University of Southern California Dissertations and Theses

Three essays on social policy: institutional development, and subjective well-being as a cause and consequence of labor market outcomes

(USC Thesis Other)

Three essays on social policy: institutional development, and subjective well-being as a cause and consequence of labor market outcomes

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content
THREE ESSAYS ON SOCIAL POLICY: INSTITUTIONAL DEVELOPMENT, AND
SUBJECTIVE WELL-BEING AS A CAUSE AND CONSEQUENCE OF LABOR MARKET
OUTCOMES

by

Kelsey J. O’Connor

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirement for the Degree
DOCTOR OF PHILOSOPHY
(ECONOMICS)

August, 2017

Copyright 2017 Kelsey J. O’Connor
ii

Dedication

For my family and friends.

iii
Acknowledgements

I would first like to thank Richard A. Easterlin. He patiently listened while I struggled to explain
my ideas and guided me to obtain workable solutions. He conveyed more knowledge and
wisdom than I could fully internalize. I will continue to learn from him for years to come. My
approaches to research and communication have immeasurably improved through his influence.
Jeffrey B. Nugent also greatly influenced my work. He too patiently offered guidance and
listened to academic and non-academic matters. We coauthored two projects, which needless to
say, I could not have done without him. He also encouraged me to present at the Western
Economic Association Annual Meetings, where I truly got started as an academic. John Strauss I
would like to thank as well. He served as an invaluable teacher and advisor. In several courses,
seminars, and meetings, I learned microeconometrics from him. I also want to thank Young
Miller and Morgan Ponder for their infinite patience while handling administrative matters. And
I thank the university itself for providing financial support.
I learned a great deal from Malgorzata Switek and Robson Morgan. We worked on
projects together with Professor Easterlin, and they provided both academic and non-academic
support. Robson also offered his services as American partner in crime when traveling to several
international conferences. I benefited from his guidance and friendship daily. I would also like to
thank Francesco Sarracino and Stefano Bartolini for organizing the Tuscany Quality of Life
Winter School, which was one of the best experiences of my life.
I would like to thank my academic cohort. I could not have made it through the first year
without their help, and without their friendship, I may not have wanted too. Mallory
Montgomery in particular, not only helped me through the first year, became one my better
friends, but also, now fallen prey to the sway of happiness research, provides instrumental
iv
research notes. I would also like to thank Rakesh Banerjee, Niree Kodaverdian, Ida Johnnsson,
Teresa Molina, Juan Ovalle, and Jorge Tarrasó. From other years, I thank Aleksandar Giga,
Ashish Sachdeva, Rosanna Smart, Petra Thiemann, and Fengyu Wu for their research assistance
and friendship.
And of course my friends and family, to which I dedicate this dissertation. It need not be
mentioned how much influence they have had. My step-brother Reid Havens calls me daily, I
want to make my little sister Kirsten Wiegand proud, and my parents made me who I am. My ex-
girlfriend, Nicole Antoinette Armstrong, was the one who encouraged me to apply to the
University of Southern California in the first place. I would also like to thank my best friends
from Denver, Nick Williams and Noble Varughese, and my roommates, Sam Gebauer and Dan
McCarty.

I needed a village, and they showed up.
For which, I will be forever grateful.

v
Table of Contents
Dedication .............................................................................................................................................. ii
Acknowledgements ............................................................................................................................ iii
List of Tables ........................................................................................................................................ vii
List of Figures ....................................................................................................................................... ix
Abstract .................................................................................................................................................... x
Chapter 1 Introduction ...................................................................................................................... 1
1.1 What is Subjective Well-Being ............................................................................................................. 5
Chapter 2 Does Oil Curse Democracy? A Long-Run Time-Series Analysis of 127
Countries ................................................................................................................................................. 8
2.1 Introduction ............................................................................................................................................... 8
2.2 Background ............................................................................................................................................. 10
2.3 Data ............................................................................................................................................................ 12
2.3.1 Variables, Data, and Sample Coverage ..................................................................................................... 12
2.3.2 Subsample Definitions .................................................................................................................................... 14
2.4 Methods .................................................................................................................................................... 17
2.5 Results ....................................................................................................................................................... 21
2.5.1 Polity, Polcon, Civil Liberties, Political Rights, and O&G Value Over Time .............................. 21
2.5.2 Statistical Analysis ............................................................................................................................................ 24
2.6 Comment on Recent Analysis ............................................................................................................ 32
2.7 Conclusion ............................................................................................................................................... 36
Chapter 3 Who Suffered Most From the Great Recession? Happiness in the United
States ...................................................................................................................................................... 38
3.1 Introduction ............................................................................................................................................ 38
3.2 Evidence From Past Literature ......................................................................................................... 40
3.3 Happiness Data and Methods ............................................................................................................ 42
3.4 Description of Well-Being Impacts .................................................................................................. 45
3.4.1 Statistical Significance of Declines by Population Group ................................................................. 45
3.4.2 Statistical Significance of Observed Trends in Happiness ............................................................... 48
3.5 Great Recession Channels – Explaining the Impacts ................................................................. 50
3.5.1 Methods and Variables to Identify Plausible Channels ..................................................................... 50
3.5.2 Plausible Channels – Average Relationship For Full Population .................................................. 52
3.5.3 Channels by Population Group .................................................................................................................... 61
3.6 Comparison With the 1980s Recession ......................................................................................... 65
3.7 Conclusion ............................................................................................................................................... 67
Chapter 4 Happier People Are Less Likely To Be Unemployed: Evidence From
Longitudinal Data In Germany ...................................................................................................... 69
4.1 Introduction ............................................................................................................................................ 69
4.2 Past Evidence for Subjective Well-Being’s Relation with Unemployment ......................... 72
4.3 Data and Empirical Methods ............................................................................................................. 75
4.3.1 Data ......................................................................................................................................................................... 75
4.3.2 Fixed Effects and Dynamic Ordinary Least Squares .......................................................................... 75
4.4 Results: SWB Predicts the Likelihood Of Unemployment ....................................................... 77
4.4.1 Main Results ........................................................................................................................................................ 77
vi
4.4.2 Heterogeneity of SWB – Unemployment Relation .............................................................................. 79
4.4.3 Robustness - Additional Control Variables and Placebo Test ........................................................ 81
4.4.4 Arellano and Bond’s System Generalized Method of Moments .................................................... 86
4.5 Mechanisms – Character Skills ......................................................................................................... 90
4.5.1 Past Evidence of Character Skills as a Mechanism .............................................................................. 91
4.5.2 Character Skills Operate as a Mechanism ............................................................................................... 92
4.5.3 Implications for Future Research ............................................................................................................... 96
4.6 Conclusions ............................................................................................................................................. 97
Chapter 5 Summary and Conclusions ...................................................................................... 100
References ......................................................................................................................................... 103
Appendix A Supplementary Material for Chapter 2 ............................................................ 112
Appendix B Supplementary Material for Chapter 3 ............................................................ 121
Appendix C Supplementary Material for Chapter 4 ............................................................ 126
C.1 Sample Attrition ................................................................................................................................... 126
C.2 German Socio-Economic Panel Variables ................................................................................... 128

vii
List of Tables

Table 2-1 Long-Run Effect of O&G Value pc on Democracy. Two Mean Group Models. Various
Indicators and Country Groups. ............................................................................................ 25
Table 2-2 Long-Run Effect of Oil & Gas Value pc on Democracy. CS-EC Fixed-Effects
Models. Various Indicators and Country Groups. ................................................................ 26
Table 2-3 Long-Run Effect of Oil & Gas Value pc on Democracy. CS-EC Mean Group with
adjustments. Various Indicators and Country Groups. ......................................................... 29
Table 2-4 Replication of Andersen and Ross (2014) Table 2, Column 1 And Sample Adjustment
Long-Run Effects on Polity .................................................................................................. 35
Table 3-1 Deviations from Long-Term Trends; Marginal Effects by Group from Ordered-Probit
Regressions with Happiness as the Choice Variable. Probability of Reporting Very Happy,
1973 – 2014........................................................................................................................... 47
Table 3-2 Deviations from Long-Term Happiness Trends; Group Ordered-Probit Regressions.
Choice Variable Happiness, 1973 – 2014. ............................................................................ 49
Table 3-3 Ordered-Probit Regressions for Full Sample with added Micro and Macro Controls,
1977 – 2012........................................................................................................................... 53
Table 3-4 Deviations from Long-Term Happiness Trends; Ordered-Probit Regressions for U.S.
Sample with added Controls and Interactions, 1977 – 2012. ............................................... 55
Table 3-5 2010 Deviation from Long-Term Trend – Change in Probability of Reporting “Very
Happy” by Population Group with Specified Controls ......................................................... 64
Table 4-1 Unemployed and Life Satisfaction Regressions. Dependent Variable: Unemployed .. 78
Table 4-2 Heterogeneity of Unemployed - Life Satisfaction Relationship. Dependent Variable:
Unemployed .......................................................................................................................... 81
Table 4-3 Unemployed and Life Satisfaction Regressions with Additional Control Variables.
Dependent Variable: Unemployed ........................................................................................ 84
Table 4-4 Placebo Test: Unemployed and Family Satisfaction Regressions. Dependent Variable:
Unemployed .......................................................................................................................... 86
Table 4-5 Arellano and Bond Type GMM Estimation. Estimation Dependent Variable:
ΔUnemployed. Relations Presented in Levels for Unemployed ........................................... 89
Table 4-6 Mechanisms: Lagged Life Satisfaction Regressions. Various years.
a
.......................... 95

Appendix Table A-1 Summary Statistics ................................................................................... 112
Appendix Table A-2 Oil & Gas Value’s Impact on Polity by Country Group CS-EC Mean Group
Regressions ......................................................................................................................... 113
Appendix Table A-3 Oil & Gas Value’s Impact on Polcon by Country Group. CS-EC Mean
Group Regressions .............................................................................................................. 114
Appendix Table A-4 Oil & Gas Value’s Impact on Civil Liberties by Country Group CS-EC
Mean Group Regressions .................................................................................................... 115
Appendix Table A-5 Oil & Gas Value’s Impact on Political Rights by Country Group CS-EC
Mean Group Regressions .................................................................................................... 116
Appendix Table A-6 Test for Integration Order of Polity, Polcon Civil Liberties, Political Rights,
and Oil & Gas Value ........................................................................................................... 117
Appendix Table A-7 Cointegration Results, Westerlund Panel Tests. Restricted Sample - 52
Countries ............................................................................................................................. 118
Appendix Table A-8 List of Countries and Group Composition ................................................ 119
viii
Appendix Table B-1 Population Shares by Demographic Group, General Social Survey ......... 121
Appendix Table B-2 Summary Statistics for Key Micro Variables ........................................... 121
Appendix Table B-3 Self-Reported Income and Unemployment as a Percentage of Total
Population, by Population Group, 2006 – 2012. ................................................................. 122
Appendix Table B-4 Key Macroeconomic Variables and Their Sources ................................... 123
Appendix Table B-5 2010 Deviation from Long-Term Trend. Change in Probability of Reporting
“Very Happy”, Full Sample, 1973 – 2014 .......................................................................... 124
Appendix Table B-6 Fixed-Effects Logit Regressions; GSS Panel 2006-2014; Dependent
Variable “Very Happy”, Main Effects by Year (Base = 2010). ......................................... 125
Appendix Table C-1 Selective Attrition Fixed-Effects Regression. Germany 1996-2013
Unemployment Regressed on Lagged Determinants
a
and Future Selection ...................... 128

ix
List of Figures
Figure 2-1 Polity, Polcon, and O&G Value p.c. Lowess Smoothed Trends (1974 – 2012) ......... 22
Figure 2-2 Civil Liberties, Political Rights, and O&G Value p.c. Lowess Smoothed Trends (1974
– 2012) .................................................................................................................................. 23
Figure 3-1 U.S. Happiness 1973 – 2014. Annual Proportions Reporting Very Happy, Pretty
Happy, or Not Too Happy .................................................................................................... 46
Figure 3-2 2010 Deviation from Long-Term Trend – Change in Probability of Reporting “Very
Happy”, Two Models by Population Group ......................................................................... 62

x
Abstract

How can social policy be used to improve societal well-being? First, quality institutions are
necessary to implement optimal policies. Second societal well-being needs to be measured to
determine optimal policies. Subjective well-being (SWB), or happiness, data are well suited for
this purpose, because SWB measures allow individuals to weigh both economic and non-
economic outcomes. For these reasons, Chapter 2 addresses the determinants of democratic
institutions, while Chapters 3 and 4 discuss well-being directly. In particular, Chapter 2
challenges the so-called political resource curse, finding no robust long-term effects of oil
abundance on Polity, civil liberties, political rights, or constraints on the executive. The chapter
extends previous empirical work by using multiple measures of democracy, the most current
time-series econometrics, and by allowing additional relationship heterogeneity, especially by
geographic region and time period. Chapter 3 documents the year 2010 as the lowest level of
recorded happiness in the U.S. since consistent measurement began in the 1970s. Disparate
declines in income and employment, associated with the Great Recession, are shown to explain
the disparate declines in SWB experienced by different demographic groups. Chapter 4 reverses
the relationship direction by showing that people reporting higher SWB are less likely to be
unemployed in Germany. The mechanisms include changes in the Big-Five personality traits.
Within-person increases in extraversion or emotional stability, for example, are associated with
increases in SWB. The results also show that the magnitude of the SWB-unemployed relation is
substantially larger for people who are currently unemployed, and that too much SWB can also
increase the likelihood of unemployment.
1
Chapter 1 Introduction

Social policy is intended to improve societal well-being. Societal well-being is perhaps best
measured with subjective well-being (SWB), or happiness, data because SWB accounts for
individuals’ feelings about their lives as a whole, including dimensions that are otherwise
difficult to measure. The questions then become, can the institutions that are responsible for
social policy be improved? And what policies should be implemented to improve SWB? The
present dissertation contributes research to further understand these broad questions. The first
analysis focuses on the development of formal institutions. The two subsequent analyses focus
directly on SWB; the first relates to its determinants, and the second, consequences.
Demonstrating SWB has positive consequences on more traditional economic outcomes should
reduce the still-present skepticism of subjective measures in economic research and
policymaking.
Chapter 2 focuses on democracy as an important institution. It challenges the so-called
political resources curse, finding that oil abundance does not have robust long-run negative
effects on democracy. Early empirical research found that countries with relatively more mineral
resources, especially oil, exhibited lower levels of economic growth. An early theory explained
this outcome stating that resources negatively affected political institutions, which in turn
hindered economic growth. Consequently much of this literature has shifted focus from
economic growth to political institutions, especially democracy. There has been substantial
support for the existence of a political resources curse (e.g. Ahmadov 2014), and a few
challenges in recent years (e.g. Haber and Menaldo 2011). In Chapter 2, two coauthors and I
contribute to the ongoing debate.
2
Our analysis improves upon comparable studies. We use four measures of democracy
including the broad measure Polity, and the three underlying measures, civil liberties, political
rights, and constraints on the executive. By using four measures, from three different
organizations
1
, we test theoretically different mechanisms and reduce concern of measurement
error that could have attenuated the oil-democracy relation. Many papers focused solely on
Polity (e.g. Haber and Menaldo 2011). We also estimate the relations separately for subsamples
of: Latin American countries, countries in the Middle East and North Africa (geography is
important in Ahmadov 2014), producers with more or less experience in the industry (experinece
is important in Omgba 2015), and the period post 1980. Andersen and Ross (2014) found that the
insignificant oil-democracy relationship demonstrated by Haber and Menaldo (2011) becomes
statistically significant and negative when estimated for the post-1980 period. We use time-series
econometrics to estimate long-run effects separately from short-run changes. Long-run effects
capture relationships that are more consistent with the slow pace of institutional change, and
separately accounting for short-run changes improves the estimation precision. We also allow for
relationship heterogeneity across countries and account for bias caused by common cross-
country effects, such as regional democratic development or world oil prices. Yet in the various
specifications, subsamples, and robustness tests we do not find consistent evidence of a within-
country democratic resource curse.
Chapters 3 and 4 contribute to the economics of happiness (or SWB) literature. A more
complete understanding of SWB - documenting it, its determinants and outcomes - will help to
set policy goals and evaluate current systems (Frey and Stutzer 2010; 2012; and Frey and Gallus
2013). SWB measures are complementary to more objective measures (Fleurbaey 2009; Stiglitz

1
Polity is from the Polity IV Project; civil liberties and political rights are from Freedom House; and constraints
the executive is from Polcon.
3
et al. 2009), but have the advantage of including whatever economic and non-economic factors
that individuals value. It is precisely because objective measures miss important dimensions of
well-being that I use SWB data in Chapter 3 to document the impacts of the Great Recession. For
example, Lucas (1987) argued that business cycles are not very important determinants of well-
being when using aggregate consumption as a measure of well-being. However when using
SWB, macroeconomic volatility was found to have ‘moderate but important’ effects on well-
being (Wolfers 2003).
2
The effects of unemployment on well-being serve as an example at the
micro level. For individuals, the non-pecuniary effects of being unemployed are larger than the
effects from income loss alone (Winkelmann and Winkelmann 1998).
Chapter 3 addresses the question, which demographic group suffered the most during the
Great Recession. For the U.S. population a whole, the year 2010 marked the lowest level of
happiness recorded since 1972 - based on the nationally representative General Social Survey
(GSS). Happiness trended downward over this period and there were further declines during the
Great Recession. However, experiences differed substantially by demographic group. For this
reason, I estimated group-specific declines, from group-specific trends, that occurred during the
years 2008 and 2010. By the year 2012, the population had recovered to pre-recession trends.
3
Of
the population groups studied, the foreign-born fared the worst with declines in the likelihood of
reporting the top happiness category (of three) by approximately fifteen percent. In contrast, the
full population reported a decline of approximately five percent and young adults (18-24), two
percent.
Declines in income and employment best explain the happiness declines in 2010. To
determine the potential mechanisms, micro and macro controls were added to the base

2
The author purposely contrasts the two results to highlight benefits of SWB.
3
During the years surrounding the Great Recession, the GSS is fielded every two years.
4
regressions that were used to estimate the declines. In the base regressions, the declines were
estimated as dummy variables, for the years 2008 and 2010, that capture the otherwise
unobserved factors that are coincident with those years. Income and employment variables are
shown to operate as mechanisms because their inclusion reduced the significance of the dummy
variables, thus accounting for the unobserved variation. Chapter 3 also discusses the other
important micro and macro controls and how the income and employment relations with
happiness changed during the Great Recession. Understanding the impact of the Great
Recession, and the channels through which it operated, could help policymakers plan for future
recessions.
In Chapter 4, people reporting higher SWB are shown to be less likely to be unemployed
in Germany (1996 – 2013), based on longitudinal data from the German Socio Economic Panel.
SWB is not only an outcome of interest, but also, a determinant of outcomes that economists
have traditionally been interested in. In recent years there has been a growing consensus that
psychological factors are important for performance, which represents a large improvement in
the field. Previously unobserved ability was thought of in terms of intelligence and did not
include psychological characteristics. Models of performance are now more likely to include
psychological characteristics using the Big-Five personality traits. However, relying on the Big-
Five traits is still limiting for empirical work because they are not broadly available. In contrast,
SWB is available in at least 162 countries and over a longer period of time (e.g. since the 1970s
in the U.S.). SWB positively affects a great many factors too, including health, productivity, and
social relations. The results from Chapter 4 suggest human capital models should include
psychological traits like SWB, and that social policy that improves SWB should also pay off in
terms of improved labor performance.
5
The SWB-unemployed relation exhibits a meaningful magnitude. A one standard
deviation increase in lagged SWB is associated with approximately a one-percentage point
decline the likelihood of being unemployed. Mechanisms include the Big-Five personality traits.
For example, within person increases in extraversion or emotional stability are associated with
increases in SWB. There are two important variants to the relation. First, it is quadratic in nature.
At high levels of SWB, people that report higher SWB are more likely to be unemployed.
Second, the magnitude of the SWB-unemployed relation is greater for people that are currently
unemployed, meaning SWB is more important for becoming reemployed than staying employed.
The results in chapter 4 are based on panel regressions of unemployed status (binary) on
lagged SWB and a host of additional controls, including lagged employment status or fixed
effects. The potential for important omitted variables is greatly reduced through the inclusion of
these controls. Concern of reverse causality is addressed by including lagged unemployment and
controls for labor market expectations. I also use a placebo test to show that systematic
measurement error that may be present in satisfaction variables does not account for the observed
relationship. Further support for the main results is provided by regressions that simultaneously:
are dynamic, include fixed effects, and allow for endogeneity in the controls (Arellano and Bond
1991).

1.1 What is Subjective Well-Being
Subjective well-being measures are based on the responses to survey questions concerning one’s
well-being. While there are different types of SWB, the focus in this dissertation is on evaluative
measures. As the name implies, evaluative measures ask the respondent to evaluate their lives,
often on a scale from zero to ten, with ten indicating high satisfaction with life. They provide
6
consistent and meaningful measures of well-being. They: show a high degree of correlation
between subject responses over a short period of time, are well explained by life circumstances,
predict future behavior, and correlate well with other subjective and objective measures of well-
being. Alternative experiential or emotional SWB measures, such as positive affect, are not as
closely related to life circumstances (Helliwell and Wang 2012).
4
For a further discussion of the
types of SWB questions and their reliability and validity see (Helliwell and Wang 2012; OECD
2013; Kapteyn et al. 2015).
The economics of happiness (or SWB) is becoming an increasingly important area of
research. From the year 2000 to 2011, there was a fivefold increase in the number of articles
published annually on topics pertaining to ‘happiness’ or ‘subjective wellbeing’ (Tay,
Kuykendall, and Diener 2015). Official statistics are being recorded by a growing number of
countries and organizations (including the United States and OECD) (Durand and Smith 2013),
and at least six Nobel laureates in economics have advocated obtaining subjective measures of
well-being (Stiglitz, Sen, and Fitoussi 2009).
For skeptics of SWB measures, several common concerns are easily addressed. First,
interpersonal comparison of utility can be avoided by instead comparing individuals over time
using fixed effects. Second, SWB measures are ordinal in nature, yet many researchers treat
them as cardinal. However, results do not vary much when treating the scales as ordinal or
cardinal (Ferrer-i-Carbonell and Frijters 2004). Moreover, ordered discrete choice specifications
can be used to address this concern directly, as in Chapter 2. Last, the SWB scale is bounded.
While true, there is still meaningful movement available for many countries and for population
groups within a country. Sub-Saharan Africa currently has a SWB score of less than five on a

4
An example of an experiential question is, “How happy were you yesterday?”.
7
scale from 0 – 10 (Helliwell, Layard, and Sachs 2017), and African Americans have experienced
a positive trend in the United States since the 1970s (Chapter 3).
Much of the literature uses SWB as an outcome variable. It has even been suggested as a
measure of utility. For example in a 2007 article, Gary Becker and Luis Rayo state, “… we
consider that maximizing happiness is closely linked, if not identical, to maximizing utility in the
standard economic way (Rayo and Becker 2007, 487).” However, SWB can be used as a
determinant as well. Indeed the United Nations publication, World Happiness Report (2013),
summarizes the evidence for the benefits of SWB on health and longevity, income, productivity,
and individual, organizational, and social behavior (De Neve et al. 2013, 56-57). More recently
the results published in a Journal of Labor Economics article, led the authors to say, “… it
appears that economists and other social scientists may need to pay more attention to emotional
well-being as a causal force (Oswald, Proto, and Sgroi 2015, 807).”
8
Chapter 2 Does Oil Curse Democracy? A Long-Run Time-Series Analysis of 127
Countries
5

2.1 Introduction
The resource curse is a topic studied intensively in both economics and political science. The
original discussion centered on why countries with large resource industries, typically measured
by the share of natural resource exports in Gross Domestic Product (GDP), seemed to have
slower rates of GDP growth than other countries. Many different reasons for this phenomenon
have been offered. One prominent view traces the growth shortfalls experienced in natural
resource-rich countries to their weak political and other institutions. The question then becomes
whether countries with greater natural resources happen to have weak institutions or if natural
resources retard the development of political institutions like democracy. Consequently, much of
the resource curse literature has shifted to focus on the latter question.
The purpose of this paper is to examine the long-term effects of changing oil and gas
values on an overall measure of democratic development (Polity), and three underlying
characteristics of democracy: constraints on executive power (Polcon), Civil Liberties, and
Political Rights. Most comparable studies have focused on Polity, yet disaggregation is necessary
for several reasons. First, because the effects of oil could differ from one element of democracy
to another. Second, their importance for economic growth (or objectives) might well differ.
Third, as will be shown, the actual trends in these different measures differ substantially, for
example Civil Liberties and Political Rights were fairly stagnant in the Middle East and North
Africa, but Polcon and Polity generally increased. Last, the use of additional measures increases
the variability in the data. For countries like the United States and Norway, change over time can
only be detected using Polcon, because these countries reached the upper limit of the other

5
This chapter was co-written by Kelsey J. O’Connor, Luisa R. Blanco, and Jeffrey B. Nugent.
9
measures many years ago. Nevertheless, we find that the long-run effects of oil abundance are
not statistically significant for any of the four democracy measures. Indeed, this finding is
supported by the results generated from multiple estimation strategies and separately for different
country groups and time periods.
Part of our contribution is the application of advanced time-series econometric
methodology. Most importantly, we separately identify the long-run effects of oil abundance
from short-run fluctuations. Long-term effects come from sustained changes over periods of
several years. They correspond better with the slow speed of institutional change, while annual
fluctuations in oil prices, e.g., are less likely to have lasting effects. Long-run effects are also
consistently estimated even in the presence of reverse causality. To identify long-run effects, two
different time-series models are used, specifically an error correction model and a distributed lag
model. Error correction models are more common and have been used before, but previous
researchers have typically not addressed two potentially serious sources of bias. In particular,
they assume the relationship between oil and democracy to be identical across countries, and also
do not control for cross-sectional dependence. In contrast, our analysis has the advantages of
allowing for heterogeneous relationships by country, and controlling for omitted common
correlated effects (or cross sectional dependence). Both the error correction and distributed lag
models control for these sources of bias, but they differ in other ways that are discussed in the
methods section.
While our analysis differs substantially from existing analyses, we deem it important to
also compare our results with those obtained in two recent studies that also focus on the long-run
effects of oil on democracy. Haber and Menaldo (2011) (henceforth “HM”) provides evidence
(like us) suggesting that oil abundance does not have a long-term effect on democratic
10
development (using Polity alone). However, Andersen and Ross (2014) (henceforth “AR”)
reevaluate the relationship using the same data and similar methodology, and offer support for
significant-negative effects of oil after 1980. This disagreement over the existence of long-run
effects highlights the current political-resource-curse debate. The present study suggests a
conclusion more like that of HM, but show that this conclusion holds not only for Polity but also
for Polcon, Civil Liberties, and Political Rights. In addition, when we use the same data and
techniques as used by HM and AR, we show that AR’s significant-negative effects (after 1980)
are not robust. We conclude therefore, that there is no robust evidence of a within-country
democratic resource curse.

2.2 Background
The oil as blessing or curse literature has generated many variants, which have been of great
interest to both economists and political scientists. For economists the most debated one
concerns the effect of oil endowments on long-term growth. Do they raise long-term growth
rates or lower them? Recent empirical analyses provide evidence that oil does not affect growth
rates with examples including James (2015) and Smith (2015).
6
However the existence of a
resource curse is still debated. One important arm has suggested that the link between natural
resources and growth may depend on whether or not various adverse links through the quality of
institutions (especially democracy) are realized (Andersen and Aslaksen 2008; Robinson, Torvik,
and Verdier 2006).
7

6
James (2015) explains that slower economic growth in resource-dependent countries is due to a slow-growing
resource sector, contrasting views that oil has sweeping negative effects, and Smith (2015) finds resource discovery
is in fact positively associated with the level of GDP per capita.
7
James (2015) provides a recent review on the literature on the resource curse. Another commonly asserted
institutional link between oil and growth is via corruption. Oil’s harmful effects on corruption have been shown
11
Much of the attention on the resource curse has shifted to analyzing the effect of oil
endowments on democracy. Following some earlier work by Barro (1998) showing that
countries with substantial oil (measured by an oil dummy) had a negative effect on democracy,
Acemoglu and Robinson (2006), Haber and Menaldo (2011), and Menaldo (2013) provided a
rationale in terms of state capabilities, which includes the ability to collect taxes. For citizens to
be willing to pay taxes, they are likely to insist that the state agrees to let them have a say in what
the state does, i.e. by adopting democratic institutions. If the state has an alternative means of
sustaining itself, such as by selling off its oil or other natural resources, it can avoid the costly
process of developing the capability to tax its citizens and having to give those citizens more say
over what the government does. Only when the oil-endowed country adopts institutions like the
effective rule of law and democracy can such countries use oil to their advantage. The political
version of the oil curse (Mahdavy 1970; Beblawi and Luciani 1987; Przeworksi et al. 2000; Ross
2001; Ross 2012) however, argues that oil rents directly retard the development of democracy
through increased rent-seeking behavior, rent grabbing, and leaders locking out the rivals.
8

The number of studies on this political version of the resource curse has grown
significantly. Ahmadov (2014) has conducted a meta–regression analysis on the basis of data
taken from 29 existing publicly available studies on the relation between oil rents and
democracy. After controlling for the many different methodological differences, time periods,
and country coverage, Ahmadov (2014) finds the overall effect of oil rents on democracy to be
small, negative, and significant, but with considerable heterogeneity across regions, being

empirically numerous times, examples including: Ades and Di Tella (1999) (one of the first); Vicente (2010), which
exploits a natural experiment; and Brollo et al. (2013) which provides plausibly causal estimates in Brazil.
8
The rationale for the rent-seeking link between oil abundance and lower levels of income and welfare is that the
rents from natural resources are likely to divert entrepreneurs from productive activities into rent-seeking ones
Torvik (2002).
12
somewhat more negative among countries of the Middle East and North Africa but positive and
highly significant in Latin America.

2.3 Data
2.3.1 Variables, Data, and Sample Coverage
Oil abundance is measured as oil and gas value per capita (Ross and Mahdavi, 2015). Oil and gas
value is defined as extraction of oil and gas multiplied by the unit price. The data are considered
to be superior to oil rents data obtained from the World Development Indicators (WDI).
9
For one
reason, oil and gas value is based on data from multiple sources (including the WDI), which
increases the sample coverage (Mohtadi, Ross, and Ruediger 2015). For purposes of estimation,
the natural log of real oil and gas value per capita are used (henceforth “O&G Value”).
As noted above, we use four complementary but distinct measures of democratic
institutions: Polity, Civil Liberties, Political Rights, and Polcon. Polity (Polity 2011) is perhaps
the most commonly used indicator of democracy. It is “an index of the competitiveness of
political participation, the openness and competitiveness of executive recruitment, and the
constraints on the chief executive” (HM, 4).
Civil Liberties and Political Rights disentangle two attributes of democracy that might
otherwise be concealed by Polity. BenYishay and Betancourt (2014) argue that Civil Liberties
are more persistent than Political Rights and tend to encourage the development of Political
Rights to a greater extent than Political Rights do for Civil Liberties. For this reason, Civil
Liberties would seem to be more important in the promotion of democracy in the long run.
Political Rights relate to free and fair elections, while Civil Liberties refer to the protection of

9
The WDI rents data are nationally reported, and as such could be biased by institutional characteristics, and
because the cost component is generally estimated at one point in time, rents do not accurately change over time.
(Mohtadi, Ross, and Ruediger 2015).
13
individual rights such as freedom of speech, freedom of assembly and equal treatment under the
law (Freedom House, 2013).
Our fourth measure of democratic institutions is Polcon (Polcon 2012). Distinct from the
other three variables, it measures “the number of independent veto points over policy outcomes
and the distribution of preferences of the actors that inhabit them (Henisz 2002, 380).” Polcon
scores rise with the number of veto players,
10
which places additional constraints on the
executive and are thought to increase regulatory stability and credibility. Several studies of oil
abundant countries have shown that government expenditures are far less volatile, procyclical,
and erratic when the chief executive is constrained by one or more veto players as captured by
Polcon (Collier and Hoeffler 2009; Elbadawi and Soto 2016; Selim and Zaki 2016).
The period of analysis (1974-2012) was based on data availability for the key variables.
In particular Civil Liberties and Political Rights only become available in 1974. The sample of
127 countries is also based on the availability of data. Countries with less than 30 years of
coverage were excluded from the sample, because the analysis involves estimation of country-
specific relationships, which requires a sufficient number of observations (Chudik et al. 2013,
26).
GDP data are primarily from the WDI (World Bank 2013). When GDP data from this
source did not extend back into the 1970s, growth rates implied by data from the Penn World
Tables (PWT) (Heston et al. 2012) or an earlier WDI publication (World Bank 1995) were used
to extrapolate backwards.

10
Veto players are most commonly the lower and upper branches of the legislature, the judiciary, other units of
the government, or subnational governments.
14
2.3.2 Subsample Definitions
Once the sample was determined for our base analysis, two further classifications were used.
First, we break the sample into countries that have mature oil industries and countries with less
experience. We also break the sample into three regional classifications, Middle East and North
Africa (MENA), Latin America and the Caribbean (LAC), and the rest of the world (ROW). The
rationale for using these categories is discussed below.

Industry Tenure Motivation:– The timing of oil discovery is likely to affect how oil abundance
affects institutions. Countries that discovered oil later may have had the time necessary to
develop sufficient state capacity to avoid the adverse effects of oil, such as: increased civil
conflict (Collier and Hoeffler 2005), reduced transparency (Mohtadi, Ross, and Ruediger 2015),
and increased fiscal volatility. Indeed the evidence supports this argument. Tsui (2011) found
that those countries that experienced major oil discoveries early in time developed democracy
significantly more slowly than countries that did not. However, countries that were at least
somewhat democratic initially did not experience the negative effects of oil.
The evidence from Omgba (2015) also suggests that mature producers are likely to be
less democratic. The main finding from Omgba (2015) is that both Polity and the Voice and
Accountability index (the most democracy-oriented measure from the World Bank’s World
Governance Indicators) are positively related to the period of time between the beginning of oil
production and a country’s political independence, which means that a country that discovered
oil later (relative to political independence) exhibits higher levels of democracy. He argues that
oil discovery before political independence negatively affects export diversity and institutional
development. Blanco, Nugent, and O’Connor (2015) also considers oil industry maturity, but
15
found somewhat nuanced effects, the number of years since peak oil discovery having a positive
effect on government stability, a negative effect on bureaucratic quality, while oil had no causal
effect on democracy itself.
To allow for different effects of oil by industry experience, we divide the sample into
Mature and non-Mature producers (also referred to as Low-Experienced). Countries that
discovered oil before the median discovery year (1959) were considered mature. The year of oil
discovery is available from The World History of Oil and Gas (Geo-Help Inc. 2011).

