Employment, late-life work, retirement, and well-being in Europe and the United States

Unemployment and underemployment, aging populations, and unsustainable retirement policies, among others, pose serious challenges for welfare states around the world (Derviş 2013). The global financial crisis and its uneven recovery have heightened these issues, with the new labor force entrants taking the hardest hit. In 2012, the youth unemployment rate (ages 15–24) peaked to 53 percent in Spain and nearly 40 percent in Portugal, compared with 16 percent in the U.S. (OECD 2013). The young are disproportionately affected even in countries with relatively lower unemployment rates such as the U.S. (Burtless 2013), and the long-term effects of delayed labor force entrance could be especially harmful. On the other end of the age distribution, unsustainable fiscal deficits in many countries are forcing the reconsideration of public pensions and retirement schemes. Given higher life expectancies and low fertility rates in the developed world, moreover, the share of older workers will continue to grow.

One solution to these challenges, recently proposed by Kemal Derviş (2013), entails more individual choice regarding employment, including a scheme whereby older workers gradually lower their working hours but remain in the labor force (and pay taxes) until the age of 70¹. Gradual retirement could be beneficial for governments, employers, and workers. Specifically, from a public finance perspective, phased-in retirement will likely increase tax revenues and reduce expenditures on pensions and retirement benefits. Moreover, while European employers view older workers as a burden (vanDalen et al. 2010), late-life employees are valuable to organizations as they have experience and competence. Burtless (2013), for example, documents the pay premium that late-life workers receive compared to their younger counterparts in the U.S., suggesting that older workers are more productive. Finally, late-life work can be beneficial for older employees as well. Research shows that late-life work has a positive effect on well-being in the U.S. (Calvo 2006) while involuntary retirement decreases well-being in the U.S. and Germany (Bonsang and Klein 2012; Calvo et al. 2009).

From a general equilibrium perspective, phased-in retirement constitutes an increase in the supply of older workers, which decreases their wages. As a result, if young and old workers are substitutes, employers may prefer to hire older workers which in turn reduces both the number of young people employed and their wages². The fiscal contributions from extended retirement could be used to mitigate any harmful effects on younger workers through apprenticeships and training programs. While general equilibrium analyses on the effects of altering employment and retirement schemes rarely factor in subjective well-being, work arrangements have differential effects for the happy and unhappy people, moreover.³

A priori, the potential individual well-being effects of changing employment and retirement policies are ambiguous. The results from smaller-scale experiments or experiences with altering these schemes from other countries may be misleading as scale and context matter. We therefore attempt to inform the policy discussions, as outlined by Derviş (2013) and in Aaron and Burtless (2013), among others, by furnishing insights from the relatively novel “science” of well-being measurement. Specifically, we explore the relationship between voluntary part-time employment, late-life work, and retirement and various well-being dimensions and job satisfaction using individual-level data from the Gallup World Poll (GWP) for a number of European countries and the U.S.

We build on several strands of research. It is well-established that the unemployed have lower well-being levels than the employed in virtually every context that relationship has been studied (Clark et al. 2001; Winkelmann and Winkelmann 1998; Hetschko et al. 2013; DiTella et al. 2001; Clark and Oswald 1994; Ravallion and Lokshin 2001). In European countries, for example, the jobless experience a life satisfaction reduction of 0.33 (on a scale of 1 “not at all satisfied” to 4 “very satisfied”) (DiTella et al. 2001) and even if compensated for the income loss, the unemployed would still be less happy compared to identical employed counterparts (Frey 2008). Clark et al. (2001) and Clark (2003) show, moreover, that unlike other adverse life events, past unemployment has lasting perverse effects on subjective well-being. The psychology literature also documents the “scarring” effects on the well-being of the long-term unemployed (Lucas 2007; Lucas et al. 2004). Indeed, long-term unemployment stands in contrast to other major life-events, such as divorce and physical injury, to which individuals adapt (i.e., return to their baseline well-being levels).

