Unemployment Scarring by Gender: Human Capital Depreciation or Stigmatization? Longitudinal Evidence from the Netherlands,

Transcription

1 Unemployment Scarring by Gender: Human Capital Depreciation or Stigmatization? Longitudinal Evidence from the Netherlands, Running head: Unemployment Scarring by Gender Irma Mooi-Reci Assistant Professor of Sociology and Research Methodology Vrije Universiteit (VU) Amsterdam Harry B. Ganzeboom Professor of Sociology and Social Research Methodology Vrije Universiteit (VU) Amsterdam Acknowledgments We gratefully acknowledge insightful comments on earlier drafts of this paper from seminar participants at Society of Labor Economics, American Sociological Association, and Research Committee on Social Stratification (RC 28). Corresponding author Irma Mooi-Reci De Boelelaan HV Amsterdam The Netherlands Tel vu.nl 1

2 ABSTRACT (182 WORDS) Much theoretical controversy surrounds the question of what drives unemployment scarring by gender: human capital depreciation or stigmatization? We address this gap by arguing that if stigma effects drive unemployment scarring, then scarring effects should exacerbate in specific labor market situations (e.g., slack labor markets) and among specific (disadvantaged) groups (e.g., age, parenthood and ethnicity groups). By contrast, little or no contextual variation would indicate that human capital depreciation effects dominate. Using a comprehensive longitudinal dataset from the Dutch Labor Supply Panel for 1980 through 2000, the analysis provides strong evidence that for women, the effects of unemployment scarring are driven mainly by human capital depreciation while for men stigma effects dominate. After combining the separate dimensions of unemployment into a single unemployment index, our results reveal that the effects of unemployment scarring upon women exacerbate with their parenthood status while for men these become more pronounced at advanced age and during economic downturns. This latter finding contradicts common expectations, while it suggests that the combined effects of unemployment occurrence, duration and incidence lead to stronger stigmatization for men than already reported. 2

3 INTRODUCTION In the last three decades various studies have made important contributions to the understanding of the processes that generate long-term wage setbacks, also known as unemployment scarring (DiPrete 1981; DiPrete and McManus 2000; Gangl 2004, 2006; Gregg 2001; Jacobson et al. 1993; Kuhn 2002). Two competing explanatory mechanisms for unemployment scarring have been proposed in this literature: human capital depreciation and stigmatization. Drawing upon human capital theory (Becker 1964), the first explanation provides a theoretical account for the enduring negative wage effects from a supply-side perspective. Specifically, the human capital explanation argues that loss of human capital during periods of unemployment reduces workers productivity and leads to lower post-unemployment wages subsequently (Gregg 2001; Rhum 1991; Stewart 2000; Mühleisen and Zimmerman 1994). Stigma (or signaling) theory (Spence 1973) provides an alternative to this explanation from a demand-side perspective. With reference to how employers make their hiring decisions, signaling theory assumes that unemployment periods are seen as a signal of unproductivity and low commitment, leading to less secure and lower paid job-offers (Arulampalam 2001; Arulampalam et al. 2000; Gregory and Jukes 2001; Jacobson et al. 1993; Omori 1997; Stevens 1997). Although important progress has been made on this topic, theoretical and empirical challenges remain. First, while existing studies provide conclusive evidence on whether unemployment scars, they still fall short in explaining how unemployment scars. More specifically, how do we know when unemployment scarring arises due to stigma effects and when it arises due to human capital depreciation effects? With exception of Omori (1997), no earlier unemployment scarring studies have made convincing attempts 3

4 to disentangle stigma from human capital effects. As a result, the mediating mechanisms for the relationship between early unemployment and later post-unemployment wages remains largely a black box. Second, earlier studies have shown that the negative impact of unemployment on workers subsequent wages can be attributed to a number of different dimensions of unemployment, such as to its occurrence, duration and repetition. These dimensions may each independently influence workers post-unemployment wages, and in different ways. Yet, while the separate dimensions of unemployment have been extensively examined, the combined effect of these dimensions on postunemployment wages has remained unexplored. Third and finally, existing work on unemployment scarring has been mainly focused on working men. Far less is known about the role of unemployment scarring among women. Are women equally disadvantaged by episodes of unemployment and if so, do existing gender wage inequalities exacerbate after unemployment periods? These questions have remained largely unexplored in the literature and are important to reveal potential double scarring effects that may advance the existing debate on gendered labor market inequalities (Budig and England 2001; Tam 1997) even further. In the present study, we address these three key issues. First, we compare the two competing explanatory hypotheses for unemployment scarring effects: human capital depreciation and the unemployment stigma hypothesis. Similar to Omori (1997), we argue that if stigma effects drive unemployment scarring, such scarring effects should exacerbate in specific labor market situations (e.g., slack labor markets) and among specific (disadvantaged) groups (e.g., age, parenthood and ethnicity groups). By contrast, little or no contextual variation would indicate that human capital depreciation effects 4

