Why Precarious Work Is Bad for Health: Social Marginality as Key Mechanisms in a Multi-National Context

The expansion of precarious work in recent decades has motivated a large body of research on its implications for health. While considerable work has focused on whether precarious work undermines health, much less is known about why it matters. To ﬁll this gap, this paper offers and tests a conceptual model whereby the effects of precarious work on health are mediated by social marginality, speciﬁcally reduced self-efﬁcacy, weaker social integration, and lower social capital. All three mechanisms are understood as both social consequences of precarious work and important determinants of health. Empirically, we use data from the European Social Survey and investigate (1) conditional direct effects of precarious work on self-rated health and (2) extent of mediation via the three mechanisms. Furthermore, we assess the generalizability of the model across ﬁve welfare state regimes that prior work has deemed to be important moderators of the health – precarious work relationship. Results indicate precarious work has signiﬁcant conditional direct effects and indirect effects through all three mediators that signiﬁcantly reduce effect of precarious work on health. This is robust in the general sample and for four of ﬁve welfare state regimes. These ﬁndings highlight a previously unexplored vector connecting precarious work to health and indicate that the effects of precarious work on perceptions of self and social relations is a key link to poorer health. The study also expands conceptualization of the broad role of socioeconomic status for health inequalities and furthers understanding of the mechanisms at work.


Introduction
The expansion of precarious work in recent decades has motivated a large body of research on its implications for health. Ambitious agendas have emerged around issues of organizational restructuring and downsizing, perceived job insecurity, and temporary employment across a range of countries (Bardasi and Francesconi 2004;Ferrie et al. 2005;Liukonnen et al. 2004;Virtanen, Janlert, and Hammarström 2011;Vives et al. 2013). Although some have noted benefits of precarious work in terms of flexibility and autonomy (e.g., Ravenelle 2019), precarious work is increasingly recognized as a key detriment of poor health and an important component of the overall relationship between work, socioeconomic status, and health and well-being (Benach et al. 2014). Epidemiological interest will likely only grow given the increasingly prevalence of precarious work, expansion of precarity into different aspects of work, and the dominance of "flexibility" in discourses around employment relations with economic globalization (Kalleberg 2009).
Theoretically, there are good reasons why precarious work would undermine health. Increased precarity in work likely fosters a wide array of insecurities that would translate into life course stressors that undermine health. Benach et al. (2014), for example, offer a conceptual model that emphasizes exposure to hazardous work conditions, psychosocial stress associated with limited control and feelings of powerlessness, and problems in the management of social and economic affairs of everyday life. Yet, beyond such statements, there has been little work that attempts to build and test a theory. In particular, empirical work on the mechanisms that link precarious work to health is particularly lacking. In this paper, we fill this gap and extend thinking by focusing on the social and social psychological consequences of precarious work, particularly those reflecting social marginality (see discussions in Kalleberg 2009;Standing 2011), as mediating the precarious work-health relationship.
At the same time, there are reasons to expect that any relationship between precarious work and health is variable across countries. Scholarship on welfare state regimes in particular argues that different regimes (1) have more or less stringent labor protections that would alter risk of exposure to precarious work and (2) have more or less extra-market supports (e.g., health care, pensions, income support, unemployment benefits) that would moderate negative effects in the face of precarious work (Bambra 2007;Kim et al. 2012). Likewise, the "varieties of capitalism" tradition emphasizes variation in policies around employment and unemployment protections and research shows important effects on employment experiences over the life course (Van Winkle and Fasang 2017). Still, welfare state regimes and variation in employment policy more generally did not emerge in a void and aspects of culture and history may also condition the meaning and experience of precarious work. For example, gender norms and the gendered division of labor imprint labor markets to shape the meaning of work and not just for women (Esping-Andersen 2009). Given structural and cultural variation, it is not clear how well any conceptual model would generalize. Moreover, in the absence of systematic empirical study across contexts, the role of broader political economic conditions in shaping the health consequences of precarious work is speculative at best. This paper examines the connections between precarious work and health by connecting two traditionally separate bodies of research. First, we articulate and empirically evaluate three mechanisms that link precarious work to health. We draw upon both classic statements on precarious work and their consequences (Standing 2011) and work in the social epidemiology of health  to focus on social marginality indexed by reduced global selfefficacy, weakened social integration, and lower social capital as mediating the relationship between precarious work and health. Second, we repeat the analyses for different welfare state regimes-Scandinavian, Continental, Anglo-Saxon, Eastern European, and Mediterranean-to test whether the association holds across welfare states and whether its strength varies in plausible ways with country context. In doing so, we extend explanation on the deleterious effects of precarious work by identifying the micro-dynamics that connect precarious work to health and showing how such dynamics relate to macro political-economic contexts. In identifying new and unexplored mechanisms linking precarious work to health, we further extend thinking on work experiences, socioeconomic status, and health inequalities more generally.

