2. Previous research

Job loss is a major life-course risk, which itself is influenced by a variety of other adult life-course risks, such as health shocks (e.g., Arrow, Reference Arrow1996) or marital disruption (e.g., Covizzi, Reference Covizzi2008). Moreover, predictors of unemployment can even be traced back to one's childhood, that is, to a stage of the individual's life-course years before he or she entered the workforce (e.g., Anyadike-Danes and McVicar, Reference Anyadike-Danes and McVicar2005; Caspi et al., Reference Caspi, Entner Wright, Moffit and Silva1998). Specifically, parental background factors (socio-economic status [SES] and family functioning) as well as childhood health and cognitive skills have been shown to impact future educational and labour market outcomes (e.g., Currie, Reference Currie2009; Elman and O’Rand, Reference Elman and O'Rand2004; Haas et al., Reference Haas, Glymour and Berkman2011).

These findings provide evidence for an early onset of cumulative (dis-)advantage processes, which are likely to be aggravated once the individual exits the educational system. Namely, next to a variety of immediate economic and non-economic consequences following a difficult transition from school to work (e.g., Jacob and Kleinert, Reference Jacob and Kleinert2008), cumulative (dis-)advantage theory also suggests potential long-run negative effects of an early failure in the labour market: ‘Previous unemployment arguably scars a worker and creates unemployment because it reduces a worker's human capital, reduces a worker's psychological readiness for work, and makes a person less attractive to prospective employers’ (DiPrete and Eirich, Reference DiPrete and Eirich2006: 287f.; also see Arulampalam et al., Reference Arulampalam, Gregg and Gregory2001).

Scars have been shown to be reflected in higher risks of future job loss among those with previous unemployment experience (‘occurrence dependence’; e.g., Doiron and Gørgens, Reference Doiron and Gørgens2008; Gregg, Reference Gregg2001). What is more, unemployment scars may also reduce the quality of jobs available to previously unemployed workers (e.g., Dieckhoff, Reference Dieckhoff2011; see Gangl, Reference Gangl2004, for an extensive discussion). A number of empirical studies suggest that unemployment depresses subsequent wages (e.g., Eliason and Storrie, Reference Eliason and Storrie2006; Gangl, Reference Gangl2006) and is associated with lower subsequent job tenure (e.g., Belzil, Reference Belzil2001; Böheim and Taylor, Reference Böheim and Taylor2002). The evidence on the role that the length of unemployment might play in future employment prospects (‘duration dependence’) is fairly ambiguous, though (e.g., Böheim and Taylor, Reference Böheim and Taylor2002; Doiron and Gørgens, Reference Doiron and Gørgens2008; Luijkx and Wolbers, Reference Luijkx and Wolbers2009). Moreover, while the number of unemployment spells seems to matter for the extent of earnings losses (Stevens, Reference Stevens1997), there appears to be no such effect on the likelihood of exiting subsequent employment (Luijkx and Wolbers, Reference Luijkx and Wolbers2009).

Previous research has also shown that, independent of individual life-course characteristics, men's and women's career paths vary widely across countries characterised by different educational systems, labour market institutions and – more generally – welfare state regimes (e.g., Buchholz et al., Reference Buchholz, Hofäcker, Mills, Blossfeld, Kurz and Hofmeister2009; Sackmann, Reference Sackmann2001). Parallel to a growing interest in the labour market for older workers (e.g., Blossfeld et al., Reference Blossfeld, Buchholz and Hofäcker2006; Szinovaczs, Reference Szinovaczs2011), recent years have witnessed a plethora of studies investigating cross-national variation in patterns of labour market entry, including youth unemployment (e.g., Breen, Reference Breen2005; Scherer, Reference Scherer2005; Wolbers, Reference Wolbers2007). One general finding is that unemployment among labour market entrants tends to be particularly high in Southern European countries, which are also characterised by potentially huge unused labour capacity in the older population (cf. Brugiavini et al., Reference Brugiavini, Croda, Mariuzzo and Börsch-Supan2005; also see Sackmann, Reference Sackmann2001).