5 dominate. The advantage of this approach is the possibility of disentangling stigma from human capital depreciation effects, which has been less common in existing literature. Another advantage is that it gives us the possibility to reveal when and for which social groups unemployment stigma effects dominate human capital depreciation effects. Secondly, we address the measurement issue by including multiple dimensions of unemployment previous unemployment occurrence, repetition and duration to investigate their individual effects on men s and women s post-unemployment wages. We also combine these three dimensions into a single unemployment index by which the full magnitude of unemployment effects is measured. This approach represents a significant contribution to the literature because it provides a more comprehensive perspective on the unemployment scarring effects while at the same time it introduces a tool for other studies that seek to compare unemployment scarring across different countries. Third, our data allow us to examine gender difference in scarring effects in a systematic way. We make use of a rich and comprehensive longitudinal dataset, the Netherlands Labor Supply Panel (OSA), which covers the period in full detail. The panel structure of the data and its clear definition of unemployment among men and women allows us to identify multiple dimensions of unemployment and examine gendered unemployment effects more effectively than earlier studies. More importantly, the data includes rich information on labor market and demographic situation, which provides an excellent opportunity to disentangle stigma from human capital effects. The analytical strategy in our study is to compare two equivalent groups of workers who differ only with respect to their route into employment: one group came into employment 5

6 via a spell of unemployment and the other group via employment. We use fixed-effects panel models that correct for time-constant unobserved heterogeneity to analyze the effects of unemployment and to disentangle human capital depreciation from stigma effects on men s and women s post-unemployment wages. By combining the separate dimensions of unemployment into a single unemployment index, our study reveals that unemployment scarring effects have an additive nature in that the combined effect of single unemployment dimensions is much larger than that of each separate dimension of unemployment individually. Our results suggest that, for women, unemployment scarring effects are driven mainly by human capital depreciation while for men stigma effects prevail. However, our results indicate that unemployment scarring effects exacerbate during economic downturns, contradicting existing expectations in existing economic literature on stigmatization. EXISTING LITERATURE ON UNEMPLOYMENT SCARRING In recent years, a considerable debate has arisen about the role played by trigger events (i.e., disruptive life-course events such as child-birth, divorce, job-displacement and unemployment) in generating patterns of inequality in the labor market (Budig and England 2001; DiPrete and McManus 2000; Gangl 2004; 2006). A rapidly growing body of research suggests that trigger events in general, and unemployment in particular, lead to highly negative and long-lasting scars in workers employment careers and postunemployment wages. Early empirical evidence on unemployment scarring comes from studies conducted in the US, where significant re-employment wage penalties vary around 15-25% percent among once displaced workers compared to a situation without 6

7 such an interruption. In this literature there is conflicting evidence whether the danger of unemployment lies primarily in its (first) occurrence, or especially in the unemployment re-occurrence over time. For instance, Jacobson, LaLonde, and Sullivan (1993) demonstrate that a single occurrence of unemployment leaves significant scars on postunemployment wages even five years after the job loss. According to Stevens (1997), who used panel data from the US Panel Study of Income and Dynamics (PSID) , the key driving force behind these long-lasting wage penalties is the number of earlier unemployment episodes that inhibit workers recovery trajectories. In line with this evidence, Gregory and Jukes (2001) find a persisting negative effect on workers reemployment wages, arising from the re-occurring unemployment periods, or as we call it in this study unemployment repetition. Interesting evidence follows from the study of Arulampalam, Booth and Taylor (2000) who using data from the British Household Panel Data (BHPS) show that especially the first employment interruption carries the highest re-employment wage penalties compared to later interruptions. More recent empirical evidence shows that unemployment scarring effects may work differently for different social groups and in different contexts. For instance using the same PSID data DiPrete and McManus (2000) find that, in general, unemployment reduces household earnings, but that these negative effects alleviate over time as result of counterbalancing effects arising from subsequent employment episodes. In addition, Gangl (2004, 2006) who compares the magnitude of unemployment scarring between US and German workers, finds that wage penalties arising from previous unemployment durations may also alleviate when institutional support in the form of unemployment benefits is provided. In particular, this study demonstrates that German workers 7

8 supported by unemployment benefits experience much lower and more temporary wage losses compared to their equivalent counterparts who live the US. These empirical findings suggest that different dimensions of unemployment entail different negative effects on workers subsequent wages. These effects have different implications for different social groups and may also alleviate when followed by counterbalancing employment periods or when supported by institutions. In this study, we complement the existing studies by examining the separate dimensions together and by disentangling when and why unemployment scars. In what follows we use argumentations from different labor market theories to derive the mechanisms that cause unemployment scarring. MECHANISMS THAT DRIVE UNEMPLOYMENT SCARRING The theoretical literature on unemployment scarring has pointed to two key complementary mechanisms: namely, loss and depreciation of (non)transferable human capital, and reduced bargaining power as result of unemployment stigma. The first perspective explains unemployment scarring in terms of human capital depreciation in which individuals marketable resources are lost during unemployment. The second perspective explains unemployment scarring in terms of social stigma where, above and beyond the human capital depreciation effects, workers previous characteristics affect post-unemployment wages via a statistical discrimination mechanism. In what follows, we will first develop expectations from the human capital hypothesis. Later, drawing on signaling theories and sociological perspectives, we derive stigma-specific expectations. 8

