THE EUROPEAN WORLD OF TEMPORARY EMPLOYMENT: Gendered and poor?

It is often said that being employed is the best strategy for an individual to prevent living in poverty and that assumption is indeed backed by a vast amount of research (e.g., Cantillon et al.2003; Atkinson et al.2005; OECD 2009). Less than a decade ago, being employed yet living in poverty was considered a marginal issue only relevant for the Anglo-Saxon countries (think about the rhetoric surrounding the so-called McJobs). Nevertheless, scholars and policymakers are nowadays increasingly worried about the incidence of in-work poverty in all European member states (Nolan and Marx 2000; Peña-Casas and Latta 2004; Andreβ and Lohmann 2008; Lohmann 2009; Brady et al.2010; Fraser et al.2011; Marx et al.2012). These concerns are paralleled by the growing emphasis on flexibility to help increase employment rates and render European labour markets more adaptable to structural changes and the emergence of new risks (Viebrock and Clasen 2009).

More flexibility on the labour market is often translated in nonstandard forms of employment like temporary employment, assumed to improve employment chances for weak labour market profiles on the one hand and to give employers the opportunity to adapt the number of employees to fluctuating levels of demand less costly on the other (Booth et al.2002; Debels 2008). However, previous research has shown that temporary employment can also have problematic consequences in terms of job security, income security, employer-provided social security benefits and on-the-job training (OECD 2002; Leschke and Watt 2008). If the growth and distribution of flexible employment is associated with negative consequences in terms of income security, this phenomenon could very well be related to the incidence of in-work poverty in Europe. However, research into the poverty risk associated with nonstandard employment relations is rather scarce (exceptions: Debels 2008; Amuedo-Dorantes and Serrano-Padial 2010).

Furthermore, some are concerned that the proliferation of flexible work as a strategy to raise employment rates will affect women and men differently and that, consequently, existing gender inequalities will be reinforced (Jepsen 2005; Hansen 2007). These worries are not ill-founded: women already have more flexible patterns of work and make transitions in and out employment more often than men, mainly due to their continuing responsibility for parental care and unpaid household work (Lewis 2006). One of the direct results of this unequal division in household work is lower pay associated with women's work. Given these circumstances, it is unclear how women will fare on the income security side when they are increasingly engaged in nonstandard work patterns. One of the key assumptions of the flexicurity literature is that being at work automatically leads to income security but the above-mentioned concerns about the working poor exemplify that this is not necessarily the case (Lewis and Plomien 2009).

Given the above, the main aim of this article is to explore the relationship between temporary employment and poverty in a European comparative perspective, including 24 countries. First, we scrutinize the poverty risk associated with temporary employment vis-à-vis permanent employment. In other words, to what extent differs temporary employment from permanent employment in terms of poverty risk? Second, the article will assess whether the ‘poverty risk profiles’ of the temporarily employed differ from the working poor in general. If this is the case, the implications for policies addressing in-work poverty and flexibility could be profound. Third, we will investigate cross-country variation in the association between temporary employment and poverty. Finally, we will look into the gender dimension of this issue and compare the poverty risk of temporarily employed women and men.

We draw data from the 2008 wave of the Survey on Living and Income Conditions (EU-SILC). We selected 24 countries: Austria, Belgium, Bulgaria, Czech Republic, Estonia, Germany, Spain, Finland, France, Greece, Hungary, Ireland, Italy, Lithuania, Latvia, The Netherlands, Norway, Poland, Portugal, Sweden, Slovenia, Slovakia, and United Kingdom. It has to be noted that we draw data from the 2007 wave for United Kingdom and France due to restrictions in the 2008 questionnaire. Luxemburg is excluded because the sample size is too small, Denmark because the data does not allow to identify a temporary employment variable for this country.

The pivotal concepts in this article are temporary employment and poverty. First, temporary employment is commonly seen as a form of nonstandard work (also referred to as atypical employment), which is then juxtaposed against the notion of ‘standard work’, i.e., permanent and fulltime employment (Kalleberg 2000). Here, we endorse the generally used definition of temporary employment as an ‘employment relationship with a limited duration’ referring to seasonal jobs, agency work, specific training contracts and fixed-term contracts. As such we are only concerned with salary and wage work, hence excluding self-employment. To employ a more exact definition would be rather tricky because of the different meaning attached to certain forms of temporary work among different countries (Campbell and Burgess 2001; Conley 2008). Some instances of temporary employment in a certain country do not always have an equivalent in other countries which obviously complicates comparing its outcomes (see the European Commission ESOPE report for further reading: Laparra 2004).

