Job loss among rich and poor in the United States and Germany: Who loses more income?

Available online at www.sciencedirect.com

Research in Social Stratification and Mobility 32 (2013) 85–103

Job loss among rich and poor in the United States and Germany: Who loses more income? Martin Ehlert ∗ Wissenschaftszentrum Berlin für Sozialforschung (Social Science Research Center Berlin), Reichpietschufer 50, 10785 Berlin, Germany Received 15 February 2012; received in revised form 15 August 2012; accepted 16 November 2012 Available online 23 November 2012

Abstract This article compares household income losses after involuntary job loss between household income quintiles in the United States and Germany. I argue that income trajectories after job loss vary between social strata in country-specific ways because of differences in the labor market, the family and the welfare state. Using panel data from the Panel Study of Income Dynamics and the German Socio-Economic Panel, I calculate household income after job loss for each household income quintile. The results show that job loss in the United States has the most severe effect on the poorest quintile whereas in Germany, the middle quintiles lose most after job loss. My analysis reveals that this is due to differences in the factors that buffer income losses between the strata: In both countries, the lower quintiles have the highest losses in earnings and family income support is comparatively low among them. In Germany, the welfare state ameliorates this because it has a higher impact on the lower quintiles than on the upper quintiles. In the United States on the other hand, the welfare state has a more equal impact among the quintiles and hence does not offset the disadvantages of the lower quintiles that the labor market and the family generate. © 2012 International Sociological Association Research Committee 28 on Social Stratification and Mobility. Published by Elsevier Ltd. All rights reserved. Keywords: Job loss; Household income mobility; Income inequality; Labor market; Welfare state; Family

1. Introduction The comparative study of social stratification in different countries is a well established field in the social sciences. In addition to classical explanations like differences in welfare policy or occupational structure, differences in economic insecurity and mobility recently gained attention (DiPrete, 2002). Involuntary job loss is one of the major causes of economic insecurity in the middle of the life-course because most people depend on labor income during this stage. Therefore, it is crucial to understand the mechanisms that shape the impact

∗

Tel.: +49 30 25491 381. E-mail address: [email protected]

of job loss on economic well-being. Previous research found that nation specific institutions influence how much income people lose through job loss (DiPrete & McManus, 2000a; Gangl, 2004). However, these studies only compared the consequences of unemployment as nation specific averages. Yet, it is likely that income losses through job loss are unevenly distributed within a country. Therefore, I analyze how people in different strata fare after job loss. This idea merges dynamic with classical study of social stratification. Research on economic insecurity has grown substantively during the last decades. Scholars studied the transition in and out of poverty (e.g. Bane & Ellwood, 1986; Layte & Whelan, 2003; Leisering & Leibfried, 1999) and income volatility (e.g. Fritzell, 1990; Gottschalk & Moffitt, 1999; McManus & DiPrete,

0276-5624/$ – see front matter © 2012 International Sociological Association Research Committee 28 on Social Stratification and Mobility. Published by Elsevier Ltd. All rights reserved.

http://dx.doi.org/10.1016/j.rssm.2012.11.001

86

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

2000). Reviewing this research however, Western, Bloome, Sosnaud, and Tach (2012) argue that analyzing events that cause income changes provides most insight about economic insecurity. The literature providing this kind of analysis generally finds that job loss and family break-up are the main triggers for income losses (Burkhauser & Duncan, 1989; DiPrete & McManus, 2000a; Kohler et al., 2012). In a cross-national perspective, welfare state institutions, the family, and the labor market lead to distinct income mobility patterns (DiPrete, 2002). Studies of income losses through job loss differ in terms of the analyzed income. While some focus on wages after the unemployment spell, others analyze household income. Both types of analyses agree that job loss and subsequent unemployment leaves “scars” in incomes. That is to say, in addition to direct loss, job loss lowers income in the long run. Analyses of earnings trajectories found that those who experienced unemployment have lower earnings in the mid and long term than those who did not become unemployed (Arulampalam, 2001; Burda & Mertens, 2001). Cross-national research showed that such scars in earnings are lower in countries that provide long and generous unemployment benefits because such benefits help the unemployed to find better jobs (Gangl, 2004, 2006). Researchers who considered household income after job loss also proved that the welfare state is important in mediating losses. DiPrete and McManus (2000a) found that Americans are worse off than Germans both initially and in the long run because they often remain below their initial income level. Later results that included the 1990s and the 2000s however suggest that this changed over time: With the growing number of long-term unemployed in Germany and the booming labor market in the United States, the differences between the countries diminished (Ehlert, 2012).1 All of these studies have in common that they study job loss and unemployment as an individual risk with little or no consideration of social stratification. This fits well into the idea of “individualization” of social inequality coined by Beck (1986). According to this idea, classes are dissolving because life courses in today’s societies are growing more dissimilar. For example, Beck claimed that labor markets become more volatile and employment insecurity nowadays affects everyone regardless of their position in the social stratification. Following this reasoning, it becomes harder if

1 Researchers also found scars after unemployment in other life domains such as life satisfaction (Clark, Georgellis, & Sanfey, 2001) or job quality (Dieckhoff, 2011).

not impossible to classify individuals into classes. Consequently, research on social stratification should focus on individuals rather than groups. Researchers adopting a dynamic perspective on social inequality often follow this suggestion, and analyze individual mobility regardless of class background. This has not remained uncontested. Mayer (1991) for example warned against a tendency to replace “inequality” by “life course”. That is to say, the structural location of life courses within social stratification should always be considered. Many researchers studying life courses consequently showed that class background is an important determinant of life chances, contrary to Beck’s claims. For example, Mayer and Carroll (1987) analyzed the impact of class background on job shift patterns. Likewise, Allmendinger and Hinz (1998) studied class mobility in different welfare states. Likewise, literature on “cumulative advantage” over the life course showed in multiple cases that a persons origin determines outcomes to a great extent (DiPrete & Eirich, 2006). Despite this continuous focus on social stratification in the field of life course studies, many analyses of economic insecurity and especially income loss after adverse events do not consider this perspective. However, social strata vary in terms of resources and options to cope with job loss and unemployment as I will explicate below. Also, the risk of becoming unemployed is unevenly distributed. People in the lower strata have a increased risk of losing a job (Giesecke & Heisig, 2010; Keys & Danziger, 2008; McGinnity & Hillmert, 2004). In the face of growing income inequalities in almost all western countries, the neglect of stratification in the analysis of income mobility does not seem appropriate. In order to understand individual dynamics in the different strata, the consequences of unemployment also have to be differentiated. A few studies connect income losses after trigger events and social stratification. Burda and Mertens (2001) compare labor earnings before and after unemployment in Germany and found different “scars” in post-unemployment wages depending on preunemployment wages. Those with higher earnings lose more compared to their prior job than those with lower earnings. Similarly, DiPrete and McManus (2000b) report that household income losses after various life course events are higher in the upper parts of the income distribution than in the lower parts in both the United States and Germany. However, they do not explore the causes of this finding. In the field of poverty dynamics, Worts, Sacker, and McDonough (2010) find that incidences of poverty are highly concentrated among already disadvantaged

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

groups in the United Kingdom and the United States. Low education and routine occupations are overrepresented among those who enter poverty. However, they do not differentiate poverty risks by trigger events. Such a differentiation is provided by Vandecasteele (2011), who estimated poverty entry risks after trigger events across EGP classes in 13 European countries. The study finds that upper classes have a higher risk of entering poverty than lower classes. Consequently, the author concludes that unemployment reduces inequalities between classes instead of increasing them. According to this study, unemployment in Europe is an “individualized” life course risk instead of a risk that hits the lower strata harder and leads to a cumulation of disadvantages. Although these studies connect economic insecurity with social stratification, few researchers relate these results to cross-national differences in the welfare state, the family, and the labor market. Aiming to fill this gap, I analyze the differences in income losses between social strata in two countries: The United States and Germany. These two countries differ in the influence the welfare state has on individuals’ lives. The German welfare state generally provides more financial help than its American counterpart (Esping-Andersen, 1990), so that household income trajectories in Germany depend more on government help than in the United States. Conversely, private incomes play a greater role in the United States than in Germany (DiPrete & McManus, 2000a). In this study, I want to analyze whether income losses due to unemployment are worse for the upper or the lower strata in the two countries. Thus, I address one of the key research questions in the field of economic insecurity identified by Western et al. (2012). I go beyond existing literature in two regards. First, I focus on the differences between countries in the stratification of outcomes to gauge the influence of different macro conditions. Second, I try to disentangle the sources of these differences by considering welfare state, family and labor market influences. The article is structured as follows. First, I introduce a theoretical framework to study income losses after job loss for different social strata and develop expectations about differences between the United States and Germany. Then, I describe the data set and the methods used. Finally, I present the results and relate them to the theoretical expectations. A discussion concludes the article. 2. Theoretical background: income mobility and stratification Cross-national differences in life course patterns have attracted much research in recent years (Mayer, 2009).

