of the findings already presented
and also enable more interesting questions to be asked. Borghans et al. (2008) have stressed the
importance of using personality to understand behaviour and incentives and there is a small but
burgeoning literature that has begun to do take up this task (Ameriks et al., 2003; Ameriks et al.,
154
2007; Bowles et al., 2001a, 2001b; Groves, 2005; Nyhus & Pons, 2005). This is an important
area of research to develop.
8.3
Conclusion
The use of subjective well-being data has enabled researchers to get closer to
understanding what makes us happy. Psychologists have been using subjective well-being data
for a considerable amount of time and the research area is well developed. Economists have only
relatively recently begun to explore the use of subjective well-being data but the progress has
been rapid. Both disciplines are concerned with a different set of questions and are equipped
with specific methods by which to answer their own set of questions. The main theme of this
thesis has been to merge the subjective well-being research of economists and psychologists by
using psychological concepts and ideas to answer questions of economic importance. This thesis
concluded that (1) rank income, and not absolute income, influenced life satisfaction; (2)
psychological therapy is undervalued and could be better for well-being than aspiring to ever
higher incomes; (3) personality is useful for understanding the contribution of demographic
characteristics to well-being, and (4) increased job status does not lead to improved health. These
findings illustrate that an interdisciplinary approach can advance our knowledge of human
happiness.
155
CHAPTER 9
9
REFERENCES
Abbott, D. H., Keverne, E. B., Bercovitch, F. B., Shively, C. A., Medoza, S. P., Saltzman, W., et
al. (2003). Are subordinates always stressed? A comparative analysis of rank differences
in cortisol levels among primates. Hormones and Behavior, 43(1), 67-82.
Abbott, D. H., Saltzman, W., Schultz-Darken, N. J., & Tannenbaum, P. L. (1998). Adaptations to
subordinate status in female marmoset monkeys. Comparative Biochemistry and
Physiology C-Toxicology & Pharmacology, 119(3), 261-274.
Adams, P., Hurd, M. D., McFadden, D., Merrill, A., & Ribeiro, T. (2003). Healthy, wealthy, and
wise? Tests for direct causal paths between health and socioeconomic status. Journal of
Econometrics, 112(1), 3-56.
Adda, J., Chandola, T., & Marmot, M. (2003). Commentary - Socio-economic status and health:
causality and pathways. Journal of Econometrics, 112(1), 57-63.
Aiken, L. S., & West, S. G. (1991). Multiple Regression: Testing and Interpreting Interactions
Thousand Oaks: Sage Publications.
Ala-Mursula, L., Vahtera, J., Linna, A., Pentti, J., & Kivimaki, M. (2005). Employee worktime
control moderates the effects of job strain and effort-reward imbalance on sickness
absence: the 10-town study. Journal of Epidemiology and Community Health, 59(10),
851-857.
Alesina, A., Di Tella, R., & MacCulloch, R. (2004). Inequality and happiness: Are Europeans
and Americans different? Journal of Public Economics, 88(9-10), 2009-2042.
Alesina, A., & Ichino, A. (2007). Gender based taxation.
156
Alloy, L. B., Abramson, L. Y., Whitehouse, W. G., & Hogan, M. E. (2006). Prospective
incidence of first onsets and recurrences of depression in individuals at high and low
cognitive risk for depression. Journal of Abnormal Psychology, 115(1), 145-156.
Allport, G. W. (1937). Personality: A Psychological Interpretation. New York: Henry Holt.
Ameriks, J., Caplin, A., & Leahy, J. (2003). Wealth accumulation and the propensity to plan.
Quarterly Journal of Economics, 118(3), 1007-1047.
Ameriks, J., Caplin, A., Leahy, J., & Tyler, T. (2007). Measuring self-control problems.
American Economic Review, 97(3), 966-972.
Anand, P., Hunter, G., Carter, I., Dowding, K., Guala, F., & Van Hees, M. (2009). The
development of capability indicators. Journal of Human Development and Capabilities,
10(1), 125-152.
Anderson, M., & Marmot, M. (2007). The effects of social status on heart disease: Evidence from
Whitehall.
Argyle, M. (1999). Causes and correlates of happiness. In D. Kahneman, E. Diener & N. Scwarz
(Eds.), Well-Being: The Foundations of Hedonic Psychology. New York: Russell Sage
Foundation
Ashforth, B. (2001). Role Transitions in Organizational Life: An Identity-based Perspective
Hillsdale: Lawrence Erlbaum Associates Inc
Barrick, M. R., Mount, M. K., & Strauss, J. P. (1993). Conscientiousness and performance of
sales representatives - test of the mediating effects of goal-setting. Journal of Applied
Psychology, 78(5), 715-722.
157
Barsky, R. B., Juster, F. T., Kimball, M. S., & Shapiro, M. D. (1997). Preference parameters and
behavioral heterogeneity: An experimental approach in the health and retirement study.
Quarterly Journal of Economics, 112(2), 537-579.
Beck, N., & Katz, J. N. (2001). Throwing out the baby with the bath water: A comment on
Green, Kim, and Yoon. International Organization, 55(2), 487- 495.
Benet-Martinez, V., & John, O. P. (1998). Los Cinco Grandes across cultures and ethnic groups:
Multitrait multimethod analyses of the Big Five in Spanish and English. Journal of
Personality and Social Psychology, 75(3), 729-750.
Bilancini, E., & Boncinelli, L. (2007). Ordinal vs cardinal status: Two examples. University of
Siena.
Blanchflower, D. G., & Oswald, A. J. (2004). Well-being over time in Britain and the USA.