An additional problem with such broad definition of the concept is that the great diversity among the temporarily employed within countries is largely ignored: it could for instance comprise positions as diverse as a civil engineer on a training contract and a low skilled manual worker doing agency work. As Barbier argues, concepts with such different meanings in different circumstances cannot simply be compared because they share some similarities (i.e., they don't have the prospect of a long lasting relationship) but ‘have to be set against the “normative frameworks” of each particular society’ (Barbier 2004: 12) However, as much as one can agree with this critique, taking fully account of it would render any comparative exercise of outcomes rather meaningless. Furthermore, the dataset at hand does not allow to distinguish different appearances of temporary work. Consequently, we proceed with the broad definition of temporary employment and make abstraction of the ‘comparison problem’ by controlling for sectorial and occupational differences in the statistical analyses.

Second, following European practice, people are considered to be poor if they live in a household with an equivalent household income below 60 percent of the median equivalent household income in the country of residence.¹ All technicalities aside, the poverty concept employed here is a relative one, referring to the minimum minimorum considered necessary to avoid social exclusion in one's country of residence (Cantillon 2011).

There is a close association between forms of employment different from the standard model and precariousness (Vosko 2008) but not all forms of nonstandard work are precarious per se. The example of part-time work in countries such as the Netherlands is a case in point in this respect. Although associated with negative externalities like lower hourly wages and limited opportunities for advancement in the job, it is in particular a women's affair and for the overwhelming majority a deliberate choice to be able to combine work and family obligations (Eurofound 2007). Moreover, Dutch women repeatedly report to be satisfied with these atypical work arrangements (Bosch et al.2010). The same patterns hold for the majority of women in most European countries (OECD 2010). In contrast, more than 80 percent of the temporarily employed report to be involuntary engaged in this kind of contractual arrangements (European Commission 2002). Furthermore, temporary employment is almost always associated with precariousness, either measured by objective (Booth et al.2002; Giesecke and Groβ 2004) or subjective standards (Burgoon and Dekker 2010). Indeed, a large body of research associates temporary employment with lower job satisfaction (Petrongolo 2004), less access to fringe benefits (OECD 2002), reduced job security (Gash and McGinnity 2007), negative health effects (Virtanen et al.2005; Gash et al.2007), social fragmentation, isolation and mental problems (Erlinghagen 2008) and–especially–lower wages compared with permanent employment (Mertens et al.2007; Elia 2010).

However, some authors argue that the use of flexible working arrangements serves as a bridge to permanent employment for jobseekers (Atkinson et al.1996). Most research shows that the majority of temporary workers indeed move into permanent jobs within 2 years, although it varies depending on the type of temporary employment (e.g., seasonal workers have less chance to make the transition to permanent employment) and on individual characteristics (OECD 2002). For instance, the odds of finding permanent employment are lower (and the risk of entering non-employment is higher) for the less educated, older workers and workers who have already been unemployed (Debels 2008; Barbieri 2009). Nevertheless, even if temporary workers can genuinely regard their job as a stepping-stone to permanent employment, they can still be expected to suffer from the detrimental effects associated with these working arrangements and be, for instance, at risk of poverty at that particular moment (Burgoon and Dekker 2010). Given the disadvantages of temporary employment and regardless of mobility patterns, we expect it to be associated with a higher poverty risk than permanent employment (Hypothesis 1).

Not everyone is exposed to the risk of becoming temporarily employed to the same extent. Empirical studies have shown that lowly skilled (Giesecke and Groβ 2004), young (Baranowska and Gebel 2010) and female workers (Petrongolo 2004) are overrepresented in temporary employment patterns. Especially for the latter, the negative consequences of temporary employment could lead to a reinforcement of existing inequalities (because the overwhelming majority of women still face the burden of parental care and household work resulting in–inter alia–wage penalties). Therefore, we expect the poverty risk to be higher for temporary working women compared to their male counterparts (Hypothesis 2).