Retirees’ well-being, meanwhile, varies across countries because of different retirement norms and the generosity of public pensions. Research shows that the retired are more satisfied with their lives than the average in the U.S., less satisfied than the average in Russia, and the same as the average in Latin America (Graham and Pettinato 2002; Graham 2009)⁴. A study of male Canadian retirees demonstrates that the relationship between life satisfaction and retirement remains relatively stable over the course of retirement (Gall et al. 1997)⁵. Early on, recent retirees experience a life satisfaction boost and retirees’ perceptions of health, satisfaction with activity and overall life satisfaction persist 6–7 years post-retirement. A study using panel data from the U.S. shows, however, that retirement can lead to three psychological well-being patterns (Wang 2007). The majority of retirees maintain the same levels of pre-retirement well-being throughout their retirement years by sticking to their patterns of thought and behavior. A second group of retirees, especially those who disliked their jobs, experience a gain in psychological well-being which lasts throughout their retirement years. Finally, a third group of retirees experience a sharp decline in well-being due to loss of identity and status followed by a well-being recovery. Much less is known, however, about how different working and retirement arrangements – such as part-time work and flexible employment– correlate with well-being for older workers (as well as for other age cohorts).

Furthermore, employment arrangements and job types play a role for subjective well-being (SWB) comparisons. Knabe and Rätzel (2010) show that those who are unemployed but expecting to return to the labor force soon are no less happy than those who are employed in jobs with medium job security, and are in fact happier than those with minimal job security⁶. Several studies find that the self-employed are more satisfied with their jobs than those working for an employer in the OECD countries (Dolan et al. 2008; Binder and Coad 2011) and in other contexts (Benz and Frey 2008). In contrast, while mixed, the evidence from Latin America seems to suggest that the self-employed are less satisfied with their lives and jobs, due to the precarious nature of self-employment in the region – self-employment there is more likely to be involuntary and in the informal sector (Graham and Pettinato 2002; Graham and Felton 2006). Indeed, of the self-employed in Latin America, only business owners are more satisfied with their lives, incomes, and jobs, while the self-employed in more precarious occupations, such as farmers, fishermen, and informal employees, are less satisfied than the average (Aguilar et al. 2013).

Moreover, social norms related to employment mediate the employment-SWB relationship (Hetschko et al. 2013). While the unemployed are unhappier than the employed, the well-being effects are mitigated by the local unemployment level or norms about the generosity of the welfare system. For example, the unemployed in Britain are less unhappy when the regional unemployment rate, which approximates the strength of the social norm, is higher (Clark and Oswald 1994; Clark 2003; Shields and Price 2005). This result holds across different countries such as Australia (Shields et al. 2009), South Africa (Powdthavee 2007), Germany (Clark et al. 2010), and Russia (Eggers et al. 2006). For Russia, Eggers et al. (2006) demonstrate that high regional unemployment rates mitigate the individual well-being effects of unemployment and precarious employment and Grogan and Koka (2013) find that men face stigma in some non-market activities, suggesting that the effects of unemployment on SWB are more severe for men than for women. Based on research in Swiss cantons, moreover, Stutzer and Lalive (2004), show that the extent of canton-level support for unemployment benefits mediates the well-being gap between the unemployed and the employed. The unemployed who transition into retirement realize gains in life satisfaction precisely because they enter into a phase in life when working is unimportant (Hetschko et al. 2013).

In addition, scholars distinguish between hedonic and evaluative well-being (Graham 2012; Kahneman and Krueger 2006; Kahneman and Deaton 2010). Evaluative well-being survey items capture how people assess their lives as a whole, through general life satisfaction questions, or via the Cantril ladder question, which asks respondents to compare their life to the best possible life they can imagine, based on an eleven-point scale or ladder, where zero is their worst possible life and ten is their best possible life. Hedonic well-being, in contrast, is about people’s affective states and encompasses day-to-day positive and negative emotions related to work commutes, one’s immediate health state, job quality, and others (Graham 2012). Hedonic well-being metrics include both positive affect (e.g., smiling, experiencing happiness or joy) and negative affect (e.g., experiencing worry, sadness, anger, or stress).

Looking at both evaluative and hedonic well-being dimensions is particularly relevant in the case of employment and late-life work. For example, research shows that the unemployed have lower evaluative well-being levels (e.g., life satisfaction) than the employed but experience similar levels of positive and negative affect: while the jobless suffer consumption losses, these are partially offset by having more leisure time (Knabe et al. 2010). Furthermore, what a person expects to achieve in the future may be as important as current circumstances. Some people are in work arrangements because of what they want to achieve over their life course, rather than because they make their day-to-day living more pleasant. The balance between such objectives can change over the life course, moreover. While people may choose arrangements which enhance their daily living, at least temporarily, at the expense of career objectives, daily experiences can affect long-term objectives. Krueger et al. (2011) find, for instance, that the time spent looking for jobs is the saddest part of the day for the unemployed, and the longer those times last, the higher the likelihood that they cease the job search.

