University of Southern California Dissertations and Theses

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

(USC Thesis Other)

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

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content i FOUR ESSAYS ON HOW POLICY, THE LABOR MARKET, AND AGE RELATE TO SUBJECTIVE WELL-BEING by Robson Morgan 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, 2016 Copyright 2016 Robson Morgan ii For my Mother and Father iii ACKNOWLEDGEMENTS I would first like to thank, Richard A. Easterlin, for going above and beyond his responsibilities as my advisor. His mentorship, encouragement, and support have made my experience as a Ph.D. student rewarding and enjoyable. Whenever I am confronted with a challenge, I inevitably end up referring to something he taught me. I would also like to thank Jeffery Nugent for always having time to give me advice even before I was accepted to the Ph.D. program. I have also benefited greatly from working with Titus Galama, Juan Saavedra, Fei Wang, and Malgorzata Switek, and Kelsey O’Connor. Thank you to all my classmates for providing me with feedback on my research and for their support as my friends. Thank you to Stefano Bartolini and Francesco Sarracino for organizing the Tuscan Winter School. I gained an incredible happiness research network and memories I will always cherish from my experience in Tuscany. And thank you to Tani and Yuki for being my home away from Minnesota. Lastly, and most importantly, I would first like to thank my family for providing me with a life filled with infinite opportunity. Thank you to my grandparents, Yoji and Yuriko Hatsukami, for making sacrifices I cannot imagine for their children and grandchildren. Thank you to my grandmother, Elaine Morgan, for instilling a positive attitude in our family. And finally, I am grateful for the constant support from my wonderful parents, Steve Morgan and Dorothy Hatsukami. iv TABLE OF CONTENTS DEDICATION ....................................................................................................................... ii ACKNOWLEDGEMENTS .................................................................................................... iii LIST OF TABLES ................................................................................................................. v LIST OF FIGURES ............................................................................................................... vi ABSTRACT ........................................................................................................................ vii CHAPTER 1. INTRODUCTION ................................................................................................................. 1 CHAPTER 2. EXPERIENCED LIFE CYCLE SATISFACTION IN EUROPE .................................... 6 2.1 INTRODUCTION ....................................................................................................................................... 6 2.2 DATA AND METHODOLOGY ..................................................................................................................... 9 2.3 RESULTS ............................................................................................................................................... 13 2.3 CONCLUSION ......................................................................................................................................... 30 CHAPTER 3. WELL-BEING IN TRANSITION: LIFE SATISFACTION IN URBAN CHINA FROM 2002 TO 2012 ................................................................................................................................... 32 3.1 INTRODUCTION ..................................................................................................................................... 32 3.2. DATA ................................................................................................................................................... 35 3.3. METHODOLOGY ................................................................................................................................... 41 3.4. RESULTS .............................................................................................................................................. 43 3.5. HISTORICAL CONTEXT ......................................................................................................................... 52 3.6. CONCLUSION ........................................................................................................................................ 54 CHAPTER 4. LABOR MARKET POLICY AND SUBJECTIVE WELL-BEING DURING THE GREAT RECESSION ........................................................................................................................ 56 4.1. INTRODUCTION .................................................................................................................................... 56 4.2. BACKGROUND ...................................................................................................................................... 59 4.3. DATA ................................................................................................................................................... 61 4.4. METHODS ............................................................................................................................................. 64 4.5. EXPECTATIONS – A MODEL .................................................................................................................. 65 4.6. RESULTS .............................................................................................................................................. 69 4.7. CONCLUSION ........................................................................................................................................ 86 CHAPTER 5. THE IMPACT OF SOCIAL ASSISTANCE PROGRAMS ON SUBJECTIVE WELL-BEING: REGRESSION DISCONTINUITY EVIDENCE FROM A CASH TRANSFER PROGRAM IN COLOMBIA ..................................................................................................................... 88 5.1. INTRODUCTION .................................................................................................................................... 88 5.2. CONTEXT – AN OVERVIEW OF FAMILIAS EN ACCIÓN AND SISBEN ..................................................... 92 5.3. DATA AND DESCRIPTIVE STATISTICS ................................................................................................... 94 5.4. METHODS ............................................................................................................................................. 99 5.5. RESULTS ............................................................................................................................................ 112 5.6. CONCLUSION ...................................................................................................................................... 117 CHAPTER 6. SUMMARY AND CONCLUSIONS ................................................................................ 127 REFERENCES ......................................................................................................................................... 130 APPENDICIES .............................................................................................................................................. APPENDIX A: SUPPLEMENTARY MATERIAL FOR CHAPTER 2 ................................................................. 137 APPENDIX B: SUPPLEMENTARY MATERIAL FOR CHAPTER 3 ................................................................. 138 APPENDIX C: SUPPLEMENTARY MATERIAL FOR CHAPTER 4 ................................................................. 140 APPENDIX D. SUPPLEMENTARY MATERIAL FOR CHAPTER 5 ................................................................. 141 v LIST OF TABLES Table 2-1: Common life cycle life satisfaction (LS) pattern characteristics .................... 15 Table 2-2: Maximum and minimum life Satisfaction over the life cycle (5 year age group dummy variables, 1-4 scale) ................................................................................... 18 Table 2-3: Testing for the U-shape in life cycle satisfaction with age and age squared variables (life satisfaction the dependent variable, scale 1-4) .......................................... 22 Table 3-1. Step 1 OLS regression results .......................................................................... 44 Table 3-2. Contribution of the change in variable values to the life satisfaction increase between 2002 and 2012 ...................................................................................... 45 Table 3-3. Step 1 OLS regression results, division by level of education ........................ 50 Table 3-4. Contribution of the change in variable values to the life satisfaction increase between 2002 and 2012, by level of education ................................................... 51 Table 4-1. Marginal effects of policy variables on peak to trough change in life satisfaction by level of education, life satisfaction is the dependent variable (1-4 scale) 71 Table 4-2. Marginal effects of policy variables on peak to trough change in life satisfaction by cohort, life satisfaction is the dependent variable (1-4 scale) ................... 73 Table 4-3. Comparing marginal effects of NRR and ALMP on peak to trough change in life satisfaction, by level of education and by cohort .................................................. 74 Table 4-4. Marginal effects of EPL and EPL-T on peak to trough changes in life satisfaction by level of education, life satisfaction is the dependent variable (1-4 scale) 76 Table 4-5. Marginal effects of EPL and EPL-T on peak to trough changes in life satisfaction by cohort, life satisfaction is the dependent variable (1-4 scale) ................... 77 Table 4-6. Marginal effects of employment protection variables on peak to trough changes in reporting future job situation will be worse over next 12 months by level of education ....................................................................................................................... 78 Table 4-7. Marginal effects of employment protection variables on peak to trough changes in reporting future job situation will be worse over next 12 months by cohort .. 79 Table 4-8. Robustness tests: marginal effects of policy variables on peak to trough change in life satisfaction by level of education, life satisfaction is the dependent variable (1-4 scale) ............................................................................................................ 81 Table 4-9. Robustness tests: marginal effects of policy variables on peak to trough change in life satisfaction by cohort, life satisfaction is the dependent variable (1-4 scale) .......................................................................................................................... 82 Table 5-1: Household Characteristics, 2006 Census data compared to 2010 CNQLS data .................................................................................................................................... 96 Table 5-2: First Stage: Program Receipt as a Function of SISBEN Score and Eligibility, Doughnut Hole Between -1 and 2 ................................................................ 112 Table 5-3R: The Effect of FIA Receipt on Outcomes, Doughnut Hole Between -1 and 2, Life Satisfaction ............................................................................................. 115 Tables referenced in results section (Tables 5-3A to 5-3Q) ........................................... 118 vi LIST OF FIGURES Figure 2-1: Mean life cycle satisfaction pattern, 17 countries .......................................... 13 Figure 2-2: Life cycle satisfaction by country .................................................................. 17 Figure 2-3: Mean life cycle satisfaction patterns by gender, 17 countries ....................... 19 Figure 2-4: Mean life cycle satisfaction patterns by education, 17 countries ................... 20 Figure 2-5: Mean Swedish life cycle satisfaction, Eurobarometer and YAPS Patterns Compared ......................................................................................................................... 24 Figure 2-6: Mean German Life Cycle Satisfaction, Eurobarometer and GSOEP Patterns Compared ............................................................................................................ 25 Figure 2-7: Mean British Life Cycle Satisfaction, Eurobarometer and BHPS Patterns Compared .......................................................................................................................... 26 Figure 2-8: Percent of population retired compared to life satisfaction by age, 17 countries ............................................................................................................................ 28 Figure 2-9: Mean male percent married less female percent married by age, 17 countries ............................................................................................................................ 29 Figure 3-1. Average life satisfaction in urban China from 2002-2012 ............................. 38 Figure 3-2. Life satisfaction patterns by level of education from 2002 to 2012 ............... 49 Figure 4-1. Theoretical relationship between unemployment support programs and SWB during recession ....................................................................................................... 66 Figure 4-2. Theoretical relationship between employment protection legislation and SWB during recession ....................................................................................................... 67 Figure 4-3. Mean changes in life satisfaction from peak to trough of Great Recession, by group ............................................................................................................................ 69 Figure 4-4. Robustness of results to omitting individual countries, groups by level of education ........................................................................................................................... 84 Figure 4-5. Robustness of results to omitting individual countries, groups by cohort ..... 85 Figure 5-1. Density of Households over SISBEN Scores, 2010 CNQLS Compared to Official 2006 Colombian Census Data. .......................................................................... 102 Figure 5-2: Baseline and 2010 FIA Participation Discontinuity Around Eligibility Cutoff .............................................................................................................................. 104 Figure 5-3. Life Satisfaction over SISBEN score ........................................................... 116 vii ABSTRACT The studies in this dissertation provide new knowledge on how life circumstance and policy relate to subjective well-being. The findings from the studies either directly evaluate policy, or provide direction for future policy. Chapter 2 estimates the average pattern of subjective well-being across the life cycle for 17 European countries using Eurobarometer data. The average pattern of experienced life cycle satisfaction for the 17 countries resembles a wave like M-shape. Other features that are shared among the majority of countries are male life satisfaction improving relative to female life satisfaction as people age, and more educated people reporting higher levels of life satisfaction throughout the entire life cycle. Although these characteristics are shared by the majority of countries, there is no uniform life satisfaction pattern shared by all countries. Chapter 3 uses annual cross-sectional data to identify what changes in peoples’ lives are driving the upward trend in subjective well-being observed in urban China from 2002 to 2012. The analysis finds that improvements in the labor market, mostly driven by a large drop in unemployment, account for 29% of the increase in subjective well-being. Increases in income, after accounting for income comparison and habituation, are not significantly related to the change in subjective well-being. Chapter 4 uses policy variation across 23 European countries to evaluate the effectiveness of four labor market policies to mitigate the negative impact of the Great Recession on subjective well- being. Policies that provide support to people who become unemployed significantly mitigated the negative effect of the Great Recession, where policies that limited the ability of firms to freely adjust their labor force significantly exacerbated the effect of the recession. Chapter 5 uses a regression discontinuity approach to evaluate the impact of a Colombian conditional cash transfer program. The program positively affected many aspects of life – improvements in self reported viii health, increased income, increased expenditure on food, goods and services, and increased frequency of formal employment. These improvements lead to higher satisfaction with food, income, work, and the ability to help others. Subjective well-being was also positively affected by program participation, and although the effect is not robustly statistically significant, the result is consistent with the improvements in other aspects of life due to participation in the conditional cash transfer program. 1 CHAPTER 1. Introduction How can policy help promote and protect well-being? Four original research papers are presented in this dissertation that provide new insights into answering this question. The topics covered include: 1) well-being over the life cycle, 2) well-being during the transition to a free labor market in urban China, 3) labor market policy and well-being during the Great Recession in Europe, and 4) the effects of a social assistance program in Colombia on well-being. In all papers, subjective well-being is used as the primary measure of well-being. In this dissertation, the term “subjective well-being” is used to refer to self reported evaluations of a person’s happiness or satisfaction with life 1 . In both of these self-report measures, a survey respondent is asked to consider their life as a whole, and then report their feelings about happiness or life satisfaction on a numerical scale. The measure is therefore a self-defined, comprehensive measure of well-being that reflects both the life circumstances of the survey respondent and their perception of their life circumstances. There are two major advantages to using subjective well-being data when evaluating the well-being effects of a policy that I will highlight here 2 . First, the researcher does not need to make assumptions about the factors that contribute to a person’s well-being. In other words, instead of researchers assuming they know what makes a person better off and measuring those aspects of life as a proxy for well-being, subjective well-being researchers simply ask people how they feel about their well-being. Some researchers might argue that it is possible to identify the preferences of people by observing their choices. In situations where individuals have choice, this 1 These measures are considered comparable because they correlate with the same explanatory variables (Helliwell et al. 2012) 2 For a more comprehensive review of the benefits of subjective well-being policy, see the World Happiness Report 2012 (Helliwell et al. 2012). 2 may be a valid approach. But, in the case of policy, individual choice is limited because the policy choice is imposed by the government, and not directly chosen by the individual. A second advantage of using subjective well-being data stems from the comprehensive nature of the measure. Most policies rarely affect only one aspect of a person’s life, and often have psychological affects that may be hard to capture using objective measures. A policy that makes the firing of workers difficult will be used as an example to illustrate this point. While people who are employed might benefit from the extra job security, finding another job if they become unemployed or want to transfer companies may be more difficult. This is because if it is harder for companies to fire workers, the implicit cost of hiring workers is greater, and thus companies will be more reluctant to hire workers. How can a researcher take into account these policy implications when evaluating the effects on well-being? Using subjective well-being is one solution. The multidimensional impact of the policy is translated to a single dimensional measure, subjective well-being, by the people who are impacted by the policy. The advantages of using subjective well-being data to evaluate policy are moot, however, if the findings from subjective well-being research fall on policymakers’ deaf ears. Recent trends in both the official measurement of subjective well-being and interest from policymakers have put this worry to rest. In 2008, a panel assembled by the then French President Nicolas Sarkozy, which included six Nobel Laureates in economics, recommended that governments start officially collecting subjective well-being data (Stiglitz, Sen, Fitoussi 2009). In 2011, the United Nations passed a resolution that invited member countries to start officially collecting subjective well- being data and use this measure as a guide to inform policy (Helliwell et al. 2013). At present day, almost all Organization for Economic Co-operation and Development (OECD) countries are 3 collecting subjective well-being data and “many national leaders are talking about the importance of well-being as a guide for their nations” (Helliwell et al. 2015). The growing interest in using subjective well-being to guide policy motivates more research in this area. The four studies presented in this dissertation provide findings that contribute knowledge on the effectiveness of policy to protect and promote well-being, and also identify segments of the population where policy intervention can be particularly useful. Chapter 2 addresses the question of what happens to subjective well-being as people progress through the life cycle. What is the pattern of subjective well-being across the life cycle, is it different in different countries, and does it vary for different groups of people? The findings from the analysis are useful for targeting policy and future research aimed at people during ages of low well-being. The findings from this study are based on a non-parametric analysis using repeated cross-sectional Eurobarometer data spanning 1973 to 2010. The analysis includes controls for cohort effects and other characteristics that are fixed throughout the life cycle. Controls for non-fixed characteristics are not included because the aim of the study is to describe and compare experienced life satisfaction over the life cycle. An advantage of this study is that the analysis uses the same methodology to analyze data that were collected using uniform methods across countries, so it can be concluded that results in this study are identified due to differences in life cycle patterns, not differences in the methodologies or data. Chapter 3 examines the underlying changes in peoples’ lives that drive the pattern of subjective well-being observed in urban China from 2002 to 2012. The analysis is motivated by the finding that, despite income in urban China quadrupling between 1990 and 2010, subjective well-being followed a U-shaped pattern during this time and did not improve overall (Easterlin et al. 2010). The analysis in this chapter pays special attention to conditions in the labor market and 4 income comparison and habituation. Conditions in the labor market and income are an important focus during this time period because urban China was transitioning from a communist to free market-based system. This transition resulted in much higher incomes, but also a less stable labor market. Findings from the analysis are especially relevant for the near future in China as the government has stated a goal to further liberalize the labor market on a national level by encouraging rural to urban migration. The data for this analysis in Chapter 3 is repeated cross- sectional survey data used in the annual reports on China’s society published by the Chinese Academy of Social Sciences. A modified version of the Oaxaca decomposition method is developed to utilize annual data and is employed to divide the increase in life satisfaction from 2002 to 2012 into segments explained by changes in different aspects of life over this time. Chapter 4 focuses on the role labor market policy played in mitigating or exacerbating the decline in subjective well-being experienced in Europe during the Great Recession. The Great Recession provides an unparalleled opportunity to test how well labor market policy can protect well-being when countries are challenged with a large negative economic shock. Four different types of labor market policies are compared. The policies included in the analysis consist of policies that provide support to those who become unemployed, and policies that aim to protect workers by limiting the ability of firms to freely adjust their workforce. The policy effects on people with different levels of education and from different birth cohorts are considered because the effects of the Great Recession differed for these groups, and the policies are expected to affect these groups differently. The results provide evidence of what types of policy were most effective in mitigating the negative effects of the Great Recession on well-being, and who these policies protected. The analysis is based on a multi-level regression model using the variation in labor market policy across 23 European countries. Micro-level data is from the Eurobarometer, and 5 macro-level data (including policy data) is from the OECD. An instrumental variable approach is used to control for the endogenous labor market policy changes that occurred during the recession. Chapter 5 addresses the question of whether or not policy can be used to promote well- being. The analysis evaluates the effectiveness of a Familias en Acción, a conditional cash transfer program in Colombia, on subjective well-being. Familias en Acción provides cash transfers to households with children under the age of 18, conditional on the household being poor and the children of the household meeting school attendance or doctor checkup requirements. Poverty is determined by a household’s government assigned poverty score lying below a specific value, creating exogenous variation in program participation around the eligibility cutoff. The analysis re-creates the poverty score and uses the cutoff in eligibility as the basis for a regression discontinuity research design. This is the first study that can make causal claims about the effects of a conditional cash transfer program on subjective well-being. This contribution is especially relevant because conditional cash transfer programs have become popular in developing countries in the past decade. Thus, this dissertation is organized as follows: Chapters 2 through 5 present the four studies mentioned in the introduction. Chapter 6 summaries the findings from the studies and discusses the policy implications. 6 CHAPTER 2. Experienced Life Cycle Satisfaction in Europe 2.1 Introduction How does life satisfaction change as people age? Is the pattern consistent across countries? Is it the same for males and females and does it vary by level of education? The main purpose of this article is to answer these questions by comparing the experience of 17 European countries. Although these questions are fundamental to fully describing and understanding the life satisfaction of a population, there is still little consensus about how life satisfaction progresses as people age (Ulloa, Moller, and Sousa-Poza 2013, Baird, Lucas, and Donnellan 2010). A clear description of how life satisfaction changes over the life cycle can also be used to motivate further research on what causes the highs and lows in life satisfaction as people age, and subsequently to direct policy towards groups of people during ages when they experience low levels of satisfaction. Previous research on life satisfaction and age can be divided into two strands of separate but related literature which are often mistakenly grouped together. The aim of one strand is focused on identifying the pure age effect of life satisfaction and the aim of the other strand is concerned with identifying actual experienced life satisfaction over the life cycle. The former strand, focusing on the pure age effect, attempts to isolate and identify the effect of aging alone on life satisfaction by holding constant all life circumstances that are a function of age – such as health, marital status, employment status, and income. Therefore, the pure age effect strand of the literature does not answer the question of what happens to life satisfaction as people age. Instead, the identified relationship between age and life satisfaction reflects the experience of a hypothetical group of people who do not experience any changes in 7 life circumstances as they age. This group would have the same average income, health, marital status, and employment status at age 20 as they would at age 50 or 80. There is a robust result for the pure age effect on life satisfaction – the relationship is U- shaped over the life cycle. That is, the relationship is most negative in middle age, and is more positive in younger and older ages. This result has been identified across many countries, at various levels of income, using repeated cross sectional data (Blanchflower and Oswald 2008, Di Tella, MacCulloch, and Oswald 2001) and using panel data (Blanchflower and Oswald 2004, Van Landeghem 2012, Clark 2007). This paper contributes to the strand of the life cycle literature that aims to identify the actual experienced life satisfaction as people age – a simple description of what happens to life satisfaction from young to old ages. The results in the experienced life cycle satisfaction strand of the literature are mixed. Two studies find that the relationship between age and life satisfaction is hill-shaped in the United States (Easterlin 2006, Mroczek and Spiro 2005). Switek (2013) finds young adults in Sweden experience a hill shape in life satisfaction from age 22 to age 40. In contrast, a U-shape is identified in the Netherlands (Latten 1989). Blanchflower and Oswald (2009) also find a U-shape in three European countries 3 , and in pooled data for eight European countries 4 . The U-shape is also identified in Germany, Britain, and Australia by Cheng, Powdthavee, and Oswald (2014), but in those same countries a hill shape pattern that occurs later in the life cycle is identified by Beatton and Frijters (2012). In Britain, McAdams, Lucas, and Donnellan (2012), however, find a sideways S-shape; life satisfaction decreases until early 30's, then increases until just before age 80 at which point life satisfaction starts declining again. Baird, 3 France, Germany, and the United Kingdom 4 The countries pooled together are Belgium, Denmark, France, Germany, Italy, Ireland, Netherlands, and the United Kingdom. The fact that a U-shape is identified when pooling the eight countries, however, does not imply that a U-shape exists in all countries. 8 Lucas, and Donnellan (2010) also find this pattern in Britain as well as a mostly flat pattern that starts to decline later in life in Germany. The varied findings of these studies could be due to actual differences in life satisfaction patterns between countries, or due to differences in data sets or methodologies. Without comparable results among a group of countries it is impossible to determine if countries share the same patterns in life satisfaction, only share certain pattern characteristics, or are completely different. A study by Deaton (2010) addresses this issue by using the 2006 wave of the Gallup World Poll which includes a life satisfaction question for all countries. Deaton finds different patterns of life cycle satisfaction among 132 countries, but a common U-shaped relationship in high income English speaking countries. The major weakness of this study, however, is the findings are based on a single cross section. Therefore, it is impossible to control for cohort effects, which can change the identified shape of life cycle satisfaction (Clark 2007). The gaps that remain in the experienced life cycle satisfaction literature that this study addresses are as follows. First, to my knowledge, no study exists that covers a handful of countries using uniform methodology and data with enough time coverage to control for cohort effects. By addressing this shortcoming of the literature, this study will help resolve the question of whether the different life cycle patterns identified between countries in previous studies are due to different life cycle patterns or due to differences in datasets or methodologies. Secondly, the literature has largely overlooked whether there are differences by gender and education level in life cycle satisfaction patterns. This study contributes by identifying the life cycle satisfaction patterns for these population components. The major findings of this study are: 1) The average life cycle pattern for all countries resembles a wave like M-shape. The M-shape is the result of the majority of countries sharing a 9 few common life cycle features – a local maximum of life satisfaction around age 30, declining life satisfaction until around age 50 followed by rising life satisfaction, and decreasing life satisfaction after age 75. 2) In the majority of countries male life satisfaction increases relative to female life satisfaction throughout the life cycle. 3) More educated people report higher life satisfaction than less educated people throughout the majority of the life cycle. 4) Although the life cycle patterns of the countries included in this study share many common characteristics, when considering the life cycle as a whole there is no common pattern shared by all countries. 2.2 Data and methodology 2.2.1 Data Eurobarometer data that span from 1973 to 2010 are used for this analysis. The dataset is in the form of repeated cross sections that are harmonized for comparability across years. Two to five waves of data collection were conducted each year. The Eurobarometer is representative of populations at the country level with sample weighting. Throughout the years the Eurobarometer has been administered, countries have been added to the survey. At the start of the survey, only nine countries were sampled. By 2010, 36 countries were included in the survey. Not all 36 countries, however, have enough time coverage to properly control for cohort effects. After testing how many years of survey coverage is necessary to control for cohort effects 5 , all countries with less than 17 years of coverage are determined to have too few years and are dropped from the analysis. This criteria leaves 17 countries with a minimum of 17 years of coverage and a maximum of 36 years of coverage. The countries included in this study 5 The average life cycle life satisfaction pattern for countries with the longest coverage, 36 years, are first identified. Then, for these same countries, years of coverage are progressively dropped until the life cycle life satisfaction patterns identified using the truncated data start to diverge from the patterns identified with the full 36 years of coverage. The cutoff point where identified patterns started to change is 17 years of coverage. 10 are France, Belgium, the Netherlands, West Germany, Italy, Luxembourg, Denmark, Ireland, Great Britain, Northern Ireland, Greece, Spain, Portugal, East Germany, Finland, Sweden, and Austria. Life satisfaction is measured as the integer response to the question: “On the whole, are you very satisfied, fairly satisfied, not very satisfied, or not at all satisfied with the life you lead? Would you say you are .......?” (coded 4-1, with 4 being very satisfied and 1 being not at all satisfied) Cohorts range from birth year 1877 to 1990. For the countries with the longest coverage across years (36 years), the most sampled cohorts are cohorts born between 1945 and 1965. For the countries that span 17 years of coverage, cohorts born between 1960 and 1980 are the most sampled. Students were dropped from the sample because it is impossible to determine the final educational outcome of a student surveyed in cross sectional data. People less than 20 years old and older than 80 years old are also dropped due to small sample size. Table A1 in the appendix A summarizes the data that is used in the analysis. 2.2.2 Methodology The large size of the Eurobarometer allows for the use of a non-parametric approach as the main analysis. A non-parametric approach is preferred because it does not limit the possible shapes of life cycle satisfaction that can be identified. The general specification used is as follows: 𝐿𝑆 # =α+𝐴𝐷 # ∗𝛽+𝑓𝑒𝑚𝑎𝑙𝑒 # ∗𝛿+𝑐𝑜ℎ𝑜𝑟𝑡𝐷 # ∗𝛾+𝑒𝑑𝑢𝐷 # ∗𝜃+𝑠𝑢𝑟𝑣𝑒𝑦𝐷 # ∗𝜉+𝜀 11 LS represents life satisfaction. AD are age dummies with age 20 as the reference group. female is a gender dummy that takes the value of 1 when the respondent is female. cohort are 10 year cohort dummies with the cohort born between 1951 and 1960 as the reference group. The oldest cohort includes more than 10 birth years and spans 1881 to 1910. eduD are education dummies that represent the highest level of education attained. Education dummies are secondary, technical, and university or more. The reference group is less than secondary school. surveyD are survey dummies, which control for survey and period effects. Age, cohort, and survey (period) dummies are purposely not set to the same length of time to avoid an over-identification issue when including variables for age, period, and cohort. Variables that affect life satisfaction and are a function of age are not included as controls because the aim of this study is to identify the average experienced level of life satisfaction across ages 20 to 80. By omitting the controls that are a function of age, the parameter estimates for the age variables will reflect the average experienced life satisfaction across ages taking into account changes in variables like health, income, and employment status that vary from younger to older ages. The gender, education 6 , and birth cohort composition of the sample changes throughout the years of the Eurobarometer. If these variables are not included as controls, they could bias the identified life cycle pattern. For example, the shape of the latter part of the life cycle is identified from the survey responses of older cohorts. If the older cohorts were relatively less satisfied with their life at older ages due to a cohort effect, not an age effect, omitting cohort controls would bias 6 Education dummies reflecting highest level of education completed are generated based on age at last year of education. For each country, histograms plotting the distribution of age at last year of education are used to create education level dummies based on ages with high final year of education rates. The rational for this method is ages with high final year of education rates will reflect ages where major education milestones are reached, such as graduating from high school or university. Some countries warrant three education dummy variables and some countries warrant four. 12 the identified life satisfaction in the latter part of the life cycle downward. Also, because gender, education level, and birth cohort do not change over the life cycle for an individual, controlling for them will not change the shape of the identified life cycle satisfaction pattern. For each country, weighted OLS regressions are run for the population as a whole, and also for population subsets by gender and level of education. The regressions are weighted so the estimates are representative of national populations. For level of education, the population is divided into more and less educated groups. The less educated group is defined as the portion of the population with less than a four-year university level education, and the more educated group is defined as the portion of the population with at least a four-year university level education. For each population of interest, the parameter estimates from the weighted OLS regressions and the sample means of the explanatory variables, excluding age dummies, are recorded. Then, using the following equation average life satisfaction for each age is calculated: 𝐿𝑆 ?@A = α+𝛽∗𝐴𝐷 ?@A +𝛿∗𝑓𝑒𝑚𝑎𝑙𝑒 +𝛾∗𝑐𝑜ℎ𝑜𝑟𝑡𝐷 +𝜃∗𝑒𝑑𝑢𝐷 +𝜉∗𝑠𝑢𝑟𝑣𝑒𝑦𝐷 In the above equation 'bar' indicates the average value of an explanatory variable for the population being analyzed. The 'hats' over the parameters indicate they are parameter estimates and the hat over life satisfaction indicates predicted life satisfaction. In the above equation 𝐴𝐷 ?@A is the only variable. A separate calculation is made for each age by inserting the relevant age dummy into the equation above. For example, to calculate the average life satisfaction for age 31, the following calculation is made: 𝐿𝑆 BC = α+𝛽 BC ∗1+𝛿∗𝑓𝑒𝑚𝑎𝑙𝑒 +𝛾∗𝑐𝑜ℎ𝑜𝑟𝑡𝐷 +𝜃∗𝑒𝑑𝑢𝐷 +𝜉∗𝑠𝑢𝑟𝑣𝑒𝑦𝐷 The pattern identified is then smoothed using a locally weighted scatterplot smoothing technique (lowess). For each country, three sets of figures are generated. The first figure plots the average life satisfaction for the entire population. The second figure plots male and female life cycle patterns, and the final figure plots life cycle patterns for more and less educated groups. 13 To represent the common aspects of life cycle satisfaction patterns among countries, three graphs are generated that plot the average life cycle patterns of the 17 countries for the entire population, population divided by gender, and population divided by education level. 2.3 Results 2.3.1 National populations Figure 2-1: Mean life cycle satisfaction pattern, 17 countries How does the average life satisfaction change as people age from 20 to 80 years old? On average, for all the countries in this study, life satisfaction follows a wave like M-shape (Figure 2-1 7 ). Life 7 Because the shape of the life satisfaction pattern, not the level, is the focus of this study, Figure 2 shows how life satisfaction progresses from age 20 as a starting point. 14 satisfaction rises to a local maximum around age 30, declines to a minimum level around age 50, rises again to a maximum at age 70, and then declines until age 80. This pattern arises due to common life cycle satisfaction pattern features shared by the majority of countries. The most common feature shared among countries is a local minimum in life satisfaction around age 50. This feature is identified in 13 of 17 of the countries as shown in column 1 of Table 2-1. It should be noted that a local minimum around age 50 does not mean the overall pattern in life satisfaction resembles the often cited U-shape. In fact, only Belgium and East Germany have a pattern over the entire life cycle that resembles a U-shape. The second most common feature shared among countries is a local maximum in life satisfaction that occurs around age 30. As illustrated in column 2 of Table 2-1, in 12 of the 17 countries a local maximum around age 30 is identified. And finally, column 3 of Table 2-1 shows in 10 of the 17 countries life satisfaction decreases after age 75. 15 16 The fact that countries share the same life cycle pattern characteristics does not mean their patterns are the same over the length of the life cycle. At the country level, when considering the life cycle as a whole, there is no individual pattern shared by all countries. Instead, as illustrated in Figure 2-2 which plots the life cycle satisfaction patterns for all countries, there are a wide variety a patterns exhibited by the 17 European countries. To test if the magnitude of change in life satisfaction over the life cycle is significant for each country, the main regressions are re-run replacing the individual age group dummies with five-year age group dummies. Five-year age groups are used to minimize the effect of outlying levels of life satisfaction at single ages 8 . The difference between the five-year age group dummies that correspond to the life cycle satisfaction low and high is tested for significance. In 16 of the 17 countries the changes in life satisfaction that occur over the life cycle are significant at the 5% level (Table 2-2). In many of the countries the changes are still significant at even much stricter significance 8 Although the identification technique using five-year age dummies is different, the new technique doesn’t change the overall identified pattern of life satisfaction. 17 Figure 2-2: Life cycle satisfaction by country 18 levels. The average magnitude of change from lowest to highest levels of satisfaction over the life cycle for all countries is 0.13 on a 1-4 scale. To put that figure in perspective, for the 17 countries in this study 0.13 is approximately the same magnitude as the difference in life satisfaction between people with secondary school as their highest level of education compared to people with at least a university degree (0.11) without controlling for income differences between the two groups 9 . 9 The parameter estimates for 'secondary' and 'university' education dummies from a pooled regression with country dummies including all 17 countries. The estimate values are 0.081 and 0.195 respectively. 19 2.3.2 Differences by gender Figure 2-3: Mean life cycle satisfaction patterns by gender, 17 countries Does life satisfaction over the life cycle progress similarly for men and women? On average, life satisfaction follows an M-shape for both men and women, but there is a difference in trend between the genders. Male life satisfaction increases compared to female life satisfaction as both genders progress through the life cycle. This relationship in Europe is illustrated in Figure 2- 3 which plots the mean life cycle satisfaction pattern of all countries by gender. Because the level of life satisfaction is important for gender comparison, Figure 2-3 plots the mean level of life satisfaction for both men and women. The experience of individual countries also reflects the common relationship of male life satisfaction increasing relative to female life satisfaction. As shown in column 4 of Table 2-1, in 20 12 of the 17 countries male life satisfaction increases relative to female life satisfaction. In 9 of the 12 countries, male life satisfaction starts out lower than female satisfaction at young ages, but surpasses female life satisfaction by the end of the life cycle. In only one country, Sweden, female life satisfaction increases relative to male life satisfaction over the life cycle (Table 2-1, column 5). 2.3.3 Differences by level of education Figure 2-4: Mean life cycle satisfaction patterns by education, 17 countries Does life cycle satisfaction differ by level of education? The mean pattern for groups of people with at least a college degree and people with less than a college degree does not differ much – both groups have M-shaped patterns. But, the level of life satisfaction is greater for the 21 more educated group. This is illustrated in Figure 2-4, where the mean life cycle satisfaction pattern for more and less educated groups of all 17 countries is plotted. The level difference between groups is also common at the country level. As columns 6 and 7 in Table 2-1 show, this relationship is true for the majority of the life cycle in all countries if not the entire life cycle. 2.3.4 The importance of non-parametric methods Although the M-shape is the mean pattern of life cycle satisfaction of the 17 European countries, the patterns found for individual countries vary. The wide variety of life cycle satisfaction patterns found for individual countries in this study suggest non parametric methods should be used in future age and life satisfaction studies. A large portion of the existing life cycle satisfaction literature is focused on either proving or disproving the U-shaped life cycle satisfaction pattern. As a result, in many studies the pattern of life satisfaction over age is usually identified using age and age squared variables. This parametric approach limits the identified pattern of life satisfaction over age to be either flat, increasing, decreasing, hill-shaped or U-shaped. If the analysis in this study is re-run using age and age squared variables, 7 of the 17 countries would be classified as having a U-shaped pattern (Table 2-3). This conclusion is misleading. The results at the country level from the primary analysis in Figure 2-2 show the life satisfaction pattern over age is more complex – only two countries show a U-shaped pattern and almost all countries show patterns that cannot be identified by only using age and age squared variables. 22 2.3.5 Validity of findings – support from panel studies The largest weakness of this study is that the data used are not panel data, but are in the form of repeated cross sections. Controlling for the age, period, cohort problem, and also age related selectivity of survey participants, is a larger problem when analyzing repeated cross 23 sectional data compared to panel data. However, the validity of this study's findings is buttressed by the fact that the results are in agreement previous panel data studies that cover the same countries. Easterlin and Switek (2015) track the trajectory of life satisfaction for young adults in Sweden and also find a local maximum around age 30 (Figure 2-5). Cheng, Powdthavee, and Oswald (2014) find a U-shape for Britain and unified Germany. Their analysis, however, is constrained to finding only a handful of simple patterns. When the analysis for the current study is re-run with the same constraint on possible patterns as the Cheng, Powdthavee, and Oswald study, a U-shaped pattern is found for both Britain and unified Germany (Table 2-3, regressions 9 and 18). Frijters and Beatton (2012) analyze the life cycle satisfaction patterns for West Germany and Britain employing a non-parametric approach, and therefore their results are more comparable to the main results of this study. In both countries, Frijters and Beatton identify what they describe as a “late wave” – relatively little change in life satisfaction until a large increase in life satisfaction around age 50 followed by a turn in life satisfaction around age 70. Figure 2-6 and Figure 2-7 show the “late wave” found in West Germany and Britain by Frijters and Beatton is similar to the results from this study. 24 Young Adult Panel Study Young Adult Panel Survey figure taken from Easterlin and Switek (2015) Figure 2-5: Mean Swedish life cycle satisfaction, Eurobarometer and YAPS Patterns Compared 25 Figures from the German Socio-Economic Panel Survey (GESOP) are take from Frijters & Beatton (2012) Figure 2-6: Mean German Life Cycle Satisfaction, Eurobarometer and GSOEP Patterns Compared 26 Figures from the British Household Panel Survey (BHPS) are take from Frijters & Beatton (2012) Figure 2-7: Mean British Life Cycle Satisfaction, Eurobarometer and BHPS Patterns Compared 27 2.3.6 Plausible mechanisms Although identifying the underlying mechanisms behind life cycle satisfaction patterns is not the primary aim of this study, some exploration into the causes of the findings are possible. A feature of the M-shape is an increase in life satisfaction around age 60. A probable cause of the increase is retirement. Figure 2-8 plots the mean percent of the population reporting that they are retired by age for all countries compared to the average life satisfaction pattern over age for all countries. An increase in the portion of the population reporting they are retired coincides with the increase in life satisfaction around age 60. The different experience of men and women is also telling. As shown in Figure 2-3, the increase in life satisfaction around age 60 is greater for men than for women. This observation is in agreement with the retirement explanation because in the time period this study covers, a higher proportion of men worked compared to women, and therefore a higher portion of men also retired from work. The retirement explanation is also supported by previous studies that find life satisfaction is positively associated with retirement (Easterlin 2003, Kim and Moen 2002, Charles 2004, Fonseca et al. 2014). There is also suggestive evidence for some mechanisms underlying the relative increase of male to female life satisfaction. The same relationship between genders is identified in the United States in a study by Easterlin (2010). The dataset used in Easterlin's study, the General Social Survey, includes variables on satisfaction with domains of life, which are used to provide an explanation of the improvement in male life satisfaction compared to female life satisfaction. Easterlin argues that different patterns of satisfaction with family between men and women contribute to the different patterns in 28 Figure 2-8: Percent of population retired compared to life satisfaction by age, 17 countries 29 life satisfaction. Females enjoy relatively higher satisfaction with family earlier in life and males enjoy relatively higher satisfaction with family later in life. The explanation for the initial advantage in family satisfaction that females enjoy over males is because females marry at younger ages than males. As both genders age, however, male family satisfaction increases relative to female family satisfaction because men are more likely to be married at older ages. This is primarily because males exit marriages more frequently at older ages due to dying earlier than their wives, leaving their widows with low levels of family satisfaction (Easterlin 2010). Figure 2-9: Mean male percent married less female percent married by age, 17 countries The explanation of declining family satisfaction shaping the life cycle satisfaction between the genders also seems fitting in Europe. The same patterns in marriage rates between the genders, a relatively higher percentage of females being married in early ages and a relatively higher 30 percentage of males being married at older ages, is also found in the 17 countries included in this study and is illustrated in Figure 2-9. Figure 2-9 plots the mean percentage of males reporting they are married less the percentage of females reporting they are married over age for all countries. At younger ages there is a larger percentage of females reporting they are married compared to males, but this relationship is reversed as people age. 2.4 Conclusion The descriptive findings in this study answer several questions that were previously unanswered in the life cycle satisfaction literature. The mean life cycle pattern of populations as a whole is M- shaped for the 17 European countries in this study. The M-shape is a reflection of common life cycle satisfaction features shared by the majority of countries: a local maximum in life satisfaction around age 30, declining life satisfaction until around age 50 followed by rising life satisfaction, and declining life satisfaction after age 75. On average both males and females share the M-shaped pattern, but the trajectory of life satisfaction differs between males and females – male life satisfaction improves relative to female life satisfaction as people progress through the life cycle. People with at least a university education and people with less education also share the M-shape on average, but people with a university education enjoy a higher level life satisfaction throughout the life cycle. At the country level, the life cycle satisfaction patterns vary. Unlike in the pure age effect strand of the life cycle satisfaction literature, where the U-shaped relationship between age and life satisfaction is frequently found (Di Tella, MacCulloch, and Oswald 2001, Blanchflower and Oswald 2004, Blanchflower and Oswald 2008, Van Landeghem 2012, Clark 2007), this study provides evidence that there is no universal life cycle pattern shared among all countries when 31 considering experienced life satisfaction. Furthermore, the wide variety of patterns found in the 17 countries in this study motivates the use of non-parametric methods in future life cycle satisfaction studies. The large contribution of this study is enabling the comparison of life cycle satisfaction patterns between a larger number of countries than previously possible. By using the exact same methodology to analyze data that were collected using uniform methods among countries, it can be concluded that the results in this study are due to differences in life cycle patterns, not differences in the methodologies or data. The facts about experienced life cycle satisfaction established in this study are meant to motivate more rigorous research investigating the underlying mechanisms behind changes in life satisfaction over the life cycle. A secondary motivation for this article is to clarify the difference between the two strands of life cycle satisfaction literature (pure age effect on life satisfaction vs. actual experienced life cycle satisfaction) for popular media. The results from this study clearly show that the experienced life cycle satisfaction pattern is not U-shaped for the vast majority of countries. But in popular media, the U-shaped relationship between age and life satisfaction that is commonly identified in the pure age effect strand is frequently misreported as the experienced life cycle pattern. A study in the pure age effect strand of the literature by Blanchflower and Oswald (2008) has been cited by The Economist (December 16 2010), The Guardian (January 29 2008), and Scientific American (January 30 2008) in articles misreporting that people experience a U-shaped pattern of life cycle satisfaction in countries around the world. These articles are representative of the popular media spreading false information by confusing the two separate stands of life cycle satisfaction literature. 32 CHAPTER 3. Well-Being in Transition: Life Satisfaction in Urban China from 2002 to 2012 10 3.1 Introduction In the past few decades, China has undergone a radical transition from a poor communist country to a rising superpower by embracing the free-market. The rapid transition has resulted in vast changes in life circumstance for over one billion people in China. No where in China has this statement been more true than in the urban centers. Since 1990, when the Chinese government started a transition from a centrally planned communist system to a free-market based system in urban China, average income more than quadrupled by 2010 (Chen et al. 2011). The average was not just moved by richer peoples’ skyrocketing incomes; almost everyone in urban China enjoyed increasing incomes and consumption after the start of the transition (Cai et al 2010). Yet, when people were asked about how they feel about their lives during this time, there was no improvement. Between 1990 and 2010, average subjective well-being in urban China did not follow income; if anything, average subjective well-being declined (Easterlin et al. 2012, Bartolini and Sarracino 2015). It is clear from the disparate income and subjective well-being trends that simply increasing income is not enough to increase subjective well-being in urban China; there are other changes in peoples’ lives that are dominating any positive effects of income. The aim of this study is to identify what changes in the lives of urban Chinese were responsible for the trend in subjective well-being in urban China from 2002 to 2012. Repeated cross-sectional micro-level data are used 10 Co-authored with Fei Wang 33 in the analysis. We develop a modified version version of the Oaxaca decomposition method to take advantage of annual data and estimate what portion of the change in subjective well-being can be attributed to changes in various aspects of peoples’ lives. Furthermore, the urban population is also divided by level of education and separately analyzed to determine if subjective well-being trends are pushed by the same underlying changes in life circumstance for these different segments of the population. We pay special attention to how conditions in the labor market and income habituation and comparison relate to the subjective well-being trend during this time. This study’s focus on conditions in the labor market is directly motivated by the findings from Easterlin et al. (2012), an article that establishes the subjective well-being trends in urban China from 1990 to 2010. The 2012 article finds that, between 1990 and 2010, subjective well- being followed a U-shaped pattern. The U-shaped pattern inversely mirrors urban unemployment rate estimates during this time, which is taken as evidence that changes in unemployment rate are moving the pattern in subjective well-being. This finding is a valuable contribution, but still questions remain about to what extent labor market conditions are determining the trend in subjective well-being. This study builds on these findings by estimating what percent of the increase in subjective well-being from 2002 to 2012 can be attributed to changes in labor market conditions. This study also focuses on income comparison and habituation. The experience in urban China during this time period presents a unique opportunity to test an explanation of why, generally, subjective well-being trends do not follow income trends – positive effects of income increases are undercut by social comparison and habituation/rising aspirations (Easterlin 2003, Clark et al. 2008). This study specifically addresses the question, can the counterbalancing forces of social comparison and income habituation, even during a time of unprecedented income growth, 34 help explain why there has been no long run increase in subjective well-being in urban China since 1990? Despite the surprising subjective well-being pattern in urban China, there have been relatively few studies addressing this issue. Of the studies that address the pattern, no studies besides Easterlin et al. (2012) examine conditions in the labor market as a potential explanation. Three studies focus primarily on simply describing the trend in subjective well-being (Burkholder 2005, Kahneman and Krueger 2006, Crabtree and Wu 2011), and two studies attempt to find the underlying factors moving subjective well-being, but don’t specifically address labor market conditions. Brockmann et al. (2009) find that increasing income inequality and rising financial dissatisfaction contributed to the decline in subjective well-being from 1990 to 2000. Bartolini and Sarracino (2015) find that social comparisons and the loss of social capital explain the trend observed between 1990 and 2007. The issue of income comparison and/or habituation as an explanation of subjective well- being trends in China has been addressed by a handful a studies. The following studies find evidence that income comparison/habituation do exist in China (Oshio et al. 2011, Appleton and Song 2011, Smyth and Qian 2008, Wang and VanderWeele 2011, Knight and Gunatilaka 2011). These studies, however, are based on cross-sectional data, and therefore cannot address to what extent income comparison/habituation relate to trends in subjective well-being. Brockmann et al. (2009) and Bartolini and Sarricino (2015) use financial dissatisfaction as a proxy for social comparison under the assumption that the financial dissatisfaction measure reflects social comparison; an assumption is buttressed by empirical findings from Germany (D’Ambrosio and Frick 2007). They indeed to find that financial dissatisfaction is an important determinant of subjective well-being change. Our study provides a complimentary method, based on the work of 35 Clark et al. (2008), of estimating the relationship between subjective well-being and income comparison and habituation using reference group and previous income. Our analysis finds improving labor market conditions can account for 29% of the increase in subjective well-being between 2002 and 2012 in urban China. This is primarily driven by a large decrease in unemployment during this time. The improving labor market is especially important for the segments of the population most vulnerable to the negative effects of the transition – people with less than a college education. We also find that any positive effect increasing income had on subjective well-being was nullified by income comparison and habituation. The net result is that changes in income has no significant relationship with the increase in subjective well-being from 2002 to 2012. We also find suggestive evidence that the increasing trend in subjective well-being can partially be explained by older, less satisfied birth cohorts exiting our sample. 3.2. Data 3.2.1 Dataset Description The data used in this analysis is from the Horizon Research Consultancy Group (HRCG). HRCG is an independent data company based in China. They have collected data since the 1990’s that are used in the annual reports on China’s society published by the Chinese Academy of Social Science. The complete HRCG dataset that we use is in the form of repeated cross sections spanning 2000-2012 and covers both urban and rural China. A multi-stage random sampling method was used to gather the data 11 and sample weights are used in the analysis to assure the data are representative of the population within the characteristics of our sample. 11 Cities of survey were determined first. Then, within each city, the central district along with other randomly selected districts were targeted for sampling. Communities within the selected districts were randomly chosen for sampling. Finally, households in a selected community were 36 For this study’s analysis, data must be gathered from an urban population that is comparable across years. To this end, certain segments of the dataset are omitted from the analysis. Because the focus of this analysis is urban China, the rural data are dropped. 12 In urban areas, Horizon surveyed twenty different cities from 2000-2012, but the coverage of these cities varied from year to year. Seven cities, however, were surveyed together frequently throughout this time period: Beijing, Shanghai, Guangzhou, Wuhan, Shenyang, Xi’an, and Chengdu. Of the 13 years of coverage, in 11 years all these seven cities were surveyed. All cities besides the seven most frequently surveyed were dropped from the analysis. The two years that did not cover these seven cities, 2000 and 2006, were also dropped. Over the entire sample, ages 18-60 were sampled every year. In some years, ages outside of this range were sampled, but to keep waves comparable across years observations outside of 18-60 were dropped from the analysis. We also drop students because it is impossible to determine their final level of education in our data. The remaining data that we use in the analysis span 2001 to 2005 and 2007 to 2012 and are representative of people aged 18 to 60 in the following seven cities: Beijing, Shanghai, Guangzhou, Wuhan, Shenyang, Xi’an, and Chengdu. 13 There are just under 2,000 total observations included in the sample we analyze each year. randomly chosen according to the rule of “sampling one household after passing by five households”. One respondent was randomly determined within each selected household to complete the survey. 12 Another reason of dropping the rural sample is that selected rural villages were generally incomparable over time, and for some years, rural surveys were not conducted. 13 The seven cities are not representative of urban China. Nevertheless, the data includes typical cities in all regions of China. Beijing, Shanghai, Guangzhou, Wuhan, Shenyang, Xi’an, and Chengdu belong to North China, East China, South China, Central-south China, Northeast China, Northwest China and Southwest China. 37 3.2.2 Variables Subjective well-being 14 is the focus of this study because it is a comprehensive measure of well-being based on what people deem to be important in their own lives. This is especially important when studying the well-being of urban China during a time when so many changes are occurring. Instead of relying on an expert to decide which life circumstance should be a proxy measure of well-being, we allow the people experiencing the changes in China to report their feeling about their lives, and then find variables that are related to the self reported measure. Subjective well-being has been validated as a meaningful and reliable measure of well-being (Stiglitz, Sen, Fitousi 2009), is recently being officially measured by almost all OECD countries, and the United Nations is even encouraging nations to used subjective well-being as a guide for policy (Helliwell et al. 2013). Life satisfaction is the subjective well-being measure we use in this study. Life satisfaction is measured by the question (translated from Chinese): “Overall, how satisfied are you with your life now? Very satisfied, fairly satisfied, neutral, fairly unsatisfied, or very unsatisfied? [choose one].” The response options are coded 5 though 1, with 5 representing very satisfied and 1 representing very unsatisfied. Neutral is coded 3. The trend in life satisfaction that we analyze in this analysis is illustrated by Figure 3-1. The increasing trend in life satisfaction between 2002 and 2012 found in our HRCG data is similar to the increasing trends reported in previous studies that use other datasets (Easterlin et al. 2012, Bartolini and Sarracino2015). 14 In this paper, the term “subjective well-being” is used to refer to self reported evaluations of a person’s happiness or satisfaction with life. These measures are considered comparable because they correlate with the same explanatory variables (Helliwell et al. 2012) 38 Source: HRCG data used in analysis Figure 3-1. Average life satisfaction in urban China from 2002-2012 The explanatory variables included in the analysis measure income, reference income, previous year income, employment status, level of education, gender, cohort, age, year, and city. In this analysis, income is measured as the categorical response to a household monthly income question. 15 The categories are: 3,000 yuan or less, 3,001-5,000 yuan, 5,001-8,000 yuan, and more than 8,000 yuan. To ensure the categories are the same across all years, which is required by our analysis, the income categories are all nominal values. Although controlling for real income is ideal, converting income to real values would result in inconsistent income categories across years. Previous studies have shown that income aspirations due to social comparison affect the relationship between income and subjective well-being (Clark et al. 2008, Vendrik 2013, D’Ambrosio and Frick 2012). To account for the increasing income aspirations of respondents 15 Household size is not controlled in analyses because only a couple of years of surveys include this variable. For the years with household size, results are similar no matter if household size is considered or not. 39 due to the rising income of a comparison group over time, we generate a reference group income variable from the data. The reference group variable is generated as follows. For each wave, we place every observation in a category defined male/female, high/low education, old/young cohorts, and their city. 16 The reference income for each observation is defined as the median income of the category in which the observation is assigned. People also habituate to their own level of income (Ditella et al 2010, Vendrik 2013, Clark et al 2008). To account for habituation effects, we generate an approximate previous year income variable. We run an ordred logit regression of income categories on gender, education dummies, city dummies and cohort dummies in the previous year. Previous year income is created by predicting an individual’s income category in the previous year using their characteristics of the survey respondent in the current year. 17 The methods we use to generate reference income and previous year income have both been used in the subjective well-being literature. 18 We use different methods to generate reference and previous year income because of the accuracy of each method. The categorical method used to generate reference income is less precise than the order logit prediction method. We argue that this is preferable for reference income under the assumption that people tend to compare themselves to others somewhat similar to themselves. The accuracy of the ordered logit prediction method is used to predict previous year income because we want to estimate the previous year category for each survey respondent as accurately as possible. The analysis of macro-level trends provides strong evidence that the labor market conditions largely shaped the pattern of life satisfaction in urban China since the early 1990s 16 There are 56 groups in each year. Groups can not be smaller as the sample size of a group would be too small to generate reliable statistics. 17 The previous income of respondents in 2007 is their predicted income in 2005. 18 Clark et al (2008) offer a review. 40 (Easterlin et al 2012). Furthermore, unemployment is commonly found to be an important determinant of subjective well-being (Blanchflower and Oswald 2004, Clark et al. 2001, Kassenboehmer and Haisken-DeNew 2009, Winkelmann and Winkelmann 1998). In our analysis, employment status is divided into three categories: employed, unemployed, and retired. Employed is defined as anyone responding they are formally or informally working. For example, a respondent formally working for a company and a housewife informally working for a household would both be considered employed. Unemployed is defined more broadly than the traditional definition; a respondent is considered unemployed if they respond they don’t have a job, lost their job, are looking for a job etc. In this broad definition, discouraged workers are also categorized as unemployed. Compared to the traditional definition of unemployment, our broad definition likely better reflects labor market conditions during this time because in the early 2000’s there were many discouraged workers in urban China (Cai et al 2008, Knight and Xue 2007, Lu and Gao 2011). In the experience of the European transition countries, older generations typically faired worse than younger generations after the fall of the USSR (Easterlin 2010). Given the parallels seen in the pattern of life satisfaction during the Chinese transition to capitalism, older cohorts may have also suffered in China. Furthermore, Wang and Zhou (2016) find that older cohorts in this sample are less satisfied with their lives due to participating in Mao’s “send-down movement.” Cohort is therefore an important variable to control for especially because the age cutoff in our sample is 60; the increase in life satisfaction could be largely explained by older, less satisfied cohorts leaving the sample by 2012. Cohort is controlled for by one-year cohort dummies. To isolate the effect of cohort, age and year are included as controls. Year dummies are included and age is controlled for by two-year age group dummies to break the age, period, cohort control 41 collinearity problem 19 . City, education level, and gender dummies are also included in the analysis as controls. 3.3. Methodology The general analysis is divided into two steps. First, the relationship between life satisfaction and explanatory variables is established by a linear regression. Second, using the parameter estimates and the change in the average values of the explanatory variables from 2002 to 2012, the change in life satisfaction from 2002 to 2012 is divided into portions related to the change in each explanatory variable. 20 Our method is essentially an Oaxaca decomposition modified to take advantage of yearly data (see Appendix B for details). Using yearly data has a few advantages over using only start and end dates (which the Oaxaca decomposition typically requires). Using yearly data provides a larger sample size and allows for this analysis to include age and cohort effects along with previous year income effects. In the first step of the analysis, data from all years is pooled together and life satisfaction is regressed on all the explanatory variables. The estimated linear equation is expressed as 𝐿𝑆 #,F =𝒙 #,F 𝜷, (3.1) where i indicates an individual, t represents a year, 𝐿𝑆 is fitted life satisfaction, x is a vector of K explanatory variables and 𝛽 denotes the parameter estimates. In the second step of the analysis, the parameter estimates from the linear regression and the change in explanatory variables are used to divide the change in life satisfaction from 2002 to 19 Different age group dummies were inserted into the analysis for robustness checks. The results did not change. 20 The sample of year 2001 will be only used to generate the previous income of year 2002. 42 2012 into portions attributed to changes in each variable. According to equation (1), the average life satisfaction of year t, 𝐿𝑆 F , can be expressed as: 𝐿𝑆 F =𝐿𝑆 F =𝒙 F 𝜷= 𝑥 J,F 𝛽 J K JLC , (3.2) where the first equality holds because the survey year variables in the linear regressions are in the form of year dummies. 21 The year dummies absorb any year to year effects that are not captured by the other control variables. The increase in life satisfaction from 2002 to 2012 is Δ𝐿𝑆 =𝐿𝑆 NOCN −𝐿𝑆 NOON , and the percent contribution of the change in 𝑥 J to Δ𝐿𝑆 is defined as: 𝑐 J = Q R,STUS VQ R,STTS W R XYZ ×100%. (3.3) This method does not impose any restriction on the sign of 𝑐 J . Therefore, if the change in a variable pushed subjective well-being down during this time, the corresponding 𝑐 J would be negative. That is to say, our model allows for counteracting effects on subjective well-being. The only restriction we impose on this size of 𝑐 J is that all the all the percent contribution variables sum to 100%. Therefore, the percent contribution results can be interpreted as follows: if the percent contribution of a variable is large and significant, 1 - variable is closely related to subjective well-being, and 2- the variable changed a lot between 2002 and 2012. Our model imposes one restrictive assumption – the relationship between explanatory variables and subjective well-being does not change over time. In our model, any changes that cannot be explained by the included explanatory variables are captured by the year dummies. This includes changes in life satisfaction that are due to changes in the relationship between the 21 In the pooled cross-sectional regression 𝐿𝑆 #,F = 𝒙 #,F 𝜷 = 𝒛 #,F 𝜸+ 𝑑 F 𝛿 F NOCN FLNOOC , where 𝑑 F is a dummy for year t, 𝛿 F will be chosen, by construction, so 𝐿𝑆 F = 𝐿𝑆 F for each t. 43 explanatory variables and life satisfaction (proof in Appendix B). For this reason, the interpretation of the year dummies is percent change in life satisfaction unexplained by our model. 3.4. Results Table 3-1 presents the OLS regression results from the first step of the analysis. The first step identifies the relationship between the explanatory variables and life satisfaction that we use to compute the percent contribution results in step two. The relationships have the expected signs and levels of significance. Males are less satisfied with life than females, more educated people are more satisfied with life, unemployed people are much less satisfied than employed people, and life satisfaction is increasing in own income. The relationship between previous income and life satisfaction is evidence of income habituation to a point. That is, if in the previous year the survey respondent had higher income up to > 8,000 yuan, they are less satisfied with their life in the current year. The previous year > 8,000 yuan dummy is positive (the opposite of what we expect), but not significant. Similar to previous year income, the relationship between reference income shows income comparison negatively affects life satisfaction up to a point. If a respondent’s reference group has higher income, the respondent has lower life satisfaction up until the reference income is > 8,000 yuan. The relationship between > 8,000 reference group income and life satisfaction is negative, but not significant and also smaller in magnitude than the relationships identified for reference income between 3,001 and 8,000 yuan. 44 Table 3-1. Step 1 OLS regression results Dependent variable: Life satisfaction (1-5 scale) Male -0.054*** [ 0.020] Middle school 0.028 [ 0.024] High School 0.129*** [ 0.041] College 0.237*** [ 0.089] Unemployed -0.469*** [ 0.027] Retired 0.076*** [ 0.025] OWN INCOME (yuan) 3,001-5,000 0.205*** [ 0.032] 5,001-8,000 0.320*** [ 0.054] > 8,000 0.327*** [ 0.050] PREVIOUS INCOME (yuan) 3,001-5,000 -0.025 [ 0.144] 5,001-8,000 -0.648** [ 0.293] >8,000 0.281 [ 0.254] REFERENCE INCOME (yuan) 3,001-5,000 -0.108*** [ 0.033] 5,001-8,000 -0.177** [ 0.086] > 8,000 -0.062 [ 0.176] Two-year age dummies Y One-year cohort dummies Y City dummies Y Year dummies Y R squared 0.1029 N 18139 The omitted variables are: female, less than middle school education, employed, and for all income categories less than or equal than 3,000 yuan. The table includes all data from 2002 to 2012. 45 The incongruent relationship identified for the > 8,000 yuan previous year and reference income are most likely due to small sample size. Only 5.9% of the sample had a previous year income of > 8,000 yuan, and only 0.25% of the sample had reference income > 8,000 yuan. Table 3-2. Contribution of the change in variable values to the life satisfaction increase between 2002 and 2012 Δ𝐿𝑆 = 0.292 Variables ∆𝑥 Contribution (%) Male 0.03 -0.6 (p = 0.0345) Education 4.5 (p = 0.0277) middle school -0.05 -0.4 high school 0.01 0.5 college 0.05 4.4 Employment status 29.0 (p = 0.0000) unemployed -0.20 31.5 retired -0.09 -2.5 Income -55.4 (p = 0.4695) Own income 3,001-5,000 0.12 8.4 5,001-8,000 0.40 44.4 > 8,000 0.18 20.4 Previous income 3,001-5,000 0.21 -1.9 5,001-8,000 0.39 -86.5 >8,000 0.15 14.0 Reference income 3,001-5,000 0.24 -9.0 5,001-8,000 0.74 -44.9 > 8,000 0.02 -0.4 Age -10.5 (p = 0.1586) Cohort 66.4 (p = 0.1570) City 9.4 (p = 0.0478) Year 57.0 (p = 0.4501) The right column of Table 3-2 shows the contribution of the change in variable values to the life satisfaction increase between 2002 and 2012. The center column shows the change in the average variable value. The contribution of one variable may involve the contribution of more 46 than one dummy. The sum of all dummies within every such variable is contribution of the variable as a whole. For example, the contribution of changing employment status as a whole is the sum of the contribution from both employment status dummies. Table 1 segments the variables involving multiple dummies with horizontal lines and lists the contribution of variables as a whole in bold. The bolded contribution is followed by p values which represent the statistical significance of the contribution of the variable as a whole 22 . From 2002 to 2012, life satisfaction increased 0.292 points on a 5 point scale. The change in employment status, the largest statistically significant variable, contributed to 29.0% of the increase in life satisfaction primarily due to the large decrease on the number of people unemployed (-0.20). One might be skeptical about such a dramatic drop in unemployment, but two things must be kept in mind. First, peak levels of unemployment are observed around 2000- 02 in urban china in other datasets, thereafter a sharp decrease in unemployment rate was recorded (Knight and Xue 2006, OECD 2010, Gustafsson and Ding 2011). Both Knight and Xue (2006) and Gustafsson and Ding (2011) report peak unemployment levels just under 12%. Secondly, the definition of unemployed used in this analysis includes discouraged workers and therefore will count more people as unemployed than the common definition. Previous studies from the subjective well-being literature support the finding that change in unemployment status is important. Unemployment is consistently found to have a substantial negative impact on subjective well-being at an aggregate level (Ditella et al. 2003, Wolfers 2003), and studies show people don't adapt to being unemployed (Lucas et al 2004). In fact, there is even reason to think the contribution of employment status change is a lower bound estimate. High unemployment rate is commonly found to have negative spillover effects for those who remain 22 Appendix-B details how p-values were calculated 47 employed (Ditella et al 2003, Bjornskov 2012, Wolfers 2003, Arampatzi et al. 2014), presumably because people who remain employed worry about loosing their job in times of high unemployment. The analysis in this paper only estimates the individual effects of unemployment, so the spillover effects of the decrease in the unemployment rate on employed people is not accounted for in the percent contribution of change in employment status. Change in population distribution between cities explains a statistically significant 9.4% increase in life satisfaction between 2002 and 2012, which implies that the population shares of happier cities have been rising. The increasing level of education in the urban population accounted for a statistically significant 4.5% increase in life satisfaction over time. Change in own income is positively related to the increase in life satisfaction, but the overall contribution of income turns negative and statistically insignificant once previous income and reference income are taken into account. While the finding that increasing incomes did not significantly contribute to the increase in life satisfaction might shock some, our result is consistent with findings on financial dissatisfaction from previous studies. Both Brockmann et al. (2009) and Bartolini and Sarracino (2015) find that financial dissatisfaction increased in urban China during rapid income growth after 1990. Also, our finding that changes in income are not related to the changes of subjective well-being are consistent with the U-shpaed pattern of subjective well- being in urban China between 1990 and 2010 (Easterlin et al. 2012) and the finding that globally, long run changes in income are not related to long run changes in subjective well-being (Easterlin et al. 2010). The effect of birth cohort change in our sample is large, 66.4%, but insignificant. Significance is difficult to interpret in this case because we control for both age and year along with birth cohort. Because all of these variables are highly correlated, the standard errors on the 48 estimated relationship between life satisfaction and cohort, year, and age are large. Because of the large standard errors, we lack the statistical power to gain a statistically significant result or accurate estimate for the contribution of cohort change. Keeping these statistical limitations in mind, the size of the contribution of cohort change implies a large part of the increase in life satisfaction between 2002 and 2012 is potentially explained by older cohorts exiting our sample. Finally, the large percent contribution of year indicate that the variables included in our model only can explain a portion of the increase in life satisfaction between 2002 and 2012. As mentioned previously, the contribution of year will reflect the change in life satisfaction that is unexplained by our model. Due to the vast changes occurring in urban China during this time, and our limited number of variables, we expect that our model cannot completely explain the change in life satisfaction during this period. The population is also divided by level of education and the analysis is repeated. This division is motivated by vastly different life satisfaction experience of groups with different levels of education from 2002 to 2012, as illustrated by Figure 2. The less educated segments of the population were worse off initially, but recovered to a point closer to the college educated segment of the population by 2012. The analysis divided by level of education will address the question, are the trends in life satisfaction for the more and less educated driven by the same factors? 49 Figure 3-2. Life satisfaction patterns by level of education from 2002 to 2012 Although the primary focus of this study is the analysis of trends between 2002 and 2012, the initial starting point of these three groups in 1990 is also of considerable importance. In 1990, before the transition started, life satisfaction levels were evenly distributed throughout the population and the average satisfaction level was higher than in 2012 (WVS 2014). If the starting points of these different education groups are viewed in terms of the transition as a whole, it is clear the less educated segments of the population suffered greatest during the initial phases of the transition from the early 1990s to 2002. 50 Table 3-3. Step 1 OLS regression results, division by level of education Dependent variables: Life satisfaction High school or below College Male -0.052** -0.057** [ 0.025] [ 0.028] Middle school 0.039* [ 0.022] High school 0.160*** [ 0.036] Unemployed -0.481*** -0.348*** [ 0.031] [ 0.023] Retired 0.067*** 0.039 [ 0.024] [ 0.082] OWN INCOME (yuan) 3,001-5,000 0.224*** 0.154*** [ 0.033] [ 0.052] 5,001-8,000 0.333*** 0.285*** [ 0.067] [ 0.062] > 8,000 0.293*** 0.340*** [ 0.038] [ 0.070] PREVIOUS INCOME (yuan) 3,001-5,000 -0.252** 0.610*** [ 0.100] [ 0.178] 5,001-8,000 -0.589 -0.520 [ 0.484] [ 0.347] > 8,001 -0.154 0.902*** [ 0.514] [ 0.275] REFERENCE INCOME (yuan) 3,001-5,000 -0.130*** -0.026 [ 0.044] [ 0.049] 5,001-8,000 -0.206* -0.042 [ 0.107] [ 0.058] > 8,000 0.116 -0.057 [ 0.130] [ 0.111] Two-year age dummies Y Y One-year cohort dummies Y Y City dummies Y Y Year dummies Y Y R squared 0.1091 0.0692 N 12968 5171 51 Table 3-3 presents the step 1 OLS regression results for the population divided by level of education. In general, the relationship between the explanatory variables and life satisfaction is similar between people with less than a college level education and people with a college education. There is one important difference between subsamples, however. The relationship between the unemployed dummy and life satisfaction is larger for the less educated group. Table 3-4. Contribution of the change in variable values to the life satisfaction increase between 2002 and 2012, by level of education High school or below Δ𝐿𝑆 = 0.347 College Δ𝐿𝑆 = 0.125 Variables ∆𝑥 % ∆𝑥 % Male 0.04 -0.6 (p=0.08) 0.00 0.2 (p=0.09) Education 2.6 (p=0.00) Middle school -0.04 -0.4 High school 0.07 3.0 Employment status 33.3 (p=0.00) 8.1 (p=0.11) unemployed -0.25 35.2 -0.04 10.2 retired -0.10 -1.9 -0.07 -2.1 Income -77.9 (p=0.29) 117.0 (p=0.51) Own income 3,001-5,000 0.22 14.0 -0.11 -13.5 5,001-8,000 0.43 41.5 0.33 75.7 > 8,000 0.12 10.2 0.30 82.2 Previous income 3,001-5,000 0.31 -22.2 0.00 0.0 5,001-8,000 0.39 -65.5 0.39 -162.0 > 8,000 0.10 -4.6 0.23 167.5 Comparison group income 3,001-5,000 0.32 -12.0 0.08 -1.7 5,001-8,000 0.67 -39.6 0.88 -29.6 > 8,000 0.01 0.3 0.03 -1.6 Age -3.4 (p=0.01) -3.0 (p=0.93) Cohort 64.8 (p=0.23) 52.3 (p=0.76) City 14.0 (p=0.00) 4.7 (p=0.14) Year 67.3 (p=0.29) -79.3 (p=0.76) 52 Table 3-4 displays the final results of the analysis divided by level of education. All the results in Table 3-4 were calculated the same way as the results in Table 3-2. Comparting the percent contribution to increase in life satisfaction of changing employment status between the two groups, the results show that changing employment status is relatively more important contributor for people with less than a college education. This is due to two factors. First, the linear regression results of Table 3-3 show the negative association between being unemployed and life satisfaction is greater for people will less than a college education. Second, the decrease in people reporting they were unemployed was much greater for the lesser educated. Changes in income were not significantly related to changes in life satisfaction for either group. While these results should be interpreted cautiously due to the lack of statistical significance, it is of interest that the percent contribution estimate for people with less than a college degree is large and negative, and for people with a college degree the estimate is large and positive. These results may indicate that increasing incomes have some benefit for the richer people in society, but come at a cost to the poorer members of society. 3.5. Historical Context The narrative that emerges from our results is improving labor market conditions, indicated by the drop in number of unemployed people, contributed most to the increasing life satisfaction in urban China from 2002 to 2012. This is primarily due to the segments of the population that had the largest increase in life satisfaction during this time - the people with less than a college degree - greatly reducing their levels of unemployment. 53 The time period this analysis covers is a middle segment of a larger transition towards a capitalist economy that still continues through the 2010s. It is important to frame the results from this analysis in a historical context in order to draw the correct conclusions from this study. The starting point of this analysis occurs after the first phase of the transition away from a communist system, the massive downsizing and diminishing of the state owned enterprises and the creation of a labor market in urban China (Knight and Song 2005). The rapid reduction of the urban labor force employed by state owned enterprises combined with the troubles of transitioning to a market to allocate labor resulted in high unemployment (Gustafsson and Ding 2011, OECD 2010) and low life satisfaction in the early 2000s compared to 1990 (Easterlin et al 2012, Bartolini and Sarracino2015). The period there after, which this study covers, was a time of labor market and life satisfaction recovery. The results for this analysis should therefore not be viewed as potential ways to make the urban Chinese population happier, but instead as a reflection of what is important in peoples lives as they recover from the adjustment pains of a large and rapid transition. While the labor market was improving during the years of this study, this trend will not necessarily continue in the near future. China is currently aggressively continuing to liberalize their labor market by relaxing internal migration restrictions. In 2013, the Chinese government announced their goal is to move 100 million more people into cities by 2020 (OECD 2015). The government plans to accomplish their goal by encouraging rural, less educated, people to move to cities – exactly the people who are most vulnerable during times of transition. As of 2010, last data on rural life satisfaction in the Horizon dataset, average life satisfaction in rural China is slightly higher than in urban China. If migration to the cities occurs too quickly resulting in a surplus of low skill labor, a massive drop in national life satisfaction is likely as relatively happy 54 rural people move to cities and struggle to find jobs. Furthermore, a large increase in supply of low skill labor from outside the city will negatively affect the job prospects of workers who currently are urban residents. If the goal of the government is to safeguard the well-being of its entire population over the upcoming years, much attention should be directed towards ways of ensuring lesser educated segments of the population in urban China can find jobs in cities. 3.6. Conclusion This study has two main findings. First, the improvement of the labor market was a large and significant contributor to the increase in life satisfaction in urban China from 2002 to 2012. Furthermore, this was especially important for the people who suffered most during the initial phase of the transition away from a communist system in urban China – people with less than a college degree. Second, when income comparison and habituation are accounted for, increasing incomes did not significantly contribute to the increase in life satisfaction. Although this finding might sound shocking to some because of the rapid income growth during this period, it is consistent with rising financial dissatisfaction in China found by other studies (Brockmann et al. 2009, Bartolini and Sarracino2015) and the finding that worldwide, long run growth in income is not related to increasing subjective well-being (Easterlin et al. 2010). Taken together, these two findings imply that China may face a subjective well-being disaster in the coming years as the government continues to liberalize the national labor market. If a massive wave of less educated workers migrate to cities from rural areas: a large drop in life satisfaction is likely if conditions in the labor market are compromised by an increase in the supply of uneducated workers. To protect the well-being of people in China, it is crucial to make sure unemployment rates don’t rise sharply as the national labor market is liberalized, even if taming 55 unemployment rates comes at the cost of slowing income growth. Results from this study and others show that simply relying on increasing incomes is unlikely to protect well-being. 56 CHAPTER 4. Labor Market Policy and Subjective Well-being during the Great Recession 4.1. Introduction Economic recessions have large negative effects on subjective well-being (Ditella et al. 2003; Wolfers 2003; Helliwell and Huang 2014; De Keulenaer et al. 2014), which can primarily be explained by increasing unemployment (Ditella et al. 2003; Wolfers 2003; Helliwell and Huang 2014). Furthermore, the negative impacts of increasing unemployment have effects beyond just those who are unemployed (Ditella et al. 2003; Helliwell and Huang 2014; Arampatzi et al. 2015). This paper addresses an important question – can labor market policy mitigate the negative impact of a recession on subjective well-being? Furthermore, which types of labor market policies are most effective, and who is most affected by these policies? Data from 23 European countries during the Great Recession are analyzed to address these questions. The labor market policies that are compared in this paper can be divided into two broad categories. The first category includes policies that provide support for people who become unemployed. This category includes two types of unemployment support – programs that assist unemployed workers to find employment (active labor market policy) and income replacement for people who become unemployed (net income replacement). The second category includes policies that restrict the ability of firms to freely adjust their workforce (employment protection legislation). Employment protection legislation for fixed term and non-fixed term contract workers are considered in this paper. The analysis is structured as follows: first, the change in subjective well-being during the Great Recession for people with different levels of education and also for different birth cohorts is 57 identified. Second, using the policy variation across countries, the mitigating effects of labor market policy on the change in subjective well-being during the Great Recession for each group is estimated for each type of policy. If protecting those who are most vulnerable in society is a goal of labor market policy, one would hope that the policy significantly mitigates the negative effect of the Great Recession for the groups of people who suffered the largest decreases in subjective well-being. Prior research has established that labor market policy has a significant relationship with subjective well-being, but the vast majority of studies focus on the relationship with level of well- being. These studies find a significant and positive relationship between the generosity of unemployment support programs and level of subjective well-being (Ditella et al. 2003; Ochsen and Welsch 2012; Boarini et al. 2013; Wulfgramm 2014) and also between level of employment protection legislation and level of subjective well-being (Ochsen and Welsch 2012; Boarini et al. 2013). Overall, there is growing evidence that more generous labor market policies promote higher levels of subjective well-being. But this study focuses on the mitigating effects of labor market policy on subjective well-being during a recession. To date, studies on how labor market policy protects subjective well-being in the context of an economic recession are limited. Only two studies consider the potential mitigating effects of labor market policy on labor market conditions associated with economic recessions. Wulfgramm (2014) finds that policies providing support for unemployed persons reduce the negative effects of becoming unemployed on subjective well-being. This finding is identified using policy variation between countries at a point in time, and also using variation in policy changes within countries over time. The second study provides evidence that unemployment support positively affects the subjective well-being of people who remain employed during economic recessions (Carr and Chung 2014). 58 These two studies make valuable contributions to evaluating labor market policy in the context of a recession, but many questions remain unanswered. Neither of these studies consider the potential mitigating effects of employment protection legislation. Also, neither study includes a model that fully captures the societal-wide mitigating benefits of labor market policy. Wulfgramm focuses only on unemployed people without considering the benefits accrued by the employed, and Carr and Chung only focus on employed people without considering direct benefits to people who are unemployed. Considering benefits outside of the labor force is also potentially important, but has not been previously addressed. Kim and Do (2013) find that a spouse becoming unemployed is significantly related to partner decreases in subjective well-being, which implies the possibility that labor market policy effects could extend beyond the labor force. This study extends the previous subjective well-being literature on the mitigating effects of labor market policy during economic recessions beyond policies that provide support for unemployed persons to also include employment protection legislation. The effects of policy are estimated on people of all occupational statuses, which captures the societal-wide effects of labor market policies. Another contribution of this paper is estimating the differential effects of policy on different groups of people. It is important to know if the groups that suffer the most during recessionary periods are benefiting from, or are being hurt, by labor market policy. The results from the analysis in this paper show that subjective well-being dropped significantly during the Great Recession for the following groups: people with less than a high school education, people with a high school education, youth, and working aged people. For all groups suffering a significant decrease in subjective well-being with the exception of youth, labor market policies significantly affected the magnitude of subjective well-being decline. But not all types of labor market policy protected well-being. Labor market policies providing support to 59 unemployed persons significantly mitigated the negative effect of the Great Recession. Employment protection legislation, however, significantly exacerbated the negative effect of the Great Recession. 4.2. Background The term “subjective well-being” is used in this paper to refer to either self-reported measures of life satisfaction or happiness. These measures are considered comparable because they correlate with the same explanatory variables (Helliwell et al. 2012). Data for both measures are based on questions from surveys where respondents are asked to evaluate their life as a whole, and then report feelings about their life on a numerical scale. Subjective well-being is therefore a comprehensive measure of well-being that individuals define based on their own preferences and life circumstances. Growing evidence supports subjective well-being as a valid and reliable measure of well-being (refer to the 2012 World Happiness Report for comprehensive list of supporting studies). Furthermore, in 2008, a commission comprised of the worlds leading social scientists, including six Nobel Laureates in economics, recommended that governments start officially collecting subjective well-being data (Stiglitz, Sen, Fitoussi 2009). At present day, almost all Organization for Economic Co-operation and Development (OECD) countries are officially collecting subjective well-being data and “many national leaders are talking about the importance of well-being as a guide for their nations” (Helliwell et al. 2015). The motivation to use subjective well-being as the measure of well-being in this paper is based on the sensitivity of subjective well-being measures to both observable and psychic changes that occur during recessions. Previous studies have found accounting for these psychic costs are important in explaining declines in well-being during recessions. This is clearly illustrated by 60 comparing the discrepant findings between studies that infer the well-being costs of economic fluctuations based on observable outcomes and studies that measure well-being using responses to subjective well-being questions. For example, Lucas (1987) finds that the risk associated with economic fluctuations is equivalent to only 0.1% of consumption. He concludes these effects on well-being are only “of second order importance.” In response to the study by Lucas and other similar studies (e.g. Romer 1996), Wolfers (2003) uses subjective well-being data and finds the costs of economic fluctuations on well-being are much larger. Ditella et al. (2003) find that subjective well-being follows macroeconomic movements and the cost of economic recessions extend beyond loss of income and increasing unemployment rates. The authors conclude that “standard economics tends to ignore what appear to be important psychic costs of recessions.” A study by Deaton (2012) specifically measures psychic costs during the Great Recession in the United States. Deaton finds that subjective well-being fell during the Great Recession accompanied with increases in worry and stress. Helliwell and Huang (2014) report similar findings using data from the United States, finding that unemployment indirectly affects people who are not unemployed. They conclude that, “[their findings] suggest that more precise estimation and understanding of the indirect effects of unemployment are essential for any cost- benefit analysis of policies designed to mitigate the economic and social effects of unemployment.” 4.3. Data This analysis uses data collected from 23 countries in Europe. Two levels of data, macro and micro-level, from two time periods are used. The two time periods are defined as pre-recession (peak) and trough of the recession. The dates of micro-level data collection are the latter half 2007 61 for the peak and the middle of 2009 for the trough. These dates are chosen based on peak and trough business cycle dates from the Center for Economic Policy Research: Business Cycle Dating Committee. Macro-level data is matched to the micro-level data. Quarterly macro level variables are paired with the micro-level data when possible, and yearly data is paired otherwise. The micro-level data is nationally representative with weighting. Weighting is also adjusted to give equal weight to each country in the analysis. The macro-level variables reflect national conditions in each country. People of all occupations, inside and outside of the labor force, are included in the sample that is analyzed. All people over the age of 15 and under the age of 65 (a common retirement age in Europe) at the peak period are included in the analysis. 4.3.1 Macro-level data The macro-level data consists of macroeconomic variables and policy variables. The macroeconomic variables included in the analysis are log GDP per capita, unemployment rate, and inflation rate. They are all taken from the OECD (2014a) and are measured in the standard way. The macroeconomic variables are included as controls to account for the different levels of recession severity experienced by different countries. The labor policy variables are measured at the peak and trough time period. The OECD carefully constructed all of the policy variables to make them as comparable across countries as possible. The policy variables that are included in the analysis are listed as follows: Net replacement rate (NRR) measures the average proportion of net in-work income that is maintained for 60 months when someone becomes unemployed (OECD 2014b). NRR is measured in percent with a scale of 0-100. 62 Active labor market policy (ALMP) is the percent of GDP spent on active labor market policy (OECD 2014c). Active labor market policy is defined as programs that help unemployed people find new jobs. This includes job placement services, benefit administration, job training, and job creation programs. Because unemployment and GDP changed during the great recession for all countries, the active labor market policy variable inserted into the analysis is converted to expenditure per unemployed person measured in thousands of 2005 US dollars. Employment protection legislation (EPL) is a synthetic measure from 0 to 6 that reflects the strictness of regulations governing the dismissal of workers in non-fixed contract jobs. It covers regulation of individual and collective dismissals (OECD 2014d). 0 represents the least and 6 represents the most strict regulations. The employment protection summary indicator for temporary work (EPL-T) covers fixed term contract jobs (OEDC 2014d). It is a measure of how freely firms can use fixed term contracts, that is, it reflects how many times firms can renew fixed term contracts, the types of jobs firms can use fixed term contracts to hire workers, and also regulations on the duration of fixed term contracts. EPL-T is also measured on a 0 to 6 scale, where 0 represents the least strict and 6 represents the most strict regulations. Not all policy variables cover all 23 countries. In the primary analysis the maximum number of countries allowed by the policy variables included in each regression are used. Robustness checks to see if the results are dependent on country selection are run accordingly. A full list of the policy variables used in the analysis can be found in the appendix C (Table C-1). 63 4.3.2 Micro-level data Micro-level data is taken from the Eurobarometer (European Commission 2007, 2009). Subjective well-being is measured as the response to a question on how the respondent feels about satisfaction with their life as a whole. The question is worded: “On the whole, are you very satisfied, fairly satisfied, not very satisfied, or not at all satisfied with the life you lead? Would you say you are .......?” (coded 4-1, with 4 being very satisfied and 1 being not at all satisfied). The analysis examines the differing experience of groups by level of education and by birth cohort. Level of education includes three groups and is defined by highest level of education achieved. The three groups are: less than a high school education, a high school education, and a four-year college degree or more. The high school group includes trade school because a small percentage of people fall into this category. Cohort is also divided into three groups defined by age at the peak of the recession. The age groups are: 15 to 24, 25 to 44, and 45 to 64 years of age. The groups are referred to as youth, middle age, and older age, accordingly. A dummy variable for each group is created. The groups are created based on ex-anti expectations about who would be affected most by the great recession. Micro level controls are included to account for the change in individual’s lives during the Great Recession that could vary in different countries. The micro level control variables included in the analysis are variables for a person’s employment status, marital status, gender, age (when estimating effects by level of education), and level of education (when estimating effects by cohort). Students were dropped from the sample when estimating policy effects by level of education because it is impossible to determine a student’s final level of educational attainment. Country dummies are included in the regression to control for country characteristics that are fixed. 64 4.4. Methods A multi-level model is used in the analysis which allows the interaction between macro-level data, specifically policy, with mico-level data, specifically group dummies. Policy variables changed in the period this study covers. If countries adjusted their labor market policy as a function of the severity of the Great Recession, the estimates of policy effects would be biased. To control for endogenous policy change, the trough policy variables are instrumented with the peak period policy variables. This IV approach is based on the fact that the Great Recession was an unpredicted event, so the policy variables during the peak period are exogenous to changes that occurred during the Great Recession. Life satisfaction is the dependent variable. The primary specification is as follows: 𝐿𝑆 #hF =𝛼+𝛾′𝑐 h +𝛽 O 𝑌 hF +𝛽 C 𝑋 #hF +𝛽′ N 𝐺 @hF +𝛿 C 𝐼′ FLC 𝐺 @hF +𝛿 N #o 𝐼′ FLC 𝑃 h,FLC 𝐺 @hF +𝜀 #hF (4.1) i = 1,….,N (individual) j = 1,….,M (country) t = 0,1 (peak and trough) 𝐿𝑆 #hF = life satisfaction of individual i in country j at time t 𝑐 h = Mx1 vector with jth entry=1 and all other entries=0 𝑌 hF – macroeconomic variables of country j at time t 𝑋 #hF – characteristics of individual i in country j at time t 𝐺 @hF = (𝐼 #hF C 𝐼 #hF N 𝐼 #hF B )′, indicators for group of individual i in country j at t 𝑃 h – policy variable of country j at time t=1 (instrumented with policy at t=0) 𝛾 =(𝛾 C ,…,𝛾 u )′ with 𝛾 C =0 (France is base group) 𝛽 N =(𝛽 NC 𝛽 NN 𝛽 NB )′ with 𝛽 NC =0 (less than high school or youth in t=0 is base group) 65 Labor market policy variables are inserted individually into the model. Each estimate of policy effects is therefore based on an individual regression. Two stage least squares, using OLS for both stages, is the method of estimation 23 . Standard errors are robust and are clustered at the country level. This model controls for country fixed effects, differential changes in macroeconomic conditions, change in individual characteristics, and group level life satisfaction. Policy variables only enter into the equation in the trough period, but the level effect of policy is captured by the country dummies. The parameter 𝛿 C represents the conditional mean change in life satisfaction for each group. The parameter of interest is 𝛿 N , which reflects the marginal effect of a policy variable on the change in life satisfaction between peak and trough for each group. This identifies if a specific labor market policy mitigated or exacerbated the negative effect of the Great Recession on life satisfaction (this study’s subjective well-being measure). A positive parameter estimate would indicate a mitigating effect on subjective well-being, and a negative estimate an exacerbating effect. 4.5. Expectations – a model The way labor market policy is expected to affect the change in subjective well-being during the Great Recession depends on the type of labor market policy. In this section, simple models representing expectations about how labor market policy interacts with changes that occur during a recessionary period are presented for the two categories of policy that are included in this analysis. The simple models are represented visually in tables, with the first column of the tables representing important changes that occur during recessions, and the second column representing how these changes affect subjective well-being. The changes that occur during a recession that 23 Frijters (2004) finds that the difference between treating subjective well-being questions as cardinal or ordinal makes little difference. OLS is preferred in this analysis to make interpretation of the parameter estimates easier. 66 are included in the model are unemployment and stress from labor market conditions. Stress from labor market conditions is assumed to affect people who are not unemployed. These changes are included based on the previous findings that unemployment during recessions negatively affects the subjective well-being of unemployed people, and also people who are not unemployed (Ditella et al. 2003; Arampatzi et al. 2014). The arrows in the model represent the magnitude and direction of changes that occur during the recession (column 1) and their effects on subjective well-being (column 2). The dark arrows represent the changes with no policy, ceteris paribus, and the lighter arrows represent how policy is expected to affect each change. There are likely many more ways labor market policy interacts with subjective well-being during recessionary periods, but the dynamics in Figure 1 and 2 are included for two reasons. First, findings from previous studies have shown that these dynamics are important. Second, some of these dynamics can be tested using the dataset analyzed in this paper. 4.5.1 Unemployment support programs Figure 4-1. Theoretical relationship between unemployment support programs and SWB during recession 67 Figure 4-1 presents the expected relationship between unemployment support programs and change in subjective well-being during recession. As represented in row 1, unemployment support programs are not expected to affect the change in unemployment during recession. But, based on the findings by Wulfgramm (2014), unemployment support is expected to mitigate the negative effect of unemployment on subjective well-being. This relationship is expected to also have effects on the changes represented in row 2. Because the losses in well-being associated with unemployment are reduced, the stress from labor market conditions is expected to also be reduced. Subsequently, the loss of subjective well-being due to stress from labor market conditions is also reduced. The relationship in row 2 are based on the findings from Carr and Chung (2014). The net expectation is that unemployment support programs will mitigate the negative effect of the Great Recession on subjective well-being. 4.5.2 Employment protection legislation Figure 4-2. Theoretical relationship between employment protection legislation and SWB during recession 68 Figure 4-2 presents the expected relationship between employment protection legislation and change in subjective well-being during recession. One of the ways employment protection is supposed to protect well-being is by ensuring workers they cannot be freely laid off. This is especially relevant for EPL, policy for protecting non-fixed term workers. Row 1 presents the expectation that employment protection reduces unemployment because firms cannot freely dismiss workers. This expectation is based on how employment protection legislation is supposed to protect workers 24 . If fewer people are negatively affected by unemployment, then the overall decrease in subjective well-being would be mitigated. But, the cost of employment protection reducing the ability of firms to freely adjust their workforce is an implied increased the cost of hiring workers. If it is more costly for firms to hire workers, it would be harder for people to get hired. Row 2 presents the theoretical effect of increased cost of hiring workers on stress from labor market conditions. The stress from labor market conditions rises with employment protection legislation because it is harder to find jobs for unemployed people. The result is a larger drop in subjective well-being. The net effect of employment protection legislation during a recession will be mitigating if the effects in row 1 dominate the effects in row 2, but the net effect will be exacerbating if row 2 dominates row 1. 24 EPL could also increase unemployment in the long run because firms are less willing to hire workers, but the time frame this study considers is less than two years, so this effect is not considered in the analysis. 69 4.6. Results Figure 4-3. Mean changes in life satisfaction from peak to trough of Great Recession, by group Which groups suffered the largest drop in subjective well-being during the great recession? Figure 4-3 plots the level of subjective well-being during the peak and the trough divided by education and birth cohort 25 . The groups by level of education that suffered the largest drops in subjective well-being during the Great Recession were people with less than a college degree. All cohort groups suffered a significant drop in subjective well-being, with the largest drop experienced by the youth group (less than 25 years old at peak). The magnitude of the decreases in subjective 25 Students are omitted from the graph plotting average life satisfaction divided by level of education because they are dropped from the primary analysis when estimating the policy effects on groups divided by level of education. 70 well-being were large. For all groups experiencing statistically significant declines in subjective well-being, the size of the decline from peak to trough is approximately at least as large as the mean level difference in subjective well-being between people with less than a high school degree and people with a high school degree during the peak period. Did labor market policy mitigate the negative impact of the Great Recession on the groups that suffered the most? The regression analysis addresses this question. The marginal effect of labor market policy variables on change in subjective well-being during the Great Recession estimates are presented for each policy. In all results tables, each row represents a separate regression. To help interpret the size of the estimates, the mean change in life satisfaction by each group is included in each regression, and also standardized beta coefficients are presented in the tables. 71 4.6.1 Policy effects by level of education Table 4-1 presents policy effects on groups defined by level of education. The positive and significant NRR and ALMP estimates presented in columns 1 and 2 indicate that unemployment support programs significantly mitigated the negative effect of the Great Recessions for the groups that suffered the most, people with less than a high school degree and people with a high school degree. This result confirms the expectations presented in Figure 4-1. None of the unemployment policies had a significant effect on the group that did not suffer a significant decline in subjective well-being during the Great Recession, people with a college 72 education. The null result for the college educated group has an encouraging interpretation. The college educated group most likely contributes the most funding to the safety net programs by paying higher taxes, so one might worry that their decreased income due to funding unemployment support would have negative affects during a recession. This negative effect could possibly dominate any positive effect of the unemployment support programs, especially because high skilled workers are least likely to loose their jobs. The null results indicate, however, that the net effect on college educated people was not negative. The negative and significant parameter estimates for employment protection legislation, presented in columns 3 and 4, indicate that both types of protection policy exacerbated the negative effects of the Great Recession for almost all groups. Unemployment controls are purposefully not included in these regressions so any positive effect employment protection had on curbing unemployment will be captured in the parameter estimates. EPL had an exacerbating effect for all groups except the college educated, and EPL-T had an exacerbating effect for all groups. The magnitude of the parameter estimates is more comparable between EPL and EPL-T because they are measured on the same scale and the regressions include the same countries. Overall, the size of the estimates for EPL are slightly larger, indicating a slightly larger exacerbating effect. This is confirmed judging by the parameter estimates and the standardized beta coefficients. 73 4.6.2 Policy effects by cohort Table 4-2 presents policy effects on groups defined by cohort. An important finding presented in the first row of results is that no policies significantly benefited the youth cohort. NRR and ALMP likely had no effect on the youth group because to qualify for these unemployment benefits, a person must contribute to the welfare state for a minimum number of months. The insignificant relationship of EPL on youth can presumably be explained by the fact that EPL applies to career type jobs, which are less relevant for many youth. EPL-T, however, applies to the types of jobs relevant for many youth, fixed term work. The effect of EPL-T is negative and significant. Excluding the youth group, the overall findings from Table 4-2 are in 74 agreement with the findings from Table 4-1 – unemployment support programs had a mitigating effect where as employment protection programs had an exacerbating effect. 4.6.3 Which matters more, NRR or ALMP? NRR and ALMP polices both provide support to unemployed persons, and both have a mitigating impact on the negative effects of the Great Recession. But which policy has a larger 75 and more robust mitigating effect? To address this question, NRR and ALMP are entered into the same regression equation. Table 4-3 presents the results by level of education and by cohort. It appears that NRR has a larger and more robust mitigating effect. When both variables are included in the same regression, the statistical significance of the ALMP parameter estimates disappear. This is true for all groups defined by level of education (column 1) or by cohort (column 2). The statistical significance of the NRR parameter estimates do not disappear for any groups. Table 4- 3 also presents the standardized beta coefficients for both NRR and ALMP parameters. Based on the standardized beta coefficients, one standard deviation change in NRR can explain much more of the change in subjective well-being than one standard deviation of ALMP. This is also true when comparing the size of the beta coefficients in Tables 4-1 and 4-2. The results for NRR and ALMP from all tables should be interpreted with caution, however, because these variables are significantly correlated. Therefore, it is hard to disentangle the individual effects of each variable. What can be concluded is that programs that help the unemployed, in general, have a mitigating effect for many of the groups, and it appears that the relationship between NRR and subjective well-being is more robust than ALMP. 4.6.4 Explaining employment protection effects The significant exacerbating effects of employment protection indicate that if employment protection had any positive effects, these effects were dominated by negative effects. This next section attempts to use the theory presented in Figure 4-2 to disentangle the positive and negative effects of employment protection. 4.6.4.1 Benefits from reducing unemployment. Table 4-4 and Table 4-5 presents results that attempt to isolate the effect described in row 1 of Figure 4-2. Specifically, that employment 76 protection could have a mitigating effect on subjective well-being by reducing the increase in unemployment during a recession. To isolate this effect, employment protection parameter estimates are compared when controls for unemployment are either included or excluded. If employment protection is mitigating the negative effect of the Great Recession on subjective well- being by limiting the increase in unemployment, then the parameter estimates will be larger when controls are included for unemployment. This is because positive effects of reduced unemployment will now be controlled for, and the parameter estimates will only reflect the relationship between employment protection and subjective well-being holding unemployment constant. Table 4-4 presents the comparisons for the analysis of groups defined by level of education, and Table 4-5 presents the comparisons for groups defined by cohort. In both tables, when unemployment controls are included, the parameter estimate for EPL is larger, indicating the relationship in row 1 of Figure 4-2 exists. The EPL-T estimates does not change the same way when unemployment controls are included. This observation indicates that the relationship in row 77 1 does not exist for EPL-T. This is not surprising, however, because EPL-T primarily protects workers by limiting the ability of firms to freely use fixed term contracts. 4.6.4.2 Costs of negative feelings about future job situation. To identify the effects employment protection on stress from labor market conditions, presented in row 2 of Figure 4-2, a new dependent variable is introduced, expectations about future job situation. This new dependent variable measures how people feel about their job situation in the coming 12 months. The question worded: What are your expectations for the next twelve months: will the next twelve months be better, worse or the same, when it comes to [Your personal job situation]? (coded 1 – worse, 2 – same, 3 – better). The variable is recoded to take the value of 1 if the respondent chooses “worse” and 0 otherwise. This variable should reflect how people feel about their ability to find a new job if they are unemployed, or their worries about becoming unemployed. The assumption is that this measure is related to overall stress people experience due to conditions in the labor market. The new variable is then inserted into equation (4.1) as the dependent variable. The regressions are re- run using the same specification used for the results presented in Tables 4-1 and 4-2. 78 The parameter estimates from the future job situation analysis are presented in Table 4-6 and Table 4-7. The second stage is estimated using a probit because the new dependent variable is binary. A positive estimate represents employment protection exacerbating negative feelings about future job situation during the great recession. Table 4-6 shows that for all groups except people with high school education, the estimates are positive and significant. Furthermore, all but one of the policy effect estimates has a positive coefficient. These findings indicate that, for the groups defined by level of education, in general employment protection had an exacerbating effect on negative feelings about future job situation. 79 Table 4-7 presents the effects of employment protection by cohort. The lack of significant coefficients for all of the groups for EPL and all but the old cohort for EPL-T implies that the effects of employment protection on views of future job do not depend much on age. It is worth noting, however, that for EPL-T all the estimated signs are positive, indicating an exacerbating effect, although not significant in the case of anyone younger than the old group. Although it is not a perfect comparison, the results from this section are consistent with previous findings that the strictness of employment protection policies are negatively related to feelings of job security (Clark and Postel-Vinay 2009). But, the results from this analysis using expected future job situation do have limitations. Expected future job situation is not a perfect measure of stress related to conditions in the labor market. Also, unlike subjective well-being, expected future job situation is a subjective measure that has not been rigorously tested for validity and reliability. Expected future job situation is included in this analysis as suggestive evidence that 80 employment protection policies do have palpable negative effects on the ability of the market to allocate labor. 4.6.4.3 Summary: costs of negative feelings about future job situation. The results from the primary analysis (Tables 4-1 and 4-2) do not impose structure on how employment protection is affecting subjective well-being. This section provides insight about the mechanisms through which employment protection interacts with subjective well-being. Overall, the results from this section provide evidence that EPL may have some benefits by limiting unemployment increases, but these benefits are dominated by the negative effects. Evidence suggests that some of these negative effects are associated with poor expectations people have about their future job situation. EPL-T did not have any benefits from limiting unemployment, and seems to be more robustly related to negative feelings about future job situation. 4.6.5 Robustness checks The findings from the main analysis have three potential weaknesses that can be addressed. First, the countries included in each regression vary depending on the policy variable included. The countries that have ALMP observations are a subset of the countries that have NRR observations. To check if country selection is pushing the main results for NRR, the countries included in the NRR regressions are limited to countries that have ALMP observations. Column 3 in Table 4-8 and Table 4-9 presents the robustness results. Column 1 in Table 4-8 and 4-9 presents the results from the primary analysis. The parameter estimates for the effect of NRR change, but based on the sign and statistical significance of the estimates, the overall message of the analysis remains the same. Therefore, it can be concluded that NRR results are robust to country selection. The countries included in the EPL and EPL-T regressions are the same, so this issue is not addressed for employment protection. 81 82 83 The second issue that can be addressed is endogenous policy choice of countries. The IV approach does control for any endogenous policy change that occurred during the Great Recession, but it is possible that countries chose their labor market policy before the recession as a function of vulnerability to a negative economic shocks. To test if this endogenous policy choice issue is biasing the estimates, the analysis is re-run omitting all macroeconomic controls and individual level controls besides group dummies and policy variables. The intuition behind this test is, without full controls, if severity of the Great Recession in each country is correlated with the policy variables, the estimates on policy effects will change. The results from regressions including full controls (columns 1, 4, 6, and 8 from Tables 4-8 and 4-9) and with minimal controls (columns 2, 5, 7, and 9 from Tables 4-8 and 4-9) are strikingly similar, supporting a conclusion that the results are not biased by endogenous policy choice. The third potential weakness of the analysis is that the policy variation is at the country level. The small number of countries included in the analysis allows for a single outlier country to have a potentially large impact on parameter estimates. A country could be an outlier because of an unusual experience during the Great Recession, or a policy variable could potentially not reflect the outlying country’s policy in a comparable way to the other countries 26 . To address the potential problem that a single outlier country is pushing the main results, a single country is omitted and the analysis re-run using the same methods used for the results presented in Tables 4- 1 and 4-2. This is repeated for each country so there are the same number of regressions re-run as number of countries included in the primary analysis. The results are presented in graphical 26 The OECD was very carful to make policy variables comparable across countries, but there are differences in policy implementation across countries that are potentially missed by the policy variables included in this study. One example is strictness of regulations governing access to benefits is not considered in NRR and ALMP. Another difference between countries could be social norms around receiving benefits. 84 Notes: Each point represents parameter estimates using the same methods used in Table 1, but omitting a single country for each point. The horizontal lines represent the original parameter estimates from Table 1. 90% confidence intervals are included for each estimate. Figure 4-4. Robustness of results to omitting individual countries, groups by level of education 85 Notes: Each point represents parameter estimates using the same methods used in Table 2, but omitting a single country for each point. The horizontal lines represent the original parameter estimates from Table 2. 90% confidence intervals are included for each estimate. Figure 4-5. Robustness of results to omitting individual countries, groups by cohort 86 format in Figure 4-4 and 4-5. In each graph, the gray lines represent the original parameter estimates for marginal effect of policy on peak to trough change in subjective well-being for each group. The points in each graph represent each new parameter estimate when omitting an individual country. Confidence intervals at the 90% level are included for each new parameter estimate. The results from Figure 4-4 and 4-5 show that in the vast majority of times a country is omitted, the parameter estimates do not significantly change from the estimates presented in Tables 4-1 and 4-2. Furthermore, the statistical significance of the estimates rarely change when omitting a country. The one notable exception to this is the statistical significance of the estimates for EPL- T seem to not be robust if Turkey is omitted. When Turkey is omitted, the significance of the parameter estimate is lost in many cases, although the estimate is not statistically different than the estimates presented in Tables 4-1 and 4-2. Therefore, the results for EPL-T should not be weighed as heavily as the results for the other policy variables included in this study. 4.7. Conclusion The results from this study contribute to our understanding of how labor market policy can mitigate, or exacerbate, the negative effects of a recession on well-being. Expectations about the mitigating effects of unemployment support policy based on previous literature are generally confirmed (Carr and Chung 2014; Wulfgramm 2014). This study provides new evidence, however, that these mitigating effects are not experienced by youth, a group that suffered a large and significant decline in subjective well-being during the Great Recession. The finding that employment protection legislation exacerbates the negative effect of the Great Recession for many groups is also new. Any benefits employment protection provides by keeping people employed 87 are dominated by negative effects. Suggestive evidence is provided that the negative effects are due to employment protection imposing rigidities on the labor market, making people less optimistic about their future job prospects. The finding that NRR has a more robust and likely larger mitigating effect than ALMP is relevant for the current debate on labor market policy in Europe. There has been a turn away from NRR unemployment insurance in favor of ALMP programs in recent decades. The findings from this study do not support this policy trend. Evidence is provided that in the short term, NRR is more effective in protecting well-being than ALMP during economic recession. These findings support the findings of Wulfgramm (2014), who comes to the same conclusion. Both this study and the study by Wulfgramm focus on short term effects, however, and the findings should not be extrapolated to the long term. It is possible that ALMP policies do have better long run outcomes. If a policy goal of a country is to protect well-being during recessionary periods, the results from this study support a Denmark style approach to labor market policy; impose minimal restrictions on the ability of firms to adjust their workforce, but provide generous support for people who become unemployed. Additionally, policies specifically targeting youth should be considered. 88 CHAPTER 5. The Impact of Social Assistance Programs on Subjective Well-being: Regression Discontinuity Evidence from a Cash Transfer Program in Colombia 27 5.1. Introduction How can governments promote increases in well-being? The growing availability of subjective well-being data provides opportunities for social scientists to explore this issue empirically without having to make any assumptions on preferences. Findings from the subjective well-being literature suggest that social assistance programs could be a way for governments to raise the subjective well-being of their populations. However, the vast majority of studies in this literature do not provide causal evidence. Furthermore, although subjective well-being tends to be lower and the majority of the world’s people live in less developed countries (Helliwell et al. 2012), the majority of studies on the subjective well-being benefits of social assistance programs are focused on developed countries. Conditional cash transfer (CCT) programs have become popular social assistance programs in developing countries over the past 20 years. These programs provide a cash transfer to families conditional on the families meeting certain requirements. Fiszbein and Schady (2009) provide an extensive and detailed review of research on programs that were implemented in developing countries. They find that conditional cash transfer programs generally raise consumption levels, reduce poverty, and increase the use of health and education services. However, the effects on health and education outcomes are mixed and there is some evidence of decreased labor market participation by recipients of the cash transfers. Despite the widespread adoption of conditional 27 Coauthored with Titus J. Galama and Juan E. Saavedra 89 cash transfer programs, there is no study to our knowledge that asses the impact on subjective well- being. This paper aims to fill this gap in the literature by evaluating the impact of Familias en Acción (FIA), the largest CCT program in Columbia, on subjective well-being. The term “subjective well-being” is used to refer to either self-reported measures of life satisfaction or of happiness. These measures are considered comparable because they correlate with the same explanatory variables (Helliwell et al. 2012). Data for both measures are based on questions from surveys where respondents are asked to evaluate their life as a whole, and then report feelings about their life on a numerical scale. Subjective well-being is therefore a comprehensive measure of well-being that individuals define based on their own preferences and life circumstances. Growing evidence supports subjective well-being as a valid and reliable measure of well-being (refer to the 2012 World Happiness Report for comprehensive list of supporting studies). Furthermore, in 2008, a commission comprised of several of the worlds leading social scientists, including six Nobel Laureates in economics, recommended that governments start officially collecting subjective well-being data (Stiglitz et al. 2009). At present day, almost all OECD countries are officially collecting subjective well-being data and “many national leaders are talking about the importance of well-being as a guide for their nations” (Helliwell et al. 2012). There is a small but growing literature examining the affect of social assistance programs on subjective well-being. A number of studies find that the generosity of social assistance is positively associated with subjective well-being in developed countries (Boarini et al. 2012, Rothstein 2010, Pacek and Radcliff 2008a and 2008b, Wulfgramm 2014, Carr and Chung 2014, Helliwell and Hwang 2008, Ditell et al. 2003). But, the vast majority of these studies do not have causal interpretations. One notable exception is the Oregon Health Insurance Experiment which 90 expanded access of public health insurance to low-income adults. An evaluation of the experiment found that it an increase in subjective well-being (Finkelstein et al. 2012). But, it is not clear that these findings from developed countries are applicable to developing countries because social assistance programs and life circumstances differ in between developed and developing countries. In the developing country context there are fewer studies, but the studies that exist provide causal evidence that social assistance programs had a positive impact on subjective well-being. In Morocco, Devoto et al. (2012) implemented a randomized controlled trial (RCT) and found that improving access to in-house water access positively impacted subjective well-being even though there was no positive effect on health outcomes observed. The increase in subjective well-being was presumably through increased social integration and increased leisure time allowed by time saved from families not having to walk for water access. Another study, which took place in Mexico and used geographic variation in implementation to identify program effects, found that replacing dirt floors with concrete floors had a positive impact on the subjective well-being of adults in the households (Cattaneo et al 2009). The study also found that child health improved, which the authors argue was the reason for the improved adults’ subjective well-being. However, the social assistance programs these studies evaluated are not as widespread as CCT programs. With well-being becoming increasingly a goal of policy making, information on the causal effects of programs, such as CCTs, that impact large segments of society is of significant importance. Two studies are more closely related to conditional cash transfers. The study closest to our study examines the effects of a randomized control trial unconditional cash transfer program in Kenya that targeted poor, rural households. The program gave a sum of money roughly equal to two months’ worth of average household consumption to participants without any restrictions. The effects of the unconditional cash transfer resulted in a .17 standard deviation increase in 91 happiness and a .15 standard deviation increase in life satisfaction (Haushofer and Shapiro 2013). This result is consistent with studies in developed countries linking exogenous income gains to increases in subjective well-being (Kuhn et al 2011, Boyde-Swan 2013) or improvements in mental well-being (Gardner and Oswald 2007). A study in Malawi that examines psychological measures that are related to subjective well-being provides evidence that the positive effect of an unconditional cash transfer might not translate to programs where cash is conditionally rewarded. Baird, De Hoop, and Özler (2013) compared the psychological effects of an unconditional cash transfer program with the effects of a conditional cash transfer program for adolescent schoolgirls. Compared to a random control group, both the unconditional and the conditional cash transfer group enjoyed a decrease in psychological stress after receiving the cash transfer, but the conditional cash transfer group maintained a higher baseline level of stress, presumably due to the stress of meeting the conditions for the cash transfer. This study focuses on the first wave of the FIA urban expansion in Colombia. The urban expansion was implemented in large cities 28 , started in 2007 and continued through 2010. Families eligible for FIA must have at least one child under the age of 18 and be under a specified level of poverty as measured by a government assigned poverty score. Eligible families receive a cash transfer, conditional on the children in their household meeting school attendance or doctor visit requirements. This study implements a regression discontinuity research design that exploits exogenous program participation around the cutoff in program eligibility. The regression discontinuity research design allows us to make causal inferences about the effects of FIA on a host of outcomes including subjective well-being. We find that FIA increased many objective and subjective dimensions of life that are positively associated with subjective well-being. Households 28 Large cities is defined as cities with a population of 100,000 or greater 92 that participated in FIA enjoyed higher incomes, spent more on items to make their lives more comfortable, enjoyed improved self reported health, and had better labor market outcomes. There was a subsequent increase in the satisfaction with a handful of life domains: income, food, work, and the ability to help others. Subjective well-being, measured as a response to an evaluative life satisfaction question asked to the household head, increased one standard deviation. The increase in subjective well-being we observe is large, but our sample seems to be too small to find a robust statistically significant increase. Despite the lack of robust statistical significance, we argue that our analysis provides evidence that FIA participation caused an increase in subjective well-being. The paper is structured as follows. Section 5.2 presents an overview of the FIA program. Section 5.3 describes the data used in our primary analysis. Section 5.4 describes our methods and includes a comparison of the data we use for our primary analysis with official baseline data taken from the Colombian census. Section 5.5 presents our results, and section 5.6 concludes. 5.2. Context – an overview of Familias en Acción and SISBEN Familias en Acción (FIA) started in 2001 as a pilot program in small cities (population less than 100,000) across Colombia. In 2007, FIA expanded to large cities (population over 100,000) and the first wave of expansion lasted until 2010. Included in the first wave of expansion were Medellin, Bogota, Barranquilla, Bucaramanga, Yopal, Monteria, Pasto, Pereira, Villavicencio, Ibague, Neiva, Santamarta, and Sincelejo. The first wave of the FIA’s expansion to large cities is the focus of this study. FIA targets poor households with children that are school aged. Eligibility for the program was based on two criteria: 1 – families must have children under the age of 18 in the household, 93 and 2 – the family’s government assigned SISBEN 29 score, a measure of poverty, must be below 11. SISBEN scores are calculated based on the education level of members of the household, the occupational status of household members, the material living standard of a household’s dwelling, and various other demographic characteristics of the household. Information about the households is gathered from the national census survey and an algorithm generates the SISBEN score. The specific algorithm used to calculate the score is kept private to prevent households from strategically answering the survey questions used to manipulate their SISBEN score. SISBEN scores range from 0 to 100 and can take any value in between. About 50% of poor households in Colombia are chosen to complete the national census survey, but households can request to be surveyed if they are not initially chosen. The SISBEN score determines eligibility for social programs in Colombia such as subsidized health care and childcare in addition to FIA. As a result, almost all poor households participate in the national census. The SISBEN scores used to determine eligibility for the first wave of the FIA urban expansion were generated from the 2006 national census and were not updated during the first wave of expansion. If a household is eligible for FIA subsidies, the child/children in the household must meet certain conditions for the household to receive benefits. These conditions depend on the age of children in a household. Children younger than one must see a doctor for checkups every two months. Children older than one, but younger than two, must see a doctor for checkups three times a year. Children older than two, but younger than seven, must see a doctor for checkup twice a year. Children older than 11 years must have at least an 80% school attendance, which is verified every two months. 29 Sistema de Potenciales Beneficiarios para programas sociales 94 The benefits also depend on the age of children. Households receive benefits for each child in the household that meets the conditions. Benefits are Col$50,000 per month for each child under age 7, and Col$25,000-60,0000 per month for each child in grades 6-11. The subsidy figure for children in grades 6-11 reported here is the average amount across regions, where the subsidy varies. The subsidy for children in grades 6-11 is paid for 10 months of the year. Children in grades 2-5 do not receive a subsidy, but if the family only has children aged 7 to 11 in the household, the children receive a Col$20,000 subsidy. The benefits of FIA are substantial for the families that qualify. GDP per capita (PPP, 2005 Col$) for the entire country in 2006 was about Col$850,500 per month, and the average monthly income of the households that received FIA support was under Col$546,415 before receiving support (Méndez et al. 2011). 5.3. Data and Descriptive Statistics Three datasets are used in this study. Data for the primary analysis are from the 2010 Colombian National Quality of Life Survey (CNQLS). We also use data from the 2006 national census and the 2007 National Planning Department FIA baseline survey to validate the accuracy of the 2010 CNQLS data and use as a baseline reference. The 2010 CNQLS and the 2006 national census are both publically available at the Colombian National Administrative Department of Statistics (DANE) website (http://www.dane.gov.co). The CNQLS is modeled after the World Bank’s Living Standards Measurement Survey (http://go.worldbank.org/WKOXNZV3X0) and is collected by the Colombian National Administrative Department of Statistics. Before 2010, in years 1993, 1997, 2003, and 2008, the CNQLS was also administered. We chose the 2010 CNQLS for three reasons. First, the 2010 CNQLS includes a life satisfaction question. Second, 2010 marks the final year of the FIA urban 95 expansion, so we are able to evaluate the effects of the full first wave of urban expansion on a population that had not been previously exposed to the program. And thirdly, the 2010 CNQLS includes all the variables necessary to re-create the household SISBEN scores (the 2008 CNQLS does not). We need to re-create the SISBEN scores for families because their official SISBEN scores are not included in the CNQLS. All of the data for the 2010 CNQLS were gathered in 2010 and are nationally representative. The data from the CNQLS is collected at the individual and at the household level. For each household, the head of the household is interviewed and provides information about him- / herself and the other members of the household. For the majority of variables, data is gathered for every member of the household. But some variables that are important to our study’s aim are only available at the household level (e.g., household expenditure on various types of goods), or are only asked of the household head (e.g., subjective satisfaction with domains of life and life satisfaction). Panel 1 of Table 5-1 compares the household level characteristics of households located in the 13 FIA urban expansion cities sampled in the 2006 national census (baseline) with the 2010 CNQLS. By almost all measures, the living levels of the households sampled in 2010 are much higher than the households sampled in 2006. There are differences between the characteristics of households in the two datasets for three primary reasons. First, unlike the 2010 96 CNQLS, the sampling in the 2006 national census was not designed to be nationally representative; as Panel A of Figure 5-1 illustrates, poorer households were sampled at a higher rate in the 2006 national census compared to the 2010 CNQLS. The Colombian government used the responses from the census to generate SISBEN scores, which are used to determine qualifications for a number of social assistance programs, so they purposefully oversampled households from poor 97 neighborhoods. Also, households could request to be surveyed in the census. Many social assistance programs require SISBEN scores, so poor families had incentive to request inclusion in the census. Second, due to previous social assistance programs that used SISBEN scores to determine eligibility, households may have manipulated their answers to appear poorer. In fact, in the census of the poor used to calculate previous SISBEN scores from 1994 and 2003, there was strong evidence of manipulation by survey respondents (Camacho and Conover 2010). The variables used to generate the SISBEN scores based on the 2006 national census were changed to address the problem of manipulation, but respondents still may have attempted to appear poorer. Third, Colombia experienced rapid economic growth between 2006 and 2010 and the government aggressively improved infrastructure. The focus of our study is on poor households with children in urban Colombia, so only a subset of households from each dataset is included in the analysis. Therefore, we limit the data in our analysis further than all households in the 13 cities that were included in the first wave of the FIA urban expansion. Additionally, we only include households with children age 18 and younger in our analyses to ensure that the comparison households that do not receive FIA support are as similar as possible to treatment households that do receive FIA support. We further limit the data to households with SISBEN scores below 24 (for reference, the FIA SISBEN eligibility cutoff is 11). This leaves 629 households in our sample. Of the households in the sample, 25.6% of households received FIA support in the year preceding the 2010 CNQLS. There were two ways of distributing payments in the first wave of the FIA urban expansion. In 8 of the 13 cities, a portion of the monthly FIA subsidy was withheld and at the end of the year the withholdings were paid to the families in a lump sum. In a minority of cities, families received the full monthly FIA subsidy. Payments were made to the mothers in the household via a wire 98 transfer to their bank account. Due to the sample size of families around the SISBEN cutoff, we group both ways of distributing payments together in our analysis. Panels B, C, and D in Table 5-1 show how household characteristics change in the 2006 national census and the 2010 CNQLS as data is progressively restricted. Panel C presents the households that are included in our analysis. It can be seen, from comparing the average characteristics of households in panel C to households in panel A, the households in our analysis are the poorer households in the 13 FIA urban expansion cities. Panel D shows the characteristics of the households that are eligible to participate in FIA. These households have the lowest living levels, which is illustrated by comparing panel D with the other panels. Panel D also provides evidence that the eligible households in the official 2006 national census data are similar to the eligible households in the 2010 CNQLS data. That is, that our treatment group in the data we use for our analysis is an accurate representation of the group of households that was eligible for FIA at baseline. The characteristics of the households in panel D are similar between the households in the 2006 national census and the 2010 CNQLS. The households in panel D are a the poorest households, so the bias towards sampling poor households in the 2006 census has less of a pronounced effect on biasing the household characteristics downward. We use the life satisfaction question from the 2010 CNQLS as the primary measure of well-being in this study. Life satisfaction is an evaluative subjective well-being measure where respondents are asked to consider their life as a whole, weighing what aspects of their life are important to their well-being, before choosing a response category that most closely reflects their evaluation. We chose an evaluative measure of well-being, as opposed to a momentary measure of well-being 30 , as the primary measure of well-being as evaluative measures are more strongly 30 Momentary SWB measures ask the respondent how they feel at a specified moment. For example, how satisfied with your life are you feeling right now? 99 correlated with life circumstances (Helliwell et al. 2012). The survey question (translated into English) that is asked of each respondent is: “How satisfied are you with the following aspects?” Life in general is listed as one of the aspects. The response options are coded: 1 –very dissatisfied, 2 – unsatisfied, 3 – satisfied, and 4 – very satisfied. Satisfaction with other life domains are asked the same way. In addition to life satisfaction and satisfaction with various life domains, the effect of FIA on other measures are considered: expenditures on categories of goods, self reported health, number of paid trips to the doctor, labor market outcomes, and characteristics of household dwelling. Details on these variables are presented in the results section or appendix D. 5.4. Methods 5.4.1. Generating and verifying household SISBEN scores The official government issued SISBEN scores that determined eligibility to the first round of FIA in large cities were generated in 2006 using data from the 2006 national census. The official government assigned SISBEN scores of households are not included in the CNQLS. To work around this issue, we use the responses to the 2010 CNQLS to compute a SISBEN score for each household. We were provided with the algorithm the government uses to generate the SISBEN score and the 2010 CNQLS includes all the variables required by the algorithm. Because we are not using the official SISBEN scores in our analysis, we need to deal with measurement error in our generated SISBEN score from the 2010 CNQLS. We are primarily concerned with three sources of measurement error. First, it is possible that respondents attempted to manipulate their official SISBEN score when filling out the 2006 national census by attempting to appear poorer, but did not attempt to appear poorer in the 2010 CNQLS. The 2010 CNQLS was 100 not used to determine qualification for social assistance programs, so the respondents did not have any incentive to untruthfully answer the questions. This type of measurement error would result in a household’s 2010 generated SISBEN score being higher than their official SISBEN score on average. Second, the official SISBEN scores were calculated based on the characteristics of households measured in 2006. The SISBEN scores generated with the 2010 CNQLS data were calculated based on characteristics of households measured four years after the 2006 national census. During this time, both economic growth and participation in FIA might have improved the living circumstances of households. This source of measurement error would also result in a household’s 2010 generated SISBEN score being higher than their official SISBEN score on average. Third, due to the four year period in between data collection, households may have experience idiosyncratic good and bad shocks moving their 2010 generated SISBEN scores higher and lower respectively than their official SISBEN scores. In the following sections, we examine how comparable the SISBEN scores generated with the 2010 CNQLS data are to the official SISBEN scores and also propose an empirical strategy that takes into account the measurement error. 5.4.2. Comparison with official baseline SISBEN data Figure 5-1 displays the densities of households over SISBEN scores for both the 2010 CNQLS and the official 2006 census data. The SISBEN scores are transformed to represent the distance from the FIA SISBEN eligibility cutoff (a score of 11). There are relatively more household observations below the FIA SISBEN cutoff in the 2006 national census data compared with the 2010 CNQLS data. This is due to sampling. The 2006 SISBEN census survey surveyed poorer households at a higher rate so poorer households could be assigned a SISBEN score 101 necessary to qualify for social assistance programs. The 2010 CNQLS, however, randomly sampled the entire population, and as a result did not sample poorer households as frequently. Despite the different sampling methods between the two datasets, we can compare some characteristics of the two densities over SISBEN scores. Both the 2006 official SISBEN and the 2010 generated SISBEN scores show a bi-modal distribution. In both graphs, two local peaks appear at either side of the FIA eligibility cutoff. There is some evidence that SISBEN scores shifted upward between 2006 and 2010 due to measurement error. In the 2006 national census data, the local peak below the eligibility cutoff is located between -6 and -5 SISBEN units below eligibility. In 2010 CNQLS data, the local peak below the eligibility cutoff is between -5 and -4 SISBEN units below eligibility. The local peaks above the eligibility cutoff show similar signs of measurement error. In the 2010 national census data, there is a local peak from 1 to 3 Panel A. Actual Scores from Baseline (2006) Census Data 102 Panel B. Generated Scores from 2010 Survey Data Notes: Households from all cities in the 2007 FIA expansion with children and a SISBEN score under 24 included in both panels in figure 1. Data for panel A is from the official 2006 Colombian census which includes official SISBEN scores. Sampling is not random in the Census data. Households can opt into being surveyed to obtain a SISBEN score. Panel B is from the 2010 CNQLS and the SISBEN scores are generated using variables from the CNQLS data. Sampling is random in the CNQLS. Figure 5-1. Density of Households over SISBEN Scores, 2010 CNQLS Compared to Official 2006 Colombian Census Data. SISBEN units above the eligibility cutoff. In the 2010 CNQLS data, the peak has shifted to 3 to 5 units above the eligibility cutoff. More evidence of SISBEN score comparability can be seen by looking at FIA program participation over SISBEN scores. The 2006 national census does not have FIA program participation data (FIA was implemented after the census data was collected), so to compare official FIA participation rates over SISBEN scores with the 2010 CNQLS participation rates, we use administrative data collected by the Colombian National Planning Department (CNPD) in 2007. The 2007 CNPD administrative data was gathered for a baseline assessment of FIA implementation. The primary concern of the CNPD was determining if there was a discontinuous 103 jump in program participation at the FIA eligibility cutoff. As a result, they primarily sampled households within two SISBEN units around the eligibility cutoff 31 . If our 2010 CNQLS SISBEN scores are accurate, we would expect to see similar participation rates over SISBEN scores between the 2007 CNPD administrative data and the 2010 CNQLS data. We can also look for evidence of SISBEN score measurement error by comparing FIA participation rates around the FIA eligibility cutoff. Panels A and B from Figure 5-2 plot the FIA participation rates over SISBEN scores for the 2007 CNPD administrative data and the 2010 CNQLS data, respectively. Figure 5-2 includes 95% confidence intervals around points that represent the average participation rate of households within a bin of with a 1 SISBEN unit width. Two lowess lines are also fit to the data above and below the eligibility cutoff. In panel A of Figure 5-2, there is a large discontinuity in program participation at the eligibility cutoff. Participation rates for eligible households are just above 0.6, and participation for ineligible households is much lower. There are a few bins that show high participation rates 31 Figure D-1 in the appendix D shows the density of households over SISBEN score in the 2007 CNPD administrative data. 104 Panel A. 2007 CNPD Administrative Data Panel B: 2010 CNQLS Data, No Doughnut Hole Figure 5-2: Baseline and 2010 FIA Participation Discontinuity Around Eligibility Cutoff 105 Figure 5-2 continued Panel C: 2010 CNQLS Data, Doughnut Hole Between -1 and 2 SISBEN points around cutoff Notes: Red lines represent 95% confidence intervals around average FIA subsidy receipt in bin with one unit width. If no confidence interval is reported, there is either not enough observations in bin or no variance. Data for panel A is from the baseline FIA survey gathered by the Colombian National Planning Department 3 months after FIA was implemented. The data includes official SISBEN scores from the 2006 Census. The sample was gathered giving priority to households around the SISBEN cutoff. Data from Panel B, C, and D is from the 2010 CNQLS. Data from both figures are limited to households from all cities included in the 2007 FIA expansion with children and a SISBEN score under 24. In all panels, lowess curves are fit to FIA program participation over SISBEN score. In Panel A and B, no doughnut hole is inserted in the data. In Panel C and D, a doughnut hole that omits data between the doughnut hole boundaries is inserted. The lowess cures are fitted to the data omitting the observations that lie in the doughnut hole. far above the cutoff, but as can be seen by the size of the confidence intervals, the sample size in these bins is very small. Overall, participation rates for ineligible households above the cutoff range between 0 and 0.25. Panel B presents the FIA participation data over generated SISBEN score from the 2010 CNQLS. Average participation rate for FIA for eligible and ineligible households are very similar to the 2007 administrative data: just over 0.6 for eligible households and between 0 and 0.25 for ineligible households. The similar participation rates of eligible and ineligible households supports the accuracy of the 2010 CNQLS generated SISBEN scores. Panel 106 B also shows evidence of measurement error around the eligibility cutoff. Specifically, there is no discontinuous jump in FIA participation at the cutoff. As explained in the previous section, we expect that the generated SISBEN from the 2010 CNQLS of a household to trend upwards due to economic growth and FIA participation, and also random change upwards and downwards for individual households due to idiosyncratic good and back shocks. Panel C presents the evidence of these types of measurement error around the eligibility cutoff. Panel C presents the same data as in Panel B, but the two lowess curves are fit to FIA participation data for households bellow -1 SISBEN unit of the cutoff and 2 SISBEN units above the cutoff. One unit below the cutoff and 2 units above the cutoff are denoted with vertical thick gray lines. Between -1 and 2 SISBEN units around the cutoff, we see program participation levels less than eligible households, but greater than ineligible households – a consequence that would be caused by the measurement error we expect. The lowess curves that omit households located between -1 and 2 SISBEN points around the cutoff in Panel C very closely resemble the lowess curves fit to the official 2007 administrative data in panel A. Overall, Figure 5-2 provides evidence that FIA participation over generated SISBEN scores in the 2010 CNQLS is very similar to the official 2007 CNPD data after we account for measurement error. In addition to verifying that the generated SISBEN scores from the 2010 CNQLS data accurately reflect FIA participation rates, we need to verify that the generated SISBEN scores are accurate predictions of actual SISBEN scores. Because we do not have the actual SISEBN scores in the 2010 CNQLS data, we verify the accuracy of the generated SISBEN scores by comparing the characteristics of households over SISBEN scores in the 2006 national census data with the 2010 CNQLS data. In other words, we want to verify that households sampled in 2010 CNQLS with similar characteristics to households sampled in the 2006 national census have similar 107 SISBEN scores (generated SISBEN vs. official SISBEN). To verify this, we compare the characteristics of households over SISBEN scores from the 2006 national census data with households in the 2010 CNQLS data. Figure D-2 in appendix D presents characteristics of households from the 2006 national census in the left column and households from the 2010 CNQLS in the right column. Each row represents a single characteristic. For almost all variables, the characteristics of households over SISBEN score are very similar in the 2006 national census and the 2010 CNQLS. Some household characteristics in the 2010 CNQLS, for example household income and if the household has a telephone, show evidence of a positive FIA treatment effect. That is, in the 2010 CNQLS data the households that are eligible for FIA show higher income and are more likely to have a phone than the households in the 2006 national census data. But, the household characteristics that we do not expect to change due to FIA program participation, for example if the household head has a secondary education or the average age of children in the household, look very similar in the 2006 national census and 2010 CNQLS data. 5.4.3. Identification strategy - implementing a “doughnut” hole regression discontinuity design The FIA SISBEN eligibility cutoff is the basis for the regression discontinuity design we use to identify the FIA program effects. This approach exploits the exogenous variation in program participation around the FIA SISBEN cutoff to effectively create a treatment and control group. The crux of the regression discontinuity method is that households just above the SISBEN cutoff are very comparable to those just below the cutoff, but one group can participate in FIA while the other cannot. The regression discontinuity approach requires a sharp discontinuity in participation around the cutoff, and due to the measurement error discussed earlier in our 2010 CNQLS 108 generated SISBEN scores, there is not a large enough discontinuity in participation at the cutoff in the raw 2010 CNQLS data. The solution to the diluted discontinuity in program participation in the 2010 CNQLS data is to create a “doughnut hole” between the on SISBEN unit below the FIA eligibility cutoff and two SISBEN units above the cutoff. Panels A and C in Figure 5-2 shows that after the doughnut hole is implemented, the program participation discontinuity in the 2010 CNQLS data closely resembles the baseline discontinuity. The households with scores inside the doughnut hole are dropped from the analysis. This is a common way to implement a regression discontinuity strategy when there is measurement error in the eligibility variable. The standard doughnut hole procedure is to create a doughnut hole that is symmetrical around the eligibility cutoff. Our approach is not symmetrical around the cutoff because the comparison between the 2007 baseline data and the 2010 CNQLS data shows evidence that the SISBEN scores in 2010 are biased upward. 5.4.4. Fuzzy RD estimation strategy Panel C in Figure 5-2 shows that not all households below the SISBEN cutoff received support from participating in FIA, and some households above the SISBEN cutoff did receive cash support FIA. Because of the households’ imperfect compliance, we use a fuzzy RD design – a two stage least squares approach. This allows us to estimate the treatment effect, not just the intent to treat effect. The equations that are the basis of the analysis are: 1. 𝐹𝐴 # =𝛾+𝛿𝑇 # +𝜌 𝑋 # −𝑐 +𝜆𝑇 # ∗(𝑋 # −𝑐)+𝑣 # 2. 𝑌 =𝛼+𝜏𝐹𝐴 # +𝛽 𝑋 # −𝑐 +𝜉𝑇 # ∗(𝑋 # −𝑐)+𝜀 # Where in the first equation: 𝐹𝐴 # is a dummy, where 𝐹𝐴 # =1 if the household received cash from FIA 109 𝑋 # is the SISBEN score of the household (SISBEN scores between -1 and 2 SISBEN units from the eligibility cutoff are omitted) c=11, the FIA SISBEN eligibility cutoff 𝑇 # is an eligibility dummy where 𝑇 # =1 if 𝑋 # <c In the second equation, 𝐹𝐴 # , is instrumented using 𝑇 # from equation 1. In addition to the basic specifications described by equations 1 and 2, our analysis includes two additional specifications that sequentially add more control variables. The second specification adds dummies for the household city. The third specification includes city dummies and variables for household characteristics. The household characteristics controlled for are: dummies for whether household has trash collection, has a sewage system, has a water source inside their dwelling, has piped water, and if the household head is male. These additional control variables were chosen because they are not likely to change as a result of receiving FIA support. All the additional control variables, with the exception of the gender of the household head which could change if a male head is replaced by a female head (or vice versa), are dwelling characteristics that are difficult to change because these variables are determined by infrastructure, not the household. We do not expect households included in our sample to change dwellings because SISBEN scores are tied to a household’s address. If the household relocates, it looses its FIA eligibility. We treat the relationship between dependent variables and SISBEN scores as linear and run multiple regressions per outcome, varying the distance around the doughnut hole boundary included in the regression. The results from the multiple regressions for each outcome allow us to determine how much the results depend on curvature of the outcome over SISBEN score, but not 110 sacrifice statistical power by only comparing households close to the eligibility cutoff. As of writing this analysis, this is the preferred regression discontinuity method. 5.4.5. Additional verification of research design The two conditions that must be satisfied to validate our regression discontinuity research design are, 1 – households did not manipulate their SISBEN score around the FIA eligibility cutoff, 2 – besides the FIA participation, all other variables are continuous around the FIA SISBEN eligibility cutoff at baseline. If the first condition is not satisfied, then we do not have exogenous variation around the eligibility cutoff. If condition 2 isn’t satisfied, we cannot determine if discontinuous jumps in outcome variables around the cutoff observed after FIA was implemented (treatment effects) are due to participation in FIA or due to baseline discontinuities in other household characteristics at the eligibility cutoff. Evidence that the first condition is satisfied is illustrated in Figure 5-1. If households were manipulating their SISBEN scores to be eligible for FIA, we would expect to see a heaping of households just below the FIA SISBEN cutoff. In both the 2006 official census data and in the 2010 CNQLS data we see fewer rather than more households just below the threshold compared to just above the threshold – the opposite of what we would expect if there was evidence of households manipulating their SISBEN score. To verify that FIA participation is the only discontinuous variable around the FIA cutoff at baseline, we examine the 2006 census data in Figure D-2 in appendix D. Specifically, we want to make sure that all households just below and just above the cutoff have similar characteristics. That is, there are no discontinuities in household characteristics around the cutoff besides 111 participation in FIA. Indeed, for all variables in Figure D-2, households just above and below the cutoff have similar characteristics in the 2006 data (baseline). 5.4.6. First stage of fuzzy regression discontinuity analysis The fussy regression discontinuity approach requires that eligibility for participation in FIA is a sufficiently strong instrument for participation in FIA. If the F-stat on the first regression of the two stage least squares is over 10, the instrumental variable is considered sufficiently strong. Table 5-2 presents the first stage regression results for all three specifications and also for regressions including data with different bandwidths around the doughnut hole. The bandwidths included in the table range from households within 10 SISBEN points from the to within 4 SISBEN points. The F-stats for each regression are close to 10 or greater than 10. The fact that the F-stats are not all above 10 indicate that the statistical analysis might suffer from problems associated with a weak instrument. The most important consequence is that the standard errors will be large when we estimate the effects of FIA participation. This is especially relevant to our study due to the limited sample size of the CNQLS. For this reason, the results need to be interpreted considering that our analysis might be statistically underpowered. The large effects of participation in FIA will be captured by our analysis, but smaller effects will likely not be statistically significant. 112 5.5. Results The estimated effects of FIA participation are presented in the following tables. Because of the large number of results reported here, most of the tables are displayed after the conclusion to reduce clutter in the text. Each table presents the estimated effect of FIA participation on a single outcome variable and includes 21 different estimates that vary by what control variables are included in the specification, and by observations included in the regressions. The panels present estimates for specifications that either include no controls (Panel A), city dummies (Panel B), and city dummies plus household characteristics (Panel C). The columns present estimates that vary by bandwidth around the doughnut hole boundary. The far left column includes all households within 10 SISBEN units of the doughnut hole boundary, and the far right column includes all 113 households within 4 SISBEN units of the boundary. Estimates with a wider bandwidth include more observations, but also include observations of households that are far away from the eligibility cutoff and therefore includes households that are less similar than household close to the cutoff. The smaller bandwidths compare households that are closer to the eligibility cutoff, and are therefore more similar, but have less observations and thus less statistical power. Figures (appendix D, Figure D-3) that plot variable values over SISBEN score are presented as compliments to the tables. The figures allow us to determine if program effects are being pushed by outlying observations or a non-linear relationship between the outcome variable and SISBEN score. Increased income is intended to be the primary and most direct benefit of participating in FIA. Our estimates of FIA participation on monthly household income are indeed robust and positive. Table 5-3A shows that the positive effect is robust to different specifications and also different bandwidths around the doughnut hole boundary. The significant estimates range from $399 to $725 (2010 USD). The increased income of households that participated in FIA also increased the head of household’s satisfaction with income (Table 5-3B). Did the households that participated in FIA spend their extra income, and if so, how did they spend it? We find evidence that participation in FIA increased household spending on a variety of items to make life more comfortable. We see a significant effect for expenditure on services (Table 5-3C), food (Table 5-3D), and durable goods (Table 5-3E). We also find a significant positive effect for expenditure on semi-durable goods (Table 5-3F), but this effect seems to be a result of an outlying observation (Figure D-3, panel F). We also see evidence that a household that participated in FIA is more likely to have a telephone (Table 5-3G), have cable TV (Table 5-3H), and has hot water (Table 5-3I). 114 Did the increased expenditure result in increased satisfaction with the corresponding domains of life? Household increased expenditure on food significantly increased the household head’s satisfaction with food (Table 5-3J). However, the increased spending on durable goods, cable TV, telephone, and hot water did not increase household’s head satisfaction with dwelling (Table 5-3K). FIA positively affected the self reported health of the household head (Table 5-3L), potentially due to increased paid trips to the doctor in the past month (Table 5-3M). Labor market outcomes also improved for the head of the household and also the partner of the head. FIA participation increased the likelihood that household head and partner of household head were formally employed (Table 5-3N for household head, Table 5-3O for partner). The improvement in labor market outcomes also has positive affects on the corresponding domain satisfaction variable: the household head’s work satisfaction was also positively affected by FIA participation (Table 5-3P). And finally, participation in FIA caused increased satisfaction with the household head’s ability to help others (Table 5-3Q). Overall, participation in FIA improved many objective and subjective outcomes. Families enjoyed extra income, which they spent on making their lives more comfortable. Beyond the extra spending allowed by increased income, FIA participation positively affected health, labor market outcomes, and satisfaction with the ability to help others. But, did these changes result in a higher level of life satisfaction for the household head? We argue that our analysis provides evidence that participating in FIA positively affected life satisfaction. Table 5-3R presents the FIA program effects on life satisfaction. For all specifications, and for all bandwidths besides 4 SISBEN units from doughnut hole boundary, the estimated effect of FIA participation is positive. But, a significant effect is only found for one 115 specification (panel C: city and household characteristic controls) and one bandwidth (6 SISBEN units). We interpret the results from the statistical analysis as evidence that participating in FIA positively affected life satisfaction for the following reasons. First, our analysis has limited statistical power. This is due to a relatively small sample size (531 households at the widest bandwidth) and a borderline weak first stage (F-stats just under 10). Second, despite the lack of statistical power, for all bandwidths around the doughnut hole greater than 5 SISBEN units, the estimates are fairly consistent: around or above 0.5 on a 1-4 scale. This is a large effect. Using the average level of life satisfaction for the ineligible group of households as a reference point, a 0.5 point increase in life satisfaction would be around a 20% increase. 116 Notes: Red lines represent 95% confidence intervals around average life satisfaction in bin with one unit width. If no confidence interval is reported, there is either not enough observations in bin or no variance. Data from the 2010 CNQLS. Doughnut hole that omits data between the doughnut hole boundaries is inserted. The lowess cures are fitted to the data omitting the observations that lie in the doughnut hole. Figure 5-3. Life Satisfaction over SISBEN score The smaller bandwidths, however, report much smaller and even negative effects. The results from the smaller bandwidths are due to patterns of life satisfaction around the eligibility cutoff that are inconsistent with the overall trend in life satisfaction over SISBEN score. Figure 5-3 displays life satisfaction over SISBEN score. Overall, the trend in life satisfaction over SISBEN score is positive on either side of the cutoff, with the eligible households reporting a discontinuously high level of life satisfaction. But, if we only focus on the households closer to the cutoff – the households included in the smaller bandwidth estimates in Table 5-3R – the trend over SISBEN scores appears to be negative or flat. We therefore focus on the wider bandwidths reported in Table 5-3R, which more actually reflect the trend. 117 Previous findings from the subjective well-being literature also support our interpretation of a positive effect on life satisfaction. Easterlin (2010) writes, “Virtually all life domain studies, however, agree that economic condition, family circumstances, health, and work are important domains determining happiness.” In our study, we observe statistically significant and positive effects on economic condition (income), health, and work domains. Furthermore, we do not observe any statistically significant declines in satisfaction for any domains of life. 5.6. Conclusion This paper provides evidence that FIA, a conditional cash transfer program in Colombia, positively affected many objective and subjective aspects of well-being. Households that participated in FIA enjoyed higher incomes, spent more on items to make their lives more comfortable, enjoyed improved self reported health, and had better labor market outcomes. As a result, participants enjoyed increases in satisfaction with a handful of life domains: income, food, work, and the ability to help others. We further argue participating in FIA positively affected life satisfaction, the subjective well-being measure used in this analysis. The results provide additional support for implementing conditional cash transfers in developing countries by showing these programs have palpable positive effects on the way people view their lives. Furthermore, the findings from this paper support the overall argument that social assistance programs can be used by governments to promote increases in well-being 118 Tables reference in results section 119 120 121 122 123 124 125 126 127 CHAPTER 6. Summary and Conclusions The findings from the studies included in this dissertation contribute new knowledge on how policy can protect and promote well-being. All the studies use subjective well-being as the primary measure of well-being. Subjective well-being studies are becoming increasingly important as subjective well-being is starting to be used to guide policy. Chapter 2 presents evidence that life cycle subjective well-being, on average for 17 European countries, follows a wave like M-shape over the life cycle. The M-shape arises because patterns in the majority of countries share the following characteristics: a local maximum in subjective well-being around age 30, declining subjective well-being until around age 50 followed by rising subjective well-being, and then declining subjective well-being after age 75. Other features that are shared across the majority of countries are male subjective well-being improving relative to female life satisfaction as people age, and more educated people reporting higher levels of subjective well-being throughout the entire life cycle. Although these characteristics are shared by the majority of countries, taking the entire life cycle into consideration, there is no uniform life satisfaction pattern shared by all countries. The finding that in most countries people experience subjective well-being troughs around age 50, and then have declining subjective well-being after age 75 motivates policy intervention at these ages. The well-being of females at older ages and lesser educated people at all ages should also be considered when designing policy. The finding that the subjective well-being patterns vary between countries motivates further research as to why these variations exist. Finding from this line of future research can help countries design effective age targeted policy. 128 The analysis in Chapter 3 demonstrates that improving labor market conditions can account for 29% of the increase in subjective well-being between 2002 and 2012 in urban China. This is primarily driven by a large decrease in unemployment during this time. The improving labor market is especially important for the segments of the population most vulnerable to the negative effects of the transition – people with less than a college education. We also find that any positive effect increasing income had on subjective well-being was nullified by income comparison and habituation. The net result is that changes in income has no significant relationship with the increase in subjective well-being from 2002 to 2012. The results present a potential well-being problem as China continues to liberalize its labor market by allowing freer rural to urban migration. The people who are most vulnerable during labor market liberalization, less educated people, will flood the urban areas in China. If they struggle to find jobs, the results from this analysis suggest that their well-being will suffer greatly. If the government wants to protect well-being, policy that promotes employment of the less educated masses is of paramount importance. Chapter 4 focuses on the Great Recession in Europe. The analysis finds that average subjective well-being decreased, primarily among people with less than a college education and among all age groups less than retirement age. The primary aim of the analysis is to examine whether different types of labor market policies mitigated or exacerbated the negative impact of the Great Recession on subjective well-being for these vulnerable groups. The results demonstrate that for all vulnerable groups with the exception of youth, labor market policies had a significant effect on subjective well-being during the Great Recession, but the effect was either mitigating or exacerbating depending on the type of labor market policy. Unemployment support that provided income replacement or programs to help unemployed workers find jobs mitigated the negative effect of the Great Recession on subjective well-being. Conversely, stricter employment protection 129 legislation exacerbated the negative effect of the Great Recession. Suggestive evidence is presented that the exacerbating effect is explained by strict employment protection legislation imposing rigidities on the labor market, making people feel less optimistic about their future job prospects. The results support a Denmark approach to labor market policy; allow firms to freely adjust their workforce, but provide generous support to those who lose their jobs. The study in Chapter 5 assesses the subjective well-being effects of Familias en Acción, a conditional cash transfer program in Colombia. The program positively affected many aspects of life – improvements in self reported health, increased income, increased expenditure on food, goods and services, and increased frequency of formal employment. These improvements lead to higher satisfaction with food, income, work, and the ability to help others. Satisfaction with life was also positively affected by program participation, and although the effect is not statistically significant, the result is consistent with the improvements in other aspects of life due to participation in Familias en Acción. The results provide additional support for implementing conditional cash transfers in developing countries by showing these programs have palpable positive effects on the way people view their lives. 130 REFERENCES Appleton, SM, and Song, L (2008) Life satisfaction in urban China: Components and determinants. World Dev 36: 2325–2340. Arampatzi, E, Burger, MJ, and Veenhoven, R (2015) Financial Distress and Happiness of Employees in Times of Economic Crisis. Appl Econ Lett 22(3):173-179 Baird, BM, Lucas, RE, and Donnellan, BM (2010). Life Satisfaction Across the Lifespan: Findings from Two Nationally Representative Panel Studies. Social Indicators Research, 99(2): 183-203. Beaton, T, and Frijters, P (2012). The mystery of the U-shaped relationship between happiness and age. Journal of Economic Behavior & Organization, 82: 525-542. Blanchflower, DG, and Oswald, AJ (2004). Well-being over time in Britain and the USA. Journal of Public Economics, 88: 1359-1386. Blanchflower, DG, and Oswald, AJ (2008). Is well-being U-shaped over the life cycle? Social Science & Medicine, 66(8): 1733–1749. Blanchflower, DG, and Oswald, AJ (2009) The U-shape without controls: a response to Glenn. Social Science & Medicine, 69: 486–488. Brockmann, H, Delhey, J, Welzel, C, and Yuan, H (2009) The China puzzle: Falling happiness in a rising economy. J Happiness Stud, 10:387–405. Boarini, R, Comola, M, Keulenaer, FD, Manchin R, and Smith, C (2013) Can Governments Boost People’s Sense of Well-being? The Impact of Selected Labour Market and Health Policies on Life Satisfaction. Soc Indic Res, 114(1):105-120 Burkholder, R (2005) Chinese far wealthier than a decade ago—but are they happier? The Gallup Organization. Available at http://www.gallup.com/poll/14548/chinese-far- wealthier-than-decade- ago-they-happier.aspx. Accessed January 27, 2006. Cai, HB, Chen, YY, and Zhou, LA (2010) Income and consumption inequality in urban China: 1992–2003. Econ Dev Cult Change, 58:385–413. Cai, F, Wang, MY, and Wang, DW, eds (2010) The China Population and Labor Yearbook: The Sustainability of Economic Growth from the Perspective of Human Resources (Brill, Leiden), Vol 2. Carr, E, and Chung, H (2014) Employment Insecurity and Life Satisfaction: The Moderating Influence of Labour Market Policies across Europe. J Eur Soc Policy, 24(4):383-399. Cattaneo, MD, Galiani, S, Gertler, PJ, Martinez, S, and Titiunik, R (2009). Housing, health, and happiness. American Economic Journal: Economic Policy, 1(1): 75-105. 131 Charles, KK (2004). Is Retirement Depressing? Labor Force Inactivity and Psychological Well- Being in Later Life. Research in Labor Economics, 23: 269-299. Chen Z, Davey, G (2008) Happiness and subjective well-being in mainland China. J Happiness Stud, 9:589–600. Chen JG, Liu, SC, and Wang TS, eds (2011) The China Economy Yearbook: Analysis and Forecast of China’s Economic Situation (Brill, Leiden), Vol 5. Cheng, TC, Powdthavee, N, and Oswald, AJ (2014) Longitudinal Evidence for a Midlife Nadir in Human Well-being: Results from Four Data Sets. IZA Discussion Paper, No. 7942, 1-26. Clark, A (2007) Born to be mild? Cohort effects don’t (fully) explain why well-being is U shaped in age. IZA Discussion Paper, No. 3170, 1–29. Clark, A, Frijters, P, and Shields, MA (2008) Relative Income, Happiness, and Utility: An Explanation for the Easterlin Paradox and Other Puzzles. Journal of Economic Literature, 46(1): 95-144. Clark, A, Georgellis, Y, and Sanfey, P (2001) Scarring: The psychological impact of past un- employment. Economica, 68:221–241. Clark, A, and Postel-Vinay, F (2009) Job Security and Job Protection. Oxford Econ Pap, 61(2):207-239. Crabtree, S, and Wu, T (2011) China’s puzzling flat line. Gallup Management Journal Available at: http://gmj.gallup.com/content/148853/china-puzzling-flat-line.aspx#1 Accessed February 1, 2012. Deaton, A (2010) Income, Aging, Health and Well-Being around the World: Evidence from the Gallup World Poll. In DA Wise (Ed.), Research Findings in the Economics of Aging (pp. 235- 263). Chicago: The University of Chicago Press. Deaton, A (2012) The Financial Crisis and the Well-being of Americans. Oxford Econ Pap, 64(1): 1-26. De Keulenaer, F, De Neve, JE, Kavetsos, G, Norton, MI, Van Landeghem, B, and Ward, GW (2014) Individual Experience of Positive and Negative Growth is Asymmetric: Evidence from Subjective Well-being Data. Harvard Bus Sch Working Pap http://www.hbs.edu/faculty/Publication%20Files/15-021_500f17e8-a78c-4301-a2be- 755a1fdd6679.pdf Devoto, F, Duflo, E, Dupas, P, Parienté, W and Pons, V (2012) Happiness on Tap: Piped Water Adoption in Urban Morocco. American Economic Journal: Economic Policy, 4(4): 68-99. 132 Ditella, R, Haisken De-New, J, and MacCulloch, R (2010) Happiness adaptation to income and to status in an individual panel. Journal of Economic Behavior & Organization, 76: 834–852. Di Tella, R, MacCulloch, RJ, and Oswald, AJ (2001) Preferences over inflation and unemployment: evidence from surveys of happiness. American Economic Review, 91(1): 335–341. Di Tella, R, MacCulloch ,RJ, and Oswald, AJ (2003) The Macroeconomics of Happiness. Rev Econ Stat, 85(4):809-827 Easterlin, RA (2003) Explaining Happiness. Proceedings from the National Academy of Sciences, 100(19): 11176-11183. Easterlin, RA (2006) Life Cycle Happiness and Its Sources: Intersections of Psychology, Economics and Demography. Journal of Economic Psychology, 27(4): 463–482. Easterlin, RA (2010) Lost in Transition: Life Satisfaction on the Road to Capitalism. In RA Easterlin, ed. Happiness, Growth, and the Life Cycle (pp. 82-110). Oxford, UK: Oxford University Press. Easterlin, RA (2010). The Cross-Over in the Life Cycle Happiness of American Women and Men. In RA Easterlin, ed. Happiness, Growth, and the Life Cycle (pp. 175-188). Oxford, UK: Oxford University Press. Easterlin, RA, Angelescu, LM, Switek, M, Sawangfa, O, and Zweig, JS (2010) The Happiness- income paradox revisited. Proceedings of the National Academy of Sciences of the United States of America, 107 (52): 22463-8. Easterlin RA, Morgan, R, Switek, M, and Wang, F (2012) China’s Life Satisfaction, 1990-2010, Proceedings of the National Academy of Sciences of the United States of America, 109 (25): 9775- 80. Easterlin, RA, and Switek, M (2015 forthcoming) Explaining Well-Being over the Life Cycle: A Look at Life Transitions during Young Adulthood European Commission: Eurobarometer 68.1. TNS Opinion & Social, Brussels [Producer]. GESIS Data Archive, Cologne. ZA4565 data file version 4.0.1, doi:10.4232/1.10988, 2007. Available at: http://zacat.gesis.org. Access data: April 2015 European Commission: Eurobarometer 71.3. TNS Opinion & Social, Brussels [Producer]. GESIS Data Archive, Cologne. ZA4973 data file version 3.0.0, doi:10.4232/1.11135, 2009. Available at: http://zacat.gesis.org. Access data: April 2015 Finkelstein, A and the Oregon Health Study Group (2012) The Oregon Health Insurance Experiment: Evidence from the First Year, Quarterly Journal of Economics, 127(3): 1057-1106. Fonseca, R, Kapteyn, A, Lee, JK, and Zamarro, G (2014) Does Retirement Make You Happy? A Simultaneous Equations Approach. Michigan Retirement Research Center Research Paper No. 133 2014-310. Available at SSRN: http://ssrn.com/abstract=2566755 or http://dx.doi.org/10.2139/ssrn.2566755 Fiszbein, A, and Schady, NR (2009) Policy research reports: Conditional cash transfers: Reducing present and future poverty: The World Bank, available at: http://econ.worldbank.org/external/default/main?theSitePK=469382&contentMDK=23027684& menuPK=574960&pagePK=64165401&piPK=64165026. Frijters, P (2004) How Important is Methodology for the Estimates of the Determinants of Happiness? Econ J, 114(497):641-659. Gardner, J, and Oswald, AJ (2007) Money and mental wellbeing: A longitudinal study of medium- sized lottery wins. Journal of health economics, 26(1): 49-60. Gustafsson, B, and Ding, S (2011) Rising Inequality in China: Challenge to a Harmonious Society, eds Li S, Sato H, Sicular T (Cambridge Univ Press, New York). Available at http://economics.uwo.ca/centres/cibc/wp2011/Gustafsson_Ding17.pdf. Accessed January 5, 2012. Haushofer, J, and Shapiro, J (2013 working paper) Household Response to Income Changes: Evidence from an Unconditional Cash Transfer Program in Kenya Helliwell, JF, and Huang, H (2014) New Measures of the Costs of Unemployment: Evidence from the Subjective Well-being of 3.3 Million Americans. Econ Inq, 52(4):1485-1502 Helliwell, JF, Layard, R, and Sachs, J (2012) World Happiness Report 2012. Sustainable Development Solutions (Network, New York). Helliwell, JF, Layard, R, and Sachs, J (2015) World Happiness Report 2015. Sustainable Development Solutions (Network, New York). Kahneman, D, and Krueger, AB (2006) Developments in the measurement of subjective well- being. J Econ Perspect, 20:3–24. Kassenboehmer, SC, and Haisken-DeNew, JP (2009) You’re fired! The causal negative effect of entry unemployment on life satisfaction. Econ J, 119:448–462. Kim, JE, and Moen, P (2002) Retirement Transitions, Gender, and Psychological Well-Being. A Life-Course, Ecological Model. Journal of Gerontology, 57B(3), 212-222. Kim, MH, and Do, YK (2013) Effect of Husbands' Employment Status on their Wives' Subjective Well-being in Korea. J Marriage Fam, 75(2):288-299. Knight, JB, and Song, L (2005) Towards a labour market in China, Oxford University Press. (Oxford ;New York). 134 Knight, J, and Xue, JJ (2006) How high is urban unemployment in China? J Chin Econ Bus Stud, 4:91–107. Knight, J, Gunatilaka, R (2011) Does economic growth raise happiness in China? Oxf Dev Stud, 39:1–24. Kuhn, P, Kooreman, P, Soetevent, A, and Kapteyn, A (2011) The effects of lottery prizes on winners and their neighbors: Evidence from the dutch postcode lottery. The American Economic Review, 101(5): 2226-2247 Latten, JJ (1989). Life-course and Satisfaction, Equal for Everyone? Social Indicators Research, 21(6): 599-610. Lucas, RE (1987) Models of business cycles. Blackwell (New York). Lucas, RE, Clark, AE, Georgellis, Y, and Diener, E (2004) Unemployment Alters the Set Point for Life Satisfaction, Psychological Science, 15(1): 8-13. McAdams, KK, Lucas, RE, and Donnellan, MB (2012) The Role of Domain Satisfaction in Explaining the Paradoxical Association Between Life Satisfaction and Age. Social Indicators Research, 109(2): 295–303. Méndez et al. (2011) Evaluación del Programa Familias en Acción en Grandes Centros Urbanos. Available at: http://www.larepublica.co/sites/default/files/larepublica/6335_Evaluacion_FeA- U_Informe_final_Diciembre_2011.pdf Mroczek, DK, and Spiro, A (2005). Change in Life Satisfaction During Adulthood: Findings from the Veterans Affairs Normative Aging Study. Journal of Personality and Social Psychology, 88(1): 189–202. Ochsen, C, Welsch, H (2012) Who benefits from labor market institutions? Evidence from surveys of life satisfaction. J Econ Psychol, 33(1):112-124 OECD (2010) OECD Economic Surveys: China (Organization for Economic Development and Cooperation, Beijing),Vol. 2010/6. OECD (2013) OECD economic surveys. China 2013. Paris: OECD. OECD (2014a) OECD StatExtracts. Quarterly Data. http://stats.oecd.org. Access date: May 2014 OECD (2014b) Directorate for Employment, Labour and Social Affairs, Benefits and Wages: Statistics. http://www.oecd.org/els/benefitsandwagesstatistics.htm. Access date: January 2013 OECD (2014c) Dataset: Public expenditure and participant stocks on LMP. Public expenditure as a percentage of GDP (Annual). http://stats.oecd.org. Access date: April 2013 135 OECD (2014d) OECD Indicators of Employment Protection. http://www.oecd.org/employment/emp/oecdindicatorsofemploymentprotection.htm. Access date: May 2014 OECD (2015) OECD Economic Surveys: China 2015. (Paris: OECD). Oshio, T, Nozaki, K, and Kobayashi, M (2011) Relative income and happiness in Asia: Evidence from nationwide surveys in China, Japan, and Korea. Soc Indic Res, 104: 351–367. Pacek, AC, and Radcliff, B (2008a) Welfare Policy and Subjective Well-Being Across Nations: An Individual-Level Assessment. Social Indicators Research, 89:179–191. Pacek, AC, and Radcliff, B, (2008b) Assessing the Welfare State: The Politics of Happiness. Perspectives on Politics, 6(2): 267-277. Romer, D (1996) Advanced Macroeconomics. (McGraw Hill, New York). Rothstein, B (2010) Happiness and the Welfare State. Social Research, 77(2): 441 – 469. Smyth, R, Nielsen, I, and Zhai, QG (2010) Personal well-being in urban China. Soc Indic Res, 95: 231–251. Stiglitz, JE, Sen, A, Fitoussi, JP (2009) Report of the Commission on the Measurement of Economic Performance and Social Progress. Available at: www.stiglitz-sen-fitoussi.fr. Accessed October 20, 2011. Ulloa, BF, Møller, V, and Sousa-Poza, A (2013) How Does Subjective Well-Being Evolve with Age? A Literature Review. Population Ageing, 6(3): 227-246. Van Landeghem, B (2012) A test for convexity of human well-being over the lifecycle: longitudinal evidence from a 20-year panel. Journal of Economic Behavior and Organization, 81(2): 571–585. Vendrik, MCM (2013) Adaptation, anticipation and social interaction in happiness: An integrated error-correction approach. Journal of Public Economics, 105: 131–149. Wang, S, and Zhou, WN (2016) The Unintended Long-term Consequences of Mao's Mass Send- down Movement: Marriage, Social Network, and Happiness, HiCN working paper no. 213. Available at: http://www.hicn.org/wordpress/wp-content/uploads/2012/06/HiCN-WP-2131.pdf. Winkelmann, L, and Winkelmann, R (1998) Why are the unemployed so unhappy? Evidence from panel data. Economica, 65:1–15. Wolfers, J (2003) Is Business Cycle Volatility Costly? Evidence from Surveys of Subjective Well- being. Int Finance, 6(1):1-26. 136 World Values Survey (2014) WVS Database, available at: http://www.worldvaluessurvey.org/. Accessed February 2015. Wulfgramm, M (2014) Life Satisfaction Effects of Unemployment in Europe: The Moderating Influence of Labour Market Policy. J Eur Soc Policy, 24(3): 258-272. Web References (non-academic articles) The Economist, December 16 2010. The U-Bend of Life. URL: http://www.economist.com/node/17722567, last accessed: 11/20/2015. The Guardian, January 29 2008. Happiness is being young or old, but middle age is misery. URL: http://www.theguardian.com/society/2008/jan/29/health.medicalresearch, last accessed: 11/20/2015. Scientific American, January 30 2008. Midlife Misery: Is There Happiness After the 40s? URL: http://www.scientificamerican.com/article/midlife-misery-is-there-happiness-after-the-40s/, last accessed: 11/20/2015. 137 APPENDIX A: Supplementary material for Chapter 2 138 APPENDIX B: Supplementary material for Chapter 3 Equivalence of Oaxaca decomposition and method from Chapter 3 The specification of the pooled cross-sectional regression used in the paper is 𝐿𝑆 #,F =𝒙 #,F 𝜷=𝒛 #,F 𝜸+ 𝑑 F 𝛿 F NOCN FLNOOC , (B.1) where 𝑑 F is a dummy for year t. To implement Oaxaca decomposition, we run regressions for the years of 2002 and 2012, respectively, based on the same specification as in equation (B.1). Then, we have 𝐿𝑆 #,NOOC =𝒛 #,NOON 𝜸 NOON +𝛼 NOON , 𝐿𝑆 #,NOCN =𝒛 #,NOCN 𝜸 NOCN +𝛼 NOCN , (B.2) where 𝛼 is the estimate of constant. Then, the Oaxaca decomposition can be expressed as Δ𝐿𝑆 =𝐿𝑆 NOCN −𝐿𝑆 NOON =𝒛 NOCN 𝜸 NOCN −𝒛 NOON 𝜸 NOON + 𝛼 NOCN −𝛼 NOON = 𝒛 NOCN −𝒛 NOON 𝜸 NOCN +𝒛 NOON 𝜸 NOCN −𝜸 NOON + 𝛼 NOCN −𝛼 NOON = 𝒛 NOCN −𝒛 NOON 𝜸 NOON +𝒛 NOCN 𝜸 NOCN −𝜸 NOON + 𝛼 NOCN −𝛼 NOON = 𝒛 NOCN −𝒛 NOON 𝜸+𝒛 NOON 𝜸−𝜸 NOON +𝒛 NOCN 𝜸 NOCN −𝜸 + 𝛼 NOCN −𝛼 NOON (B.3) where the second and third lines of equation (B.3) represent the Oaxaca decomposition with different base years, and the last line is an improved version of Oaxaca decomposition which avoids the issue of double base years. We assume 𝜸 is obtained from equation (B.1). According to the approach in the paper, Δ𝐿𝑆 =𝐿𝑆 NOCN −𝐿𝑆 NOON = 𝒛 NOCN −𝒛 NOON 𝜸+ 𝛿 NOCN −𝛿 NOON , (B.4) 139 By comparing equations (B.3) and (B.4), we can find that, the contribution of survey year dummies in (B.4), 𝛿 NOCN −𝛿 NOON , is equivalent to the contribution of the regression coefficients in (B.3), and the rest parts, the contribution of the change in the values of variables are identical between (B.3) and (B.4). The shortcoming of the approach in the paper compared to Oaxaca decomposition is that the former cannot distinguish the contribution of the change in each regression coefficient. However, the former could be acceptable if the total contribution of regression coefficients is relatively small, which is the case in the paper, or the study cares more about the contribution of the change in variable values. Calculation of p-values reported in Tables 3-2 and 3-4 The p-value of the percent contribution of variable 𝑥 J is derived from the following test. H0: percent contribution of 𝑥 J = 0. H1: percent contribution of 𝑥 J is not 0. As the contribution of 𝑥 J is calculated from 𝑐 J = Q R,STUS VQ R,STTS W R XYZ ×100%, it is equivalent to test if 𝛽 J is 0, conditional on the changes in life satisfaction (Δ𝐿𝑆) and in explanatory variables (𝑥 J,NOCN −𝑥 J,NOON ), or treating the two as constant. If the percent contribution is calculated from the summation of many 𝑐 J (e.g., the percent contribution of age is equal to the summation of contributions of each age dummy.), i.e., 𝑐 J = Q R,STUS VQ R,STTS W R XYZ ×100%, then the p-value is from the test whether 𝑥 J,NOCN −𝑥 J,NOON 𝛽 J is 0, assuming Δ𝐿𝑆 and each 𝑥 J,NOCN −𝑥 J,NOON are constant (or conditional on them). 140 APPENDIX C: Supplementary material for Chapter 4 141 APPENDIX D. Supplementary material for Chapter 5 Figure D-1: Density of Households over SISBEN Scores, 2007 CNPD Administrative Data 142 Figure D-2: Baseline and 2010 Continuity of Variables around Eligibility Cutoff Panel A: House Has Electricity 2006 Census Data 2010 CNQLS Data Panel B: House Has Sewage System 2006 Census Data 2010 CNQLS Data Panel C: House Has Gas Connection 2006 Census Data 2010 CNQLS Data 143 Panel D: House Has Telephone 2006 Census Data 2010 CNQLS Data Panel E: House Has Trash Collection 2006 Census Data 2010 CNQLS Data Panel F: House Has Piped Water 2006 Census Data 2010 CNQLS Data 144 Panel G: House Has Toilet 2006 Census Data 2010 CNQLS Data Panel H: House Has Shower 2006 Census Data 2010 CNQLS Data Panel I: House Has Refrigerator 2006 Census Data 2010 CNQLS Data 145 Panel J: House Has Heater 2006 Census Data 2010 CNQLS Data Panel K: House has Washing Machine 2006 Census Data 2010 CNQLS Data Panel L: House has Air Conditioning 2006 Census Data 2010 CNQLS Data 146 Panel M: House has Cable TV 2006 Census Data 2010 CNQLS Data Panel N: House has Hot Water 2006 Census Data 2010 CNQLS Data Panel O: House has Indoor Water Source 2006 Census Data 2010 CNQLS Data 147 Panel P: House has Toilet Attached to Sewage System 2006 Census Data 2010 CNQLS Data Panel Q: Number of Toilets in House 2006 Census Data 2010 CNQLS Data Panel R: House has Shared Phone, Conditional on not having Private Phone 2006 Census Data 2010 CNQLS Data 148 Panel S: Stove is Connected to Public Gas Network 2006 Census Data 2010 CNQLS Data Panel T: House has Finished Floor Materials 2006 Census Data 2010 CNQLS Data Panel U: Average Age of Children in Household 2006 Census Data 2010 CNQLS Data 149 Panel V: Total Number of People in Household 2006 Census Data 2010 CNQLS Data Panel W: Head of Household Has Secondary Education 2006 Census Data 2010 CNQLS Data Panel X: Monthly Household Income (2010 USD) 2006 Census Data 2010 CNQLS Data 150 Panel Y: Share of employed people in household 2006 Census Data 2010 CNQLS Data Panel Z: Partner of Household Head has Secondary Education 2006 Census Data 2010 CNQLS Data Panel AA: Number of Dormitories per Person 2006 Census Data 2010 CNQLS Data 151 Panel AB: Share of Employed People Formally Working 2006 Census Data 2010 CNQLS Data Notes: Households from all cities in the 2007 FIA expansion with children and a SISBEN score under 24 included in all panels in figure 3. Data for the left column is from the official 2006 Colombian census. Data from the right column is from the 2010 CNQLS. Red lines represent 95% confidence intervals around average variable value in bin with one unit width. If no confidence interval is reported, there is either not enough observations in bin or no variance. Due to computational limitations and a dataset with over 8 million observations, lowess curves are fit to the average bin values in the left column (2006 Census data). The lowess curves are fit to the raw data in the right column (2010 CNQLS data). 152 Figure D-3: Outcome Variables over SISBEN, 2010 CNQLS, Doughnut Hole Between -1 and 2 Panel A: Monthly Household Income (2010 USD) Panel B: Household Head Satisfaction with Income (1-4 scale) Panel C: Household Expenditure on Services in Past Month (2010 USD) 153 Panel D: Household Expenditure on Food in Past 7 Days (2010 USD) Panel E: Household Expenditure on Durable Goods in Past Year (2010 USD) Panel F: Household Expenditure on Semi-durable Goods in Past 3 Months (2010 USD) 154 Panel G: Household has a Telephone Panel H: Household has Cable TV Panel I: Household has Hot Water 155 Panel J: Household Head Satisfaction with Food (1-4 scale) Panel K: Household Head Satisfaction with Dwelling (1-4 scale) Panel L: Household Head Self-reported Health (1-4 scale) 156 Panel M: Household Head Paid Trip to Doctor in Past Month Panel N: Household Head is Formally Employed Panel O: Partner of Household Head is Formally Employed 157 Panel P: Household Head Satisfaction with Work (1-4 scale) Panel Q: Household Head Satisfaction with Ability to Help Others (1-4 scale) Panel R: Household Head Satisfaction with Life (1-4 scale)

Abstract (if available)

Abstract The studies in this dissertation provide new knowledge on how life circumstance and policy relate to subjective well-being. The findings from the studies either directly evaluate policy, or provide direction for future policy. Chapter 2 estimates the average pattern of subjective well-being across the life cycle for 17 European countries using Eurobarometer data. The average pattern of experienced life cycle satisfaction for the 17 countries resembles a wave like M-shape. Other features that are shared among the majority of countries are male life satisfaction improving relative to female life satisfaction as people age, and more educated people reporting higher levels of life satisfaction throughout the entire life cycle. Although these characteristics are shared by the majority of countries, there is no uniform life satisfaction pattern shared by all countries. Chapter 3 uses annual cross-sectional data to identify what changes in peoples’ lives are driving the upward trend in subjective well-being observed in urban China from 2002 to 2012. The analysis finds that improvements in the labor market, mostly driven by a large drop in unemployment, account for 29% of the increase in subjective well-being. Increases in income, after accounting for income comparison and habituation, are not significantly related to the change in subjective well-being. Chapter 4 uses policy variation across 23 European countries to evaluate the effectiveness of four labor market policies to mitigate the negative impact of the Great Recession on subjective well-being. Policies that provide support to people who become unemployed significantly mitigated the negative effect of the Great Recession, where policies that limited the ability of firms to freely adjust their labor force significantly exacerbated the effect of the recession. Chapter 5 uses a regression discontinuity approach to evaluate the impact of a Colombian conditional cash transfer program. The program positively affected many aspects of life—improvements in self reported health, increased income, increased expenditure on food, goods and services, and increased frequency of formal employment. These improvements lead to higher satisfaction with food, income, work, and the ability to help others. Subjective well-being was also positively affected by program participation, and although the effect is not robustly statistically significant, the result is consistent with the improvements in other aspects of life due to participation in the conditional cash transfer program.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

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

PDF

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

PDF

Explaining well-being: essays on the socio-economic factors accompanying life satisfaction

PDF

Essays on the economics of subjective well-being in transition countries

PDF

Beyond revealed preferences: how gender, relative socioeconomic status, and social norms drive happiness and behavior

PDF

Essays on health and well-being

PDF

Adaptation, assets, and aspiration. Three essays on the economics of subjective well-being

PDF

Essays on behavioral decision making and perceptions: the role of information, aspirations and reference groups on persuasion, risk seeking and life satisfaction

PDF

Essay on monetary policy, macroprudential policy, and financial integration

PDF

Essays on development and health economics: social media and education policy

PDF

The role of public policy in the decisions of parents and caregivers: an examination of work, fertility, and informal caregiving

PDF

The determinants and measurement of human capital

PDF

Essays on macroeconomics and income distribution

PDF

Essays on health insurance programs and policies

PDF

Essays on family and labor economics

PDF

Essays on macroeconomics of health and labor

PDF

Three essays on health & aging

PDF

Three essays on the evaluation of long-term care insurance policies

PDF

Essays on health and aging with focus on the spillover of human capital

PDF

Essays in labor economics: demographic determinants of labor supply

Asset Metadata

Creator Morgan, Robson (author)

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

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Publication Date 06/17/2016

Defense Date 05/04/2016

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

Tag China,Happiness,life cycle,life satisfaction,OAI-PMH Harvest,policy,subjective well-being

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Easterlin, Richard A. (committee chair), Kapteyn, Arie (committee member), Myers, Dowell (committee member)

Creator Email rhmorgan@usc.edu,robsonhmorgan@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-253158

Unique identifier UC11281107

Identifier etd-MorganRobs-4444.pdf (filename),usctheses-c40-253158 (legacy record id)

Legacy Identifier etd-MorganRobs-4444-0.pdf

Dmrecord 253158

Document Type Dissertation

Format application/pdf (imt)

Rights Morgan, Robson

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