A Pecuniary Explanation for the Heterogeneous Effects of Unemployment on. Happiness. Jianbo Luo *

A Pecuniary Explanation for the Heterogeneous Effects of Unemployment on Happiness Jianbo Luo * ABSTRACT. Why unemployment has heterogeneous effects on subjective well-being remains unexplained. Using German Socio-Economic Panel (GSOEP) data, I find that both subjective and objective income measures mediate the negative effects of unemployment and explain its heterogeneous effects on happiness. This finding suggests that the root cause of the negative effects of unemployment is pecuniary. Policy implications for taxation and unemployment insurance are discussed. Key Words: unemployment; subjective well-being; heterogeneity; minimum required income; subjective financial evaluation JEL code: D6; E2; I3; J6 * Jianbo Luo: Department of Economics, State University of New York at Buffalo, Buffalo, NY, USA <jianbolu@buffalo.edu> 1

I. Introduction Why, after income is controlled for, unemployment substantially reduces subjective well-being (SWB 1 ) remains a hot topic in the happiness literature. Various non-pecuniary factors are proposed to explain this phenomenon, such as the social work norm, psychological distress, etc (see Winkelmann [2014] and the review section in Luo [2016b] for more details). An exception is Luo (2016b). He uses combined cross-sectional datasets World Values Survey (WVS) and European Values Study (EVS) to empirically demonstrate that the root cause is pecuniary. First, he suggests that controlling only actual income cannot capture all the pecuniary effects. Two thirds of the negative effects of unemployment disappear after including actual income and subjective financial evaluation (household financial satisfaction, in his paper), an additional control for pecuniary costs. Second, non-pecuniary theories have less explanatory power than, but are compatible with, pecuniary factors. All of the above explanations, however, cannot explain the heterogeneous effects of unemployment on SWB. Winkelmann (2009), and Gielen and van Ours (2014) find that about half of those unemployed do not experience SWB loss. Binder and Coad (2015a, 2015b) try to explain the heterogeneity. They find that SWB itself is a buffer to unemployment. In other words, no factors, except SWB itself, are identified to explain the heterogeneity. This paper finds that both subjective and objective income measures mediate the negative effects of unemployment and explain its heterogeneous effects on happiness. The approach in this paper complements that of Luo (2016b), who uses cross-sectional data and domain satisfaction as control in his analysis. The above findings confirm Luo s (2016b) conclusion that the root cause is pecuniary. More details follow and table I provides a summary of this paper s findings. First, by employing German Socio-Economic Panel (GSOEP) data, this paper controls individual level fixed effects (FE) in regressions. Individual FE help to eliminate the endogeneity generated by time-invariant omitted variables, such as personality. For example, personality is found to significantly impact both SWB (Feist et al. 1995) and 1 In this paper SWB, happiness, and life satisfaction are interchangeable. 2

several right hand variables in happiness regression, such as wealth and health (Graham, Eggers, and Sukhtankar 2004). Ferrer-i-Carbonell and Frijters (2004) find that whether or not individual FE are controlled gives different results. As a consequence, recent literature in unemployment and SWB often uses the individual FE model. Second, this paper uses various income correlated domain satisfactions (satisfaction with household income/standard of living/personal income) and finds that all these domain satisfactions (DS) are able to mediate the negative effects of unemployment. A potential argument is that DS is significantly correlated with overall SWB, so that the coefficients of other variables will be mediated. However, I show that income uncorrelated DS (with environment) does not affect the unemployment coefficient. Third, another potential argument is that DS is a special measure since it is regarded as important components of overall SWB (Michalos 1985, and van Praag, Frijters, and Ferrer-i-Carbonell 2003). I use subjective concerns to test this argument. Again, income correlated concern (about economic status) is able to mediate the negative effects, whereas income uncorrelated ones (about environment/peace) are unable to. Fourth, and most importantly, the subjective income measures (both DS and concern) explain the heterogeneity of unemployment. Winkelmann (2009), and Gielen and van Ours (2014) find that about half of those unemployed do not experience SWB loss. By collapsing those unemployed into different groups according to their DS/concern changes before and after unemployment, this paper verifies that positive DS/concern change may render the unemployment effects positive. On the contrary, the effects remain negative if the unemployed are collapsed by income uncorrelated DS/concern changes. This is the additional and stronger evidence that the root cause is pecuniary: loosely speaking, explaining the heterogeneity effects is the sufficient and necessary condition of being the root cause. Fifth, and equally importantly, objective foundations are provided. The link between DS/concern change and objective income is discussed. Moreover, the objective income can also mediate unemployment effects and explain the heterogeneity. (1) Generally, the more income after unemployment, or the higher living standard before unemployment, the more the DS/concern changes positively. (2) Taking advantage of the rich variables in GSOEP and the panel setting, two objective income measures are tested, showing that 3