Journal of Public Economics, 88(7-8), 1359-1386.
Blanchflower, D. G., & Oswald, A. J. (2008). Hypertension and happiness across nations.
Journal of Health Economics, 27(2), 218-233.
Booth, A. L., & van Ours, J. C. (2008). Job satisfaction and family happiness: The part-time
work puzzle. Economic Journal, 118(526), F77-F99.
Borghans, L., Duckworth, A. L., Heckman, J. J., & ter Weel, B. (2008). The economics and
psychology of personality traits. Journal of Human Resources, 43(4), 972-1059.
Borkenau, P., & Ostendorf, F. (1990). Comparing exploratory and confirmatory factor-analysis a study on the 5-factor model of personality. Personality and Individual Differences,
11(5), 515-524.
Bower, P., Byford, S., Sibbald, B., Ward, E., King, M., Lloyd, R., et al. (2000). Randomised
controlled trial of non-directive counselling, cognitive-behaviour therapy, and usual
158
general practitioner care for patients with depression. II: Cost effectiveness. British
Medical Journal, 321(7273), 1389-1392.
Bowles, S., Gintis, H., & Osborne, M. O. (2001a). The determinants of earnings: A behavioral
approach. Journal of Economic Literature, 39(4), 1137-1176.
Bowles, S., Gintis, H., & Osborne, M. O. (2001b). Incentive-enhancing preferences: Personality,
behavior, and earnings. American Economic Review, 91(2), 155-158.
Brickman, P., Coates, D., & Janoff-Bulman, R. (1978). Lottery winners and accident victims - is
happiness relative? Journal of Personality and Social Psychology, 36(8), 917-927.
Brown, G. D. A., Gardner, J., Oswald, A. J., & Qian, J. (2008). Does wage rank affect
employees' well-being? Industrial Relations, 47(3), 355-389.
Buunk, B. P., Taylor, S. E., Dakof, G. A., Collins, R. L., & Vanyperen, N. W. (1990). The
affective consequences of social-comparison - Either direction has its ups and downs.
Journal of Personality and Social Psychology, 59(6), 1238-1249.
Cantarero, D., & Pascual, M. (2005). Socio-economic status and health: Evidence from the
ECHP. Economics Bulletin, 9(9), 17.
Cardozo, B. L., Vergara, A., Agani, F., & Gotway, C. A. (2000). Mental health, social
functioning, and attitudes of Kosovar Albanians following the war in Kosovo. JAMAJournal of the American Medical Association, 284(5), 569-577.
Cattell, C. R. (1946). Description and measurement of personality. Yonkers-on-Hudson: NY:
World Book.
Clark, A. E. (2003). Unemployment as a social norm: Psychological evidence from panel data.
Journal of Labor Economics, 21(2), 323-351.
159
Clark, A. E., Diener, E., Georgellis, Y., & Lucas, R. E. (2008). Lags and leads in life satisfaction:
A test of the baseline hypothesis. Economic Journal, 118(529), F222-F243.
Clark, A. E., Etile, F., Postel-Vinay, F., Senik, C., & Van der Straeten, K. (2005). Heterogeneity
in reported well-being: Evidence from twelve European countries. Economic Journal,
115(502), C118-C132.
Clark, A. E., Frijters, P., & Shields, M. A. (2008). Relative income, happiness, and utility: An
explanation for the Easterlin paradox and other puzzles. Journal of Economic Literature,
46(1), 95-144.
Clark, A. E., Georgellis, Y., & Sanfey, P. (2001). Scarring: The psychological impact of past
unemployment. Economica, 68(270), 221-241.
Clark, A. E., Kristensen, N., & Westergard-Nielsen, N. (2009a). Economic satisfaction and
income rank in small neighbourhoods. Journal of the European Economic Association,
7(2-3), 519-527.
Clark, A. E., Kristensen, N., & Westergard-Nielsen, N. (2009b). Job satisfaction and co-worker
wages: Status or signal? Economic Journal, 119(536), 430-447.
Clark, A. E., Masclet, D., & Villeval, M. C. (in press). Effort and comparison income:
Experimental and survey evidence Industrial & Labor Relations Review.
Clark, A. E., & Oswald, A. J. (1994). Unhappiness and unemployment. Economic Journal,
104(424), 648-659.
Clark, A. E., & Oswald, A. J. (1996). Satisfaction and comparison income. Journal of Public
Economics, 61(3), 359-381.
Clark, A. E., & Oswald, A. J. (1998). Comparison-concave utility and following behaviour in
social and economic settings. Journal of Public Economics, 70(1), 133-155.
160
Clark, A. E., & Oswald, A. J. (2002). A simple statistical method for measuring how life events
affect happiness. International Journal of Epidemiology, 31(6), 1139-1144.
Cohen, J. (1992). A power primer. Psychological Bulletin, 112(1), 155-159.
Costa, P. T., & McCrae, R. R. (1980). Influence of extraversion and neuroticism on subjective
well-being - happy and unhappy people. Journal of Personality and Social Psychology,
38(4), 668-678.
Costa, P. T., & McCrae, R. R. (1988). Personality in adulthood - a 6-year longitudinal-study of
self-reports and spouse ratings on the Neo Personality-Inventory. Journal of Personality
and Social Psychology, 54(5), 853-863.
Creed, P. A., Lehmann, K., & Hood, M. (2009). The relationship between core self-evaluations,
employment commitment and well-being in the unemployed. Personality and Individual
Differences, 47(4), 310-315.
Cutler, D., Deaton, A., & Lleras-Muney, A. (2006). The determinants of mortality. Journal of
Economic Perspectives, 20(3), 97-120.