Geographic Differences:– Treating the Middle East and North African and Latin American
groups separately is important because there are important differences between them in their
historical experience, culture, and socio-economic conditions. Not surprisingly, and as noted
above, much of the existing literature has shown that the relationship between democracy and oil
varies substantially across different parts of the world. In particular, the Middle East and North
Africa has long been identified as a region with both substantial amounts of oil and very little
democracy. When oil was discovered in the Middle East and North Africa between 1930 and
1965, these countries were extremely low on educational attainment, state capacity, and
urbanization, which helps explain their low initial levels of democracy. While democracy has
been on the rise throughout much of the world since then, its level has remained low in MENA
(Huntington 1991; Przeworksi et al. 2000).
The low levels of democracy in MENA were partially attributed to oil. Barro (1998) used
pooled international cross-sections to show that urbanization, education, life expectancy and
lagged democracy all had positive effects on subsequent democracy and that an oil dummy
variable had a significant negative effect. More recent studies have continued to highlight the
16
failure of MENA countries to democratize over time (Ross 2012, 75; Hertog 2007). Herb (2005)
also showed that MENA countries had the lowest democracy scores in the world, but attributed
much of this to the influence of regional neighbors because it also applied to MENA countries
without oil. The effects of regional neighbors are accounted for in our analysis, which is
discussed further in the methods section.
Other scholars, however, have pointed to another distinctive pattern in Latin America.
For example, Dunning (2008) pointed to the fact that in Latin America oil seemed to be
positively related to the adoption of democracy. There are several possible explanations for this
apparent exception. First, countries with greater oil rents may face less resistance to
democratization from wealthy elites. Wealthy elites are fearful that democracy will bring greater
taxation and expropriation of private wealth, especially in high income-inequality nations like
those in Latin America. However, oil wealth reduces government’s need for non-resource
revenues, and therefore elites’ fear of democracy. Additional explanations for Latin America’s
exceptionalism may be (1) that their governments generated mass political support through
skilled control and redistribution of the resource rents, or (2) that they simply have been less
dependent on natural resource rents than the Middle Eastern and North African countries, for
example (Sinnott, Nash, and Torre 2010).
The evidence supports Latin America’s exceptionalism. Ross (2012, 74, 85–86) concedes
this point, but notes the explanation is unclear. Sinnott, Nash, and Torre (2010) provides support
for Dunning's (2008) theory, but also discusses how the relationship may have changed, because
the recent resource booms seem to have weakened, not strengthened, democracy in Latin
America. The authors call attention to the need for further research based on more recent data.
17
Appendix Table A-1 includes summary statistics for the sample as a whole, and for each
country group separately. The list of countries in each group is included in Appendix Table A-8.

2.4 Methods
To estimate the effects of O&G Value on institutional development, we use two forms of an
error-correcting autoregressive distributed lag (ARDL) model and a distributed lag model.
Before introducing the specific models, it is helpful to understand the assumed data generating
process presented below.
𝐼𝑛𝑠𝑡
!"
= 𝜆
!"
𝐼𝑛𝑠𝑡
!,!!!
!
j=1
+ 𝛿
!"
𝑂𝑖𝑙
!,!!!
q
j=0
+𝜇
!
+𝜀
!"
(2-1)
𝐼𝑛𝑠𝑡
!"
represents the institutional measure (Polity, Polcon, Civil Liberties, or Political Rights) for
country i at time t. Each institutional level is assumed to depend on a set of its past levels, on
𝑂𝑖𝑙
!,!
(O&G Value), and its past levels. Country heterogeneity is allowed by using country-
specific coefficients and fixed effects 𝜇
!
. O&G Value is used as the sole explanatory variable
because oil is theorized to have sweeping economic and social effects (as in Ross, 2012), and
other controls might have the effect of capturing some of the possible channels through which
O&G Value may affect institutions. As such, our empirical method aims to estimate the full
effects of oil on institutions, although admittedly not the channels through which they operate.
For estimation, we reparameterized Equation (2-1) into an error correction form. In error
correction form, the long-run relationship between oil and institutions can be consistently
estimated separately from short-term fluctuations. The error correction form is appropriate
because annual fluctuations in O&G Value are unlikely to have lasting impacts on institutions
while sustained increases over a few years may have such effects. Estimates from Equation (2-1)
18
could be confounded by noise from annual fluctuations, while Equation (2-1) in differences
would limit the interpretation of the results to the short-term changes.
An error correction form of an ARDL model is presented below as Equation 2-2.
𝛥𝐼𝑛𝑠𝑡
!"
= 𝜙
!
𝐼𝑛𝑠𝑡
!,!!!
− 𝜃
!
𝑂𝑖𝑙
!"
+ 𝜆
!"
∗
𝛥𝐼𝑛𝑠𝑡
!,!!!
!!!
j=1
+ 𝛿
!"
∗
𝛥𝑂𝑖𝑙
!,!!!
q-1
j=0
+𝜇
!
+𝜀
!"
(2-2)
Where: 𝜙
!
= − (1− 𝜆
!"
!
j=1
); 𝜃
!
= 𝛿
!"
q
j=0
/(1− 𝜆
!" !
);
𝜆
!"
∗
=− 𝜆
!"
!
m=j+1
; and 𝛿
!"
∗
=− 𝛿
!"
!
m=j+1
.
The short-run relations are captured by 𝛿
!"
∗
, while the long-term relationship is captured by 𝜃
!
.
With a large change in 𝑂𝑖𝑙
!"
, the response in the dependent variable might overshoot the long-run
equilibrium relationship. When this happens, the error correction term, 𝜙
!
, serves to bring the
relationship back to the long-run one (hence the term “error correction”). The lag orders p and q
were set to three, using Akaike and Schwarz’s Bayesian information criteria.
11

For an error correction model to be appropriate, (1) the error correction term should be
statistically significant, negative, but greater than negative two, and (2) there must be a long-run
cointegrating relationship between the level variables.
12
Both institutional development and
O&G Value are likely to be persistent and have trends of integration order one (I(1)), and the
cointegrating relationship, 𝜃
!
, must exist for the term 𝐼𝑛𝑠𝑡
!,!!!
− 𝜃
!
𝑂𝑖𝑙
!"
to be stationary.
Stationarity is necessary for the equation to balance when 𝜙
!
is statistically significant. Results
for the necessary tests are discussed further in the results section.

11
The lag order was selected separately by country and democracy variable and then averaged. Averages were
between two and three for each variable and then rounded up to three. The results are also not very sensitive to lag
order. We find similar results when using two lags instead of three lags.
12
The first condition is met in all estimations (see the coefficient on the lagged dependent variable in Appendix
Tables A-2 – A-5 for estimates of the error correction term). Cointegration is discussed at the end of the Results
section.
19
ARDL models have several advantages. The long-run effects are consistent, regardless of
whether the level variables are stationary or trended, and they are consistent in the presence of
reverse causality (Chudik et al. 2013, 26; Pesaran 2015). We use two different error correcting
ARDL models that differ in the amount of restrictions applied. The most restrictive model
assumes the relationships are common across panels (basic dynamic model with panel fixed
effects, called “DFE”). We focus on the most flexible model, the mean group (MG) model,
which simply estimates the ARDL model separately for each panel (allowing for heterogeneous
short-term and long-term effects) and then reports the mean coefficients across panels. The MG
model loses efficiency compared to the DFE model but is more appropriate when the estimated
relationships are heterogeneous across countries.
There is another ARDL model that is a hybrid of the MG and DFE models, called the
pooled mean group (PMG). It allows the short-run effects to differ across countries, but assumes
the long-term effects are common. The PMG model should be more efficient than the MG, but
less efficient than the DFE and possibly still subject to bias. Consequently we focus on the more
consistent MG model, but present the most efficient (DFE) results for robustness purposes, and
also because DFE models are common in the related literature (e.g. HM).
Our specifications also correct for cross-sectional dependence using a common correlated
effects approach, which simply adds cross-sectional means of the dependent and independent
variables and their lagged values to the ARDL specification (as suggested by Chudik and
Pesaran, 2013). Cross-sectional dependence arises when the error terms are correlated across
observations. In the present analysis, cross-sectional dependence could be caused by an omitted
common correlated effect, e.g. oil prices. Controls for oil prices, subtracting the cross-sectional
20
mean of each variable, clustering errors, or simply using year fixed effects could partially
account for common correlated effects, but these approaches have limitations (Westerlund 2007).
Equation 2-3 below presents the same error correction model as in Equation 2-2, but adds
the modification for cross-sectional dependence, referred to as a cross section augmented error
correction (CS-EC) model.
𝛥𝐼𝑛𝑠𝑡
!"
= 𝜙
!
𝐼𝑛𝑠𝑡
!,!!!
− 𝜃
!
𝑂𝑖𝑙
!"
+ 𝜆
!"
∗
𝛥𝐼𝑛𝑠𝑡
!,!!!
!!!
j=1
+ 𝛿
!"
∗
𝛥𝑂𝑖𝑙
!,!!!
q-1
j=0
+𝜇
!

+ 𝜓
!"
𝑍
!!!
!
!!!
+𝜀
!"
(2-3)
Where 𝑍
!
= (𝐼𝑛𝑠𝑡
!
,𝑂𝚤𝑙
!
); 𝐼𝑛𝑠𝑡
!
= 𝑁
!!
𝐼𝑛𝑠𝑡
!"
!
!!!
; 𝑂𝚤𝑙
!
= 𝑁
!!
𝑂𝑖𝑙
!"
!
!!!

Using the common correlated effects approach, the effects of O&G Value are estimated from the
relationship between idiosyncratic deviations from mean O&G Value levels and idiosyncratic
institutional deviations from the mean. 𝑍
!
captures the average relationships, observed and
unobserved.
13
Average levels of factors that affect institutions are accounted for, such as the
effects of neighbors, regional institutional change, economic development, and urbanization.
Factors affecting O&G Value, such as oil prices are also captured. Lagged cross-sectional means
capture any serial correlation in omitted factors. The lag order is set as the integer value of the
cube root of the number of time periods, in our case three lags are used (Chudik et al. 2013).
We also use a cross section augmented distributed lag (CS-DL) model. It differs from the
CS-EC models, is considered complementary, and possibly superior when using highly persistent
variables, such as democracy. The specification of this model (Equation 2-4) is similar to the CS-
EC model, but the dependent variable is modeled in levels, and the specification does not include
lags of the dependent variable as explanatory variables (in levels or changes).

13
Bove, Elia, and Smith (2015, 8) identify another benefit of the common correlated effects approach. It has
similarities to counterfactual analysis because the cross-sectional averages are similar to predicted counterfactuals.
21
𝐼𝑛𝑠𝑡
!"
= 𝜃
!
𝑂𝑖𝑙
!"
+ 𝛿
!"
∗
𝛥𝑂𝑖𝑙
!,!!!
q-1
j=0
+𝜇
!
+ 𝜓
!
!
𝐼𝑛𝑠𝑡
!
+ 𝜓
!"
!
𝑂𝚤𝑙
!!!
!
!!!
+𝜀
!"
(2-4)
As in the CS-EC model, the lag order on 𝑂𝚤𝑙
!!!
is set as the integer value of the cube root of the
number of time periods (three). However, unlike the CS-EC model, the lag order on the cross-
sectional mean of 𝐼𝑛𝑠𝑡
!
is set to zero (Chudik et al. 2013, 15).
Each of the models, two CS-EC and one CS-DL model, are used in estimating the results.
It should be remembered that the CS-EC models are each specified in error correction form and
differ simply in whether or not the short-run and long-run relationships are assumed to be
homogeneous across countries. Of the CS-EC models, the mean group, cross-sectionally
augmented, error correction model (MG CS-EC) is the least restrictive, allowing for
heterogeneous short- and long-run relationships. The DFE CS-EC model is the most restrictive,
assuming homogeneous short- and long-run relationships. The MG CS-DL model differs from
the MG CS-EC model because it uses the distributed lag specification (instead of an error
correction), but is similar in that it allows for heterogeneous short- and long-run relationships.

2.5 Results
2.5.1 Polity, Polcon, Civil Liberties, Political Rights, and O&G Value Over Time
Figure 2-1 illustrates how the levels of Polity, Polcon, and O&G Value have moved over time in
the full sample of 127 countries and in each of the country groups identified above. The series
have been smoothed to help reduce annual fluctuations and more closely represent long-run
trends. In most of the country groups, O&G Value declined from the mid to late 1970s until the
early 1990s when O&G Value began to increase again. Both Polity and Polcon generally
increased like democracy. However, Polity started at a higher level than Polcon and increased at
22
a greater rate. Figure 2-2 presents corresponding trends for Civil Liberties and Political Rights,
which like Polity and Polcon were generally increasing.

Figure 2-1 Polity, Polcon, and O&G Value p.c. Lowess Smoothed Trends (1974 – 2012)
Note: Polity rescaled to range from 0-1.
Source: Geo-Help Inc. (2011), Polcon (2012), Polity (2011), Ross and Mahdavi (2015).
Polcon
Polity
Oil (scale-->)
2 4 6 8 10
ln(Mean O&G Val. pc)
0 .2 .4 .6 .8 1
Institutions (Scale 0-1)
1970 1980 1990 2000 2010
Full Sample (127 Countries)
Polcon
Polity
Oil (scale-->)
2 4 6 8 10
ln(Mean O&G Val. pc)
0 .2 .4 .6 .8 1
Institutions (Scale 0-1)
1970 1980 1990 2000 2010
Rest of World (86 Countries)
Polcon
Polity
Oil (scale-->)
2 4 6 8 10
ln(Mean O&G Val. pc)
0 .2 .4 .6 .8 1
Institutions (Scale 0-1)
1970 1980 1990 2000 2010
Middle East and North Africa (18 Countries)
Polcon
Polity
Oil (scale-->)
2 4 6 8 10
ln(Mean O&G Val. pc)
0 .2 .4 .6 .8 1
Institutions (Scale 0-1)
1970 1980 1990 2000 2010
Latin America and Carribean (23 Countries)
Polcon
Polity
Oil (scale-->)
2 4 6 8 10
ln(Mean O&G Val. pc)
0 .2 .4 .6 .8 1
Institutions (Scale 0-1)
1970 1980 1990 2000 2010
Mature Producers (41 Countries)
Polcon
Polity
Oil (scale-->)
2 4 6 8 10
ln(Mean O&G Val. pc)
0 .2 .4 .6 .8 1
Institutions (Scale 0-1)
1970 1980 1990 2000 2010
Less-Experienced Producers (43 Countries)
23

Figure 2-2 Civil Liberties, Political Rights, and O&G Value p.c. Lowess Smoothed Trends
(1974 – 2012)
Note: Civil Liberties and Political Rights rescaled to range from 0-1.
Source: Freedom House (2013), Geo-Help Inc. (2011), Ross and Mahdavi (2015).
PR
CL
Oil (scale-->)
2 4 6 8 10
ln(Mean O&G Val. pc)
0 .2 .4 .6 .8 1
Institutions (Scale 0-1)
1970 1980 1990 2000 2010
Full Sample (127 Countries)
PR
CL
Oil (scale-->)
2 4 6 8 10
ln(Mean O&G Val. pc)
0 .2 .4 .6 .8 1
Institutions (Scale 0-1)
1970 1980 1990 2000 2010
Rest of World (86 Countries)
PR
CL
Oil (scale-->)
2 4 6 8 10
ln(Mean O&G Val. pc)
0 .2 .4 .6 .8 1
Institutions (Scale 0-1)
1970 1980 1990 2000 2010
Middle East and North Africa (18 Countries)
PR
CL
Oil (scale-->)
2 4 6 8 10
ln(Mean O&G Val. pc)
0 .2 .4 .6 .8 1
Institutions (Scale 0-1)
1970 1980 1990 2000 2010
Latin America and Carribean (23 Countries)
PR
CL
Oil (scale-->)
2 4 6 8 10
ln(Mean O&G Val. pc)
0 .2 .4 .6 .8 1
Institutions (Scale 0-1)
1970 1980 1990 2000 2010
Mature Producers (41 Countries)
PR
CL
Oil (scale-->)
2 4 6 8 10
ln(Mean O&G Val. pc)
0 .2 .4 .6 .8 1
Institutions (Scale 0-1)
1970 1980 1990 2000 2010
Less-Experienced Producers (43 Countries)
24
Differences between the democratic measures highlight why it is important to consider
more than Polity alone. Latin America and the Caribbean provide an example. Although Polity
would predict stagnant democratic change during the 2000s, Polcon would suggest it declined.
Similarly, in the Middle East and North African countries, both Polity and Polcon increased
throughout the period, while Civil Liberties and Political Rights largely declined until the mid-
1990s.
The best visual support for a relationship between O&G Value and democratic
institutions appears to be for a negative effect of O&G Value on Polcon in Latin America and the
Caribbean. However, they both end the period at higher levels, which would suggest a positive
relation. It is also important to remember that the mean institutional and oil variables also serve
as controls. A statistical analysis is needed to identify the long-run effects of idiosyncratic
changes in oil on democratic institutions, to which we now turn.

2.5.2 Statistical Analysis
Table 2-1 presents the long-run effects of O&G Value on each of the four institutional variables.
The absence of a relationship is demonstrated using the MG CS-EC and MG CS-DL
specifications discussed in the methods section. The column heads indicate the models used, and
the full regression results for the MG CS-EC models are available in Appendix Tables A-2, A-3,
A-4, and A-5). The results are consistently insignificant for the full sample and in most of the
country groups. If the political-resource curse were true, certainly one would expect it show up
among the producers. Yet collectively, the Mature and non-Mature groups cover the sample of
producers, and even there, the effects are nil. In MENA, too, the effects are insignificant.
25
There is some support for a negative O&G Value-democracy relationship in the Latin
America and Caribbean group, however that is somewhat unexpected. As discussed above, Latin
America is thought to be an exception to the political-resource curse (Dunning 2008; Ahmadov
2014). The relationship is also not robust. It only holds at five percent significance for the MG
CS-DL model, and even then, not for Polcon, which would seem to be supported by the Figure 2-
2.
Table 2-1 Long-Run Effect of O&G Value pc on Democracy. Two Mean Group Models.
Various Indicators and Country Groups.

The DFE CS-EC model results are still generally statistically insignificant, even with increased
estimation efficiency (by assuming the short- and long-run relations are the same across
countries). Table 2-2 presents the long-run effects of oil on each institutional variable from the
DFE CS-EC model. In each country group the effects are insignificant except for one. The long-
Polity Polcon Civil Liberties Political Rights
Country Group CS-EC CS-DL CS-EC CS-DL CS-EC CS-DL CS-EC CS-DL
Full -0.040 -0.226 0.003 -0.004 0.044 0.043 -0.083 -0.051
[-0.196] [-1.039] [0.323] [-0.367] [0.819] [0.934] [-0.860] [-0.856]
Mature 0.164 -1.446 0.029 -0.000 0.268 -0.049 0.070 -0.030
[0.093] [-1.435] [0.632] [-0.004] [0.873] [-0.209] [0.140] [-0.090]
Non-Mature -0.063 -0.056 -0.001 -0.012 0.021 0.007 -0.029 0.009
[-0.390] [-0.229] [-0.067] [-0.962] [0.399] [0.175] [-0.254] [0.130]
Latin America -1.582 -1.921* -0.029 -0.042 -0.326+ -0.245* -0.510+ -0.428*
[-1.469] [-2.148] [-0.385] [-0.957] [-1.737] [-2.041] [-1.837] [-2.072]
Middle East and -0.609 -0.145 0.000 0.000 0.067 0.050 0.032 0.101
North Africa [-0.957] [-0.725] [0.006] [0.017] [0.582] [0.641] [0.198] [0.684]
Rest of World -0.243 -0.036 -0.006 -0.009 0.047 0.013 -0.085 -0.004
[-1.125] [-0.271] [-0.511] [-0.813] [0.841] [0.318] [-1.047] [-0.058]
Note: Mean group allows heterogeneous short- and long-run relations by country.
CS-EC – cross-sectionally augmented error correction model (Equation 3 in the text). CS-DL – cross-sectionally
augmented distributed lag models (Equation 4 in the text). Oil & Gas value per capita are used in natural log
form. Polity scaled from -10 (autocracy) to 10 (democracy). Polcon scaled 0-1, with increasing values indicating
greater constraints. Civil liberties and Political Rights are scaled from 1 to 7, with 7 indicating greater civil
liberties or political rights. Group composition presented in Table A. 7. The complete CS-EC regression results
are presented in Appendix Table A. 2.
t statistics in brackets; + p<0.10 * p<0.05 ** p<.01 *** p<.001.
Source: Freedom House (2013), Geo-Help Inc. (2011), Heston et al. (2012), Polcon (2012), Polity (2011), Ross
and Mahdavi (2015), World Bank (1995; 2013).

26
run effects are negative and significant in the Mature group. Although this result may be
consistent with expectations (Tsui 2011; Omgba 2015), there are two reasons to doubt it. First it
is subject to bias compared to results of the MG CS-EC model. Second, it does not persist
through the robustness checks as will be shown. For Latin America and the Caribbean too, the
significant relationship is no longer present. Overall, the results suggest there is no long-run
relationship between the O&G Value and Polity, Polcon, Civil Liberties or Political Rights
during the period 1974 - 2012. In what follows we provide several robustness checks and the
cointegration results.
Table 2-2 Long-Run Effect of Oil & Gas Value pc on Democracy. CS-EC Fixed-Effects
Models. Various Indicators and Country Groups.

Country Group CL PR Polity Polcon
Full -0.012 -0.021 -0.009 -0.003
[-0.520] [-0.695] [-0.114] [-1.086]
Mature -0.328* -0.443* 0.001 -0.046*
[-2.490] [-2.454] [0.001] [-2.266]
Non-Mature -0.027 -0.037 -0.017 -0.002
[-0.905] [-1.126] [-0.208] [-0.697]
Latin America 0.044 -0.331 -0.000 -0.051+
[0.208] [-1.153] [-0.000] [-1.863]
Middle East and 0.008 0.021 0.016 -0.007
North Africa [0.350] [0.717] [0.011] [-0.833]
Rest of World -0.022 -0.023 0.006 -0.003
[-0.807] [-0.647] [0.033] [-0.807]
Note: Fixed effects models allow for heterogeneous intercepts by country, but restrict
slopes to be common across countries. CS-EC – cross-sectionally augmented error
correction model. Oil & Gas value per capita are used in natural log form. Polity
scaled from -10 (autocracy) to 10 (democracy). Polcon scaled 0-1, with increasing
values indicating greater constraints. Civil liberties and Political Rights are scaled
from 1 to 7 with more 7 indicating greater civil liberties or political rights. Group
composition presented in Table A. 7. The specification is based on Equation 3 in the
text, but simplified to restrict the short- and long-run relations to be the same across
countries.
t statistics in brackets; + p<0.10 * p<0.05 ** p<.01 *** p<.001.
Source: Freedom House (2013), Geo-Help Inc. (2011), Heston et al. (2012), Polcon
(2012), Polity (2011), Ross and Mahdavi (2015), World Bank (1995; 2013).

27
Robustness:– The robustness tests we conduct include specification changes to the base MG
CS-EC models.
14
The changes are as follows: (a) substitution of the oil measure from the natural
log of oil and gas value per capita to oil and gas value as a percent of GDP, (b) changing the
sample from 1974 – 2012 to post-1980 as in most of the AR results, and (c) the addition of non-
oil GDP per capita as a control.
15

We expected the results to be similar across the different tests, but with some changes.
Oil and gas value as a share of GDP could be argued to be less a measure of oil abundance and
more a measure of dependence, in which case, they might be expected to be more negatively
associated with democratic institutions than O&G Value (natural logs of oil and gas value per
capita). Second, AR argue that oil should be more negatively associated in the post-1980 period.
Lastly, including non-oil GDP may affect the results because GDP per capita has a modernizing
influence and could be one of the channels through which oil affects institutions. If oil reduces
growth, and growth is positively related to institutional development, then controlling GDP
would reduce any negative effects of oil on institutions. Specifically, our expectations are that
the changes (a) and (b) would result in more negative relations, while change (c) would result in
a more positive relation.
We present the results in Table 2-3. Estimates obtained from the alternative specifications
are shown one change at a time. The first column reports the results from the base MG CS-EC

14
The MG CS-EC models were used as the base model for robustness tests because the EC models are more
common in the literature than CS-DL models. Nevertheless, unreported results conducting the same robustness tests
using the MG CS-DL model provide similar conclusions.
15
We previously excluded GDP per capita as a control because oil production is directly included in GDP, and
because oil may also affect non-oil GDP. To account for the first problem we estimated GDP per capita excluding
the portion due to oil revenues and used the adjusted non-oil measure of GDP per capita as a control. If oil also
affects non-oil GDP, we may expect the relation between oil and institutions to change, which is discussed further
below.
28
model in Table 2-1. The results from change (a) are reported in the second column and changes
(b) and (c) in the third and fourth columns, respectively.
The long-run effect of O&G Value on democracy is statistically insignificant in the full
sample and in each of the country groups, with a couple of exceptions. There is a significant
negative relationship once again in Latin America. During the post-1980 period the negative
long-run effects of O&G Value on Civil Liberties increase as expected. There are also negative
long-term effects of O&G Value on Polity in the Middle East and North African countries (in
post-1980 and with non-oil GDP controlled), but they are significant only at the ten percent
level. Although there are a few somewhat significant results, they are not consistent across
democracy variables or the robustness checks. As mentioned, the robustness tests are based only
on the MG CS-EC model, however, comparable results from the MG CS-DL model support the
same story.
Of all the previous results there was not one showing a significant relationship for the full
sample. The negative effects in Latin America and the Caribbean are unexpected (compared to
MENA, for example), and they generally do not hold using the more efficient DFE CS-EC
model or under the robustness checks. The significant effects in the Mature producers when
using the DFE CS-EC are supported by theory but are not robust.

Cointegration:– Cointegration testing is not necessary in the present case, but discussed for
completeness. Testing is unnecessary because we did not find robust long-term effects of O&G
Value on democracy. If we had, cointegration would have been necessary to assure the
relationship was not spurious. That is because the regression of one non-stationary (e.g. unit-
root) series on another is likely to yield a statistically significant result even for two unrelated
29
series. However, if a linear combination of two unit-root processes is stationary, then the
variables are said to be cointegrated and the relationship is not spurious (Engle and Granger
1987). The implication is that cointegration testing should first be conducted to identify whether
a long-run relationship exists. If one does exist, then the error correction models can be used to
identify whether it is statistically significant and in what direction.

Table 2-3 Long-Run Effect of Oil & Gas Value pc on Democracy. CS-EC Mean Group with
adjustments. Various Indicators and Country Groups.

Panel A: Polity
Country Group Base Oil Rent % of GDP Post 1980
Full -0.040 9.891 -0.089
[-0.196] [1.281] [-0.506]
Mature 0.164 -0.387 -0.217
[0.093] [-0.751] [-0.143]
Non-Mature -0.063 14.696 0.209
[-0.390] [0.823] [0.835]
Latin America -1.582 0.635 -1.194
[-1.469] [1.085] [-1.107]
Middle East and -0.609 -2.682 -0.936+
North Africa [-0.957] [-1.295] [-1.838]
Rest of World -0.243 9.670 -0.266
[-1.125] [0.992] [-1.179]
Panel B: Polcon
Country Group Base Oil Rent % of GDP Post 1980
Full 0.003 -0.629 -0.002
[0.323] [-0.684] [-0.203]
Mature 0.029 0.047 -0.076
[0.632] [0.630] [-1.280]
Non-Mature -0.001 -1.528 -0.007
[-0.067] [-0.806] [-0.386]
Latin America -0.029 -0.015 -0.041
[-0.385] [-0.573] [-0.700]
Middle East and 0.000 0.012 0.001
North Africa [0.006] [0.412] [0.046]
Rest of World -0.006 -0.607 0.002
[-0.511] [-0.761] [0.190]
GDP Control
0.083
[0.300]
1.014
[0.679]
-0.033
[-0.086]
-1.344
[-1.057]
-1.439+
[-1.887]
-0.089
[-0.263]
GDP Control
-0.025
[-0.802]
-0.012
[-0.131]
0.003
[0.084]
-0.016
[-0.135]
-0.042
[-0.785]
0.010
[0.427]
30
Table 2 3 Continued Long-Run Effect of Oil & Gas Value on Democracy. CS-EC Mean
Group with adjustments. Various Indicators and Country Groups.

In the present case, O&G Value and each of the democracy variables are shown below to exhibit
non-stationary unit-root processes in the full sample, and many of the subsamples, implying that
cointegration tests would be necessary to identify a non-spurious relationship. Appendix Table
A-6 presents the test results for integration order and stationarity. In most instances, the results
Panel C: Civil Liberties
Country Group Base Oil Rent % of GDP Post 1980
Full 0.044 4.714 0.013
[0.819] [1.173] [0.276]
Mature 0.268 1.288 -0.135
[0.873] [1.210] [-0.398]
Non-Mature 0.021 21.075 0.078
[0.399] [1.155] [1.359]
Latin America -0.326+ 0.140 -0.533**
[-1.737] [1.426] [-2.599]
Middle East and 0.067 -0.640 0.015
North Africa [0.582] [-1.002] [0.112]
Rest of World 0.047 5.720 -0.022
[0.841] [1.217] [-0.400]
Panel D: Political Rights
Country Group Base Oil Rent % of GDP Post 1980
Full -0.083 1.915 -0.084
[-0.860] [0.617] [-0.839]
Mature 0.070 0.529 0.499
[0.140] [0.727] [0.683]
Non-Mature -0.029 8.526 0.012
[-0.254] [0.741] [0.111]
Latin America -0.510+ -0.005 -0.431
[-1.837] [-0.073] [-1.159]
Middle East and 0.032 1.506 0.016
North Africa [0.198] [0.979] [0.090]
Rest of World -0.085 2.327 -0.073
[-1.047] [0.587] [-1.024]
GDP Control
-0.147
[-1.568]
-0.268
[-0.645]
-0.058
[-0.670]
0.017
[0.061]
-0.331
[-1.579]
-0.062
[-0.603]
GDP Control
-0.093
[-0.667]
-0.087
[-0.155]
0.103
[0.441]
0.064
[0.136]
0.071
[0.318]
-0.117
[-1.072]
Note: See notes from Table 1. Base model from Table 1 (MG CS-EC).
t statistics in brackets; + p<0.10 * p<0.05 ** p<.01 *** p<.001.
Source: Freedom House (2013), Geo-Help Inc. (2011), Heston et al. (2012), Polcon
(2012), Polity (2011), Ross and Mahdavi (2015), World Bank (1995; 2013).
31
suggest the likelihood of unit-root processes. The first section of each panel shows the results
from unit root tests of the variables in levels. They are performed using Dickey Fuller, Fisher
type panel tests (Choi 2001), where the null hypothesis is that all panels contain a unit root. The
second section of each panel shows the test results performed on the variables in first differences.
Generally, we fail to reject unit root in levels, but reject unit root in first differences. However,
there are some exceptions. For example, in the Latin America and Caribbean group, the tests
show O&G Value may exhibit a unit-root process, yet the democracy variables are likely to be
stationary in levels. When O&G Value and democracy are of different integration orders, it is
unlikely they could be cointegrated.
There is conditional support for a long-run (cointegrating) relationship in the full sample
for Polcon, Civil Liberties, and Political Rights, but not Polity. That means any significant long-
run effects of O&G Value on Polity are likely to be spurious (supported by HM cointegration
testing), but long-run effects of O&G Value on the other democracy variables are possible.
Appendix Table A-7 presents the results of the Westerlund panel cointegration tests.
Bootstrapped critical values were used to allow for cross-sectional dependence (Westerlund
2007). In the full sample, we reject no cointegration for Polcon, Civil Liberties, and Political
Rights, but not for Polity. The results also vary by subgroup. The least support for a long-run
relationship is in Latin America and the Caribbean where three out of four democracy variables
do not appear to be cointegrated. However, three out of four democracy variables show some
support for a long-run relation in the non-Mature group.
It should be noted, however, that the sample for cointegration testing was greatly reduced
from 127 to 52 countries due to exclusion of countries that did not exhibit measurable change in
either O&G Value or the democracy variables. Non-producers, initially full democracies, and
32
unchanging autocracies had to be excluded because, mechanically, no within-country
relationship can exist when there is no change within country. This sample restriction should
increase the likelihood of finding of a long-run relationship, suggesting the cointegration
evidence should be interpreted with caution. However, as mentioned above, because we did not
find evidence of long-term effects of O&G Value on democracy, support for a long-run
relationship, or cointegration, is unnecessary anyway.