they can mediate the negative effects of unemployment. Luo (2016b) uses financial satisfaction as additional control because DS is influenced by social comparison, adaptation, aspiration and expectation, all of which influence SWB (Michalos 1985). Actual income itself cannot capture such effects. Two income measures are chosen according to the function of DS. The first is lead and lag income, which captures adaptation and expectation. The second is minimum required income, the minimum net household income that one needs to support his or her current living standard. Stutzer (2004) regards this measure as a proxy for individual s income aspiration and finds that it is correlated with comparisons to one s past and to others. Regression results demonstrate that either of the two measures substantially reduces the coefficient of unemployment. (3) The minimum required income is able to explain the heterogeneity. Luo (2016a) proposes the fixed cost of living theory, which is able to explain the negative effects of unemployment. His theory suggests that individuals balance work and leisure only after the unavoidable expenses of current living standard (i.e., the fixed cost of living) are met. The simplest utility function is u(c, l) = (c F)l, where c is consumption (current income as a proxy), l is leisure, and F is the fixed cost of living (minimum required income as a proxy). When an individual is unemployed, the utility falls to u(c, l) = F(l + L) where L is previous working hours, assuming c = 0 after unemployment for simplicity. The utility gap is FL cl. After objective income loss is controlled for ( cl), there still remains an unexplained happiness decrease ( FL). Thus, controlling F should mediate the effects of unemployment. Moreover, the heterogeneity of SWB change could be explained by the change of c F. In formal regressions, those falling from c > F before unemployment to c < F after unemployment experience largest SWB decrease. On the contrary, those remaining c > F before and after unemployment (more precisely, from c > F to c F) experience SWB increase. Sixth, I demonstrate how the unemployment induced by company shutting down affects the above analysis. The literature suggests a backward causality between SWB and income/unemployment. For example, Graham, Eggers, and Sukhtankar (2004) find that happiness itself increases future income. Kassenboehmer and Haisken-DeNew (2009) suggest that individuals who are dissatisfied with their jobs may choose to become voluntarily unemployed. Some papers, such as Kassenboehmer and Haisken-DeNew 4

(2009), and Baetschmann, Staub, and Winkelmann (2015), use company closure to obtain exogenous unemployment. Such exogenous shock actually affects income and employment status simultaneously. Accordingly, I use this extended specification as a control of backward causality and see how the above analysis changes. Seventh, using nationally representative data, this paper shows that the root cause of unemployment s negative effects is pecuniary in Germany. Luo s (2016b) analysis is based on data from 98 countries. Each country has limited sample size. Moreover, SWB is influenced by different cultures (Oishi et al. 1999). This paper shows that the conclusion holds in a relatively narrow cultural background and for a large national data set. Furthermore, people in Germany and Europe are relatively supportive to a welfare state. We can expect that unemployment has more negative consequences in nations (e.g., USA) where people put more emphasis on self-responsibility. Finally, this paper has strong policy implications. The pecuniary root cause emphasizes the importance of unemployment benefits. The heterogeneity shows that the benefits should be means-tested, since those with sufficient income from other sources may not experience SWB loss from unemployment. Other supports, such as job search assistance, should focus on those with least income sources. Moreover, this finding justifies the progressive taxation: those with a large income from wealth or other sources might not experience SWB decrease from job loss so that they may choose to be unemployed voluntarily (Smith and Razzell 1975) 2. The paper proceeds as follows. Section II provides background information, including previous study and data source introduction. Section III describes empirical strategy. Section IV shows the results and section V concludes. II. Background 1. Previous studies Unemployment and SWB 2 Smith and Razzell (1975) also find that lottery winners choose not to work because their labor income faces a high marginal tax rate, due to their large capital income derived from investment. This implies to tax labor income and other sources of income separately. 5

The happiness literature finds that unemployment substantially decreases SWB, after actual income is controlled for (see Frey and Stutzer 2002 for a review). The general consensus is that the SWB loss is caused by non-pecuniary costs. Various factors are identified, such as social work norms, psychological effects (Winkelmann [2014] and the review section in Luo [2016b] contain more details). However, Luo (2016b) empirically shows that the root cause is pecuniary, using combined cross-sectional datasets World Values Survey (WVS) and European Values Study (EVS). He first suggests that actual income cannot capture all the pecuniary effects. On the contrary, financial satisfaction captures the effects of income comparisons to others, to one s past, to aspirations and to expectations; these comparisons all impact SWB (Michalos 1985). Including financial satisfaction as an additional control of pecuniary factors mediates 2/3 of unemployment s negative effects. He then categorizes the non-pecuniary factors into two subgroups: those with and without explanatory power in previous empirical studies. He finds that those with explanatory power (social work norms, social capital, expectation, psychological effects, and stigma) mediate fewer effects of unemployment. Moreover, they are compatible with the pecuniary explanation. For example, social work norms can be regarded as the norms to support one s family financially. When men are more obligated financially, unemployment hurts more for them. Based on above findings, Luo suggests that the root cause is pecuniary. Inspired by the above findings, Luo (2016a) proposes the fixed cost of living theory to explain the above findings. In his theory, individual balances work and leisure only after the unavoidable expenses (i.e., the fixed cost of living) are met. Such expenses, determined by an individual s living standard, include house related costs (real estate tax or renting), utility bills, etc. There are various associated utility functions, with the simplest one u(c, l) = (c F)l, where c is consumption (current income as a proxy), l is leisure, and F is the fixed cost of living (minimum required income as a proxy). When an individual is unemployed, the utility falls to u(c, l) = F(l + L) where L is previous working hours, assuming c = 0 after unemployment for simplicity. The utility gap is FL cl. After actual income is controlled for ( cl), the happiness decrease ( FL) remains unexplained. He tries to explain more puzzles in the literature using this theory. 6