Origins of Precarious Work
Scholars have documented significant changes in the global character of work over the last three decades, particularly the emergence and expansion of precarity in the nature of paid employment (Kalleberg 2009;Osterman and Shulman 2011;Standing 2011). Beginning in the mid 1970s, neoliberal globalization arose as the dominant logic of finance, economics, and management. Effects were multifaceted but its macro-level expression included heightened economic integration and a more globalized marketplace, increased competition (particularly from countries with much lower aggregate labor costs), concomitant greater opportunities for outsourcing of labor (typically to lower wage countries), and corresponding immigration that introduced new pools of (lower wage) labor (Kalleberg 2009). Technological advances accentuated such processes by forcing companies to become more competitive and providing mechanisms whereby they could compete with one another. The end result was a pervasive, if not global, managerial turn toward greater flexibility in labor relations. As Kalleberg (2009, 3) describes, The standard employment relationship, in which workers were assumed to work full-time for a particular employer at the employer's place of work, often progressing upward on job ladders within internal labor markets . . . [gave way to] . . . various types of corporate restructuring, which in turn led to a growth in precarious work and transformations in the nature of the employment relationship.
As the logic of labor underwent radical change, "flexibility" became the overarching principle of worker-employer relationships (Kalleberg, Reskin, and Hudson 2000). The main mechanism was the growth of nonstandard work arrangements, part-time, and contingent work. This included a range of practices, but most fundamentally involved a combination of "downsizing" whereby the number of permanent staff was reduced and outsourcing whereby firms increasingly purchased goods and services from other organizations and individuals that they themselves would traditionally have produced (Osterman and Schuman 2011).
But "flexibility" reconfigured other aspects of the labor relationship and produced externalities for the worker's socioeconomic position. Flexibility results in declines in attachment to employers whereby people spend less time with any given employer, increases in long-term unemployment, growth in perceived job insecurity, and shifts in financial risks from employers to employees (Kalleberg 2009). Given this, conceptualization of precarious work expanded beyond precarious employment to incorporate multiple aspects of employment conditions. Rodgers and Rodgers (1989), for example, conceptualized precarious work as including low degree of certainty of continuing work, limited control over work conditions, limited protections provided by regulation, social expectations, and benefits, and an inadequate income (see also Tompa et al. 2007). Kalleberg, Reskin, and Hudson (2000) focus on "non-standard jobs" also situated precarious employment within a larger context of employment conditions or job quality and showed that non-standard work was strongly associated with significant increases in the likelihood of low pay, no health benefits, and no employer pension-key dimensions of "bad jobs." Vives et al. (2010) further describe an "employment precariousness" scale (EPRES) organized around temporariness of work, low power in negotiations around employment conditions, vulnerability to authoritarian treatment, low or insufficient wages, limited rights, and poor social security benefits. Standing (2011) emphasizes similar themes when he characterizes the "precariat" as workers who face multifaceted insecurity-employment insecurity that increases risk of arbitrary job loss, job insecurity that lessens worker's ability to maintain an employment niche (i.e., the type of work one does is reasonably stable), and skill reproduction insecurity whereby people cannot gain or improve occupational skills or use the skills they have in the workplace. Additionally, income insecurity undermines the maintenance of everyday life and precarious workers often have few, if any viable channels for occupational, if not social, mobility. In the aggregate, this corpus of work highlights how precarious work involves a complex combination of precarious employment and poor employment quality with particular, but not exclusive, emphasis on how precarious employment opens the door for multiple aspects of "bad jobs" (e.g., Kalleberg, Reskin, and Hudson 2000). Against this backdrop, precarity can be understood as a characteristic of work at a particular point in time or something that arises over time and hence is an important determinant of the overall character of work over the life course (Ritschard, Bussi, and O'Reilly 2018).