2.6 Comment on Recent Analysis
Illustrated by the quotes shown below, from Andersen and Ross (2014) (AR) and Haber and
Menaldo (2011) (HM), these two references are essential for understanding the current debate
over the existence of a political resource curse:
“The Haber–Menaldo article has had a powerful impact on the resource curse debate,
calling into question widely held beliefs about the politically malignant effects of petroleum
wealth” (AR, p. 994). Using the Haber and Menaldo data and models, AR showed that from
1800 to the 1970s there was no evidence of a resource curse (confirming the finding of HM) but
that “since the late 1970s—the period that is the focus of most other studies—oil wealth has
strongly inhibited democratization (AR, 994).”
Our primary analysis, presented above, and analysis of the valuable data prepared by HM
lead us to agree largely with HM, that there was no long-run relation between oil and Polity,
even for the post-1980 period. Specifically, we replicate the most convincing analysis performed
by AR, Table 2, p.1006, which itself first replicated the analysis of HM. AR then expand HM’s
analysis by adding an interaction term between their oil measure and a post-1980 dummy
variable.
33
The convincing rationale for the addition of the post-1980 dummy was the fact that by
1980, in most developing countries, the relationship between governments and oil companies had
changed radically due to the nationalization movement during the 1970s that was prompted by
the preceding jump in oil prices (1971-1974). In the AR analysis (see Table 2, Column 2 in their
paper) the effect of oil on the change in Polity was positive but small prior to 1980. Post-1980,
however, the effect was statistically significant and negative, because the large negative
interaction term for the post-1980 period more than offsets the small positive main effect of oil.
The argument seems intuitive, given that the early developers like the US have continued to rely
on private oil companies and Norway was already democratic long before its oil boom. By
contrast, in the countries of the Middle East and in many other developing countries, oil
production was nationalized and most of the countries remained highly autocratic.
For the AR theory to be persuasive, however, their empirical finding should be robust to
changes in the empirical specification. Yet, as we show in Table 2-4 below, in fact it is not. In
Column (1) we first replicate the AR analysis for estimating the long-run effect of oil on Polity
with the HM data set for the full period 1800-2006, including the interaction term for the post-
1980 period. The results match AR’s finding of a negative long-run effect of oil during the post
period (Long Run Effect (Post), obtained as the sum of the main effect and interaction term
divided by the negative of the coefficient on the lagged value of Polity). In Column (2), we
present the results for the corresponding specification based on the post-1980 period alone, and
after dropping the no longer necessary interaction term. In contrast to AR’s results, the main
effect of oil (Lag Total Oil Income), and hence the long-run effect, is positive and statistically
34
significant at the five percent level. This result is inconsistent with AR’s theory, suggesting that
the long-run effect of oil is not robust.
16

Two methodological issues also warrant notice. First, for there to be a long-run
relationship between Polity and oil, the two variables must be cointegrated because, as shown by,
HM they are integrated variables (HM, 14). While we agree with AR that we might not expect a
long-term relationship, or cointegration, over the whole period (AR, 1003), because of the
structural break attributable to oil nationalization, that does not preclude the necessary
cointegrating relationship for the post-1980 period alone.
Second, AR discusses that the theoretical relationship between oil and institutions is one
between countries not within countries. This is to say, countries with more oil may be expected
to develop democratic institutions at a different rate than those with less oil. The within-country
relationship, in contrast, estimates the relationship between oil and governance institutions as
they change over time within a country. While the initial research on the subject focused on the
between-country relationship (e.g. Barro 1998), an important problem was recognized with this
research. Between-country analysis omits country fixed effects and suffers from omitted variable
bias. It is an unfortunate statistical limitation that we cannot estimate between-country
relationships and at the same time also include country fixed effects.

16
Additional results are consistent with the finding reported in Column (2) of our Table 2-4. The long-run effect
of oil is not robustly negative and statistically significant even during the post-1980 period. For example, in AR’s
Table 2, Column (4), AR replicates HM’s analysis using fiscal reliance as an alternative measure to total oil income.
They again find the interaction term for the post-1980 period to be negative and significant, however, the total effect
during the post-1980 period (the addition of the main effect and interaction term) is not statistically significant. We
also replicated the AR analysis in Tables 4 and 5 using the post-1980 period alone, and similarly found the effect of
total oil income on Polity was not negative and statistically significant. These results are available on request.
35
Table 2-4 Replication of Andersen and Ross (2014) Table 2, Column 1 And Sample
Adjustment Long-Run Effects on Polity

(1) (2)
Long Run Effect (Post)
Lag Polity
Lag Total Oil Income
Lag Total Oil Inc. X Post
Δ Total Oil Income
Δ Total Oil Inc. X Post
Lag log(GPD pc)
Lag Civil War
Lag Reg. Democracy
Lag World Democracy
Δ log(GPD pc)
Δ Reg. Democracy
Δ World Democracy
DKSE DKSE Post80
-1.140* 0.752*
[-2.033] [2.642]
-0.087*** -0.157***
[-11.613] [-5.488]
0.045* 0.118*
[2.035] [2.507]
-0.144***
[-3.748]
-0.023 -0.163
[-1.008] [-1.517]
-0.340***
[-3.457]
-0.276 -2.313***
[-0.870] [-4.538]
0.063 -0.252
[0.140] [-0.363]
0.025*** 0.045*
[3.411] [2.610]
0.059* 0.657***
[2.048] [6.910]
1.322 2.013
[0.762] [0.863]
0.375*** 0.464***
[5.368] [4.538]
-0.278* 1.482***
[-2.568] [6.300]
Constant 2.552 0.000
[1.358] [.]
Total Observations 10195 3891
Number of Countries 163 163
Note: Same model as used in Andersen-Ross (2014). Dependent variable: Δ Polity. Driscoll Kray standard errors
are used to allow for cross-sectional dependence. t statistics in brackets; * p<0.05 ** p<.01 *** p<.001.
Source: See Haber and Menaldo (2011).
36
2.7 Conclusion
We find no robust long-run effect of oil abundance on democratic institutions using four
different indicators associated with democracy (Polity, Polcon, Civil Liberties, and Political
Rights), estimating our model using different methods and samples, including non-oil GDP in
our model, restricting the sample to the post-1980 period, and considering oil and gas value both
as a share of GDP and per capita. Our results are consistent with Haber and Menaldo (2011), but
contrast evidence supporting a negative relation, Andersen and Ross (2014) (AR) and Ahmadov
(2014) being the most notable. The meta-study Ahmadov (2014) finds the overall effect of oil on
democracy to be small, negative, and significant, but with considerable heterogeneity across
regions - somewhat more negative among countries of the Middle East and North Africa but
positive and highly significant in Latin America. The present study does not find robust evidence
for long-term (within country) effects of oil in Latin America, MENA, among Mature producers,
or non-Mature producers. There is some support for a negative relationship in Latin America and
the Caribbean and the Mature producers, but the results are not robust.
We expanded the empirical work done by Haber and Menaldo (2011) (HM) to include
different aspects of democracy: checks on the executive (Polcon), Civil Liberties and Political
Rights, and also to use the most updated time-series econometric methods. Like HM, the
regressions separately estimated long-run effects from short-run, but our models also allowed for
country-relationship heterogeneity, corrected for correlation across countries, and used an
alternative model that is better suited for persistent dependent variables like democracy. In
general, the results from the different models were insignificant. Moreover, the cointegration
results support only conditional evidence of a long-term relationship between O&G Value and
37
the democracy measures. We also replicated the results from AR and show that a simple change
to their model challenges the robustness of their findings.
It is important to note that, although we are unable to identify robust long-run negative
effects of O&G Value on democracy within countries on average, it may still be the case that oil
has significant negative effects in individual countries. AR also points out that the oil curse
theory is about between-country levels of oil. However it is difficult to test this theory.
Unfortunately, between-country analyses omit important characteristics including informal
institutions (for example). Country case studies, that investigate both the dynamics of the oil
industry and the development of political institutions, may be necessary to further inform this
debate.
Last, regional democratic development is statistically significant, positive, and robust.
Although not discussed in body of the paper, this result is shown by the coefficient on the cross-
sectional mean on each of the institutional variables (Appendix Tables A-2 – A-5), and
replicated in each model variant. The models are not simply underpowered or unable to
consistently identify a robust relationship. The cross-sectional means pick up the positive trend
in democratic development occurring around the world.

38
Chapter 3 Who Suffered Most From the Great Recession? Happiness in the United States

3.1 Introduction
The Great Recession (“Recession”) was the most severe recession in the United States since the
Great Depression. Annual GDP per capita growth was negative during the years 2008 and 2009
(World Bank 2015). The annual unemployment rate reached its highest levels since 1982 and
remained above 7.0 percent until 2014 (BLS 2015a). The median house price declined by 12.6
percent from 2007 – 2009 and had still not recovered by 2012 (Census 2015). Each measure
represents a significant negative shock to the American people, but how were they affected? Did
some population groups fare better than others, and can we explain why? To answer these
questions, I provide evidence from nationally representative surveys, from the General Social
Survey (GSS), of self-reported evaluations of one’s life, commonly referred to as subjective
well-being (SWB) or more simply, happiness. Data from the GSS Panel (2006-2014), which
tracks the same individuals over time, was also used to supplement the main analysis.
In economics, the well-being impacts of past business cycles have been most commonly
measured in terms of economic growth, unemployment, and inflation (e.g. the misery index), but
in more recent years there has been a growing interest in measures of SWB (Stiglitz et al. 2009).
SWB may be better placed than more traditional economic metrics for this purpose. For example,
Lucas (1987) argues that business cycles are not very important when considering their effects
on aggregate consumption. In response, Wolfers (2003) uses SWB and finds macroeconomic
volatility to have ‘moderate but important’ effects on well-being. Wolfers (2003), and others
focusing on SWB, contribute to the economics of happiness, which is a relatively new area of
research but one that is becoming increasingly important.
39
This study is the first to document the SWB impacts of the Great Recession, measured as
deviations from long-term trends, disaggregated by population group, and to provide statistical
evidence for the mechanisms affecting happiness in the United States during this period. To
understand the effects of the Recession, estimates of group-specific deviations from group-
specific trends are necessary for two reasons. First, the happiness trends are generally negative,
but not strictly. They vary especially by race and gender (Blanchflower and Oswald 2004;
Stevenson and Wolfers 2008b, 2009 and 2012; Herbst 2011). Second, different population
groups report different average happiness levels, and as different trends suggest, they are subject
to different long-term forces that may have persisted through the Recession.
In addition to the varying trends in happiness, the results show each population group
reported significant declines during the Great Recession. For the population as a whole, 2010
marks the lowest level of reported happiness in the United States since consistent measurement
began in the 1970s. The declines during 2010 vary substantially however. The foreign-born, who
were the greatest impacted, reported a decline more than three times greater than the full
population. Men were impacted more than women, young adults less than people older than 24,
and Hispanics more than non-Hispanics. Comparison with the 1980s recession shows that the
duration of the Great Recession’s well-being impacts was longer, but that the 1980s’ impact was
deeper. The 1980s’ depth is partially explained by a greater decline in women’s happiness,
however, the overall mechanisms are not yet well understood. In contrast, the declines reported
in 2010 can be statistically explained by declining income and rising unemployment. The large
decline reported by the foreign-born in 2010 is not surprising when considering they reported
declines in both income and employment that were each among the largest for the groups
studied. The conclusion that declining income best explains the declines in happiness during the
40
Great Recession is further supported by robustness checks, including panel analysis with
individual fixed effects.
The results suggest recessions have a large impact on well-being (cf. Lucas 1987), and
the mechanisms are not surprising. Declining family income affects consumption, the ability to
meet financial obligations, and has many indirect effects. Unemployment similarly has many
consequences, not only through income, but also large non-pecuniary effects. Counter-cyclical
income and employment support may be the most effective for mitigating the well-being effects
of future recessions, and policy makers may want to target certain populations.

3.2 Evidence From Past Literature
A review of the past evidence points to income and unemployment as key variables to account
for the Great Recession’s impact. Two closely related studies, Graham et al. (2010) and Deaton
(2011), each show unemployment and income measures (including stock prices) are correlated
with SWB during the Great Recession in the United States. Unemployment, short-term changes
in income, and to a lesser extent inflation have been consistently shown to be related to SWB in
a broad context (di Tella et al. 2001 and 2003; Stevenson and Wolfers 2008a; Easterlin et al.
2010; Diener et al. 2012; Dolan et al. 2008; and Winkelmann and Winkelmann 1998), and during
economic crises (Wolfers 2003; Bjørnskov 2014; Arampatzi et al. 2015). Thus the expectation is
that the Great Recession directly reduced SWB through increased unemployment and reduced
income.
What other factors might be important, and were there any that mitigated the income and
employment shocks? There is some evidence that welfare-state policies mitigated the effects.
Morgan (2015) shows that greater net income replacement rates reduced the SWB declines
41
reported by European nations during the Great Recession, and generous labor market policy
helps to reduce the negative association between SWB and unemployment (Carr and Chung
2014; Wulfgramm 2014). However, not all policies are beneficial. Morgan (2015) shows
employment protection legislation exacerbated the well-being effects of the Great Recession in
Europe, and Bjørnskov (2014) shows “that wellbeing losses during crises are substantially larger
in countries with tighter regulations of credit, labour or product markets (Bjørnskov 2014, 175).”
Concerning different population groups, young adults are expected to be affected more by
recessions (Bell and Blanchflower 2011). Better-educated people and married people are happier
than their counterparts (Dolan et al. 2008) and they may have also fared better through additional
support or better coping mechanisms. In contrast, parents are less happy in the United States
(Herbst and Ifcher 2014), and this association may have increased during the Recession through
additional income needs or concern for their children’s future.
The most closely related studies, Graham et al. (2010) and Deaton (2011), provide some
helpful insight, but comparability is limited. As mentioned, they point toward income and
unemployment as potential channels, and similar to the present study, each shows SWB
declining from early 2008 into 2009. However, they show SWB trending upward beginning in
2009, and recovery by the end of 2009 for Graham et al. (2010) and 2010 for Deaton (2011). In
contrast, SWB does not recover until 2012 in the present study. This difference can be explained
primarily by different benchmarks. They measure recovery to the early 2008 SWB levels, while I
measure recovery to long-term trend levels, and 2008 was below trend.
What is more important for comparison, the data used in Graham et al. (2010) and
Deaton (2011) have limitations. Both papers use SWB data from the Gallup Healthways Well-
Being Index, which is a daily survey beginning in 2008. The first limitation relates to the daily
42
survey, which may be overly sensitive to day-to-day events, some that may be important, and
“some that have only dubious implications for well-being (Deaton 2011. 23).” Second, the
Gallup SWB data are biased downward by the presence and placement of political questions in
the survey that also varies over the study period. Deaton (2011) implements corrections for the
political question bias, but the analysis depends on the corrections’ validity. Last, their analyses
are necessarily limited to focus on short-term relationships because the survey begins in 2008.
Free from the limitations associated with Gallup’s daily data, the present analysis is better placed
to address the effects of the Great Recession on happiness in a long-term context.

3.3 Happiness Data and Methods
The General Social Survey (NORC 2015) is the primary source of happiness data for time-trend
analysis in the U.S. In 30 waves it covers the 42-year period from 1972 – 2014. The survey
collects demographic, economic, and attitudinal information for more than 1500 people per
wave. Unlike daily surveys, the waves are fielded over a period of several months (typically
February – April). It should be noted, however, that there have been changes that could affect
time trends (i.e. sample composition), but consistent with the past literature, population weights
were applied and problematic samples dropped (e.g. 1972).
17
The GSS measures happiness as the
response to the question, “Taken all together, how would you say things are these days–would
you say that you are very happy, pretty happy, or not too happy?” This happiness question
represents one of many SWB questions. Similar to life satisfaction, it is more evaluative in
nature.

17
See Appendix B for a discussion of the population weights and which samples have been dropped.
43
The impacts of the Great Recession were estimated as group-specific deviations from
group-specific long-term trends, using individual-level happiness regressions, with repeated
cross-sectional data from the GSS. Each regression has two population groups that were selected
based on fixed characteristics.
18
Deviations in happiness were estimated for women compared to
men, African Americans compared to whites and other races, young adults ages 18-24 (also
referred to as youth) compared to older people
19
, foreign-born compared to native-born, and non-
white Hispanics compared to non-Hispanics. The regressions use dummy variables for the years
2008 and 2010 (referred to as “Recession dummies”), fixed characteristics (e.g. birth-cohort),
group indicators, a linear trend, a dummy variable for past recession years, and group
interactions with trend and the Recession dummies, to obtain group-specific trends and
deviations-from-trend. By excluding additional control variables, the Recession dummies capture
the full short-term impacts of the Great Recession and any additional effects experienced during
2008 and 2010. This model is referred to as the base model and will be built upon in subsequent
analysis. The particular estimating equations and control variables are listed in the table
footnotes (presented in ordinary least squares form “OLS” for simplicity).
Consistent with the past literature, the regressions are performed using an ordered probit
specification to account for the ordinal nature of the happiness data (e.g. Stevenson and Wolfers
2009; Ifcher and Zarghamee 2014). Unlike OLS, ordered probit regressions do not make the

18
Based on fixed characteristics, the group composition should remain the same over time. Selective migration
could still affect the group composition, especially for the foreign-born, but if we assume those affected most during
the Recession were the most likely to move, then the Recession’s impacts were understated not exaggerated.
19
Identification of the effect of being a youth during the Recession depends on the birth-cohort variables. The
youngest birth-cohort is defined as those born in 1986 or later (1986-cohort), and in 2010, the entire youth group
belongs to the 1986-cohort. As a consequence the youth variable interacted with 2010 is directly collinear with this
birth-cohort variable in 2010, and identification relies on the 1986-cohort variable’s association in alternative years
(2004, 2006, 2008, 2012, and 2014). To determine if identification for the effect of being a youth during 2008 and
2010 is a problem (in 2008 youth belong to two cohorts), I estimated the same specification for youth without
including the birth-cohort variables, and found similar results. The sample sizes and five alternative years provide
sufficient variation for consistent identification.
44
assumption that people treat the difference between “very happy” and “pretty happy” the same as
the difference between “pretty happy” and “not too happy.”
20
Ordered probit regressions
estimate the probability of each response category as a discrete ordered choice. The resulting
coefficients, however, do not apply linearly. So to ease interpretation of the results, I also
provide the marginal effects for the probabilities of responding “very happy”, which are locally
linear and can be interpreted like OLS coefficients. A marginal effect also shows the total effect
for a group (i.e. it includes the main effect and interaction term for the group of interest).
The paper focuses on explaining the declines in reported happiness in 2010. The GSS
was not conducted in 2009, and the 2008 survey was fielded prior to much of the Recession’s
effects.
21
Although the Recession officially began in December of 2007, much of the economic
decline occurred later. The collapse of Lehman Brothers, the largest bankruptcy in U.S. history,
occurred on September 15, 2008. The largest percentage decline in GDP occurred from the third
quarter 2008 to the fourth quarter 2008 (BEA 2015a). The official unemployment rate did not
exceed 7.0 percent until December 2008, where it remained until December 2013 (BLS 2015b).
For this reason, the happiness figures for 2008 are presented along with 2010, but the discussion
focuses on 2010.
It is important to note that self-reported income used throughout the paper is total family
income, from all sources, before taxes, not conditional on employment, and adjusted for inflation
and household size. Previous researchers using the GSS have also used family income (e.g.
Stevenson and Wolfers 2009; Ifcher and Zarghamee 2014), because missing values for

20
A further discussion of this approach can be found in Ferrer-i-Carbonell and Frijters (2004).
21
The 2008 General Social Survey was fielded between April and September. Although the Recession officially
began in December of 2007, by the time of fielding, the self reported economic factors had not changed
significantly. The unemployed population share had only increased slightly (2008: 3.4 percent, 2006: 3.3 percent),
and self-reported real family income, per household equivalent, had not significantly declined (2008: 33,826, 2006:
33,776). Note the GSS based unemployment information is for unemployed people as a percent of the total
population not the labor force.
45
individual income greatly exceeds those for family income (40 percent compared to 10 percent).
Analysis using personal income is discussed in the robustness section and shows the main result
does not depend on income measures.

3.4 Description of Well-Being Impacts
A smaller share of Americans report being “very happy” today as compared to the early 1970s,
and the lowest recorded share is for the year 2010. Although the Great Recession officially ended
in 2009 (NBER 2014), happiness did not recover to pre-period trends until 2012. Figure 3-1
illustrates the negative trend, the low mean score in 2010, and the subsequent recovery.

3.4.1 Statistical Significance of Declines by Population Group
The size of the declines during the Recession, and how they compare across groups, is
summarized in Table 3-1, based on the results in Table 3-2 discussed below. Each group is
statistically less likely to report being “very happy” during 2008 and 2010, and the declines are
usually statistically greater in 2010. The foreign-born were greatest impacted. With a 15.1
percentage point reduced probability of reporting “very happy” in 2010, they reported a
substantially larger decline than the full sample, which reported a corresponding decline of 4.6
percentage points. The next largest decline, 8.9 percentage points, was for Hispanics, which is
not surprising because more than one third are foreign-born. In contrast, youth (ages 18-24)
reported the smallest decline of only 2.1 percentage points, even though they were expected to be
one of most affected groups (e.g. Bell and Blanchflower 2011). Compared to their reference
groups, men reported a statistically greater decline (in 2010), and so did: blacks (2008), non-
youth (2010), foreign-born (2008 and 2010), and Hispanics (2008 and 2010).
46

The statistical significance of differences between groups is obtained from the group interactions
in the ordered-probit-regression results presented in Table 3-2. For example, the coefficient on
“2010 X Group” in column 3 provides an estimate of blacks’ experience in 2010 relative to
whites and other races, and in this instance, blacks do not experience a statistically significant
difference. The row labels for “Group” refer to the group listed in the column head (i.e. in
column 2 women are the main group while men are the omitted group). “Year 2008” should be
interpreted as the deviation from “Trend” for the omitted group. The 2010 decline for a specified
group (reported in table 3-1) is the non-linear combination of “Year 2010” and “2010 X Group.”

Figure 3-1 U.S. Happiness 1973 – 2014. Annual Proportions Reporting Very Happy, Pretty
Happy, or Not Too Happy
Source: NORC, General Social Survey (Release 2, April 17, 2015)
Notes: Happiness Scored: Very Happy = 3, Pretty Happy = 2, Not Too Happy = 1
No controls, sample weights and adjustments applied, see paper for details.
Very Happy
Pretty Happy
Not Too Happy
Mean Score (scale -->)
2 2.1 2.2 2.3 2.4 2.5
Mean Happiness (Scale 1-3)
.1 .2 .3 .4 .5 .6
1975 1985 1995 2005 2015
Population Share
47
Table 3-1 Deviations from Long-Term Trends; Marginal Effects by Group from Ordered-
Probit Regressions with Happiness as the Choice Variable. Probability of Reporting Very
Happy, 1973 – 2014.
Panel A
Full
Sample Women Men Black
White and
Other
2008 -0.020***

-0.010* -0.031*** -0.087*** -0.011***

[-5.312]

[-1.739] [-3.181] [-7.953] [-3.082]
2010 -0.046***

-0.032*** -0.062*** -0.034*** -0.048***

[-12.245]

[-5.000] [-5.799] [-2.965] [-13.208]
Trend -0.002***

-0.003*** -0.002** 0.000 -0.003***

[-3.310]

[-4.072] [-2.019] [0.471] [-3.707]

Difference in Year Effects
2008-2010 0.026***

0.022*** 0.031*** -0.053*** 0.037***

[47.215]

[18.374] [25.309] [25.970] [56.150]
Panel B
Youth
25 and
Older
Foreign-
Born
Native-
Born Hispanic nonHispanic
2008 -0.023** -0.020*** -0.068*** -0.014*** -0.076*** -0.014***

[-2.348] [-5.026] [-5.271] [-3.489] [-4.951] [-4.008]
2010 -0.021* -0.049*** -0.151*** -0.031*** -0.089*** -0.036***

[-1.834] [-13.177] [-9.810] [-7.476] [-5.113] [-8.329]
Trend -0.002** -0.002*** 0.001 -0.003*** 0.003 -0.005***

[-2.346] [-3.286] [0.536] [-3.496] [1.033] [-4.335]

Difference in Year Effects
2008-2010 -0.002 0.030*** 0.083*** 0.017*** 0.013** 0.023***

[0.378] [33.586] [21.850] [29.906] [2.178] [19.804]

The results for the foreign-born and youth may be surprising, but self-reported declines in
income and employment provide plausible explanations. Data from the GSS (reported in
Appendix Table B-3) show the foreign-born experienced both a substantial decline in income (21
percent from 2006 to 2010) and increase in unemployment (5.2 percentage points). Youth, in
Source: Author’s calculations based on data reported in text.
Notes: Marginal Effects, probability of reporting “very happy”, are estimated from corresponding
regressions in Table 2. The two other reporting categories are: “pretty happy”, and “not too happy”.
Nativity data are available beginning in 1977. Hispanic data are available beginning in 2000.
t statistics in brackets (clustered by year); * p<.10 ** p<.05 *** p<.01.
48
contrast, reported the smallest increase in unemployment from 2006 to 2010 at 1.8 percentage
points. Youth also reported a large decline in income, but as mentioned, income is measured as
total family income, and it is not clear whom youth are including in “family” income. Moreover,
Graham et al. (2010) also states young people (19-35) responded less to events during the
Recession than older people.

3.4.2 Statistical Significance of Observed Trends in Happiness
Although the focus is on the Great Recession, a few of the long-term trends warrant notice. The
first is for women, who report declining levels of happiness in both absolute terms and relative to
men. The absolute trend is shown in Table A-1 by the “Trend” marginal effect. Specifically it
means that the probability of women reporting “very happy” declined on average by 0.3
percentage points per year over the period 1973 – 2014.
22
Given the host of improvements in
objective indicators for women, it may be surprising that the decline was greater than for men
(shown by the negative and statistically significant “Trend X Group” coefficient in Table 3-2,
column 2). Comparable results and potential explanations are discussed in Stevenson and
Wolfers (2009) and Herbst (2011).
During this period, most groups report a negative trend, however that is not true for
blacks, foreign-born, and Hispanics. In the present study blacks report a positive, though
statistically insignificant trend, and past studies have shown a significant-positive trend. The
difference is likely because the present study extends the analysis from 2008 to 2014 (c.f.
Stevenson and Wolfers, 2012), and the trend has flattened out in recent years. Significant or not,
a positive trend in the U.S. is unusual. Blacks’ long-term trend has been discussed in the

22
On average women are approximately 12.3 percentage points less likely to report being “very happy” in 2014
than they were in 1973.
49
literature (Stevenson and Wolfers 2008b, and 2012), while the trends for the foreign-born and
Hispanics should be the subject of future research.
Table 3-2 Deviations from Long-Term Happiness Trends; Group Ordered-Probit
Regressions. Choice Variable Happiness, 1973 – 2014.
Group Full Sample Women Black Youth Foreign-Born Hispanic
(1) (2) (3) (4) (5) (6)
Women 0.072*** 0.150*** 0.072*** 0.072*** 0.060*** 0.023

[4.047] [4.928] [4.051] [4.050] [3.334] [0.817]
Black -0.328*** -0.328*** -0.513*** -0.329*** -0.298*** -0.225***

[-10.403] [-10.409] [-8.520] [-10.428] [-10.197] [-6.580]
Youth -0.107*** -0.106*** -0.108*** -0.105** -0.109*** -0.147**

[-3.856] [-3.804] [-3.876] [-2.413] [-3.374] [-2.388]
Group

-0.250*** -0.871***

[-4.649] [-2.802]
Past Recession -0.049** -0.049** -0.049** -0.049** -0.055* -0.077***

[-2.130] [-2.164] [-2.110] [-2.125] [-1.826] [-5.408]
Trend -0.007*** -0.005** -0.007*** -0.007*** -0.008*** -0.014***

[-3.304] [-2.014] [-3.698] [-3.284] [-3.493] [-4.283]
Year 2008 -0.055*** -0.086*** -0.029*** -0.054*** -0.039*** -0.037***

[-5.358] [-3.209] [-3.095] [-5.062] [-3.503] [-4.067]
Year 2010 -0.126*** -0.172*** -0.128*** -0.134*** -0.085*** -0.099***
[-12.491] [-5.887] [-13.432] [-13.384] [-7.535] [-8.594]
Trend X Group -0.004** 0.009*** -0.000 0.010*** 0.024***

[-2.298] [4.117] [-0.095] [4.989] [2.578]
2008 X Group

0.059 -0.256*** -0.012 -0.154*** -0.184***

[1.544] [-7.932] [-0.411] [-4.451] [-4.041]
2010 X Group

0.086** 0.016 0.074** -0.344*** -0.160***
[2.025] [0.445] [2.215] [-8.663] [-2.911]
Observations 32945 32945 32945 32945 28803 11371
R2 (psuedo) 0.010 0.010 0.010 0.010 0.009 0.008

Source: Author’s calculations based on data reported in text.
Notes: Omitted groups are: men, white and other races, ages 25 and older, native-born, and non-
Hispanics. Additional control variables include age, age squared, ten-year birth-cohort, and mother's
and father's education. The estimated regression (specified in OLS) is: ℎ𝑎𝑝𝑝𝑦
!"#
= 𝛼
!
+ 𝜷
!
𝒙
!"
+
𝛿𝑝𝑟𝑒𝑐𝑒𝑠
!
+ 𝝀
!
′𝒕𝒊𝒎𝒆
!
+ 𝝀
!
′𝒕𝒊𝒎𝒆
!
𝑋 𝑔𝑟𝑜𝑢𝑝
!
+ 𝜀
!"#
. ℎ𝑎𝑝𝑝𝑦
!"#
is reported happiness for individual i
belonging to one of two groups g in year t; 𝒙
!"
is a vector of individual characteristics; 𝑝𝑟𝑒𝑐𝑒𝑠
!
is a
dummy variable for past recessions; 𝒕𝒊𝒎𝒆
𝒕
′ is the vector (1 𝑡𝑟𝑒𝑛𝑑
!
𝑑
!"
𝑑
!"
), where 𝑡𝑟𝑒𝑛𝑑
!
= 𝑦𝑒𝑎𝑟
!
−
1972 and 𝑑
!
are dummy variables for the years 2008 and 2010; 𝑔𝑟𝑜𝑢𝑝
!
is a dummy variable for the
demographic group of interest. The coefficients of vector 𝝀
!
are the main effects common to all groups,
and the marginal effects are obtained from the nonlinear combination of the main effect and the
interaction coefficient (i.e. 𝝀
𝟎
and 𝝀
𝟏
). Nativity data are available beginning in 1977. Hispanic data are
available beginning in 2000.
t statistics in brackets (clustered by year); * p<.10 ** p<.05 *** p<.01.
50

3.5 Great Recession Channels – Explaining the Impacts
Declines in income and employment provide plausible explanations for the declines in happiness
reported during the Great Recession, but were other factors important? Did GDP per capita or the
aggregate unemployment rate affect happiness beyond their direct effects on individual income
and employment? Were other individual characteristics important? What about housing prices?
The following sections identify the plausible channels through which the Great Recession
operated, and the statistical methods to obtain the results.

3.5.1 Methods and Variables to Identify Plausible Channels
To identify plausible channels, regional and quarter-of-interview controls, personal
characteristics, macro variables, and interactions with certain micro controls are sequentially
added to the base model. As mentioned, the base model incudes fixed-individual characteristics,
a dummy for past recessions, a linear trend, group indicators, Recession dummies, and
interactions to obtain group-specific deviations from long-term trends. The additional control
variables include traditional micro characteristics that affect happiness (Dolan et al. 2008) and
certain macro-economic variables. In particular the macro pathways include: log GDP per capita
and lagged log GDP per capita, the unemployment rate, log median house price, the inflation
rate, income inequality (Gini coefficient), and government assistance (social expenditures).
Lagged log GDP per capita is included because GDP per capita and GDP per capita growth have
both been shown to be important variables in the literature, and adding both log GDP per capita
and its lag is statistically more flexible than GDP per capita or GDP per capita growth separately.
In this context, the unemployment rate could be interpreted as affecting feelings of job security,
51
because controls for individual employment status are also included. Income inequality could be
interpreted as affecting trust and feelings of fairness (Oishi et al. 2011). GDP and the
unemployment rate were measured at the census division level, the median house price at the
census region, and the others at the country level. The specific variables and their sources are
detailed in the Appendix Table B-4.
When adding control variables, if the statistical significance of a Recession dummy is
reduced, then the added variable helps account for the previously unidentified effects associated
with the Recession years. In the next step, key micro-control variables are interacted with the
Recession dummies. Interactions are important because they allow for the relationships of the
interacted variables to change during the Great Recession. The relationships could change
because people’s preferences change, the economic and social context changed, and because the
source of variation is likely due to the Great Recession. With interactions the original Recession
dummies (main effects) capture only the remaining variation during that year that is not
associated with that channel.
The sample has been restricted to people reporting family income, employment status,
and each of the micro characteristics of interest. Nativity in particular affects the sample because
it was not added to the GSS until 1977. The Gini coefficient also limits the period to 2012
because it was not available for 2014 at time of writing. The analysis based on Hispanic origin is
restricted further to the period beginning in 2000, because Hispanic origin was not available
before. The base model used to describe the initial declines is an exception. It uses the longest
period available, from 1973 – 2014, for each group except those based on nativity and Hispanic
origin.
23

23
If sample period is a concern, there are two models that retain the same main conclusions with adjusted sample
periods. First, the base model with added location and quarter of interview controls, uses the period 1977-2012, and
52
As with the descriptive analysis the regressions are conducted using an ordered probit
specification, and the particular estimating equations are listed in the table footnotes (presented
in OLS form for simplicity). In what follows the analysis is first performed for the population as
a whole, including robustness checks. Then group specific deviations are estimated as outlined
earlier.