Unemployment and heterogeneity Although in general unemployment reduces SWB, Winkelmann (2009) and Gielen and van Ours (2014) find that about half of those unemployed do not experience SWB loss, which prompts further studies. Binder and Coad (2015a, 2015b), both using British Household Panel Survey data and quantile regression, find that individuals with higher SWB suffer less from unemployment. In other words, no factors, except SWB itself, are identified to explain the heterogeneity. 2. Data and summary statistics The German Socio-Economic Panel (GSOEP) is a national representative longitudinal study of private households, conducted every year since 1984 by German Institute for Economic Research (DIW Berlin). This dataset contains rich information, including both objective and subjective aspects. See Haisken-DeNew and Frick (2005) for more details. I use the waves from 1984 to 2012, and restrict the age from 16 to 65, inclusively. The dependent variable, overall life satisfaction (LS), derives from the question: how satisfied are you with your life, all things considered? The answers range from 0 to 10, where [0] for completely dissatisfied and [10] for completely satisfied. The appendix contains more detailed definition regarding the variables. The pecuniary factors include monthly net household income, household income satisfaction, and economic concern. Incomes are deflated by consumer price index, with the base year 2006. As a common practice, I use the log income. Household income satisfaction arises from the question: how satisfied are you with your household income? [0] for completely dissatisfied and [10] for completely satisfied. Economic concern comes from: how concerned are you about your own economic situation? [1] for very concerned, [2] for somewhat concerned, and [3] for not concerned at all. Employment status is 3-scale dummies, where [0] for employed (Emp), [1] for not in labor force (NLF), and [2] for unemployed (Unemp). Demographic controls include years of education, marital status (6-scale dummies), disability (2-scale dummies where [1] for disabled), age, age square ( ), survey year (29-scale dummies), and residential state 7

(16-scale dummies). I also include 7 dummy variables indicating the number of household members aged 0-1/2-4/5-7/8-10/11-12/13-15/16-18 years old. For example, the number of members aged 0-1 years is a 4-scale dummies, where [n] means currently there are n members aged 0-1 years in this household. All the controls above are frequently used in the literature of SWB. I use only disability as a proxy of health status, because self-rated health status is available for limited years, while using health satisfaction may be criticized by using domain satisfaction as control. I also try the regressions using self-rated health or health satisfaction. They return qualitatively similar results. Invalid responses of gender are excluded. But gender will not be controlled in fixed effects regression since there is no variation. After the data clean (age restriction and exclusion of invalid responses), the basic sample contains 43,872 individuals and 331,718 observations. The following variables are not involved in the data clean process: minimum required household income, living standard satisfaction, personal income satisfaction, environment satisfaction, environment concern, and peace concern. Some of them are available in limited years. See the appendix for more details on variable definitions. Table II provides the summary statistics by the employment status. Compared to those employed, the unemployed are less satisfied with their lives and various domain satisfactions correlated with income (household income/living standard/personal income satisfaction), are more concerned about economics status, have less monthly income and lower living standard (proxy by minimum income), are older and less likely to be married, and are more likely to be disabled. For the three groups, the patterns of their income related domain satisfactions and concerns are very similar to that of the life satisfaction. On the contrary, the scores are almost the same for the domains and concerns uncorrelated with income (environment and peace). This is the preliminary evidence that unemployment affects life satisfaction through pecuniary factors. III. Empirical Strategy I use individual level fixed effects ordinary least squares (OLS-FE) model in this paper because the results of OLS regression enables easy interpretation. Moreover, 8

Ferrer-i-Carbonell and Frijters (2004) verify that treating ordinal SWB scales as cardinal by OLS gives similar results, whereas controlling for individual fixed effects produces substantial difference. The specification is the traditional happiness regression LS = α + β ES + γ EXP + θ X + ε where LS is the life satisfaction of individual i in year t, α captures individual fixed effect, ES is employment status (3-scale dummies), EXP includes explanatory variables (various subjective evaluations or objective income measurements, in different specifications), X is demographic controls (years of education, marital status, disability, age, age square, survey year, residential state, and dummies for number of household members in different ages), and ε is random error. I first replicate Luo s (2016b) approach that financial satisfaction substantially mediates the negative effects of unemployment. I test different income correlated domain satisfactions (satisfaction with household income/living standard/personal income). Income uncorrelated domain satisfaction (satisfaction with environment) is also tested as a comparison. Then I verify how income correlated subjective evaluation (economic concern) and uncorrelated ones (environment/peace concern) affect unemployment. Next, I try to explain the heterogeneity of unemployment by subjective income measures. In individual fixed effects model, the unemployment coefficient comes from the life satisfaction change ( LS), i.e., the difference in LS between employment and unemployment, ceteris paribus, given the reference group is the employed. Previous studies try to explain LS by absolute level explanatory variables, in traditional or quantile regressions. But they do not return a promising result. In this paper, I try to explain LS by the change of explanatory variables, EXP. Specifically, I expand the employment status dummy to capture both unemployment and EXP. For example, when the explanatory variable is 11-scale (0 to 10) domain satisfactions (DS), the employment status is expanded to 22-scale, where [0] for employed, [1] for not in labor force, [2] for unemployed AND DS = 10,, [22] for unemployed AND DS = 10. Similarly, when the explanatory variable is 3-scale (1 to 3) subjective concerns (CON), the employment status is 7-scale, where [0] for employed, [1] for not in labor force, [2] for unemployed AND CON = 2,, [6] for unemployed 9