87

This research generally builds on the idea that macro structures such as welfare state institutions or labor markets structure life courses (Mayer, 1997; Mayer & Müller, 1986). Building on this research, DiPrete (2002) proposed the “mobility regimes” framework to study cross-national differences in the impact of trigger events on household income. According to this approach, nation specific mobility regimes consist of two groups of mechanisms that influence the impact of events. First, there are mechanisms that have direct influence on income mobility. Second, there are mechanisms that provide opportunity structures for “counter mobility events”, that is to say, events that ameliorate the loss. In the case of job loss, the first group includes welfare state transfers such as unemployment insurance as well as existing additional incomes within the household. The second group comprises two coping strategies: individual re-employment and increases in the labor market participation of other household members. For simplicity, I refer to both as “income buffering mechanisms”. Before I begin the elaboration of my theoretical argument, I briefly discuss why I use quintiles of household income as an indicator of social stratification in my analyses. In the social sciences, there is a long debate about adequate measurements of social stratification and social inequality. Broadly speaking, there are two camps: first, those who use job characteristics to determine a person’s social position and group people in classes (e.g. Goldthorpe, 2010) and second, those who group people using resources like income on the individual or household level (e.g. Sørensen, 2000). For my analysis, the advantage of the resource approach is its ability to take the whole household into account whereas the class approach remains on the individual level. Clearly, in male-breadwinner societies, the class of the household head approximates the household’s class. However, the growth of female labor force participation and dual earner-families made households more diverse. If class should mirror living conditions, household income is a better indicator than an individual’s job characteristics (DiPrete, 2003). Therefore, I use quintiles of size adjusted household income as a starting point to test whether income buffering mechanisms have different impacts across social strata. Building on the literature on the sociology of the life course, I conceptualize an individual’s position in the social stratification is the result of processes of accumulation over the life course (DiPrete & Eirich, 2006; Mayer, 2004). These processes have a consequence for income buffering because some of the advantages and disadvantages they cumulate have direct influence on income buffering mechanisms. As a result, job

88

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

losses have different effects on income depending on an individual’s position within social stratification. The cumulation of advantages and disadvantages occur on the individual level and the household level and affect buffering mechanisms stemming from the labor market, the family, and the welfare state. In the following, I explore this argument on different levels beginning with individual labor market risks and labor market structure. Then I turn to the household level and finally to the welfare state. 2.1. Individual factors and labor market structure Income buffering through re-employment depends on two factors: duration of unemployment and wages in the new job. In other words, the incidence and the impact of the counter mobility strategy. Labor market theories generally assume that the two are connected. However, there is disagreement about the direction of this correlation. Human Capital Theory for example (Becker, 1975) predicts a negative relationship: longer unemployment durations lead to lower post-unemployment wages. Search Theory (Mortensen & Pissarides, 1999) on the other hand predicts a positive relationship. Both theories begin with the same reasoning: During unemployment, a person receives job offers at a certain rate and decides which offers to take. The rate at which job offers appear depends on the demand for the individual specific marketable skills, which are often called “human capital”. Thus, people with sought after skills receive more offers and therefore have shorter unemployment durations and higher post-unemployment wages because they can choose a good job out of the many offers. However, if a person stays unemployed for a longer period of time, Human Capital Theory predicts that skills devaluate because they remain unused. Hence, from this perspective, longer spells of unemployment lead to lower re-employment probabilities and lower post-unemployment wages. Accordingly, people with long spells of unemployment receive fewer and worse job offers.2 Search theory on the other hand assumes that post-unemployment wages depend on the quality of the match between employer and employee. However, job offers that match a person’s skills do not occur frequently. Therefore, an unemployed person has to search for some time until a well-paid job is

2

Signaling Theory (Spence, 1973) arrives at a similar conclusion, albeit it assumes a different mechanism: The theory posits that unemployment is seen as a signal for low productivity, which employers use when screening applicants.

found. Hence, according to Search Theory, longer unemployment durations lead to higher post-unemployment wages. Previous research found that the impact of these mechanisms are mediated by labor market structure (Gangl, 2003). Given that labor market structure differs greatly between the United States and Germany, I expect that the importance of the mechanisms varies between the two countries. This has consequences for the variation of post-unemployment outcomes between the strata. According to Marsden (1990), Germany’s labor market is dominated by occupational labor markets (OLM) in which skills and credentials are occupation specific. This enables employees to move between firms within the same sector. For example, if one company closes down, employees move to another firm and use their credentials to enter into a job that is comparable to their prior one. Thus, their human capital is less likely to devaluate. In United States on the other hand, internal labor markets (ILM) dominate. In firms with ILMs, skills are firm specific. Thus, after job loss, employees cannot use their skills in other firms and their human capital depreciates. Hence, even high-skilled workers in the United Sates may have to accept entry-level jobs after unemployment whereas their German counterparts can use their credentials to re-enter at a higher level.3 This difference leads to my first hypothesis: in the long run, individual earnings trajectories should differ more in the United States than in Germany: workers from the upper part of the social stratification should have higher long term losses compared to their prior earnings in the United States than in Germany (Hypothesis 1). However, people in higher social strata in Germany may have very specific skills and job offers for them are infrequent. Therefore they have to search longer in order to find a suitable job. The longer duration of unemployment benefits enables them to do so (Gangl, 2003). This leads to a refinement of Hypothesis 1: In Germany, many unemployed from high strata have longer spells without earnings after job loss, but their earnings rise quickly in the long run (Hypothesis 1a). Thus, the configuration of the welfare state influences labor market mobility (DiPrete, De Graaf, Luijkx, Tåhlin, & Blossfeld, 1997).

3 Clearly, Germany has a mix of portable skills within industries or occupation and firm specific skills (Estevez-Abe, Iversen, & Soskice, 2001). Yet, in comparison to the United States, the main difference is that there are skills that workers can use in different firms because of the vocational training system.

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

2.2. Family The income buffering capacity of the family is generally determined by the labor market participation of the partner and its possible expansion.4 Since most men in couple households work full time, it is mainly the variance in women’s labor force participation that drives cross-national differences in the family income buffer. In both the United States and Germany, the number of dual-earner couples is rising. However, this process has been faster in the United States (OECD, 2007). Also, in the United States, it is more common for both partners to work full time. In Germany on the other hand, combinations of full-time working men and part-time working women are common (OECD, 2011). This reflects cross-national differences in gender role expectations: In Germany, norms that favor a traditional separation of male breadwinners and female homemakers are still more powerful than in the United States (Grunow, Hofmeister, & Buchholz, 2006). These country-specific patterns lead to two gender differences: first, if women in couples lose their jobs, there is almost always a second income. Second, if men in the United States lose their jobs, they are more likely to have a working partner than in Germany. Thus, less of total household income is lost than in Germany. Women’s job losses on the other hand influence total household income more in the United States than in Germany. Therefore, I have to consider men and women separately in my analyses. In addition to existing additional incomes in a couple household, the family buffer also reflects the increase in hours or incomes on part of women whose partners become unemployed. This is known as the “added worker effect” (Lundberg, 1985; McGinnity, 2002). The magnitude of this effect depends on women’s chances on the labor market: Higher educated women have more opportunities to increase their earnings after their partner’s job losses. However, women in couples who are already working may have difficulties increasing their earnings because they cannot change their work hours. Because of these “hours constraints”, the only chance of increasing their income is job change (Altonji & Paxson, 1992; Reynolds & Aletraris, 2010). The impact of the family income buffer presumably differs across social strata because of educational 4 Note that I only focus on family support from within the household. Clearly, private transfers and parental wealth are also important determinants of mobility over the life course (Pfeffer, 2011). Yet, the impact of financial help by family members outside the household is beyond the scope of this study.