9 Unemployment and Reduced Human Capital In human capital models, workers wage losses after a spell of unemployment are related to changes in their human capital (Becker 1964; 1993). According to this line of argument, human capital can be divided into a generic part, which is acquired through education and is transferable across employers, and a specific part, which is acquired through experience and on-the-job training in a certain firm or sector and is nontransferable across employers (Becker 1993). A direct implication of this distinction is the argument that interruption of the investment in generic and specific training may lead to lower levels of productivity, both instantaneously and in the long run. Specific human capital embodied in a worker s job tenure is lost when unemployment occurs. In particular, when specific skills are tied to a particular industry or sector, and are nontransferable, this may lead to decreased wage bargaining power (Sattinger 1993) once unemployment occurs. In addition, generic human capital may depreciate over long spells of unemployment. Not only initial loss of workers specific human capital but its gradual depreciation over an unemployment spell influences workers market-enhancing productivity. The human capital explanation of unemployment scarring thus points in particular to the duration of previous unemployment as a key factor. Several studies have argued that the number of previous job losses may also have a negative effect upon worker s human capital (Jacobson et al. 1993; Stevens 1997). These authors suggest that in case of multiple job loss, workers do not have sufficient time to recover from the effects of unemployment, because the recovery periods are interrupted by new and recurrent unemployment periods. This repeating process of 9

10 human capital depreciation is extremely precarious as it may lead to low-pay-no-pay cycles (Gregg and Tominay 2004). Taking these arguments together, we anticipate that unemployment affects postunemployment wages through more than one dimension. Our expectation is that previous unemployment occurrence, duration and repetition all constitute important dimensions of unemployment that together can set off a chain of negative wage outcomes in the future. We call this the additive effects of unemployment. We also anticipate that the combined effect of these dimensions may cause an cumulative scarring effect: the simple occurrence of a single unemployment episode may already lead workers into a negative wage cycle, but this may then be accelerated when the episode lasts longer or is repeated. Overall we hypothesize: Hypothesis 1: All else equal, previous unemployment occurrence, duration and repetition will have an additive and cumulative negative effect on post-unemployment wages. Unemployment and Stigma across Social Groups and Labor Market Circumstances Unemployment stigma is the most often cited competing explanation for unemployment scarring. According to signaling theory (Spence 1973), employers must use signals when a direct evaluation of a worker s productivity is unavailable. Within a signaling framework, employers hiring decisions are taken against a background of uncertainty about workers productive capabilities. Whenever such uncertainty exists, employers rely on the observable characteristics of workers such as their past employment history, but also on observables such as their past employment history, age, gender, ethnicity, and family situation which serve as a statistical screening device in the hiring process 10

11 (Lockwood 1991; Eliason 1995). What makes signaling theory important for the explanation of unemployment scarring is the argument that an episode of unemployment is considered an additional signal of low commitment and low productivity by the employer. Above and beyond the wage setback caused by previous unemployment periods, we anticipate that the strength of stigma effects may vary across groups with disadvantaged labor market characteristics (i.e., age, parenthood status and ethnicity) or during specific labor market circumstances (i.e., slack labor markets). In the following discussion, we will take a look at these observable characteristics in the process of stigma development. We start with age as a potential stigmatizing characteristic. Past studies find that a spell of unemployment that occurs when the individual is older increases the likelihood that (s)he will once again experience unemployment in the future and that this can have a stigmatizing effect in the case of older workers (Stevens 1997). According to an explanation proposed by Arulampalam (2001) employers expect younger workers (i.e., younger than 25 years old) to show a pronounced job-shopping behavior that may be characterized by short periods of unemployment. Such early spells of unemployment produce less of a stigma. Conversely, research suggests that if workers experience unemployment at older ages (> 45 years old), they will not only be considered as less productive by employers, but will also have less time to recover from prior spells of unemployment because of shorter employment periods thereafter. This argument brings us to the following hypothesis: 11