Building on the extant literature, we pose two research questions: (i) What is the relationship between employment status, work arrangements, and SWB (including job satisfaction)? and (ii) Is there an additional well-being effect for late-life workers? Our main contribution to the existing knowledge is the detailed exploration of the relationship between voluntary part-time work and different subjective well-being and job satisfaction dimensions, especially as related to late-life work. We find that voluntary part-time workers are happier, experience less stress and anger, and are more satisfied with their work than other workers. In addition, the propensity score matching results show that older workers who remain in the labor force under voluntary part-time or full-time arrangements have higher well-being than comparable retirees.

Different employment and retirement schemes may be better suited for people depending on their career objectives, innate well-being levels, and on where they are in the life cycle. Understanding how employment, retirement, and late-life work relate to well-being can contribute to ongoing public policy discussions in the U.S., Europe, and beyond. Given aging populations and the increasing role of work in retirement in several countries, understanding the well-being effects of flexible work arrangements such as late-life work may increase in relevance. Employment and late-life work can provide social contacts and interactions, personal growth, autonomy, and sense of purpose. The latter may be particularly relevant for older cohorts when the opportunities for active engagement decrease (Fisher, 1995).

The analysis sample comprises: France, Germany, Greece, Italy, Portugal, Spain, Sweden, Turkey, United States, and the United Kingdom. These countries represent a range of welfare state programs, tax administration systems, and cultural and institutional norms.

The data span 2009–2012 and are from the Gallup World Poll (GWP), which is an annual survey run by the Gallup organization in 160 countries, representing more than 98 percent of the world's adult population. GWP is probability-based and nationally representative (of populations aged 15 and over), polling about 1,000 individuals in each country, with samples ranging from about 500 in Puerto Rico to 4,000 in China, India, and Russia. The data are collected using telephone or in-person interviews using the same survey methodology across countries, making results comparable cross-sectionally and over time. The surveys were conducted via landline or cellular phone in France, Germany, Italy, Portugal, Spain, Sweden, United States, and United Kingdom. The data were collected via face-to-face interviews in Turkey and Greece. Because different respondents are polled each year, we have pooled cross-sections rather than a panel.

There is now a wide consensus that SWB metrics are valid, reliable, and psychometrically sound (Diener et al. 2012; Diener et al. 1999) and comparable across countries (Helliwell et al. 2010). Still, we recognize and attempt to correct for certain methodological issues. Research from the UK, for example, indicates that respondents are significantly more likely to report lower life satisfaction and happiness and higher anxiety in person than over the phone (Dolan and Kavetsos 2012). As the differences in interview mode are at the country level, country dummies in our regressions should capture this bias.⁷

We use a range of dependent variables to investigate the complex effects of employment and retirement on well-being. We measure evaluative well-being with the best possible life (BPL) question, and include measures of both positive and negative hedonic affect. For the first hedonic dimension, we use a binary indicator about experiencing happiness the day before, and for the second, we use separate variables about experiencing stress and anger the previous day, as the correlates for stress and anger often differ. Research across a range of disciplines finds that the stress associated with positive career advancement is not linked to negative health outcomes, but stress associated with circumstances beyond one’s control is associated with shorter life spans (Velasquez-Manoff 2013)⁸. Studies also suggest that the process of acquiring freedom and capabilities often correlates with higher levels of evaluative well-being and higher levels of stress at the same time (Graham and Nikolova 2013).

Similarly to evaluative and hedonic subjective well-being, we investigate whether job satisfaction has hedonic and evaluative aspects. The few extant studies on the link between meaningful work and job satisfaction (Erdogan et al. 2012) show that people value interesting work and autonomy (i.e., the process of work) not because they lead to higher incomes but because they are important procedural aspects of work (Benz and Frey 2008). General job satisfaction questions likely capture day-to-day aspects of one’s work life, while questions about whether one’s job is the ideal one for them are more about their career path as a whole and are more likely to be reflective of eudemonic aspects.

Our focal independent variables are the individual-level employment status indicators, which Gallup included as part of the GWP starting in 2009. The employment data are collected using identical questions and standardized calculations enabling the comparison of employment statistics across countries and over time. Based on a series of questions about employment and work, Gallup researchers classify individuals in one of six employment categories: employed full-time for an employer; employed full-time for self; employed part-time, do not want to work full-time (i.e., voluntarily employed part-time); employed part-time, want to work full-time (i.e., involuntarily employed part-time); unemployed; out of the work force. The dataset distinguishes between voluntary and involuntary part-time work thus allowing us to account for individual choice regarding working time.

Those who are employed full-time (either for an employer or for self) work for at least 30 hours a week, while part-time work is less than 30 hours a week. The unemployed are those who reported that they did not work in the past seven days but were actively looking for a job and were able to begin work. The “out of the work force” category includes respondents who are not employed, are not looking for work, and/or are not able to start work. Such respondents may be homemakers, retired, students, and disabled individuals.

We first used regression analysis to explore the relationship between employment status and SWB (including job satisfaction). We focused on whether and how different SWB dimensions relate to employment status and how this relationship varies by age. The dependent variables are: Best Possible Life (BPL), experienced happiness yesterday; job satisfaction; whether the respondent thinks her current job is her ideal job; experienced stress yesterday and experienced anger yesterday. All regressions are for 2009–2012 as there are no individual-level employment status data prior to 2009. When BPL is the dependent variable, we use an ordered logit specification and logits for the binary metrics.

where X_1itc — X_5itc are the employment status variables (with the reference category being the full-time employees), T′_itc is a vector of observed individual-level variables such as gender, age, marital status, income and others, Z′_itc is a vector of person-specific observed household-level variables such as household size, household location (i.e., rural or urban), and others; κ_c are country dummies, τ_t are time dummies (year of survey), and ϵ_itc is the stochastic error term.

Second, to study the nuances of the SWB-employment relationship across age groups, in separate regressions, we add interaction terms (of age cohorts and employment categories). We included the following age groups: under 45 years of age; ages 46–65; and age 66 and older. The reference category for all interactions is the young.

Third, we employ propensity score matching (PSM) to compare retired individuals with observably similar late-life workers, as a means to assess the effect of retirement on well-being. Self-selection and endogeneity are the two main methodological problems related to discerning the causal effects of retirement and employment status on well-being. First, self-selection relates to the fact that those who choose late-life work or work under flexible employment arrangements (e.g., voluntary part-time work) may be unobservably different from those who select into retirement or those who work under traditional employment arrangements. For example, late-life workers may have different skills and motivation or face different levels of family pressure than those who choose to retire. Second, the direction of causality between happiness and employment could run both ways: different SWB levels may influence decisions about retirement and employment and employment and retirement affect well-being. We use the PSM technique in an attempt to mitigate these issues. While it controls for selections on observables, is semi-parametric, and does not require assumptions for the outcome equation or the error term, PSM may not eliminate all bias; it demands sufficient data to pick the right pre-treatment covariates⁹; and using a small sample may yield incorrect estimates (Peikes et al. 2008). Given the unavailability of experimental or panel data, PSM is the best available alternative in our case.¹⁰

When deciding to retire or continue working, an individual i considers whether the expected utility of retirement (U^R) is greater than the expected utility of late-life work (U^L) and retires if U^R > U^L (Latif 2011). In addition, latent retirement behavior is given by R* = ηZ +ϵ, where R = 1 if R* > 0 and R = 0 if R*≤0, Z is a vector of individual characteristics determining retirement, and ϵ is the error term.

The estimator is based on the following assumptions:

While PSM is useful in teasing out the causal relationship between retirement and SWB, the results should be treated with caution, as the matching is as good as the available matching covariates. Similarly, the t-statistics reported in Table 10 are computed using assumptions such as normal distributions of the variables, among others, and are very sensitive to sample size.

The top panel in Table 10 shows unequivocal differences between the retired and their non-retired voluntary part-time counterparts (who are of similar age, gender, education, religiosity, marital status and live in the same country). While the latter had on average higher evaluative well-being than the retired, the difference is statistically insignificant (t-stat = 1.160). There is a statistically significant difference in favor of voluntary part-time workers for both happiness yesterday and smiling. There are no significant differences between the groups with respect to experiencing stress and anger the day before. Late-life workers are also more satisfied with their personal health and their freedom in life but the latter difference is statistically insignificant.

The second panel in Table 10 uses a different control group for the retirees –identical full-time workers. Based on these results, non-retired full-timers have significantly higher evaluative well-being (BPL) and positive hedonic well-being (smiling and happiness) than their retired counterparts. While most of the other differences are not statistically significant, the full-timers had higher health satisfaction. There are no statistically significant differences between the retirees and their non-retired self-employed counterparts (Table 10 – third panel), meanwhile. This finding could be due to the fact that both the retired and the self-employed older workers face similar freedoms and constraints related to free time and activities. In addition, older workers for whom part-time work is not available may be choosing self-employment in preparation for retirement and may therefore have similar well-being outcomes as their retired counterparts. Finally, the bottom panel shows that the involuntarily employed part-time non-retirees are more satisfied with their health than retirees. These results furnish empirical evidence that late-life work has positive well-being effects, especially in terms voluntary work or full-time work for an employer.

The control groups – i.e., full-time workers, part-time workers, and the self-employed— are proxies for an underlying latent variable – desired working time. The extant studies have assumed that because they prefer shorter working hours, older workers may retire earlier than planned if their employer does allow with such flexibility. Using longitudinal data from the UK, Bell and Rutherford (2013) show that while the overemployed older workers prefer to reduce their working hours, there is a substantial number of underemployed old works who would like to work longer hours. In addition, the self-employed older workers are much more likely to want to adjust their work hours and wages than the employed (Bell and Rutherford 2013). Our data allow us to complement this analysis by providing well-being comparisons between voluntary and involuntary part-time workers. In many countries, older workers face discrimination and involuntarily choose early retirement (Dorn and Sousa-Poza 2008). To supplement their income, they become “necessity” entrepreneurs and experience dissatisfaction (Block and Koellinger 2009). While our findings suggest that the retired fare worse than their voluntarily full-time and part-time counterparts along a number of well-being dimensions, there is no well-being premium for involuntary work and self-employment compared to retirement.

Building on research on the well-being effects of unemployment, late-life work, and job satisfaction, we explored the effects of different employment status and retirement arrangements on well-being. Our analysis covered a range of age cohorts and a number of European countries and the United States. While we confirm the well-known negative effects of unemployment on well-being, we have novel results on the different employment arrangements. The most important of these are the seeming well-being benefits to voluntary part-time employment as well as to remaining in the workforce beyond retirement age. Both sets of results were more robust in countries where part-time work is more likely to be the norm and/or where post retirement age work was likely to be voluntary.

Voluntary part-time workers had higher levels of evaluative and hedonic well-being, as well as lower levels of stress and anger than full-time workers. They were also more likely to report that they were satisfied with their jobs. Older workers who are self-employed, meanwhile, were more likely to report having an ideal job than were other cohorts. Furthermore, our propensity score matching estimation, which compared retired workers with observably similar late-life workers, found significant benefits for those who continued working. Older cohorts who remained in the labor force under full-time and voluntary part-time arrangements had higher levels of well-being and health satisfaction and lower levels of stress and anger than their retired counterparts.

Our results challenge the traditional model of full-time work and timely retirement, and suggest that flexible work arrangements (either full- or part-time) later in life may have significant well-being benefits. We hope that these results will inform the discussion about the potential benefits of flexible work and retirement arrangements, both in terms of fiscal burdens and the well-being of citizens. In addition to the administrative and implementation costs, changing employment and retirement schemes and allowing more labor market flexibility will likely have transactions costs, at least in the short run. For example, both employers and employees may face bargaining costs related to negotiating the flexible work arrangements which are optimal for both sides. Employers may moreover incur additional costs related to restructuring tasks to accommodate part-time work. Yet, the long-run fiscal benefits, as well as those to aggregate well-being – which our results suggest are quite large – make such policy changes worth considering in many countries. While this analysis is limited to Europe and the U.S., it may also have implications for other countries.

¹While there have been extensive labor market reforms in Europe in the past two decades, they have primarily focused on temporary and fixed-contract work (Dawson and Veliziotis 2013).

²If young and old workers are complements, however, extending the retirement age may in fact increase the demand for younger workers, their employment levels, and their wages.

³Specifically, our analyses using quantile regressions show that the well-being of the happiest individuals does not depend on employment. Quantile evaluative well-being regression results are available upon request.

⁴Wottiez and Theeuwes (1998) show that early retirees (ages 43–63) in Holland have higher life satisfaction levels than workers but normal-aged retirees may be unhappier than workers. In Denmark, retirement seems to have no effect on women’s well-being but reduces men’s well-being (Jaeger and Holm 2004). Latif (2011) finds a positive effect of retirement on hedonic happiness in Canada; using panel data from Germany, Bonsang and Klein (2012) find that while the effect of retirement on life satisfaction is negligible and income satisfaction decreases, satisfaction with free time increases. Studies document the negative well-being effects of involuntary retirement in Germany (Bonsang and Klein, 2012) and the U.S. (Bender 2012; Calvo et al. 2009). Calvo (2006) shows that late-life work is positive for well-being overall, but undesirable work (i.e., jobs that have excessive demands or cause dissatisfaction) impacts moods and mortality.

⁵Life satisfaction and morale among male retirees, for example, increase during the first years of retirement (Kim and Moen 2002; Gall et al. 1997) possibly because retirees take advantage of their free time to pursue their interests and hobbies. Continuous retirement, however, leads to an increase in depressive symptoms among males in the Untied States (Kim and Moen 2002).

⁶See Dawson and Veliziotis (2013) on the relationship between temporary employment, job security, and SWB in Britain.

⁷Research from the British Household Panel shows that a partner’s presence during the interview increased the likelihood of underreporting job satisfaction while the presence of children increased the likelihood of overreporting it for women (Conti and Pudney 2011). We do not have a variable for whether the partner or children were present at the time of the interview and could only imperfectly control for audience effects by including controls for household size, whether the respondent is de facto married, and has children in the household.

⁸This phenomenon was originally identified in research by Michael Marmot, a British epidemiologist, who found that high status civil servants lived much longer than low status civil servants; both worked in a very rigid hierarchy with little avenue for advancement.

⁹While PSM matches only on observables, it also matches on unobservables to the extent that they are correlated with the matching covariates (Stuart 2010).

¹⁰Blundell and Costa Dias (2000) Caliendo and Kopeinig (2008) Deheja and Wahba (2002), Rosenbaum and Rubin (1983), Smith and Todd (2005), Stuart (2010), and Todd (2006) furnish detailed overviews of the theoretical, practical, and methodological details pertaining to Propensity Score Matching (PSM).

¹¹We used Stata’s user-written -psmatch2- module to implement the matching (Leuven and Sianesi 2003).

¹²Summary statistics of employment status by age and country are available upon request.

¹³Detailed statistics for each well-being dimension by country are available upon request.

¹⁴Respondents in Sweden and the U.S. had the highest average BPL scores (7.4 and 7.3, respectively), while those in Portugal (5.2) and Turkey (5.5) had the lowest. The voluntarily employed part-time workers in the U.S. and Sweden had the highest average BPL scores, while participants in Portugal scored the lowest on this well-being dimension. Results available upon request.

¹⁵Overall, the UK, Germany, and the U.S. had the highest proportions of respondents reporting happiness, while Greece had the lowest. Among the group of voluntarily employed part-time workers, the U.S. and the UK had the highest proportions of respondents reporting having experienced happiness yesterday, while Greece had the lowest proportion. Results available upon request.

¹⁶Among the voluntarily employed part-time, the most stressed respondents were again in Turkey, while the least stressed were in Sweden. Results available upon request.

¹⁷The Swedes (93 percent) and Germans (91 percent) were the most satisfied, while the Greeks (78 percent) and the Turks (77 percent) had the lowest job satisfaction rates. The U.S., the UK, Spain, Sweden, and Germany were among the countries with the highest proportion of voluntary part-time workers who were satisfied with their jobs. Results available upon request.

¹⁸Respondents in Sweden, Portugal, and Germany most likely to report so, while those in Greece, Spain, and Turkey least likely to report so. Among the voluntarily part-time workers, those in the US and Sweden were most likely to report that their current job was the ideal one for them, while those in Italy and Greece were least likely to report so. Results available upon request.

¹⁹The BPL regression equations are estimated using both standard logistic regressions as well as using heterogeneous choice models, which control for across-group heterogeneity inherent in logistic models (Williams 2009). We only report the standard ordered logit results as the heterogeneous choice results are similar.

²⁰The job satisfaction regressions may suffer from selection bias as job satisfaction is observed only for those who work. We performed a Heckman probit correction (using the number of children in the household as the additional variable in the selection equation). The rho statistics were not statistically significant, suggesting that unobservables affecting labor market participation do not correlate with unobservables affecting job satisfaction. Likewise, the likelihood ratio test had a p-value of 0.430 for the job satisfaction regression and a p-value of 0.124 for the ideal job regression; we do not reject the null hypothesis that the outcome equation (job satisfaction) is uncorrelated with the selection equation. In other words, selection bias is not a problem in this instance, and we can be confident in our logit estimates.

²¹We checked whether the balancing property was satisfied using t-tests for the equality of means for the key covariates after matching (See Table 1 for BPL). We also calculated the mean standardized bias before and after matching (Table 2). The balancing tests for the rest of the dependent variables are available upon request. In all cases, the t-tests are statistically insignificant and the mean bias is at most 8.5, suggesting that the matching was successful.