Social Psychological Consequences of Precarious Work
Beyond efforts to delineate its nature and understand its origins, there has also been some effort to articulate the social and psychological consequences of exposure to precarious work. Standing (2011, 68) for example argues that the precariat experience "the four As-anger, anomie, anxiety, and alienation." The anger stems from frustration at the lack of viable pathways for social mobility and a meaningful life. Anomie is "a feeling of passivity born of despair" and has its origins in the inability to secure a better socioeconomic foothold. Precarious workers also live with anxiety due to the chronic insecurity that comes from unstable work and financial vulnerability. And alienation arises from knowing that work is not for self-fulfillment or for personal pride but instead is simply done for others at their demand. In sum, precarious work is seen to broadly undermine positive affect that may ultimately prove detrimental to self and psyche (Virtanen et al. 2005). At the same time, the description of the four As has limitations and is ultimately more heuristic than proscriptive. In one respect, they are largely hypothetical, offering possible consequences of exposure to precarious work rather than showing that such affect emerges. At the same time, their scope is largely psychological without conceptual clarity. As such, they provide limited insight into social and sociological consequences of precarious work. The latter is particularly significant for efforts to link precarious work to health and well-being given ambiguity of meaning (e.g., how is alienation measured) and conceptual overlap with the outcome (e.g., anger and anxiety are measures of psychological well-being).

Precarious Work and Health
Perhaps because other aspects of workplace experiences, relationships, and their consequences have been studied independently and in isolation, early studies of precarious work tended to focus on either contract status or job insecurity (perceived or experienced). And although there are good reasons to view temporary or ephemeral work as a detriment to health (see discussions in Tompa et al. 2007;Vives et al. 2010), research has been more equivocal. For example, studies of temporary employment-including part-time, contingent, subsidized, and insecure employment-and health find negative (e.g., Ferrie et al. 2005), null (e.g., Bardasi and Francesconi 2004), or mixed results (e.g., Liukonnen et al. 2004). Studies of job insecurity are equally ambiguous. A meta-analysis of 72 published studies for example found only a weak, average association that was highly heterogeneous (Sverke, Hellgren, and Naswall 2002), while a more recent multi-national assessment reported highly variable and typically weak associations (László et al. 2010).
One potential explanation for varied effects in prior work is variation in conceptualization and measurement. While early studies focused on temporary work and job insecurity, others argue for a broad conceptualization. Tompa et al. (2007) for example characterize precarious work as work with some degree of uncertainty of continuity of employment, limited control over work processes, low income and benefits, subordination in the workplace, limited social support, greater exposure to physical hazards, and diminished opportunities for training and career advancement. The EPRES approach described earlier is similarly multidimensional and incorporates issues of instability, powerlessness, vulnerability, and insufficiency of wages in its characterization of precarious work (Vives et al. 2010; see also Lewchuk 2017; Van Aerden et al. 2016). We build upon such studies by using a multi-dimension construct that reflects major dimensions of precarious work and item response scaling (IRT). Specifically, we incorporate information on both the current job and employment history to capture non-indefinite contract status, low control over job activities, low influence on the direction of the work organization, gaps in employment, and financial vulnerability. This conceptualization is reasonably comprehensive and overlaps with other multidimensional approaches (e.g., Lewchuk 2017;Tompa et al. 2007; Van Aerden et al. 2016;Vives et al. 2010). At the same time, other dimensions-including exposure to toxic work conditions, lack of control over the when and where of work, lack of social mobility or even a clear mechanism of social mobility, and a lack of collective representation-are not captured by the present operationalization. General discussion of measurement strategies is found in Benach et al. (2014).

The Political-Economic Context of Precarious Work
Research on precarious work and health has also occurred in a number of countries and contexts, and this too may contribute to a lack of consistency in results across studies given potential heterogeneity with respect to national context. Of primary importance may be the welfare state context as it shapes both exposure to precarious work and extra-market supports that could buffer otherwise negative consequences of precarious work. While early work distinguished socialdemocratic, corporatist-conservative, and Liberal welfare states (Esping-Andersen 1990), more recent work in social epidemiology offers a four-fold typology of Liberal/Anglo-Saxon, Conservative/Corporatist/Bismarkian/Continental, Nordic/Social democrat/Scandinavian, and Southern/Mediterranean (Bambra and Eikemo 2009). Adding Eastern European countries with their own distinct political economic contexts produces five distinct regimes.