12 Hypothesis 2a. All else equal, due to stigma, workers who experience unemployment during older ages will experience stronger wage penalties then those who experience unemployment during younger ages. Another characteristic that may negatively influence employers hiring and wage decisions is workers ethnicity and migration status. Two explanations have been put forward in order to understand the often observed wage gap between natives and nonnatives. The first explanation draws from the assimilation argument that relates immigrants lower wages to lack of educational qualifications and language proficiency (Chiswick 1978; Lalonde and Topel 1997). The second explanation relates to stigma, or statistical discrimination. If employers have little information about the productivity of an immigrant applicant from a specific ethnic group then employers may discriminate on statistical grounds. The uncertainty about the productivity of an immigrant employee creates an extra risk for the firm that becomes expressed in the form of wage differences between natives and non-natives. Existing literature confirms conclusively that firms invest less in immigrant s on-the-job training, which leads not only to lower status and lower paid jobs but also to a continuously growing skill gap between natives and immigrants over time (Nielsen et al. 2004). In this study we extend this literature by testing whether stigma effects related to unemployment are stronger among non-natives then among native workers. We assume that a previous unemployment episode for migrants will reinforce the stereotype related to immigrants as less productive and more frail to recurrent unemployment periods. This will lead to higher wage penalties between natives and non-natives over time. We therefore expect that: 12

13 Hypothesis 2b: All else equal, due to stigma, previously unemployed immigrants will suffer stronger wage penalties then native workers. Stigma effects from unemployment may also vary with the business cycle. In times of economic downturn with high unemployment rates and few open vacancies, the probability of finding a job decreases. Following arguments from the economic literature, a spell of unemployment during a recession can be regarded less of a negative signal to a firm than one experienced during an economic upturn (Blanchard and Diamond 1994; Lockwood 1991). This leads to the expectation that when screening applications during economic downturns firms often attribute the occurrence of unemployment to firm closures and reorganizations rather than to workers own (lack of) motivation or performance. Consequently, if stigma effects exist then unemployment scarring from stigma should be higher during periods with lower unemployment rates than in periods with higher unemployment rates. This argument builds upon two assumptions. First, in a slack labor market when there are more workers competing for a job, firms receive multiple applications. Second, when firms receive multiple applications they choose the worker with the thinnest unemployment record. These ranking models, however, are themselves based on the assumption that the pool of applicants consists of previously unemployed workers. What happens if we allow for continuously employed workers to compete for the same jobs as the unemployed workers? In general, we would then expect firms to prefer those in continuous employment over those unemployed. After all, employed workers are less costly, more efficient and more productive in terms of output and existing networks that may prove more profitable for the company. However, for 13

14 firms, choosing continuous employed over previously unemployed comes at the price of higher wage offers. In this study, we extend the ranking model by including both continuously employed and previously unemployed workers in our analyses. Following these arguments we posit that: Hypothesis 2c. All else equal, due to stigma, previously unemployed workers will suffer more severe wage penalties during economic downturns compared to workers who remain in continuous employment. There is an extensive literature that suggests that stigma effects may also relate to one s parenthood status (Budig and England 2001; Budig and Hodges 2010; Correll et al. 2007; Gangl and Ziefle 2009). Most studies suggest that employers discriminate in terms of hiring decisions, promotion opportunities and wages only against mothers, but not against fathers (Correll et al. 2007). According to Correll et al. (2007) employer discrimination against working mothers is related to the cultural understanding of the motherhood role. Especially in societies in which women are expected to take care of their children, employers unconsciously deem working mothers as less productive and less competent (Ridgeway and Correll 2004). It is especially this cultural expectation that leads to discrimination against mothers in hiring, promotion and wage decisions. By contrast, good fathers are expected to work hard and are seen as more committed to their work. Fathers are therefore expected to experience an advantage over those men without children. A recent study by Budig and Hodges (2010) shows that the motherhood wage penalty varies with the age of children. Specifically, the study finds that the older the children, the higher the hourly wage penalty experienced by the mother. In this study 14

15 we extend this evidence by investigating whether the combination of being unemployed and being a parent (especially a mother) exacerbates existing gender inequalities even further. We argue that mothers who experience unemployment may be doubly stigmatized by employers for not only being a bad but also a unproductive mother. Given the arguments above, this would not hold for fathers. In sum, we expect that: Hypothesis 2d: All else equal, due to stigma, unemployed mothers will exhibit stronger wage penalties than fathers or women without children. Unemployment Scarring and Gender Does unemployment scarring differ between men and women? There exists very little research on the scarring effects among women. Part of this limitation is related to the lack of data that can precisely distinguish episodes of non-employment from episodes of unemployment, especially among women. Existing studies on the wage penalties mostly focus on the so-called gender wage gap and merely emphasize the difference in pay between men and women. Over the past decades, an extensive body of research has emerged about this gender wage gap (see England (2005) for a literature review). Interestingly, most economic research relates the gender wage gap to the process of human capital depreciation. The idea is that women s careers are more often characterized by work interruptions due to caring responsibilities, and therefore women have shorter spells of work experience compared to men. From a firm s perspective, this makes women less competitive and less productive relative to men, which leads to job offers that pay less (Becker 1993; Polachek 1981; Tam 1997). By contrast, the sociological literature favors two other explanations for the gender wage gap. The first is 15

16 the devaluation thesis, which claims that employers ascribe lower value to work, and thus pay lower wages, when it is conducted by women. This claim has been empirically supported by several studies that demonstrate that both men and women earn less in female-dominated occupations (England and Folbre 1999; England 2005; Tam 1997). The second explanation relates to women s choice of motherhood-friendly jobs. These jobs are characterized by reduced working hours, flexible hours, short commuting times and on-site day care facilities and are mostly situated in lower-paying industries (Budig and England 2001; Drobnic, Blossfeld and Rohwer 1999; Gangl and Ziefle 2009; Glass 2004). In this study we examine whether the gender wage gap is echoed in an unemployment recovery gap, and in particular whether unemployed women are more disadvantaged by periods of unemployment then men. We expect that unemployment should reinforce existing gender wage penalties. In particular we expect that: Hypothesis 3: All else equal, previous unemployment occurrence, duration and repetition will have more severe effects on re-employment wages among previously unemployed women then previously unemployed men. DATA, MEASURES AND METHODS Data The data for our study come from the Netherlands Labor Supply Panel (OSA). The OSA is a panel study that is continually refreshed and is targeted at a representative sample of 4,000 to 5,000 respondents in each wave. The first wave was interviewed in 1985 (with a retrospective component reaching back to 1980) and then re-approached in 1986 with further biannual waves until In all waves, panel attrition (around 35% in each 16

17 wave) was compensated by adding fresh respondents using a stratified sampling design. The data includes a wealth of information about respondents family background, their education, and incomes. In particular, in each wave the current earnings situation of the respondent was examined. Moreover, the data provides detailed information about respondents labor market situation with start and end dates of all unemployment episodes, making it possible to trace transitions in a dynamic way. To be included in our analysis, respondents had to be between 21 and 54 years old, be employed at the moment of interview and have positive earnings in a given year. Given these requirements, we start with a sample containing 22,938 wage observations over 10,410 workers, 1,329 of whom experienced unemployment at the time of interview and another 1,971 workers who experienced unemployment between two interview dates. In Table 1A of Appendix A, we show the distribution of wage observations starting with workers with no wage observations to those with consistently nine wage observations over the nine waves. For the purpose of fixed effects models, which we will explain in the next section in more detail, we need at least two wage observations per worker. This means that workers with less than two wage observations are dropped from our analyses leaving us with an effective sample of 11,703 valid wage observations. To identify unemployment episodes in our data we have followed a two-step approach. First, we have used a respondent s reported labor force status at the date of interview, distinguishing between (1) employed, (2) self-employed, (3) unemployed, (4) non-participating, (5) in military service and (6) in education. Unemployment is explicitly defined in the questionnaires as currently out of labor and searching actively for a job. Second, the OSA survey asks respondents to report the start and end dates of any change 17

18 in labor force status that occurred between the current and last interview date. Using this retrospective information, we can record also the unemployment spells that occurred between two interview dates. Table 1 shows the distribution of unemployment at the time of interview (column A) and between the interview dates (column B) by year. In both columns the proportion of unemployed individuals decreases over time. Some of this decrease may be due to attrition, but the high unemployment rate during the mid 1980s and 1990s more likely reflects trends of the business cycle in the Netherlands during the same period. It is interesting to note the sizable loss of unemployment episodes that would occur when disregarding the retrospective information about job losses between the interview dates (column B relative to column A). TABLE 1 ABOUT HERE Using aggregate data from Statistics Netherlands, in Figure 1, we depict the distribution of unemployment rates separately for men and women over the period After a difficult economic period in the 1970s and the beginnings of the 1980s, the Dutch economy recovered in the late 1980s into what is now referred to as the Dutch employment miracle engineered by the Wassenaar agreement of 1982 (Visser and Hemerijck 1997). Unemployment during this period was reduced by strict wage moderations and active labor market policies, which also led to increased (part-time) labor market participation of women. During the mid 1990s, the unemployment rate rose 18

19 again, but this recovered in the late nineties. This trend very closely matches the percentage of the unemployed in our sample. FIGURE 1 ABOUT HERE In Figure 2, we show the distribution of age at first unemployment in our sample. This figure indicates the presence of two groups of men and women in our sample. First, we have a group of men and women who experience their first unemployment around their 20s and 30s, at which point the high proportion indicates a possible mismatch between their attained education and the labor force entry. Second, we find a group of women who experience their first unemployment in their mid-30s or 40s, which is likely to be related to fertility and child-rearing. FIGURE 2 ABOUT HERE Measures Post-unemployment wages. In this study, the dependent variable is the log of hourly wages at time t for individual i. This variable is constructed by dividing the monthly net wages by the hours of work and then taking the logarithm, which is a standard indicator also used in other studies that estimate scarring effects (Gregory and Jukes 2001; Arulampalam et al. 2000). To standardize the variation, we have divided hourly wages by the mean of hourly wages in each particular wave. 19

20 Measuring unemployment. To test the expectations about the scarring effects of unemployment, the following indicators are constructed. Previous unemployment occurrence is measured by constructing a lagged binary unemployment variable, which takes the value of 1 if the worker was unemployed in the previous wave and 0 if continuously employed. Unemployment duration refers to the most recent unemployment spell that is recorded before the interview date. The reference category of 0 covers workers with no unemployment spells during the observation period. Unemployment repetition is measured by counting the number of unemployment episodes over the observation period. To disentangle the effect of multiple unemployment episodes from the effect of a first unemployment spell we have constructed two variables. First, the variable first unemployment is a lagged binary variable, which takes the value of 1 if a worker s unemployment spell in the previous wave was his or her first unemployment. The reference category refers to those in continuous employment. Second, the variable multiple unemployment spells is constructed to capture the effect of unemployment repetition and is a counting variable that refers to the maximum number of unemployment episodes within a worker over the observation period. The reference category refers to those in continuous employment. To capture the combined effects of these unemployment indicators, we have constructed an unemployment index. To do so, we have first standardized the individual unemployment indicators to have a mean of 0 and a variance of 1 (for men and women combined). We then averaged the four standardized unemployment indicators into a single index (Cronbach s alpha of reliability = 0.701). In Figure 1A of Appendix A, we show the kernel distribution of the standardized unemployment scale for men and women separately. Most unemployment 20

21 experienced by women is their first episode, whereas for men multiple unemployment episodes are much more common. Demographic and human capital measures. The measure of education refers to level of education attained by each individual. This variable distinguishes between five levels: (1) elementary education [LO]; (2) lower intermediate education [VBO-MAVO]; (3) higher intermediate secondary education [MBO-HAVO-VWO]; (4) vocational college [HBO] and (5) university degree [WO]. Age (ranging from 21-54) is entered in the model as proxy for work experience. The variable age squared is incorporated to control for a curvilinear relationship between accumulation of work experience and wages. To assess whether workers recover from unemployment, we construct the variable employment duration after unemployment, which is measured as the number of consecutive months between the start of employment after a period of unemployment and the end of that employment period. Workers who have remained in continuous employment are scored 0. To control for differences in wages that relate to differences in the demographic situation we have included the variables of marital status (1 = married / cohabiting workers and 0 otherwise) and the number of children living at home (ranging between 0 and 10). In addition, we create an alternate variable of parents with children older/younger than 12 yrs to measure whether the effects of unemployment vary with the age of their children. This variable distinguishes between five categories: (0) no children; (1) women with children younger than 12 years of age; (2) women with children older than 12 years of age; (3) men with children older than 12 years of age; (4) men with children older than 12 years of age. We also included a dummy variable for respondents 21

22 ethnicity (1= non-dutch; 0 = Dutch) to control for wage differences that arise due to one s immigrant background. Job characteristics and business cycle measures. We also check different jobrelated characteristics that may influence post-unemployment wages such as the number of working hours (ranging between 12 and 40); type of contract (1 = permanent and 0 temporary contracts); working sector (1 = public and 0 private) and the level of occupational status using the International Socio-Economic Index (ISEI) scale of Ganzeboom et al. (1992). Finally, in order to check period and business cycle effects, we include a variable indicating the annual rate of unemployment separately for men and women as reported by Statistics Netherlands (2010) 1. Since we expect men and women to differ in their labor market experience and to have different labor market behaviors, we have conducted the analyses for men and women separately. A detailed description of the variables is provided in Table 2A of Appendix A. Methods To address scarring in terms of wage penalties, a log-wage linear regression panel model is fitted. We apply a fixed-effects model, which eliminates biases that occur by the failure to include controls for unmeasured personal characteristics such as motivation to work or ability to keep a job. In fixed-effects models, comparisons within individuals are conducted by averaging at least two wage observations and by modeling their deviations from this average. Since the unobserved heterogeneity in fixed-effects models is assumed to be time constant, any difference with its mean results in 0 and is dropped from the model. The model yields the following linear specification: 22

23 ln w = β x + α + e (1) it it i it Where ln(wit) is the natural logarithm of hourly wage at time t for individual i. xit refers to a vector of observable variables on individual characteristics, β refers to a transposed vector that accounts for coefficients associated with the observable characteristics. Finally, αi refers to the time-invariant individual specific errors that capture the unobserved heterogeneity and the eit is the equation error term. To examine our human capital depreciation hypothesis we start with a model that includes the separate indicators of unemployment, namely: unemployment occurrence, duration and repetition. In this case, wage equation (1) is extended to the following specification: ln w = β x + γ u + α + e (2) it it i, t 1 i it Where ui,t-1 refers to the vector of (lagged) unemployment dimensions whereas γ refers to a vector that captures the coefficients associated with each separate dimension of unemployment. Next, interaction effects are added to the model to examine the stigma hypothesis. The wage equation (2) therefore extends to the following specification: ln w β + α + e (3) it = xit + γ ui, t 1 + λ (ux) i, t 1, t i it 23

24 where, uxi,t-1 refers to the vector of interactions between disadvantageous micro and macro level labor market conditions (i.e., age, parenthood status, ethnicity, slack labor markets) and the unemployment index with λ as the pertinent vector of coefficients. EMPIRICAL RESULTS Testing the Human Capital Depreciation Hypothesis It is important to recall that a central expectation of our first hypothesis was that each specific indicator of unemployment would have a long-term negative effect on postunemployment wages leading to wage inequalities between those with and without previous unemployment spells. The descriptive results in Figure 3 provide an initial clue concerning the cumulative scarring effect, showing that unemployed workers not only have lower wages compared to their continuously employed companions, but that these wage differentials grow larger over time. FIGURE 3 ABOUT HERE In Table 2 we present baseline results from fixed-effects models for men and women separately. These models do not control for any differences in terms of the job characteristics, human capital or macro fluctuations, but provide information about the gross effects of unemployment indicators on the post-unemployment wages of men and women. Model 1 in Table 2 indicates that for women, unemployment occurrence in the previous wave and a first unemployment spell are key to scarring, inflicting a wage penalty of respectively 12.1% and 10.7%. In contrast, for men all the three different 24

25 dimensions of unemployment seem to affect their wages negatively. It is interesting to note that for both men and women, the first spell of unemployment cuts the deepest scar in re-employment wages, not the number of subsequent spells of unemployment. This result is in line with earlier results established in the UK (Arulampalam et al. 2000) and implies that once unemployed, workers will suffer wage penalties over longer periods. TABLE 2 We next explore whether these results continue to persist once we control for human capital and macro variables. Results from four fixed-effect regression models that test for human capital depreciation effects are presented in Table 3. By including individual-level and business cycle differences, Models 1 and 2 test the human capital depreciation hypothesis. In Models 3 and 4, we included the unemployment index to measure the combined effect of unemployment on workers post-unemployment wages. TABLE 3 With regard to the human capital depreciation hypothesis, we have argued that the previous unemployment occurrence, duration and repetition will have an additive negative effect on post-unemployment wages due to loss and depreciation of human capital. Results from Model 1 and 2 repeat the findings in Table 2, but this time take into account the listed background variables. Specifically, Models 1 and 2 in Table 3 indicate, that for women, an initial spell of unemployment inflicts an average wage scar of about 11.3% compared to women in continuous employment. In addition, women who experienced unemployment one wave earlier earn on average 12.8% lower hourly wages. This wage gap continues to grow by 2% for each 10 additional months in unemployment. For men, a first unemployment spell inflicts hourly wages that are 8.1% lower compared 25

26 to those earned by men in continuous employment and this gap grows larger by 2% for every 10 additional months in unemployment. Beyond this wage gap, the penalty grows larger by 2.4% for every additional unemployment experience. These results imply that the effects of unemployment are additive and cumulative in nature, in that an initial incidence of unemployment not only gives rise to a wage penalty, but this penalty grows larger for every additional month in unemployment for both men and women. How large is unemployment scarring when we combine the different unemployment indicators? To answer this, we have added the unemployment index in Model 3 and 4 in Table 3. The index gives us the opportunity to look at unemployment on a continuous scale where cases vary from workers with few to those with extensive unemployment scores. As the index has been standardized, the coefficients reflect effects per unit standard deviation. According to these estimations and as expected, the full cost of unemployment experiences is higher for women (0.055) than for men (0.033). This implies that a one standard deviation increase in the unemployment index is associated with a drop in re-employment wages of 1.8%. More specific, the standard deviation of the log of hourly wages was estimated 0.33 which constitutes a negative change (0.055*0.33) of 1.8% in the subsequent wages of women. This is slightly larger than the estimated effect for men of around 1.1% (0.033*0.33). Testing for Stigma Effects To examine the unemployment stigma hypotheses 2a-2d, we turn to Models 1 and 2 in Table 4. Using explanations from the stigma model, we hypothesized that unemployment scarring should be higher among older workers, working mothers, immigrants, and would 26

27 also higher in times of economic downturns. As mentioned earlier, the existence of interaction effects should point to the existence of possible stigma effects, while an absence of interaction effect would favor the human capital depreciation hypothesis. TABLE 4 The results from Models 1 and 2 hold a number of interesting implications. First, surprisingly, for women we find no evidence for the expectation that unemployment scarring effects are driven by stigma effects. All four interaction effects between the unemployment and the different groups remain non-significant, which indicates that the wage penalties observed earlier (in Table 3) can be ascribed to women s human capital depreciation that grows larger if unemployment lasts for longer periods. Second, in Model 2, we find strong evidence that unemployment scarring may be the result of stigma effects for men. More specifically, we find that in weak labor markets with high unemployment rates and stiff competition between job seekers, previously unemployed men earn on average wages that are 1.9% lower, compared to those of men in continuous employment. This implies that unemployment during lax labor markets becomes a stronger negative signal for men compared to those who manage to stay in the labor force during economic downturns. This significant effect for men invalidates expectation that employers are less punitive towards those who experience unemployment during difficult economic periods. Third, in our theoretical model we expected to find larger scarring effects among workers with observable characteristics that have the likelihood to transmit negative signals to employers such as older age, ethnic background and parenthood status. Results from the interaction terms in the Models 1 and 2 in Table 4 show that unemployment 27

28 scarring effects exacerbate by 5.3 percent if men experience unemployment above their 40s, while this effect remains negative but not-significant for women. This result can be taken as evidence for the existing view that younger men are less scarred by unemployment, as employers will be less concerned by a pronounced job shopping behavior among them. Unemployment at an older age on the other hand restricts workers wages not only by raising doubts about their productivity, but also by restricting the subsequent employment periods which are needed to compensate the effects from earlier unemployment spells. Although statistically not significant, according to our estimations, being unemployed and non-native may also lead to stronger wage penalties among unemployed men and women, while being an unemployed mother (but not father) may lead to some wage penalties compared to unemployed persons without children. This difference is in line with existing literature that argues that cultural expectations attributed to working fathers as committed and responsible may have a positive effect upon the hiring decision of firms. To explore this latter in more depth and test hypotheses 2d and 3 about the parenthood and gender effects of unemployment, we conduct additional analyses based on a pooled sample shown in Models 1 and 2 of Table 3A in Appendix A. Interaction effects between the unemployment index and gender as shown in Model 1, indicate no significant gender differences in unemployment scarring. In Model 2, however, when we differentiate between mothers and fathers with children of older and younger ages, an interesting result emerges. Specifically, estimates from the pooled regression show that compared to workers without children and no employment interruptions, previously unemployed mothers with children older than 12 years to experience by far the highest 28

29 wage penalties (6.4%) followed by previously unemployed mothers with children younger than 12 years old (2.9%). This result is in line with recent results by Budig and Hodges (2010) for the US, that estimate a wage penalty of around 5 percent for mothers with children older than 12 years situated above the 0.10 wage quintiles. This negative wage effect does not hold for fathers. An explanation for this finding may be related to mothers cumulated disadvantage effects. More specifically, women with older children are more likely to have experienced more interruptions in their work history and therefore suffer more strongly from human capital depreciation effects and lower productivity effects than workers without children who do not experience career interruptions. This implies that above and beyond the negative wage effects related to unemployment spells, being a mother and having children, especially older than 12 years, exacerbates the wage inequalities even more. This confirms our expectation that existing gender wage inequalities exacerbate during unemployment periods but invalidates the existing view that this is due to stigmatization for women. SUMMARY AND CONCLUSION The present study addresses three uncertainties about the relationship between unemployment and subsequent wages. First, while many studies exist on unemployment scarring, it remains unclear when unemployment scarring arises due to stigma and when due to human capital depreciation effects. Second, different studies on unemployment have used different indicators for unemployment making the comparability of unemployment effects difficult to assess. Finally, whether gender effects in scarring exist remains unclear in existing research. Using a longitudinal dataset and after developing an 29

30 unemployment index that combines the major unemployment dimensions, our findings suggest that for women, unemployment scarring effects arise mainly due to human capital depreciation, while for men, they are due to stigma effects. More importantly, we find that for men, unemployment scarring effects are exacerbated by age and in particular during economic downturns. This latter finding invalidates existing expectations in the literature that unemployment should be less scarring when it is experienced during unfavorable economic times (Blanchard and Diamond 1994; Omori 1997). This discrepancy between our findings and those in other studies deserves a closer look. First, existing studies have focused on unemployment effects between short versus long-term unemployed, while in our study we compare unemployed workers to those who have remained in the labor force. Second, our findings are not inconsistent with the theoretical predictions (Blanchard and Diamond 1994) that argue that in slack labor markets when there are more workers than available vacancies, employers prefer workers who remain in the labor force rather than those who have experienced unemployment. Another contribution made by our study relates to the unraveling of gender differences in unemployment scarring. Our analyses suggest that wage penalties are more nuanced for men, but more pronounced for women. More importantly, we provide a first explicit demonstration that gender differences in unemployment scarring are highly apparent among mothers with children, especially when those children are older than 12 years of age. What can we learn from the findings of this particular study? First and foremost, we have shown that it is fruitful to study the effects of different unemployment dimensions simultaneously. The different dimensions of unemployment are measured in 30