3.5.2 Plausible Channels – Average Relationship For Full Population
The primary mechanisms affecting happiness during the Great Recession are income and
unemployment. The results are based on Tables 3-3 and 3-4. Table 3-3 presents the first set of
results with sequentially added controls, and Table 3-4 further adds micro-control Recession
interactions. Adding micro controls, especially income, labeled ln(Eqv. Inc.), and employment
status, reduces the decline reported by the population as a whole, and is enough to reduce the
significance of past recessions, though not the Great Recession. The interactions with the
Recession dummies are necessary to statistically account for the Great Recession’s effects –
discussed below with Table 3-4. The common trend can be statistically accounted for by adding
marital status. This result makes sense as marriage is positively associated with happiness, and
the married-population share declined over the period from approximately seventy to fifty
percent (shown in the Appendix Table B-1).
In general, the coefficients in Table 3-3 are in the expected direction and statistically
significant. Women are happier; blacks less happy; higher education is positively associated with
happiness; and republicans, religious people, married couples, and non-parents are all happier.

shows similar deviations during 2008 and 2010 (in Table 3-3). Second, the base model with added controls, but
excluding nativity and the Gini coefficient, to retain the period 1973-2014, provides similar explanations (discussed
in the robustness section).
53
Table 3-3 Ordered-Probit Regressions for Full Sample with added Micro and Macro
Controls, 1977 – 2012.
(1) (2) (3) (4) (5) (6) (7) (8)
Year 2008 -0.055*** -0.114*** -0.109*** -0.126*** -0.123*** -0.125*** -0.170*** -0.148***

[-5.045] [-5.913] [-5.707] [-6.838] [-6.562] [-7.556] [-5.575] [-6.138]
Year 2010 -0.126*** -0.150*** -0.143*** -0.147*** -0.145*** -0.116*** -0.076** -0.066**

[-11.650] [-9.098] [-8.736] [-8.536] [-8.350] [-7.407] [-2.155] [-2.322]
Trend -0.007*** -0.011*** -0.011*** -0.007** -0.007** -0.006* -0.002 -0.000

[-3.120] [-3.432] [-3.365] [-2.162] [-2.107] [-1.867] [-0.418] [-0.013]
Past Recession -0.054* -0.068** -0.065** -0.055 -0.055 -0.047 -0.039* -0.022

[-1.818] [-2.100] [-2.143] [-1.551] [-1.548] [-1.393] [-1.902] [-0.956]
Women 0.060*** 0.044** 0.050*** 0.078*** 0.082*** 0.094*** 0.093*** 0.093***

[3.314] [2.514] [2.795] [3.582] [3.727] [4.527] [4.505] [4.507]
Black -0.299*** -0.303*** -0.236*** -0.164*** -0.153*** -0.126*** -0.127*** -0.128***

[-10.151] [-8.929] [-6.403] [-4.089] [-3.874] [-3.243] [-3.247] [-3.245]
Youth -0.110*** -0.084** -0.043 0.047 0.041 0.048 0.046 0.046

[-3.409] [-2.286] [-1.259] [1.202] [1.065] [1.204] [1.155] [1.160]
Foreign -0.024 -0.025 -0.046 -0.076* -0.075* -0.064 -0.063 -0.063

[-0.647] [-0.631] [-1.143] [-1.872] [-1.846] [-1.593] [-1.579] [-1.576]
High School and Less

-0.086*** -0.094*** -0.090*** -0.065** -0.064** -0.065**

[-2.588] [-2.872] [-2.797] [-2.029] [-2.018] [-2.015]
Bach. Deg. +

0.131*** 0.109*** 0.104*** 0.072** 0.073** 0.073**

[3.537] [2.984] [2.866] [1.972] [2.026] [2.019]
Republican

0.037*** 0.031*** 0.031*** 0.028*** 0.028*** 0.028***

[6.686] [5.554] [5.643] [5.160] [5.150] [5.149]
Religious

0.205*** 0.157*** 0.162*** 0.156*** 0.156*** 0.156***

[5.797] [4.454] [4.578] [4.472] [4.487] [4.484]
Never Married

-0.424*** -0.475*** -0.416*** -0.418*** -0.417***

[-7.917] [-8.776] [-7.725] [-7.840] [-7.807]
Unmarried

-0.592*** -0.597*** -0.554*** -0.557*** -0.556***

[-27.883] [-28.218] [-26.304] [-26.444] [-26.456]
Parent

-0.090*** -0.059*** -0.059*** -0.059***

[-4.480] [-3.151] [-3.174] [-3.141]
Empl Part

-0.061** -0.060** -0.060**

[-2.157] [-2.121] [-2.124]
Unemployed

-0.340*** -0.333*** -0.334***

[-5.590] [-5.374] [-5.374]
Out Workforce

-0.009 -0.009 -0.009

[-0.273] [-0.289] [-0.291]
ln(Eqv. Inc.)

0.124*** 0.122*** 0.123***

[11.783] [11.524] [11.672]
ln(GDPpc) -0.277

[-0.680]
Lag ln(GDPpc)

0.472

[1.464]
Unemp. Rate

-0.018** -0.021***

[-2.115] [-3.421]
Gini

-2.380** -2.524***

[-2.539] [-2.691]
ln(House Price)

-0.094

[-0.978]
Inflation

0.003

[0.632]
Location and Quarter Controls No Yes Yes Yes Yes Yes Yes Yes
Observations 28803 24161 24161 24161 24161 24161 24161 24161
R2 (psuedo) 0.008 0.010 0.017 0.039 0.039 0.045 0.045 0.045

Source: Author’s calculations based on data reported in text.
Notes: Additional control variables include age, age squared, ten-year birth-cohort, and mother's and
father's education. Location and quarter controls are dummies for census division, rural location, and
quarter of interview. The estimated regression (specified in OLS) is: ℎ𝑎𝑝𝑝𝑦
!"
= 𝛼
!
+ 𝜷
!
𝒙
!"
+𝜸
!
𝒚
!
+
𝛿𝑝𝑟𝑒𝑐𝑒𝑠
!
+ 𝜆
!
𝑡𝑟𝑒𝑛𝑑
!
+𝜆
!
𝑑
!"
+ 𝜆
!
𝑑
!"
+ 𝜀
!"
. ℎ𝑎𝑝𝑝𝑦
!"
is reported happiness for individual i in year t;
𝒙
!"
is a vector of individual characteristics; 𝒚
!
is a vector of macro economic variables, 𝑝𝑟𝑒𝑐𝑒𝑠
!
is a
dummy variable for past recessions; 𝑡𝑟𝑒𝑛𝑑
!
= 𝑦𝑒𝑎𝑟
!
−1972 and 𝑑
!
are dummy variables for the years
2008 and 2010. For specification 1, the full sample is used (1973 – 2014). For specifications 2 – 8, the
sample is restricted to the years 1977 – 2012, because nativity data are available beginning in 1977, and
the Gini coefficient is not available for 2014.
t statistics in brackets (clustered by year); * p<.10 ** p<.05 *** p<.01.
54
The happiness association with being foreign-born or a young adult depends on other covariates.
At the macro level, income inequality and the unemployment rate play the largest role.
Remember income inequality could be interpreted as affecting trust and feelings of fairness
(Oishi et al. 2011), and the unemployment rate can be interpreted as affecting feelings of job
security when individual employment status is also controlled. In contrast to what one might
expect, housing prices at the census division level do not help explain the declines. Social
expenditures, GDP per capita, and inflation were also dropped because they are not statistically
important. Social expenditures were not presented because they reduced comparability across
columns (they are only available beginning in 1980).
24

Table 3-4 presents the more important results. The happiness declines from long-term
trends are accounted for by the added micro-control interactions, which as explained below, is
shown by the Year 2010 dummy (Panel A) being reduced in magnitude and significance.
Specifically, full-time employed people are not statistically less happy than trend in 2010, and
people report the trend level of happiness when excluding the effects of income.
25

The first column of Table 3-4 presents the results from the base model with location and
quarter-of-interview controls added. The subsequent columns include the macro and micro
controls from Table 3-3, column 8, and add interactions with key micro variables. In column 2,
employment status is interacted with the Recession dummies. Because the omitted category is
employed full-time, the Recession dummies capture the effect of being employed full-time
during the years 2008 and 2010. Thus, the insignificant Year 2010 dummy (column 2) means

24
Social expenditures may still effect the transmission of the Recession’s effects, but self-reported income
includes government transfers. For this reason we cannot identify the full effects of social expenditures while
income is controlled.
25
Like all regression results, this result is conditional on the other controls included in the regression. The
robustness section below discusses the effects of employment and income from the base model without additional
micro and macro controls.
55

Table 3-4 Deviations from Long-Term Happiness Trends; Ordered-Probit Regressions for
U.S. Sample with added Controls and Interactions, 1977 – 2012.
Panel A (1) (2) (3) (4) (5) (6) (7) (8)
Past Recession -0.068** -0.022 -0.022 -0.022 -0.021 -0.020 -0.021 -0.021

[-2.100] [-0.935] [-0.957] [-0.937] [-0.930] [-0.880] [-0.929] [-0.896]
Trend -0.011*** -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000

[-3.432] [-0.008] [-0.015] [-0.003] [-0.004] [-0.074] [-0.006] [-0.048]
Year 2008 -0.114*** -0.163*** -0.789*** -0.854*** -0.722*** -0.578*** -0.716*** -0.354**

[-5.913] [-5.872] [-6.385] [-6.713] [-5.714] [-4.335] [-5.614] [-2.559]
Year 2010 -0.150*** -0.043 -0.023 0.061 0.155 0.230 0.149 0.273*
[-9.098] [-1.228] [-0.172] [0.437] [1.138] [1.599] [1.085] [1.873]
Empl Part X 2008 0.073** 0.096*** 0.096*** 0.110*** 0.095*** 0.113***

[2.304]

[3.033] [3.072] [3.509] [3.037] [3.600]
Unemployed X 2008

-0.065

-0.020 -0.021 0.011 -0.021 0.029

[-1.031]

[-0.311] [-0.321] [0.184] [-0.326] [0.471]
Out Workforce X 2008

0.027

0.047 0.049 0.043 0.049 0.052

[0.841]

[1.427] [1.476] [1.286] [1.495] [1.592]
ln(Eqv. Inc.) X 2008

0.063*** 0.067*** 0.057*** 0.046*** 0.057*** 0.038***
[5.617] [5.938] [5.130] [3.968] [5.135] [3.227]
Empl Part X 2010 -0.027 -0.034 -0.033 -0.035 -0.033 -0.034

[-0.840]

[-1.096] [-1.072] [-1.122] [-1.083] [-1.086]
Unemployed X 2010

0.233***

0.227*** 0.236*** 0.238*** 0.237*** 0.236***

[3.622]

[3.466] [3.578] [3.662] [3.600] [3.619]
Out Workforce X 2010

-0.104***

-0.110*** -0.106*** -0.111*** -0.107*** -0.109***

[-3.342]

[-3.530] [-3.351] [-3.517] [-3.406] [-3.494]
ln(Eqv. Inc.) X 2010

-0.004 -0.010 -0.017 -0.022* -0.017 -0.023*
[-0.370] [-0.851] [-1.437] [-1.832] [-1.491] [-1.929]
High School X 2008 -0.057** -0.053** -0.057** -0.045*

[-2.421] [-2.326] [-2.379] [-1.908]
High School X 2010

-0.053** -0.055** -0.053** -0.052**
[-2.170] [-2.314] [-2.149] [-2.122]
Nev.Mar. X 2008 -0.160*** -0.270***

[-3.654]

[-6.038]
Nev.Mar. X 2010

-0.092*

-0.114**

[-1.914]

[-2.333]
UnMar. X 2008

0.013

0.005

[0.514]

[0.183]
UnMar. X 2010

0.030

0.027

[1.100]

[0.988]
Parent X 2008 -0.009 -0.163***

[-0.343] [-7.752]
Parent X 2010

0.018 -0.038*
[0.632] [-1.763]

56
Table 3-4 Continued

Panel B (1) (2) (3) (4) (5) (6) (7) (8)
High School and Less

-0.065** -0.065** -0.064** -0.058* -0.058* -0.058* -0.058*

[-2.021] [-2.014] [-2.014] [-1.727] [-1.720] [-1.719] [-1.717]
Bach. Deg. +

0.074** 0.073** 0.074** 0.074** 0.074** 0.074** 0.075**

[2.035] [2.003] [2.019] [2.033] [2.027] [2.035] [2.056]
Never Married

-0.416*** -0.417*** -0.416*** -0.416*** -0.402*** -0.416*** -0.396***

[-7.804] [-7.808] [-7.806] [-7.803] [-7.212] [-7.812] [-7.053]
Unmarried

-0.556*** -0.556*** -0.555*** -0.555*** -0.558*** -0.555*** -0.558***

[-26.424] [-26.378] [-26.347] [-26.286] [-23.976] [-26.278] [-23.893]
Parent

-0.058*** -0.058*** -0.058*** -0.058*** -0.057*** -0.059*** -0.047**

[-3.111] [-3.143] [-3.112] [-3.115] [-3.089] [-2.950] [-2.444]
Empl Part -0.063** -0.061** -0.064** -0.064** -0.065** -0.064** -0.065**

[-2.041] [-2.146] [-2.087] [-2.094] [-2.120] [-2.094] [-2.132]
Unemployed

-0.357*** -0.334*** -0.358*** -0.359*** -0.360*** -0.359*** -0.360***

[-5.339] [-5.372] [-5.362] [-5.370] [-5.409] [-5.372] [-5.408]
Out Workforce

-0.005 -0.010 -0.006 -0.007 -0.006 -0.007 -0.006

[-0.141] [-0.323] [-0.190] [-0.202] [-0.169] [-0.200] [-0.182]
ln(Eqv. Inc.)

0.123*** 0.120*** 0.120*** 0.121*** 0.122*** 0.121*** 0.123***
[11.705] [10.504] [10.477] [10.506] [10.525] [10.491] [10.530]
Macro Controls No Yes Yes Yes Yes Yes Yes Yes
Observations 24161 24161 24161 24161 24161 24161 24161 24161
R2 (psuedo) 0.010 0.045 0.045 0.045 0.045 0.046 0.045 0.046

full-time employed people did not report a decline from long-term trends in 2010. In column 3,
income is interacted with the Recession dummies, and the 2010 dummy indicates that after
accounting for income changes, people are not statistically less happy than trend levels.
Remember the income measure is adjusted family income, from all sources, and not conditional
on employment.
Column 4 indicates that people with high school or less education are less happy than
those with more education (Panel B), and education is more important during the Recession (see
Source: Author’s calculations based on data reported in text.
Notes: Additional control variables include age, age squared, ten-year birth-cohort, mother's and father's
education, census division, rural location, quarter of interview, woman, black, youth, foreign born,
republican, and religious. Macro controls include the unemployment rate, and Gini coefficient. The
estimated regression (specified in OLS) is: ℎ𝑎𝑝𝑝𝑦
!"
= 𝛼
!
+ 𝜷
!
𝒙
!"
+𝜸
!
𝒚
!
+𝛿𝑝𝑟𝑒𝑐𝑒𝑠
!
+ 𝜆
!
𝑡𝑟𝑒𝑛𝑑
!
+
𝜆
!
𝑑
!"
+ 𝜆
!
𝑑
!"
+𝝀
𝟑
′𝒄
!"
𝑑
!"
+𝝀
𝟒
′𝒄
!"
𝑑
!"
+ 𝜀
!"
. ℎ𝑎𝑝𝑝𝑦
!"
is reported happiness for individual i in year t;
𝒙
!"
is a vector of individual characteristics; 𝒚
!
is a vector of macro economic variables, 𝑝𝑟𝑒𝑐𝑒𝑠
!
is a
dummy variable for past recessions; 𝑡𝑟𝑒𝑛𝑑
!
= 𝑦𝑒𝑎𝑟
!
−1972 and 𝑑
!
are dummy variables for the years
2008 and 2010. 𝒄
!"
is a vector of individual variables or channels that may explain the Great Recession.
They are also included in 𝒙
!"
. The sample is restricted to the years 1977 – 2012, because nativity data
are available beginning in 1977, and the Gini coefficient is not available for 2014.
t statistics in brackets (clustered by year); * p<.10 ** p<.05 *** p<.01.
57
the negative coefficient on the high-school Recession interactions in Panel A). Lower educated
people may be more vulnerable to the effects of the Recession or have inferior support systems.
Moving across the columns, married people are happier on average (Panel B), and never married
people are even worse off during the Recession (column 6, Panel A). Marriage could mitigate the
negative effects of the Recession, but unmarried people (separated, divorced, and widowed)
people were not differentially affected during the Recession. Column 7 shows parents (married
and unmarried) were also not differentially affected during the Great Recession (insignificant
parent-Recession interactions), but when controlling for marriage during the Recession (column
8), parents do report a larger negative relationship with SWB. As a reminder, the coefficient on
2010 is for the omitted category with continuous controls accounted for separately, which means
the positive and significant coefficient on 2010, in column 8, shows that married people, who are
full-time employed, have no kids, have more than a high school education, and excluding income
effects, showed an increase in happiness during 2010 (at 10 percent significance).
It is interesting to note that the declines in happiness observed in 2008 are not well
explained. As discussed earlier, the survey in 2008 preceded much of the economic decline, and
it is likely for this reason that the decline in 2008 cannot be explained by economic factors. Note
too, however, that education, marital status, and parental status also fail to explain the impacts in
2008. It is possible that Americans perceived uncertainty in anticipation of the economic declines
and that reduced their happiness.

Robustness Checks—Additional results emphasize the importance of income during the Great
Recession. In the previous analysis, the deviations were measured from a linear trend, and the
decline during 2010 was statistically explained with micro, macro, and micro interactions.
58
However, the models may face problems with endogeneity associated with behaviorally chosen
variables, and it is possible that the long-term trends are non-linear. For these reasons, two
robustness checks were used.
The first check uses regressions that separately add the income and employment status
interactions to the base model. This reduces endogeneity concerns because the variables resulting
from behavioral choice are excluded, and the main effects of income and employment should
capture any endogenous relationship that is not specific to 2008 or 2010. As an added benefit, the
full period (1973 – 2014) is retained when excluding nativity and the Gini coefficient from the
regressions. The second check uses a cubic trend in place of the linear trend.
The 2010 marginal effects, or probability of reporting “very happy” are reported in
Appendix Table B-5. Remember the marginal effects are associated with the 2010 dummy or
main effect excluding the interaction terms. Without additional controls, the interactions between
income and the Recession dummies are sufficient to account for the decline in happiness
reported in 2010 (shown in column 3), and this result does not depend on a linear trend (column
5). In contrast, full-time employed people are statistically less happy in 2010 (column 2).
Without controls, they report a smaller decline in 2010 than the average person, but the decline is
still statistically significant. Reduced income is the most important channel affecting happiness
during the Great Recession, and this result holds under multiple scenarios. However, it is
important to remember that income and unemployment are not independent of each other.
Changes in adjusted family income may result from changes in: personal wages, family member
wages, household composition, government transfers, or have been caused by unemployment or
underemployment.
59
As mentioned, adjusted family income was relied upon because personal income data
was more likely to be missing (40 percent compared to 10 percent). However, it may be expected
that the happiness-income relation depends on the source of income. To determine if the income
measure drives the key results, an additional robustness test was used. In column 6 of Appendix
Table B-5, real personal income and its interaction with the Recession dummies was added to the
base model. Results for the comparable analysis using adjusted family income are presented in
column 3. Comparing the two estimates, the results are visibly different, but neither is
statistically significant. Like adjusted family income, reduced personal income in 2010 can
account for the average reported decline in happiness in 2010.

Happiness Changes by Individual.—The interpretation of the long-term analysis is limited to
comparisons of different people. To measure the effects of variable changes over time for a given
person, longitudinal or panel data are necessary. Using the relatively new GSS Panel data
(covering the period 2006-2014)
26
I further tested the mechanisms affecting happiness during the
Great Recession using a fixed-effects logit specification. The main conclusion is the same.
Declining income statistically explains the happiness declines in 2010. However, this result may
be considered more robust, because individual fixed effects capture omitted time-invariant
factors.
The panel analysis is similar to the long-term but differs in a few important aspects. Year
dummies are added for each year in the sample excluding 2010, making it the reference period,
not long-term trends. Then similar to the robustness checks, income and employment status are

26
In 2006 the General Social Survey added a longitudinal component that tracks the same people over time, and
there are now three separate overlapping panels, each with three waves, that collectively cover the years 2006-2014.
The first panel was fielded in 2006, 2008 and 2010, the second panel 2008, 2010, and 2012, while the last panel (to-
date) was fielded in 2010, 2012, and 2014.
60
separately added to see if they can explain the year effects. 2010 was used as the reference period
because it had the lowest level of happiness and was the only year that each GSS panel was
fielded (see footnote 26). Fixed-effects logit specifications are used with the binary variable
“very happy” because ordered probit models are not possible with fixed effects (Cameron and
Trivedi 2005, 796).
27

Appendix Table B-6 presents the results. Column 1 includes only the year dummies
(fixed effects are also included with the model). Compared with 2010, each year is positively
associated with the probability of reporting “very happy”. Note the estimates do not have a linear
interpretation, but consistent with previous findings they are increasing away from 2010 (i.e.
people are happiest in 2006 and 2014). Column 2 adds controls for income and employment
status. Each year is still statistically significant and positive. Accounting for the period-average
income and employment relationships are insufficient to account for the 2010 decline in
happiness.
Column 3 excludes the income control, but interacts employment status with each year.
The results show that a full-time employed person (the omitted category) or someone with no
change in employment status
28
reports comparable happiness in 2008 and 2010 (2008 is not
statistically different from 2010). However, they are happier in the other years when compared to
2010.
Column 4 presents the main effects for each year when income-year interactions are used.
Excluding the effects of income, individuals are not statistically more likely to report “very
happy” during 2006, 2008, or 2012. They are equally happy in 2010, which is consistent with the

27
There are ordered logit estimation techniques that allow fixed effects (e.g. Ferrer-i-Carbonell and Frijters, 2004)
but the binary-response logit model is consistent and simpler to implement.
28
The effect of not changing employment status is treated the same as being in the reference group because fixed-
effects models estimate the effects of changes in independent variables. Also, anyone that did not report a change in
“very happy” over the period is dropped from the regression.
61
long-term analysis. However, the 2014 main effect is statistically significant, which indicates
people are happier in 2014 than in 2010, even when excluding the effects of differences in
income.

3.5.3 Channels by Population Group
The explanations of the Great Recession’s effects for various population groups are similar to
those for the population as a whole. Declining income statistically accounts for the declines in
happiness reported by each group in 2010, with one exception. Rising unemployment is also
important.
A summary of the results can be illustrated with the reported declines before and after
accounting for the plausible channels. In Figure 3-2 the darker bars correspond to the 2010
declines in the probability of reporting “very happy”. The first estimates are from the base model
and repeated from Table 3-1. Note that as before, the deviations are negative and statistically
significant at five percent for each population group except youth (18-24). The lighter grey bars
are for the base model, but with additional channels controlled. Specifically, the light-grey
deviations show the “effects” of 2010 excluding income’s association with happiness. Notice the
confidence intervals greatly increase, and for men and blacks, what were statistically significant
and negative deviations are now positive and significant. The 2010-reported decline in happiness
is accounted for with micro and macro controls, and income-interactions. Women are, however,
an exception when using income interactions. The supporting estimates are presented in Table 3-
5 and discussed below, along with alternative models.
62

Table 3-5 provides a summary of the 2010 change in the likelihood a population group will
report being “very happy”. The estimates are marginal effects from ordered probit regressions for
the 2010 dummy or main effect. The first row repeats the declines reported in Table 3-1.
Subsequent rows show how the declines change as controls are added to the base model. For the
full sample, the rows correspond to the columns in Tables 3-3 and 3-4 (specific columns are
listed in the footnotes). Compared to the base model, the model for the row labeled “Micro
Controls” adds dummies (without interactions) for: census division, rural location, quarter of
interview, education level, republican, religious, marital status, parent, employment status, and

Figure 3-2 2010 Deviation from Long-Term Trend – Change in Probability of Reporting
“Very Happy”, Two Models by Population Group
Source: Author calculations based on data and analysis reported in text.
Notes: The value of nearly -5 for “Full” (Base) indicates that the full sample was approximately 5 percentage points
less likely to report being “very happy” in the year 2010, when compared to long-term trends. “Base” corresponds
to the estimates reported in Table 3-1, based on the regressions in Table 3-2. “Inc. Int.” shows the deviations
excluding the effects of declining income. The figures correspond to the estimates reported in Table 3-5, Income
Int. row, which for the full sample, are in turn based on Table 3-4, column 3. Error bars represent 95 percent
confidence intervals. They have been truncated when extending beyond -17 or 17 to reduce the size of the figure.
−15 −10 −5 0 5 10 15
Deviation in Very Happy (percentage points)
Full
Women
Men
Black
White & Other
Youth
25 & Older
Foreign
Native
Hispanic
non −Hispanic
Base
Income Interaction
63
income. The macro controls include the unemployment rate, and Gini coefficient. The
statistically insignificant marginal effect for Hispanics in row 2 means the micro controls are
sufficient to account for their 2010 decline in reported happiness.
29

The rows beginning with “Empl. Status Int.” add Recession year interactions with: 1)
employment status (employed full-time is omitted); 2) income; 3) employment status and
income; and 4) high school education (less than high school omitted), employment status, and
income. The row “Income Int.” presents the estimates associated with figure 2. With the
interactions, the reported marginal effects represent the deviations for the omitted category or
excluding the effects of income. The last row, for example, is based on the model with micro and
macro controls, and interactions with employment status, income, and high school or less
education (Table 3-4, column 5), and the marginal effect is for people employed full-time, with
more than a high school education, and excluding the association with income.
Moving down the rows in Table 3-5, column 1, the reduced magnitude and significance
of the 2010 “effects” shows that the unexplained decline in happiness reported in 2010 can be
accounted for with micro and macro controls (shown by the reduced magnitude), but requires
income or unemployment interactions to completely account for the decline (reduced average
decline reported by several groups, specifically: whites and other races, youth, and those older
than 24. For other groups, reduced feelings of job security and increasing income inequality
(macro controls) are sufficient (women, blacks, native-born, and non-Hispanics). In stark
contrast, women do not report a decline in 2010 with unemployment interactions, but they do

29
Remember the analysis for Hispanic and non-Hispanic groups only covers the period 2000 – 2012 based on
availability of the Hispanic variable and Gini coefficient. It is likely due to the shorter period that Hispanics only
require controls for the period-average relationship between income and unemployment to statistically account for
the declines in 2010.
64
when controlling for income interactions – even full-time employed women. This result is
unexpected and should be explored further in future analysis.

Table 3-5 2010 Deviation from Long-Term Trend – Change in Probability of Reporting
“Very Happy” by Population Group with Specified Controls
Panel A (1) (2) (3) (4) (5) (6)

Full
Sample Women Men Black
White and
Other
Base -0.046***

-0.032*** -0.062*** -0.034*** -0.048***

[-12.245]

[-5.000] [-5.799] [-2.965] [-13.208]
Micro Controls -0.040***

-0.020*** -0.063*** -0.030** -0.042***

[-7.318]

[-3.123] [-6.931] [-2.512] [-6.357]
Macro Controls -0.023**

-0.003 -0.046*** -0.016 -0.024**

[-2.315]

[-0.280] [-4.144] [-1.235] [-2.205]
Empl. Status Int. -0.015

0.008 -0.038*** 0.016 -0.018

[-1.224]

[0.556] [-2.749] [1.054] [-1.368]
Income Int. -0.003

-0.065** 0.100*** 0.139*** -0.018

[-0.167]

[-2.181] [8.402] [5.968] [-0.997]
Empl. & Inc. Int. 0.007

-0.059** 0.146*** 0.314*** -0.015

[0.466]

[-2.061] [8.880] [6.129] [-0.834]
High School, Empl. & 0.019

-0.061* 0.150*** 0.324*** -0.009
Inc. Int. [1.363]

[-1.865] [8.775] [4.128] [-0.497]

Panel B (1) (2) (3) (4) (5) (6)
Youth
25 and
Older
Foreign-
Born
Native-
Born Hispanic nonHispanic
Base -0.021* -0.049*** -0.151*** -0.031*** -0.089*** -0.036***

[-1.834] [-13.177] [-9.810] [-7.476] [-5.113] [-8.329]
Micro Controls -0.044*** -0.041*** -0.127*** -0.029*** -0.012 -0.035***

[-2.774] [-5.706] [-4.935] [-4.594] [-1.448] [-5.579]
Macro Controls -0.032* -0.022** -0.109*** -0.010 -0.028 -0.051

[-1.652] [-2.133] [-4.803] [-0.861] [-0.452] [-0.832]
Empl. Status Int. -0.029 -0.014 -0.057** -0.009 0.004 -0.062

[-1.123] [-1.160] [-2.321] [-0.652] [0.056] [-0.916]
Income Int. -0.014 0.002 -0.161 0.004 -0.021 0.014

[-0.153] [0.275] [-0.893] [0.317] [-0.428] [0.940]
Empl. & Inc. Int. 0.022 0.006 -0.017 0.005 -0.022 0.005

[0.279] [0.736] [-0.122] [0.323] [-0.392] [0.159]
High School, Empl. & na 0.019*** 0.143* 0.007 na 0.010
Inc. Int.

[2.861] [1.712] [0.519]

[0.334]

Source: Author’s calculations based on data reported in text.
Notes: The first row of each panel corresponds with Table 1. The subsequent rows correspond as
follows: second row - Table 3-3, column 6; third - Table 3-3, column 8; fourth - Table 3-4, column 2;
fifth – Table 3-4, column 3; sixth – Table 3-4, column 4; seventh – Table 3-4, column 5. Results were
omitted if the bin size was too small (e.g. there was less than 30 youth in the sample who were full-time
employed and also had a high school or less education in 2010). For the same reason specifications
including interactions for parents and marital status were excluded.
t statistics in brackets (clustered by year); * p<.10 ** p<.05 *** p<.01.
65
Among some of the other interesting results, remember the foreign-born showed the
largest decline in the likelihood of reporting “very happy” during 2010 (15.1 percentage points).
Table 3-5, panel B, column 3, shows that even the full-time-employed foreign-born reported a
10.9 percentage point decline, which is substantial because the next largest decline, excluding for
Hispanics,
30
was for men at only 6.2 percentage points (Table 3-1, panel A). The 10.9-point
decline for the full-time-employed foreign-born is 75 percent greater than the average for men
(employed and unemployed). However, adding the income interaction is sufficient to account
for the foreign-born’s reported decline in happiness (Table 3-5, panel B, column 3, the decline is
large but no longer statistically significant). Income interactions are important for other groups
too, especially men and blacks who become statistically happier in 2010 when excluding the
effects of income.
The figures by population group are based on the same analysis that was applied to the
full sample, but with added group interactions. Full regression results that form the basis for
Table 3-5 (analogous to Tables 3-3 and 3-4) are available upon request.

3.6 Comparison With the 1980s Recession
The early 1980s was another period of significant economic decline. Periods of 1980, 1981, and
1982 were officially recognized as recessions (NBER 2014), and in some ways this period may
have been worse than the Great Recession. The annual unemployment rates, at more than 9.5
percent, were the same or higher in 1982 and 1983 compared to 2009 and 2010. However, the
annual unemployment rate did decline below 8.0 percent in 1984, but remained above 8.0
percent during the Great Recession until 2013 (BLS 2015a). Which period had a greater impact

30
Hispanics were not referenced due to the large overlap between Hispanics and the foreign-born. In the
weighted GSS sample, more than one third of Hispanics are foreign-born and similarly, more than one third of the
foreign-born are Hispanic, though the exact figures depend on the survey year.
66
on the American people? And did the 1980s recession affect SWB through the same channels,
especially unemployment and income?
The 1980s results, which are available upon request, show 1983 was in fact associated
with a larger decline in happiness than 2010. The population was 6.0 percentage points less
likely to report being “very happy” in 1983, and only 4.6 percentage points less likely in 2010.
This result is partially explained by women’s happiness. Women reported substantially larger
declines in 1983 (5.5 percentage points) than in 2010 (3.2 percentage points). This result is
unexpected given women faced larger decreases in employment and income during the Great
Recession.
31
The foreign-born again reported greater declines in happiness than the rest of the
population, but the difference was not as extreme (1983: 7.2 percentage point decline, 2010: 15.1
percentage points).
32

The 1980s recession warrants further analysis. Unlike 2010, the decline in 1983 cannot
be explained with declining income and rising unemployment. Even married people, who are
full-time employed, have no kids, have more than a high school education, and excluding the
effects of income, are statistically less happy in 1983 than trend levels. The two recessions also
differ in duration. Figure 3-1 illustrates how long happiness was below trend during the Great
Recession, and how short the deviation was during the 1980s. Statistically the Great Recession’s
impacts started in 2008, at a 2.0 percent point decline in happiness from trend, while the 1980s
happiness decline began in 1982 at only 1.0 percentage point (the decline in 1980 was not

31
Compared to 1980, women’s 1983 self reported income was approximately the same, and their unemployment
share increased by approximately 0.9 percentage points.
32
The analysis for the 1980s mirrors that for the Great Recession. The models employed were the same, except
the Great Recession dummies were for the years 2008 and 2010, and the 1980s dummies were for the years 1980,
1982, and 1983. The past recession dummy was also changed. The 1980s recession years were swapped for the
Great Recession years.
67
statistically significant and the GSS was not fielded in 1981). To compare the recessions’
impacts, future analysis should also account for their duration.

3.7 Conclusion
Surveys from mid-2010, one year after the official end of the Great Recession, mark the lowest
level of happiness in the United States since consistent measurement began in the early 1970s.
Declining income and employment from the Recession best explain the drop in happiness during
2010. Of the population groups studied, the foreign-born reported declines in both income and
employment that were among the largest, and correspondingly, the largest decline in happiness,
which was more than three times greater than the full population. Compared to women, men
reported greater declines in happiness, income, and employment. Contrary to expectations,
young adults (18-24) reported a smaller decline in happiness than older people, which is likely
because they reported one of the smallest increases in unemployment. The most important macro
relationships during this period were associated with the rising unemployment rate and income
inequality. The other macro variables, GDP, inflation, house prices, and social expenditures did
not statistically affect happiness when individual characteristics were also controlled. To help
summarize the results, Figure 3-2 illustrates the initial declines by population group and
estimates of the declines excluding the effects of changing income. As a reminder, supporting
income and employment data from the GSS are presented in Appendix Table B-3.
The mechanisms are not surprising. As discussed in the literature section, past work on
economic crises has pointed to income loss and unemployment as the drivers of declining well-
being. In general, income losses have larger negative effects on well-being than the positive
effects of gaining an equivalent amount (Kahneman and Tversky, 1979), and during the
68
Recession this effect was likely amplified. Individuals faced reduced consumption and increased
stress associated with meeting financial obligations, especially mortgages. Income loss also
affects factors not strictly related. Take spousal job loss as an example; it will reduce family
income and possibly affect marital satisfaction. Underemployment is another - it is likely to
reduce income and job quality, thus affecting job satisfaction. The effects of unemployment on
well-being are also far reaching. Beyond its effects on income, there are substantial non-
pecuniary costs. Winkelmann and Winkelmann (1998) show that the non-pecuniary effects of
unemployment on life satisfaction are larger than from the loss of income alone. The
Introduction to the journal this analysis is published in also discusses the effects of job loss
during the Great Recession, including the non-pecuniary effects (Kalleberg and von Wachter
2017).
The results are based on estimates of group-specific deviations from group-specific
trends covering a period of approximately 40 years, with various micro and macro controls to
explain the deviations, and supplemented by panel-data analysis with individual fixed effects.
The analysis differs substantively from the two closest studies, (Graham, Chattopadhyay, and
Picon 2010) and (Deaton 2011). Without their data limitations, the present analysis is better
placed to document the effects of the Recession on the SWB of different populations in a long-
term context.