Darity, W., & Goldsmith, A. H. (1996). Social psychology, unemployment and macroeconomics.
Journal of Economic Perspectives, 10(1), 121-140.
Deaton, A. (2003). Health, inequality, and economic development. Journal of Economic
Literature, 41(1), 113-158.
DeNeve, K. M., & Cooper, H. (1998). The happy personality: A meta-analysis of 137 personality
traits and subjective well-being. Psychological Bulletin, 124(2), 197-229.
Di Tella, R., & MacCulloch, R. (2006). Some uses of happiness data in economics. Journal of
Economic Perspectives, 20(1), 25-46.
161
Di Tella, R., MacCulloch, R. J., & Oswald, A. J. (2001). Preferences over inflation and
unemployment: Evidence from surveys of happiness. American Economic Review, 91(1),
335-341.
Di Tella, R., MacCulloch, R. J., & Oswald, A. J. (2003). The macroeconomics of happiness.
Review of Economics and Statistics, 85(4), 809-827.
Diener, E. (1984). Subjective well-being. Psychological Bulletin, 95(3), 542-575.
Diener, E. (2000). Subjective well-being - The science of happiness and a proposal for a national
index. American Psychologist, 55(1), 34-43.
Diener, E., & Biswas-Diener, R. (2002). Will money increase subjective well-being? Social
Indicators Research, 57(2), 119-169.
Diener, E., & Lucas, R. E. (1999). Personality and subjective well-being. In D. Kahneman, E.
Diener & N. Scwarz (Eds.), Well-Being: The Foundations of Hedonic Psychology. New
York: Russell Sage Foundation
Diener, E., & Oishi, S. (2000). Money and happiness: Income and subjective well-being across
nations. In E. Diener & E. M. Suh (Eds.), Culture and Subjective Well-Being (pp. 185218). Cambridge: MIT Press.
Diener, E., Sandvik, E., Seidlitz, L., & Diener, M. (1993). The relationship between income and
subjective well-being - relative or absolute? Social Indicators Research, 28(3), 195-223.
Diener, E., Suh, E. M., Lucas, R. E., & Smith, H. L. (1999). Subjective well-being: Three
decades of progress. Psychological Bulletin, 125(2), 276-302.
Dohmen, T., Falk, A., Huffman, D., & Sunde, U. (2008). Representative trust and reciprocity:
Prevalence and determinants. Economic Inquiry, 46(1), 84-90.
162
Duesenberry, J. S. (1949). Income, Saving, and the Theory of Consumer Behavior: Harvard
University Press.
Duleep, H. O. (1986). Measuring the effect of income on adult mortality using longitudinal
administrative record data. Journal of Human Resources, 21(2), 238-251.
Duncan, G. J., Daly, M. C., McDonough, P., & Williams, D. R. (2002). Optimal indicators of
socioeconomic status for health research. American Journal of Public Health, 92(7),
1151-1157.
Dunn, E. W., Aknin, L. B., & Norton, M. I. (2008). Spending money on others promotes
happiness. Science, 319(5870), 1687-1688.
Easterlin, R. A. (1974). Does economic growth improve the human lot? Some empirical
evidence. In P. A. David & M. W. Reder (Eds.), Nations and Households in Economic
Growth: Essays in Honour of Moses Abramovitz. New York: Academic Press.
Easterlin, R. A. (1995). Will raising the incomes of all increase the happiness of all? Journal of
Economic Behavior & Organization, 27(1), 35-47.
Easterlin, R. A. (2001). Income and happiness: Towards a unified theory. Economic Journal,
111(473), 465-484.
Easterlin, R. A. (2005). A puzzle for adaptive theory. Journal of Economic Behavior &
Organization, 56(4), 513-521.
Ekman, P., Friesen, W. V., & Davidson, R. J. (1990). The Duchenne smile - emotional
expression and brain physiology - 2. Journal of Personality and Social Psychology, 58(2),
342-353.
Elster, J. (1998). Emotions and economic theory. Journal of Economic Literature, 36(1), 47-74.
163
Emmons, R. A. (1992). Abstract versus concrete goals - personal striving level, physical illness,
and psychological well-being. Journal of Personality and Social Psychology, 62(2), 292300.
Epstein, S. (1979). Stability of behavior. 1. Predicting most of the people much of the time.
Journal of Personality and Social Psychology, 37(7), 1097-1126.
Ettner, S. L. (1996). New evidence on the relationship between income and health. Journal of
Health Economics, 15(1), 67-85.
Falk, A., & Knell, M. (2004). Choosing the Joneses: Endogenous goals and reference standards.
Scandinavian Journal of Economics, 106(3), 417-435.
Feather, N. T., & Davenport, P. R. (1981). Unemployment and depressive affect - a motivational
and attributional analysis. Journal of Personality and Social Psychology, 41(3), 422-436.
Feldman, J. J., Makuc, D. M., Kleinman, J. C., & Cornonihuntley, J. (1989). National trends in
educational differentials in mortality. American Journal of Epidemiology, 129(5), 919933.
Ferrer-i-Carbonell, A. (2005). Income and well-being: an empirical analysis of the comparison
income effect. Journal of Public Economics, 89(5-6), 997-1019.
Ferrer-i-Carbonell, A., & Frijters, P. (2004). How important is methodology for the estimates of
the determinants of happiness? Economic Journal, 114(497), 641-659.
Ferrer-i-Carbonell, A., & van Praag, B. M. S. (2002). The subjective costs of health losses due to
chronic diseases. An alternative model for monetary appraisal. Health Economics, 11(8),
709-722.
Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7(2).
164
Finkelstein, A., Luttmer, E. F. P., & Notowidigdo, M. J. (2008). What good is wealth without
health? The effect of health on the marginal utility of consumption. NBER Working Paper
No. 14089.
Fischer, C. S. (2008). What wealth-happiness paradox? A short note on the American case.
Journal of Happiness Studies, 9(2), 219-226.
Fischer, J. A. V., & Sousa-Poza, A. (2009). Does job satisfaction improve the health of workers?
New evidence using panel data and objective measures of health. Health Economics,
18(1), 71-89.
Fliessbach, K., Weber, B., Trautner, P., Dohmen, T., Sunde, U., Elger, C. E., et al. (2007). Social
comparison affects reward-related brain activity in the human ventral striatum. Science,
318(5854), 1305-1308.
Fordyce, M. W. (1988). A review of research on the happiness measures - a 60 second index of
happiness and mental-health. Social Indicators Research, 20(4), 355-381.
Frank, R. H. (1999). Luxury Fever: Why Money Fails to Satisfy in an Era of Excess. New York:
The Free Press.
Frederick, S., & Loewenstein, G. (1999). Hedonic adaptation. In D. Kahneman, E. Diener & N.
Scwarz (Eds.), Well-Being: The Foundations of Hedonic Psychology. New York: Russell
Sage Foundation
Frey, B. S., & Stutzer, A. (2002). What can economists learn from happiness research? Journal
of Economic Literature, 40(2), 402-435.
Frijters, P., Haisken-DeNew, J., & Shields, M. A. (2004). Money does matter! Evidence from
increasing real income and life satisfaction in East Germany following reunification.
American Economic Review, 94(3), 730-740.
165
Frijters, P., Haisken-DeNew, J., & Shields, M. A. (2005). Socio-economic status, health shocks,
life satisfaction and mortality: Evidence from an increasing mixed proportional hazard
model.
Fuchs, V. R. (2004). Reflections on the socio-economic correlates of health. Journal of Health
Economics, 23(4), 653-661.
Gardner, J., & Oswald, A. (2004). How is mortality affected by money, marriage, and stress?
Journal of Health Economics, 23(6), 1181-1207.
Gardner, J., & Oswald, A. J. (2006). Do divorcing couples become happier by breaking up?
Journal of the Royal Statistical Society Series a-Statistics in Society, 169, 319-336.
Gardner, J., & Oswald, A. J. (2007). Money and mental wellbeing: A longitudinal study of
medium-sized lottery wins. Journal of Health Economics, 26(1), 49-60.
Gerlitz, J., & Schupp, J. (2005). Zur erhebung der Big-Fivebasierten personlichkeitsmerkmale im
SOEP - Collecting the Big-Five Personality in SOEP. Berlin: DIW Berlin.
Gigerenzer, G., & Selton, R. (2001). Bounded Rationality: The Adaptive Toolbox Cambridge,
Massachusetts: The MIT Press.
Gigerenzer, G., & Todd, P. M. (1999). Simple Heuristics That Make Us Smart USA: Oxford
University Press.
Graham, C., Eggers, A., & Sukhtankar, S. (2004). Does happiness pay? An exploration based on
panel data from Russia. Journal of Economic Behavior & Organization, 55(3), 319-342.
Griffin, J. M., Fuhrer, R., Stansfeld, S. A., & Marmot, M. (2002). The importance of low control
at work and home on depression and anxiety: do these effects vary by gender and social
class? Social Science & Medicine, 54(5), 783-798.
166
Groves, M. O. (2005). How important is your personality? Labor market returns to personality
for women in the US and UK. Journal of Economic Psychology, 26(6), 827-841.
Gutierrez, J. L. G., Jimenez, B. M., Hernandez, E. G., & Puente, C. P. (2005). Personality and
subjective well-being: Big Five correlates and demographic variables. Personality and
Individual Differences, 38(7), 1561-1569.
Hagerty, M. R. (2000). Social comparisons of income in one's community: Evidence from
national surveys of income and happiness. Journal of Personality and Social Psychology,
78(4), 764-771.
Hagerty, M. R., & Veenhoven, R. (2003). Wealth and happiness revisited - Growing national
income does go with greater happiness. Social Indicators Research, 64(1), 1-27.
Haisken-DeNew, J., & Frick, R. (1998). Desktop companion to the German Socio-Economic
Panel Study (GSOEP). Berlin: DIW Berlin.
Haring, M. J., Stock, W. A., & Okun, M. A. (1984). A research synthesis of gender and socialclass as correlates of subjective well-being. Human Relations, 37(8), 645-657.
Hayes, N., & Joseph, S. (2003). Big 5 correlates of three measures of subjective well-being.
Personality and Individual Differences, 34(4), 723-727.
Headey, B., & Wearing, A. (1989). Personality, life events, and subjective well-being - toward a
dynamic equilibrium-model. Journal of Personality and Social Psychology, 57(4), 731739.
Helson, H. (1964). Adaptation Level Theory. New York: Harper & Row
Hogan, R. (2005). In defense of personality measurement: New wine for old whiners. Human
Performance, 18(4), 331-341.
167
Howell, R. T., & Howell, C. J. (2008). The relation of economic status to subjective well-being
in developing countries: A meta-analysis. Psychological Bulletin, 134(4), 536-560.
Inglehart, R., Foa, R., Peterson, C., & Welzel, C. (2008). Development, freedom, and rising
happiness: A global perspective (1981-2007). Perspectives on Psychological Science,
3(4), 264-285.