AND CON = 2. The explanatory variable is not included in such regression to avoid redundancy. The corresponding regression equation becomes LS = α + β ES + θ X + ε where employment status is 22-scale (for DS) or 7-scale (for CON) dummies. Similar to LS, EXP should capture the change in explanatory variable between employment and unemployment: EXP = EXP EXP. When a person is unemployed, EXP is his or her current level explanatory variable. The problem, however, is if an individual has multiple observations as employed, how to define the employment benchmark to obtain EXP? To preserve as many unemployment observations as possible, I define the benchmark observation as the last employment before the first unemployment. If such employment observation does not exist, then the benchmark is the first employment in all the observations. For example, if a person s employment sequence is Emp, Emp, NLF, Unemp, and Emp, then the benchmark is the second Emp. Instead, if a person s employment sequence is Unemp, Emp, Emp, and Unemp, then the first Emp acts as benchmark. Under this definition, unemployed individuals in any given year do not contribute to the heterogeneity analysis. In this analysis, if an explanatory variable is able to explain the heterogeneity of unemployment, then the coefficient should increase and become positive as the value of employment dummy increases ( EXP increases). On the contrary, if a variable has no explanatory power, the coefficients will remain negative. The reasons of EXP are also discussed. Then, I try to demonstrate whether various objective income measures, aside from current household income, are able to mediate the effects of unemployment and even explain the heterogeneity effects. (1) The first income measure is the lag and lead income. Luo (2016b) uses financial satisfaction as additional control because it captures the effects of income comparisons to others, to one s past, to aspirations and to expectations; these comparisons all impact SWB (Michalos 1985). Lead and lag income captures the effects of adaptation and expectation. Moreover, recent researches, such as Clark et al. (2008), Powdthavee (2012), and Clark and Georgellis (2013), find that various life events, including unemployment, have lead (anticipation) and lag effects on SWB. If the root 10

cause of unemployment s negative effects is pecuniary, I expect the lead and lag income to reduce the coefficient of unemployment. Denoting current year as t, the income from t 4 to t + 4 is controlled in this specification, since the normal practice is to control 4-5 years before and after an event (for example, Clark et al. 2008, Clark and Georgellis 2013, and Powdthavee 2009). Unfortunately, there is no formal test to determine whether this income measure is able to explain the heterogeneity of unemployment. (2) The second objective income measure is the minimum required income for current living standard. This income measure also fits Luo s (2016b) rationale for using financial satisfaction, because Stutzer (2004) regards this measure as individual s aspirations and finds that it is correlated with the adaptation and social comparison. If unemployment hurts an individual by income adaptation and social comparison of income, including this measure should mediate the effects of unemployment. Moreover, Luo s (2016a) fixed cost of living theory suggests that individual balance work and leisure only after the unavoidable expenses of current living standard (i.e., fixed cost of living) are met. An example of utility function is u(c, l) = (c F)l, where c is consumption (current income as a proxy), l is leisure, and F is the fixed cost of living (minimum income as a proxy). Thus controlling F should mediate the effects of unemployment. Furthermore, his theory suggests that the heterogeneity of SWB change could be tested by the change of c F. Specifically, the employment status dummy is expanded: [0] for Emp, [1] for NLF, [2] to [5] for Unemp, where [2] denotes c > F and c < F, [3] c F and c < F, [4] c > F and c F, and [5] c F and c F. A drawback of this specification is that the variable minimum income for current living standard is available for only 2002, 2007, and 2012. We should interpret the results derived from this small sample with caution. Finally, the above analysis is repeated in an extended specification, focusing on those company shutting down induced unemployment. This specification is a control of backward causality between income/unemployment and SWB. A caveat is that there are only 2,534 observations of company shut down unemployment. 11

IV. Results 1. Subjective income measure and mediating effects Table III shows the mediating effects of domain satisfactions. Column (1) is the result without income or domain satisfactions. The coefficient of Unemp is -0.73, indicating that falling from Emp to Unemp reduces 0.73 of 11-scale life satisfaction. After actual household income is controlled in column (2), the effects remain similar (- 0.66). The coefficient of log household income is 0.32 and statistically significant. The above findings are consistent with the literature that log income s coefficient in happiness regression is about 0.3, and unemployment hurts even after actual income is controlled for. The coefficients of other variables are not listed due to space constraints. The interpretations are consistent with the literature: marriage increases happiness, SWB is U- shaped in age, and disability reduces SWB. The years of education have no significant impacts on SWB. After age 16, the change of education within a person is relatively small. The small variation may reduce the statistical significance. To verify this hypothesis, I run regression without individual fixed effects (the results are not shown here). The education coefficient becomes significant. Column (3), however, shows that the effects of unemployment decrease substantially to -0.41 after household income satisfaction is controlled for. About half of the negative effects of unemployment are mediated by actual income and subjective income measures. For other income correlated domain satisfactions, namely living standard satisfaction in column (4) and personal income satisfaction in column (5), the results are similar. Note the coefficient of log income decreases substantially or even becomes insignificant, consistent with the literature that subjective income measure mediates the effects of actual income (for example, Johnson and Krueger 2006). Several specifications are tested while not listed due to space constraint. First, the OLS without individual fixed effects shows that the negative effects of Unemp are much larger (-1.17) without controlling for pecuniary factors. The log income and household income satisfaction explains about 2/3 of the negative effects of Unemp, similar to the result of Luo (2016b). Second, I run the OLS-FE model for men and women separately. 12