Previous research on the working poor showed that the variation in poverty rates for the working population (without differentiating between permanent and temporary employment) stems from a combined effect of welfare state policies, the role of the family and individual characteristics (Lohmann and Marx 2008; Marx and Verbist 2008). On the individual and the family level, several ‘risk profiles’, i.e., constellations of characteristics prone to working poverty, have been identified. First of all, in a context where double earnership (and the associated living standard) has become the norm, single earner households with dependent children are most affected by in-work poverty. Households with children entail greater needs and single earners are restricted in their available resources to fulfil those needs (Bardone and Guio 2005). Adding an extra income to the household's resources whittles down the poverty risk to a great extent, even if the extra income stems from low waged jobs. This is especially relevant in the case of temporary employment because these jobs tend to be more often low waged than permanent employment. Previous work has regarded the influence of low wage work on in-work poverty and showed that low wages an sich do not necessarily lead to working poverty, but the combination with other risk factors lead to the observation that ‘while most low-paid workers are not in poor households, most workers in poor households are themselves low paid’ (Nolan and Marx 2000: 105). Low pay is thus a factor that should be relevant especially in the case of temporary employment. Second, adding up to this ‘household effect’ is the finding that dual earner households consist disproportionally of couples where both partners have a higher level of education and, as a consequence, higher earnings. This educational homogamy has the opposite effect for the lower educated (Cantillon et al.2001). They tend to face an accumulation of disadvantages at the household level by combining higher risks of being unemployed with a higher risk of ending up with a low earning partner (Lohmann and Marx 2008). Finally, age is also a determining factor. In most European countries, young workers have a higher poverty risk because of the insider/outsider tendencies prevalent on the labour market (Bardone and Guio 2005). In the Southern countries, however, in-work poverty is more concentrated among older workers (Airio 2008).

Thus, age, education and the composition of the household are risk factors leading to poverty among the working population, especially when these risk-enhancing factors are accumulated at the household level. We expect that the same determinants to become working poor are at play in the case of temporary employment (Hypothesis 3).

These determinants are not exogenous but influenced by the institutional configuration of welfare states. To capture this influence on the relationship between temporary employment and in-work poverty in a straightforward way (it is not possible to discuss the institutional structure of 24 countries in detail), we group our countries in clusters based on their flexibility-security nexus (Muffels et al.2002) drawing on the ‘welfare regimes’ approach (Esping-Andersen 1990; Arts and Gelissen 2002), and the ‘employment regimes’ literature (Gallie 2007; Gautié and Schmitt 2010).

What can be expected from the interconnectedness between clusters of welfare states and working poverty in the case of temporary employment can be illustrated by briefly summarizing findings from previous research.

In the segmented labour market of the Southern countries (Portugal, Spain, Greece, and Italy) we expect high levels of in-work poverty among the temporarily employed. Standard workers are strongly protected at the expense of non-core workers (such as the temporarily employed) which makes it especially difficult for young adults to make a decent living from employment (Esping-Andersen 1999). Due to the residual system of social protection they are likely to fall back on within-family solidarity for social protection (e.g., they stay longer at home), leading to a shift of the poverty risk from young adults to older working family members (Lohmann and Marx 2008). Furthermore, because of the scarcity of dual earner policies (e.g., public childcare) female employment rates are rather low making it harder to achieve dual earnership (Goerne 2011) and likely exacerbating the child effect.

The Northern cluster (Sweden, Finland and Norway² ) is in many respects the antipode of the Southern regime. Characterized by an inclusive labour market with a strong emphasis on centralized and collective wage bargaining (and a compressed wage distribution), an emphasis on high levels of employment and a generous and universal system of social protection, one would expect relatively low levels of working poor among the temporarily employed. Strong degrees of defamilization externalized by inter alia extensive dual earner policies lead to high female employment rates and, thus, high levels of dual earnership. This should whittle down the detrimental effect of having children on living in poverty. Furthermore, young adults are expected to become self-reliant at early age. Because families are thus not additionally burdened, the risk is not shifted to the older workers and young adults are expected to have an above average poverty risk (Lohmann and Marx 2008).

Continental welfare states (Belgium, the Netherlands, Germany, France, and Austria) are more ambiguous in their policies. Similar to the Southern countries, their labour markets are segmented and young workers, as newcomers and thus outsiders, find themselves in a weaker position when trying to secure permanent employment (Baranowska and Gebel 2010). Because continental welfare states realize considerable degrees of decommodification (e.g., via minimum income protection schemes), young adults do not fall back on family solidarity to the same extent as in the Southern countries. Moreover, social security rights are mainly earnings/contributions related and, hence, disadvantageous for young workers in the periphery of the labour market, expected to lead to a higher poverty risk for them.