John, O. P., & Srivastava, S. (1999). The Big Five trait taxonomy: History, measurement, and
theoretical perspectives. In L. A. Pervin & O. P. John (Eds.), Handbook of Personality:
Theory and Research (2nd Edition ed.). New York: London: Guildford Press.
Johns, H., & Ormerod, P. (2007). Happiness, Economics and Public Policy: The Institute of
Economic Affairs.
Johnson, N. J., Sorlie, P. D., & Backlund, E. (1999). The impact of specific occupation on
mortality in the US National Longitudinal Mortality Study. Demography, 36(3), 355-367.
Judge, T. A., & Ilies, R. (2002). Relationship of personality to performance motivation: A metaanalytic review. Journal of Applied Psychology, 87(4), 797-807.
Kahneman, D., Fredrickson, B. L., Schreiber, C. A., & Redelmeier, D. A. (1993). When more
pain is preferred to less - adding a better end. Psychological Science, 4(6), 401-405.
Kahneman, D., Knetsch, J. L., & Thaler, R. H. (1990). Experimental tests of the endowment
effect and the Coase theorem. Journal of Political Economy, 98(6), 1325-1348.
Kahneman, D., Krueger, A. B., Schkade, D., Schwarz, N., & Stone, A. (2004). Toward national
well-being accounts. American Economic Review, 94(2), 429-434.
Kahneman, D., & Tversky, A. (1979). Prospect theory - analysis of decision under risk.
Econometrica, 47(2), 263-291.
168
Kahneman, D., Wakker, P. P., & Sarin, R. (1997). Back to Bentham? - Explorations of
experienced utility. Quarterly Journal of Economics, 112(2), 375-405.
Kapteyn, A. (1994). The measurement of household cost-functions - revealed preference versus
subjective measures. Journal of Population Economics, 7(4), 333-350.
Kenny, C. (1999). Does growth cause happiness, or does happiness cause growth? Kyklos, 52(1),
3-25.
Koivumaa-Honkanen, H., Honkanen, R., Viinamaki, H., Heikkila, K., Kaprio, J., & Koskenvuo,
M. (2001). Life satisfaction and suicide: A 20-year follow-up study. American Journal of
Psychiatry, 158(3), 433-439.
Kornienko, T. (2004). A cognitive basis for cardinal utility. University of Stirling.
Lahelma, E., & Valkonen, T. (1990). Health and social inequities in Finland and elsewhere.
Social Science & Medicine, 31(3), 257-265.
Layard, R. (2006a). Happiness: Lessons from a New Science. London: Penguin.
Layard, R. (2006b). Health policy - The case for psychological treatment centres. British Medical
Journal, 332(7548), 1030-1032.
Layard, R., Nickell, S. J., & Mayraz, G. (2008). The marginal utility of income. Journal of
Public Economics, 92(8-9), 1846-1857.
Lepper, H. S. (1998). Use of other-reports to validate subjective well-being measures. Social
Indicators Research, 44(3), 367-379.
Loewenstein, G. (2000). Emotions in economic theory and economic behavior. American
Economic Review, 90(2), 426-432.
Lorgelly, P. K., & Lindley, J. (2008). What is the relationship between income inequality and
health? Evidence from the BHPS. Health Economics, 17(2), 249-265.
169
Lucas, R. E. (2007). Adaptation and the set-point model of subjective well-being - Does
happiness change after major life events? Current Directions in Psychological Science,
16(2), 75-79.
Lucas, R. E., Clark, A. E., Georgellis, Y., & Diener, E. (2004). Unemployment alters the set
point for life satisfaction. Psychological Science, 15(1), 8-13.
Lucas, R. E., Diener, E., & Suh, E. (1996). Discriminant validity of well-being measures. Journal
of Personality and Social Psychology, 71(3), 616-628.
Lucas, R. E., & Dyrenforth, P. S. (2006). Does the existence of social relationships matter for
subjective well-being? In E. J. Finkel & K. D. Vohs (Eds.), Self and Relationships:
Connecting Intrapersonal and Interpersonal Processes. New York: NY: Guildford Press.
Lucas, R. E., & Schimmack, U. (2009). Income and well-being: How big is the gap between the
rich and the poor? Journal of Research in Personality, 43(1), 75-78.
Lunney, M., & Oliphant, K. (2008). Tort Law: Text and Materials (Third Edition ed.). New
York: Oxford University Press.
Luttmer, E. F. P. (2005). Neighbors as negatives: Relative earnings and well-being. Quarterly
Journal of Economics, 120(3), 963-1002.
Lykken, D., & Tellegen, A. (1996). Happiness is a stochastic phenomenon. Psychological
Science, 7(3), 186-189.
Mackay, J., Bruffell, M., Cherry, J., Hughes, A., & Tillett, M. (2006). Guidelines for the
assessment of general damages in personal injury cases. New York: Oxford University
Press.
Macleod, J., Smith, G. D., Metcalfe, C., & Hart, C. (2005). Is subjective social status a more
important determinant of health than objective social status? Evidence from a
170
prospective observational study of Scottish men. Social Science & Medicine, 61(9), 19161929.
Marks, N., & Shah, H. (2005). A well-being manifesto for a flourishing society. In F. A.
Huppert, N. Baylis & B. Keverne (Eds.), The Science of Well-Being. Oxford: Oxford
University Press.
Marmot, M. G. (2004). Status Syndrome: How your social standing directly affects your health
and life expectancy. London: Bloomsbury Publishing PLC
Marmot, M. G., Shipley, M. J., & Rose, G. (1984). Inequalities in Death - Specific Explanations
of a General Pattern. Lancet, 1(8384), 1003-1006.