69
Chapter 4 Happier People Are Less Likely To Be Unemployed: Evidence From
Longitudinal Data In Germany

4.1 Introduction
Traditional empirical models of labor market performance omit important determinants. In
theory wages depend on a worker’s marginal productivity, which has been modeled most
frequently using human capital variables like education and experience. However, more recent
theoretical and empirical work has begun incorporating psychological measure, especially
“character skills” (Borghans et al. 2008). Character skills affect labor market outcomes
independently of traditional human capital measures; in some cases they are more important than
cognitive ability (Heckman and Kautz 2013). Character skills are typically measured using the
Big-Five personality traits, but, the Big-Five are usually unavailable in national surveys. This
paper demonstrates that subjective well-being (SWB) can be used to predict unemployment, and
that SWB measures, like happiness, are closely related to character skills. SWB represents an
alternative psychological measure that predicts performance, and one that is more widely
available than personality traits.
While the burgeoning economics of happiness literature has focused mostly on the
determinants of SWB, there is a small subset that evaluates the outcomes of SWB. The findings
from this subset show SWB positively determines many factors that affect labor market
outcomes including: productivity, health, education, and social capital (De Neve et al. 2013).
However, the mechanisms through which SWB positively affect outcomes are not often
identified. In the present context, it is helpful to think of SWB as an index that summarizes
multiple dimensions of an individual, including both life circumstances and psychological
attributes. Improvements in an individual’s productive traits are likely to improve performance
70
as well as SWB. By controlling for life circumstances, this paper focuses on the effects of SWB
on unemployment that are due to changes in otherwise unobservable psychological
characteristics.
The results in the present paper show: people reporting higher SWB are less likely to be
unemployed, and personality operates as one of the mechanisms. Changes in the Big-Five
personality traits are related to changes in SWB. For example, within-person increases in
extraversion or emotional stability are associated with increases in SWB. The SWB-unemployed
relation has a meaningful magnitude as well. A one standard deviation increase in lagged SWB is
associated with approximately a one-percentage point decline in the probability of being
unemployed in Germany (1996-2013). The results also show two important variants to the
relation. The magnitude is substantially larger for people who are currently unemployed than
employed people (a novel result). Second, at high levels of SWB, greater SWB is associated with
a greater likelihood of being unemployed. SWB can be a bad thing, because the SWB-
unemployed relation is quadratic (consistent with Krause, 2013).
The SWB-unemployed relation is estimated using separate dynamic and fixed-effects
regressions with a rich set of controls that account for many potential sources of bias. In the
dynamic specification, unemployment is regressed on lagged unemployment and lagged SWB;
reverse causality could only operate through an individual’s expectations of being unemployed,
which is captured using controls for individuals’ job-market expectations and job satisfaction.
The individual fixed effects capture fixed unobserved factors. The other controls include: typical
socio-economic variables, self-reported health, unemployment history, industry and occupation
categories, and macro and seasonal effects. SWB data come from the German Socio-Economic
Panel (GSOEP).
71
Additional robustness is established using a placebo test and the Arellano and Bond
(1991) (AB) generalized methods of moments (GMM) approach. The placebo test shows that
family satisfaction is not significantly related with being unemployed and thus rules out
measurement error or omitted variables related to satisfaction questions as a mechanism. The AB
method also supports the main results, showing increasing SWB reduces the likelihood of being
unemployed, albeit with a larger magnitude than in the main analysis. The AB method has the
benefits of simultaneously controlling for dynamics and fixed effects, and allowing for
endogeneity in the regressors, but depends on strong assumptions.
This paper adds to the settings and methods used in similar papers. There are no studies
with directly comparable results. The samples used in similar papers are typically constrained to
initially unemployed people. I am not aware of any similar studies that appropriately use fixed-
effects or the same robustness checks. The most similar study is Krause (2013). Krause (2013)
uses a sample of initially unemployed people in Germany, and a similarly rich set of controls but
without fixed effects. The present paper also provides empirical evidence to support personality
traits as one of the mechanisms through which SWB affects outcomes – there is only a limited
amount of similar evidence in the literature (e.g. Krause 2013).
In Chapter 4, Section 4.2 reviews evidence from the literature related to the effects of
SWB on income and employment. Section 4.3 presents the data and primary methods. Section
4.4 presents the main results and robustness tests. Section 4.5 points to character skills as the
mechanism that explains the effects of SWB on unemployment and draws implications for future
research using SWB as a determinant. Section 4.6 concludes Chapter 4.
72
4.2 Past Evidence for Subjective Well-Being’s Relation with Unemployment
Findings from the economics of happiness literature show SWB is a determinant of positive
labor market outcomes. Important mechanisms include productivity, health, and character skills.
Productivity is shown in an experimental setting, the details of which are discussed below
(Oswald, Proto, and Sgroi 2015). The positive relation with health is partially attributed to
“improve[d] immune, cardiovascular, and endocrine functioning (De Neve et al. 2013, 55).”
(Tay, Kuykendall, and Diener 2015)
33
The evidence relating to character skills or personality
traits is discussed in Section 4.5.
Positive effects of SWB are shown both in the psychology and economics literatures. A
review of the evidence from psychology led Boehm and Lyubomirsky to conclude, “happiness
often precedes measures of success and […] positive affect leads to improved workplace
outcomes. (Boehm and Lyubomirsky 2008, 101)” The focus here, however, is on the methods
and results from economics studies. For an introduction to the reverse relationship that focuses
on unemployment as a determinant of SWB, see reviews in Layard, Clark, and Senik (2012) and
Krause (2014).
Among the most notable of the related economics studies are Oswald, Proto, and Sgroi
(2015) and De Neve and Oswald (2012). Oswald, Proto, and Sgroi (2015) is unique for showing
that increasing SWB will causally increase productivity, which they accomplished by inducing
positive affect in three lab experiments.
3435
To extend the lab results, the authors also
implemented a fourth trial, which showed that, participants who reported having experienced a

33
However, not all studies find significant effects on health (c.f. Liu et al. 2016).
34
In two experiments Oswald et al. induce positive affect by showing a randomly selected group of participants a
comedy clip. In a third experiment, the treatment group is provided chocolate, fruit, and drinks. Productivity,
measured on a standardized task, is increased in each case. In experiment three they spent approximately $2 per
person to raise productivity by 20 percent for a short period of concentrated work.
35
Similar lab experiments are rare; however, Ifcher and Zarghamee (2011) also randomly induce positive affect
and show that it reduces time preferences over money.
73
family tragedy (akin to a natural experiment) also reported lower happiness and had lower
productivities. Collectively, results from the three lab experiments and the fourth trial suggest
that increases in emotional- and evaluative-SWB are likely to improve productivity.
36
This
conclusion is further supported by De Neve and Oswald (2012), which shows that emotional-
and evaluative-SWB are positively correlated to log income seven or more years later, using data
from a representative U.S. panel (National Longitudinal Study of Adolescent to Adult Health).
Their results are net of the effects from traditional human capital variables, including IQ, and
also sibling fixed effects.
37

There are a couple other notable studies that show positive effects of SWB. On income
for example, in the U.S. (Mohanty 2009b; Mohanty and Ullah 2012), Russia (Graham, Eggers,
and Sukhtankar 2004), and in more locations generally (De Neve et al. 2013, 56-57). Binder and
Coad (2010) establishes a positive relationship between lagged well-being and each of the
variables: marriage, health, income, and employment status in Britain. They use a vector auto
regresison model that accounts for fixed effects, but it is not clear if they address Ncikell Bias
(Nickell 1981). Two studies that use a prospective analytical approach include, Marks and
Fleming (1999) and Roberts, Caspi, and Moffitt (2003). Each show positive effects of lagged
SWB on work experiences years later (in Australia and New Zealand).
The most similar study is that of Krause (2013). There are only a few studies that I am
aware of that use SWB to predict employment. She finds that SWB has an inverted U-shape
relation with reemployment and reentry wage, using a sample of initially unemployed people in
Germany during 2007 – 2009 (using the IZA Evaluation Dataset S). It is interesting to note that

36
Remember positive affect differs from evaluative SWB, yet family tragedy is likely to have longer lasting
effects on evaluative SWB than positive affect (Helliwell and Wang 2012).
37
When entered together life satisfaction has a larger standardized effect on log income than positive affect,
though admittedly the life satisfaction value was observed closer to the income observation.
74
the results do not hold for women. In her study males exiting into self-employment drive the
main effect on reemployment. While fixed effects are not possible because there are only two
time periods, she is able to control for labor market histories and future job prospects. The
applicability of the findings from Krause (2013) for the current setting is however limited for
two reasons. Krause (2013) focuses on a sample of unemployed people, which depends on the
institutional environment and macroeconomic conditions. She also obtains the effects of SWB on
reemployment, not the effects I investigate, on unemployment generally.
The studies that discuss search behavior similarly have limited comparability because
they discuss reemployment, but they are still informative. The findings from these studies show
unhappier people search for jobs more, which indicates a negative channel through which SWB
affects employment. However, the negative effects may not be enough to offset the otherwise
positive effects of SWB. The estimated net effects from the related literature are mixed (Clark
2003; Mavridis 2012; Krause 2013; Gielen and van Ours 2014). Each study uses fairly
convincing identification strategies, but do not account for omitted fixed determinants of
employment.
The present analysis contributes by estimating the effects of SWB on the likelihood of
being unemployed in a unique setting – the German labor force, including employed and
unemployed people, from 1996 – 2013. Indeed I show that there is a differential effect of SWB
on unemployment when currently unemployed, which is something previous studies were unable
to analyze. My study also controls for additional sources of bias through the use of fixed-effects,
by using a placebo test to rule out measurement error as a driver of the results, and by supporting
the results with Arellano and Bond (1991) type regressions.
75
4.3 Data and Empirical Methods
4.3.1 Data
The SWB and employment data are obtained from the German Socio-Economic Panel (GSOEP),
a nationally representative survey that is made available by the German Institute for Economic
Research (DIW), Berlin.
38
The GSOEP represents one of richest sources of longitudinal SWB
data. It began in 1984 with approximately 18,000 individuals, and with replenishment samples,
allows for repeated observations of more than 35,000 individuals totaling an approximate
280,000 SWB observations for the present study. The particular SWB measure for this chapter is
referred to as life satisfaction, based on the question, “How satisfied are you with your life, all
things considered?”, which is evaluated on a scale from zero to ten.
The analyses are based on reduced samples that exclude people: who are out of the labor
force throughout the entire sample period (e.g. retirees at initial observation), or if they are
missing important variables, such as employment status, income, educational attainment, or
SWB. People without at least three observations are also necessarily excluded in order to use
lagged observations and fixed effects. People who left the sample (attrits)

are discussed further in
Appendix C.1. The period was also restricted to post-1995 due to the availability of a consistent
question on labor market expectations.

4.3.2 Fixed Effects and Dynamic Ordinary Least Squares
Unemployment is assumed to depend on lagged unemployment, lagged SWB, and additional
controls. This data generating process (DGP) is listed below as equation (1). 𝑈
!"
is a dummy

38
German law restricts the sample to 95 percent of the original households (for users in non European Economic
Area countries).
76
variable that takes the value of one if person i is unemployed at time t. 𝑆𝑊𝐵
!"!!
is standardized
39

life satisfaction for person i at period t-1. 𝑿
𝟏,𝒊𝒕
is a vector of exogenous controls such as age,
month, and year effects. Characteristics affected by the individual’s choices, including income
and educational attainment, have been included in 𝑿
𝟐,𝒊𝒕!𝟏
. They are lagged to reduce concerns of
reverse causality.
𝑈
!"
= 𝜌𝑈
!"!!
+𝛿𝑆𝑊𝐵
!"!!
+𝜷
!
!
𝑿
𝟏,𝒊𝒕
+𝜷
!
!
𝑿
𝟐,𝒊𝒕!𝟏
+𝛼
!
+𝜀
!"
(4-1)
Fixed effects, 𝛼
!
, account for all fixed-characteristics observed and unobserved, including
genetics and childhood experience. Including lagged unemployment accounts for unemployment
dynamics, especially persistence, and also allows for additional controls that are conditional on
being employed (e.g. industry and occupation).
Equation 4-1 cannot be estimated by ordinary least squares (OLS) without a serious
source of bias. Nickell Bias (Nickell 1981) arises when including both lagged unemployment and
fixed effects.
40
However equation 4-1 can be estimated when excluding the fixed effects or
lagged unemployment. For this reason, I use fixed-effects (FE) estimation excluding lagged
unemployment, and dynamic OLS (DOLS), which includes lagged unemployment but excludes
the fixed effects. Fixed effects are often preferred, but in the case of employment outcomes,
dynamics may be more relevant (Ashenfelter 1978). Estimates from the two methods should also
bound the true estimate (Angrist and Pischke 2009, 184).
41
Under certain conditions it is possible

39
To standardize, the sample mean is subtracted and this total is divided by the standard deviation.
Standardization has also been used in past SWB research (e.g. Stevenson and Wolfers 2009).
40
Fixed-effects models are typically estimated by subtracting from each variable its mean value over time, and in
the case of a dynamic panel, the mean of the lagged dependent variable is correlated with the mean error term. In
other words, de-meaning introduces a source of endogeneity.
41
If fixed effects represent the true DGP, but a dynamic model is used, then the resulting estimate will
be biased downward. However, if the true DGP is dynamic but fixed effects are used, then the estimate
will be biased upward.
77
to use an alternative approach to account for both fixed effects and dynamics (e.g Arellano and
Bond 1991). I discuss this approach in Section 4.4.4.
The controls include: current labor market indicators (employment status, unemployment
history, income, and industry and occupation categories), traditional human capital variables
(education, and potential experience and its square) (cf. Mincer 1974), socio-characteristics
(religious, parental, and marital status), age categories, self-reported health, month of interview,
and year by region (German state).

4.4 Results: SWB Predicts the Likelihood Of Unemployment
4.4.1 Main Results
Lagged SWB is negatively associated with being unemployed in Germany. The magnitude is
meaningful as well. Table 4.1 presents the results. Recall that unemployed status is binary and
life satisfaction has been standardized. Consequently the coefficients on lagged SWB reflect
differences in the probability of being unemployed with approximately one standard deviation
difference in lagged SWB. Columns 2 and 4 show that a one standard deviation increase in
lagged SWB is associated with a decline in the probability of being unemployed between 0.10
and 0.13 percentage points. As mentioned in the methods section, the DOLS and FE estimates
should set bounds on the true estimate, and indeed, the FE estimate is larger (smaller negative
number) as expected. The DOLS estimations include lagged employment status, omitting
employed fulltime as the comparison group, but exclude fixed effects. In contrast, the fixed-
effects (FE) regressions exclude employment status. Column 1 presents the OLS results for
comparison.
78
A great many potential sources of bias have been accounted for using the DOLS and FE
approaches. Anything that affects current unemployment that also affects lagged employment
status is captured, which includes many traditional human capital characteristics, both observed
and unobserved. Likewise, unobserved ability, family characteristics, and all other fixed
characteristics are captured using fixed effects. That includes sources of measurement error to
the extent they are fixed, such as reporting bias. Time-varying factors that occur at the German
state level are also accounted for with state-by-year dummies.
Table 4-1 Unemployed and Life Satisfaction Regressions. Dependent Variable:
Unemployed

(1) (2)
OLS DOLS
(3)
DOLS
(4) (5)
FE FE
Life Sat. (ST) -0.033*** -0.013***
(0.001) (0.001)
-0.007***
(0.001)
-0.010*** -0.008***
(0.001) (0.001)
Unemployed
Retired
In School
NonWorking
Temp. NonWork
Other Status
0.457***
(0.005)
0.024***
(0.002)
0.031***
(0.004)
0.018***
(0.002)
0.048***
(0.003)
0.050***
(0.003)
0.437***
(0.005)
0.016***
(0.002)
0.022***
(0.004)
0.007***
(0.002)
0.041***
(0.003)
0.044***
(0.003)
Socio-Econ. Controls - - Yes - Yes
Constant
Observations
-0.059*** -0.058***
(0.008) (0.006)
233574 233574
0.350***
(0.014)
233574
0.043 0.086*
(0.036) (0.044)
233574 233574
# of People 35078 35078 35078 35078 35078
Adj. R Sq. 0.050 0.249 0.257 0.014 0.014
Notes: Controls are lagged excluding male, age, month of interview, and year by
region. Additional controls include: (All columns) male, age, month of interview, year
by region, education, potential experience and its square. Socio-Econ Controls include
lagged: log of adjusted family income, self-reported health, marital status, parental
status, and presence of a young child. Dummies for missing observations associated
with income, and health were also included when those variables were included.
Standard errors in parentheses (clustered by individual)
Significance: * p<0.10 ** p<0.05 *** p<.01

79

The SWB-unemployed relationship remains statistically significant, with a reduced magnitude,
as additional time-varying controls are added. Socio-economic controls including adjusted
family income, self-reported health, marital and parental status were added in columns 3, and 5.
The decline in magnitudes makes sense as each of the additional socio-economic controls
mediate the effects of SWB on unemployment. Mediation occurs because people that report
higher SWB have better health, are more productive, and are more likely to get married (De
Neve et al. 2013). Self-reported health is an especially important mediator as they are both self-
reported and highly correlated.

4.4.2 Heterogeneity of SWB – Unemployment Relation
There are reasons to believe the SWB – unemployed relation may vary across different
dimensions. Krause (2013) found SWB was associated with reemployment in a quadratic manner
and also that her SWB-reemployment relation was stronger for men. In particular she found that
beyond a certain point, happier people were more likely to stay unemployed, and that happier
men were more likely to exit unemployment into self-employment. One thing Krause is unable
to test is whether the SWB-unemployment relation differs by employment status. It is plausible
that the effect of SWB on becoming reemployed is different than on the effect of becoming
unemployed. In three additional scenarios I allow for: 1) a quadratic SWB – unemployed
relation, 2) differences by gender, and 3) differences by lagged unemployed status.
The scenario results reveal some important relations, presented in Table 4-2, but in each
case, the main effect of SWB remains statistically significant and negative. The magnitudes are
also similar to the main analysis that includes socio-economic controls. The first scenario results
80
(columns 1 and 2) reveal a strong quadratic SWB – unemployment relation. The marginal effect
of increasing SWB on unemployment becomes positive (unemployment more likely) at
standardized SWB values greater than 0.83 (DOLS) and 1.50 (FE).
42
Greater SWB reduces the
likelihood of unemployment, but too much SWB leads to a higher likelihood of being
unemployed. Krause (2013) suggests that it may be that unemployed people that are relatively
highly satisfied with their lives may have less incentive to become reemployed. The analyses by
gender, in columns 3 and 4, do not reveal different relations for men and women. The coefficient
on the Male X SWB interaction term is not statistically significant.
The SWB-unemployed relation is substantively different for currently unemployed and
employed people. The results from scenario 3 are presented in column 5. For currently
unemployed people, a one standard deviation increase in lagged SWB is associated with a 1.8
percentage point decline in the likelihood of being unemployed, which is three times larger than
the same relation for someone who is not unemployed. The result (1.8 = 100*(-0.012 -0.006))
comes from adding the statistically significant interaction term (Unemployed X Life
Satisfaction) to the main effect (Life Sat. ST). Only the DOLS analysis is used because, as
mentioned, estimating fixed effects and lagged unemployment in the same regression requires
instrumental variable analysis to avoid bias.

42
In the DOLS case, the partial derivative of unemployed status with respect to lagged SWB ( ∂U
it
/ ∂SWB
it-1
)
= -0.005 + 0.006 SWB
it-1
. Setting equal to zero, the turning point is 0.005/0.006 = 0.833.
81
Table 4-2 Heterogeneity of Unemployed - Life Satisfaction Relationship. Dependent
Variable: Unemployed

4.4.3 Robustness - Additional Control Variables and Placebo Test
The primary analysis discussed above controls for fixed unobserved characteristics and time-
varying characteristics affecting concurrent employment outcomes and income. However, there
may still be other omitted time-varying factors or systematic measurement error in lagged SWB
that causes bias. To address these issues additional control variables are added and a placebo (or
falsification) test is used. The results support the conclusion that happier people are less likely to
(1) (2) (3) (4) (5)
DOLS FE DOLS FE DOLS
Life Sat. (ST)
Life Sat. Squared
Male X Life Sat.
Unemployed X Life Sat.
-0.005***
(0.001)
0.003***
(0.000)
-0.006***
(0.001)
0.002***
(0.001)
-0.006***
(0.001)
-0.002
(0.001)
-0.007***
(0.001)
-0.002
(0.002)
-0.006***
(0.001)
-0.012***
(0.004)
Unemployed
Male
0.436***
(0.005)
0.005***
(0.001)
0.437***
(0.005)
0.005***
(0.001)
0.430***
(0.005)
0.004***
(0.001)
Socio-Econ. Controls
Constant
Observations
Yes
0.353***
(0.014)
233574
Yes
0.102**
(0.046)
233574
Yes
0.354***
(0.014)
233574
Yes
0.107**
(0.046)
233574
Yes
0.356***
(0.014)
233574
# of People
Adj. R Sq.
35078
0.257
35078
0.015
35078
0.257
35078
0.014
35078
0.257
Notes: Controls are lagged excluding male, age, month of interview, and year by
region. Additional controls include: (All columns) age, month of interview, year by
region, education, potential experience and its square. Socio-Econ Controls include
lagged: log of adjusted family income, self-reported health, marital status, parental
status, and presence of a young child. Dummies for missing observations associated
with income, and health were also included when those variables were included.
Standard errors in parentheses (clustered by individual)
Significance: * p<0.10 ** p<0.05 *** p<.01

82
be unemployed, in a robust fashion, but with the added qualification that job satisfaction
represents one of the important mechanisms.

Additional Variables:– It is important to account for channels operating through SWB that may
affect the likelihood of being unemployed directly. For example, an individual may have been
unemployed in the past, which reduced SWB (Clark, Georgellis, and Sanfey 2001) and also
reduced job opportunities directly. Alternatively an individual could be less happy because while
currently employed, they know they are going to lose their job, or because they are not satisfied
with their job and plan to quit (Clark 2001). Social capital (or networks) is another important
channel. Happier people have greater social capital (Guven 2011) and social networks could lead
to improved labor outcomes. To account for these channels, variables for unemployment history,
job satisfaction and expectations, and social capital were added as controls. The length of time
(tenure) in a current position was also added, as well as industry and occupation categories. The
additional variables are defined in the Appendix C.2 and described briefly here.
Unemployment history is defined as the total length of time a person has been
unemployed (in years). “JobFind Difficult” accounts for an individual’s expectations of poor
labor market conditions. Scaled 1 – 3 it takes greater values if the person believes it would be
difficult to find a comparable job in the event they needed to. Job satisfaction is scaled from 0 –
10, taking greater values as satisfaction increases. Social capital is measured as a set of binary
variables that take the value of one if the individual participated in an events or activities at least
once in the last month. Events and activities include whether the participant: met with friends or
relatives, visited with people in the community or at a bar, volunteered, participated in religious
or sports events, and attended cultural or other entertainment events. Social capital variables
83
were omitted from the main analysis because the relevant questions were asked in only half of
the years used for the main analysis.
The results presented in Table 4-3 show a declining magnitude for the lagged SWB-
unemployed relation when including the additional controls, but it generally remains statistically
significant. The relationship declines to -0.005 (Table 4-3, column 1) from -0.007 (Table 4-1,
column 3) when adding controls for unemployment history, tenure, and industry and occupation
categories. Further adding job market expectations and satisfaction reduces the relation to a
significant -0.002 (Table 4-3, column 2). Adding job satisfaction as a control is a particularly
strong test. Job satisfaction is highly correlated with life satisfaction (35%), which makes
identifying separate relationships for each variable challenging. It requires a high degree of
statistical power. Note that once again only the DOLS analysis is used when the included
variables depend on employment status.
Adding the social capital variables does not change the relations much, even though the
sample is reduced to approximately a third of the original. The DOLS and FE relations remain
the same (compare Table 4-1, columns 3 and 5, with Table 4-3, columns 3 and 4). However,
reducing the sample size when adding the additional job variables reduces the precision and
statistical significance of the SWB-unemployed relation (contrast columns 2 and 5). As
mentioned there is a high degree of collinearity of between life satisfaction and job satisfaction. I
interpret the results as, happier people are less likely to be unemployed, but being happy at work
is most important.

84
Table 4-3 Unemployed and Life Satisfaction Regressions with Additional Control
Variables. Dependent Variable: Unemployed

The relationships are in line with expectations. Those people who have been in their job longer
are less likely to be unemployed in the future (Tenure). People who have been unemployed a
greater amount of time are more likely to be unemployed again (Unempl. Hist.) as expected
(Clark et al. 2008). Expecting it to be difficult to find a comparable job is negatively associated
(1)
DOLS
(2)
DOLS
(3)
FE
(4)
DOLS
(5)
DOLS
(6)
DOLS
Life Sat. (ST) -0.005***
(0.001)
-0.002***
(0.001)
-0.008***
(0.002)
-0.007***
(0.001)
-0.002
(0.001)
-0.001
(0.001)
Unemployed 0.387***
(0.005)
0.287***
(0.009)
0.431***
(0.008)
0.289***
(0.014)
0.289***
(0.014)
Tenure (yrs)
Unmpl. Hist
JobFind Difficult
-0.001***
(0.000)
0.017***
(0.001)
-0.001***
(0.000)
0.016***
(0.001)
0.038***
(0.002)
-0.001***
(0.000)
0.015***
(0.001)
0.042***
(0.003)
-0.001***
(0.000)
0.015***
(0.001)
0.042***
(0.003)
Job Satisfaction -0.004***
(0.000)
-0.004***
(0.001)
-0.004***
(0.001)
Socio-Econ. Controls
Industry & Occ.
Social Capital
Yes
Yes
-
Yes
Yes
-
Yes
-
Yes
Yes
-
Yes
Yes
Yes
-
Yes
Yes
Yes
Constant
Observations
0.204***
(0.014)
233574
0.215***
(0.015)
233574
0.131*
(0.071)
79870
0.352***
(0.022)
79870 79870 79870
# of People 35078 35078 28590 28590 28590 28590
Adj. R Sq. 0.274 0.280 0.024 0.257 0.279 0.279
Soc. Cap. F stat. 0.939 7.088 3.728
Notes: Controls are lagged excluding male, age, month of interview, and year by
region. Social Capital (Joint) is the linear combination of dummies for participation in
various events that are described further in Appendix B.
Additional controls include: (All columns) male, age, month of interview, year by
region, education, potential experience and its square. Socio-Econ Controls include
lagged: log of adjusted family income, self-reported health, marital status, parental
status, and presence of a young child. Industry and Occupation dummies are described
further in Appendix C.2. Dummies for missing observations associated with income,
health, and job satisfaction and expections were also included when those variables
were included. Standard errors in parentheses (clustered by individual)
Significance: * p<0.10 ** p<0.05 *** p<.01

85
(JobFind Difficult). People who are more satisfied in their jobs are also less likely to be
unemployed. Greater social capital is related statistically to a lower likelihood of future
unemployment in the DOLS regressions (Table 4-3, columns 4 and 6), but not the FE regression
(column 3). Given the large number of social capital variables, the individual coefficients have
been omitted from Table 4-3, but an F-stat for the joint significance is reported. The relationship
magnitudes of the additional control variables should not be strongly interpreted. The variables
were included to test the robustness of the SWB-unemployed relation, but are often collinear and
potentially endogenous.

Placebo Test:–Some readers may be concerned that there is a systematic measurement error or
an omitted variable related to satisfaction questions that drive the results. Placebo or falsification
tests can be used to test this challenge. In particular the test replicates the main analyses using
family satisfaction as an alternative explanatory variable to life satisfaction. Family satisfaction
is used for two reasons. First, in theory family satisfaction is not a determinant of unemployment
to the degree that life satisfaction is. Second, if measurement error present in satisfaction
questions drives the life satisfaction – unemployed relation, then the same measurement error is
likely to be present in family satisfaction.
Table 4-4 presents the results of the placebo test. Lagged family satisfaction is not related
to unemployment once socio-economic controls or fixed effects have been included. This
insignificant relation means that the life satisfaction – unemployed relation is not driven by
measurement error or omitted variables related to satisfaction questions. Columns 1 – 5 directly
correspond to the columns 1 – 5 of Table 4-1. They differ only by sample size and the use of
family satisfaction instead of life satisfaction. Column 6 shows that life satisfaction is still a
86
significant determinant in the same sample and using the same specification as used for family
satisfaction (column 5). The sample is reduced based on the availability of the family satisfaction
question.

Table 4-4 Placebo Test: Unemployed and Family Satisfaction Regressions. Dependent
Variable: Unemployed

4.4.4 Arellano and Bond’s System Generalized Method of Moments
The results from the alternative Arellano and Bond approach confirm happier people are less
likely to be unemployed. The approach has some benefits compared to the main analysis, but
depends on certain conditions. Dynamics and fixed effects can both be accounted for in the same
specification and the method allows for endogeneity of the regressors. The relationships are also
considered causal. However, for the present paper, Arellano and Bond is used to support the
(1) (2) (3) (4) (5) (6)
OLS DOLS DOLS FE FE FE
Fam. Sat. (ST)
Life Sat. (ST)
-0.009*** -0.004*** 0.001 0.000 0.001
(0.001) (0.001) (0.001) (0.001) (0.001)
-0.006***
(0.001)
Socio-Econ. Controls
Constant
Observations
- - Yes - Yes Yes
-0.077*** -0.059*** 0.371*** 0.124 0.115 0.095
(0.010) (0.008) (0.019) (0.086) (0.095) (0.095)
92430 92430 92430 92430 92430 92430
# of People 26999 26999 26999 26999 26999 26999
Notes: Controls are lagged excluding male, age, month of interview, and year by
region. Additional controls include: (All columns) male, age, month of interview, year
by region, education, potential experience and its square. Socio-Econ Controls include
lagged: log of adjusted family income, self-reported health, marital status, parental
status, and presence of a young child. Dummies for missing observations associated
with income, and health were also included when those variables were included.
Standard errors in parentheses (clustered by individual)
Significance: * p<0.10 ** p<0.05 *** p<.01

87
main results and not to advance a statistically identified causal relationship because it is
debatable whether the necessary conditions are met.

Method Description:–To understand Arellano and Bond (1991) (referred to as “AB”) it is
useful to begin with Anderson and Hsiao (1981). To account for fixed effects Anderson and
Hsiao apply first differences to Equation 4-1. First differences, however, cause the lagged
differenced dependent variable to be related to the differenced error term. To overcome this
problem, lagged differenced unemployment is predicted using an instrumental variable approach.
Equation 4-2 presents the second-stage specification:
𝛥𝑈
!"
= 𝜌𝛥𝑈
!"!!
+𝛿𝛥𝑆𝑊𝐵
!"!!
+𝜷
!
!
𝛥𝑿
𝟏,𝒊𝒕
+𝜷
!
!
𝛥𝑿
𝟐,!𝒕!𝟏
+𝛥𝜀
!"
(4-2)
Where 𝛥𝑈
!"
= 𝑈
!"
−𝑈
!"!!
and 𝛥𝑈
!"!!
is the predicted value for 𝛥𝑈
!"!!
. Anderson and Hsiao
(1981) suggest 𝑈
!"!!
as a valid instrument for 𝛥𝑈
!"!!
. 𝑈
!"!!
is relevant because it correlated with
𝛥𝑈
!"!!
and it is excludable (not correlated with 𝛥𝜀
!"
) if there is no autocorrelation in the level
equation 1 (i.e. 𝑐𝑜𝑣 𝜀
!"
,𝜀
!"!!
= 0). 𝛥𝑆𝑊𝐵
!"!!
and 𝛥𝑿
𝟐,𝒊𝒕!𝟏
are also allowed to be endogenous
and predicted in the same way as 𝛥𝑈
!"!!
. Building on this approach, AB recognized that further
lags could be used as additional instruments. For example, both 𝑈
!!
and 𝑈
!!
are valid instruments
for 𝛥𝑈
!!
. For 𝛥𝑈
!!
even more instruments are available, namely: 𝑈
!!
, 𝑈
!!
, and 𝑈
!!
. In this way,
an additional instrument is added for each time period. Estimation of this structure is then
performed using GMM. See Arellano and Bond (1991) for further details. The AB approach also
allows for traditional exogenous instruments. To increase efficiency I also use “death of a
partner” as an excluded instrument.
In the present context, there are challenges facing the AB approach. There is
autocorrelation in the level equation 1 (i.e. 𝑐𝑜𝑣 𝜀
!"
,𝜀
!"!!
≠ 0), because there is serial
88
correlation present in unemployment, which means 𝑈
!"!!
will not be excludable. To reduce this
problem, I use AB instruments that begin with lag order three (e.g. 𝑈
!"!!
), instead of two, for all
of the endogenous variables. In this way the condition for the level relation is relaxed to
𝑐𝑜𝑣 𝜀
!"
,𝜀
!"!!
= 0. However, there may be different orders of serial correlation in the variables
treated as endogenous (𝛥𝑈
!"!!
, 𝛥𝑆𝑊𝐵
!"!!
, and 𝛥𝑿
𝟐,𝒊𝒕!𝟏
) and tests are necessary to provide
assurance that the exclusion restrictions are being met. I use the typical Hansen J test of over-
identifying restrictions, and also a test that is specific to AB type models. Because the excluded
instruments are lags of the endogenous variables, one can test their excludability by the degree of
autocorrelation in the predicted residuals. It is expected that there is first-order autocorrelation in
the predicted residuals from Equation 4-2 (i.e. 𝑐𝑜𝑣 𝛥𝜀
!"
,𝛥𝜀
!"!!
≠ 0) due to the mechanical
relation between 𝛥𝑈
!"!!
and 𝑈
!"!!
. However, there should be no second-order autocorrelation
(i.e 𝑐𝑜𝑣 𝛥𝜀
!"
,𝛥𝜀
!"!!
= 0) for the instruments to be valid.