Finally, the liberal cluster (Ireland, United Kingdom) is characterized by deregulation of the labour market, decentralized wage bargaining (reflected in high earnings inequality and a high incidence of low-waged work) and a meagre role of the state in protecting citizens against social risks (Gallie 2007). Similar to the Nordics, the focus is on getting people into the labour market, however not by providing dual earner policies and high levels of decommodification but via labour market flexibility (Fraser 2011). Since there is no strong employment regulation, it can be expected that the prevalence of temporary employment will be rather limited on the one hand, and that those who are temporarily employed will have more employment opportunities, especially if they are low skilled, on the other. This in turn should increase the chances of multi-earnership for this vulnerable group which is expected to mitigate the detrimental effect of low skill, despite the high prevalence of low wage work (Lloyd et al.2008).

The inclusion of post-communist countries in our sample complicates the matter because we are playing on rather uncharted territory, despite a growing body of research on the classification of these countries in regime typologies (Deacon 1993; Fenger 2007; Draxler and Van Vliet 2010). Although they share a common history of communist rule and a centrally planned economic system, these countries have been characterized by hybridization of their social protection and employment systems since the fall of the Berlin wall (Eichhorst and Hemerijck 2010). Although all former socialist economies have been affected by labour market flexibilization and deregulation (often under auspices of IMF and World Bank, see Cerami 2010) and have also moved away from a centralized wage bargaining system towards a more liberal system (Cazes 2002), most authors observe clear differences between the Baltic countries on the one hand and the Central and Eastern European (CEE) countries on the other (Bohle and Greskovic 2007; Fenger 2007; Peña-Casas 2007). Here we adhere common practice and consider both as two distinct clusters based on their broad institutional similarities (Keune 2006). The Baltic countries (Estonia, Latvia, and Lithuania) have taken the Anglo-Saxon route including a strong emphasis on labour market deregulation without investing much in inclusive labour market policies, and a focus on means-testing and targeting in the social security system consequently leading to low levels of decommodification and social spending. Similar to the liberal countries, they report rather high female employment rates despite the absence of extensive dual earner policies (Ghysels and Van Lancker 2011).

The other Central and Eastern European countries (Poland, Czech Republic, Slovakia, Slovenia, Hungary, Bulgaria, and Romania) have in some important social measures returned to employment-related Bismarckian-style social insurance albeit combined with Anglo-Saxon elements of privatization, such as market-based services, and a tightening of eligibility conditions (Cazes and Nesporova 2003; Keune 2006; Cerami 2008). With regards to labour market flexibility, they constitute some middle group in the European concerto of welfare states, clearly distinct from the ‘extremes’ of the Southern and Liberal countries. The rather stringent employment protection legislation for standard workers in the Baltic countries is often circumvented by employers (Saar 2005) making the labour market flexibility de facto similar to the liberal countries. Obviously, this broad overview obfuscates large differences between these countries (Deacon 1993). Our empirical analysis will learn more on the appropriateness and the intrinsic value of our choice to regard these countries as clusters sui generis, and we will return to this when discussing the results.

All in all, the combined effect of social policy and labour market regulation is expected to lead to cross-country variation in the relative importance of the individual and household-level determinants to become working poor for the temporarily employed (Hypothesis 4).

The sample drawn from EU-SILC is constrained to contractual workers (permanent or temporary) in private households at active age (16–64) which leaves us with 119.895 observations. The dependent variable is a binary indicator, coded 1 for workers living in poverty and 0 for not having a poverty risk.

The set of explanatory variables reflect individual, household and job characteristics. Type of contract (1 = temporary employment, 0 = permanent employment) and gender (1 = female, 0 = male) are dummy-coded. Age is grouped in three intervals reflecting young, prime age and older workers (16–29, 30–49, 50–64) while also three educational levels are distinguished (low, medium and high based on the ISCED-97 classification). For household characteristics, we include the number of children (coded with three dummies and no children as reference), living with a partner (1 = yes) including marriage and cohabitation and living in a dual (=1) or single earner household (=0). We also include the gross monthly pay. To make wages comparable between countries and account for differences in affluence and purchasing power, the variable is z-standardized. Finally, welfare regimes are coded as dummy variables.