Marx, K. (1952). Wage Labour and Capital. Moscow: Progress Publishers.
Maslow, A. H. (1954). Motivation and Personality: Harper & Row.
Matzler, K., Renzl, B., Muller, J., Herting, S., & Mooradian, T. A. (2008). Personality traits and
knowledge sharing. Journal of Economic Psychology, 29(3), 301-313.
McBride, M. (2001). Relative-income effects on subjective well-being in the cross-section.
Journal of Economic Behavior & Organization, 45(3), 251-278.
McCrae, R. R., & Costa, P. T. (1987). Validation of the 5-factor model of personality across
instruments and observers. Journal of Personality and Social Psychology, 52(1), 81-90.
McCrae, R. R., & Costa, P. T. (1990). Personality in Adulthood. New York: The Guildford Press.
Mccrae, R. R., & Costa, P. T. (1991). Adding liebe und arbeit - the full 5-factor model and wellbeing. Personality and Social Psychology Bulletin, 17(2), 227-232.
McDonough, P., Duncan, G. J., Williams, D., & House, J. (1997). Income dynamics and adult
mortality in the United States, 1972 through 1989. American Journal of Public Health,
87(9), 1476-1483.
171
McGregor, I., & Little, B. R. (1998). Personal projects, happiness, and meaning: On doing well
and being yourself. Journal of Personality and Social Psychology, 74(2), 494-512.
McKee-Ryan, F. M., Song, Z. L., Wanberg, C. R., & Kinicki, A. J. (2005). Psychological and
physical well-being during unemployment: A meta-analytic study. Journal of Applied
Psychology, 90(1), 53-76.
Mellers, B. A. (1986). Fair allocations of salaries and taxes. Journal of Experimental
Psychology: Human Perception and Performance, 12(1), 80-91.
Menchik, P. L. (1993). Economic-status as a determinant of mortality among black-and-white
older men - does poverty kill? Population Studies-A Journal of Demography, 47(3), 427436.
Michaud, C. M., Murray, C. J. L., & Bloom, B. R. (2001). Burden of disease: Implications for
future research. JAMA-Journal of the American Medical Association, 285(5), 535-539.
Mongrain, M., & Blackburn, S. (2005). Cognitive vulnerability, lifetime risk, and the recurrence
of major depression in graduate students. Cognitive Therapy and Research, 29(6), 747768.
Moore, S., & Shepherd, J. P. (2006). The cost of fear: shadow pricing the intangible costs of
crime. Applied Economics, 38(3), 293-300.
Moulton, B. R. (1990). An Illustration of a pitfall in estimating the effects of aggregate variables
on micro units. Review of Economics and Statistics, 72(2), 334-338.
Mueller, G., & Plug, E. J. S. (2006). Estimating the effect of personality on male and female
earnings. Industrial & Labor Relations Review, 60(1), 3-22.
Mundlak, Y. (1978). Pooling of time-series and cross-section data. Econometrica, 46(1), 69-85.
Myers, D. G., & Diener, E. (1995). Who is happy? Psychological Science, 6(1), 10-19.
172
Neumark, D., & Postlewaite, A. (1998). Relative income concerns and the rise in married
women's employment. Journal of Public Economics, 70(1), 157-183.
Niedrich, R. W., Sharma, S., & Wedell, D. H. (2001). Reference price and price perceptions: A
comparison of alternative models. Journal of Consumer Research, 28(3), 339-354.
Nordenmark, M., Strandh, M., & Layte, R. (2006). The impact of unemployment benefit system
on the mental well-being of the unemployed in Sweden, Ireland and Great Britain.
European Societies, 8(1), 83-110.
Nyhus, E. K., & Pons, E. (2005). The effects of personality on earnings. Journal of Economic
Psychology, 26(3), 363-384.
O'Cleirigh, C., Ironson, G., Weiss, A., & Costa, P. T. (2007). Conscientiousness predicts disease
progression (CD4 number and viral load) in people living with HIV. Health Psychology,
26(4), 473-480.
Offer, A. (2006). The Challenge of Affluence New York: Oxford University Press.
Oswald, A. J. (1997). Happiness and economic performance. Economic Journal, 107(445), 18151831.
Oswald, A. J., & Powdthavee, N. (2008a). Death, happiness, and the calculation of compensatory
damages. Journal of Legal Studies, 37, S217-S251.
Oswald, A. J., & Powdthavee, N. (2008b). Does happiness adapt? A longitudinal study of
disability with implications for economists and judges. Journal of Public Economics,
92(5-6), 1061-1077.
Palmore, E. (1969). Predicting longevity: A follow up controlling for age. The Gerontologist,
9(4), 247-250.
173
Parducci, A. (1965). Category judgment: a range-frequency model. Psychological Review, 72(6),
12.
Parducci, A. (1995). Happiness, Pleasure, and Judgment: The Contextual Theory and Its
Applications. Mahwah, NJ: Lawrence Erlbaum Associates.
Pavot, W., & Diener, E. (1993a). The affective and cognitive context of self-reported measures
of subjective well-being. Social Indicators Research, 28(1), 1-20.
Pavot, W., & Diener, E. (1993b). Review of the satisfaction with life scale. Psychological
Assessment, 5(2), 19.
Pervin, L. A., & John, O. P. (Eds.). (1999). Handbook of personality (2nd ed.). New York:
Guildford Press.
Pevalin, D. J., & Ermisch, J. (2004). Cohabiting unions, repartnering and mental health.
Psychological Medicine, 34(8), 1553-1559.