Unemployment affects men more negatively than women (-0.91 and -0.56) without controlling for pecuniary factors. After log income and household income satisfaction are included, the negative effects of Unemp are still larger for men (-0.51 and -0.31). The literature uses stronger work norm for men to explain this finding, while Luo (2016b) suggest a financially supporting norm: males are more obligated to support their families financially. A potential objection to this approach is that domain satisfaction (DS) and overall life satisfaction are both subject to the mood effects. For example, Schwarz and Strack (1999) suggest that global evaluation could be based on one s current mood. It is natural to suspect that DS is also affected by mood, and including DS in the regression will mediate the effects of unemployment. If this objection holds, an income uncorrelated DS will also generate similar result. I use environmental satisfaction to test this objection. Since this variable is available in limited years, I first replicate the specification (2) in column (6). The coefficient of unemployment is -0.60 and the coefficient of log income is 0.32, similar to column (2). In column (7), environment satisfaction is included. The coefficient of DS is large and significant, consistent with the literature that domain satisfactions are important components of overall life satisfaction (for example, van Praag, Frijters, and Ferrer-i- Carbonell 2003, and Powdthavee 2012). However, the coefficients of unemployment and log income remain almost the same, i.e., income uncorrelated DS does not mediate the effects of income or employment. If unemployment s effects are reduced by income correlated DS only, it is the evidence that unemployment affects SWB through pecuniary channels. Another potential objection is that DS is a special control since it is an important component of overall life satisfaction. I use various subjective concerns to test this argument. Table IV lists the result. Column (1) shows that the negative effects of unemployment and positive effects of log income again are mediated, after concern about economic status is included (comparing with table III column [2]). However, when the concerns are uncorrelated with income, concerns about environment/peace in column (2)/(3), the coefficients of unemployment and income remain almost the same as table III 13

column (2). The results confirm that unemployment affects SWB through pecuniary factors. There are two caveats of this approach. First, compared to income correlated DS, the concern about economic status mediates less effects of unemployment and income. The DS may capture more variations due to its 11-scale measurement. To test this hypothesis, I reorganize the household income DS into 3-scale concern: DS [1] to [4] as very concerned [1], [5] to [7] as somewhat concerned [2], and [8] to [10] as not concerned at all [3]. Thus, the distribution of the reorganized DS is similar to that of economic concern. Table IV column (4) lists the regression result. The reorganized DS mediates slightly more effects of unemployment yet much more effects of log income. Maybe the concerns as negative evaluations have less power in explaining positive overall life satisfaction, compared to positive evaluations (domain satisfactions). The happiness literature categorizes the SWB measure into three subgroups: global evaluation (such as life satisfaction), positive affect, and negative affect (for example, Busseri and Sadava 2011). This paper may suggest that evaluation be separated into positive and negative evaluations. The second caveat is that the coefficients of environment/peace concerns are statistical significant but small. It could be the reason that such concerns seldom mediate the effects of unemployment and income. This caveat again implies that negative evaluation is different with positive one. 2. Subjective income measure and heterogeneity Table V shows the results of heterogeneity analysis. Column (1) demonstrates that the negative effects decrease as DS (household income satisfaction, in this column) increases. Unemployment brings positive effects after DS 4. In this table, log household income is not controlled for; including it in regressions generates similar results. Column (2) provides the LS of each group. As expected, LS increases steadily as DS increases, explaining the result of column (1). Column (3) shows the number of observations of each group. The distribution is an asymmetric bell shape: many more unemployed are located in DS < 0 than in DS > 0 (or in LS < 0 than in LS > 0), explaining why in mean regressions the unemployment is associated with large SWB 14

decreases. Note that the observations for DS values 9 and 10 are too small to obtain significant result. The regressions of living standard (column 4) and personal income (column 5) satisfactions generate similar but different patterns. For living standard, the effects of unemployment become insignificant since DS = 1. For personal income, the effects are much more devastating for those DS < 0. It suggests that unemployment hurts more in the domain of personal income because unemployment directly decreases individual s labor income, whereas it hurts less in the domain of living standard since in Germany there are various supports for those unemployed. However, the pattern for income uncorrelated environment DS is different, as shown in column (6). Although the coefficients increase slowly as DS increases, there is no clear trend for coefficients from DS = 0 to 6. Moreover, from DS = 7 to 10, although the coefficients are not significant due to small observation, they are generally negative. That is, satisfaction with environment does not explain the heterogeneity of unemployment. The categorization based on subjective concerns generates similar results, as in columns (7) to (9). The 3-scale concerns produce 5-scale EXP, from -2 to 2. The advantages of fewer categories are that there are more observations in each category and that they reveal clearer patterns. For income-related concern (about economics status), the coefficients increase steadily and become positive in the end. For environment/peace concerns, however, the trend is not clear, and the coefficients remain negative and significant. The above evidence demonstrates that subjective income evaluations are able to explain the heterogeneity effects of unemployment, whereas income uncorrelated evaluations fail to do it. The coefficients of EXP = 0 are negative, consistent with previous findings that subjective income measures are unable to fully mediate the negative effects. However, since pecuniary costs explain the heterogeneity effects, such costs should be the root cause of why the unemployed are so unhappy. Arbitrarily speaking, explaining the heterogeneity effects should be the sufficient and necessary condition of being the root cause. 15