Negative long-term consequences of poor macro-economic conditions at labour market entry have been observed in a number of studies (e.g., Kahn, Reference Kahn2010; Raaum and Røed, Reference Raaum and Røed2006). With regard to unemployment scarring, though, Lupi and Ordine (Reference Lupi and Ordine2002) argue that in a high youth unemployment environment, such as Southern Europe, individual unemployment experiences tend to be perceived as ‘normal’ and do not necessarily signal that the worker is of poor quality, as is the case in Northern Europe, where youth unemployment is less common (see Gibbons and Katz, Reference Gibbons and Katz1991, for related evidence from mass layoffs). Moreover, next to differences in macro-economic conditions in general, and the level of unemployment in a particular labour market in particular, the institutional regulation of job loss is likely to matter to the extent to which unemployment is associated with cumulative disadvantages (e.g., Dieckhoff, Reference Dieckhoff2011; see Gangl, Reference Gangl2006, for an elaborate theoretical framework). The effect of unemployment insurance (UI), for example, ‘goes significantly beyond preserving economic status through earnings-related benefits during unemployment spells, but rather presumes that the welfare state has a positive effect on workers’ capacity to recover from unemployment – and thus on preventing career scars and negatively cumulating career dynamics in the longer run’ (Gangl, Reference Gangl2004: 1323; italics in the original; also see Belzil, Reference Belzil2001). Further institutional features of labour markets which might affect scar effects are amongst others a country's employment protection legislation (EPL) (e.g., Gangl, Reference Gangl2006) or active labour market policy (ALMP), (e.g., Strandh and Nordlund, Reference Strandh and Nordlund2008) programmes.

Comparing such regulations and programmes cross-nationally reveals a pattern of institutional labour market regimes that follows ‘classic’ welfare state typologies (cf. Gangl, Reference Gangl2006; Tatsiramos, Reference Tatsiramos2009; also see Gallie and Paugam, Reference Gallie and Paugam2000). Accordingly, the countries covered by our study can be categorised as social-democratic (or universal), conservative and Mediterranean. Social-democratic regimes are generally characterised by a combination of flexible labour markets, strongly commodifying UI systems providing almost universal access to generous unemployment benefits and ALMPs investing significantly in worker retraining (see Elvander, Reference Elvander2002 for a more nuanced comparison of the Nordic labour market regimes). Conservative welfare states also have a tradition of strong UI, but are somewhat weaker in their ALMPs and much stronger in their regulation of labour markets through EPL than their social-democratic counterparts. Under the Mediterranean welfare regime, EPL is strongest whereas ALMPs are weaker and UI is generally less encompassing and less generous than elsewhere (but note that Spain resembles the ‘conservative’ group of countries in terms of UI generosity; cf. Tatsiramos, Reference Tatsiramos2009).

There is ample evidence demonstrating that these differences in labour market regimes correspond not only to differences in aggregate labour market outcomes, but also to differences in unemployment scarring. At the macro-level, higher employment rates (lower unemployment rates, respectively) have been shown to be positively associated with measures of policy generosity, especially ALMPs, and negatively associated with institutions and policies inducing labour market rigidities, such as strong EPLs (see, for example, Botero et al., Reference Botero, Djankov, La Porta, Lopez-de-Silanes and Shleifer2004; Bruno and Rovelli, Reference Bruno and Rovelli2010). At the micro-level, a beneficial effect of UI on subsequent employment stability has been observed in countries with relatively generous benefit systems (albeit at the cost of a prolonged period of unemployment; e.g., Gangl, Reference Gangl2004; Tatsiramos, Reference Tatsiramos2009). Moreover, Gangl (Reference Gangl2006: 1009) suggests institutional complementarities between UI and EPL, providing evidence that the ‘positive effects of UI systems [are] enhanced by relatively dynamic labour markets that provide ample job opportunities and permeable labour market boundaries’. Finally, ALMP programmes aiming to maintain or enhance unemployed workers’ human capital have been shown to increase individuals’ probability of reaching their pre-unemployment incomes and to reduce their risk of exiting the labour market (Strandh and Nordlund, Reference Strandh and Nordlund2008).

It is clearly beyond the scope of the present analysis to account for the complexities in the evolution, shifts and changing roles of specific labour market institutions in European welfare states after World War II (see Bonoli, Reference Bonoli, Clasen and Clegg2011; Emmenegger, Reference Emmenegger2011; Nickell et al., Reference Nickell, Nunziata and Ochel2005 for such accounts) and their potential role in shaping the effects of previous unemployment experiences on individuals’ subsequent labour market outcomes. The insights obtained from the studies reported above still lead us to hypothesise that Scandinavian-type, i.e., social-democratic, labour markets constitute the most favourable context for workers to recover from unemployment (cf. Gangl, Reference Gangl2006).

3. Data and method

This study uses data from SHARELIFE, the third wave of the Survey of Health, Ageing and Retirement in Europe (SHARE; a detailed description of methodological issues is given in Schröder, Reference Schröder2011; also see Börsch-Supan et al., Reference Börsch-Supan, Hank, Jürges, Schröder and Harkness2010, Börsch-Supan et al., Reference Börsch-Supan, Hank, Jürges, Schröder and Harkness2013). Individual life histories of respondents aged fifty or older, who had already participated in at least one of the previous SHARE waves, were collected between October 2008 and August 2009. Eleven countries are represented in our analysis: Austria, Belgium, Denmark, France, (West) Germany, Greece, Italy, the Netherlands, Spain, Sweden, and Switzerland (see Blom and Schröder, Reference Blom, Schröder and Schröder2011, for a detailed account of retention rates by country, age and gender). Respondents of formerly socialist countries (Czech Republic, Poland and East Germany) as well as individuals younger than age sixty at the time of the SHARELIFE interview are excluded, which results in an analytic sample of roughly 13,000 observations; see Table 1 for descriptive sample statistics.

TABLE 1.

Descriptive sample statistics

Notes: ^a Ranging from ‘none or very few (0–10 books)’ [1] to ‘enough to fill two or more bookcases’ [5]. ^b Ranging from ‘much better’ [1] to ‘much worse’ [5] in comparison to other children in the respondent's class; ^c Ranging from ‘poor’ [1] to ‘excellent’ [5].

Source: SHARELIFE. Authors’ calculations.

SHARELIFE provides basic information on the work histories of respondents who reported to have ever done any paid work, beginning with the year when they left full-time education until the date of the interview. The definition of unemployment in the data rests solely on the individual's self-reported activity status (‘Which of these best describes your situation?’), irrespective of whether searching for a job and independent of receipt of unemployment benefits. We might therefore provide a conservative measure of unemployment occurrence, particularly among individuals with a low labour force attachment. Women with younger children living in a traditional ‘male breadwinner’ social context, for example, have been suggested to report periods of unemployment solely as having been a homemaker (see Jürges, Reference Jürges2007). Moreover, SHARELIFE allows us to identify periods of unemployment only if the gap between two jobs (or between leaving full-time education and the respondent's first job, respectively) was longer than six months. Errors in the retrospective reporting of unemployment spells should thus be a relatively minor issue in our analysis, because recall error has been suggested to be a problem when shorter spells are considered (e.g., Manzoni et al., Reference Manzoni, Luijkx and Muffels2011).

We observe individuals across their entire working life. Our main variables of interest are three binary indicators of unemployment periods (none vs. any) during different major phases of the individual's career: early (that is, within the first three years after having left full-time education; cf. Luijkx and Wolbers, Reference Luijkx and Wolbers2009), prime-age working years (between the end of ‘early’ and age forty-nine), and late (age fifty or over; cf. Blossfeld et al., Reference Blossfeld, Buchholz and Hofäcker2006). We use these variables to perform separate analyses of unemployment risks in each career phase. To ensure that an individual had been at risk of becoming unemployed at all, she must have reported at least one employment spell in her early, prime-age, and/or late career, respectively, in order to be included in the analytic sample for a particular phase (resulting in the different sample sizes displayed in Table 1). Note that this restriction leads to a substantial decline in the proportion of women in the ‘late career’ sample. Information on individuals’ previous experience of unemployment enters the models for unemployment during prime-age and later years as an explanatory variable. We interpret significant associations of unemployment in t–1 with the risk of experiencing a period of unemployment in phase t of one's career as indication of scarring effects.Footnote ¹

All control variables refer to measures which were determined before the individual first entered the labour market. That is, we estimate the total scarring effect excluding potentially confounding factors which themselves might have been affected by earlier experiences of unemployment, such as health, parenthood or marital status (e.g., Hansen, Reference Hansen2005; Kreyenfeld, Reference Kreyenfeld2010; Schröder, Reference Schröder2013). Next to gender and years of education (derived from SHARE's wave 1 and 2) as basic socio-demographic variables, we focus on information regarding respondents’ childhood conditions at age ten: parents’ socio-economic status – operationalised by the number of rooms per person and the number of books in the household (ranging from ‘none or very few (0–10 books)’ [1] to ‘enough to fill two or more bookcases’ [5]), a composite measure of self-rated maths and language skills (ranging from ‘much better’ [1] to ‘much worse’ [5] in comparison to other children in the respondent's class), and self-rated health (ranging from ‘poor’ [1] to ‘excellent’ [5]); see Haas, Reference Haas2007, for an evaluation of the reliability of retrospective childhood health reports.

Finally, we roughly account for the historical period during which the SHARELIFE respondents graduated (pre-1950, 1950s, 1960 or later) as well as for the respective welfare state context, grouping the countries covered by our study into three clusters labelled as social-democratic (DK, SE), conservative (AT, BE, CH, DE_west, FR, NL), and Mediterranean (ES, GR, IT). Moreover, in a supplementary analysis, we draw on statistics provided by the International Labour Organization to take into consideration the labour market situation in each country at the time of individuals’ labour market entry. That is, for all cohorts who left education in 1929 or later, we account for the unemployment rate in the year when the respondent completed her full-time education. Unfortunately, we were not able to obtain youth unemployment rates, and even information on overall unemployment rates has not been available for all years in all countries, resulting in a significant decline of the sample for this part of the analysis.