Plumper, T., & Troeger, V. E. (2007). Efficient estimation of time-invariant and rarely changing
variables in finite sample panel analyses with unit fixed effects. Political Analysis, 15(2),
124-139.
Powdthavee, N. (2007). Are there geographical variations in the psychological cost of
unemployment in South Africa? Social Indicators Research, 80(3), 629-652.
Powdthavee, N. (2008). Putting a price tag on friends, relatives, and neighbours: Using surveys
of life satisfaction to value social relationships. Journal of Socio-Economics, 37, 14591480.
Powdthavee, N. (in press). How important is rank to individual perception of economic
hierarchy? A within-community analysis. Journal of Economic Inequality.
174
Qian, J., & Brown, G. D. A. (2007). Loss aversion, anchoring, and similarity sampling in price
perception: Towards an integrated model. University of Warwick.
Rablen, M. D., & Oswald, A. J. (2008). Mortality and immortality: The Nobel Prize as an
experiment into the effect of status upon longevity. Journal of Health Economics, 27(6),
1462-1471.
Ralph, J. A., & Mineka, S. (1998). Attributional style and self-esteem: The prediction of
emotional distress following a midterm exam. Journal of Abnormal Psychology, 107(2),
203-215.
Read, D. (2007). Experienced utility: Utility theory from Jeremy Bentham to Daniel Kahneman.
Thinking & Reasoning, 13(1), 45-61.
Redelmeier, D. A., & Singh, S. M. (2001). Survival in academy award-winning actors and
actresses. Annals of Internal Medicine, 134(10), 955-962.
Revelle, W., & Scherer, K. R. (2008). Personality and emotion. Oxford Companion to the
Affective Sciences.
Rogers, C. R. (1959). A theory of therapy, personality and interpersonal relationships as
developed in the client-centered framework. In Psychology: A Study of a Science (Vol. 3,
Formulations of the person and the social context, pp. 181-256). New York: McGraw-Hill
Book Company.
Rotter, J. B. (1966). Generalized expectancies for internal versus external control of
reinforcement. Psychological Monographs, 80(1), 1-28.
Ruhm, C. J. (2000). Are recessions good for your health? Quarterly Journal of Economics,
115(2), 617-650.
175
Russell, J. (1987). Mental-health of unemployed men in different parts of England and Wales.
British Medical Journal, 295(6603), 922-922.
Ryff, C. D., Love, G. D., Urry, H. L., Muller, D., Rosenkranz, M. A., Friedman, E. M., et al.
(2006). Psychological well-being and ill-being: Do they have distinct or mirrored
biological correlates? Psychotherapy and Psychosomatics, 75(2), 85-95.
Sales, S. M., & House, J. (1971). Job dissatisfaction as a possible risk factor in coronary heart
disease. Journal of Chronic Diseases, 23(12), 861-873.
Samuelson, L. (2004). Information-based relative consumption effects. Econometrica, 72(1), 93118.
Samuelson, P. A. (1938). A note on the pure theory of consumer's behaviour. Economica, 5, 11.
Sandvik, E., Diener, E., & Seidlitz, L. (1993). Subjective well-being - the convergence and
stability of self-report and non-self-report measures. Journal of Personality, 61(3), 318342.
Sanjuan, P., Perez, A., Rueda, B., & Ruiz, A. (2008). Interactive effects of attributional styles for
positive and negative events on psychological distress. Personality and Individual
Differences, 45(2), 187-190.
Sapolsky, R. M. (2004). Social status and health in humans and other animals. Annual Review of
Anthropology, 33, 393-418.
Sapolsky, R. M., Romero, L. M., & Munck, A. U. (2000). How do glucocorticoids influence
stress responses? Integrating permissive, suppressive, stimulatory, and preparative
actions. Endocrine Reviews, 21(1), 55-89.
176
Schmitt, P., Shupp, R., Swope, K., & Mayer, J. (2008). Pre-commitment and personality:
Behavioral explanations in ultimatum games. Journal of Economic Behavior &
Organization, 66(3-4), 597-605.
Seidl, C. (1994). How sensible is the Leyden individual welfare function of income. European
Economic Review, 38(8), 1633-1659.
Seligman, M. E. P., & Csikszentmihalyi, M. (2000). Positive psychology - an introduction.
American Psychologist, 55(1), 5-14.
Seligman, M. E. P., Steen, T. A., Park, N., & Peterson, C. (2005). Positive psychology progress Empirical validation of interventions. American Psychologist, 60(5), 410-421.
Semykina, A., & Linz, S. J. (2007). Gender differences in personality and earnings: Evidence
from Russia. Journal of Economic Psychology, 28(3), 387-410.
Senik, C. (2004). When information dominates comparison - Learning from Russian subjective
panel data. Journal of Public Economics, 88(9-10), 2099-2123.
Shields, M. A., & Price, S. W. (2005). Exploring the economic and social determinants of
psychological well-being and perceived social support in England. Journal of the Royal
Statistical Society Series A-Statistics in Society, 168, 513-537.
Smith, A. (1976). The Theory of Moral Sentiments. Oxford: Clarendon Press.
Smith, D. M., Langa, K. M., Kabeto, M. U., & Ubel, P. A. (2005). Health, wealth, and happiness.
Psychological Science, 16(9), 663-666.
Smith, J. P. (1999). Healthy bodies and thick wallets: The dual relation between health and
economic status. Journal of Economic Perspectives, 13(2), 145-166.
177
Smith, R. H., Diener, E., & Wedell, D. H. (1989). Intrapersonal and social-comparison
determinants of happiness - a Range-Frequency analysis. Journal of Personality and
Social Psychology, 56(3), 317-325.