3. Preliminary explanations Explaining why some people obtain higher household income satisfaction after unemployment is beyond the scope of this paper. But I try to provide some preliminary analysis based on simple regressions (the results are not shown here). These regressions are restricted on those unemployed, with the dependent variable DS where DS is household income satisfaction. A simple regression demonstrates that the more a person believe he or she can find a job easily, the higher DS. This is consistent with Gielen and van Ours s (2014) suggestion that individuals who believe they are likely to find a job very soon might regard the short-term unemployment as a rest and enjoy the leisure. Note, however, that expectation itself cannot explain the heterogeneity: if the unemployed are collapsed into different groups according to their job prospective, all the coefficients remain negative and significant. Moreover, the expectation explanation is compatible with the pecuniary one: those who can obtain jobs soon will not suffer from income loss after being employed. I also try to link the subjective and objective income measures. I regress DS on current household income (c ) and minimum required income (F ). The higher the income, the higher DS. The lower the unavoidable expense F, the higher the DS, although the coefficient is not significant. I also regress DS on previous household income (c ) and minimum required income (F ). This time the coefficient of c is negative (though insignificant), i.e., the higher the income during employment, the more negative the DS (and so the LS). The coefficient of F is positive and significant. Higher F refers to higher living standard. Maybe such families have more savings or other income resources as a buffer to deal with unemployment. The opposite signs between c and c (and between F and F ) are consistent with the literature of income adaptation (Frederick and Loewenstein 1999). The positive effects of c and F imply that money increases the SWB of those unemployed. If actual income cannot capture all the pecuniary effects (as suggested by Luo 2016b and this paper), then the previous finding in quantile regressions (Binder and Coad 2015a, 2015b find that higher SWB itself mediates the effects of unemployment) may actually capture 16

the buffering effects of income. The following subsection shows that objective income measures themselves are able to mediate the effects of Unemp and explain the heterogeneity. 4. Objective income measures Table VI gives the regressions about objective income measures. Column (1) shows that after log income of t 4 to t + 4 is included in the regression, the negative effects reduce to -0.54, compared with -0.66 in column (2) of table III. The coefficients are negative for previous and positive for future income, consistent with the literature about adaptation (Frederick and Loewenstein 1999) and expectation (Knabe and Rätzel 2011). Column (2) controls minimum income for current living (F), providing similar results. The coefficient of minimum income F is negative, consistent with the theory of Luo (2016a) and the empirical finding of Stutzer (2004). Both the lead and lag income and the minimum income mediate less effects of unemployment than domain satisfactions (column [3] to [5] of table III) but more than subjective concern (column [1] of table IV). The minimum income F is able to explain the heterogeneity, as shown in column (3). The unemployed are collapsed into 4 categories, according to the change of c F, where consumption c is measured by actual income. For the group with c F > 0 before unemployment but c F < 0 after unemployment, unemployment reduces their income to the level that they cannot meet their current unavoidable expenses. Unsurprisingly, the negative effects -1.31 is the largest for them. For the group with c F < 0 for both before and after unemployment (more precisely, from c F 0 to c F < 0), the effects -0.50 are less negative. On the contrary, for the group who maintains c F > 0 for both periods (more precisely, from c F > 0 to c F 0), the effects become positive and significant at 0.72. This finding explains the heterogeneity and again confirms that money is the buffer to unemployment. A seemingly surprising finding is the negative effects for the group with c F 0 before and c F 0 after unemployment. The reason is that this group has two special characteristics. First, the average household income increases substantially from 1,514 to 2,082 Euro, while other three groups all experience income decrease after unemployment. 17

Since unemployment benefit cannot fully compensate the income loss from labor, such increase may come from extra working of other family members, which is likely to stigmatize those unemployed. Second, the unavoidable expenses F drop from 1,968 to 1,467 Euro, implying sharp decrease in living standard. The deterioration of living conditions should decrease the SWB of those unemployed. A caveat for the analysis involving minimum income is that the variable is only available in 2002, 2007, and 2012, which reduces the sample size for those unemployed. There are only 591 observations in the 4 groups of column (3). Column (4) shows the sample size change more clearly. If the log minimum income is included as additional control of the whole sample, then the result is similar while sample size reduces to 32,462 observations. 5. Company closure as an exogenous shock Table VII shows the results of the extended specification of company shutting down. For simplicity, the unemployed who enter unemployment due to other reasons are excluded from the sample. Note that only 2,534 unemployed observations remain, and the results should be interpreted with caution. Columns (1) to (3) give the results of mediating effects of pecuniary factors. Actual income (column [2]) slightly mediates the effects, compared with column (1). Household income satisfaction substantially reduces the unemployment coefficient, as shown in column (3). The coefficients in these three specifications are larger than those of the whole sample (columns [1] to [3] in table III), consistent with the literature that company closure brings further SWB decrease (Kassenboehmer and Haisken-DeNew 2009, and Baetschmann, Staub, and Winkelmann 2015). Other domain satisfactions (DS) and subjective concerns give the same conclusion as before: income correlated DS and concern substantially mediate the effects of unemployment, while income uncorrelated ones have negligible impact. When correlated with income, subjective concern is able to explain the heterogeneity, as shown in column (4). Concerns about environment (column [5]) and peace (not shown here) do not have such explanatory power. Due to small sample of unemployment, 18

collapsing by DS results in no observations in some subgroups and insignificant coefficients. The effects of objective income measures are shown in column (6) and (7). Lead and lag income has negligible effects (column [6]). Log minimum required income again mediates the effects of Unemp (column [7]). Heterogeneity analysis cannot be conducted for objective measures. The minimum required income is only available in 2002, 2007, and 2012. Collapsing the unemployed by F for both periods (before and after unemployment) results in none observations. V. Conclusion The happiness literature finds that unemployment largely decreases SWB, even after controlling for income. However, about half of those unemployed do not experience SWB loss. So far no factors, except SWB itself, are identified to explain the heterogeneity. Using German Socio-Economic Panel (GSOEP) data, this paper finds that both subjective and objective income measures mediate the negative effects of unemployment and explain its heterogeneous effects on happiness. This finding confirms Luo s (2016b) suggestion that the root cause of negative effects of unemployment on happiness is pecuniary. This paper has strong policy implications. Since the root cause of the happiness decline is pecuniary, unemployment benefits can play an important role in improving welfare among job losers. The heterogeneous effects of unemployment imply that such benefits should be means-tested and that other supports, such as job search assistance, should focus on those with the fewest sources of income. Moreover, this heterogeneity justifies progressive taxation because those with a large income from wealth or other sources may choose to be unemployed voluntarily (Smith and Razzell 1975). 19

Appendix This Appendix contains definitions of variables. Life Satisfaction: How satisfied are you with your life, all things considered? Please answer on a scale from 0 to 10. [0] for completely dissatisfied and [10] for completely satisfied. Monthly Net Income: The net monthly income of all of the members of your household. That means the income after deductions for taxes and social security, including regular income such as pensions, housing allowances, child benefits, grants for higher education, maintenance payments, etc. Household Income/Living Standard/Personal Income/Environment Satisfaction: How satisfied are you today with the following areas of your life? Your household income/your overall standard of living/your personal income/the environmental conditions in your area. [0] for completely dissatisfied and [10] for completely satisfied. Economic/Environment/Peace Concern: How concerned are you about the following issues? Your own economic situation/environmental protection/maintaining peace. [1] for vey concerned, [2] for somewhat concerned, and [3] for not concerned at all. Minimum Required Income: What would you personally consider the minimum net household income you would need in your current living situation? We are referring here to the net monthly income that your household would need to get by. 20

Reference Baetschmann, Gregori, Kevin E. Staub, and Rainer Winkelmann. 2015. Consistent Estimation of the Fixed Effects Ordered Logit Model. Journal of the Royal Statistical Society: Series A (Statistics in Society). doi:10.1111/rssa.12090. Binder, Martin, and Alex Coad. 2015a. Unemployment Impacts Differently on the Extremes of the Distribution of a Comprehensive Well-Being Measure. Applied Economics Letters 22 (8): 619 27. doi:10.1080/13504851.2014.962219.. 2015b. Heterogeneity in the Relationship Between Unemployment and Subjective Wellbeing: A Quantile Approach. Economica 82 (328): 865 91. doi:10.1111/ecca.12150. Busseri, Michael A., and Stan W. Sadava. 2011. A Review of the Tripartite Structure of Subjective Well-Being: Implications for Conceptualization, Operationalization, Analysis, and Synthesis. Personality and Social Psychology Review 15 (3): 290 314. doi:10.1177/1088868310391271. Clark, Andrew E., Ed Diener, Yannis Georgellis, and Richard E. Lucas. 2008. Lags And Leads in Life Satisfaction: A Test of the Baseline Hypothesis. The Economic Journal 118 (529): F222 43. doi:10.1111/j.1468-0297.2008.02150.x. Clark, Andrew E., and Yannis Georgellis. 2013. Back to Baseline in Britain: Adaptation in the British Household Panel Survey. Economica 80 (319): 496 512. doi:10.1111/ecca.12007. Feist, Gregory J., Todd E. Bodner, John F. Jacobs, Marilyn Miles, and Vickie Tan. 1995. Integrating Top-down and Bottom-up Structural Models of Subjective Well-Being: A Longitudinal Investigation. Journal of Personality and Social Psychology 68 (1): 138 50. doi:10.1037/0022-3514.68.1.138. Ferrer-i-Carbonell, Ada, and Paul Frijters. 2004. How Important Is Methodology for the Estimates of the Determinants of Happiness? The Economic Journal 114 (497): 641 59. doi:10.1111/j.1468-0297.2004.00235.x. Frederick, Shane, and George Loewenstein. 1999. Hedonic Adaptation. In Well-Being: The Foundations of Hedonic Psychology, edited by D. Kahneman, E. Diener, and N. Schwarz, 302 29. New York, NY, US: Russell Sage Foundation. Frey, Bruno S., and Alois Stutzer. 2002. What Can Economists Learn from Happiness Research? Journal of Economic Literature 40 (2): 402 35. 21

Gielen, Anne C., and Jan C. van Ours. 2014. Unhappiness and Job Finding. Economica 81 (323): 544 65. doi:10.1111/ecca.12089. Graham, Carol, Andrew Eggers, and Sandip Sukhtankar. 2004. Does Happiness Pay? An Exploration Based on Panel Data from Russia. Journal of Economic Behavior & Organization 55 (3): 319 42. doi:10.1016/j.jebo.2003.09.002. Haisken-DeNew, John P., and Joachim R. Frick. 2005. Desktop Companion to the German Socio-Economic Panel (GSOEP). Technical report. https://www.diw.de/documents/dokumentenarchiv/17/diw_01.c.38951.de/dtc.4 09713.pdf. Johnson, Wendy, and Robert F. Krueger. 2006. How Money Buys Happiness: Genetic and Environmental Processes Linking Finances and Life Satisfaction. Journal of Personality and Social Psychology 90 (4): 680 91. doi:10.1037/0022-3514.90.4.680. Kassenboehmer, Sonja C., and John P. Haisken-DeNew. 2009. You re Fired! The Causal Negative Effect of Entry Unemployment on Life Satisfaction. The Economic Journal 119 (536): 448 62. doi:10.1111/j.1468-0297.2008.02246.x. Knabe, Andreas, and Steffen Rätzel. 2011. Scarring or Scaring? The Psychological Impact of Past Unemployment and Future Unemployment Risk. Economica 78 (310): 283 93. doi:10.1111/j.1468-0335.2009.00816.x. Luo, Jianbo. 2016a. The Fixed Cost of Living and Its Applications. Mimeo.. 2016b. Unemployment and Unhappiness: The Role of Pecuniary Factors. SSRN Scholarly Paper ID 2813844. Rochester, NY: Social Science Research Network. http://papers.ssrn.com/abstract=2813844. Michalos, Alex C. 1985. Multiple Discrepancies Theory (MDT). Social Indicators Research 16 (4): 347 413. Oishi, Shigehiro, Edward F. Diener, Richard E. Lucas, and Eunkook M. Suh. 1999. Cross-Cultural Variations in Predictors of Life Satisfaction: Perspectives from Needs and Values. Personality and Social Psychology Bulletin 25 (8): 980 90. doi:10.1177/01461672992511006. Powdthavee, Nattavudh. 2009. What Happens to People before and after Disability? Focusing Effects, Lead Effects, and Adaptation in Different Areas of Life. Social Science & Medicine, Part Special Issue: New approaches to researching patient safety, 69 (12): 1834 44. doi:10.1016/j.socscimed.2009.09.023.. 2012. Jobless, Friendless and Broke: What Happens to Different Areas of Life Before and After Unemployment? Economica 79 (315): 557 75. doi:10.1111/j.1468-0335.2011.00905.x. Schwarz, Norbert, and Fritz Strack. 1999. Reports of Subjective Well-Being: Judgmental Processes and Their Methodological Implications. In Well- 22

Being: The Foundations of Hedonic Psychology, edited by D. Kahneman, E. Diener, and N. Schwarz, 61 84. New York, NY, US: Russell Sage Foundation. Smith, Stephen, and Peter Razzell. 1975. The Pools Winners. London: Caliban Books. Stutzer, Alois. 2004. The Role of Income Aspirations in Individual Happiness. Journal of Economic Behavior & Organization 54 (1): 89 109. doi:10.1016/j.jebo.2003.04.003. van Praag, B. M. S, P Frijters, and A Ferrer-i-Carbonell. 2003. The Anatomy of Subjective Well-Being. Journal of Economic Behavior & Organization 51 (1): 29 49. doi:10.1016/s0167-2681(02)00140-3. Winkelmann, Rainer. 2009. Unemployment, Social Capital, and Subjective Well-Being. Journal of Happiness Studies 10 (4): 421 30. doi:10.1007/s10902-008-9097-2.. 2014. Unemployment and Happiness. IZA World of Labor. doi:10.15185/izawol.94. 23

Table I. Summary of the findings Subjective Measure Domain Satisfaction Concerns About Objective Measure Income Related Hh Income; Living Standard; Personal Income Economic Situation Income Uncorrelated Environment Environment; Peace Lag and Forward Income Minimum required Income Mediate Unemployment Effect? Yes No Yes Yes Explain Heterogeneity of Unemployment? Yes No No Yes 24

Table II. Summary statistics (1) (2) (3) Group Emp Unemp Not in labor force Life Satisfaction (10-scale) 7.1 5.7 7.0 Hh Monthly Net Income (Euro) 2918 1737 2411 Hh Income Satisfaction (10-scale) 6.5 4.2 6.2 Economic Concern (3-scale) 2.1 1.5 2.1 Age 40 43 44 Education (years) 12 11 11 Married (%) 64 56 68 Disabled (%) 4.8 10.2 16.7 Male (%) 55 51 30 Observation 231,161 20,546 80,011 Less observations Living Stand Satis (10-scale) 7.1 5.6 6.9 Personal Inc Satis (10-scale) 6.2 2.9 4.6 Environment Satis (10-scale) 6.3 5.9 6.4 Environment Concern (3-scale) 1.8 1.8 1.7 Peace Concern (3-scale) 1.8 1.7 1.7 Minimum Required Income (Euro) 2359 1591 1925 GSOEP. Age 16 to 65, inclusively. The sample size is smaller for variables below less observations. For domain satisfactions and various concerns, the higher the score, the more satisfied and less concerned. 25