Results:–Greater life satisfaction is statistically associated with a lower likelihood of being
unemployed across each of five different AB type regressions. The results are presented in Table
4-5. The relationship magnitudes increased substantially from approximately 0.01, in the main
results, to approximately 0.04. The range of AB results is from -0.035 to -0.052, based on
specifications that differ in several important ways. However, only the specifications in columns
3 and 4 pass the tests necessary for consistency, suggesting the 0.04 magnitude is the most
reliable. Column 5 uses the column-4 analysis but with added socio-economic controls that are
also allowed to be endogenous.
As mentioned, I present the AB results as support for the main analysis, but not to
advance a statistically identified causal relationship. There is substantial serial correlation in both
89
unemployment and SWB, which challenges the excludability of the AB-type instruments. While
the AR(2) and Hansen J test results support the AB models used in columns 3 and 4, the
evidence from tests of exclusion restrictions is suggestive in nature, not conclusive. The
remainder of the section describes the different specifications and test results in Table 4-5.
Table 4-5 Arellano and Bond Type GMM Estimation. Estimation Dependent Variable:
ΔUnemployed. Relations Presented in Levels for Unemployed

The different specifications are used primarily to obtain excludable instruments. Column 1
presents the results from the AB approach described above. The results from the AR(2) test
results reveal the presence of second order autocorrelation in the error terms, indicating the
model is not valid. Although the Hansen J suggests the instruments are excludable, the AR(2)
(1) (2) (3) (4) (5)
AB AB Factor GMM Factor GMM Factor GMM
L Life Sat. (ST)
L2 Life Sat. (ST)
L Unemployed
L2 Unemployed
-0.035*** -0.035*** -0.040*** -0.044*** -0.052***
(0.008) (0.008) (0.014) (0.016) (0.016)
-0.005 0.005 0.005 0.009
(0.007) (0.011) (0.012) (0.011)
0.569*** 0.238*** 0.428*** 0.436*** 0.324***
(0.029) (0.038) (0.084) (0.122) (0.080)
0.240*** 0.215*** 0.171** 0.236***
(0.029) (0.061) (0.081) (0.054)
Socio-Econ. Controls
Observations
- - - - Yes
165974 165974 165974 165974 165974
Hansen J p-value
AR(1) p-value
AR(2) p-value
0.233 0.658 0.287 0.120 0.414
0.000 0.000 0.000 0.000 0.000
0.000 0.000 0.208 0.608 0.048
# of Instruments 731 910 225 132 237
Notes: AB is the traditional Arellano and Bond one-step GMM estimation. Factor
GMM replaces the GMM-style instruments with their principal components Controls
are lagged excluding male, age, month of interview, year, and region. Additional
controls include: (All columns) male, age, month of interview, year, and region,
education, potential experience and its square. Socio-Econ Controls include lagged:
log of adjusted family income, self-reported health, marital status, parental status, and
presence of a young child. Dummies for missing observations associated with income,
and health were also included when those variables were included. Standard errors in
parentheses (clustered by individual)
Significance: * p<0.10 ** p<0.05 *** p<.01

90
test is more reliable in the present case because the Hansen J test suffers from low power when
there are many instruments (in this case more than 700). Column 2 adds the second lag of life
satisfaction and unemployment as endogenos variables to help account for the serial correlation
that is causing the AR(2) test to fail. However, while the Hansen J test improves, the AR(2) still
fails.
The estimates in columns 3 and 4 are more reliable, because the AR(2) and Hansen J tests
each suggest the models are correctly specified. Using Factor GMM, the endogenous variables
are predicted using the factors of lagged values, but not the lagged values themselves, which
seems to capture the previous serial correlation. Factor GMM replaces the AB instruments with
their principal components. In particular, the 910 instruments used in column 2 are reduced to
225 components in column 3. Reducing the number of instruments improves efficiency and
reduces problems associated with weak instruments. In column 4 the instruments are further
reduced by limiting the number of lags that are used to obtain components. The original GMM
instruments include lags three and higher depending on the period of observation (the number of
lags used for an endogenous variable = t - 3 for t ≥ 4). In column 4, only lags three, four, and five
are allowed, and the components derived from these instruments number 132. See Kapetanios
and Marcellino (2010) for an explanation of Factor GMM. Estimation is implemented using the
Stata user written command xtabond2 (Roodman 2009).

4.5 Mechanisms – Character Skills
Character skills are likely to be one of the mechanisms through which SWB affects
unemployment. ‘Character skills’ is a broad term used to describe “soft skills, personality traits,
non-cognitive skills, non-cognitive abilities, character, and socio-emotional skills (Heckman and
91
Kautz 2013, 10).” They are usually measured using personality traits, especially the Big-Five
personality measures: Openness to Experience, Conscientiousness, Extraversion, Agreeableness,
and Neuroticism.
43
Other traits include optimism, locus of control, and reciprocity.
Character skills are interrelated with SWB. They are both determinants and outcomes of
each other (Roberts, Caspi, and Moffitt 2003; Mohanty 2009b; Specht, Egloff, and Schmukle
2012; and Soto 2015). One study states, “that personality traits and well-being aspects
reciprocally influence each other over time. (Soto 2015, 45).” A similar statement is also found
in Roberts et al. (2003, 590). The relationship between SWB and the Big-Five personality traits
is further supported using the GSOEP data. Specht, Egloff, and Schmukle (2012) shows that
changes in the Big-Five affect changes in life satisfaction, and also that the relationship holds in
reverse - changes in life satisfaction affect the Big-Five.

4.5.1 Past Evidence of Character Skills as a Mechanism
Character skills have become the subject of growing research in economics. Formal economic
theory is being developed to include character skills but it is still in its infancy (Borghans et al.
2008). Intuitively from human capital theory, labor market outcomes depend on character skills
because personality contributes to individuals’ productive traits. Indeed performance and job
sorting are partially explained by varying personality traits. For example, “[e]xtroversion was
positively linked to job performance in management and sales occupations, (Mueller and Plug
2006, 6).”
Empirically, character skills predict a variety of outcomes even when traditional human
capital measures, including IQ, have been controlled (Heckman and Kautz 2013, 4). In the SWB

43
The Big-Five measures have become relatively well accepted, “[t]here are reliable ways to measure them, and
there are proven ways to enhance them and to evaluate efforts to foster them. (Heckman and Kautz 2013, 5)”
92
literature, character skills are often suggested as a mechanism through which SWB affects labor
outcomes: later life wages can be explained by “higher degrees of optimism and extraversion and
less neuroticism (De Neve and Oswald 2012, 19957).” Mohanty (2009b) attributes positive
attitude toward oneself as a channel, which is similar to locus of control. Likewise Krause (2013)
uses “the concept of locus of control and the personality traits of neuroticism and extraversion”
to explain happiness’ effect on reemployment. Roberts, Caspi, and Moffitt (2003) also describes
personality measures similar to neuroticism and social closeness as the most important traits
predicting work outcomes.
While personality traits are often thought to be relatively fixed over time, recent literature
has shown that is not the case. Indeed one of the studies that is responsible for the former view
does not say personality traits are unchangeable (Costa and McCrae 1994, 22). Among their
conclusions they say, “People surely grow and change, but they do so on the foundation of
enduring dispositions (Costa and McCrae 1994, 36).” Recent research suggests dispositions
change too, at least as measured using the Big-Five personality traits, which were shown to
change throughout adulthood. “The Big Five traits are complex phenomena subject to a variety
of developmental influences (Srivastava et al. 2003, 1041).” Furthermore, (Heckman and Kautz
2013) summarize the evidence for lasting benefits of interventions that aim to improve character
skills, both in early childhood and workplace based. “[T]he answer to the question of whether the
change in personality is possible must be a definite yes. (Borghans et al. 2008, 1020)”

4.5.2 Character Skills Operate as a Mechanism
Character skills are plausible mechanisms that help to explain the relationship between lagged
SWB and unemployment. However, cognitive ability does not represent an omitted variable or
93
mechanism operating in the present context. Lagged SWB is partially explained by the Big-Five
personality traits, optimism, reciprocity, and external locus of control, but not by two different
measures of cognitive ability. The results hold whether controlling for employment variables or
fixed effects. Not only do personality traits changer over time for a given individual, but they
explain corresponding within-person changes in SWB. The results are presented in Table 4-6,
but first a discussion of the methods is warranted.
There are many factors through which SWB could affect unemployment. As discussed in
the results section, many of the control variables represent channels that mediate the effects of
SWB. The remaining mechanisms must affect lagged SWB independently of the controls used in
the main analysis. To test plausible factors, lagged SWB is regressed on the control variables
used in equation 1 and additional variables representing plausible mechanism. Equation 4-3
presents the OLS specification as an example that corresponds to the main DOLS estimates.
𝑴
𝒊𝒕!𝟏
is a vector of variables that may operate as mechanisms. Analogous fixed-effects
regressions are also used.
𝑆𝑊𝐵
!"!!
= 𝛾
!
𝑈
!"!!
+𝜷
!
!
𝑿
𝟏,𝒊𝒕
+𝜷
!
!
𝑿
𝟐,𝒊𝒕!𝟏
+𝜸
𝟐
!
𝑴
𝒊𝒕!𝟏
+𝜇
!"
(4-3)
In Table 4-6, the OLS specifications correspond to the Table 4-3, column 2, DOLS regression.
They include lagged socio-economic controls and all of the same employment variables. The FE
regressions correspond to specification in Table 4-1, column 5, by excluding the lagged
employment variables but including individual fixed effects. The personality and intelligence
variables are standardized to facilitate comparison of their coefficients. The specific definitions
are included in Appendix C.2. The columns test different variables because they are not all
available in the same years.
94
The results, presented in Table 4-6, support the explanations that SWB affects labor
market outcomes through increased extraversion and less neuroticism (De Neve and Oswald,
2012; Krause, 2013). Indeed each of the Big-Five personality traits are statistically related to
SWB in the expected ways and consistent with past literature (Soto 2015). The pattern is similar
whether using fixed effects or controlling for employment variables. Neuroticism (or emotional
instability) is negatively related with SWB. The other Big-Five personality traits are positively
and significantly related. While optimism is not included among the Big-Five, it shows the
expected, highly significant and positive relationship with SWB. Changes in optimism also have
the largest relationship with changes in SWB. Neuroticism has the next largest relationship.
Table 4-6, columns 3 and 4, show the results when using reciprocity and locus of control.
The results may again be expected. Reciprocity is broken into positive reciprocity, when a person
wants to repay good deeds, and negative, when someone desires to “get even” following negative
events. It was not clear before hand what direction reciprocity would have with SWB
44
, but it is
not surprising that positive reciprocity is positively related to SWB, while negative has the
opposite direction. Locus of control also has two dimensions that are treated separately. Greater
external locus of control means the person believes outcomes are controlled more by more
external factors. It is negatively related to SWB as expected. Internal locus of control relates
more to personal control. It is omitted because internal consistency is low (Heineck and Anger
2010).

44
To my knowledge there is little SWB research in economics that discusses reciprocity.
95
Table 4-6 Mechanisms: Lagged Life Satisfaction Regressions. Various years.
a

(1) (2) (3) (4) (5) (6)
OLS FE OLS FE OLS FE
Agreeable
Conscientious
Neuroticism
Extraversion
Openness
Optimism
0.049*** 0.025**
(0.007) (0.011)
0.029*** 0.044***
(0.008) (0.012)
-0.093*** -0.109***
(0.007) (0.012)
0.042*** 0.041***
(0.007) (0.013)
0.030*** 0.029**
(0.008) (0.012)
0.211*** 0.153***
(0.007) (0.010)
Reciprocity (pos.)
Reciprocity (neg.)
Ext. Locus of Control
0.080*** 0.076***
(0.009) (0.013)
-0.027*** -0.016
(0.008) (0.012)
-0.127*** -0.089***
(0.009) (0.013)
Symbol Test
Word Fluency Test
-0.031 -0.059
(0.031) (0.040)
0.003 0.033
(0.027) (0.039)
Socio-Econ. Controls
Employment Variables
Constant
Observations
Yes Yes Yes Yes Yes Yes
Yes - Yes - Yes -
-3.304*** -3.398*** -3.515*** -3.634*** -5.905*** -3.290
(0.219) (0.945) (0.290) (1.166) (0.582) (4.828)
20288 20288 12947 12947 1792 1792
# of People
Adj. R Sq.
10144 10144 6789 6789 896 896
0.401 0.111 0.365 0.080 0.323 0.094
a. Columns (1, 2): 2005 and 2009; (3, 4) 2005 and 2010; and (5, 6) 2006 and 2012.
Notes: Additional controls include: (all columns) male, age, month of interview, year,
and region, education, potential experience and its square. Socio-Econ Controls
include lagged: log of adjusted family income, self-reported health, marital status,
parental status, and presence of a young child. Employment Variables include lagged:
employment categories, tenure, unemployment history, JobFind Difficult, Job
Satisfaction, and Industry and Occupation categories. Dummies for missing
observations associated with income, health, job satisfaction and expectations were
also included when those variables were included.
Standard errors in parentheses (clustered by individual)
Significance: * p<0.10 ** p<0.05 *** p<.01

96
Columns 5 and 6 include measures of cognitive ability as alternative mechanisms.
Cognitive ability is measured using two tests with each representing different aspects of
intelligence. The word fluency test is most comparable to measures of crystalized intelligence,
which increases as the individual attains more knowledge through experience or education. It is
also affected by fluid intelligence, which is related to the speed of cognition. The second test is
the symbol correspondence test, which is most closely related to fluid intelligence.
Intelligence does not have a significant relationship with SWB when controlling for
either the employment variables (column 5) or fixed effects (column 6). Note there is a high
degree of correlation between the two and their relationships with SWB should not be interpreted
independently. However, jointly they are also statistically insignificant (with a max F-stat of 1.2
in the FE regression).

4.5.3 Implications for Future Research
The results suggest traditional models of labor market performance should be expanded. Both
character skills and SWB are important determinants of labor market outcomes and labor market
equations should be adjusted accordingly. “Focusing on one dimension of human skills to the
exclusion of other dimensions misses fundamental aspects of human performance and
development (Heckman and Kautz, 2013, 5).” Although economic theories of wage
determination, for example, have traditionally focused on cognitive ability, character skills
should not be ignored. “For many outcomes, the predictive power of character skills rivals that of
measures of cognitive ability (Heckman and Kautz 2013, 23).” Another study concludes that,
“…the effects of psychological capital variables in predicting wages are even stronger than the
effects of traditional human capital variables (Mohanty 2009a, 357).”
97
The literature supporting character skills as determinants of performance is greater than
for SWB, but when measures of character skills are absent, SWB may suffice as a proxy.
Obtaining personality measures in general surveys can be time-prohibitive, and many surveys do
not include them. In such an instance, Guven (2011) suggests residual SWB as a proxy. Residual
SWB is the portion of SWB that is not explained by observable life circumstances. Guven (2011)
shows that residual SWB is the strongest predictor of optimism and suggests this finding as an
“invaluable opportunity” to researchers. “… [O]ne can easily calculate residual happiness in
these datasets and can use them as a variable that is highly correlated with optimism (Guven
2011, 180).” One can also use raw SWB, in contrast to residual SWB, with the same
interpretation so long as they include controls for life circumstacnes.
While SWB may serve as a reasonable proxy for character skills, it is important to
remember they are distinct. Self-esteem serves as an example. It is another trait that may be
considered a character skill, and conceptually, is closely enough related to happiness to warrant
direct comparison. In a theoretical and empirical investigation of the similarities between
happiness and self-esteem the authors determine that while they are strongly correlated, they
represent separate psychological constructs (Lyubomirsky, Tkach, and DiMatteo 2006). Also
remember that the effects of SWB are not mediated solely by character skills, SWB affects
additional channels too (e.g. positive affect’s effect on productivity, Oswald et al. 2015).

4.6 Conclusions
Recent research has begun to use psychological variables to measure important dimensions of
labor performance. Support for so called “character skills” as drivers of numerous outcomes is
increasing in both empirical studies and economic theory. Character skills are typically measured
98
using personality traits, such as the Big-Five traits, but personality and alternative measures are
not frequently available. This paper suggests subjective well-being (SWB) as a more widely
available psychological measure that can be used in analyzing the determinants of
unemployment.
This paper contributes to both the labor economics and the economics of happiness
literature. The main results demonstrate that people reporting higher SWB are less likely to be
unemployed, and the relationships have meaningful magnitudes as well. A one standard
deviation improvement in SWB is associated with approximately a one-percentage point
decrease in the likelihood of being unemployed in Germany. The effects can be explained in part
by character skills. Additional analysis shows that improvements in an individual’s character
skills, such as reduced neuroticism, are associated with increased SWB in a fixed-effects
regression with complete socio-economic controls.
There are two important variations to the SWB-unemployed relation. First, too much
SWB can be a bad thing. The SWB-unemployed relation is quadratic and the marginal effect of
increasing SWB on unemployment becomes positive (unemployment more likely) at
standardized SWB values greater than 0.83 (DOLS) and 1.50 (FE). Second, higher SWB has a
greater impact on getting hired than on staying employed. The SWB-unemployed relation is
three times larger for people who are currently unemployed. Obtaining the relation for employed
people is a new finding because previous studies have worked with samples of unemployed
people (e.g. Krause 2013).
This study identifies the effects of SWB net of interrelationships with traditional human
capital variables. These channels are captured when including a full range of controls in separate
dynamic and fixed-effects regressions. It is difficult to conceive of any omitted channels or
99
alternative sources of bias. Identification comes from time-varying shocks to SWB, excluding
the effects of individuals’ job satisfaction, job expectations, unemployment history, family
income, industry and occupation, social networks, and self-reported health. A placebo test is
used to demonstrate the results are not driven by systematic measurement error or omitted
variables specifically related to satisfaction type questions. Finally, Arellano and Bond (1991)
GMM regressions support the main results. AB regressions are used to simultaneously include
fixed-effect and dynamics, and to account for endogeneity. Most of the literature uses dynamic
regressions, some condition on a similarly rich set of observables, but there is limited use of
fixed effects.
Future research should include variables like SWB in analyzing dimensions of labor
performance. There is substantial support for the use of personality traits in labor equations (e.g.
Borghans et al. 2008; and Heckman and Kautz 2013), but when personality is not available,
SWB is demonstrated here as an alternative. Another implication is for policy makers. Policies
focused on character skill development will not only improve well-being, but also labor market
performance.

100
Chapter 5 Summary and Conclusions

The goal of social policy is to improve societal well-being. The focus of this dissertation is on
democratic development, as indirectly important for well-being, and second, subjective well-
being (SWB). Quality institutions are necessary to implement SWB-improving policies, and we
also need to know what can be done to improve SWB. SWB measures are valuable because they
include economic and non-economic factors that are otherwise difficult to measure. Moreover
greater SWB is associated with improved outcomes that economists are traditionally interested
in. Chapter 4 demonstrates that greater SWB leads to a lower likelihood of being unemployed.
In Chapter 2, two coauthors and I contribute to the political variant of the resource curse
debate generally, and especially in relation to two prominent papers (Haber and Menaldo 2011;
Andersen and Ross 2014). Proponents of the resource curse suggest it exists under certain
conditions, such as in the Middle East and North Africa, or in the period post 1980; however,
under these conditions we do not find a significant relation using the most up-to-date time-series
econometrics. We also estimate the relation for two groups comprised solely of producers,
mature and less-experienced producers; yet even there, we do not find evidence to support the
curse.
The results of Chapter 2 suggest increasing oil abundance need not hinder democratic
development based on a sample of 127 countries. Countries with increasing oil abundance face
challenges, such as increased revenue volatility or rent-seeking behavior, that may make policy
planning difficult or increase corruption. Yet the evidence suggests the challenges have been
manageable to the extent they did not hinder development of democratic institutions.
Consequently, increasing oil abundance does not likely hinder social policy.
101
Chapter 3 shows the Great Recession had a large effect on SWB, primarily operating
through reduced income and employment, and the effects differed substantially for different
demographic groups. The results also provide an example of why social policy should be
concerned with SWB. While young adults (18-24) are vulnerable to job loss, the SWB data
reveal they were not the greatest impacted. The foreign born faced a decline about three-times
larger than the full population and more than seven-times greater than young adults. The
unemployment rate increased by less than two percentage points for young adults and for the
foreign-born, by more than five percentage points. The SWB data suggest that while vulnerable
to job loss, young adults are better able to cope, perhaps by enrolling in school and or living with
their parents. Policymakers can learn from young-adults’ experience, and perhaps provide better
income support to undertake high-quality training programs. The results are based on
microeconometric regressions, including macro-controls, and data from the General Social
Survey for the years 1973-2014.
Chapter 4 establishes happier people are less likely to be unemployed using longitudinal
data in Germany for the years 1996-2013 from the German Socio-Economic Panel (GSOEP).
The relationship: has a meaning magnitude, is quadratic, has a larger magnitude for the
unemployed, cannot be explained by omitted fixed variables, nor is it likely to be explained by
reverse causality or measurement error. The panel structure controls for fixed characteristics and
unemployment dynamics. The GSOEP data also includes a large number of characteristics that
are used to further substantiate the main results, including family satisfaction (used in the
placebo test), labor market expectations (used to account, in part, for reverse causality), and the
Big-Five personality traits. The analysis demonstrates that personality changes over the period
2005-2009 in a way that is positively correlated with changes in SWB.
102
The results in Chapters 3 and 4 suggest the possibility of generating a virtuous cycle in
social policy. Improving SWB is associated with lower unemployment, which is in turn
positively associated with greater SWB. The cycle could be initiated through character skill
development in occupational and educational structures. The results in Chapter 4 demonstrate
that character skills, such as extraversion and emotional stability, are positively related to SWB.
Greater SWB is likely to improve employment outcomes for many reasons. However, I focus on
character skills (using personality as a measurable proxy) in part because there is complementary
research that shows character skills can be trained and that they exhibit a positive relationship
with labor market outcomes. Additional research should be conducted to understand which
occupational and educational structures are better for SWB and to develop better training
programs.
Social policy should be informed by SWB data. SWB measures reveal otherwise
unobserved outcomes. It was not anticipated that young adults would represent the success story
during the Great Recession, but SWB data indicate they were affected the least. Greater SWB
also leads to a lower likelihood of being unemployed. Policies aimed at improving SWB will
have the indirect effect of improving labor market outcomes as well. Last, the institutions
necessary to implement SWB-improving policies are not likely to be affected by increasing oil
abundance.