Beyond the traditional welfare state typology, other scholars have developed complementary political-economic models that flesh out variation across countries in labor and extra-labor policies. 1 Building off the "varieties of capitalism" model, important research focuses on how variation in policies around employment and unemployment protections foster systematic variation in work and trajectories of work over the life course (e.g., Van Winkle and Fasang 2017). There are a number of dimensions. First, countries vary in terms of active labor market policies that provide supports for more rapid returns to paid employment in the event of job loss (Kluve et al. 2007). Such policies are particularly developed in Scandinavian countries and have been shown to reduce involuntary job moves that are a signature of precarious work. Second, employment protection legislation (EPL) restricts employer's rights to terminate employees and limits the use of limited duration or other forms of non-standard work. Although critical for offsetting the expansion of precarious work, such protections are highly heterogeneous across European countries-strongest in Scandinavian and Mediterranean contexts, but quite weak in Anglo-Saxon and Eastern European ones (OECD 2019).
Third, labor market segmentation indexes the variable connections to paid labor with "insiders" in standard employment and "outsiders," including women, minorities, and low skill labor, exposed to more precarious work. Importantly, different structures of social welfare minimize or mitigate such exposures (Häusermann and Schwander 2012). Continental and Mediterranean regimes exacerbate labor market and job inequalities (e.g., earnings, job conditions, social mobility), undermine social protection, and weaken political integration. In contrast, Liberal regimes exacerbate job and labor market inequalities, but have stronger social protections due to lingering policies and institutions of social welfare. Even more distinct, insider-outsider differentiation is more muted in Nordic regimes along all dimensions. Although not empirically studied, Eastern European countries may show the most extreme insider-outsider differentiation due to rapid privatizing in the transition to capitalism, the elimination of social protections, and the elimination of collective representation that came with the adoption of "shock doctrine" in many former Communist countries (Sachs 1994).
Finally, the basis of benefit calculation and allocation, benefits that would mitigate the consequences of precarious work, are also variable across countries and connected to welfare state regimes (Bambra and Eikmo 2009). Benefits are mostly universal in Scandinavian contexts, means tested in Anglo-Saxon ones, insurance based (often through an employer) in Continental countries, rudimentary and fragmentary in Mediterranean contexts, and almost non-existent in the Eastern European regime. Given financial vulnerabilities associated with precarious work, variation in how benefits are calculated, determined, and allocated should be an important contingency in shaping the personal and social consequences of precarious work.
Both the traditional welfare state regime and varieties of capitalism model suggest a rough hierarchy with respect to labor market protections/interventions around precarious work (e.g., Bambra 2007;Benach et al. 2014;Kim et al. 2012). At the low end, the transition to capitalism in many Eastern European countries removed traditional labor protections, gutted public services, and privatized a range of social welfare (Stuckler, King, and McKee 2009). In such contexts, one might expect the effects of precarious work to be particularly pronounced. Although less extreme, Liberal regimes are similar in that they have minimal social welfare, dominance of the market is encouraged, and protections against the negative effects of precarious work are few. Still, residuals of the traditional post-war welfare philosophy remain, particularly with respect to health care (e.g., the National Health Service in the UK, and the subsidized health care of Canada, Ireland, and Australia). Markets are buffered more in Continental regimes and welfare is more extensive. Mediterranean contexts are seen as more fragmented and diverse, but strong cultures of family and the voluntary sector may substitute for state allocated welfare provision (Galasso and Profeta 2018). Finally, the Scandinavian regime with its generous social transfers, commitment to strong employment and good incomes, and universal healthcare is seen as the strongest context for mitigating the negative health consequences of precarious work (Kim et al. 2012). While largely speculative on the specific issue of precarious work and health, the described hierarchy of welfare versus market dominance provides a useful heuristic for systematic empirical inquiry.

Social Marginality as Mechanisms
In articulating the social links between precarious work and health, we take initial guidance from Standing's (2011) thesis on the social and psychological attributes of the "precariat" and focus on self-efficacy, social integration, and social capital. Broadly conceived, these index social marginality in that they indicate how people are separated, socially and psychologically, from conventional society, social institutions, and social activity. As origins, Standing (2011, 44-45) describes the "precariat" as experiencing both "job" insecurity and "skill reproduction" insecurity. Job insecurity involves an inability to "retain a niche in employment, plus barriers to skill dilution and opportunities for upward mobility in terms of status and income." Skill reproduction insecurity involves inadequate "opportunities to gain skills, . . . employment training and so on, as well as opportunities to make use of competencies." Such descriptions echo classic research in the social psychology of work where the type of work that people do translates into differing levels of "occupational self-direction"-a sense of agency and efficacy for occupational tasks (Kohn 1989). Importantly, occupational self-direction further generalizes to self and society: those whose work involves simpler tasks, those who experience greater supervision, and those whose work is more routinized are less likely to be self-confident and are more likely to be fatalistic, anxious, and conformist in their ideas (Kohn and Schooler 1982). A defining feature of precarious work is a lack of autonomy and capacity for self-direction in work tasks that reflects both the impermanence of work and the organizational structures that have evolved around temporary employees and ephemeral placements (Standing 2011). Consistent with this, research on self-efficacy and health highlights two meditational pathways. Self-efficacy can be "pre-intention" and mitigate exposure to risk and foster health enhancing behaviors (Schwarzer and Renner 2000) or it can be "post-intention" and serve as a coping mechanism in the face of risks and stress (Benight and Bandura 2004). Through either mechanism, lower self-efficacy through precarious work should undermine health.