89

homogamy. The partner’s potential to provide income rises with the level of education. This is connected to stratification because education of partners within a household is often highly correlated. Such educational homogamy emerges from segregated marriage markets: Most individuals meet their partners during schooling (Blossfeld, 2009). Hence, if there is educational homogamy, highly educated individuals also live in families that provide a high income buffering potential through highly educated partners. In international comparison, the United States and Germany exhibit about the same rate of homogamous couples (Blossfeld & Timm, 2003). This leads to another hypothesis: The family should have a higher impact for individuals in higher strata than for lower strata in both countries (Hypothesis 2). 2.3. Welfare state The impact of the welfare state on incomes after job loss also differs depending on the prior life course and the position within the social stratification. According to Esping-Andersen (1990), welfare states do not only decomodify, i.e. ameliorate the losses, but also stratify. That is to say, welfare state benefits are not equally accessible for all. For example, eligibility rules make benefits conditional on certain characteristics such as having paid into the insurance system for a certain amount of time. Also, there may be ceilings in benefit payout or means tests that limit benefits to certain amounts, depending on prior incomes and other incomes inside the household. Thus, welfare state institutions are life course sensitive (Leisering, 2003). Therefore, I explore the extent of stratification in the two welfare states studied here.5 In the United States, unemployment insurance and minimum income schemes are administered by the states. A wide variety of regulations exist. Generally, unemployment benefits replace about 50% of former wages up to an earnings ceiling for up to six months. Many states apply earnings requirements that exclude low-wage and part-time workers. Beyond unemployment benefits there is almost no further protection. Until 1996, Aid to Families with Dependent Children (AFDC) paid comparatively low benefits to unemployed families. Since 1996, AFDC was replaced by Temporary Assistance for Needy Families (TANF) that brought stricter eligibility requirements and a limitation of benefit receipt to five years. On the other hand, reforms

5 For more detailed information, see the overviews of the respective welfare state institutions by Grell (2011) and Wörz (2011).

90

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

expanded the “Earned Income Tax Credit” (EITC), a negative income tax for low-income households. Hence, low-income households with labor income do not have to pay income taxes and even receive a refund. Thus, the American system helps high-wage earners less than low-wage earners because of the ceiling in unemployment benefit payout. On the other hand, earnings requirements and the lack of benefits if people are ineligible for unemployment insurance decrease the advantage of low-wage workers. Therefore, I expect that lower social strata should gain more from welfare state intervention after job loss than higher strata. However, the difference between low and high strata is presumably not large because of the disadvantages for the lowest strata. In the long run after job loss, EITC should help low-income households. Therefore, some welfare state impact should exist even some years after job loss, especially for the lower strata if they are working. In Germany, unemployment insurance provides 67% of former net wages if people live with children and 60% if they live without children for up to one year (pre-1998: 68% and 63% respectively). Like in the United States, there is a maximum benefit amount. Contrary to the United States, there is no earnings requirement. The minimum income schemes covering the unemployed who do not receive benefits are more encompassing. Longterm unemployed received up to 57% of their former wages until the Hartz Reforms in 2004. The reforms turned the benefits for long-term unemployed into a lump-sum payment. In addition, social assistance provides a universal minimum income. Hence, although the Hartz Reforms made these systems less generous, there is still an unlimited safety net for long-term unemployed. The German system should lead to a higher impact of the welfare state for lower strata than for higher strata. Because of the unlimited safety net at the bottom, Germans with low income are presumably better secured by the welfare state than U.S. Americans. In other words, there is a floor effect built in the German welfare state that its American counterpart lacks. Overall, the difference in welfare state impact should be higher between the strata in Germany than in the United States (Hypothesis 3).

both data sets, a set of comparable variables is available through the “Cross-National Equivalent File” (CNEF) (Frick, Jenkins, Lillard, Lipps, & Wooden, 2007). The PSID was initially a yearly survey, but after 1997 data has been collected biennially. Hence, from 1997 onward, I measure only two-year changes in income. In the CNEF, PSID data is available from 1980 to 2007.6 The GSOEP, on the other hand, offers yearly data from 1984 to 2011, which is completely available in the CNEF. I restrict the analysis in Germany to pre-unification territory, thus excluding the new states after 1990 so as to avoid changes in the population over time.7 Thus, my observation period spans from 1980 to 2007 in the United States and from 1984 to 2011 in Germany. The dependent variables individual earnings, pre-government household income, and post-government household income are from the CNEF. To ensure comparability over time, I deflate the incomes using the consumer price index provided in the CNEF. I then adjust pre and post-government household income for household size using the new OECD equivalence scale.8 Job loss is defined as moving from work to unemployment. In the PSID, respondents were asked for how many weeks of the previous calender year they were working or not working and actively looking for a job. To construct a comparable measure in the GSOEP, I use data from the activity calendar. During the survey, the respondents are asked to mark the months and the corresponding labor market activity. Because it is possible to report more than one status in a single month, I apply a state space proposed by Gangl (2003, p. 56) and deleted months of unemployment in which the respondent also marked some form of employment. Then, so as to render the data comparable with the question in the PSID, I sum up the number of months each calendar year. This obviously removes the information about the timing of job loss, that is at which point in the year the event occurred. In comparison to other studies, this is obviously a coarse measure of job loss. However, because information on the timing of unemployment is not available in the PSID, the comparable measure I chose seems the best compromise. Also, the dependent variable income is only available as a yearly total.

3. Data and method The analyses are based on micro data from two household panels – the “Panel Study of Income Dynamics” (PSID) for the United States and the “German SocioEconomic Panel Study” (GSOEP) for Germany. For

6 For reasons of comparability over time, I exclude the 1990 LatinoSample and the 1997 immigrant sample. 7 Also, I exclude the 2002 high-income sample and the 2006 refreshment sample. 8 To account for the economies of scale of a household, the head is weighted with 1, other adults with 0.5, and children with 0.3.

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

Using the variables described above, I define the event “job loss” as more than two months of unemployment preceded by more than seven months of work and positive wages. I deliberately exclude shorter spells because they are likely to be labor market churning or voluntary job change. To further exclude transitions to unemployment that are not due to involuntary job loss, I include persons aged 25–55 only. Before and after this age bracket, unemployment may occur because of transitions from or to education or because of early retirement. Job loss is an event that can occur several times during the life course. I therefore construct a data set consisting of seven year episodes (two before, four after) for each event. I also add a comparison group that contains all person years that are not in one of those episodes and that span over at least five consecutive years. Additionally, I restrict this comparison group to individuals who are employed and earn positive wages at the beginning of the control episode. Finally, I delete all incomplete episodes and those with missing information on one of the variables. Thus, I use balanced panels of person-years. In order to estimate the change in the living situation of those who become unemployed, I decided to analyze percentage changes in income. Compared to absolute changes in income, this operationalization assumes that the severity of changes depends on prior income. For example, a reduction of $10,000 is a different event for those earning $30,000 a year (≈−33%) compared to those earning $100,000 a year (=−10%). Yet, a 50% reduction is severe for both groups. Because changes in percentages can have huge positive outliers, I top-coded all changes above +1000%. To estimate the impact of job loss on household income, I use a difference-in-difference approach with statistical matching. Hence, I compare income changes through job loss with income changes of those who did not become unemployed during the same period. Expressed formally in a stylized way, I estimate: (Yi,t1 − Yi,t0 ) − (Yj,t1 − Yj,t0 ) where Y is the income, i is a person who loses a job at t1 and j is a person who does not lose a job during the same period. This technique controls for all influences that are constant within the individuals and also period effects such as common income shocks during a recession. However, the comparison to the control group who did not become unemployed may yield spurious results if the two groups differ in characteristics that are also important for income trends and chances on the labor

91

Table 1 Coarsenings of variables used for CEM. Variable

Coarsening

Education

US: < High School vs. High School vs. > High School Ger.: School w/o voc. training vs. voc. training vs. university 25–35 vs. 35–45 vs. 45–55 Below vs. above 5 years Single vs. multi-person household Yes vs. no Women vs. men As described above

Age Tenure in prev. job Household size Children Sex Household income quintiles Year of job loss Minority