103
References

Acemoglu, Daron, and James A. Robinson. 2006. Economic Origins of Dictaitorship and
Democracy. Cambridge, Mass. U.S.: Cambridge University Press.
Ades, Alberto, and Rafael Di Tella. 1999. “Rents , Competition , and Corruption.” American
Economic Review 89 (4): 982–93.
Ahmadov, Anar K. 2014. “Oil, Democracy, and Context: A Meta-Analysis.” Comparative
Political Studies 47 (9): 1238–67. doi:10.1177/0010414013495358.
Andersen, Jørgen Juel, and Silje Aslaksen. 2008. “Constitutions and the Resource Curse.”
Journal of Development Economics 87 (2): 227–46. doi:10.1016/j.jdeveco.2007.12.005.
Andersen, Jørgen Juel, and Michael L. Ross. 2014. “The Big Oil Change: A Closer Look at the
Haber-Menaldo Analysis.” Comparative Political Studies 47 (7): 993–1021.
doi:10.1177/0010414013488557.
Anderson, T W, and Cheng Hsiao. 1981. “Estimation of Dynamic Models with Error
Components.” Journal of the American Statistical Association 76 (375): 598–606.
doi:10.2307/2287517.
Anger, Silke, and Daniel D Schnitzlein. 2016. “Cognitive Skills, Non-Cognitive Skills, and
Family Background: Evidence from Sibling Correlations.” SOEPpapers on
Multidisciplinary Panel Data Research 834 (March 24, 2016). doi:10.1007/s00148-003-
0163-0.
Angrist, J. D., and J. S. Pischke. 2009. Mostly Harmless Econometrics : An Empiricist’s
Companion. Princeton University Press. doi:10.1017/CBO9781107415324.004.
Arampatzi, Efstratia, Martijn J. Burger, and Ruut Veenhoven. 2015. “Financial Distress and
Happiness of Employees in Times of Economic Crisis.” Applied Economics Letters 22 (3):
173–79. doi:10.1080/13504851.2014.931916.
Arellano, Manuel, and Stephen Bond. 1991. “Some Tests of Specification for Panel Data: Monte
Carlo Evidence and an Application to Employment Equations.” The Review of Economic
Studies 58 (2): 277–97. doi:10.2307/2297968.
Ashenfelter, Orley. 1978. “Estimating the Effect of Training Programs on Earnings.” The Review
of Economics and Statistics 60 (1): 47–57.
Barro, Robert J. 1998. “The Interplay Between Political and Economic Development.” In
Determinants of Economic Growth, 49–88. Cambridge, Mass. U.S.: MIT Press.
Beblawi, H., and G. Luciani. 1987. “The Rentier State in the Arab World.” In Nation State and
Integration in the Arab World. London: Croom Helm.
Bell, David N F, and David G. Blanchflower. 2011. “Young People and the Great Recession.”
Oxford Review of Economic Policy 27 (2): 241–67. doi:10.1093/oxrep/grr011.
BenYishay, Ariel, and Roger Betancourt. 2014. “Unbundling Democracy: Political Rights and
Civil Liberties.” Journal of Comparative Economics 42. Association for Comparative
Economic Studies: 552–68. doi:10.1016/j.jce.2014.06.001.
Binder, Martin, and Alex Coad. 2010. “An Examination of the Dynamics of Well-Being and Life
Events Using Vector Autoregressions.” Journal of Economic Behavior and Organization 76
(2). Elsevier B.V.: 352–71. doi:10.1016/j.jebo.2010.06.006.
Bjørnskov, Christian. 2014. “Do Economic Reforms Alleviate Subjective Well-Being Losses of
Economic Crises?” Journal of Happiness Studies 15 (1): 163–82. doi:10.1007/s10902-013-
9442-y.
Blanchflower, David G., and Andrew J. Oswald. 2004. “Well-Being over Time in Britain and the
104
USA.” Journal of Public Economics 88 (7–8): 1359–86. doi:10.1016/S0047-
2727(02)00168-8.
Blanco, Luisa R., Jeffrey B. Nugent, and Kelsey J. O’Connor. 2015. “Oil Curse and Institutional
Changes: Which Institutions Are Most Vulnerable to the Curse and under What
Circumstances?” Contemporary Economic Policy 33 (2): 229–49. doi:10.1111/coep.12077.
Boehm, J. K., and S. Lyubomirsky. 2008. “Does Happiness Promote Career Success?” Journal
of Career Assessment 16 (1): 101–16. doi:10.1177/1069072707308140.
Borghans, Lex, Angela Lee Duckworth, James J Heckman, and Bas Weel. 2008. “The
Economics and Psychology of Personality Traits.” Journal of Human Resources 43 (4):
972–1059. http://www.jstor.org/stable/40057376.
Bove, Vincenzo, Leandro Elia, and Ron P Smith. 2015. “The Relationship Between Panel and
Synthetic Control Estimators of the Effect of Civil War.” Unpublished Working Paper.
Brollo, Fernanda, Tommaso Nannicini, Roberto Perotti, and Guido Tabellini. 2013. “The
Political Resource Curse.” American Economic Review 103 (5): 1759–96.
Cameron, Adrian Colin, and P. K. Trivedi. 2005. Microeconometrics: Methods and Applications.
New York: Cambridge University Press.
Carr, Ewan, and Heejung Chung. 2014. “Job Insecurity and Life Satisfaction : The Moderating
Influence of Labour Market Policies across Europe.” Journal of European Social Policy 24
(4): 383–99. doi:10.1177/0958928714538219.
Choi, In. 2001. “Unit Root Tests for Panel Data.” Journal of International Money and Finance
20 (2): 249–72. doi:10.1016/S0261-5606(00)00048-6.
Chudik, Alexander, Kamiar Mohaddes, M. Hashem Pesaran, and Mehdi Raissi. 2013. “Debt ,
Inflation and Growth: Robust Estimation of Long-Run E§ects in Dynamic Panel Data
Models.” Unpublished Working Paper.
Chudik, Alexander, and M. Hashem Pesaran. 2013. “Common Correlated Effects Estimation of
Heterogeneous Dynamic Panel Data Models with Weakly Exogenous Regressors.” Federal
Reserve Bank of Dallas Globalization and Monetary Policy Institute Working Paper No.
146.
Clark, Andrew E. 2001. “What Really Matters in a Job? Hedonic Measurement Using Quit
Data.” Labour Economics 8 (2): 223–42. doi:10.1016/S0927-5371(01)00031-8.
———. 2003. “Unemployment as a Social Norm: Psychological Evidence from Panel Data.”
Journal of Labor Economics 21 (2): 323–51. doi:10.1086/345560.
Clark, Andrew E., Ed Diener, Yannis Georgellis, and Richard E. Lucas. 2008. “Lags and Leads
in Life Satisfaction : A Test of the Baseline Hypothesis.” Economic Journal 118 (529):
F222–43.
Clark, Andrew E., Yannis Georgellis, and Peter Sanfey. 2001. “Scarring: The Psychological
Impact of Past Unemployment.” Economica 68 (270): 221–41. doi:10.1111/1468-
0335.00243.
Collier, Paul, and Anke Hoeffler. 2005. “Resource Rents, Governance, and Conflict.” Journal of
Conflict Resolution 49 (4): 625–33. doi:10.1177/0022002705277551.
———. 2009. “Testing the Neocon Agenda: Democracy in Resource-Rich Societies.” European
Economic Review 53: 293–308. doi:10.1016/j.euroecorev.2008.05.006.
Costa, Paul T., and Robert R. McCrae. 1994. “Set Like Plaster? Evidence for the Stability of
Adult Personality.” In Can Personality Change?, edited by Todd F. Heatherton and Joel L.
Weinberger, 21–40. Washington DC: American Psychological Association.
De Neve, Jan-Emmanuel, Ed Diener, Louis Tay, and Cody Xuereb. 2013. “The Objective
105
Benefits of Subjective Well-Being.” In World Happiness Report 2013, edited by John F.
Helliwell, Richard Layard, and Jeffrey Sachs, 56–79. New York: UN Sustainable
Development Solutions Network. http://worldhappiness.report.
De Neve, Jan-Emmanuel, and Andrew J. Oswald. 2012. “Estimating the Influence of Life
Satisfaction and Positive Affect on Later Income Using Sibling Fixed Effects.” Proceedings
of the National Academy of Sciences of the United States of America 109 (49): 19953–
19958. doi:10.1073/pnas.1211437109.
Deaton, Angus. 2011. “The Financial Crisis and the Well-Being of Americans.” Oxford
Economic Papers 64 (1): 1–26. doi:10.1093/oep/gpr051.
Di Tella, Rafael, Robert J. MacCulloch, and Andrew J. Oswald. 2001. “Preferences Over
Inflation and Unemployment: Evidence From Surveys of Happiness.” American Economic
Review 91 (1): 335–41.
———. 2003. “The Macroeconomics of Happiness.” The Review of Economics and Statistics 85
(4): 809–27. doi:10.1162/003465303772815745.
Diener, Ed, Louis Tay, and Shigehiro Oishi. 2013. “Rising Income and the Subjective Well-
Being of Nations.” Journal of Personality and Social Psychology 104 (2): 267–76.
doi:10.1037/a0030487.
Dolan, Paul, Tessa Peasgood, and Mathew White. 2008. “Do We Really Know What Makes Us
Happy? A Review of the Economic Literature on the Factors Associated with Subjective
Well-Being.” Journal of Economic Psychology 29 (1): 94–122.
doi:10.1016/j.joep.2007.09.001.
Dunning, Thad. 2008. Crude Democracy: Natural Resource Wealth and Political Regimes. New
York: Cambridge University Press.
Durand, Martine, and Conal Smith. 2013. “The OECD Approach to Measuring Subjective Well-
Being.” In World Happiness Report 2013, 112–37. New York: UN Sustainable
Development Solutions Network. http://worldhappiness.report.
Easterlin, Richard A., Laura Angelescu McVey, Malgorzata Switek, Onnicha Sawangfa, and
Jacqueline Smith Zweig. 2010. “The Happiness-Income Paradox Revisited.” Proceedings of
the National Academy of Sciences of the United States of America 107 (52): 22463–68.
doi:10.1073/pnas.1015962107.
Elbadawi, Ibrahim, and Raimundo Soto. 2016. “Resource Rents, Political Institutions and
Economic Growth.” In Understanding and Avoiding the Oil Curse in Resource-Rich Arab
Economies, edited by Ibrahim Elbadawi and Hoda Selim, 187–224. New York: Cambridge
University Press.
Engle, Robert F., and C. W. J. Granger. 1987. “Co-Integration and Error Correction :
Representation , Estimation , and Testing.” Econometrica 55 (2): 251–76.
Ferrer-i-Carbonell, Ada, and Paul Frijters. 2004. “How Important Is Methodology for the
Estimate of the Determinants of Hapiness?” The Economic Journal 114 (497): 641–59.
Freedom House. 2013. Country Status and Ratings, 1973-2013 (FINAL)_0.xls. Accessed May
2013, online: http://www.freedomhouse.org/report-types/freedom-world.
Geo-Help Inc. 2011. The World History of Oil and Gas. Accessed August 2011, online:
http://www.geohelp.net/world.html.
Gielen, Anne C., and Jan C. van Ours. 2014. “Unhappiness and Job Finding.” Economica 81
(323): 544–65. doi:10.1111/ecca.12089.
Graham, Carol, Soymya Chattopadhyay, and Mario Picon. 2010. “Adapting to Adversity:
Happiness and the 2009 Economic Crisis in the United States.” Social Research 77 (2):
106
715–48.
Graham, Carol, Andrew Eggers, and Sandip Sukhtankar. 2004. “Does Happiness Pay? An
Exploration Based on Panel Data from Russia.” Journal of Economic Behavior and
Organization 55 (3): 319–42. doi:10.1016/j.jebo.2003.09.002.
Guven, Cahit. 2011. “Are Happier People Better Citizens?” Kyklos 64 (2): 178–92.
doi:10.1111/j.1467-6435.2011.00501.x.
Haber, Sephen, and Victor Menaldo. 2011. “Do Natural Resources Fuel Authoritarianism? A
Preappraisal of the Resource Curse.” American Political Science Review 105 (1): 1–26.
doi:10.1017/S0003055410000584.
Heckman, James J., and Tim Kautz. 2013. “Fostering and Measuring Skills: Interventions That
Improve Character and Cognition.” NBER Working Paper Series 19656.
Heineck, Guido, and Silke Anger. 2010. “The Returns to Cognitive Abilities and Personality
Traits in Germany.” Labour Economics 17 (3). Elsevier B.V.: 535–46.
doi:10.1016/j.labeco.2009.06.001.
Helliwell, John F., Richard Layard, and Jeffrey Sachs. 2017. “World Happiness Report 2017.”
New York. http://worldhappiness.report/#happiness2017.
Helliwell, John F., and Shun Wang. 2012. “The State of World Happiness.” In World Happiness
Report, edited by John F. Helliwell, Richard Layard, and Jeffrey Sachs, 10–57. UN
Sustainable Development Solutions Network.
Henisz, Witold J. 2002. “The Institutional Environment for Infrastructure Investment.” Industrial
and Corporate Change 11 (2): 355–89. doi:10.1093/icc/11.2.355.
Heston, A., R. Summers, and B. Aten. 2012. Penn World Table, Version 7.1, Center for
International Comparisons of Production, Income and Prices at the University of
Pennsylvania.
Herb, Michael. 2005. “No Representation without Taxation? Rents, Development, and
Democracy.” Comparative Politics 37 (3): 297–316. doi:10.2307/20072891.
Herbst, Chris M. 2011. “‘Paradoxical’ Decline? Another Look at the Relative Reduction in
Female Happiness.” Journal of Economic Psychology 32 (5). Elsevier B.V.: 773–88.
doi:10.1016/j.joep.2011.07.001.
Herbst, Chris M., and John Ifcher. 2014. “The Increasing Happiness of Parents.” Unpublished
Working Paper, Arizona State University.
Hertog, Steffen. 2007. “Shaping the Saudi State: Human Agency’S Shifting Role in Rentier-
State Formation.” International Journal of Middle East Studies 39 (4): 539–63.
doi:10.1017/S0020743807071073.
Huntington, S. P. 1991. The Third Wave: Democratization in the Late Twentieth Century.
Norman: University of Oklahoma Press.
Ifcher, John, and Homa Zarghamee. 2011. “Happiness and Time Preference: The Effect of
Positive Affect in a Random-Assignment Experiment.” American Economic Review 101
(7): 3109–29. doi:10.1257/aer.101.7.3109.
———. 2014. “The Happiness of Single Mothers: Evidence from the General Social Survey.”
Journal of Happiness Studies 15 (5): 1219–38. doi:10.1007/s10902-013-9472-5.
James, Alexander. 2015. “The Resource Curse: A Statistical Mirage?” Journal of Development
Economics 114. Elsevier B.V.: 55–63. doi:10.1016/j.jdeveco.2014.10.006.
Kahneman, Daniel, and Amos Tversky. 1979. “Prospect Theory : An Analysis of Decision under
Risk.” Econometrica 47 (2): 263–92.
Kalleberg, Arne L., and Till M. von Wachter. 2017. “The U.S . Labor Market During and After
107
the Great Recession: Continuities and Transformations.” RSF: The Russell Sage Foundation
Journal of the Social Sciences 3 (3): 1–19.
Kapetanios, George, and Massimiliano Marcellino. 2010. “Factor-GMM Estimation with Large
Sets of Possibly Weak Instruments.” Computational Statistics and Data Analysis 54 (11).
Elsevier B.V.: 2655–75. doi:10.1016/j.csda.2010.04.008.
Kapteyn, Arie, Jinkook Lee, Caroline Tassot, Hana Vonkova, and Gema Zamarro. 2015.
“Dimensions of Subjective Well-Being.” Social Indicators Research 123 (3): 625–60.
doi:10.1007/s11205-014-0753-0.
Krause, Annabelle. 2013. “Don’t Worry, Be Happy? Happiness and Reemployment.” Journal of
Economic Behavior and Organization 96. Elsevier B.V.: 1–20.
doi:10.1016/j.jebo.2013.09.002.
———. 2014. “Happiness and Work.” IZA Discussion Paper 8435.
Kroh, Martin. 2011. “Documentation of Sample Sizes and Panel Attrition in the German Socio
Economic Panel (SOEP)(1984 until 2010).” Data Documentation 59. Berliin.
http://panel.gsoep.de/soep-docs/surveypapers/diw_ssp0177.pdf.
Layard, Richard, Andrew E. Clark, and Claudia Senik. 2012. “The Causes of Happiness and
Misery.” In World Happiness Report, edited by John F. Helliwell, Richard Layard, and
Jeffrey Sachs, 58–89. UN Sustainable Development Solutions Network.
http://worldhappiness.report.
Liu, Bette, Sarah Floud, Kirstin Pirie, Jane Green, Richard Peto, and Valerie Beral. 2016. “Does
Happiness Itself Directly Affect Mortality? The Prospective UK Million Women Study.”
The Lancet 387 (10021): 874–81. doi:10.1016/S0140-6736(15)01087-9.
Lucas, Robert E. 1987. Models of Business Cycles. Oxford [Oxfordshire]; New York: B.
Blackwell.
Lyubomirsky, Sonja, Chris Tkach, and M. Robin DiMatteo. 2006. “What Are the Differences
between Happiness and Self-Esteem?” Social Indicators Research 78 (3): 363–404.
doi:10.1007/s11205-005-0213-y.
Mahdavy, H. 1970. “The Patterns and Problems of Economic Development in Rentier States:
The Case of Iran.” In Studies in Economic History of the Middle East, 427–67. London:
Oxford University Press.
Marks, Gary N., and Nicole Fleming. 1999. “Influences and Consequences of Well-Being among
Australian Young People: 1980-1995.” Social Indicators Research 46 (3): 301–23.
doi:10.1023/A:1006928507272.
Mavridis, Dimitris. 2012. “Can Subjective Well-Being Predict Unemployment Duration?”
Unpublished Working Paper, Paris School of Economics.
Menaldo, Victor. 2013. “The Institutions Curse: The Theory and Evidence of Oil in Weak
States.”
Mincer, Jacob. 1974. Schooling, Experience, and Earnings. New York: National Bureau of
Economic Research.
Mohanty, Madhu S. 2009a. “Effects of Positive Attitude on Earnings: Evidence from the US
Longitudinal Data.” Journal of Socio-Economics 38 (2): 357–71.
doi:10.1016/j.socec.2008.07.012.
———. 2009b. “Effects of Positive Attitude on Happiness and Wage: Evidence from the US
Data.” Journal of Economic Psychology 30 (6). Elsevier B.V.: 884–97.
doi:10.1016/j.joep.2009.08.010.
Mohanty, Madhu S., and Aman Ullah. 2012. “Direct and Indirect Effects of Happiness on Wage:
108
A Simultaneous Equations Approach.” Journal of Socio-Economics 41. Elsevier Inc.: 143–
52. doi:10.1016/j.socec.2011.12.004.
Mohtadi, Hamid, Michael L. Ross, and Stefan Ruediger. 2015. “Do Natural Resources Inhibit
Tranparency?” Working Paper No. 906. Economic Research Forum.
Morgan, Robson. 2015. “Labor Market Policy and Subjective Well-Being during the Great
Recession.” Unpublished Working Paper, University of Southern California.
Mueller, Gerrit, and Erik Plug. 2006. “Estimating the Effect of Personality on Male and Female
Earnings.” Industrial and Labor Relations Review 60 (1): 3–22.
National Bureau of Economic Research (NBER). 2014. US Business Cycle Expansions and
Contractions, Accessed: December 2014, online:
http://www.nber.org/cycles/cyclesmain.html
National Opinion Research Center (NORC). 2015. GSS 1972-2014 Cross-Sectional Cumulative
Data (Release 2, April 17, 2015). General Social Survey: National Opinion Research
Center. University of Chicago.
Nickell, Stephen. 1981. “Biases in Dynamic Models with Fixed Effects.” Econometrica 49 (6):
1417–26.
Nijman, Theo, and Marno Verbeek. 1992. “Nonresponse in Panel Data : The Impact on
Estimates of a Life Cycle Consumption Function.” Journal of Applied Econometrics 7 (3):
243–57.
Organization for Economic Co-operation and Development (OECD). 2013. OECD Guidelines on
Measuring Subjective Well-Being. OECD Publishing. doi:10.1787/9789264191655-en.
Organization for Economic Co-operation and Development (OECD). 2015. OECD-Note-
Equivalence Scales. Accessed: May 2015, online: http://www.oecd.org/eco/growth/OECD-
Note-EquivalenceScales.pdf
Oishi, Shigehiro, Selin Kesebir, and Ed Diener. 2011. “Income Inequality and Happiness.”
Psychological Science 22 (9): 1095–1100. doi:10.1177/0956797611417262.
Omgba, Luc Desire. 2015. “Why Do Some Oil-Producing Countries Succeed in Democracy
While Others Fail?” World Development 76: 180–89. doi:10.1016/j.worlddev.2015.07.005.
Oswald, Andrew J., Eugenio Proto, and Daniel Sgroi. 2015. “Happiness and Productivity.”
Journal of Labor Economics 33 (4): 789–822. doi:10.1086/298035.
Pesaran, M. Hashem. 2015. Time Series and Panel Data Econometrics. Oxford: Oxford
University Press.
Polcon. 2012. Political Constraint Index Dataset. Version: polcon2012.dta. Accessed May 2013,
online: http://mgmt5.wharton.upenn.edu/henisz/POLCON/ContactInfo.html
Polity. 2011. Polity IV Annual Time-Series, 1800-2011.” Version: p4v2011, Polity IV Project,
Political Regime Characteristics and Transitions. Accessed May 2013, online:
http://www.systemicpeace.org/inscr/inscr.htm
Przeworksi, A., M. E. Alvarez, J. A. Cheibub, and F. Limongi. 2000. Democracy and
Development: Political Institutions and Well-Being in the World, 1850-1990. New York:
Cambridge University Press.
Rayo, Luis, and Gary S. Becker. 2007. “Habits, Peers, and Happiness: An Evolutionary
Perspective.” American Economic Review 97 (2): 487–91. doi:10.1257/aer.97.2.487.
Roberts, Brent W, Avshalom Caspi, and Terrie E Moffitt. 2003. “Work Experiences and
Personality Development in Young Adulthood.” Journal of Personality and Social
Psychology 84 (3): 582–93. doi:10.1037/0022-3514.84.3.582.
Robinson, James A., Ragnar Torvik, and Thierry Verdier. 2006. “Political Foundations of the
109
Resource Curse.” Journal of Development Economics 79: 447–68.
doi:10.1016/j.jdeveco.2013.09.004.
Roodman, David. 2009. “How to Do xtabond2: An Introduction to Difference and System GMM
in Stata.” Stata Journal 9 (1): 86–136. doi:The Stata Journal.
Ross, Michael L. 2001. “Does Oil Hinder Democracy?” World Politics 53 (April): 325–61.
———. 2012. The Oil Curse: How Petroleum Wealth Shapes the Development of Nations.
Princeton: Princeton University Press.
Ross, M., and P. Mahdavi. 2015. Oil and Gas Data, 1932-2014. Version 2. Accessed March
2016, online: http://dx.doi.org/10.7910/DVN/ZTPW0Y, Harvard Dataverse.
Sacks, Daniel W., Betsey Stevenson, and Justin Wolfers. 2012. “The New Stylized Facts about
Income and The New Stylized Facts about Income and Subjective Well-Being.” IZA
Discussion Paper, no. 7105.
Selim, Hoda, and Chahir Zaki. 2016. “The Institutional Curse of Natural Resources in the Arab
World.” In Understanding and Avoiding the Oil Curse in Resource-Rich Arab Economies,
edited by Ibrahim Elbadawi and Hoda Selim, 322–72. New York: Cambridge University
Press.
Sinnott, Emily, John Nash, and Augusto De La Torre. 2010. Natural Resources in Latin America
and the Carribbean: Beyond Booms and Busts? Washington DC: World Bank.
Smith, Brock. 2015. “The Resource Curse Exorcised: Evidence from a Panel of Countries.”
Journal of Development Economics 116. Elsevier B.V.: 57–73.
doi:10.1016/j.jdeveco.2015.04.001.
Soto, Christopher J. 2015. “Is Happiness Good for Your Personality? Concurrent and
Prospective Relations of the Big Five With Subjective Well-Being.” Journal of Personality
83 (1): 45–55. doi:10.1111/jopy.12081.
Specht, J., B. Egloff, and S. C. Schmukle. 2012. “Examining Mechanisms of Personality
Maturation: The Impact of Life Satisfaction on the Development of the Big Five Personality
Traits.” Social Psychological and Personality Science 4 (2): 181–89.
doi:10.1177/1948550612448197.
Srivastava, Sanjay, Oliver P. John, Samuel D. Gosling, and Jeff Potter. 2003. “Development of
Personality in Early and Middle Adulthood: Set like Plaster or Persistent Change?” Journal
of Personality and Social Psychology 84 (5): 1041–53. doi:10.1037/0022-3514.84.5.1041.
Stevenson, Betsey, and Justin Wolfers. 2008a. “Economic Growth and Subjective Well-Being:
Reassessing the Easterlin Paradox.” Brookings Papers on Economic Activity 2008 (1): 1–
87. doi:10.1353/eca.0.0001.
———. 2008b. “Happiness Inequality in the United States.” The Journal of Legal Studies 37
(S2): S33–79. doi:10.1086/592004.
———. 2009. “The Paradox of Declining Female Happiness.” Economic Policy 1 (2): 190–225.
doi:10.1257/pol.1.2.190.
———. 2012. “Subjective and Objective Indicators of Racial Progress Subjective and Objective
Indicators of Racial Progress.” Journal of Legal Studies 41 (2): 459–93.
doi:10.1007/BF00300445.
Stiglitz, Joseph E., Amartya Sen, and Jean-Paul Fitoussi. 2009. “Report by the Commission on
the Measurement of Economic Performance and Social Progress.” Paris.
http://www.stiglitzsen-fitoussi.fr/en/index.htm.
Tay, Louis, Lauren Kuykendall, and Ed Diener. 2015. “Satisfaction and Happiness – The Bright
Side of Quality of Life Louis.” In Global Handbook of Quality of Life: Exploration of Well-
110
Being of Nations and Continents, edited by Wolfgang Glatzer, Laura Camfield, Valerie
MØller, and Mariano Rojas, 839–53. Springer Netherlands. doi:10.1007/978-94-017-9178-
6_39.
Torvik, Ragnar. 2002. “Natural Resources, Rent Seeking and Welfare.” Journal of Development
Economics 67 (2): 455–70. doi:10.1016/S0304-3878(01)00195-X.
Tsui, Kevin K. 2011. “More Oil, Less Democracy: Theory and Evidence from Crude Oil
Discoveries.” The Economic Journal 121 (551): 89–115. doi:10.1111/j.1468-
0297.2009.02327.x.
U.S. Bureau of Economic Analysis (BEA). 2015a. Table 1.1.6 Real Gross Domestic Product,
Chained Dollars: Annual and Seasonally Adjusted Quarterly. Accessed August 2015.
——. 2015b. Regional Economic Accounts: Current Gross Domestic Product by State (NAICS
and SIC), and Population. Accessed July 2015, online:
http://www.bea.gov/regional/downloadzip.cfm
U.S. Bureau of the Census (Census). 2014. Table F-4. Gini Indexes for Families, by Race and
Hispanic Origin of Householder: 1947 to 2013. Current Population Survey, Annual Social
and Economic Supplements.
——. 2015. Median and Average Sales Price of Houses Sold by Region. New Residential Sales
Historical Data. Accessed: May 2015 online:
https://www.census.gov/construction/nrs/pdf/pricerega.pdf
U.S. Bureau of Labor Statistics (BLS). 2015a. Annual Unemployment Rate: Not seasonally
adjusted for population 16 years and over. Accessed July 2015, online:
http://data.bls.gov/timeseries/LNU04000000?years_option=all_years&periods_option=spec
ific_periods&periods=Annual+Data
——. 2015b. Monthly Unemployment Rate: Seasonally adjusted for population 16 years and
over. Accessed August 2015, online: http://data.bls.gov/timeseries/LNS14000000
——. 2015c. Annual Unemployment Rate: by Census Division. Local Area Unemployment
Statistics: Accessed July 2015, online: http://www.bls.gov/lau/#data
Vicente, Pedro C. 2010. “Does Oil Corrupt? Evidence from a Natural Experiment in West
Africa.” Journal of Development Economics 92 (1): 28–38.
doi:10.1016/j.jdeveco.2009.01.005.
Westerlund, Joakim. 2007. “Testing for Error Correction in Panel Data.” Oxford Bulletin of
Economics and Statistics 69 (6): 709–48. doi:10.1111/j.1468-0084.2007.00477.x.
Winkelmann, Liliana, and Rainer Winkelmann. 1998. “Why Are the Unemployed So Unhappy?
Evidence from Panel Data.” Economica 65 (257): 1–15. doi:10.1111/1468-0335.00111.
Wolfers, Justin. 2003. “Is Business Cycle Volatility Costly? Evidence from Surveys of
Subjective Well-Being.” International Finance 6 (1): 1–26. doi:10.1111/1468-2362.00112.
Wooldridge, Jeffrey M. 2010. Econometric Analysis of Cross Section And Panel Data. 2nded.
Cambridge, Mass. U.S.: MIT Press.
World Bank. 1995. Handbook of World Development Indicators, United Nations Projections and
Perspectives Studies Branch, Department of Economic and Social Information and Policy
Analysis.
World Bank. 2011. "The Wealth of Nations Dataset” in: The Changing Wealth of Nations, The
World Bank. Accessed May 2013, online: http://data.worldbank.org/data-catalog/wealth-of-
nations.
World Bank. 2013. World Development Indicators. Accessed April 2013, online:
http://data.worldbank.org/data-catalog/world-development-indicators.
111
World Bank. 2015. World Development Indicators. Retrieved on July 2015, online:
http://data.worldbank.org/data-catalog/world-development-indicators.
Wulfgramm, Melike. 2014. “Life Satisfaction Effects of Unemployment in Europe: The
Moderating Influence of Labour Market Policy.” Journal of European Social Policy 24 (3):
258–72. doi:10.1177/0958928714525817.

112

Appendix A Supplementary Material for Chapter 2

Appendix Table A-1 Summary Statistics

Polity Polcon CL PR
Group Countries 1980 2005 1980 2005 1980 2005 1980 2005
Full Sample 127 -1.65 3.83 0.15 0.26 3.64 4.65 3.63 4.51
Mature 41 -1.05 3.73 0.17 0.26 3.90 4.56 3.93 4.49
Less-Experience 43 -1.56 3.81 0.17 0.28 3.63 4.86 3.58 4.72
Non-Producer 43 -2.33 3.93 0.10 0.23 3.40 4.53 3.40 4.33
Latin America 23 -0.57 7.17 0.17 0.30 4.09 5.13 4.00 5.22
MENA 18 -6.56 -3.61 0.04 0.16 2.78 2.89 2.78 2.56
ROW 86 -0.92 4.49 0.17 0.27 3.70 4.90 3.71 4.73
Oil & Gas Value pc GDP pc
Group Countries 1980 2005 1980 2005
Full Sample 127 2007 1438 5084 7456
Mature 41 5875 3278 8656 10476
Less-Experience 43 327 1122 5034 8971
Non-Producer 43 0 0 1728 3061
Latin America 23 656 700 2876 3527
MENA 18 11912 6165 10156 9472
ROW 86 296 647 4612 8085
Note: Oil was discovered before 1959 in Mature countries.
Source: Freedom House (2013), Geo-Help Inc. (2011), Heston et al. (2012), Polcon
(2012), Polity (2011), Ross and Mahdavi (2015), World Bank (1995; 2013).

113
Appendix Table A-2 Oil & Gas Value’s Impact on Polity by Country Group CS-EC Mean
Group Regressions

(1)
Full Sample
(2)
Mature
(3)
Low-Exp
(4)
Latin America
(5)
MENA
(6)
ROW
Lag Polity
Lag Δ Polity
Lag2 Δ Polity
Lag ln(O&G Value pc)
Δ ln(O&G Value pc)
LagΔ ln(O&G Value pc)
Lag2Δ ln(O&G Value pc)
Polity Mean
Lag Polity Mean
Lag2 Polity Mean
Lag3 Polity Mean
ln(O&G) mean
Lag ln(O&G) mean
Lag2 ln(O&G) mean
Lag3 ln(O&G) mean
-0.499***
[-12.203]
0.158***
[6.194]
0.038+
[1.719]
-0.020
[-0.196]
-0.155
[-1.637]
-0.148
[-1.013]
-0.172
[-1.559]
1.040***
[4.878]
-0.451
[-1.638]
0.077
[0.322]
0.126
[0.791]
0.202
[0.996]
-0.089
[-0.327]
0.151
[0.644]
-0.605**
[-2.756]
-0.279***
[-4.023]
0.030
[0.487]
-0.002
[-0.036]
0.046
[0.093]
-1.048+
[-1.868]
-0.487
[-0.853]
-0.678
[-1.032]
0.822***
[3.396]
-0.467
[-1.521]
-0.159
[-0.615]
0.426+
[1.907]
1.029*
[2.068]
-0.689
[-1.025]
0.217
[0.469]
-0.903
[-1.349]
-0.489***
[-8.285]
0.189***
[4.621]
0.071+
[1.869]
-0.031
[-0.389]
-0.057
[-0.428]
-0.146
[-0.682]
-0.094
[-1.146]
0.951***
[3.452]
-0.527
[-1.562]
0.176
[0.473]
0.142
[0.584]
0.065
[0.414]
-0.102
[-0.402]
0.158
[0.537]
-0.251
[-1.283]
-0.454***
[-5.063]
0.102
[1.194]
0.054
[0.660]
-0.718
[-1.609]
-0.542*
[-1.995]
-0.536
[-0.634]
-0.553
[-0.958]
1.064***
[3.906]
-0.539+
[-1.701]
-0.032
[-0.103]
0.111
[0.548]
0.728
[0.923]
1.107
[0.584]
0.199
[0.276]
-0.968
[-0.855]
-0.421***
[-4.056]
0.148+
[1.867]
0.193**
[2.820]
-0.257
[-0.887]
-0.109
[-0.362]
0.056
[0.259]
0.095
[0.941]
0.998*
[1.999]
-0.711
[-1.326]
0.048
[0.170]
0.451+
[1.880]
0.194
[0.619]
0.021
[0.061]
0.103
[0.330]
0.205
[0.562]
-0.461***
[-9.056]
0.114***
[4.674]
-0.021
[-0.919]
-0.112
[-1.123]
-0.233*
[-2.015]
-0.103
[-0.624]
-0.108
[-1.355]
0.973***
[4.346]
-0.481
[-1.389]
0.215
[0.685]
-0.001
[-0.005]
0.272
[1.166]
-0.164
[-0.634]
0.100
[0.387]
-0.378*
[-2.023]
Constant
Long-Run Effect
Observations
-1.295
[-1.512]
-0.040
[-0.196]
4438
0.812
[0.336]
0.164
[0.093]
1434
-1.449*
[-2.524]
-0.063
[-0.390]
1500
4.500
[1.375]
-1.582
[-1.469]
805
-0.853
[-0.420]
-0.609
[-0.957]
629
-1.645*
[-2.028]
-0.243
[-1.125]
3004
Periods - Avg. 35 35 35 35 35 35
Countries 127 41 43 23 18 86
Note: Mean group allows heterogeneous short- and long-run relations by country.
CS-ECM – cross-sectionally augmented error correction model (Equation 3 in the
text). Oil & Gas value per capita are used in natural log form. Polity scaled from -10
(autocracy) to 10 (democracy). Group composition presented in Table A8.
t statistics in brackets; * p<0.05 ** p<.01 *** p<.001
Source: Geo-Help Inc. (2011), Heston et al. (2012), Polity (2011), Ross and Mahdavi
(2015), World Bank (1995; 2013).

114
Appendix Table A-3 Oil & Gas Value’s Impact on Polcon by Country Group. CS-EC Mean
Group Regressions

(1)
Full Sample
(2)
Mature
(3)
Low-Exp
(4)
Latin America
(5)
MENA
(6)
ROW
Lag Polcon
Lag Δ Polcon
Lag2 Δ Polcon
Lag ln(O&G Value pc)
Δ ln(O&G Value pc)
LagΔ ln(O&G Value pc)
Lag2Δ ln(O&G Value pc)
Polcon Mean
Lag Polcon Mean
Lag2 Polcon Mean
Lag3 Polcon Mean
ln(O&G) mean
Lag ln(O&G) mean
Lag2 ln(O&G) mean
Lag3 ln(O&G) mean
-0.572***
[-13.796]
0.165***
[7.031]
0.087***
[4.521]
0.002
[0.323]
0.002
[0.327]
-0.000
[-0.032]
0.009
[1.075]
0.950***
[4.501]
-0.376
[-1.566]
0.194
[0.822]
-0.093
[-0.532]
0.004
[0.342]
0.004
[0.301]
-0.008
[-0.528]
-0.004
[-0.307]
-0.523***
[-10.275]
0.182***
[4.981]
0.124***
[4.240]
0.015
[0.637]
0.034
[1.243]
0.053
[1.503]
0.006
[0.267]
1.098***
[4.352]
-0.497+
[-1.747]
-0.127
[-0.404]
0.205
[0.895]
-0.032
[-0.919]
-0.034
[-0.819]
0.044
[0.969]
-0.009
[-0.350]
-0.640***
[-6.854]
0.179***
[4.253]
0.112*
[2.548]
-0.001
[-0.067]
-0.014*
[-2.162]
-0.012+
[-1.765]
0.001
[0.073]
0.998***
[4.170]
-0.526*
[-2.358]
0.271
[1.015]
-0.016
[-0.069]
0.003
[0.314]
0.004
[0.296]
-0.005
[-0.285]
-0.003
[-0.242]
-0.594***
[-10.917]
0.198***
[6.520]
0.137***
[4.072]
-0.017
[-0.381]
-0.014
[-0.343]
0.012
[0.684]
0.029
[1.302]
1.048***
[3.778]
-0.485+
[-1.836]
-0.156
[-0.540]
0.223
[1.013]
0.020
[0.224]
-0.006
[-0.116]
-0.041
[-0.794]
0.038
[1.019]
-0.362***
[-4.471]
0.170***
[3.817]
0.145***
[3.394]
0.000
[0.006]
0.009
[0.734]
-0.011
[-0.808]
0.004
[0.564]
0.893**
[2.937]
-0.705**
[-2.996]
0.082
[0.452]
0.180
[1.070]
-0.009
[-0.517]
0.027
[1.302]
-0.023
[-0.976]
0.003
[0.267]
-0.611***
[-10.965]
0.167***
[5.688]
0.120***
[4.749]
-0.004
[-0.513]
0.006
[0.932]
0.006
[0.970]
0.013
[1.582]
1.023***
[4.697]
-0.511*
[-1.996]
0.164
[0.698]
-0.039
[-0.204]
-0.003
[-0.219]
0.005
[0.329]
-0.002
[-0.126]
0.005
[0.355]
Constant
Long-Run Effect
Observations
-0.028
[-0.569]
0.003
[0.323]
4557
0.075
[0.491]
0.029
[0.632]
1475
-0.006
[-0.085]
-0.001
[-0.067]
1537
0.156
[0.406]
-0.029
[-0.385]
828
-0.008
[-0.111]
0.000
[0.006]
643
0.029
[0.434]
-0.006
[-0.511]
3086
Periods - Avg. 36 36 36 36 36 36
Countries 127 41 43 23 18 86
Note: Mean group allows heterogeneous short- and long-run relations by country.
CS-ECM – cross-sectionally augmented error correction model (Equation 3 in the
text). Oil & Gas value per capita are used in natural log form. Polcon scaled 0-1, with
increasing values indicating greater constraints. Group composition presented in
Table A8.
t statistics in brackets; * p<0.05 ** p<.01 *** p<.001
Source: Geo-Help Inc. (2011), Heston et al. (2012), Polcon (2012), Ross and
Mahdavi (2015), World Bank (1995; 2013).

115
Appendix Table A-4 Oil & Gas Value’s Impact on Civil Liberties by Country Group CS-
EC Mean Group Regressions

(1)
Full Sample
(2)
Mature
(3)
Low-Exp
(4)
Latin America
(5)
MENA
(6)
ROW
Lag Political Rights
Lag Δ Political Rights
Lag2 Δ Political Rights
Lag ln(O&G Value pc)
Δ ln(O&G Value pc)
LagΔ ln(O&G Value pc)
Lag2Δ ln(O&G Value pc)
PR Mean
Lag PR Mean
Lag2 PR Mean
Lag3 PR Mean
ln(O&G) mean
Lag ln(O&G) mean
Lag2 ln(O&G) mean
Lag3 ln(O&G) mean
-0.451***
[-15.656]
0.154***
[7.221]
0.114***
[5.661]
-0.038
[-0.853]
0.012
[0.311]
-0.032
[-0.867]
0.028
[0.706]
0.837***
[4.415]
-0.607*
[-2.472]
0.024
[0.095]
0.168
[0.924]
0.004
[0.050]
0.090
[0.984]
-0.067
[-0.782]
-0.014
[-0.190]
-0.488***
[-9.563]
0.190***
[5.970]
0.070*
[2.288]
0.034
[0.140]
0.050
[0.270]
-0.249
[-1.199]
0.201
[1.521]
1.111***
[4.268]
-0.549*
[-2.021]
0.199
[0.717]
-0.085
[-0.365]
-0.181
[-1.029]
0.254
[1.345]
-0.447+
[-1.949]
0.217
[1.491]
-0.538***
[-8.249]
0.181***
[4.149]
0.152***
[3.520]
-0.015
[-0.253]
0.002
[0.059]
-0.047
[-0.822]
-0.026
[-0.421]
0.971***
[3.602]
-0.629+
[-1.940]
0.311
[1.218]
-0.002
[-0.008]
-0.016
[-0.263]
0.006
[0.078]
0.001
[0.010]
-0.008
[-0.138]
-0.494***
[-7.071]
0.051
[1.038]
0.036
[0.823]
-0.252+
[-1.884]
-0.125
[-0.647]
0.158
[0.963]
0.095
[0.870]
1.000***
[3.736]
-0.469
[-1.599]
0.112
[0.755]
-0.107
[-0.491]
0.134
[0.368]
-0.213
[-0.551]
0.056
[0.266]
0.158
[0.653]
-0.768***
[-6.575]
0.288**
[3.075]
0.183**
[2.822]
0.024
[0.198]
0.110
[1.572]
-0.087
[-0.708]
0.043
[0.341]
0.886***
[3.629]
-0.433*
[-2.389]
0.163
[0.739]
0.246
[1.224]
-0.120
[-1.146]
0.114
[0.729]
-0.118
[-0.533]
0.064
[0.374]
-0.477***
[-11.486]
0.135***
[5.206]
0.091***
[3.888]
-0.040
[-1.044]
-0.004
[-0.161]
-0.001
[-0.025]
0.008
[0.214]
0.915***
[4.836]
-0.530*
[-2.457]
0.028
[0.105]
0.149
[0.804]
0.015
[0.182]
0.016
[0.175]
0.007
[0.104]
-0.022
[-0.311]
Constant
Long-Run Effect
Observations
-0.077
[-0.116]
-0.083
[-0.860]
4566
-0.252
[-0.205]
0.070
[0.140]
1475
-0.993
[-1.230]
-0.029
[-0.254]
1544
1.671
[0.925]
-0.510+
[-1.837]
828
-0.430
[-0.442]
0.032
[0.198]
648
-0.658
[-0.964]
-0.085
[-1.047]
3090
Periods - Avg. 36 36 36 36 36 36
Countries 127 41 43 23 18 86
Note: Oil & Gas value per capita are used in natural log form. Civil liberties are
scaled from 1 to 7 with more 7 indicating greater civil liberties. Group composition
presented in Table A8.
t statistics in brackets; * p<0.05 ** p<.01 *** p<.001
Source: Freedom House (2013), Geo-Help Inc. (2011), Heston et al. (2012), Ross and
Mahdavi (2015), World Bank (1995; 2013).