Besides these variables, we also control for other individual and workplace characteristics to exclude as much as possible competing explanations for our results. Ethnicity is coded with two dummies reflecting the country of origin (EU or non-EU migrants with natives as reference group). Furthermore, we control for the gendered composition of the economic sector (measured according to 8 aggregated NACE classifications). Finally, we control for working time with a dummy (1 = working part-time) and for work experience by including a linear variable (number of years) and its quadratic form.

Because we are dealing with hierarchical data (individuals are nested in countries) and our dependent variable is a binary indicator, we apply a multilevel logistic regression models with country as the higher level variable. A multilevel design takes the hierarchical structure of our data explicitly into account and yields less biased standard errors than a regular logistic regression model (Hox 2002). We estimate several models with stepwise inclusion of explanatory variables to test our hypotheses. We use the Maximum Likelihood procedure as our estimation method, and the deviance (−2*LogLikelihood) to estimate the fit of the models.

Figure 1 shows the share of the temporary workforce across 24 European countries. Looking at the incidence of temporary employment in Europe, we clearly observe great variety between the identified clusters. The lowest rates are found in the flexible labour markets of the Baltic and Anglo-Saxon countries (less than 5 percent), the highest in the segmented labour market of the Southern countries where almost a fifth of the workforce is temporary employed.³ Most countries, however, represent rates between 8 and 14 percent.. As expected from the theoretical section and our discussion on clustering, the CEE countries are the most heterogeneous with figures ranging from 3 percent in Romania to 24 percent in Poland.

The prevalence of temporary employment also differs within welfare states: Not everyone is exposed to the risk of becoming temporarily employed to the same extent (Kahn 2010). Table 1 shows the share and the composition of the temporary workforce across clusters. Turning to the gender distribution of the temporarily employed, we notice that differences between men and women are not really pronounced: the shares of temporarily working men and women are in the same order of magnitude across regimes. While in most of the cases slightly more women than men are temporarily employed, this pattern is reversed in the Baltic and CEE countries. Despite the fact that is often reported that women are overrepresented in temporary employment (Petrongolo 2004), the differ ences turn out to be rather modest. It is however important to note that temporary arrangements tend to be concentrated in certain sectors, such as the distribution sector, hotels and restaurants and public administration (Conley 2003). At this sectorial level, the distribution is much more gendered. For instance, we find in our dataset that about 43 percent of the temporarily employed women are working in the public sector against only 17 percent of the men. We will control for these sectorial differences when we engage in multivariate analyses.

In all regimes, the highest proportion of temporary workers is found among the youngest cohort with especially large differences with other cohorts in the Southern, Northern, continental and CEE countries while the differences are less pronounced in the liberal and Baltic regimes. This shows that temporary employment might facilitate labour market entrance for young people in the more protected labour markets with insider–outsider characteristics (Bukodi et al.2008). It is noteworthy that, as an exception, a large share (12 percent) of the workers between 50 and 64 are working on temporarily basis in the CEE countries. We also observe that, the low skilled have more chance to work on a temporary basis, although the extent differs between welfare states. In the Southern and CEE countries almost a quarter of the less educated are temporarily employed, while the Baltics and the liberal regimes again resemble each other again with a proportion of only 5 percent. It is interesting to notice that the risk to become temporary employed is much more equally distributed among age categories and educational levels in countries with low overall shares of tempwork.

Now that we have examined the incidence and variation of temporary work in Europe, we link these findings to the poverty figures. Table 2 reports total poverty rates for permanent and temporary employees (first and second column) and differentiated by gender, age and education for the latter. First of all, we clearly observe that temporary employment is associated with a higher poverty risk across all regimes. Surprisingly, the poverty rates for the temporarily employed are highest for the Northern countries and lowest for the Anglo-Saxon ones while the working poor rate for the permanent employed is also lower for the Anglo-Saxon than for the Baltic and Southern cluster. Although the confidence intervals show that we have to be careful in our interpretation of descriptive poverty figures, the perception of in-work poverty as an exclusive Anglo-Saxon phenomenon is clearly misleading.

Second, low education is associated with a higher poverty risk than medium and high educational levels, although the difference is lowest in the Anglo-Saxon countries. Third, we find strong variation by welfare regimes for the age cohorts most at risk. The youngest cohorts face the highest risk in the Northern countries while the oldest workers face the highest poverty risk in the Southern and Baltic countries. The other clusters, however, show a mixed pattern. Finally, we observe that poverty rates for temporary employed women are lower than for men, except in the Continental cluster. All in all, this descriptive account shows that the patterns of in-work poverty generally vary across country clusters as predicted in the theoretical section earlier: low skilled temporary workers face a lesser risk in the Anglo-Saxon countries, the young have a higher risk in the Northern and the oldest cohort has the highest poverty risk in the Southern countries.

We begin our analysis with the estimation of an empty model (baseline model, not shown) to analyse the between-country variance without considering any control or explanatory variables on the individual level. We calculate the intra class correlation coefficient (ICC) as 0.06, indicating that only 6 percent of the residual variation in poverty risk among temporary and permanent workers can be explained by country-level differences. This is an interesting observation, because it means that almost all variation in the odds of living in poverty in our sample is attributable to differences between individuals. This does not mean, however, that the country-level is negligible, only that we cannot explain much by looking at pure country-differences and that individual (and household characteristics) are of major importance to explain in-work poverty. To explore whether a multilevel approach is appropriate given the small ICC, we compared the empty model with a standard logistic regression model using a likelihood-ratio test. This showed that we can reject the null hypothesis that the variance at the country-level is equal to zero (p<0.001), pointing to the relevance of using a multilevel approach. All the models are presented in Tables 3 and 4.

In the first model, permanent and temporary employed are compared including control variables but excluding explanatory variables.⁴ The second model includes all individual and household characteristics whereby the third model adds wages. Compared with permanent workers, the odds for the temporarily employed to live in poverty are increased with a factor 2.31 (95% CI: 2.15–2.49). This result remains robust when controlling for compositional effects in the second model. Although we find significant effects of age (the youngest cohort), education (the low skilled) and composition of the household (having children and living in a single earner household) to become working poor, the difference in poverty risk between permanent and temporary employment is reduced with only 3 percent (OR: 2.22; 95% CI: 2.05–2.40). In other words, individual and household characteristics only explain the poverty gap between temporary and permanent employment to a very small extent. They do explain a reasonable amount of the poverty risk among working people in general: the deviance falls to 37.429 which is a drop of 16 percent.

The poverty difference between permanent and temporary employment is significantly reduced with 43 percent (OR: 1.31; 95% CI: 1.20–1.43), however, when we control for wage differentials (Model 3). It thus seems that, if we assume equal pay for temporary and permanent employment, the poverty differences between the two types of contract are whittled down. Interestingly, including wages also leads to a change in the effect of gender. Model 2 showed no significant difference between men and women, while it turns out that the latter have a lower poverty risk taking wages into account (OR: 0.55; 95% CI: 0.51–0.60), net of the type of contract. This result can be explained if we assert that the wage variable captures the gender wage gap. The deviance drops with an additional 17 percent which means a significantly better fit.

Our hypothesis predicting a higher poverty risk associated with temporary employment vis-à-vis permanent employment is confirmed. The model shows that the major driver of the poverty differences is connected to the lower remuneration of temporary work. We also find that age, education and household composition are determinants of becoming poor for workers, net of the type of contract. We predicted (Hypothesis 3) that the same determinants would be at play in the specific case of temporary employment. We investigate this more in-depth in the next section.

To investigate whether the ‘risk profiles’ of the temporarily employed are similar to the working poor in general, we estimate separate models for subsamples of permanent (108.455 observations) and temporary workers (11.405 observations) respectively (Table 3). Looking at Models 4 and 5, we observe that age, education, gender, and the composition of the household are determinants of becoming working poor for temporary and permanent workers alike, controlled for ethnicity, working time, job experience and sector. Model 5 shows that young temporary workers have a higher risk of living at-risk-of-poverty than the reference group (OR: 1.34; 95% CI: 1.09–1.64) while the older cohort does not exhibit an increased risk. Being high skilled reduces the poverty risk (OR: 0.57; 95% CI: 0.45–0.72) while the poverty risk soars with low skill (OR: 1.25; 95% CI: 1.05–1.49). Furthermore, living in a dual earner household protects a tempworker against poverty (OR: 0.11; 95% CI: 0.10–0.13) compared with a single earner household, and the more children in the household, the higher the poverty risk. Quite straightforward, the higher one's wage, the lower the risk of being poor (OR: 0.10; 95% CI: 0.08–0.11). As being said, the same holds for the permanent workers in Model 4. The only notable difference is found in the effect of the partner variable reflecting the effect of having a non-working spouse (because we control for dual earnership): the poverty risk increases for permanent workers, while the effect is not significant (although pointing in the same direction) for temporary workers. In other words, what is relevant to stay out of poverty is living in a multi-earner household, not having a partner as such. Finally, we observe that temporary working men are confronted with a higher poverty risk than women (OR: 0.56; 95% CI: 0.47–0.66). A preliminary refutation of hypothesis 2 we will discuss further below.

To summarize, the temporarily working poor are by no means a distinct category of the working poor because the same risk factors are determinative: being young, having low skills or living in a single earner household and/or in a household with greater needs (dependent children, non-working spouse).

Table 2 demonstrated that not the Anglo-Saxon but the Northern welfare states show the highest poverty rates for temporary workers, closely followed by the Baltic and Southern countries while the CEE and continental clusters occupy an intermediate position. Controlled for individual, household and work-related characteristics however, the pattern changes (Model 7). While the effects of the explanatory covariates remain stable, pure regime effects can be read from the regime dummies included in the model. Compared with the Southern cluster (reference category), the Baltic countries are identified as underachievers (OR: 2.29; 95% CI: 1.12–4.56) while the continental cluster does significantly better. The figures for the Northern, CEE and Anglo-Saxon clusters do not show such significant difference. In other words, the initial poor performance of the Northern cluster (see Table 2) does not stem from the institutional configuration of the Scandinavian welfare states but from the composition of the temporary workforce.

We hypothesized that the effect of (some of) the risk factors identified in the previous models would vary across regimes (Hypothesis 4). To test this in a straightforward way, we estimate separate models for each welfare regime. This way we are able to analyse whether the behaviour of the poverty determinants in a specific welfare regime differs from the general pattern observed in Models 5 to 7. A drawback of this approach is that we are not able to compare the magnitude of the effects between clusters. Because we are only interested in the broad patterns per regime, however, we can simply ignore this problem. The estimates of age, education and household composition of the six separate models are reported in Table 4.

Looking at the figures from a birds-eye view, we generally observe the predicted variation across regimes. First of all, low skilled temporary workers have a higher poverty risk due to their disadvantaged position on the labour market and the phenomenon of educational homogamy. Theory predicts, however, that they will have more employment opportunities in flexible labour markets which should lead to a lower poverty risk (because income from work can be pooled more often in multi-earner households). Indeed, the negative effect of low skill is reversed in the Anglo-Saxon but not in the Baltic cluster. We also find that being low skilled is especially problematic in the CEE countries. Second, it was expected that young workers have a higher risk in the Northern and continental welfare states (albeit for different reasons) while older workers should be in particular at risk in the Southern countries. We indeed find that young temporary workers are significantly more at risk in the continental welfare states. Although the estimates have the expected sign, the effect of age is not significant in the Northern and Southern countries. Third, having children increases the needs of households which in turn leads to a higher poverty risk. Due to extensive dual earner policies such as public childcare provisions, the child effect is whittled down in the Northern cluster. Not surprisingly given the high female employment rates in these countries, the same phenomenon can be observed for the Anglo-Saxon and Baltic countries. Because they are characterized by a lack of public childcare provisions and low levels of decommodification, it must be the case that the necessity for both parents to work forces them to find a care solution on the private market or via informal channels (Sigle-Rushton and Waldfogel 2007). The observation that in the continental cluster the effect of having 1 child is not significantly different from having no children for temporary workers while having more children clearly leads to a higher poverty risk, illustrates its policy ambiguity. Fourth, having a non-working spouse lowers the poverty risk of tempworkers in the Northern countries. The level of benefits for the unemployed or inactive is high enough to lift households above the poverty threshold. In the same line of reasoning, the opposite results we observe in the Southern cluster can be traced back to the lack of decommodifying measures in these countries, additionally burdening single earners. Finally, and importantly, we find that living in a dual earner household protects the temporarily employed against the risk of living in poverty across all European welfare regimes.

As for the former socialist economies, we are indeed to a certain extent playing on uncharted territory In the Baltic countries, there is no variation by education or age in the probability to become poor for tempworkers. In the CEE countries, having children and being low skilled leads to higher chances of becoming poor for the temporarily employed while age does not yield significant effects. These results are different from the other country clusters although it is difficult to assess whether this is the reflection of a genuine distinct policy constellation or simply conceals more divergent patterns in different countries. Some authors argue, for instance, that Poland has more in common with Spain (compare the temporary employment rates in Figure 1) than with other transition countries (Goerne 2011). Others (Crowley and Stanojević 2011), then, make the case for ‘Slovenian exceptionalism’ as it presumably resembles best the corporatist-continental welfare regime. All in all, further research is called for to disentangle the policy configuration of the new EU member states in a more detailed manner, going beyond the black box of clustering (one example being the study by Gebel and Baranowska 2008).

We assumed (Hypothesis 2) that temporary working women will be more at risk of poverty than their male counterparts. Model 5 showed, however, that the opposite seems to be the case. Women have a lower poverty risk than men, all else being equal (OR: 0.56; 95% CI: 0.47–0.66). One of the explanations has to be sought, ironically, in the still prevailing social reality of the male breadwinner model. Although the decline of the male breadwinner model has been observed throughout the European Union, women are still in majority secondary earners responsible for the bulk of parental care (Lewis 2001). In the case of temporary employment, however, this second income is sufficient to lift the household above the poverty threshold. Because temporary working men are more often primary earners in a single earner household, their income is more often not sufficient to achieve income security. Indeed, in the whole sample of temporary workers, 68 percent of women live in dual earner households while only 56 percent of the men do. Differences between welfare regimes range from 77 percent of women and 56 percent of men living in dual earner households in the Southern countries to only 48 and 31 percent in the Northern regime.⁵ Further analysis confirms this interpretation.⁶ The consequence is that once women face detrimental events such as partnership dissolution, their poverty risk soars, especially with children present in the household Misra et al.2007).

The final question is whether this phenomenon differs between countries. It could be that the circumstances shaped by country policies shape different gender outcomes. Therefore we extend the model to allow both the intercept and the slope to vary in order to test whether the effect of gender differs not only within but also between countries. Model 6 gives the results, and we observe that the variance of the gender variable at the country level is 0.005. The covariance estimate of −0.038 means that in countries with a higher than average intercept (meaning that temporary workers living in those countries have higher odds to live in poverty, taking all other variables into account), the effect of gender seems to be less marked. However, to test whether this result is significant we have to perform a likelihood ratio test (comparing this model with the previous model with only a fixed gender effect). The result is 0.448 on 2 degrees of freedom [−2*(−2710,695 minus −2710,471) = 0.448]. The 5 percent of a chi-squared distribution on 2 degrees of freedom is 5.99. Ergo, we have to conclude that the effect of gender does not differ significantly across countries and welfare regimes. The second hypothesis is thus not confirmed: temporary working women do not face higher poverty risks than temporary working men; the opposite is true.

In this article, we have demonstrated that temporary employment is ceteris paribus associated with a higher poverty risk than permanent employment. The major cause of this difference is the wage gap between both employment arrangements. If we assume that both temporary and permanent workers are equally paid, the poverty gap largely attenuates. Policies enforcing equal pay for tempworkers should thus reduce these differences in poverty risk. However, low wages do not cause in-work poverty as such. We showed that individual and socio-economic household characteristics such as age, education and the composition of the household are the determinants of living at risk of poverty, net of the type of contract. This means that it does not make sense to target policies to prevent in-work poverty specifically at the temporarily employed: they are no different from the working poor in general. Instead, our results show that encouraging dual earnership and enhancing the employability of vulnerable workers, overrepresented in temporary employment arrangements, should reduce the poverty risk of all workers alike. Although individual and household characteristics are the nuts and bolts of explaining the incidence and extent of in-work poverty among the temporarily employed, we found that differences in the institutional configuration across welfare regimes influence the magnitude of the risk factors. In the continental countries, the problem of in-work poverty amongst the temporarily employed is presumably less severe than in other countries. In contrast, they are worse off in the Baltics although in these countries temporary employment comprises only a limited share of the workforce. Finally, we devoted special attention to the gender dimension of the linkage between temporary employment and poverty. Unexpectedly, we find that women working with a temporary contract have less chance to end up in poverty compared with their male counterparts. This is because temporary working women are more often than men secondary earners in dual earner households. Their extra income generally suffices to lift a household above the poverty threshold, while this is not the case for the temporarily employed men in their role as prime earners. Ironically, one could say that temporary working women are shielded from poverty because of the historical legacy of the male breadwinner model.

Wim Van Lancker is a PhD candidate in the Herman Deleeck Centre for Social Policy at the University of Antwerp, Belgium. His current research focuses on the social distribution of family policy in comparative perspective with attention for its effects on the redistributive capacity of welfare states and female labour market participation. He has published on basic income, poverty, employment issues, social Europe, and family policy