Four year intervals US: Black vs. non-black Ger.: Migration background vs. no migration background

market. To render the control group more similar to the treatment group, I use coarsened exact matching (CEM) (Iacus, King, & Porro, 2012). This technique conducts exact matching based on coarsened variables. That is to say, I recode each variable used to match observations into meaningful sets of values such as age groups or income quartiles. CEM then constructs strata consisting of unique combinations of these values. Both those who become unemployed and the control group are assigned to one of these stratums. Within each stratum, CEM constructs weights for the control cases that reflect their number. For example, in strata with many control cases and few unemployed, the control cases receive a small weight. Thus, after reweighting, the unemployment group and the control group are identical in terms of the coarsened variables. Observations in the treatment or control group for which there is no match in the data set are pruned from the analysis. Compared to matching approaches based on distance metrics like propensity score matching or mahalanobis distance matching, this technique has the advantage of including all interactions between the covariates used for matching by default. Simulations show that CEM usually performs equal or better than standard matching approaches (King, Nielsen, Coberley, Pope, & Wells, 2011). Table 1 summarizes the coarsening of the variables used for matching. All variables used are measured before unemployment and the weights are assigned to the whole episode. Note that the coarsening takes single-parent households into account because it uses all possible interactions of the household and the children variable. In the regression models, I additionally control for changes in household size to address the influence of changes in partnership status. Because the

92

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

Table 2 Averages of individual characteristics comparing men with and without unemployment. US men

1st Quintile Age HH size % with partner Yrs. of educ. Partner’s yrs. of educ. Individuals Episodes Episodes w/o match Person years 2nd Quintile Age HH size % with partner Yrs. of educ. Partner’s yrs. of educ. Individuals Episodes Episodes w/o match Person years 3rd Quintile Age HH size % with partner Yrs. of educ. Partner’s yrs. of educ. Individuals Episodes Episodes w/o match Person years 4th/5th Quintile Age HH size % with partner Yrs. of educ. Partner’s yrs. of educ. Individuals Episodes Episodes w/o match Person years

W. Germany men

With UE

Without UE

With UE

Without UE

36.4 3.6 66.9 11.6 11.6 272 341 7 1858

37.8 3.9 79.1 12 12.1 911 5024 1887 25,775

40.3 3.8 85.6 10.4 10.1 195 238 5 1428

40.4 4 91.8 10.8 10.2 777 4430 1382 26,580

37.2 3.4 71.6 12.1 12.3 226 292 7 1604

38.2 3.7 84.3 12.8 12.7 1042 7734 2909 39716

41.9 3.1 77.8 10.8 10.9 148 181 10 1086

40.9 3.7 92 11.1 10.8 906 6375 3256 38250

37.4 3.1 74.5 12.7 12.9 190 227 2 1253

39.1 3.5 86.2 13.4 13.3 952 7992 3289 41,007

41.6 2.9 78.5 11.1 10.9 116 131 4 786

41.4 3.3 86.8 11.7 11.3 972 6422 3647 38,532

39.6 2.7 72.3 13.9 13.4 252 288 5 1559

40.4 3 84.8 14.5 14.3 1501 14109 4353 72,986

40.2 2.4 75.6 12.4 12.1 122 145 5 870

41.7 2.8 84.3 13.1 12.3 1790 12418 7383 74,508

Sources: CNEF, PSID, and GSOEP, author’s calculations.

observations within the episodes are not independent, I use clustered standard errors on the level of the primary sampling units. 4. Results

Furthermore, the tables show the differences between those with an unemployment episode and the control cases.9 Because there is a strong gender difference in incidences and consequences of job loss, I conduct the following analyses for men and women separately (cf. DiPrete & McManus, 2000a). Since there are very few

4.1. Descriptives Before I present the income trajectories after job loss, I first depict the composition of the household income quintiles in the two countries in Tables 2 and 3.

9 More technically, I compare the characteristics of the episodes. Individuals in the control group may have also experienced job loss at some point before or after the control episode. In this case they appear in both groups at different points during their life course.

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

93

Table 3 Averages of individual characteristics comparing women with and without unemployment. US women

1st Quintile Age HH size % with partner Yrs. of educ. Partner’s yrs. of educ. Individuals Episodes Episodes w/o match Person years 2nd Quintile Age HH size % with partner Yrs. of educ. Partner’s yrs. of educ. Individuals Episodes Episodes w/o match Person years 3rd Quintile Age HH size % with partner Yrs. of educ. Partner’s yrs. of educ. Individuals Episodes Episodes w/o match Person years 4th/5th Quintile Age HH size % with partner Yrs. of educ. Partner’s yrs. of educ. Individuals Episodes Episodes w/o match Person years

W. Germany women

With UE

Without UE

37.5 3.4 34.7 12 12 294 343 6 1762

38.4 3.5 45.1 12.2 12 1286 5901 2008 29,703

40.8 2.9 54.2 10.9 10.5 108 115 6 690

40.4 3.4 70.4 10.9 10.8 712 2786 1505 16,716

37.6 3.5 60.7 12.6 12.6 195 227 2 1210

38.5 3.4 64.1 12.8 12.6 1105 7190 3098 36,296

39 3.1 77.9 10.7 10.8 125 139 1 834

40.8 3.4 81 11 11 807 4423 2373 26,538

38 3.2 72 13 12.8 155 172 3 913

39 3.3 74.4 13.3 13.3 992 7007 3642 35,722

41 3 81.2 11 11.1 119 134 5 804

40.8 3.1 82.4 11.4 11.4 856 4921 2700 29,526

39 2.8 75.4 13.9 14.3 195 220 2 1173

40 2.9 78.9 14.3 14.4 1408 11,850 4430 60,712

40.7 2.7 87.3 11.7 12 185 208 3 1248

41.2 2.8 87.7 12.5 12.8 1605 10,355 4300 62,130

With UE

Without UE

Sources: CNEF, PSID, and GSOEP, author’s calculations.

instances of job loss in the highest two quintiles in both countries, I analyze them jointly. The bottom four rows for each quintile provide information about the number of cases. The number of episodes is equal to the number of job loss events. Compared to that, the number of individuals is always smaller indicating that some individuals experience multiple events. Additionally, I provide the number of episodes pruned from the analysis during the application of CEM because there was not matching episode in the control group.

Tables 2 and 3 depict that low educated people and single households are overrepresented in the group experiencing unemployment. Americans who lose their jobs are slightly younger than those who do not, whereas there is no such age pattern in Germany. This reflects the American labor market structure where early careers are generally less stable than in Germany with its vocational training system (Allmendinger, 1989). Beyond the difference in age, people who lose their jobs also have fewer years of

94

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

education than those in the control group.10 This confirms that job loss mainly hits already disadvantaged groups (Giesecke & Heisig, 2010; Keys & Danziger, 2008). The greatest difference between individuals with and without job loss is the percentage living with a partner. Men and low-income women who lose their jobs differ from the control group by about 10 percentage points. By contrast, high-income women show no difference. Hence, for men and low-income women who lose their jobs, there is a lower chance that family income support is available than for those who are continuously employed. One possible explanation for this is that some characteristics make people more marriageable and more successful on the labor market simultaneously. Another explanation could be the “marriage premium”, i.e. the finding that men earn more after marriage (Korenman & Neumark, 1991; Pollmann-Schult, 2011). The two main mechanisms that seem to cause this premium — married men’s higher orientation towards money and employer discrimination in favor of married couples — may also cause disadvantages for people living in single households. These findings indicate that statistical matching is necessary. CEM balances the two groups on basis of these variables. Between the quintiles, household characteristics differ much as well. First, household size is much larger among the lower quintiles in both countries. Hence, lowincome households often have to care for more people than high-income households. Clearly, this is partly due to the weighting of household income by household size. Yet, this reflects even more that the upper income groups command more disposable income per person. Second, women in the lower quintiles are more likely to be without a partner in both the United States and Germany whereas men exhibit similar rates of partnerships across the quintiles. Hence, single women are numerous among low-income households and even more numerous among women in low-income households who lose their jobs. The fact that women in the lowest quintile have about the same household sizes as women in the other quintiles indicates that many of those women are single parents. The finding that low-income women often live in single households while most low-income men have partners contrasts the expectations stated in the theory section about social status and living in a couple

10 Clearly, years of education is not an optimal indicator for education in Germany because of the segregated educational system. However, the results are similar using different indicators. Hence, I employ this indicator for comparative reasons.

household. When using household income as an indicator for social stratification, low-status women are more likely to live in single-adult households than men. Many of these women are single mothers who have high poverty risks because of difficulties in reconciling work and family (Brady & Burroway, 2012). Furthermore, single women’s low income is often the result of divorce or break-up (Andreß, Borgloh, Bröckel, Giesselmann, & Hummelsheim, 2006; Radenacker, 2011). Additionally, education differs between the quintiles. As expected, average years of education are higher among high-income households. Also, the individuals’ average education does not seem to differ much from their partners’ average education. Hence, in all four income groups there is a high degree of educational homogamy. Therefore, as expected, income quintiles do not only capture individual resources acquired over the life course but also resources within the household. To analyze the impact these differences have on household income after job loss, I now turn to the next step in the analysis.

4.2. Income losses Fig. 1 presents the matched difference-in-difference estimates for income changes after job loss for different parts of the income distribution. Each panel depicts the quintile-specific income changes for one income category. I observe each episode for four years after job loss. In the graphs, I label the respective years as t0 for the year of job loss and t+2/4 for two and four years after job loss. Additional estimates not presented here show that the differences between the quintiles remain similar if I analyze longer periods. Note that I consider the trajectories of all individuals after job loss regardless of their employment status. Unlike other analyses of wage scars that only include those who returned to the labor market, my approach provides a better impression of the income situation of those affected by the trigger event. Fig. 1 shows that there is a cross-national difference in quintile-specific earnings changes as expected. In the United States, earnings losses are slightly higher in the lower quintile in the year of job loss. In the years thereafter however, the losses in the upper quintiles become larger relative to the lower quintiles. This is especially visible in comparison to the second quintile. Losses in the first quintile however remain large. A closer look at the data reveals that this is mainly due to strong income growth in the control group: people in the first quintile who lose their jobs miss out on the opportunities for income growth. In Germany, on the other hand, people

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

95

United States − Men

1st 2nd 3rd 4th/5th

t0

1st 2nd 3rd 4th/5th

1st 2nd 3rd 4th/5th

t+2

Estimated Changes (%) −60 −40 −20 0

Pre−Government Income

Estimated Changes (%) −60 −40 −20 0

Individual Earnings

1st 2nd 3rd 4th/5th

t+4

t0

1st 2nd 3rd 4th/5th

t+2

1st 2nd 3rd 4th/5th

t+4

Estimated Changes (%) −40 −30 −20 −10 0

Post−Government Income

1st 2nd 3rd 4th/5th

t0

1st 2nd 3rd 4th/5th

1st 2nd 3rd 4th/5th

t+2

t+4

Germany − Men

1st 2nd 3rd 4th/5th

t0

1st 2nd 3rd 4th/5th

t+2

1st 2nd 3rd 4th/5th

t+4

Estimated Changes (%) −60 −40 −20 0

Pre−Government Income

Estimated Changes (%) −60 −40 −20 0

Individual Earnings

1st 2nd 3rd 4th/5th

t0

Estimated Changes (%) −40 −30 −20 −10 0

Post−Government Income

1st 2nd 3rd 4th/5th

t0

1st 2nd 3rd 4th/5th

t+2

1st 2nd 3rd 4th/5th

t+4

Fig. 1. Estimated income trajectories after job loss, men. Sources: CNEF, PSID, and GSOEP, author’s calculations.

1st 2nd 3rd 4th/5th

t+2

1st 2nd 3rd 4th/5th

t+4

96

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

from the upper quartiles have lower earnings losses than the lower quintiles both in the short and in the long run. These observations are partly in line with Hypothesis 1. The labor market structure in the United States apparently leads to long-lasting earnings losses for those who had high incomes because their skills are mostly firm specific and many workers have to start over again in the new firm’s internal labor markets. This is at least true in comparison between the second quintile and those in the upper quintiles. In the first quintile however, losses remain high. Apparently, there are different dynamics in this group. In Germany on the other hand, high-income workers are able to return to well-paid positions after unemployment because their skills are often transferable between firms. However, If Hypothesis 1 is true, I should observe this pattern for all strata in Germany. Yet, those coming from low-income households have lower wage growth. This is mainly due to longer unemployment durations in the lower quintiles. In this group, many unemployed are lowly educated and consequently have lower re-employment probabilities. Hypothesis 1a however can not be confirmed. In the theoretical section, I argued that people coming from the upper quartiles should have higher initial wage losses because they have longer job search durations until they find a suitable job. Yet, their initial losses are even smaller than the other quintile’s losses. Presumably, people coming from the upper quintiles also receive more job offers and can pick a good match while many people from the lower quintiles have worse labor market prospects. Next, I turn to the actual life situation of the affected households by considering post-government income in Fig. 1. The estimates show that low-income men in the United States fare worst both in the short and in the long term. In Germany, on the other hand, men in the lowest and in the highest quintiles have low losses while the middle quintiles face high losses. Thus, in the United States, job loss has the worst effect on those who are already disadvantaged. In Germany, on the other hand, the middle quintiles face the greatest losses whereas high and low-income men are better off. Thus, postgovernment household income changes mirror earnings changes to some extent. Yet, in the United States, the losses of the lowest quintile increase relative to the other groups. In Germany, on the other hand, the lowest quintile has a better relative position in household income than in individual earnings. Apparently, the family and the welfare state influence the stratification of the outcomes in different ways. To analyze this influence, I now try to isolate these effects. To gauge the effects of the family and the welfare state, I follow an approach proposed by DiPrete

and McManus (2000a). They calculate the difference between losses in household income before and after taxes and transfers to measure the impact of the welfare state.11 I extend this approach to include the impact of the family by computing the difference between losses in individual earnings and pre-government household income. Formally, the two effects are defined as follows:  Family effect = δLE − δPrG where δX signifies the estimated changes from the models of labor earnings (LE), pre-government household income (PrG), and post-government household income (PoG). The welfare state effect is likewise calculated using the estimated changes from the models of pregovernment household income and post-government household income:  Welfare state effect = δPrG − δPoG Note that these indicators must be interpreted while keeping in mind the actual losses. Hence, they do not measure the potential effects of the welfare state and the family but the actual effects given the circumstances. For example, if losses in labor earnings are zero, because of rapid re-employment and a well paid new job, the measurement of the effects becomes impossible. Fig. 2 presents family and welfare state buffers for men in different income quintiles and at different points after job loss. Note that the magnitude of the buffers as presented here cannot be compared between the countries because income losses before taxes and transfers are higher in Germany. Hence, the estimates for the buffering effect are higher simply because there is more to buffer. Analyses correcting for this yield that the family buffer is more important in the United States whereas the welfare state buffer is more important in Germany after job loss (DiPrete & McManus, 2000a; Ehlert, 2012). Yet, in order to compare the effect within a country, the raw difference between the income losses is better suited. The results for the family buffer in Fig. 2 confirm hypothesis 2 that the family has a lower impact on the incomes if men in the lower strata than on men in the upper strata the United States. Note that the buffering effects are measured relative to former income. Hence, in absolute numbers, the upper quintiles’ advantage is 11 DiPrete and McManus (2000a) calculated the effect as a fraction of total income loss to make the effect comparable. However, this method has drawbacks once total losses become small. In this case, very small differences in the estimates can lead to substantial variation in the buffering effects.

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

Family Buffer (%) 5 10 15

15 10

0

5 0

Family Buffer (%)

20

Germany − Men

20

United States − Men

1st

2nd

3rd 4th/5th

1st

t0

2nd

3rd 4th/5th

1st

t+2

2nd

3rd 4th/5th

1st

t+4

2nd

3rd 4th/5th

1st

t0

2nd

3rd 4th/5th

1st

t+2

2nd

3rd 4th/5th

t+4

0

0

10

20

30

Welfare State Buffer (%) 10 20 30

40

Germany − Men

40

United States − Men Welfare State Buffer (%)

97

1st

2nd

t0

3rd 4th/5th

1st

2nd

3rd 4th/5th

t+2

1st

2nd

3rd 4th/5th

t+4

1st

2nd

t0

3rd 4th/5th

1st

2nd

3rd 4th/5th

1st

t+2

2nd

3rd 4th/5th

t+4

Fig. 2. Family and welfare state buffers at different parts of the household income distribution for men. Sources: CNEF, PSID, and GSOEP, author’s calculations.

even greater than measured with the buffering effects. In Germany, however, this is only true in the year of job loss. Two and four years afterward, the family buffer for German men in the lowest quintiles becomes more important. Yet, still in the middle quintiles the family buffer remains more important than in the first quintile. In the descriptives presented in Table 2, I show that men in the lower quintiles often have partners and consequently potential access to the family buffer. Further analyses not presented here reveal that most of these partners are inactive prior to job loss in the lowest quintile. Apparently couples following the traditional male-breadwinner norm are concentrated in the lower part of the income distribution. However, these women enter the labor market once their partners become unemployed. Yet, it takes some time before the inactive women in low-income households find a job and contribute to household income. Clearly, they contribute to income buffering even with little earnings because total household income losses are small in real terms. On the other hand, German women in high-income households are maybe not able to increase their hours further in their current job and hence to provide additional income.

As hypothesis 3 predicted, the effects of the German welfare state differ much more by quintile than in the United States because German institutions provide a floor effect. In the year of job loss, the German buffer is about 15 percentage points higher in the first quintile than in the fourth and fifth quintile, whereas in the United States these groups differ only by about 5 percentage points. This pattern also exists two and four years after job loss in Germany. In the United States, the only salient effect two years after job loss appears for the first quintile. This can be explained by the EITC, which offers tax refunds for low-income households if they have labor income. Thus, the stratified welfare state and partners’ earnings potentials explain why German men in the lowest quintile fare better in the long run than their earnings trajectories would suggest. In the United States on the other hand, the income buffers do not change the pattern observed for earnings. Instead, they even polarize household income losses: the lack of an income floor in the welfare state and the low family support even amplify the differences in earnings losses and make lowincome men worse off than high-income men. Thus, in the United States, job loss is part of cumulative

98

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

disadvantage for low income groups. In Germany, on the other hand, the welfare state and the family prevent such a trend. Instead, those in the middle of the social stratification lose most household income after job loss because they have worse labor market chances and do not profit from the floor effect. Next, I analyze women’s incomes after job loss. Fig. 3 shows that women’s individual earnings trajectories in the United States do not follow the same pattern observed for men. Women from low-income households have the highest losses in earnings both in the short and the long run. Women from high-income households on the other hand have low losses and recover rapidly. One explanation for this difference between genders in the United States could be the high proportion of single mothers in the lowest quintile. Presumably, these women face more difficulties securing a job that enables them to reconcile work and family after unemployment. In Germany, women show the same trend: high-income women recover faster than low-income women. Thus, earnings trends are similar in Germany for the two genders and thus partly in line with hypothesis 1. However, again, high-income women seem to profit from the labor market structure whereas low-income women face reemployment difficulties. The fact that the many women in this group are single mothers presumably adds to the observed labor market disadvantages. The pre and post-government trajectories in Fig. 3 depict that American women in low-income households lose more after job loss than their high-income counterparts. In pre-government household income, such a pattern is also visible in Germany. Yet it disappears in post-government household income losses. Thus, lowincome households in the United States are much more affected by women’s unemployment than high-income households. This mirrors the high share of single-adult households in the lowest quintile and the high dependency on two incomes of couple households in this quintile. In Germany, where this is also visible at a lower level, the welfare state seems to buffer this disadvantage of the lowest quintile. However, four years after job loss, women’s post-government income losses in the lowest quintile even grow while earnings and pre-government income stagnate on a low level. Apparently, these women face difficulties returning to the labor market. In the long run, as welfare state benefits become lower, this also affects post-government income. To delve deeper into the effect of the family and the welfare state, I again turn to the buffering effects. Casual inspection of the family buffers for women in the lower panels of Fig. 4 reveals that the patterns found in the stratification of family income support are roughly

similar to men’s patterns in the year of job loss. In both countries, the family has a smaller impact for women in the lowest quintile. In the long run however, the magnitude of the family buffer for women becomes larger for the lower quintiles in both countries. A closer look into the data shows that some women in the first quintile seem to move in with a partner after job loss. Although I already included household size in the models, there is apparently an effect that goes beyond adding a partner. Separating the models into those with and without a working partner accounts for a part of the effect (estimates not presented here). However, even those without a working partner have a substantial family buffer. Apparently, these women are able to acquire non-labor income such as private transfers from family or former partners outside the household. The welfare state buffer for women in Fig. 4 is more similar in the two countries in the year of job loss than the welfare state buffer for men. Both countries show a strong stratification. One reason for the higher stratification in the effect for American women compared to American men is that the United States had a minimum income program targeted at families with children called AFDC until 1996. Later this program has been replaced by the less generous TANF. Yet, some single mothers profited from the safety net that AFDC provided. In the long run however, there is almost no welfare state effect in the United States, whereas in Germany, I still find the impact of the welfare state four years after job loss. This analysis proves that women’s household income trajectories after job losses are more complex than men’s. This is mainly due to the variety of roles women have in households — from sole income providers to housewives. These roles differ within and across social strata. While women as sole providers of income are often concentrated in the lower part of the income distribution, dual earner couples and male breadwinner households can be found in all strata. Their occurrence depends on gender norms as well as negotiations within the household. Men on the other hand almost always work full time and are mostly not influenced by their partner’s labor market behavior (Blossfeld & Drobniˇc, 2001). Future analyses of household income trajectories should therefore analyze women’s labor force participation in greater detail. 5. Discussion The aim of this article is to analyze household income losses after job loss for different parts of the social stratification. Previous research found that the impact of job loss on household income varies cross-nationally

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

99

1st 2nd 3rd 4th/5th

t0

1st 2nd 3rd 4th/5th

t+2

1st 2nd 3rd 4th/5th

Estimated Changes (%) −60 −40 −20 0

Estimated Changes (%) −80−60−40−20 0 20 40

United States − Women Individual Earnings Pre−Government Income

t+4

1st 2nd 3rd 4th/5th

t0

1st 2nd 3rd 4th/5th

t+2

1st 2nd 3rd 4th/5th

t+4

Estimated Changes (%) −30 −20 −10 0 10

Post−Government Income

1st 2nd 3rd 4th/5th

t0

1st 2nd 3rd 4th/5th

t+2

1st 2nd 3rd 4th/5th

t+4

1st 2nd 3rd 4th/5th

t0

1st 2nd 3rd 4th/5th

t+2

1st 2nd 3rd 4th/5th

t+4

Estimated Changes (%) −60 −40 −20 0

Estimated Changes (%) −80−60−40−20 0 20 40

Germany − Women Individual Earnings Pre−Government Income

1st 2nd 3rd 4th/5th

t0

Estimated Changes (%) −30 −20 −10 0 10

Post−Government Income

1st 2nd 3rd 4th/5th

t0

1st 2nd 3rd 4th/5th

t+2

1st 2nd 3rd 4th/5th

t+4

Fig. 3. Estimated income trajectories after job loss, women. Sources: CNEF, PSID, and GSOEP, author’s calculations.

1st 2nd 3rd 4th/5th

t+2

1st 2nd 3rd 4th/5th

t+4

100

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

30 20

Family Buffer (%)

0

10

30 20 10 0

Family Buffer (%)

40

Germany − Women

40

United States − Women

1st

2nd

3rd 4th/5th

1st

t0

2nd

3rd 4th/5th

1st

t+2

2nd

3rd 4th/5th

1st

t+4

2nd

1st

2nd

3rd 4th/5th

1st

t+2

2nd

3rd 4th/5th

t+4

20 10 0

0

10

20

Welfare State Buffer (%)

30

Germany − Women

30

United States − Women Welfare State Buffer (%)

3rd 4th/5th

t0

1st

2nd

t0

3rd 4th/5th

1st

2nd

3rd 4th/5th

t+2

1st

2nd

3rd 4th/5th

t+4

1st

2nd

t0

3rd 4th/5th

1st

2nd

3rd 4th/5th

t+2

1st

2nd

3rd 4th/5th

t+4

Fig. 4. Estimated welfare state and family buffers at different parts of the household income distribution for women. Sources: CNEF, PSID, and GSOEP, author’s calculations.

with the configurations of the labor market, the family and the welfare state (DiPrete, 2002). I extend this research by arguing that the influence of these three factors varies between social strata because access to the income buffering mechanisms that the market, the family and the welfare state provide depends on the individuals’ positions within social stratification. To test this, I compare income losses after involuntary job loss in two countries: the United States and Germany. To measure stratification, I use household income quintiles. The empirical results back my presumptions: social strata differ in income losses relative to their prior incomes in both countries. Moreover, these differences follow country specific patterns. In the United States, job loss causes the largest losses in the poorest quintile whereas in Germany, the middle quintiles lose most after job loss. The analysis shows that this result emerged from differences in the labor market, the family and the welfare state. However, the impact of the labor market turned out to be different than anticipated in the theoretical considerations. I expected earnings losses to be high for people with high income in the United States because workers’ skills in the American labor market are often

firm specific and job entrants consequently begin on a lower level compared to their previous job. In Germany on the other hand, portable skills and longer job search durations through long lasting unemployment benefits should facilitate re-entry on a similar level, which should decrease losses for all strata in the long run (DiPrete et al., 1997; Marsden, 1990). It turned out however, that people in the poorest quintile have the highest earnings losses in the United States. Likewise in Germany, the lower two quintiles face higher losses than the upper quintiles. In both countries, this group faces difficulties finding a new job and also misses out on income growth compared to those without job loss. In the upper quintiles however, my expectations receive some support: Americans coming from high income positions have persistent long-term losses while their German counterparts show an upward income trend in the long run. The empirical results confirmed the expectations about the impact of the family and the welfare state. The impact of family income support proved to be higher among the upper strata than among the lower strata in both countries. This is obviously detrimental for poorer households who already face long term losses in

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

earnings. Hence, in household income before taxes and transfers, the differences between the strata even grow. However, in Germany, the welfare state offsets this effect by providing relatively more support to low-income households compared to high-income households. Also, in Germany, low-income men who become unemployed often profit from the labor market entry of their inactive partners. In the United States on the other hand, the impact of the welfare state is more equal between the strata and hence does not smooth the inequalities generated by the market and the family. My analysis also reveals gender differences: In the lower income quintiles, many women live in single-adult households, especially in the United States. Although the American welfare state buffers income losses more for low-income women than for low-income men, job loss also causes a strong reduction in household income for women in the poorest quintile. In Germany, on the other hand, the welfare state buffers much of the initial losses in the poorest quintile, but these women face severe long term consequences because they are not able to increase their incomes on the labor market. The analysis also shows that women’s income trajectories after job loss are more complex than men’s. This is presumably due to the differences in labor market attachment among women that do not always correlate with household income quintiles. As mentioned in the introduction, literature on income losses after job loss in different income percentiles found that people with high income lose more than those with low income (Burda & Mertens, 2001; DiPrete & McManus, 2000b). My estimation strategy differs from both previous results. Departing from DiPrete and McManus (2000b), I apply a difference in difference design that takes into account income growth in the control group. Households in the poorest quintile have a high chance of income growth if they are not affected by job loss. Those with job loss miss out on this growth and consequently have higher losses in the long run compared to the control group. Burda and Mertens (2001) on the other hand only considered people who returned to a job and left out long-term unemployed. Since this group is concentrated in the lower strata, their sample of low-income workers is selective. In sum, American women and men who are already deprived lose more of their former income through job loss than those in the upper strata. Therefore, job loss adds to the cumulation of disadvantages for them. In Germany, on the other hand, the welfare state impedes such a trend. The floor effect through a universal minimum income prevents these households from falling deeper. Thus, I can not reproduce the finding

101

by Vandecasteele (2011) that job loss reduces inequalities because the effect of job loss is more severe for the upper classes than for the lower classes. Obviously this is partly due to a different measurement. Yet, in addition to that, my results show that the effect of job loss on different strata varies cross-nationally. Future research should investigate whether the patterns found here can be reproduced comparing more countries. Acknowledgements This article has its origins in the research project “The Economic Consequences of Key Life Risks in Germany and the US and Their Evolution since the 1980s”, which was supported by Deutsche Forschungsgemeinschaft (DFG) (grant no. KO 2239/2). I thank the project members Jens Alber, Britta Grell, Jan Paul Heisig, Ulrich Kohler, Anke Radenacker, and Markus Wörz for their collaboration. I also thank Merlin Schaeffer, Martin Schröder, two anonymous referees and the participants of the SIMLife conference for helpful comments. The data used were kindly provided by the Deutsches Institut für Wirtschaftsforschung, Berlin (GSOEP); the Institute for Social Research, Survey Research Center, University of Michigan, Ann Arbor, Michigan (PSID); and the Department of Policy Analysis and Management, Cornell University, Ithaca, New York (CNEF). References Allmendinger, J. (1989). Educational systems and labor market outcomes. European Sociological Review, 5(3), 231–250. Allmendinger, J., & Hinz, T. (1998). Occupational careers under different welfare regimes: West Germany, Great Britain and Sweden. In L. Leisering, & R. Walker (Eds.), The dynamics of modern society (pp. 63–84). Bristol: Policy Press. Altonji, J. G., & Paxson, C. H. (1992). Labor supply, hours constraints, and job mobility. The Journal of Human Resources, 27(2), 256–278. Andreß, H.-J., Borgloh, B., Bröckel, M., Giesselmann, M., & Hummelsheim, D. (2006). The economic consequences of partnership dissolution—A comparative analysis of panel studies from Belgium, Germany, Great Britain, Italy, and Sweden. European Sociological Review, 22(5), 533–560. Arulampalam, W. (2001). Is unemployment really scarring? Effects of unemployment experiences on wages. The Economic Journal, 111(475), 585–606. Bane, M. J., & Ellwood, D. T. (1986). Slipping into and out of poverty: The dynamics of spells. The Journal of Human Resources, 21(1), 1–23. Beck, U. (1986). Risikogesellschaft. Auf dem Weg in eine andere Moderne. Frankfurt/Main: Suhrkamp Verlag. Becker, G. S. (1975). Human capital. A theoretical and empirical analysis, with special reference to education. New York, KU 48218: Columbia University Press.

102

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103

Blossfeld, H.-P. (2009). Educational assortative marriage in comparative perspective. Annual Review of Sociology, 35(1), 513–530. Blossfeld, H.-P., & Drobniˇc, S. (2001). Careers of couples in contemporary societies: From male breadwinner to dual-earner families. Oxford: Oxford University Press. Blossfeld, H.-P., & Timm, A. (2003). Who marries whom in West Germany. In H.-P. Blossfeld, & A. Timm (Eds.), Who marries whom?: Educational systems as marriage markets in modern societies. Dordrecht: Kluwer Academic Publishers, pp. 19–36. Brady, D., & Burroway, R. (2012). Targeting, universalism, and single-mother poverty: A multilevel analysis across 18 affluent democracies. Demography, 49(2), 719–746. Burda, M. C., & Mertens, A. (2001). Estimating wage losses of displaced workers in Germany. Labour Economics, 8(1), 15–41. Burkhauser, R. V., & Duncan, G. J. (1989). Economic risks of gender roles: Income loss and life events over the life course. Social Science Quarterly, 70(1), 3–23. Clark, A., Georgellis, Y., & Sanfey, P. (2001). Scarring: The psychological impact of past unemployment. Economica, 68(270), 221–241. Dieckhoff, M. (2011). The effect of unemployment on subsequent job quality in Europe: A comparative study of four countries. Acta Sociologica, 54(3), 233–249. DiPrete, T. A. (2002). Life course risks, mobility regimes, and mobility consequences: A comparison of Sweden, Germany, and the United States. American Journal of Sociology, 108(2), 267–309. DiPrete, T. A. (2003). Do cross-national differences in household standard of living mobility parallel cross-national differences in occupational mobility? A comparison of Germany, Sweden and the USA. Current Sociology, 51(5), 483–498. DiPrete, T. A., De Graaf, P. M., Luijkx, R., Tåhlin, M., & Blossfeld, H.-P. (1997). Collectivist versus individualist mobility regimes? Structural change and job mobility in four countries. American Journal of Sociology, 103(2), 318–358. DiPrete, T. A., & Eirich, G. M. (2006). Cumulative advantage as a mechanism for inequality: A review of theoretical and empirical developments. Annual Review of Sociology, 32(1), 271–297. DiPrete, T. A., & McManus, P. A. (2000a). Family change, employment transitions, and the welfare state: Household income dynamics in the United States and Germany. American Sociological Review, 65(3), 343–370. DiPrete, T. A., & McManus, P. A. (2000b). The welfare state, the labor market, and the stability of household income: A comparison of the US and Germany. EurAmerica, 30(4), 1–46. Ehlert, M. (2012). Buffering income loss due to unemployment: Family and welfare state influences on income after job loss in the United States and Western Germany. Social Science Research, 41(4), 843–860. Esping-Andersen, G. (1990). The three worlds of welfare capitalism. Cambridge: Polity Press. Estevez-Abe, M., Iversen, T., & Soskice, D. (2001). Social protection and the formation of skills: A reinterpretation of the welfare state. In P. A. Hall, & D. W. Soskice (Eds.), Varieties of capitalism: The institutional foundations of comparative advantage (pp. 145–183). Oxford: Oxford University Press. Frick, J. R., Jenkins, S. P., Lillard, D. R., Lipps, O., & Wooden, M. (2007). The cross-national equivalence file (CNEF) and its member country household panel studies. Schmollers Jahrbuch, 127, 627–654. Fritzell, J. (1990). The dynamics of income distribution: Economic mobility in Sweden in comparison with the United States. Social Science Research, 19(1), 17–46.

Gangl, M. (2003). Unemployment dynamics in the United States and West Germany. Heidelberg: Physica. Gangl, M. (2004). Welfare states and the scar effects of unemployment: A comparative analysis of the United States and West Germany. American Journal of Sociology, 109(6), 1319–1364. Gangl, M. (2006). Scar effects of unemployment: An assessment of institutional complementarities. American Sociological Review, 71(6), 986–1013. Giesecke, J., & Heisig, J. P. (2010). Destabilisierung und destandardisierung, aber für wen? die entwicklung der westdeutschen arbeitsplatzmobilität seit 1984. KZfSS Kölner Zeitschrift für Soziologie und Sozialpsychologie, 62(3), 403–435. Goldthorpe, J. H. (2010). Analysing social inequality: A critique of two recent contributions from economics and epidemiology. European Sociological Review, 26(6), 731–744. Gottschalk, P., & Moffitt, R. (1999). Changes in job instability and insecurity using monthly survey data. Journal of Labor Economics, 17(S4), S91–S126. Grell, B. (2011). Unemployment compensation in the U.S.: Provisions and institutional changes since the 1980. WZB discussion paper SP I 2011-202, WZB, Berlin. Grunow, D., Hofmeister, H., & Buchholz, S. (2006). Late 20thcentury persistence and decline of the female homemaker in Germany and the United States. International Sociology, 21(1), 101–131. Iacus, S. M., King, G., & Porro, G. (2012). Causal inference without balance checking: Coarsened exact matching. Political Analysis, 20(1), 1–24. Keys, B. J., & Danziger, S. (2008). Hurt the worst: The risk of unemployment among disadvantaged and advantaged male workers, 1968–2003. In K. S. Newman (Ed.), Laid off, laid low: Political and economic consequences of employment insecurity (pp. 56–73). New York: Columbia University Press. King, G., Nielsen, R., Coberley, C., Pope, J. E., & Wells, A. (2011). Comparative effectiveness of matching methods for causal inference. Unpublished working paper. http://gking.harvard.edu/gking/files/psparadox.pdf Kohler, U., Ehlert, M., Grell, B., Heisig, J. P., Radenacker, A., & Wörz, M. (2012). Verarmungsrisiken nach kritischen Lebensereignissen in Deutschland und den USA. KZfSS Kölner Zeitschrift für Soziologie und Sozialpsychologie, 64(2), 223–245. Korenman, S., & Neumark, D. (1991). Does marriage really make men more productive? The Journal of Human Resources, 26(2), 282–307. Layte, R., & Whelan, C. (2003). Moving in and out of poverty. European Societies, 5(2), 167. Leisering, L. (2003). Government and the life course. In J. T. Mortimer, & M. J. Shanahan (Eds.), Handbook of the life course (pp. 205–225). New York: Kluwer Academic/Plenum Publishers. Leisering, L., & Leibfried, S. (1999). Time and poverty in Western Welfare States: United Germany in perspective. Cambridge: Cambridge University Press. Lundberg, S. (1985). The added worker effect. Journal of Labor Economics, 3(1), 11–37. Marsden, D. (1990). Institutions and labour market mobility: Occupational and internal labour markets in Britain, France, Italy and West Germany. In R. Brunetta, & C. Dell’Aringa (Eds.), Labour relations and economic performance (pp. 414–438). Basingstoke: Macmillan.

M. Ehlert / Research in Social Stratification and Mobility 32 (2013) 85–103 Mayer, K. U. (1991). Soziale Ungleichheit und die Differenzierung von Lebensverläufen. In W. Zapf (Ed.), Die Modernisierung moderner Gesellschaften. Verhandlungen des 25. Deutschen Soziologentages in Frankfurt am Main 1990 (pp. 667–687). Frankfurt/Main: Campus Verlag. Mayer, K. U. (1997). Notes on the comparative political economy of life courses. In L. Mjøset, F. Engelstad, G. Brochmann, R. Kalleberg, & A. Leira (Eds.), Methodological issues in comparative social science. Vol. 16 of comparative social research (pp. 203–226). Greenwich: JAI Press. Mayer, K. U. (2004). Whose lives? How history, societies, and institutions define and shape life courses. Research in Human Development, 1(3), 161–187. Mayer, K. U. (2009). New directions in life course research. Annual Review of Sociology, 1(3), 413. Mayer, K. U., & Carroll, G. R. (1987). Jobs and classes: Structural constraints on career mobility. European Sociological Review, 3(1), 14–38. Mayer, K. U., & Müller, W. (1986). The state and the structure of the life course. In A. B. Sørensen, F. E. Weinert, & L. R. Sherrod (Eds.), Human development and the life course: Multidisciplinary perspectives (pp. 217–245). Hillsdale: Lawrence Erlbaum Associates. McGinnity, F. (2002). The labour-force participation of the wives of unemployed men: Comparing Britain and West Germany using longitudinal data. European Sociological Review, 18(4), 473–488. McGinnity, F., & Hillmert, S. (2004). Persisting class inequality? European Societies, 6(3), 383–408. McManus, P. A., & DiPrete, T. A. (2000). Market, family, and state sources of income instability in Germany and the United States. Social Science Research, 29(3), 405–440. Mortensen, D. T., & Pissarides, C. A. (1999). New developments in models of search in the labor market. In O. Ashenfelter, & D. Card (Eds.), Handbook of Labor Economics. Vol. 3 (pp. 2567–2627). Amsterdam: Elsevier. OECD (2007). Babies and Bosses—Reconciling work and family life: A synthesis of findings for OECD countries. Paris: OECD.

103

OECD (2011). OECD family database. http://www.oecd.org/ document/4/0,3746,en 2649 34819 37836996 1 1 1 1,00.html. http://www.oecd.org/document/4/0,3746,en 2649 34819 37836996 1 1 1 1,00.html Pfeffer, F. T. (2011). Status attainment and wealth in the United States and Germany. In T. M. Smeeding, M. Jännti, & R. Erikson (Eds.), Persistence, privilege, and parenting: The comparative study of intergenerational mobility. New York: Russell Sage Foundation. Pollmann-Schult, M. (2011). Marriage and earnings: Why do married men earn more than single men? European Sociological Review, 27(2), 147–163. Radenacker, A. (2011). Economic consequences of family break-up. income before and after family break-up of women in Germany and the United States. Schmollers Jahrbuch: Journal of Applied Social Science Studies, 131(2), 225–234. Reynolds, J., & Aletraris, L. (2010). Mostly mismatched with a chance of settling. Work and Occupations, 37(4), 476–511. Sørensen, A. B. (2000). Toward a sounder basis for class analysis. American Journal of Sociology, 105(6), 1523–1558. Spence, M. (1973). Job market signaling. The Quarterly Journal of Economics, 87(3), 355–374. Vandecasteele, L. (2011). Life course risks or cumulative disadvantage? The structuring effect of social stratification determinants and life course events on poverty transitions in Europe. European Sociological Review, 27(2), 246–263. Western, B., Bloome, D., Sosnaud, B., & Tach, L. (2012). Economic insecurity and social stratification. Annual Review of Sociology, 38(1) Worts, D., Sacker, A., & McDonough, P. (2010). Falling short of the promise: Poverty vulnerability in the United States and Britain, 1993–2003. American Journal of Sociology, 116(1), 232–271. Wörz, M. (2011). Unemployment compensation in Germany. Provisions and institutional changes since the 1980. WZB discussion paper SP I 2011-206, WZB, Berlin.