Snyder, S. E., & Evans, W. N. (2006). The effect of income on mortality: Evidence from the
social security notch. Review of Economics and Statistics, 88(3), 482-495.
Srivastava, S., John, O. P., Gosling, S. D., & Potter, J. (2003). Development of personality in
early and middle adulthood: Set like plaster or persistent change? Journal of Personality
and Social Psychology, 84(5), 1041-1053.
Stevenson, B., & Wolfers, J. (2008). Economic growth and subjective well-being: Reassessing
the Easterlin paradox. Brookings Papers on Economic Activity(1), 1-102.
Stewart, N., Brown, G. D. A., & Chater, N. (2005). Absolute identification by relative judgment.
Psychological Review, 112(4), 881-911.
Stewart, N., Chater, N., & Brown, G. D. A. (2006). Decision by sampling. Cognitive Psychology,
53(1), 1-26.
Swope, K., Cadigan, J., Schmitt, P., & Shupp, R. (2008). Personality preferences in laboratory
economics experiments. The Journal of Socio-Economics, 37, 998-1009.
Taylor, S. E., & Lobel, M. (1989). Social-comparison activity under threat - downward
evaluation and upward contacts. Psychological Review, 96(4), 569-575.
Thaler, R. (1980). Toward a positive theory of consumer choice. Journal of Economic Behavior
& Organization, 1(1), 39-60.
Turner, J. B. (1995). Economic context and the health effects of unemployment. Journal of
Health and Social Behavior, 36(4), 213-229.
178
Tversky, A., & Kahneman, D. (1981). The framing of decisions and the psychology of choice.
Science, 211(4481), 453-458.
Vaggi, G., & Groenewegen, P. D. (2003). A Concise History of Economic Thought: From
Merchantilism to Monetarism. New York: Palgrave Macmillan.
Van Praag, B. M. S. (1971). Welfare function of income in Belgium - empirical investigation.
Econometrica, 39(4), 185-&.
Van Praag, B. M. S., & Baarsma, B. E. (2005). Using happiness surveys to value intangibles: The
case of airport noise. Economic Journal, 115(500), 224-246.
Van Praag, B. M. S., & Frijiters, P. (1999). The measurement of welfare and well-being: The
Leyden approach. In D. Kahneman, E. Diener & N. Scwarz (Eds.), Well-Being: The
Foundations of Hedonic Psychology. New York: Russell Sage Foundation.
Van Praag, B. M. S., Frijters, P., & Ferrer-i-Carbonell, A. (2003). The anatomy of subjective
well-being. Journal of Economic Behavior & Organization, 51(1), 29-49.
Vandestadt, H., Kapteyn, A., & Vandegeer, S. (1985). The relativity of utility - evidence from
panel data. Review of Economics and Statistics, 67(2), 179-187.
Veblen, T. (1899). The Theory of the Leisure class. New York Macmillan.
Veenhoven, R. (1991). Is happiness relative? Social Indicators Research, 24(1), 1-34.
Vendrik, M. C. M., & Woltjer, G. B. (2007). Happiness and loss aversion: Is utility concave or
convex in relative income? Journal of Public Economics, 91(7-8), 1423-1448.
Ward, E., King, M., Lloyd, M., Bower, P., Sibbald, B., Farrelly, S., et al. (2000). Randomised
controlled trial of non-directive counselling, cognitive-behaviour therapy, and usual
general practitioner care for patients with depression. I: Clinical effectiveness. British
Medical Journal, 321(7273), 1383-1388.
179
West, P. (1991). Rethinking the health selection explanation for health inequalities. Social
Science & Medicine, 32(4), 373-384.
Wilkinson, R. (1986). Income and mortality. In R. Wilkinson (Ed.), Class and health: research
and longitudinal data. London: Tavistock.
Wilkinson, R. (2001). Mind The Gap: Hierarchies, Health and Human Evolution. New Haven:
Yale University Press.
Wilkinson, W. (2007). In Pursuit of Happiness Research: Is it Reliable? What does it imply for
Policy? : CATO Institute.
Wilson, S. R., & Benner, L. A. (1971). Effects of self-esteem and situation upon comparison
choices during ability evaluation. Sociometry, 34(3), 381-&.
Wilson, W. (1967). Correlates of avowed happiness. Psychological Bulletin, 67(4), 294-306.
Winefield, A. H., & Tiggemann, M. (1989). Unemployment duration and affective well-being in
the young. Journal of Occupational Psychology, 62(4), 327-336.
Winkelmann, L., & Winkelmann, R. (1998). Why are the unemployed so unhappy? Evidence
from panel data. Economica, 65(257), 1-15.
Winkelmann, L., & Winkelmann, R. (2008). Personality, work, and satisfaction: Evidence from
the German Socio-Economic Panel. The Journal of Positive Psychology, 3(4), 266-275.
Winter, D. G., & Barenbaum, N. B. (1999). History of modern personality theory and research.
In L. A. Pervin & O. P. John (Eds.), Handbook of personality (Second ed.). New York:
Guildford Press.
Wolfson, M., Rowe, G., Gentleman, J. F., & Tomiak, M. (1993). Career earnings and death - a
longitudinal analysis of older Canadian men. Journals of Gerontology, 48(4), S167-S179.
180
Wood, A. M., Joseph, S., & Maltby, J. (2008). Gratitude uniquely predicts satisfaction with life:
Incremental validity above the domains and facets of the five factor model. Personality
and Individual Differences, 45(1), 49-54.