116
Appendix Table A-5 Oil & Gas Value’s Impact on Political Rights by Country Group CS-
EC Mean Group Regressions

(1)
Full Sample
(2)
Mature
(3)
Low-Exp
(4)
Latin America
(5)
MENA
(6)
ROW
Lag Political Rights
Lag Δ Political Rights
Lag ln(O&G Value pc)
Δ ln(O&G Value pc)
LagΔ ln(O&G Value pc)
PR Mean
Lag PR Mean
Lag2 PR Mean
Lag3 PR Mean
ln(O&G) mean
Lag ln(O&G) mean
Lag2 ln(O&G) mean
Lag3 ln(O&G) mean
-0.390***
[-16.179]
0.119***
[6.758]
0.002
[0.050]
0.024
[0.661]
-0.056+
[-1.711]
0.893***
[4.918]
-0.612*
[-2.505]
0.110
[0.440]
0.032
[0.182]
-0.024
[-0.343]
0.085
[0.938]
-0.080
[-1.040]
-0.025
[-0.430]
-0.427***
[-9.525]
0.154***
[5.246]
0.045
[0.222]
0.010
[0.060]
-0.359+
[-1.877]
1.049***
[4.130]
-0.549*
[-2.098]
0.227
[0.816]
-0.175
[-0.755]
-0.102
[-0.621]
0.315+
[1.730]
-0.366+
[-1.758]
0.004
[0.073]
-0.439***
[-8.610]
0.138***
[4.276]
-0.016
[-0.457]
0.021
[0.774]
-0.049
[-1.506]
0.973***
[4.026]
-0.683*
[-2.267]
0.428+
[1.728]
-0.223
[-1.299]
-0.034
[-0.646]
0.061
[0.781]
-0.046
[-0.611]
0.013
[0.228]
-0.467***
[-8.228]
0.041
[1.025]
-0.188+
[-1.700]
-0.164
[-0.941]
0.076
[0.656]
1.045***
[4.035]
-0.480+
[-1.696]
0.071
[0.421]
-0.117
[-0.565]
0.244
[0.702]
-0.194
[-0.600]
0.094
[0.473]
-0.037
[-0.252]
-0.623***
[-7.185]
0.210***
[3.737]
0.052
[0.677]
0.166*
[2.202]
-0.110
[-1.059]
0.876***
[3.672]
-0.570*
[-2.472]
0.248
[1.001]
0.077
[0.484]
-0.125
[-1.167]
0.175
[1.062]
-0.151
[-0.775]
0.046
[0.436]
-0.427***
[-12.722]
0.105***
[4.944]
-0.007
[-0.246]
-0.001
[-0.038]
-0.029
[-0.859]
0.922***
[4.851]
-0.521*
[-2.446]
0.074
[0.280]
0.065
[0.359]
0.030
[0.411]
0.021
[0.234]
-0.039
[-0.559]
-0.030
[-0.482]
Constant
Long-Run Effect
Observations
-0.476
[-0.797]
0.005
[0.050]
4571
-0.077
[-0.071]
0.104
[0.223]
1476
-0.624
[-0.983]
-0.037
[-0.456]
1547
1.139
[0.742]
-0.402+
[-1.692]
828
-0.234
[-0.294]
0.083
[0.693]
648
-0.863
[-1.528]
-0.017
[-0.246]
3095
Periods - Avg. 36 36 36 36 36 36
Countries 127 41 43 23 18 86
Note: Oil & Gas value per capita are used in natural log form. Political Rights are
scaled from 1 to 7 with 7 indicating greater political rights. Group composition
presented in Table A8.
t statistics in brackets; * p<0.05 ** p<.01 *** p<.001
Source: Freedom House (2013), Geo-Help Inc. (2011), Heston et al. (2012), Ross and
Mahdavi (2015), World Bank (1995; 2013)

117
Appendix Table A-6 Test for Integration Order of Polity, Polcon Civil Liberties, Political
Rights, and Oil & Gas Value

Variable Test Statistic (p-values) Full Sample Mature Low-Exp. LAC MENA ROW
O&G Value Chi-Squared P 1.000 1.000 0.106 0.997 1.000 1.000
Inverse Logit L 1.000 1.000 0.601 0.978 1.000 0.959
Inverse Normal Z 1.000 1.000 0.758 0.992 1.000 0.966
Modified Inverse Chi-Sq. 1.000 1.000 0.101 0.988 0.998 1.000
Unit Root Support Yes Yes Yes Yes Yes Yes
Δ O&G Value Chi-Squared P 0.000 0.004 0.000 0.898 0.014 0.000
Inverse Logit L 0.000 0.000 0.000 0.035 0.004 0.000
Inverse Normal Z 0.000 0.000 0.000 0.030 0.002 0.000
Modified Inverse Chi-Sq. 0.000 0.001 0.000 0.888 0.007 0.000
Stationary in Diffs Yes Yes Yes Yes/No Yes Yes
Variable Test Statistic (p-values) Full Sample Mature Low-Exp. LAC MENA ROW
Polity Chi-Squared P 0.112 0.999 0.001 0.000 0.914 0.969
Inverse Logit L 0.027 0.637 0.029 0.000 0.818 0.481
Inverse Normal Z 0.394 0.625 0.729 0.023 0.801 0.650
Modified Inverse Chi-Sq. 0.110 0.996 0.000 0.000 0.901 0.962
Unit Root Support Yes Yes Low No Yes Yes
Δ Polity Chi-Squared P 0.000 0.075 0.000 0.000 0.041 0.000
Inverse Logit L 0.000 0.000 0.000 0.000 0.001 0.000
Inverse Normal Z 0.000 0.000 0.000 0.000 0.001 0.000
Modified Inverse Chi-Sq. 0.000 0.068 0.000 0.000 0.029 0.000
Stationary in Diffs Yes Yes Yes Yes Yes Yes
Variable Test Statistic (p-values) Full Sample Mature Low-Exp. LAC MENA ROW
Polcon Chi-Squared P 0.000 0.834 0.001 0.053 0.999 0.000
Inverse Logit L 0.011 0.554 0.009 0.118 0.625 0.013
Inverse Normal Z 0.312 0.577 0.363 0.155 0.643 0.428
Modified Inverse Chi-Sq. 0.000 0.834 0.000 0.042 0.993 0.000
Unit Root Support Low Yes Low Yes Yes Low
Δ Polcon Chi-Squared P 0.000 0.000 0.000 0.000 0.251 0.000
Inverse Logit L 0.000 0.000 0.000 0.000 0.002 0.000
Inverse Normal Z 0.000 0.000 0.000 0.000 0.001 0.000
Modified Inverse Chi-Sq. 0.000 0.000 0.000 0.000 0.268 0.000
Stationary in Diffs Yes Yes Yes Yes Yes/No Yes
Variable Test Statistic (p-values) Full Sample Mature Low-Exp. LAC MENA ROW
CL Chi-Squared P 0.063 0.584 0.018 0.000 0.217 0.982
Inverse Logit L 0.045 0.343 0.020 0.000 0.426 0.574
Inverse Normal Z 0.186 0.327 0.235 0.015 0.393 0.591
Modified Inverse Chi-Sq. 0.058 0.603 0.012 0.000 0.229 0.976
Unit Root Support Low Yes Low No Yes Yes
Δ CL Chi-Squared P 0.000 0.000 0.000 0.000 0.000 0.000
Inverse Logit L 0.000 0.000 0.000 0.000 0.000 0.000
Inverse Normal Z 0.000 0.000 0.000 0.000 0.000 0.000
Modified Inverse Chi-Sq. 0.000 0.000 0.000 0.000 0.000 0.000
Stationary in Diffs Yes Yes Yes Yes Yes Yes
Variable Test Statistic (p-values) Full Sample Mature Low-Exp. LAC MENA ROW
PR Chi-Squared P 0.000 0.988 0.000 0.000 0.887 0.000
Inverse Logit L 0.000 0.903 0.000 0.000 0.943 0.005
Inverse Normal Z 0.064 0.904 0.024 0.001 0.939 0.155
Modified Inverse Chi-Sq. 0.000 0.980 0.000 0.000 0.877 0.000
Unit Root Support No Yes No No Yes No
Δ PR Chi-Squared P 0.000 0.000 0.000 0.000 0.000 0.000
Inverse Logit L 0.000 0.000 0.000 0.000 0.000 0.000
Inverse Normal Z 0.000 0.000 0.000 0.000 0.000 0.000
Modified Inverse Chi-Sq. 0.000 0.000 0.000 0.000 0.000 0.000
Stationary in Diffs Yes Yes Yes Yes Yes Yes
Note: Dickey Fuller, Fisher style panel unit root tests (Choi 2001). The null hypothesis is that all panels contain a unit
root. Oil & Gas value per capita are used in natural log form.
Source: Freedom House (2013), Geo-Help Inc. (2011), Heston et al. (2012), Ross and Mahdavi (2015), World Bank
(1995; 2013)

118
Appendix Table A-7 Cointegration Results, Westerlund Panel Tests. Restricted Sample - 52
Countries

Polity Full Sample Mature Low-Exp. LAC MENA ROW
Group Mean Test t 0.54 0.70 0.22 0.98 0.14 0.26
Group Mean Test a 0.16 0.74 0.08 1.00 0.24 0.12
Panel Test t 0.28 0.34 0.46 0.86 0.32 0.28
Panel Test a 0.26 0.24 0.26 0.68 0.24 0.22
Cointegration Support No No Some No No No
Polcon Full Sample Mature Low-Exp. LAC MENA ROW
Group Mean Test t 0.12 0.10 0.46 0.64 0.42 0.06
Group Mean Test a 0.20 0.06 0.22 0.46 0.46 0.10
Panel Test t 0.06 0.16 0.08 0.34 0.34 0.06
Panel Test a 0.04 0.10 0.06 0.28 0.32 0.06
Cointegration Support Yes Some Yes No No Yes
Civil Liberties Full Sample Mature Low-Exp. LAC MENA ROW
Group Mean Test t 0.04 0.04 0.12 0.00 0.14 0.30
Group Mean Test a 0.08 0.04 0.12 0.10 0.00 0.28
Panel Test t 0.10 0.02 0.18 0.08 0.32 0.26
Panel Test a 0.04 0.00 0.22 0.06 0.04 0.28
Cointegration Support Yes Yes No Yes Yes No
Political Rights Full Sample Mature Low-Exp. LAC MENA ROW
Group Mean Test t 0.00 0.04 0.04 0.20 0.02 0.02
Group Mean Test a 0.04 0.06 0.08 0.30 0.02 0.22
Panel Test t 0.12 0.04 0.26 0.18 0.06 0.24
Panel Test a 0.04 0.02 0.02 0.14 0.00 0.04
Cointegration Support Yes Yes Yes No Yes Yes
Note: Group tests, rejecting H0 suggests that there is cointegration between Oil & Gas Value
the relevant institutional variable for at least one country.
Panel tests, rejecting HO suggests that the panel as a whole is cointegrated.
Critical values were bootstrapped to allow for cross-sectional dependence.
Source: Freedom House (2013), Geo-Help Inc. (2011), Heston et al. (2012), Ross and Mahdavi
(2015), World Bank (1995; 2013).

119
Appendix Table A-8 List of Countries and Group Composition

Mature Producers LAC MENA ROW Less-Exp. Producers LAC MENA ROW
Albania X
Algeria X
Angola X
Argentina X
Australia X
Austria X
Bahrain X
Bolivia X
Brazil X
Canada X
Colombia X
Congo, Rep. X
Congo, Dem. Rep. X
Cuba X
Egypt, Arab Rep. X
Gabon X
Hungary X
India X
Indonesia X
Iran, Islamic Rep. X
Italy X
Japan X
Kuwait X
Libya X
Malaysia X
Mexico X
Netherlands X
Nigeria X
Peru X
Poland X
Qatar X
Romania X
Saudi Arabia X
Syrian Arab Republic X
Thailand X
Trinidad and Tobago X
Turkey X
United Arab Emirates X
United Kingdom X
United States X
Venezuela, RB X
Total Mature 9 11 21
Bangladesh X
Belgium X
Benin X
Bulgaria X
Cameroon X
Chad X
Chile X
China X
Cote d'Ivoire X
Denmark X
Ecuador X
Equatorial Guinea X
France X
Ghana X
Greece X
Guatemala X
Ireland X
Israel X
Jordan X
Korea, Rep. X
Mauritania X
Mongolia X
Morocco X
Mozambique X
New Zealand X
Niger X
Norway X
Oman X
Pakistan X
Papua New Guinea X
Philippines X
Rwanda X
Senegal X
South Africa X
Spain X
Sudan X
Sweden X
Switzerland X
Taiwan X
Tanzania X
Tunisia X
Vietnam X
Yemen X
Total Non-Mature 3 7 33
Source: Author Calculations, Geo-Help Inc. (2011), Ross and Mahdavi (2015).
120
Appendix Table A-8 Continued List of Countries and Group Composition

Non-Producers LAC MENA ROW Totals LAC MENA ROW
Bhutan X
Botswana X
Burkina Faso X
Burundi X
Central African Republic X
Comoro Is. X
Costa Rica X
Cyprus X
Dominican Republic X
El Salvador X
Ethiopia X
Fiji X
Finland X
Gambia, The X
Guinea X
Guinea-Bissau X
Guyana X
Haiti X
Honduras X
Jamaica X
Kenya X
Lao PDR X
Lesotho X
Liberia X
Madagascar X
Malawi X
Mali X
Mauritius X
Nepal X
Nicaragua X
Panama X
Paraguay X
Portugal X
Sierra Leone X
Singapore X
Somalia X
Sri Lanka X
Swaziland X
Togo X
Uganda X
Uruguay X
Zambia X
Zimbabwe X
Total Non-Producers 11 0 32
Mature 9 11 21
Less-Experience 3 7 33
Non-Producer 11 0 32
Total 23 18 86
Source: Author Calculations, Geo-Help Inc. (2011), Ross and Mahdavi (2015).

121
Appendix B Supplementary Material for Chapter 3

General Social Survey Sample Weights and Restrictions.—The following samples were
dropped: the African American oversample in the 1982 and 1987 GSS; surveys that were
conducted in Spanish (and could not have been completed in English); the 1972 and 1985
surveys (because the question preceding happiness changed); and observations from split-ballot
experiments that were conducted in 1980, 1986, and 1987. The GSS weight WTSSALL was
applied to ensure samples approximated the national population. This strategy was employed by
Herbst and Ifcher (2014). [Paraphrased from Footnote 2 on page 5 of Herbst and Ifcher (2014)].

Appendix Table B-1 Population Shares by Demographic Group, General Social Survey

1977 2014
Group Obs. Pop. Share Obs. Pop. Share
Women 1530 53% 2446 54%
Black 1530 11% 2446 15%
Youth 1524 15% 2436 9%
Foreign-Born 1529 7% 2446 13%
Hispanic na na 2438 14%
High School and Less 1526 84% 2446 63%
Married 1530 69% 2443 52%
Parent 1517 74% 2440 71%

Source: Author calculations from NORC, General Social Survey (Release 2, April 17, 2015)

Appendix Table B-2 Summary Statistics for Key Micro Variables
1977 2014
Variable Obs Mean Std. Dev. Obs Mean Std. Dev.
Happy (1-3 scale) 1528 2.25 0.64 2438 2.21 0.63
Age (yrs) 1524

16.86 2436 47.46 17.39
F-Educ. (yrs) 1092 9.05 4.18 1830 12.05 4.06
M-Educ. (yrs) 1249 9.58 3.66 2198 11.86 3.75
Income
a
(2000$s) 1372 25623 19968 2206 32559 27965

a. Income is measured as total family income per equivalent household size. Specifically: household
income (GSS variable coninc) divided by equivalent household size (GSS household composition and
OECD-modified equivalence scale, OECD 2015).
Source: Author calculations from NORC, General Social Survey (Release 2, April 17, 2015)

122
Appendix Table B-3 Self-Reported Income and Unemployment as a Percentage of Total
Population, by Population Group, 2006 – 2012.
Panel A Full Sample

Men Women

Unempl. Income Unempl. Income Unempl. Income
2006 3.3% 33776 3.6% 36475 3.1% 31370
2008 3.4% 33826

4.7% 36653 2.3% 31163
2010 7.3% 29501

10.0% 31471 5.1% 27798
2012 5.2% 32474

6.5% 35194 4.1% 30122
Abs. Change (2010-2006) 4.0% -4275 6.4% -5004 2.0% -3572
Percent Change (10-06) 119.4% -12.7%

179.9% -13.7% 62.8% -11.4%

Panel B Black nonBlack Youth nonYouth

Unempl. Income Unempl. Income Unempl. Income Unempl. Income
2006 5.5% 24150 3.0% 35203 6.5% 23707 2.9% 34922
2008 8.2% 21495 2.7% 35799 6.5% 20927 3.0% 35430
2010 9.3% 21138 6.9% 30950 8.3% 17653 7.1% 30809
2012 8.6% 21244 4.6% 34487 8.3% 22297 4.8% 33740
Abs. Change (2010-2006) 3.8% -3013 4.0% -4253 1.8% -6054 4.3% -4113
Percent Change (10-06) 68.7% -12.5% 133.1% -12.1% 28.5% -25.5% 148.9% -11.8%

Panel C Native Foreign Hispanic nonHispanic

Unempl. Income Unempl. Income Unempl. Income Unempl. Income
2006 3.5% 34086 1.3% 33146 6.5% 26862 3.0% 34467
2008 3.6% 33795 1.5% 34118 2.9% 25027 3.5% 34796
2010 7.4% 29981 6.5% 26196 10.5% 16385 6.8% 31378
2012 5.5% 32420 2.8% 33194 6.7% 25289 5.0% 33451
Abs. Change (2010-2006) 3.9% -4105 5.2% -6950 4.0% -10477 3.8% -3089
Percent Change (10-06) 110.9% -12.0% 395.5% -21.0% 62.4% -39.0% 126.0% -9.0%

Source: Author calculations from NORC, General Social Survey (Release 2, April 17, 2015)
Notes: Income is total real family income (from all sources) adjusted for household size. The unemployment
population share is the unemployed share of the total population, not the labor force.

123
Appendix Table B-4 Key Macroeconomic Variables and Their Sources

Value

Variable Unit 1977 2014 Aggregation Source
Real GDP pc
a
2000$s 24428 45873 By Census
Division
Bureau of Economic Analysis: Regional
Economic Accounts
Unemployment
Rate
% of labor
force
7.03 6.18 By Census
Division
Bureau of Labor Statistics: Local Area
Unemployment Statistics
Inflation Rate
(CPI)
% 6.49 1.62 U.S. Aggregate World Development Indicators
Social
Expenditure
b

% of GDP 12.80 18.70 U.S. Aggregate OECD Social Expenditure Database
(SOCX)
Median House
Price
2000$s 127765 240226 By Census Region U.S. Census Bureau New Residential Sales
Historical Data: Sales Price by Houses
Sold
Gini
Coefficient
c

na 0.36 0.45 U.S. Aggregate U.S. Census Bureau, Current Population
Survey, Annual Social and Economic
Supplements. Table F-4. Gini Indexes for
Families, by Race and Hispanic Origin of
Householder: 1947 to 2013

Past Recession
d
Dummy 0 0 U.S. Aggregate NBER Recession Dating Committee
a. NAICs Current GDP per capita by state was extended back using SIC GDP per capita by state. GDP
was then aggregated and adjusted for population by census division, and then inflation.
b. 1977 Social Expenditure value is for 1980. Data is available beginning in 1980.
c. 2014 Gini Coefficient is for 2013. 2014 is not yet available.
d. Coded: 1 for the years: 1974, 1975, 1980-1983, 1990, 1991, 2001, and 2002.
Source: Listed in table.
124
Appendix Table B-5 2010 Deviation from Long-Term Trend. Change in Probability of
Reporting “Very Happy”, Full Sample, 1973 – 2014
(1) (2) (3) (4) (5) (6)
Year 2010 -0.046*** -0.029*** -0.001 -0.044*** -0.001 -0.015

[-12.245] [-5.401] [-0.320] [-14.013] [-0.323] [-1.456]
Observations 32945 32945 30227 32945 30227 21191

Source: Author’s calculations based on data reported in text.
Notes: Marginal effects correspond to 𝜆
!
from the ordered probit regressions, detailed below. Column 1
replicates the base model (i.e. table 2, column 1). Column 2 includes the main effect for employment status
and its interaction with 2008 and 2010 (excludes income). Column 3 includes the main effect for income and
its interaction with 2008 and 2010 (excludes employment status). Column 4 uses a cubic trend, but no
additional control variables. Column 5 uses the cubic trend with the main effect for income and interaction.
Column 6 repeats column 3, but uses real personal income as opposed to adjusted family income.

All models include the control variables: woman, black, youth, age, age squared, ten-year birth-cohort, and
mother's and father's education. The estimated regression (specified in OLS) is: ℎ𝑎𝑝𝑝𝑦
!"
= 𝛼
!
+ 𝜷
!
𝒙
!"
+
𝛿𝑝𝑟𝑒𝑐𝑒𝑠
!
+ 𝜆
!
𝑡𝑟𝑒𝑛𝑑
!
+𝜆
!
𝑑
!"
+ 𝜆
!
𝑑
!"
+𝝀
𝟑
′𝒄
!"
𝑑
!"
+𝝀
𝟒
′𝒄
!"
𝑑
!"
+ 𝜀
!"
. ℎ𝑎𝑝𝑝𝑦
!"
is reported happiness for
individual i in year t; 𝒙
!"
is a vector of individual characteristics; 𝑝𝑟𝑒𝑐𝑒𝑠
!
is a dummy variable for past
recessions; 𝑡𝑟𝑒𝑛𝑑
!
= 𝑦𝑒𝑎𝑟
!
−1972 and 𝑑
!
are dummy variables for the years 2008 and 2010. 𝒄
!"
is a vector
of individual variables or channels that may explain the Great Recession. They are also included in 𝒙
!"
.
t statistics in brackets (clustered by year); * p<.10 ** p<.05 *** p<.01.

125
Appendix Table B-6 Fixed-Effects Logit Regressions; GSS Panel 2006-2014; Dependent
Variable “Very Happy”, Main Effects by Year (Base = 2010).

(1) (2) (3) (4)
2006 0.357*** 0.339*** 0.433*** -0.779

[2.995] [2.826] [2.588] [-0.619]
2008 0.163* 0.144* 0.132 -0.096

[1.907] [1.685] [1.068] [-0.107]
2012 0.154* 0.144* 0.317*** 0.656

[1.883] [1.758] [2.764] [0.763]
2014 0.397*** 0.396*** 0.544*** 2.334**

[3.429] [3.394] [3.276] [2.284]
Employment &
Income no yes no no
Employment by Year no no yes no
Income by Year no no no yes

Source: Author’s calculations based on data reported in text.
Notes: Overlapping Panel – described in footnote 10. 1,195 individuals observed 3 times for a total of
3,585 observations. Column 2 adds controls for income (ln(Eqv. Inc.)) and employment status. Omitted
category is full-time employed. Column 3 interacts employment status with year, but excludes income
controls). Column 4 interacts income with year (excludes employment controls).
Reported year main effects correspond to 𝝀
𝝉
in the following specification (in OLS form for simplicity):
𝑉𝑒𝑟𝑦ℎ𝑎𝑝𝑝𝑦
!"
=𝛼
!
+𝜷
!
𝒙
!"
+ ∑ (𝝀
𝝉
+𝜸
𝝉
!
𝒙
𝒊𝒕
)𝑑
!
𝟐
𝝉!𝟏
+∑ (𝝀
𝝉
+𝜸
𝝉
!
𝒙
𝒊𝒕
)𝑑
!
𝟓
𝝉!𝟒
+ 𝜀
!"
. 𝑉𝑒𝑟𝑦ℎ𝑎𝑝𝑝𝑦
!"
takes
the value of 1 if individual i reports being “very happy” in year t; 𝒙
!"
is a vector of individual
characteristics. 𝑑
!
are dummy variables for each period (2006 = 1; 2008 = 2; … 2014=5).
t statistics in brackets (clustered by individual); * p<.10 ** p<.05 *** p<.01.

126
Appendix C Supplementary Material for Chapter 4

C.1 Sample Attrition
The Chapter 4 regression estimates are consistent even in the presence of attrition. This
conclusion is based on the results from a test discussed below.

Background and Theory:—In general, sample attrition is a concern when exploiting the panel
dimension of a data set. If people who leave the panel have different characteristics (e.g. more
likely to become unemployed), then attrition is considered selective, and the relationship
between happiness and unemployment could be biased. This problem could be an issue in the
GSOEP, less than 50 percent of the original 1984 sample remains after 2007 (Kroh 2011).
However, that does not mean attrition is selective.
45
To test for selection bias caused by attrition,
I used a test suggested by Wooldridge (2010).
Wooldridge (2010) shows that sample selection (e.g. from attrition) is only a problem in a
fixed-effects context when selection is related to the idosyncratic errors. More precisely, in
reference to Equation C-1 (𝑈
!"
=𝜸
!
𝑿
𝒊𝒕
+ 𝜂
!
+𝜈
!"
), let selection be represented by the indicator
𝑠
!"
, where 𝑠
!"
= 1 if 𝑿
𝒊𝒕
,𝑈
!"
is observed. For selection to not be a problem, 𝜈
!"
needs to be
mean independent of 𝑠
!"
, given (𝑿
𝒊
,𝜂
!
) for all possible t. If this condition is met, fixed-effects
estimation “on the unbalanced panel is consistent and asymptotically normal. (Wooldridge 2010,
830).”
Although the “assumption [above] may be a reasonable approximation, especially for
short panels (Wooldridge 2010, 830),” there is a simple test. Add 𝑠
!"!!
to Equation C-1, estimate

45
Sample selection due to attrition in the GSOEP data is not generally a concern. For example, the 2008
Economic Journal and 2014 Economica papers, which also uses GSOEP data, do not address attrition (Clark et al.
2008; Gielen and van Ours 2014).

127
the new equation using fixed effects on the unbalanced panel, and do a t test for the significance
of 𝑠
!"!!
. Under the null hypotheses, 𝜈
!"
is uncorrelated with 𝑠
!"
for all periods, and so selection in
the future period should not be significant in the equation at time t. This test is similar to what
was originally suggested by Nijman and Verbeek (1992), and the description is paraphrased from
Wooldridge (2010, 832).

Results:—The results presented in Appendix Table C-1 suggest selection is not a problem in the
GSOEP context. The above condition appears to be met – being present in the sample one period
ahead, “Lead Selection” (𝑠
!"!!
), is not correlated with being unemployed, conditional on the
covariates and fixed effects. Note that the test suggests sample selection is not a problem for the
unbalanced panel, yet under a slightly stronger assumption, “it is also valid to use a standard
fixed effects analyiss on any balanced subset of the unbalanced panel… (Wooldridge 2010,
831).” Given this result and the need for at least three periods (for fixed-effects and lagged
variables), the main analysis uses an unbalanced panel including everyone that is observed for at
least three consecutive periods.
128
Appendix Table C-1 Selective Attrition Fixed-Effects Regression. Germany 1996-2013
Unemployment Regressed on Lagged Determinants
a
and Future Selection

C.2 German Socio-Economic Panel Variables
Ln(Eqv. Inc): The natural log of total family income, per equivalent household size. Family
income is from all sources, before taxes, not conditional on employment, and adjusted for
inflation, which is then divided by equivalent household size using the OECD-modified
equivalence scale (OECD 2015).

Pot. Exp.: Potential Experience equals: age – years of education – five.

Tenure (yrs): “Length of time with Firm”

Unmpl. Hist: Total length of time a person has been unemployed (in years).

JobFind Difficult: If you were currently looking for a new job: Is it or would it be easy, difficult
or almost impossible to find an appropriate position? Scaled (1 – 3) with almost impossible
taking the value of three.

(1)
Selection Lead
Life Satisfaction
ln(Eqv. Inc.)
0.001
(0.002)
-0.005***
(0.000)
-0.003
(0.002)
Self Reported Health -0.005***
(0.001)
Constant
Observations
Adj. R Sq.
0.119***
(0.044)
235026
0.014
a. Year by state, age, and month of interview not lagged.
Notes: Selection lead = 1 if the employment status and determinants are
observed in period t+1. Additional controls include: (year by state, age, and
month of interview, education, marital status, parent dummy, child less than
5, potential experience and its square.
Standard errors in parentheses (clustered by individual).
Significance: * p<0.10 ** p<0.05 *** p<0.01

129
Job Satisfaction: How satisfied are you today with the following areas of your life? Job. Scaled
(0-10).

Health Good: Would you say your own health, in general, is excellent, good, fair, or poor?
Scaled (1-4) with excellent scored as a four.

Industry and Occupation Categories
1 Digit Industry Code, based on collapsing the GSOEP industry categorization into ten broad
categories in order to match the variable used in the Cross-national Equivalent File (Ohio State
University). The original GSOEP categorization is based on the Statistical Classification of
Economic Activities in the European Community, commonly referred to as NACE.

2 Digit Occupation Code, based on the International Labour Organization’s International
Standard Classification of Occupations, version ISCO-88.

Death of Partner: Has your family situation changed after [year end of previous calendar year]?
Spouse/partner has died. Coded 1 if yes.

Note dummies were created for missing values associated with income, health, marital status,
and job satisfaction/expectations. If income was missing for example, then Ln(Eqv. Inc) would
be set to zero and Inc. Missing would equal one.

Personality Traits

Optimism When you think about the future, are you – optimistic, more optimistic than
pessimistic, more pessimistic than optimisitc, or pessimistic? (scored 1- 4 with 4 representing
optimistic).

Questions are answered on 7-point Likert-type scales (1 – “disagree completely” to 7 – “agree
completely”), and then averaged within a construct to obtain a single construct score. Variable
definitions are from (Heineck and Anger 2010).

Big-Five Traits I see myself as someone who…
Openness is original, comes up with new ideas
to experience values artistic experiences
has an active imagination
Conscientiousness does a thorough job
does things effectively and efficiently
tends to be lazy (reversed)
Extraversion is communicative, talkative
is outgoing, sociable
is reserved (reversed)
Agreeableness is sometimes somewhat rude to others (reversed)
130
has a forgiving nature
is considerate and kind to others
Neuroticism worries a lot
gets nervous easily
is relaxed, handles stress well (reversed)

Additional Traits
Locus of Control 1. Compared to other people, I have not achieved what I deserve.
(External) 2. What a person achieves in life is above all a question of fate or
luck.

3. I frequently have the experience that other people have a
controlling influence over my life.

4. The opportunities that I have in life are determined by the social
conditions.
5. Inborn abilities are more important than any efforts one can
make.
6. I have little control over the things that happen in my life.
Reciprocity 1. If someone does me a favor, I am prepared to return it.
(Positive) 2. I go out of my way to help somebody who has been kind to me
before.

3. I am ready to undergo personal costs to help somebody who
helped me before.
Reciprocity
(Negative)
1. If I suffer a serious wrong, I will take revenge as soon as
possible, no matter what the cost.
2. If somebody puts me in a difficult position, I will do the same to
him/her.
3. If somebody offends me, I will offend him/her back.

Cogntive Ability (description repeated from SOEPpapers: Anger and Schnitzlein, 2016)

Symbol Corespondence Test
Coded as the correct number of entries in 90 seconds. (GSOEP Var: f99z90r).
Description: “The symbol correspondence test is conceptually related to the mechanics of
cognition or fluid intelligence and encompasses general abilities. It was developed after the
symbol digit modalities test (Smith, 1995) and involves asking respondents within 90 seconds to
assign with a keyboard as many correct numbers as possible to symbols, which are consecutively
displayed on a screen, while the correspondence list is permanently visible to them (Anger and
Schnitzlein 2016, 8).”

Word Fluency Test
Coded as the total number of animals mentioned in 90 seconds (GSOEP Var: f96t90g).
Description: “The word fluency test is conceptually related to the pragmatics of cognition or
crystallized intelligence. This test involves the fulfillment of specific tasks that improve in
accordance with previously acquired knowledge and skills. The word fluency test implemented
in the SOEP is based on the animal-naming task (Lindenberger and Baltes, 1995): respondents
131
name as many different animals as possible within 90 seconds. Whereas verbal fluency is based
on learning, speed of cognition is related to an individual’s innate abilities (Cattell, 1987) (Anger
and Schnitzlein 2016, 9).”

Abstract (if available)

Abstract How can social policy be used to improve societal well-being? First, quality institutions are necessary to implement optimal policies. Second societal well-being needs to be measured to determine optimal policies. Subjective well-being (SWB), or happiness, data are well suited for this purpose, because SWB measures allow individuals to weigh both economic and non-economic outcomes. For these reasons, Chapter 2 addresses the determinants of democratic institutions, while Chapters 3 and 4 discuss well-being directly. In particular, Chapter 2 challenges the so-called political resource curse, finding no robust long-term effects of oil abundance on Polity, civil liberties, political rights, or constraints on the executive. The chapter extends previous empirical work by using multiple measures of democracy, the most current time-series econometrics, and by allowing additional relationship heterogeneity, especially by geographic region and time period. Chapter 3 documents the year 2010 as the lowest level of recorded happiness in the U.S. since consistent measurement began in the 1970s. Disparate declines in income and employment, associated with the Great Recession, are shown to explain the disparate declines in SWB experienced by different demographic groups. Chapter 4 reverses the relationship direction by showing that people reporting higher SWB are less likely to be unemployed in Germany. The mechanisms include changes in the Big-Five personality traits. Within-person increases in extraversion or emotional stability, for example, are associated with increases in SWB. The results also show that the magnitude of the SWB-unemployed relation is substantially larger for people who are currently unemployed, and that too much SWB can also increase the likelihood of unemployment.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Four essays on how policy, the labor market, and age relate to subjective well-being

PDF

What leads to a happy life? Subjective well-being in Alaska, China, and Australia

PDF

Three essays on the causes and consequences of China’s governance reforms

PDF

Three essays on economics of early life health in developing countries

Asset Metadata

Creator O'Connor, Kelsey J. (author)

Core Title Three essays on social policy: institutional development, and subjective well-being as a cause and consequence of labor market outcomes

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Publication Date 06/05/2017

Defense Date 03/02/2017

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Big-Five personality traits,democracy,demographic groups,General Social Survey,German Socio Economic Panel,great recession,Happiness,life satisfaction,long-run relationships,natural resources,OAI-PMH Harvest,oil,resource curse,subjective well-being,Unemployment

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Easterlin, Richard (committee chair), Nugent, Jeffrey (committee member), Stone, Arthur (committee member)

Creator Email kelseyoc@gmail.com,kelseyoc@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-379897

Unique identifier UC11258078

Identifier etd-OConnorKel-5378.pdf (filename),usctheses-c40-379897 (legacy record id)

Legacy Identifier etd-OConnorKel-5378.pdf

Dmrecord 379897

Document Type Dissertation

Rights O'Connor, Kelsey J.

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA