Don’t worry, be happy? Happiness and reemployment

Journal of Economic Behavior & Organization 96 (2013) 1–20

Contents lists available at ScienceDirect

Journal of Economic Behavior & Organization journal homepage: www.elsevier.com/locate/jebo

Don’t worry, be happy? Happiness and reemployment夽 Annabelle Krause ∗ Institute for the Study of Labor (IZA), Schaumburg-Lippe-Strasse 5-9, 53113 Bonn, Germany

a r t i c l e

i n f o

Article history: Received 4 January 2013 Received in revised form 28 August 2013 Accepted 8 September 2013 Available online 20 September 2013 JEL classification: J60 J64 I31 Keywords: Unemployment Reemployment Happiness Job search

a b s t r a c t This study investigates the effect of unemployed individuals’ happiness on their future labor market outcomes. It therefore acknowledges the possibility that happiness could also be a driver of behavior and influence life’s outcomes. I use rich survey data from 2007 to 2009 of entrants into unemployment in Germany (the IZA Evaluation Dataset S) to calculate residual happiness, which displays higher (or lower) satisfaction levels than would be predicted by a number of demographic and socioeconomic characteristics. I find a statistically significant inverted U-shaped effect of residual happiness on an unemployed individual’s future reemployment probability and reentry wage, even after controlling for demographic and socioeconomic characteristics, labor market histories and future job prospects. Further investigation offers three mechanisms that have not been previously shown in this context: (a) happiness is mainly a predictor for exit into self-employment rather than regular employment; (b) only male unemployed experience an effect of happiness on reemployment; and (c) the concept of locus of control and the personality traits of neuroticism and extraversion are main drivers of the baseline effect on regular reemployment and are able to explain the effect on reemployment for males. © 2013 Elsevier B.V. All rights reserved.

1. Introduction “Well-being is important because there appears to be an increasing gap between the information contained in aggregate GDP data and what counts for common people’s well-being” (Stiglitz et al., 2009, p. 12). This growing attention of policymakers highlighted by the Stiglitz–Sen–Fitoussi Commission follows the surge of academic interest in happiness research after the pioneering work of Easterlin (1974).1 Yet, to date economic research has primarily focused on subjective well-being as an outcome variable (see Frey and Stutzer, 2002, for an overview) where due to its important social implications the effect of unemployment has received particular attention, with a broad consensus that unemployment leads to a decline in happiness (e.g., Clark and Oswald, 1994; Winkelmann and Winkelmann, 1995, 1998; Gerlach and Stephan, 1996; Korpi,

夽 I would like to thank Costanza Biavaschi, Marco Caliendo, Alfonso Flores-Lagunes, Anne Gielen, Carol Graham, Daniel S. Hamermesh, Peter J. Kuhn, Simon Lüchinger, Andrew Oswald, Ulf Rinne, Michael Rosholm, Simone Schüller, Klaus F. Zimmermann, two anonymous referees, the editors and participants at the IZA Brown Bag Seminar, the 4th CIER/IZA Annual Workshop on Research in Labor Economics, the 2nd Potsdam PhD Workshop in Empirical Economics, the 18th Annual Society of Labor Economists (SOLE) Meetings in Boston and the HEIRs Conference Public Happiness in Rome for helpful discussion and comments. This study uses the IZA Evaluation Dataset S, which was created by IZA with financial support of the Deutsche Post Foundation. The IZA Evaluation Dataset S consists of survey information on individuals who entered unemployment in Germany between June 2007 and May 2008 (see Caliendo et al., 2011a). All remaining errors are my own. ∗ Tel.: +49 228 3894 527. E-mail address: [email protected] I use the terms happiness, subjective well-being and life satisfaction interchangeably in this paper, as with most economists, see, e.g., Graham et al. (2004). 1

0167-2681/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jebo.2013.09.002

2

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

1997; Clark et al., 2001; Di Tella et al., 2001; Böckerman and Ilmakunnas, 2006; Kassenboehmer and Haisken-DeNew, 2009; Winkelmann, 2009; Ohtake, 2012).2 This paper contributes to the literature on unemployment and happiness by adopting a different perspective: it acknowledges the possibility that happiness could also be a driver – and not only a response – of behavior and hence might influence life’s outcomes. In fact, while there is no doubt that people do certain things to remain happy or become happier, it is probably as likely that happier people also behave differently because they have different well-being levels. Adding to the literature on unemployment and happiness, the paper investigates the following questions: does individual happiness influence an unemployed individual’s future reemployment probability? Does individual happiness influence an unemployed individual’s future reentry wage? The interest in looking at the relationship between happiness and post-unemployment outcomes is threefold. First, unemployment constantly represents an important topic in terms of academic research and public policy, given that it leads to high psychological distress, it reduces general output, and paternal unemployment exerts a negative effect on children’s schooling efforts (Andersen, 2013). Second and related to the previous point, for an effective policy design it is important to understand the factors facilitating reemployment and whether the unemployment–happiness relationship is exclusively a one-way street. Third, given the lack of adaptation in life satisfaction with respect to unemployment compared to other life events (Clark et al., 2008), the relationship between happiness and exit from unemployment appears to be of particular importance. The paper provides a deeper understanding of the impact of life satisfaction and new insights concerning the determinants of reemployment and reentry wages. To carry the analysis, I use a rather unique and unexplored survey dataset that provides rich information on entrants into unemployment in Germany (the IZA Evaluation Dataset S). I focus on outcomes one year after unemployment entry, which is the standard definition of long-term unemployment and hence, a time-frame of particular interest. In fact, it is important that individuals avoid passing into long-term unemployment for several reasons at the societal and individual level. On the one hand, high long-term unemployment will lead to increasing inequality and higher aggregate unemployment within the whole economy (Machin and Manning, 1999). On the other hand, evidence suggests that individuals suffer from long-term unemployment with respect to their labor market opportunities and physical and mental well-being (Machin and Manning, 1999), and individuals who have been longer unemployed are less likely to find a job (Shimer, 2008). Moreover, and given that I use data on Germany, 12 months is the maximum period during which prime-aged unemployed individuals are entitled to unemployment benefits receipt in Germany.3 Given the panel structure of the dataset, I am able to observe individual happiness and the outcome variables for the same individuals 12 months apart, hence at the border of entering long-term unemployment. Moreover, the use of the IZA Evaluation Dataset S helps reducing endogeneity bias. A problem of endogeneity can arise due to omitted variable bias and reverse causality. If an unobserved variable influences life satisfaction and future employment probabilities, such as the knowledge about a future job, one would falsely interpret an effect from life satisfaction as being causal, despite the other factor actually determining the pattern in the relationship. For a clear causal effect of happiness, one would need a random assignment or experimental data. Given the impossibility to randomly allocate happiness, the IZA Evaluation Dataset S provides an alternative with rich observational data that are particularly suitable for studying this topic. It contains a large number of unemployed compared to surveys of the whole population and moreover, information about labor market histories, search behavior and other variables such as the subjective probability of finding a job. The latter helps to account for a form of self-esteem or own assessment of employability and consequently serves to reduce the reverse causality issue. Related to that, I am able to exclude those individuals who have not been looking for a job, and importantly those who have been looking and report to have already found a job, and thus the sample will only comprise actual job seekers. This also reduces reverse causality bias by minimizing the issue arising from the knowledge about a future job which leads to an increase in happiness and future reemployment probability, but is usually unobserved by the researcher. In addition, the data provide information about personality traits such as the locus of control, neuroticism and extraversion. Being able to account for personality traits is important as De Neve and Oswald (2012) show that they can be mediators of effects from happiness on income. A further advantage is that the respondents all have been unemployed for the same amount of time, around two months on average, and thus different unemployment durations do not influence their happiness levels and a discouraged worker effect should be small or non-existent. Methodologically, the empirical strategy is based on using “residual happiness” rather than absolute happiness as an explanatory variable, much in the spirit of Graham et al. (2004) to capture a sort of constant factor of happiness. The idea is to investigate whether people who had higher (positive residual) or lower (negative residual) happiness levels than their socioeconomic and demographic characteristics would predict having different labor market outcomes one year later. This residual element of happiness is interpreted as some sort of underlying inner disposition or cognitive bias (e.g., Cummins and Nistico, 2002), and therefore may capture psychological differences between the respondents (and some random noise).

2 A study using time use data finds that while the unemployed feel less happy while performing an activity, they compensate this decrease in well-being by the amount of time the employed do not have and therefore, average experienced utility does not differ between the unemployed and employed (Knabe et al., 2010). This finding extends the literature to more detailed evidence on the effect of unemployment on happiness. 3 These rules vary by age in connection with former employment duration. After these 12 months, unemployed individuals are entitled to a form of social insurance.

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

3

For this purpose, I estimate a life satisfaction regression in a first step to predict the residuals afterwards. In a second step, these residuals serve as explanatory variables in the main regressions predicting reemployment and hourly wages. The empirical findings show a statistically significant inverted U-shaped effect of residual happiness on an unemployed individual’s future reemployment probability and reentry wage, where the effect on reentry wages even seems to have a cubic shape. Further investigation shows that the reemployment result is mainly driven by males and exits into selfemployment, which is a remarkable and novel pattern. The main effect on reemployment and the gender difference can be explained by the concept of locus of control and the personality traits of neuroticism and extraversion. The non-linear effects on wages and self-employment are robust even after controlling for personality traits. Finally, the main results are also robust to using an exclusion restriction in the two-step procedure, as well as to selection bias due to attrition. The remainder of this paper is organized as follows. Section 2 introduces some theoretical considerations and discusses possible channels of the relationship between happiness and labor outcomes. Section 3 describes the data and sample. Section 4 provides the results of the empirical analysis, and Section 5 concludes. 2. Happiness as a driver of job search and labor outcomes There are still only a few papers that use happiness as a determinant rather than an outcome (see, e.g., Kenny, 1999; Graham et al., 2004; Guven, 2011, 2012; Goudie et al., 2012; De Neve and Oswald, 2012). Using residual happiness, Guven (2011) finds an inverted U-shaped effect of residual happiness on social capital, and Graham et al. (2004) find that individuals with higher residual happiness initially make more money and are in better health five years later. Psychologists and economists have also considered positive affect as an explanatory variable, with their findings suggesting that it engenders success (for a detailed overview, see, e.g., Lyubomirsky et al., 2005 and increases the present value of a future payment (Ifcher and Zarghamee, 2011). Studies connecting happiness, job search and labor outcomes include Clark et al. (2008), finding that future unemployment reduces current well-being, which can be interpreted as a lead or anticipation effect. Clark (2003) and Mavridis (2010) find that those with a higher drop in mental well-being when becoming unemployed are less likely to remain unemployed one year later and to have a shorter unemployment duration, whereas Gielen and van Ours (2012) find that this drop in life satisfaction does not stimulate job finding. Marks and Fleming (1999) find that those with higher lagged subjective wellbeing levels are more likely to remain employed and to be reemployed. Psychologists find that high trait positive affect leads to greater success at obtaining follow-up job interviews (Burger and Caldwell, 2000), and that higher well-being at the age of 18 predicts higher levels of occupational attainment (Roberts et al., 2003). Overall, the findings in the related literature suggest that higher happiness leads to “better” outcomes. In theory, the standard model of job search (McCall, 1970; Mortensen, 1970) suggests that an individual’s reemployment probability depends on both the probability of receiving a job offer and accepting it, usually displayed by the individual’s reservation wage. Determinants of the reservation wage are the expected wage distribution, possible search costs, the job offer arrival rate and unemployment benefits (or more generally, gains during jobless periods). Factors determining the job offer arrival rate include the general state of the labor market, the individual’s job search effort (if effort is endogenized), education and experience. How would an individual’s well-being enter this model? Hermalin and Isen (2008) incorporate current emotional state into an economic modeling and decision-making framework, based on the idea that a dynamic recurring relationship between affect at the beginning of a period influences preferences that determine decisions or behavior, which in turn determine affect at the end of a period. With respect to reemployment, their theoretical framework suggests that employers prefer workers with initial happiness levels greater than some cutoff value, given that their work effort should be higher.4 If the happiness level is not high enough, the employers try to induce it, e.g., by offering the employee a signing bonus and thereby boosting the state of affect. In terms of the search model, the job offer arrival rate would therefore increase with happiness, since a happier worker is assumed to be more valuable for the employer through assumed higher productivity and possibly better teamwork abilities. This would reflect a direct channel from happiness to employment, displaying a sort of unobserved characteristic for the hiring probability besides qualification, experience and possibly other factors. Besides this direct impact from the employer’s side, there are several indirect channels from the worker’s side through which happiness can affect reemployment, with the most obvious probably being job search effort. However, the direction of this effect is theoretically ambiguous: on the one hand, a very unhappy individual may suffer intensely from unemployment and tries hard to get out of it. This increases the job search intensity and/or decreases the reservation wage, both of which would lead to a higher reemployment probability. On the other hand, higher subjective well-being may display more resilience and higher motivation to search.5 In this case, higher happiness would increase the prospective employment probability through higher job search effort. Other channels include health and social contacts, which are both positively related to happiness and reemployment (see, e.g., Verkley and Stolk, 1989).

4 There is experimental evidence showing that positive affect can increase intrinsic motivation (e.g., Isen and Reeve, 2005). See also Oswald et al. (2009) for an experiment with respect to happiness and productivity. 5 As Lynch (1989) points out for the empirical analysis of reemployment probabilities of young unemployed, motivation is an unobserved and omitted factor which might bias the estimates.

4

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

A second outcome in the empirical analysis relates to the reemployed individual’s wage. What would the association be between happiness and future wages? It appears similar to that discussed for the reemployment probability, namely that employers may see higher potential or prospective productivity in happier job candidates, which would result in higher wage offers. From the workers’ perspective, happier candidates may exert greater bargaining power or abilities through higher selfesteem or knowledge about their higher productivity, and likewise reentry wages would increase with happiness. Therefore, theory predicts that the happier the unemployed individual, the higher their wage when reentering the labor market. Given that the data I use for the empirical analysis do not include information about job offer probabilities or the demand side of the market, I only investigate the supply side. The main hypotheses tested can be summarized as follows: (a) the happiness of unemployed individuals is either negatively or positively related to job search activities; (b) the happiness of unemployed individuals exhibits either a positive, negative or non-linear effect on future reemployment; and (c) the happiness of unemployed individuals exhibits a positive effect on future hourly wages. 3. Data and sample I use data from the IZA Evaluation Dataset S (Caliendo et al., 2011a), which is a survey of almost 18,000 individuals who entered unemployment in Germany between June 2007 and May 2008. One cohort of respondents was interviewed each month. In total, there are 12 cohorts, with each including around 1500 respondents. Individuals are interviewed shortly after entering unemployment, on average two months after unemployment entry. The data are a panel including three waves, and therefore individuals are followed over time. The second wave took place one year after the first unemployment entry (June 2008–May 2009), and the third wave took place three years after the first unemployment entry (June 2010–May 2011). The analysis is based on the first two waves. One advantage of the data lies in their specific focus on entrants into unemployment. Therefore, the IZA Evaluation Dataset S is highly appropriate for studying the processes of job search and labor market reintegration. Similar household surveys are generally designed to be representative of the whole population (e.g., the German Socio-Economic Panel Study, SOEP), which has an important drawback when studying unemployed individuals given that sample sizes decrease substantially. The data address a large variety of topics such as the individual’s detailed search behavior (number of applications, search channels, reservation wages, etc.), ethnic and social networks, psychological factors and life satisfaction. The exact wording of the life satisfaction question is “How satisfied are you with your life as a whole these days?” and is measured on a scale of 0–10, where 10 represents maximum satisfaction. Existing literature has shown self-reported life satisfaction to be a valid and consistent measure of subjective well-being. Self-reports and other measures such as interview ratings, peer reports and the average daily ratio of pleasant to unpleasant moods show a strong convergence (e.g., Diener and Lucas, 2000). Other objective validity has been shown through, e.g., brain-science data (Urry et al., 2004) and compensating-differentials quality of life measures (Oswald and Wu, 2010). Moreover, Lepper (1998) shows that subjective well-being measures are fairly stable over time, and are not substantially influenced by mood states or interview circumstances. The sample is selected with respect to the following characteristics. All individuals in the first wave must still be unemployed, thus I exclude those who are already reemployed at the time of the first interview. Given that the interview takes place on average around two months after unemployment entry, around 25% of the individuals in the first wave have already exited unemployment. Furthermore, respondents who claim not to have searched for a job since unemployment entry are also excluded; indeed, most of them had already found a job. Moreover, I exclude those individuals who searched for a job but claimed at the time of the interview to certainly have a prospective job. I thereby minimize the potential bias arising from reverse causality driven by knowledge about future job prospects, which causes individual happiness and future reemployment probability to increase simultaneously. The selected sample is a balanced panel of the first and second wave, and after excluding observations with missing information, I am left with a sample of 2542 individuals per wave. Table 1 displays summary statistics of the main variables. The information stems from the first interview, apart from the employment status, hourly wage and information about life satisfaction by employment status, which are derived from the second wave. The mean of the life satisfaction of the newly unemployed is 6.1 in the first wave, which is slightly higher than results from other studies using SOEP data, where the life satisfaction of the unemployed lies rather below 6 (e.g., Winkelmann and Winkelmann, 1998; Gielen and van Ours, 2012). This could be due to the early interview timing with respect to unemployment entry for all respondents. Considering the evolution of life satisfaction after one year, it confirms findings in the literature that individuals’ life satisfaction increases when they are reemployed, in this case on average by one point to around 7. Individuals who are unemployed in the second wave suffer more than in the first wave, with their life satisfaction decreasing to around 5.5, confirming evidence that there is no adaptation to unemployment. Almost 60% of the sample are employed one year after unemployment entry, reporting an hourly wage of 8.30 Euros. The average age is 38 years, and slightly fewer than half of the sample are men. Around 17% of the sample are either first or second generation migrants, and around 30% live in East Germany. 51% are married, most respondents have an intermediate school degree and vocational degree and every fifth respondent has a degree from a technical college or university.6 The average last hourly

6 There are three types of secondary schools in the German school system: (a) a lower secondary school (Hauptschule), which is designed to prepare pupils for manual professions, (b) an intermediate secondary school (Realschule), which prepares students for administrative and lower white-collar jobs, and (c) an upper secondary school (Gymnasium), which allows for access to higher education and universities.

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

5

Table 1 Descriptive statistics of main variables. Variable Life satisfaction (wave 1) Life satisfaction of the employed (wave 2) Life satisfaction of the unemployed (wave 2) Employed (wave 2) Hourly wage (wave 2) (Euros) – if employed Age Male Native 1st generation migrant 2nd generation migrant Eastern Germany Married No formal educational degree Secondary school (9 yrs) Secondary school (10 yrs) Technical college entrance qualification (11–12 yrs) General qualification for university entrance (12–13 yrs) No formal vocational degree Apprenticeship (dual system) Specialized vocational school University, technical college Net hourly wage of last job (Euros) Duration of last job (in months) Number of applications sent Number of search channels used Search for full-time job Reason for termination of previous job: Layoff Quit End of temporary contract Employer and employee agreed on termination of contract Firm closure End of self-employment Parental leave Care for person in need Other reason # of observations

Mean 6.143 7.078 5.486 0.587 8.294 38.245 0.467 0.827 0.092 0.081 0.288 0.507 0.010 0.277 0.421 0.058 0.235 0.086 0.591 0.140 0.183 7.486 52.601 15.411 5.273 0.642 0.439 0.107 0.219 0.082 0.074 0.013 0.018 0.001 0.047

Std. dev. (2.126) (1.775) (2.383) (0.492) (8.288) (9.868) (0.499) (0.378) (0.289) (0.274) (0.453) (0.500) (0.099) (0.447) (0.494) (0.233) (0.424) (0.280) (0.492) (0.347) (0.387) (3.977) (77.847) (19.268) (1.614) (0.480) (0.496) (0.309) (0.413) (0.275) (0.262) (0.115) (0.132) (0.028) (0.212)

2542

Source: IZA Evaluation Dataset S, own calculations. Notes: All variables display characteristics from wave 1 if not indicated otherwise. Differing number of observations: life satisfaction of the employed (wave 2): 1493; life satisfaction of the unemployed (wave 2): 780; hourly wage (wave 2) (Euros) – if employed: 1385.

wage was 7.50 Euros, and the average duration of the last job prior to unemployment entry was 52.6 months. On average, the individuals have written 15 applications since unemployment entry and use around five search channels (out of ten possibilities, including other search channels). 64% of the sample only look for a full-time position, as opposed to a part-time position or both types simultaneously. The most common reason for terminating the last job was layoff, accounting for around 44% of the sample. Two other prevalent reasons were quitting and the end of a temporary contract. 4. Empirical analysis 4.1. Residual happiness In order to calculate residual or unexplained happiness, I first estimate an OLS life satisfaction regression with several independent variables from the first wave.7 After this estimation, I predict a residual i for each individual i. The residuals present a measure for unexplained happiness laying above or below what would be expected by the observable individual characteristics. This variable may be interpreted as a proxy for inner individual disposition or cognitive bias, but also contains some noise. The distinction between positive and negative residuals is particularly interesting in this regard, since a positive residual can be regarded as a positive cognitive bias and a negative residual as a negative cognitive bias or a higher degree

7 Results from an ordered probit estimation are similar. Economists are more likely than psychologists to be worried about satisfaction scores only being ordinally meaningful. However, ordinal and cardinal estimations of life satisfaction usually generate very similar results (Ferrer-i-Carbonell and Frijters, 2004; Frey and Stutzer, 2000a).

6

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

of pessimism.8 If, for example, a respondent rated his or her life satisfaction as 7, but the prediction of this individual’s life satisfaction based on the estimation with this sample is 5, this respondent would have a positive residual. In a second step, these residuals serve as explanatory variables in the main regressions predicting reemployment and hourly wages. If panel data are available, it is nowadays standard in the literature to use fixed effects models for happiness estimations in order to avoid biases arising from unobserved time-invariant factors that determine both the independent variables and happiness. One could estimate a standard fixed effects model and include the fixed component and idiosyncratic error component in the measure of residual happiness, and also compare results only using one of these components. While I have panel data for my analysis, I estimate a cross-sectional model for the first wave and not a fixed effects model, since I only use two waves in my analysis and am interested in how residual happiness is related with future outcomes (see also Guven, 2011; Graham et al., 2004). Moreover, as shown by Guven (2011), there could be a problem due to the serial correlation of residuals in panel models. Nevertheless, it would be important for future research to investigate whether results largely differ between the cross-sectional and fixed effects residual happiness approach.9 The life satisfaction equation appears as follows: Wi = ˇ0 + Xi ˇ1 + i ,

(1)

where Wi is individual life satisfaction, Xi display individual, household and regional characteristics, and i are the residuals. I include demographic and socioeconomic control variables such as gender, age, education, marital status, children, and migration status in the estimation, as well as wage and duration of the last job, the amount of unemployment benefits received by the individual, and the employment status of the spouse or partner. Moreover, I control for the federal state’s unemployment rate and the reason for the last job’s termination. Geographical dummies for German federal states, interview cohorts and the amount of time between unemployment entry and interview display additional control variables. Table 2 shows the results of the life satisfaction regression, which are generally similar to the results with representative samples of the society or working population. In this case, the sample consists only of unemployed individuals, with one advantage that they have all been unemployed for a similar amount of time, which is usually not the case in other datasets. Men are significantly less happy, and happiness is U-shaped with age. Having a disability, being married to a spouse without a full or part-time job, or being single all have a statistically significant negative effect on life satisfaction.10 Having a spouse with a full-time position is associated with higher happiness. Second generation migrants are significantly less happy than natives, and the past hourly wage positively affects happiness. Compared to having had a job for under a year, having had a job for up to ten years or longer has a significant positive effect on the happiness of newly unemployed individuals. The reason for the end of the last job does not play an important role in this estimation. Fig. 1 plots the relationship between the residuals of the aforementioned regression and the employment probability in the second wave, suggesting a non-linear connection. For the most part, it increases until a certain point, when it experiences a sharp decrease at very positive residuals. The lowest reemployment probability is found for individuals with the highest unexplained happiness. Essentially, the graph suggests that both individuals who are very unhappy or very happy have a lower reemployment probability than individuals in between, pointing to an inversely U-shaped relationship. One possible explanation is lack of motivation, either because the person is depressed and the driving force is missing or the person is so happy with the situation that there is no motivation to change it. In this respect, job search effort could reflect a very important channel. Table 3 shows the distribution of the means of various job search variables, comparing individuals with positive and negative residuals. Moreover, I conduct a t-test between the two subsamples. It becomes apparent that individuals with higher residual happiness are significantly more likely on average to be employed one year later, reflecting the largely increasing relationship between the residuals and reemployment, yet neglecting the sharp decrease at the highest values. However, the difference is only significant at the 10% level. The other numbers in the table suggest that happier individuals exert less job search effort; for instance, writing significantly fewer applications and using significantly fewer search channels. In terms of the use of specific search channels, there is no significant difference between most of them, apart from searching via the job information system of the employment agency and sending out speculative applications, with happier individuals less likely to use both such channels. However, happier individuals appear to be less likely to search for a full-time position, which could be one reason why they are searching less, given that the pressure may be lower. The first hypothesis stated in Section 2 can be confirmed insofar that there seems to be a difference in search behaviors between happier and less happy individuals. The numbers show that the relationship is negative, namely less happy individuals search more. However, as seen in Fig. 1, this behavior does not seem to lead to a linear relationship with reemployment probabilities.

8 Residuals are also used in, e.g., the wage literature using wage residuals as proxies for match quality or the worker’s surplus (see, e.g.,Gielen and van Ours, 2010). 9 I will come back to this issue in Section 4.3 when I present results including personality traits in the life satisfaction estimation. This should reduce the bias arising from not estimating fixed effects regressions, since personality is assumed to account for a substantial part of individual unobserved heterogeneity (for a discussion of this topic, see Boyce, 2010). 10 The negative coefficient on the married dummy may seem counterintuitive at first, but is due to the other control variables of the spouse’s employment status and the single dummy.

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

7

Table 2 Life satisfaction estimation first wave. Life satisfaction in wave 1 Male Age Age squared Disabled Marital status (Reference: divorced/widowed) Married Single Partner (Reference: no partner) Employment status spouse (Reference: not full-time/part-time employed) Full-time employed Part-time employed Employment status partner (Reference: not full-time/part-time employed) Full-time employed Part-time employed Children in household Number of children in household Educational degree (Reference: No Degree) Secondary school (9 yrs) Secondary school (10 yrs) Technical college entrance qualification (11–12 yrs) General qualification for university entrance (12–13 yrs) Vocational degree (Reference: no degree) Apprenticeship (dual system) Specialized vocational school University, technical college Migrant status (Reference: native) 1st generation migrant 2nd generation migrant Log net hourly wage of last job (Euros) Duration of last job (Reference: until 1 yr) 1–5 years 5–10 years More than 10 years 0 month Log of unemployment benefits State unemployment rate Termination of previous job (Reference: temporary contract) Quit Layoff Employer and employee agreed Firm closure End of self-employment Parental leave Care for person in need Other Constant # of observations R2

−0.222 −0.096 0.100 −0.318

(0.091)** (0.038)** (0.050)** (0.193)*

−0.546 −0.544 −0.264

(0.165)*** (0.138)*** (0.218)

0.898 0.322

(0.143)*** (0.253)

−0.126 0.647 0.221 0.111

(0.251) (0.417) (0.171) (0.096)

0.601 0.705 0.525 0.792

(0.494) (0.490) (0.512) (0.498)

0.087 −0.004 −0.143

(0.177) (0.202) (0.207)

0.027 −0.343 0.486

(0.151) (0.168)** (0.109)***

0.145 0.374 0.307 −0.451 0.015 −0.082

(0.098) (0.138)*** (0.146)** (0.463) (0.016) (0.209)

0.088 −0.103 0.198 0.075 −0.224 0.280 −2.617 −0.200 7.429 2542 0.115

(0.159) (0.111) (0.176) (0.179) (0.385) (0.242) (1.736) (0.210) (1.947)***

Sources: IZA Evaluation Dataset S, own calculations. State unemployment rates from the federal unemployment agency. Notes: OLS regressions. Robust standard errors in parentheses. Further control variables include dummies for German federal states, interview cohorts, time between unemployment entry and interview. * Significant at 10%. ** Significant at 5%. *** Significant at 1%.

4.2. Main results 4.2.1. Reemployment and hourly wage The second step in the empirical analysis is to investigate whether residual happiness has any additional effects on the reemployment probability and hourly wages, after controlling for usual determinants. Due to the usage of a generated regressor, all standard errors in these estimations are corrected along the lines of Murphy and Topel (1985). Table 4 shows the main results when adding residual happiness as a regressor along with several other control variables as well as search behavior. Columns (1)–(6) refer to the results regarding reemployment. To detect any non-linearities, I include squared terms and quintile dummies besides the full values of the residual variable. Column (1) presents linear effects, indicating a positive and significant effect of increasing residual happiness on the future reemployment probability. While the effect is only significant at the 10% value, the p-value is 0.058, and thus it is very close to the sensitive value of 0.05. The residual

8

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

Fig. 1. Residual happiness and future reemployment probability. Note: Based on results from a locally weighted regression. Source: IZA Evaluation Dataset S, own calculations.

is subsequently divided into negative (and positive) residuals by setting the positive (or negative) residuals to zero. The negative residual variable displays the absolute values rather than the negative numbers, which means that a negative sign denotes a positive effect of an increasing residual on the reemployment probability. The positive linear effect of residual happiness is driven by individuals who are less happy than would have been predicted. No significant positive effect of positive residual happiness alone can be detected. Interestingly, when adding a squared term of the positive and negative residual, the inversely U-shaped effect becomes apparent for the positive residual fraction, with a residual of 1.6 representing the turning point. This quadratic effect is not driven by outliers, since there are more than 500 observations involving a residual of 1.6 and above. This means that having a residual 1.6 points higher than predicted by a number of variables maximizes an individual’s reemployment probability. There is no non-linear effect for individuals with negative residuals. Table 3 Job search effort.

Employed in second wave Number of applications sent Number of search channels used Search for full-time job Search channel used: Newspaper advertisement Advertisement posted Job information system Informal search (friends and relatives) Agent of employment agency Internet Private agent with agency voucher Private agent without agency voucher Speculative application Other search channel # of observations

Negative residual

Positive residual

p-Value of t-test

0.569 (0.495) 17.177 (22.869) 5.356 (1.615) 0.673 (0.469)

0.603 (0.490) 13.934 (15.485) 5.203 (1.610) 0.616 (0.486)

0.087

0.877 (0.329) 0.147 (0.354) 0.670 (0.470) 0.862 (0.345) 0.732 (0.443) 0.894 (0.308) 0.096 (0.295) 0.179 (0.383) 0.698 (0.459) 0.205 (0.404)

0.872 (0.334) 0.131 (0.337) 0.634 (0.482) 0.844 (0.363) 0.720 (0.449) 0.889 (0.314) 0.094 (0.292) 0.161 (0.368) 0.657 (0.475) 0.202 (0.401)

0.739

1158

1384

Source: IZA Evaluation Dataset S, own calculations. Note: All variables display characteristics from wave 1 except being employed at wave 2.

0.000 0.017 0.003

0.244 0.055 0.206 0.477 0.727 0.869 0.238 0.028 0.848

Table 4 Main results. Employed at the second interview (1)

i

(2)

(3)

Logarithm of hourly wage at the second interview (4)

(5)

(6)

0.041 (0.021)*

i if i > 0

(7)

(8)

(9)

(10)

i if i > 0

−0.024 (0.008)*** 0.248 (0.110)** −0.078 (0.035)**

2i if i > 0 |i | if i < 0

  2i  if i < 0

0.067 (0.027)** −0.018 (0.009)** −0.153 (0.087)*

−0.081 (0.021)***

0.017

0.014 (0.004)***

(0.020) 1st quintile of i Dummy 2nd quintile of i Dummy 3rd quintile of i Dummy 4th quintile of i 5th quintile of i Dummy

−0.315 (0.135)** −0.168 (0.135) 0.079 (0.137) Reference −0.092 (0.136) 2542 −1621.621

2542 −1618.417

2542 −1619.008

2542 −1618.046

2542 −1616.605

−0.122 (0.035)*** −0.116 (0.034)*** −0.024 (0.033) Reference −0.070 (0.034)** 1383 0.324

1383 0.322

1383 0.325

1383 0.324

1383 0.328

1383 0.331

Sources: IZA Evaluation Dataset S, own calculations. State unemployment rates from the federal employment agency. Notes: Logit ((1)–(6)) and OLS ((7)–(12)) regressions. Parameter estimates are shown. Murphy and Topel (1985) standard errors in parentheses. i = residual happiness. Positive (negative) residual happiness contains the residual values while setting the negative (positive) values to zero. Further control variables are dummies for German federal states, interview cohorts, time between unemployment entry and interview, state unemployment rate wave 1 and wave 2, reason for termination of previous job, male, migrant status, age and age squared, marital status, disability, (number of) children in household, employment status of spouse/partner, duration and logarithm of hourly wage of last employment, logarithm of unemployment benefits, educational and vocational degrees, search variables of wave 1 (number of search channels and applications, search for full-time or part-time job). * Significant at 10%. ** Significant at 5%. *** Significant at 1%.

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

−0.084 (0.033)**

2542 −1619.874

(12)

0.019 (0.010)*

0.021 (0.042)

|i | if i < 0

# of observations Log likelihood/R2

(11)

0.015 (0.005)***

9

10

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

Finally, four dummies for quintiles of the happiness distribution are added with the fourth quintile being the reference group. Again, this demonstrates a non-linear effect by all dummies having a negative sign. In this case, statistical significance is only given at the lower spectrum of the distribution. In summary, these results suggest a positive significant effect of residual life satisfaction, particularly at the lower part of the distribution, whereas the linear effect turns non-linear inversely Ushaped in the upper part of the distribution. The effect at the top of the residual distribution may typify individuals who are voluntarily unemployed or did not try to change their life situation, since they were already very satisfied with it, whereas the individuals at the bottom of the distribution may be too depressed to change their situation. I can confirm the hypothesis that happiness is related to future reemployment, and more so the assumed non-linear shape. Columns (7)–(12) in Table 4 refer to the results regarding the smaller sample of individuals who found a job in the second wave, with the dependent variable being their logarithmic hourly wage at the new job. Columns (7)–(9) highlight a statistically significant positive effect of residual happiness on future hourly wages. However, as can be seen in columns (10) and (11), the effect is not linear, as the squared terms of positive and negative residual happiness are negative. Therefore, similar to the probability of reemployment, the highest values of positive residual happiness lead to lower hourly wages. The turning point at the positive residual values is similar to before, at a residual value of 1.8. The turning point at the negative residual values is at −2.9, with only around 100 individuals falling into that category. Therefore the effect of happiness on future hourly wages seems to even have a cubic shape, where wages increase again at the lowest residual values. The quintile dummies confirm the former results showing a non-linear shape, but not confirming the positive effect for the individuals at the bottom of the residual distribution. Given that past hourly wage and education are controlled for in the regression, employers appreciate or expect something additional from happier individuals for them to be paid higher wages accordingly. Moreover, happier individuals might also be better bargainers. The negative effect at the top could be explained by individuals with the highest residual happiness not caring much about wages, such that they do not bargain as intensely. Oishi et al. (2007) also find that the highest levels of income are not reported by the most satisfied individuals, but rather by moderately satisfied individuals and Binder and Coad (2011) find a decreasing importance of, e.g., income with increasing quantiles of happiness using quantile regressions. I can therefore confirm the formulated hypothesis about a positive relationship between happiness and wages. However, once non-linearities are allowed for, the results suggest that especially individuals at the top, but also at the bottom of the distribution contribute to a non-linear shape of the effect. The first question arising at this point concerns the mechanisms behind such effects, therefore the forthcoming section attempts to provide further insights. However, the mechanisms shown in the following mostly focus on reemployment rather than wages. The channels for these two outcome variables appear not to be similar.

4.3. Potential mechanisms The following analyses test differential effects for subgroups such as men and women, as well as distinguishing between regular reemployment and self-employment as outcomes variables. Moreover, I extend estimations to variables capturing aspects of personality, which include the subjective probability of finding a job within the next six months possibly displaying a sort of self-esteem or optimism, the locus of control, neuroticism and extraversion.11

4.3.1. Male vs. female Table 5 shows the results for reemployment separately by gender, and to the best of my knowledge, such differential effects for men and women have never been shown. Interestingly, the results suggest that the male unemployed are driving the main results, as the effects for women are not statistically significant and substantially smaller than for men. This result displays a rather unexpected pattern. Moreover, the results also suggest a slight non-linearity at the bottom, with the reemployment probability for the most unhappy men increasing slightly. This would be in line with the prediction that the more dissatisfied the individuals, the more intensely they try to exit unemployment. However, the effect is only significant at the 10% level and not close to the 5% mark. Therefore, it is difficult to generalize this finding. Why should happiness only be a driver for unemployed males with respect to their reemployment probability? It could be that this selected sample displays a non-representative selection for males and females, in the sense that men still exhibit a different labor market attachment than women, and therefore their satisfaction levels have differential impacts. Additionally, the male residual happiness distribution has longer tails (women may reply more carefully or avoid outliers), and thus effects at the bottom and top can be driven by the male responses. The male and female reemployment rate is virtually the same. As it will be shown in Table 8, unobserved personality traits can basically account for the differences between men and women. However, it would be interesting for further research to consider whether such a differential pattern also exists in other settings, not only restricted to unemployed individuals.12

11 Readers should keep in mind that the different subsamples and specifications allow for a different distribution of the dependent variable, see Williams (2009). 12 Graham and Chattopadhyay (2013) consider gender differences with respect to well-being around the world. However, well-being serves as an outcome variable rather than a driver of behavior in their study.

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

11

Table 5 Employed at the second interview – gender subsamples. Male Sample (1)

i

Female Sample

(2)

(3)

(4)

i if i > 0

(6)

(7)

(8)

(9)

(10)

−0.003 (0.032) −0.034 (0.061)

0.081 (0.061)

|i | if i < 0

−0.142 (0.049)***

i if i > 0 2i

(5)

0.080 (0.032)**

−0.012 (0.049) 0.371 (0.158)** −0.096 (0.048)**

if i > 0

|i | if i < 0

  2i  if i < 0

0.103 (0.160) −0.050 (0.054) −0.367 (0.139)***

0.066 (0.123)

0.056

−0.020

(0.033)* 1185 1185 1185 1185 # of observations 1185 Log likelihood −734.274 −736.715 −733.243 −734.776 −731.686

(0.029) 1346 1346 1346 1346 1346 −846.948 −846.801 −846.919 −846.379 −846.679

Sources: IZA Evaluation Dataset S, own calculations. State unemployment rates from the federal unemployment agency. Notes: Logit regressions. Parameter estimates are shown. Murphy and Topel (1985) standard errors in parentheses. i = residual happiness. Positive (negative) residual happiness contains the residual values while setting the negative (positive) values to zero. Further control variables: see Table 4. Three observations are not used in the second stage of the male subsample due to perfect prediction of failure by the reason for termination of previous job “parental leave”. Eight observations are not used in the second stage of the female subsample due to perfect prediction of success by the dummy of the federal state “Bremen”. * Significant at 10%. ** Significant at 5%. *** Significant at 1%.

4.3.2. Regular employment vs. self-employment Table 6 shows the results when differentiating regular employment and self-employment in the second wave, with both possibilities being jointly analyzed thus far. The results appear quite intriguing, with columns (1)–(5) showing the estimations whereby the dependent variable is equal to one if an individual became reemployed, excluding the self-employed. Compared to the main results, all coefficients decrease; moreover, most statistical significance disappears. The nonlinear shape at the top is still apparent, yet is only significant at the 10% level. Columns (6)–(10) show the results when only considering Table 6 Regular employment vs. self-employment. Regular employed at the second interview (1)

i

(2)

(3)

(4)

0.027 (0.022)

i if i > 0

(6)

(7)

(8)

0.000 (0.043)

(10)

0.215 (0.080*** −0.065 (0.034)*

i if i > 0

−0.282 (0.083)*** 0.186 (0.111)* −0.063 (0.035)*

2i if i > 0 |i | if i < 0

  2i  if i < 0 2341 −1516.194

(9)

0.163 (0.048)***

|i | if i < 0

# of observations Log Likelihood

Self-employed at the second interview (5)

2341 −1516.993

2341 −1515.088

2341 −1515.281

0.804 (0.239)*** −0.213 (0.082)** −0.106 (0.090)

−0.559 (0.205)**

0.010

0.075

(0.020)

(0.041)*

2341 −1514.971

1249 −434.383

1249 −437.424

1249 −434.150

1249 −433.982

1249 −433.054

Sources: IZA Evaluation Dataset S, own calculations. State unemployment rates from the federal unemployment agency. Notes: Logit regressions. Parameter estimates are shown. Murphy and Topel (1985) standard errors in parentheses. i = residual happiness. Positive (negative) residual happiness contains the residual values while setting the negative (positive) values to zero. Further control variables: see Table 4. One observation is not used in the second stage of the self-employed subsample due to perfect prediction of failure by the reason for termination of previous job “care for person in need”. * Significant at 10%. ** Significant at 5%. *** Significant at 1%.

12

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

Table 7 Personality in Life satisfaction estimation first wave. Life Satisfaction in wave 1 (1) Measure of self-esteem

(2)

Locus of control index std.

(4)

−0.354 (0.044)* 0.126 (0.044)*

0.267 (0.089)* 0.381 (0.049)* −0.234 (0.047)* 0.044 (0.046)

2535 0.148

2407 0.179

0.477 (0.045)*

Neuroticism index std. Extraversion index std. # of observations R2

(3)

0.460 (0.090)*

2413 0.126

2542 0.161

Sources: IZA Evaluation Dataset S, own calculations. State unemployment rates from the federal unemployment agency. Notes: OLS regressions. Robust standard errors in parentheses. Further control variables: see Table 2. * Significant at 1%.

self-employment, with all coefficients increasing compared to the main results, being up to four times larger than the coefficients for regular employment. Moreover, they are all statistically significant, which suggests that happiness is mainly a driver for those individuals exiting from unemployment into self-employment. As with the male sample, the squared term on negative residual happiness is also slightly positively significant, suggesting a slight increase in the entry into self-employment for the most unhappy unemployed individuals. Given that statistical significance is rather low, it is difficult to ascertain whether this is a systematic pattern. The turning point for the self-employed is at a residual happiness value of 1.9, which is slightly higher than for the whole sample. This result can represent a valuable contribution given the increasing interest and literature regarding personality and entrepreneurship (see, e.g., Caliendo et al., 2011b; Caliendo and Kritikos, 2012). The results on gender and self-employment appear to be interrelated, as men are more likely to be self-employed. However, further differentiating the male sample by self-employment and employment shows that there is still a happiness effect for men with respect to regular employment (that is not apparent for women).13 4.3.3. Probability of finding a job and personality One advantage of the dataset that I use for this analysis is the variety of topics covered; therefore, the estimations can be extended to variables that are rarely available, which to my knowledge has not been achieved in such a way. This will be undertaken with three sorts of variables: (a) the subjective probability of finding a job; (b) the concept of locus of control; and (c) the personality traits of neuroticism and extraversion. Table 8 displays the main results with a step-wise inclusion of these variables, as well as including all three sorts of variables in the subsample estimations of gender and future self-employment. Importantly, all variables are included already in the first stage. Table 7 displays the latter results. The first variable is supposed to capture a sort of self-esteem or optimism, and refers to the question of whether the respondent rates his or her probability of finding a job within the next six months as very high, rather high, rather low or very low. I include a dummy variable indicating a very high probability of finding a job. This variable may be interpreted as optimism (Caliendo et al., 2010) or self-esteem, with Lyubomirsky et al. (2006) showing that happiness and self-esteem are distinct concepts. It could also be a realistic assessment of the person’s employability and should even more account for a possible reverse causality bias. There is information available concerning this variable for 2413 individuals in my sample. Table 7 shows that there is a highly significant relationship between a high probability of finding a job and happiness. Column (1) in Table 8 displays the results for the second stage. A subjective high probability of finding a job within the next six months has a strong positive effect on the reemployment probability. Most coefficients on residual happiness decrease, but the main result of an inversely U-shaped effect is robust. This offers further reassurance that the results do not seem to be driven by a reverse causality issue, since the main results are robust. One possibility to minimize the bias arising from not being able to include individual fixed effects in the estimation is to include personality traits. Moreover, these traits could possibly be able to explain certain effects. There are a number of personality questions in the questionnaire, some of them referring to the locus of control. This is a concept involving the subjective belief of whether life’s outcomes are outside one’s control and can rather be attributed to fate or luck (external), or alternatively whether life’s outcomes depend on one’s own decisions and behavior (internal). The locus of control is an important trait in this regard, since it has been shown to be related to happiness as well as labor market outcomes. Individuals with an internal locus of control have been found to be associated with higher happiness (Verme, 2009; Becker et al., 2012), and external individuals have been associated with a lower probability of full-time employment (Braakmann, 2009) and lower reservation wages (Caliendo et al., 2010), whereas internal individuals exert higher job search effort (Caliendo et al.,

13

Results are not shown.

Table 8 Mechanisms.

Reemployed

Reemployed

Reemployed

Hourly wage

Reemployed male sample

Reemployed female sample

Reemployed regular

Reemployed self-employed

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

0.026 (0.022) No

0.035 (0.022) No

Locus of control index std.

0.027 (0.022) 0.506 (0.097)*** No

No

Neuroticism index std.

No

0.144 (0.044)*** No

Extraversion index std.

No

No

−0.045 (0.046) 0.049 (0.045)

0.013 (0.023) 0.463 (0.099)*** 0.133 (0.049)*** 0.015 (0.050) 0.009 (0.047)

0.009 (0.005)* 0.026 (0.024) 0.023 (0.012)* −0.008 (0.012) 0.018 (0.011)

0.055 (0.033)* 0.315 (0.142)** 0.150 (0.073)** 0.047 (0.074) 0.103 (0.071)

−0.041 (0.034) 0.634 (0.148)*** 0.158 (0.073)** −0.024 (0.072) −0.086 (0.068)

0.008 (0.023) 0.480 (0.102)*** 0.096 (0.050)* 0.035 (0.051) −0.017 (0.049)

0.062 (0.050) 0.206 (0.214) 0.478 (0.101)*** −0.118 (0.102) 0.244 (0.099)**

i if i > 0

−0.016 (0.044) Yes No No

0.007 (0.043) No Yes No

0.018 (0.042) No No Yes

−0.026 (0.045) Yes Yes Yes

0.010 (0.011) Yes Yes Yes

0.040 (0.065) Yes Yes Yes

−0.110 (0.067)* Yes Yes Yes

−0.033 (0.045) Yes Yes Yes

0.048 (0.093) Yes Yes Yes

−0.072 (0.034)** Yes No No

−0.057 (0.034)* No Yes No

−0.072 (0.034)** No No Yes

−0.048 (0.035) Yes Yes Yes

−0.015 (0.008)* Yes Yes Yes

−0.110 (0.052)** Yes Yes Yes

0.030 (0.053) Yes Yes Yes

−0.041 (0.036) Yes Yes Yes

−0.124 (0.083) Yes Yes Yes

0.231 (0.115)** −0.086 (0.037)** Yes No No

0.072 (0.107) −0.022 (0.034) No Yes No

0.181 (0.110)* −0.057 (0.036) No No Yes

0.046 (0.115) −0.025 (0.038) Yes Yes Yes

0.050 (0.024)** −0.015 (0.007)** Yes Yes Yes

0.217 (0.163) −0.059 (0.050) Yes Yes Yes

−0.144 (0.165) 0.013 (0.056) Yes Yes Yes

0.024 (0.117) −0.019 (0.037) Yes Yes Yes

0.553 (0.275)** −0.200 (0.102)* Yes Yes Yes

−0.160 (0.090)* 0.021

−0.066 (0.086) 0.002

−0.120 (0.087) 0.012

−0.086 (0.091) 0.009

−0.064 (0.022)*** 0.012

−0.236 (0.136)* 0.031

0.063 (0.135) −0.009

−0.072 (0.093) 0.008

−0.226 (0.186) 0.026

(0.020) Yes No No

(0.019) No Yes No

(0.020) No No Yes

(0.021) Yes Yes Yes

(0.005)*** Yes Yes Yes

(0.032) Yes Yes Yes

(0.033) Yes Yes Yes

(0.021) Yes Yes Yes

(0.039) Yes Yes Yes

2413

2542

2535

2407

1296

1136

1260

2219

1193

i Measure of optimism

Measure of optimism LOC Neuroticism, extraversion |i | if i < 0 Measure of optimism LOC Neuroticism, extraversion

i if i > 0  if i > 0 2 i

Measure of optimism LOC Neuroticism, extraversion |i | if i < 0

  2i  if i < 0 Measure of optimism LOC Neuroticism, extraversion # of observations

13

Sources: IZA Evaluation Dataset S, own calculations. State unemployment rates from the federal unemployment agency. Notes: Logit ((1)–(4), (6)–(9)) and OLS (5) regressions. Each column includes results of five different regressions separated by a line. Parameter estimates are shown. Murphy and Topel (1985) standard errors in parentheses. i = residual happiness. Measure of optimism is a dummy indicating “the probability of finding a job within six months is very high” as opposed to “rather high”, “rather low” and “very low”. All measures included in the second stage are already included in the first stage of these estimations (i.e. locus of control is included in the first stage of the estimations in column (2) and (4)–(9)). Positive (negative) residual happiness contains the residual values while setting the negative (positive) values to zero. Further control variables: see Table 4. Three observations are not used in the second stage of the male subsample due to perfect prediction of failure by the reason for termination of previous job “parental leave”. Eight observations are not used in the second stage of the female subsample due to perfect prediction of success by the dummy of the federal state “Bremen”. One observation is not used in the second stage of the self-employed subsample due to perfect prediction of failure by the reason for termination of previous job “care for person in need”. * Significant at 10%. ** Significant at 5%. *** Significant at 1%.

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

Reemployed

14

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

2010). Moreover, to the best of my knowledge the locus of control has never been included as a further explanatory variable when studying happiness as a determinant, and thus it is important to analyze this possible mediating pathway. There is information available concerning this variable for all individuals in my sample. Table 7 shows that there is a highly significant relationship between a higher internal locus of control and life satisfaction.14 Column (2) in Table 8 displays the results for the second stage, indicating that the more internal the locus of control, the higher the reemployment probability. Moreover, all coefficients on residual happiness decrease and only the coefficient on negative residual happiness is still slightly statistically significant. These results suggest that the main effect of happiness on reemployment is driven by differences in personality with respect to internality. Besides this personality trait, there are also variables in the dataset referring to the “Big Five” classification of personality traits (see, e.g., Costa and McCrae, 1992). This concept distinguishes between five broad dimensions of personality, namely openness, conscientiousness, extraversion, agreeableness and neuroticism. Unfortunately, complete information is only available about extraversion and neuroticism for all respondents in the first wave. Given that these seem to be the main predictors of life satisfaction out of the five traits (Diener and Lucas, 1999), not including information about the others should not harm the analysis. Moreover, De Neve and Oswald (2012) show that extraversion and neuroticism serve as mediating pathways of life satisfaction affecting income, and thus it seems crucial to include exactly these two personality dimensions. Neuroticism is found to be negatively related to life satisfaction, whereas extraversion exhibits a positive relationship (DeNeve and Cooper, 1998). Regarding labor market outcomes, Uysal and Pohlmeier (2011) find neuroticism to have a negative effect on the instantaneous probability of finding a job. There is information about these variables for 2535 individuals in my sample. Column (3) in Table 7 shows the results when including extraversion and neuroticism in the life satisfaction estimation.15 In line with previous findings, neuroticism has a negative effect and extraversion a positive one. The results presented in column (3) in Table 8 show that neuroticism and extraversion do not exhibit any statistical significant effect on the reemployment probability one year later. However, all coefficients on residual happiness decrease and most of them also lose their statistical significance. As a conclusion, extraversion and neuroticism can explain the main effect from happiness on future reemployment. They have slightly less explanatory power than the locus of control, but both concepts seem to capture psychological differences between the respondents related to their happiness and labor market outcomes. Column (4) in Table 8 shows the results including all three sorts of variables that confirm the previous findings. It is certainly important to ascertain whether the results regarding the hourly wage, gender subsamples and selfemployment still hold. Columns (5)–(9) show the results when controlling for a high subjective probability of finding a job within the next six months, locus of control, neuroticism and extraversion in the first and second stage. These results show that happiness seems to have an effect on hourly wages up and above these personality traits, since the main effects are robust. Moreover, the relationship between the personality traits and wages is not so clear. Interestingly, the unexpected pattern between men and women seems to be mainly driven by personality differences, given that the coefficients for male unemployed all decrease and only show some statistical significance for the negative residual values, such that higher negative values leading to a lower reemployment probability. For women, the coefficient on positive residual happiness is negative and even slightly statistically significant, albeit almost below the 10% level. This would suggest a rather negative relationship between happiness and reemployment for women. However, without any stronger statistical relationship, it is difficult to identify any deep gender differences regarding how happiness impacts labor market outcomes. In any case, it would be important for future research to identify any gender differences when investigating happiness as a driver and its relationship with personality, which seems to represent a strong mechanism for differences between men and women. Columns (8) and (9) show the results regarding regular reemployment and self-employment. Interestingly, the upper panel of column (9) shows that extraversion exhibits a positive statistically significant effect on the exit to self-employment, whereas there is no such effect on regular reemployment. Moreover, the positive relationship between happiness and future selfemployment can be explained by the personality variables; however, the non-linearity for the positive residual values is robust, since the squared term of the positive residual happiness is at the border of the 5% significance level. Just as is found with respect to wages, happiness seems to have an effect on exit into self-employment up and above personality. 4.4. Robustness checks 4.4.1. Exclusion restriction An important issue neglected thus far is the identification of the two-stage estimation. Using an exclusion restriction that appears with a non-zero coefficient in the first stage yet does not appear in the second stage will yield an identified equation.

14 Constructing the locus of control index relies heavily on Caliendo et al. (2010). Respondents are asked ten statements related to attitudes toward life and the future and are supposed to agree on a scale from 1 to 7. Caliendo et al. (2010) performed a factor analysis that attributed certain items to the internal locus of control concept and certain others to the external one. For the full index, all items are standardized and the aggregated external ones are subtracted from the aggregated internal items. The full index is then standardized once more and enters the regression as such. A higher value refers to a more internal locus of control. 15 Respondents are asked to rate themselves on a scale from 1 to 7 with respect to certain personality characteristics. The standardized index of neuroticism consists of three items regarding worrying, nervousness and being relaxed. The standardized index of extraversion consist of items regarding being talkative, sociable and shy/reserved. Given that higher values of being relaxed and shy/reserved indicate the opposite of neuroticism and extraversion, these two variables enter the indices with a reversed scale.

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

15

Table 9 Exclusion restriction in first stage. First stage: OLS life satisfaction estimation first wave No. of households in living area belonging to: Upper social class

Control Variables

0.001 (0.001)** −0.001 (0.000)** 0.000 (0.000) 0.000 (0.000) −0.000 (0.001) Yes

# of observations R2

2542 0.117

Upper-middle social class Middle social class Lower-middle social class Lower social class

Second stage: logit/OLS estimations Employed at the second interview (1)

i

(2)

(3)

Logarithm of hourly wage at the second interview (4)

i if i > 0

(6)

(7)

(8)

0.025 (0.042)

(10)

0.020 (0.010)** −0.083 (0.033)**

i if i > 0

−0.025 (0.008)*** 0.237 (0.110)** −0.073 (0.035)**

2i if i > 0 |i | if i < 0

  2i  if i < 0 2542 −1619.804

(9)

0.016 (0.005)***

|i | if i < 0

# of observations Log likelihood/R2

(5)

0.042 (0.022)*

2542 −1621.566

2542 −1618.471

2542 −1619.292

0.073 (0.027)*** −0.020 (0.009)** −0.145 (0.087)

−0.083 (0.022)***

0.015

0.014

(0.020)

(0.005)***

2542 −1618.175

1383 0.325

1383 0.322

1383 0.325

1383 0.324

1383 0.329

Sources: IZA Evaluation Dataset S, own calculations. State unemployment rates from the federal unemployment agency. Notes: Logit/OLS regressions. Parameter estimates are shown. Murphy and Topel (1985) standard errors in parentheses. i = residual happiness. Positive (negative) residual happiness contains the residual values while setting the negative (positive) values to zero. Further control variables in first stage: see Table 2. Further control variables in second stage: see Table 4. * Significant at 10%. ** Significant at 5%. *** Significant at 1%.

To the best of my knowledge, this paper is the first to test the robustness of the results with respect to an exclusion restriction when using residual happiness as an explanatory variable. This variable should affect happiness, but not reemployment. Here, the living area’s social class (the number of households in a living area belonging to upper, upper-middle, middle, lowermiddle or lower social class) serves as an exclusion restriction. The variable is comprised of information gathered by the survey institute through actual site inspections. The variable is assumed to determine life satisfaction, but not directly the reemployment probability. Given that most of the variables in this dataset are somehow related to reemployment, it appears to be a reasonable fit. The variable displays the number of households belonging to a certain social class in the living area – defined as a neighborhood of around 500 households. Regarding the relationship with happiness, this may also tackle a relative aspect (Luttmer, 2005). It is constructed out of factors such as household income, purchasing power parity and quality of the residential area, defined by, e.g., distance to parks and the development structure of buildings. The choice of the exclusion restriction is supported by evidence showing that neighborhood quality does not determine eventual earnings, unemployment likelihood and welfare participation, but well-being (Oreopoulos, 2003; Ludwig et al., 2012). Moreover, residential mobility in Germany is rather low, with moving for employment-related reasons only accounting for a small share of around 10%, where commuting may be the preferred option (Caldera Sánchez and Andrews, 2011). Therefore, sorting due to employment prospects should pose no serious problem. There could be some correlation between the neighborhood’s social class and the individual’s own vocational degree, and in turn with the reemployment probability. However, individual

16

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

Table 10 Probability of responding in the second wave. Interview wave 2 Interview cohort (Reference: cohort 1) Cohort 2 Cohort 3 Cohort 4 Cohort 5 Cohort 6 Cohort 7 Cohort 8 Cohort 9 Cohort 10 Cohort 11 Cohort 12 Time between unemployment entry and interview (Reference: 1 month) 2 months 3 months 4 months Net hourly wage of last job (Euros) Duration of last job (Reference: until 1 year) 1–5 years 5–10 years More than 10 years 0 months Male Age Age squared Disabled Marital status (Reference: divorced/widowed) Married Single Partner (Reference: no partner) Educational degree (Reference: no degree) Secondary school (9 yrs) Secondary school (10 yrs) Technical college entrance qualification (11–12 yrs) General qualification for university entrance (12–13 yrs) Vocational degree (Reference: no degree) Apprenticeship (dual system) Specialized vocational school University, technical college Children in household Migrant status (Reference: native) 1st generation migrant 2nd generation migrant Constant # of observations Log likelihood

0.813 0.984 1.181 0.463 1.164 1.043 0.895 0.056 1.224 1.117 1.182

(0.163)*** (0.168)*** (0.162)*** (0.157)*** (0.151)*** (0.157)*** (0.166)*** (0.160) (0.160)*** (0.153)*** (0.152)***

−0.129 −0.226 −0.360 0.054

(0.082) (0.100)** (0.218)* (0.079)

−0.063 0.029 −0.055 −0.272 0.023 0.027 −0.008 −0.010

(0.071) (0.104) (0.109) (0.298) (0.065) (0.027) (0.036) (0.126)

0.076 0.113 −0.165

(0.099) (0.103) (0.118)

−0.001 0.164 0.353 0.531

(0.291) (0.293) (0.319) (0.303)*

0.207 0.306 0.379 0.207

(0.107)* (0.132)** (0.147)*** (0.075)***

−0.402 −0.025 −2.035

(0.105)*** (0.108) (0.580)***

4741 −3067.200

Source: IZA Evaluation Dataset S, own calculations. Notes: Logit regressions. Parameter estimates are shown. Robust standard errors in parentheses. Further control variables include dummies for German federal states. * Significant at 10%. ** Significant at 5%. *** Significant at 1%.

educational and vocational degrees are added as control variables, and thus any correlation of such a kind should be taken into account in any case.16 Table 9 displays the results for the first and second stage. The results in the upper panel show that living in an area with a higher number of households belonging to the upper social class significantly raises life satisfaction, whereas a higher number of upper-middle households significantly decreases it. A larger number of middle, lower-middle and lower class households does not significantly influence life satisfaction. The lower panel of Table 9 shows that the main results are very robust to including an exclusion restriction.

16 Two variables that display more exogenous shocks available in the data are the information about whether the respondent’s father was dead when the respondent was 15 and whether the respondent is widowed. Including these as instruments yield robust results. However, possible objections include that individuals adapt to these kind of life events (Clark et al., 2008) and that there are only 87 observations with a dead father during adolescence and 37 observations of widowed respondents.

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

17

Table 11 Panel mortality correction. Employed at the second interview (1)

i

(2)

(3)

(4)

(5)

0.032 (0.022)

i if i > 0

−0.001 (0.044)

(7)

(8)

−0.074 (0.034)*

(10)

−0.024 (0.008)** 0.296 (0.116)* −0.103 (0.038)**

if i > 0

|i | if i < 0

  2i  if i < 0 2542 −3022.943

(9)

0.013 (0.011)

i if i > 0

# of observations Log likelihood/R2

(6) 0.013 (0.005)*

|i | if i < 0

2i

Logarithm of hourly wage at the second interview

2542 −3025.099

2542 −3020.064

2542 −3017.123

0.069 (0.026)* −0.021 (0.009)* −0.112 (0.091)

−0.087 (0.023)***

0.009

0.015

(0.020)

(0.005)**

2542 −3019.851

1383 0.314

1383 0.312

1383 0.315

1383 0.314

1383 0.320

Sources: IZA Evaluation Dataset S, own calculations. State unemployment rates from the federal unemployment agency. Notes: Logit/OLS regressions weighted using inverse probability weights. Parameter estimates are shown. Murphy and Topel (1985) standard errors in parentheses. i = residual happiness. Positive (negative) residual happiness contains the residual values while setting the negative (positive) values to zero. Further control variables: see Table 4. * Significant at 5%. ** Significant at 1%.

4.4.2. Attrition Panel mortality is a common problem related to longitudinal datasets. Attrition may lead to selection bias, prompting me to check the main results for robustness as follows. With respect to the dataset that I use for this analysis, around 50% of the original sample could be reached for a second interview. In order to control for possible attrition bias, I apply inverse √ probability weighting. Assuming the selection process is based on observables this procedure is N-consistent (Wooldridge, 2002). This method involves two steps, the first step of which involves estimating a logit model of the probability of replying in the second wave on several independent characteristics of the first wave. In the second step, I calculate inverse probabilities for each individual with the fitted probabilities to reply in the second wave. The main estimation results are weighted using these inverse probability weights, which take higher dropout rates with respect to certain individual characteristics into account. Table 10 shows the results of a logit estimation, with the probability of replying in the second wave being the dependent variable. Compared to the first cohort, most cohorts are significantly more likely to reply in the second wave. The same is true for higher vocational degrees and whether children are present in the household. The larger the timelag between the actual unemployment entry and the first interview, – there is an average time gap of two months – the lower the probability of giving a second interview. Furthermore, first generation migrants are also more likely to drop out. Other characteristics such as information about the last job, gender, geographical distribution, age and marital status are not relevant for the selection process. Moreover, I do not include life satisfaction, residual happiness, very high probability of finding a job, locus of control, neuroticism and extraversion in the estimation shown here, given that they do not contribute to the selection process. Table 11 shows the main results correcting for panel mortality, with the first five columns displaying the results for future reemployment and the last five columns the results for future wages. The effects with respect to reemployment slightly decrease, and the main effect of residual happiness loses its statistically significance, whereas the nonlinear effect for positive residual happiness is particularly robust to attrition bias. Further analysis shows that dropping around 20 observations with weights above 4.5 (the overall mean is 2.07 with a standard deviation of 0.84, and the median is 1.8) leads to very robust main results. In summary, there appears to be some selection bias in the sample affecting the reemployment results which, however, is driven by outliers with very large weights. The results of future wages are very robust with respect to attrition as shown by columns (6)–(10). Therefore, correcting for attrition bias does not seem to alter the key findings of the main analysis. 5. Conclusions This study investigates the effect of unemployed individuals’ happiness on their future labor market outcomes. In particular, I use an inflow sample into unemployment in Germany to calculate residual happiness, which displays higher (or

18

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

lower) satisfaction levels than would be predicted by a number of demographic and socioeconomic characteristics. In a second step, I analyze the effect of this residual on future labor market outcomes. By accounting for the individual’s labor market history and information about future job prospects, I am able to reduce reverse causality bias. Even though I am able to reduce these worries regarding reverse causality bias and to include otherwise unobserved personality traits, final worries about causality remain. Unless I do not work with experimental data or a strong instrument, I do not want to claim with certainty that the results are clear causal effects. I am confident however, that many issues have been resolved and the results also relate to other findings in the literature (Oishi et al., 2007; Binder and Coad, 2011). I find a statistically significant inverted U-shaped effect of residual happiness on an unemployed individual’s future reemployment probability and reentry wage, even after controlling for demographic and socioeconomic characteristics. The effect on wages even seems to have a cubic shape. Further investigation offers three mechanisms that have not been previously shown in this context: (a) happiness is mainly a predictor for exit into self-employment; (b) only male unemployed experience an effect of happiness on reemployment; and (c) the concept of locus of control and the personality traits of neuroticism and extraversion are main drivers of the baseline effect on regular reemployment and are able to explain the effect on reemployment for males. The non-linear effects on wages and self-employment are robust to the inclusion of personality traits. The result regarding self-employment is a new and interesting finding that may have implications for the literature on entrepreneurship. However, this study is only representative of the selected unemployment population in Germany, and thus results are not necessarily generalizable to the whole German population. On that note, it is important to assess whether results can be generalized to other countries. Clearly, national labor market institutions differ, however, there is no obvious reason to believe that the observed pattern should be substantially different in other countries. For a final and robust answer, future research should investigate the relationship in other countries. Moreover, investigating effects of happiness in other contexts could shed light upon whether significant gender differences also exist outside the labor market. As personality appears to be the main driver of the effect on regular reemployment as well as the observed gender difference, accounting for unobserved personality traits seems crucial in detecting any mediating pathways when investigating the impact of happiness – also by gender. The inverted U-shaped effect in this study suggests that the optimal level of happiness is not necessarily the highest if reemployment and higher reentry wages are considered desirable outcomes for the unemployed individual and society. Being too happy or too unhappy may lead to the loss of motivation and resilience to pursue one’s life in a conscious and healthy manner due to different forms of lethargy – either because there is no desire for change or due to a form of depression. In the same spirit, psychologists have found high levels of happiness to be optimal for success in the domains of volunteer work and personal relationships, whereas only a moderately high level of happiness is optimal for achievement outcomes such as income and education. Oishi et al. (2007) state that a slight dissatisfaction can serve as motivation to achieve more, earn more money, and in other words, to (self-)improve, as confirmed by the findings of this paper. Even though these findings may be rather small steps toward actual policy implications, potential policies should focus on the least and most satisfied individuals as they appear to be the group at the highest risk of long-term unemployment. However, as the channels for the least and most satisfied are likely to be entirely different, one would need to first prevent the risk of a form of depression and second the risk of insufficient pressure during the job search process, by monitoring these unemployed individuals. Policies should therefore be carefully designed, also taking job quality into consideration as the most and least satisfied individuals also appear to be less successful with respect to wages later on. Thus, maximizing happiness should not necessarily be the goal that future policy-makers should consider (Frey and Stutzer, 2000b). Instead, optimizing happiness appears to be the enduring and desirable long-term ambition. References Andersen, S.H., 2013. Common genes or exogenous shock? Paternal unemployment on children’s schooling efforts. European Sociological Review 29 (3), 477–488. Böckerman, P., Ilmakunnas, P., 2006. Elusive effects of unemployment on happiness. Social Indicators Research 79 (1), 159–169. Becker, A., Deckers, T., Dohmen, T., Falk, A., Kosse, F., 2012. The relationship between economic preferences and psychological personality measures. Annual Review of Economics 4, 453–478. Binder, M., Coad, A., 2011. From average Joe’s happiness to miserable Jane and cheerful John: using quantile regressions to analyze the full subjective well-being distribution. Journal of Economic Behavior and Organization 79 (3), 275–290. Boyce, C.J., 2010. Understanding fixed effects in human well-being. Journal of Economic Psychology 13 (1), 1–16. Braakmann, N.,2009. The role of psychological traits for the gender gap in full-time employment and wages: evidence from Germany. In: SOEP papers on Multidisciplinary Panel Data Research 162. DIW Berlin. Burger, J.M., Caldwell, D.F., 2000. Personality, social activities, job-search behavior and interview success: distinguishing between PANAS trait positive affect and NEO extraversion. Motivation and Emotion 24 (1), 51–62. Caldera Sánchez, A., Andrews, D., 2011. Residential mobility and public policy in OECD countries. OECD Journal: Economic Studies 2011/1, 185–206. Caliendo, M., Cobb-Clark, D., Uhlendorff, A.,2010. Locus of control and job search strategies. In: IZA Discussion Paper 4750. Institute for the Study of Labor (IZA), Bonn. Caliendo, M., Falk, A., Kaiser, L.C., Schneider, H., Uhlendorff, A., van den Berg, G.J., Zimmermann, K.F., 2011a. The IZA evaluation dataset: towards evidencebased labour policy-making. International Journal of Manpower 32 (7), 731–752. Caliendo, M., Fossen, F., Kritikos, A.,2011b. Personality characteristics and the decision to become and stay self-employed. In: IZA Discussion Paper 5566. Institute for the Study of Labor (IZA), Bonn.

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

19

Caliendo, M., Kritikos, A., 2012. Searching for the entrepreneurial personality: new evidence and avenues for further research. Journal of Economic Psychology 33 (2), 319–324. Clark, A., 2003. Unemployment as a social norm: psychological evidence from panel data. Journal of Labor Economics 21 (2), 323–351. Clark, A., Georgellis, Y., Sanfey, P., 2001. Scarring: the psychological impact of past unemployment. Economica 68 (270), 221–241. Clark, A., Oswald, A., 1994. Unhappiness and unemployment. The Economic Journal 104, 648–659. Clark, A.E., Diener, E., Georgellis, Y., Lucas, R.E., 2008. Lags and leads in life satisfaction: a test of the baseline hypothesis. The Economic Journal 118, F222–F243. Costa, P.T., McCrae, R.R., 1992. Revised NEO Personality Inventory (NEO-PI-R) and NEO Five-factor Inventory (NEO-FFI) Manual. Psychological Assessment Resources, Odessa, FL. Cummins, R.A., Nistico, H., 2002. Maintaining life satisfaction: the role of positive cognitive bias. Journal of Happiness Studies 3 (1), 37–69. De Neve, J.-E., Oswald, A., 2012. Estimating the influence of life satisfaction and positive affect on later income using sibling fixed effects. Proceedings of the National Academy of Sciences of the United States of America 109 (49), 19953–19958. DeNeve, K.M., Cooper, H., 1998. The happy personality: a meta-analysis of 137 personality traits and subjective well-being. Psychological Bulletin 124 (2), 197–229. Di Tella, R., MacCulloch, R.J., Oswald, A., 2001. Preferences over inflation and unemployment: evidence from surveys of happiness. American Economic Review 91, 335–341. Diener, E., Lucas, R., 2000. Subjective emotional well-being. In: Lewis, M., Haviland, J. (Eds.), Handbook of Emotions. , 2nd ed. Guilford, New York, pp. 325–337. Diener, E., Lucas, R.E., 1999. Personality and subjective well-being. In: Kahnemann, D., Diener, E., Schwarz, N. (Eds.), Well-Being: The Foundations of Hedonic Psychology. Russell Sage, New York, pp. 213–229. Easterlin, R., 1974. Does economic growth improve the human lot? Some empirical evidence. In: David, P., Reder, M. (Eds.), In Nations and Households in Economic Growth: Essays in Honor of Moses Abramovitz. Academic Press, New York, London, pp. 89–125. Ferrer-i-Carbonell, A., Frijters, P., 2004. How important is methodology for the estimates of the determinants of happiness? The Economic Journal 114, 641–659. Frey, B.S., Stutzer, A., 2000a. Happiness, economy and institutions. The Economic Journal 110, 918–938. Frey, B.S., Stutzer, A., 2000b. Maximizing happiness? German Economic Review 1 (2), 145–167. Frey, B.S., Stutzer, A., 2002. What can economists learn from happiness research? Journal of Economic Literature 40, 402–435. Gerlach, K., Stephan, G., 1996. A paper on unhappiness and unemployment in Germany. Economics Letters 52, 325–330. Gielen, A.C., van Ours, J.C., 2010. Layoffs, quits and wage negotiations. Economics Letters 109, 108–111. Gielen, A.C., van Ours, J.C., 2012. Unhappiness and job finding, In: IZA Discussion Paper 6320. Institute for the Study of Labor (IZA), Bonn. Goudie, R.J.B., Mukherjee, S., Neve, J.-E.D., Oswald, A.J., Wu, S.,2012. Happiness as a driver of risk-avoiding behavior. In: Discussion Paper 1126. Centre for Economic Performance, London. Graham, C., Chattopadhyay, S., 2013. Gender and well-being around the world. International Journal of Happiness and Development 1 (2), 212–232. Graham, C., Eggers, A., Sukhtankar, S., 2004. Does happiness pay? an exploration based on panel data from Russia. Journal of Economic Behavior and Organization 55, 319–342. Guven, C., 2011. Are happier people better citizens? Kyklos 64 (2), 178–192. Guven, C., 2012. Reversing the question: does happiness affect consumption and savings behavior? Journal of Economic Psychology 33, 701–717. Hermalin, B.E., Isen, A.M., 2008. A model of the effect of affect on economic decision making. Quantitative Marketing and Economics 6, 17–40. Ifcher, J., Zarghamee, H., 2011. Happiness and time preference: the effect of positive affect in a random-assignment experiment. American Economic Review 101 (7), 3109–3129. Isen, A.M., Reeve, J., 2005. The influence of positive affect on intrinsic and extrinsic motivation: facilitating enjoyment of play, responsible work behavior, and self-control. Motivation and Emotion 29 (4), 297–325. Kassenboehmer, S., Haisken-DeNew, J., 2009. You’re fired! The causal negative effect of unemployment on life satisfaction. The Economic Journal 119, 448–462. Kenny, C., 1999. Does growth cause happiness, or does happiness cause growth? Kyklos 52 (1), 3–23. Knabe, A., Rätzel, S., Schöb, R., Weimann, J., 2010. Dissatisfied with life but having a good day: time-use and well-being of the unemployed. The Economic Journal 120, 867–889. Korpi, T., 1997. Is utility related to employment status? Employment, unemployment, labor market policies and subjective well-being among Swedish youth. Labour Economics 4 (2), 125–147. Lepper, H.S., 1998. Use of other-reports to validate subjective well-being measures. Social Indicators Research 44, 367–379. Ludwig, J., Duncan, G.J., Gennetian, L.A., Katz, L.F., Kessler, R.C., Kling, J.R., Sanbonmatsu, L., 2012. Neighborhood effects on the long-term well-being of low-income adults. Science 337, 1505–1510. Luttmer, E.F.P., 2005. Neighbors as negatives: relative earnings and well being. Quarterly Journal of Economics 120 (3), 963–1002. Lynch, L.M., 1989. The youth labor market in the eighties: determinants of re-employment probabilities for young men and women. The Review of Economics and Statistics 71 (1), 37–45. Lyubomirsky, S., King, L., Diener, E., 2005. The benefits of frequent positive affect: does happiness lead to success? Psychological Bulletin 131, 803–855. Lyubomirsky, S., Tkach, C., Dimatteo, M.R., 2006. What are the differences between happiness and self-esteem? Social Indicators Research 78, 363–404. Machin, S., Manning, A., 1999. The causes and consequences of longterm unemployment in Europe. In: Ashenfelter, O., Card, D. (Eds.), Handbook of Labor Economics. Elsevier, New York, pp. 3085–3139. Marks, G.N., Fleming, N., 1999. Influences and consequences of well-being among Australian young people: 1980–1995. Social Indicators Research 46, 301–323. Mavridis, D.,2010. Can subjective well-being predict unemployment length? In: Policy Research Working Paper 5293. The World Bank. McCall, J., 1970. Economics of information and job search. Quarterly Journal of Economics 84 (1), 113–126. Mortensen, D.T., 1970. Job search, the duration of unemployment, and the Phillips curve. American Economic Review 60 (5), 847–862. Murphy, K.M., Topel, R.H., 1985. Estimation and inference in two-step econometric models. Journal of Business and Economic Statistics 3 (4), 370–379. Ohtake, F., 2012. Unemployment and happiness. Japan Labor Review 9 (2), 59–74. Oishi, S., Diener, E., Lucas, R.E., 2007. The optimum level of well-being – can people be too happy? Perspectives on Psychological Science 2 (4), 346–360. Oreopoulos, P., 2003. The long-run consequences of living in a poor neighborhood. Quarterly Journal of Economics 118 (4), 1533–1575. Oswald, A., Proto, E., Sgroi, D.,2009. Happiness and productivity. In: IZA Discussion Paper 4645. Institute for the Study of Labor (IZA), Bonn. Oswald, A., Wu, S., 2010. Objective confirmation of subjective measures of human well-being: evidence from the U.S.A. Science 327 (5965), 576–579. Roberts, B.W., Caspi, A., Moffitt, T.E., 2003. Work experiences and personality development in young adulthood. Journal of Personality and Social Psychology 84 (3), 582–593. Shimer, R., 2008. The probability of finding a job. American Economic Review: Papers and Proceedings 98 (2), 268–273. Stiglitz, J.E., Sen, A., Fitoussi, J.-P., 2009. Report by the Commission on the Measurement of Economic Performance and Social Progress. www.stiglitz-sen-fitoussi.fr Urry, H.L., Nitschke, J.B., Dolski, I., Jackson, D.C., Dalton, K.M., Mueller, C.J., Rosenkranz, M.A., Ryff, C.D., Singer, B.H., Davidson, R.J., 2004. Making a life worth living – neural correlates of well-being. Psychological Science 15 (6), 367–372.

20

A. Krause / Journal of Economic Behavior & Organization 96 (2013) 1–20

Uysal, S.D., Pohlmeier, W., 2011. Unemployment duration and personality. Journal of Economic Psychology 32, 980–992. Verkley, H., Stolk, J., 1989. Does happiness lead into idleness? In: Veenhoven, R. (Ed.), How Harmful is Happiness? Consequences of Enjoying Life or Not. University Press, Rotterdam, The Netherlands. Verme, P., 2009. Happiness, freedom and control. Journal of Economic Behavior and Organization 71, 146–161. Williams, R., 2009. Using heterogeneous choice models to compare logit and probit coefficients across groups. Sociological Methods and Research 37 (4), 531–559. Winkelmann, L., Winkelmann, R.,1995. Unemployment: Where Does it Hurt? In: Discussion Paper 1093. Center for Economic Policy Research. Winkelmann, L., Winkelmann, R., 1998. Why are the unemployed so unhappy? Evidence from panel data. Economica 65, 1–15. Winkelmann, R., 2009. Unemployment, social capital, and subjective well-being. Journal of Happiness Studies 10 (4), 421–430. Wooldridge, J., 2002. Econometric Analysis of Cross Section and Panel Data. MIT Press, Cambridge, MA.