A second mechanism is limited social integration. Berkman et al. (2000) elaborate a conceptual model linking social integration, broadly conceived, to health. The crux of their model is social structural conditions (i.e., culture, socioeconomic conditions, political context) that shape both the structure and characteristics of social networks. Embeddedness in different types of social networks provide opportunities for social participation (e.g., social influences, social engagement, and person-to-person contact) that impact health by influencing health behaviors, psychological well-being, and physiological functioning. LaRocco, House, and French (1980) for example show that different types of social ties-to supervisors, co-workers, family, and friends-produce different types of social support that buffers the effects of occupational stress on mental health. Likewise, considerable work shows that network-based norms affect different types of health behaviors, including risk behaviors such as smoking, substance use, and poor diet (e.g., Krosnick and Judd 1982). Our focus on precarious work fits easily into such models given that it is both an element of social structure, reflecting both the organization of labor markets and broader socioeconomic differentiation, and an important determinant of the quality and structure of social networks, both within and beyond the workplace. In particular, the ephemeral nature of precarious work, diminished opportunities for meaningful workplace participation, and socioeconomic vulnerabilities should have externalities for both the frequency of interactions with others and self-perceptions of such interactions (Standing 2011). In the former case, the triangulation of governance (i.e., company-labor broker-worker) creates social distance between "real" employees and temporary workers and this will undermine the development of social ties. Equally important, competition becomes institutionalized as "real" workers struggle to maintain their position in the firm while the precariat compete to gain entry into more permanent work. In the latter case, cyclical unemployment and financial vulnerability should undermine both willingness and ability to form ties due to perceived stigma and lower capacity for social interactions beyond the workplace and hence contribute to the general marginalization of precarious workers (Handler and Hasenfeld 2006).
A final mechanism is decreased social capital. Coleman's (1988) seminal statement on the concept of social capital emphasized social ties between actors that are infused with obligations and expectations. Given this, precarious work should have pervasive, negative effects on social capital. Precarious work undermines ties to others, both within the workplace and to the broader community, by virtue of the negative social externalities that come with unemployment and the demonization of the unemployed (Standing 2011, 70). With inadequate wages, precarious work also introduces problems in the management of everyday life that likely undermine the desire for higher order needs of belonging, esteem, and self-actualization. This would ultimately foster negative social psychological orientations toward others, including trust and the perception of others as resources, by virtue of fragmented or underdeveloped networks, heightened competition, and diminished connections between self and society.
In social epidemiological and public health research, social capital has three plausible pathways to better health (Kawachi and Berkman 2000). First, social capital may influence health-related behaviors by either promoting the flow of pro-health information or by exerting social control over deviant health-related behavior. Second, social capital can influence access to services and amenities that are resources for the maintenance of good health. Finally, social capital can provide affective support and serve as a source of self-esteem and mutual respect. An important element here is trust and how trust fosters both reciprocity and exchange and the perception of reciprocity and exchange, which makes people feel better about themselves and their communities. Through all these channels, embeddedness in precarious work should erode social capital and further undermine health.  Figure 1 summarizes our basic conceptual model in a reduced form. The model begins with sociodemographic and socioeconomic background that includes age, gender, nativity, and marital status, as well as a proxy for sedentary lifestyle (i.e., amount of television watched per day) and religiosity (i.e., number of services attended per week). These are complemented with measures of socioeconomic status (SES) in one's family of origins indexed by family social class and parent's educational attainment, as well as the respondent's socioeconomic position that includes educational attainment, labor market position (e.g., being in school, unemployed, retired, etc.), poverty status, and class position. Although causal order with respect to precarious work is ambiguous for some of these variables, we treat them as exogenous theoretically and empirically to provide an unambiguous pathway from precarious work to health through the mediating variables. Our model represents this possibility by showing direct effects on the mediator set of social marginality-self-efficacy, social integration, and social capital-which collectively and directly affect health. Importantly, we allow the background variables to influence health but view the mediator set as largely accounting for the effects of precarious work on health.

Data, Measures, and Methods
This study is based on data from round 6 of the European Social Survey (ESS) fielded in 2012, subsequent to the global economic recession that wreaked havoc on employment and employment relations and hence perhaps indicating the new era of labor in the globalized economy. Data were collected from

Measures
Our focal outcome is self-rated health (hereafter SRH). SRH is coded from 1 = "poor" to 5 = "excellent." A wealth of research attests to the validity of SRH as a measure of health status with some of the most convincing evidence being its predictive capacity for subsequent mortality (Idler and Benyamini 1997). There is similar support for its cross-national validity particularly in the European context (Baćak and Olafsdottir 2017). Although there is not widespread consensus on what constitutes precarious work, we focus on five dimensions highlighted in previous research. The more traditional measure of precarious work, reflective of precarious employment, is the type of employment contract that one has. Non-indefinite work differentiates those with contracts of indefinite duration with those with contracts of limited duration or no formal contract. Non-indefinite work, however, opens the door for a host of further precarities, what are often considered indicators of employment quality (Kalleberg, Reskin, and Hudson 2000;Standing 2011;Vives et al. 2010). Low job control indexes respondents who fall within the bottom third of the distribution on the item "how much respondents can 'decide how your own daily work is/was organized?' (ranging from 0 = 'I have/had no influence' to 10 = 'I have/had complete control')." Low organizational influence is measured similarly based on whether respondents can/could "influence policy decisions about the activities of the organization? (ranging from 0 = 'I have/had no influence' to 10 = 'I have/had complete control')." Gaps in employment is based on whether respondents report "ever being unemployed and seeking work for a period of more than three months (yes coded 1; no coded 0)." The final aspect measures financial vulnerability based on a question asking respondents whether they find it "difficult" or "very difficult" to live on one's income (yes coded 1; no coded 0).
We used IRT scaling to create a cumulative measure of precarious work. Although traditionally used in psychometrics to assess differences in ability based on a set of test items, IRT has particular value in modelling phenomena that are captured by a set of binary outcomes (Embretson and Reise 2013;van der Linden and Hambleton 2013). The relationship between a set of attributes (e.g., non-indefinite contract) and the underlying latent trait (i.e., extent of precarious work) is modeled such that higher values on the attribute determine location on a continuous latent variable. Here, an item characteristic curve (ICC) relates the probability that a person is characterized by a particular attribute to a particular location on the latent variable. The latent variable has two parameters: the difficulty parameter represents the location of an attribute on the latent variable scale, while the discrimination parameter indicates the correlation between the latent trait and the attribute. Higher discrimination indicates that the attribute is useful at distinguishing low versus high levels of the latent variable, in our case precarious work. In a multivariate context, different items are arrayed in relation to the latent variable and hence make stronger or weaker contributions to greater precarity in work. In addition to providing greater precision to the measure, IRT has the value of allowing reliability (i.e., unique within-sample variance) to vary across subsamples (e.g., welfare state regimes) and hence provides a measure most appropriate for the context under examination. A focus on a continuous or graded measure is also consistent with arguments that precarious work involves multiple dimensions.
The characteristics of the IRT measure of precarious work for the full sample is shown in figure 2. Specifically, the figure shows the item characteristics curves that link each individual item to the latent variable due to their respective "difficulties." It is significant that the greatest difficulty is for contract status (1.44) as this indicates that higher values of precarious work most "require" non-indefinite contract status. There is similar differentiation for the other items with difficulties of 1.24 for low job control, 0.999 for financial vulnerability, 0.979 for low organizational influence, and 0.736 for employment gaps. The empirical realization of the latent variable ranges from −0.791 to 1.795 with higher values indicating greater precarity. There is, however, significant variation across welfare state regimes with precarious work ranging from −0.516 to 2.302 for Scandinavian countries, −0.673 to 2.063 for Continental countries, −0.620 to 2.02 for Anglo-Saxon countries, −0.983 to 1.518 for Eastern European countries, and − 0.836 to 1.859 for Mediterranean countries. Although IRT is a particularly powerful measurement strategy given the nature of the measures, several different operationalizations of precarious work, including a simple summative index and a single construct factor score, produced substantively similar results.
The set of mediating variables includes self-efficacy, social integration, and social capital. Global self-efficacy is a summative index comprised of responses to five questions: "In general, [respondent] feels very positive about him/herself," "[Respondent] feels free to decide how to live his/her life," "At times, [Respondent] feels like a failure (reversed)," "[Respondent] feels accomplishment about things done," and "when things go wrong in life, it takes a long time to get back to normal (reversed)." Social integration is measured with three items including the number of people you can talk to about intimate or personal matters (from 0 to 10 or more), how often one meets socially with friends, relatives, or colleagues (from "never" to "every day"), and respondent's perception of how often they take part in social activities compared to others of the same age (from "much less than most" to "much more than most"). As these items are on different scales, we created a normalized factor score to capture variance. Finally, social capital is a summative index comprised of whether the respondent thinks "people in the community help one another" and "feel that people treat you with respect." Correlations among the three constructs is moderate, ranging from 0.22 to 0.30.
Our control variables fall into one of three categories. First, standard control variables to address unobserved heterogeneity include age in years, gender, nativity, marital status, television watching per day in minutes, and religiosity based on frequency of attending services. Second, we capture socioeconomic status in the respondent's family of origin with two measures. Family social class uses the Erickson-Goldthorpe-Porocarero (1979) schema is based on the occupation of the respondent's father when there were 14 years of age. It differentiates "professional/technical jobs," "high administrators," "clerical workers," "sales workers," "service workers," "skilled workers," "semi-skilled workers," "unskilled workers," and "farm workers." As the categories are clearly hierarchical with respect to conventional notions of prestige or status but are not formally scaled, we model effects based on a set of dummy variables indexing group membership. We also include a measure of parent's educational attainment that is the highest level of educational attainment of the respondent's mother or father ranging from "less than lower secondary" to "tertiary education." The combination of parental occupation and education has a long and valuable history in research on stratification and social mobility (Hauser and Warren 1997). 2 A third set of variables measure the respondent's current socioeconomic position. First, we control for respondent's educational attainment. Labor market position is measured by the respondent's main activity during the week of the survey and differentiates those "employed," "in school," "unemployed," "retired," not in the labor force due to a disability, and "other [not in labor force]." Low income indexes respondents who fall in the bottom decile of their country's income distribution. 3 Finally, we measure respondent's class position using Rose and Harrison's (2010) revised European socioeconomic classification (ESEC). ESEC social class differentiates respondents as to whether they are (1) large employers, higher grade professional, or administrative/managerial occupations; (2) lower grade professionals, administrative and managerial occupations, and higher grade technician and supervisory occupations; (3) intermediate occupations; (4) small employers and self-employed occupations (excluding agriculture); (5) self-employed occupations including agriculture; (6) lower supervisory and lower technician occupations; (7) lower services, sales, and clerical occupations; (8) lower technical occupations; and (9) routine occupations. As a set, these measures have considerably more depth than those used in prior research on precarious work and health, while at the same time capturing a number of dimensions key to traditional conceptualizations of socioeconomic status.

Methods
Our main analytic strategy is ordinary least squares (OLS) regression with country fixed effects. We focus on six analytic samples: the full sample with follow-up stratification by welfare state regime that differentiates Scandinavian (n = 5,367), Continental (n = 5,087), Anglo-Saxon (n = 3,100), Eastern European (n = 8,877), and Mediterranean (n = 3,368) contexts. Standard errors are clustered at the country level to adjust for potential non-independence of units. An advantage of OLS regression given our interest in comparative analyses is that we can formally test differences in effects sizes across samples using Clogg's z-test (Clogg, Petkova, and Haritou 1995), something we would not be able to do if we used logistic regression (Mood 2010).
Our theoretical model is evaluated using mediation analysis. Baron and Kenny (1986; see also Kenny, Kashy, and Bolger 1998) offered the seminal normal theory (NT) approach based on the estimation of three causal steps. In the first step, the dependent variable (i.e., self-rated health) is regressed on the focal independent variable (i.e., extent of precarious work) with select controls. Second, each mediating variable-self-efficacy, social integration, and social capital-is regressed on the focal independent variable with select controls. Third, the dependent variable is regressed simultaneously on both the mediating variable(s) and the focal independent variable. Under the logic of statistical mediation, step one should show a statistically significant relationship between precarious work and health, step two should show that all mediating variables have statistically significant relationships with precarious work, and step three should show (1) that the mediator variables have statistically significant associations with health and (2) that there is a significantly smaller association between precarious work and health with the mediator variables included in the model. Statistical significance of the indirect effects is a z-test based on the product of coefficient from the second regression and the coefficient linking the mediator variable(s) to the dependent variable. The relevant standard error is calculated as a 2 sb 2 + b 2 sa 2 + sa 2 sb 2 where a and b are the unstandardized regression coefficients and sa and sb are the corresponding standard errors (Kenny, Kashy, and Bolger 1998). 4 All analyses were conducted using Stata 15.1.

Results
Our analyses begin with the full sample with coefficients shown in table 2. As a preliminary, the various coefficients for the sociodemographic, behavioral, and socioeconomic controls are very consistent with prior work. SRH declines with age, is lower among females, is poorer among those with more sedentary lifestyles, and is better among those who are married. There is some evidence of family class differentials in that those from unskilled labor backgrounds have poorer health. SRH is also better with greater educational attainment, among those who are contemporaneously employed, and among those not with low incomes.
Independent of these effects, precarious work has a strong, negative association with SRH. Before including the set of mediating variables, work precarity is associated with poorer SRH (b = −0.077, p < .01). To give some perspective on the size of this effect, estimation of marginal effects shows that variation in health associated with precarious work is two times larger than that for the entire range for ESEC social class, about thirty percent greater than the    Robust standard errors in parentheses. * * * p < .001, * * p < .01, * p < .05.
effect for low income, almost three times larger than the effect for being contemporaneously unemployed, and slightly larger, between twenty and fifty percent depending upon how low educational attainment is measured, than the effect of differences in educational attainment. Precarious work also associated with lower self-efficacy (b = −0.653, p < .001), weaker social integration (b = −0.086, p < .001), and less social capital (b = −0.332, p < .001). With the mediating variables included, all have statistically significant effects on health (bs = 0.055, 0.092, and 0.023 for self-efficacy, social integration, and social capital, respectively) and the coefficient for precarious work is significantly reduced and no longer statistically significant (b = −0.026, ns). Finally, z-tests indicate that all three mediating variables have statistically significant, indirect effects on precarious work on SRH (zs = 7.221, 4.030, and 3.186 for selfefficacy, social integration, and social capital, respectively). In sum, these analyses indicate that the mediator set explains 66% of the initial effect of precarious work on health ([−0.077 -(−0.026)]/−0.077 = 0.662). We next assess similarity and difference in estimates across welfare state regimes (see table 3). To save space, we focus on the most theoretically relevant coefficients linking precarious work to social marginality to health with the full set of coefficients shown in the online supplementary material for Appendix tables A1-A5. Beginning with Scandinavian countries, precarious work has a significant association with poorer health (b = −0.101) and is also associated with lower self-efficacy ( The situation is somewhat different among Eastern European countries. Here, there is no conditional direct effect of precarious work (b = −0.017, ns), although precarious work still undermines self-efficacy (b = −0.507, p < .001) and social integration (b = −0.084, p < .01). In the final model, all three social marginality indicators have effects on health and both self-efficacy and social integration have statistically significant indirect effects. Yet given the lack of any conditional direct effect, they cannot contribute to statistical explanation for precarious work.
The conceptual model, however, has more support among Mediterranean countries. The initial effect of precarious work is negative and statistically significant (b = −0.079, p < .001). Precarious work also has a negative effect   Robust standard errors in parentheses. * * * p < .001, * * p < .01, * p < .05. . The formal test also shows significant indirect effects for self-efficacy and social capital. As a final issue, the overall consistency of the model parameters across welfare state regimes is assessed by examining statistical significance of differences in the magnitude of coefficients across contexts using the Clogg test (Clogg, Petkova, and Heritou 1995). Using a critical value of 3.2 that takes into account the multiple comparisons per measure, the results are very clear. There is no situation where pair-wise comparisons show statistically significant differences for specific coefficients, although it is also true that the direct and indirect effects are smaller in magnitude for the Eastern European context for 25 of 28 comparisons. We further repeated all of the above separately by gender given prior theoretical and empirical work on the gendering of precarious work, health, and the relationship between the two (e.g., Menéndez et al. 2007) but find almost no evidence of gender variation in the key processes under study. 5 Figure 3 summarizes things by graphically showing the direct and mediated effects for all five welfare state regimes. Two things are noteworthy. Extent of statistical explanation is clear in that the point estimates for the mediated effect generally fall outside the confidence interval of the conditional direct effect. At the same time, the fact that confidence intervals overlap and point estimates for any given regime typically fall within the confidence intervals of the other regimes indicates substantive similarity in effects across contexts. That said, it is also clear that the model in general appears to work least well in Eastern European countries given that there is no direct effect, no substantive mediation, and only two significant indirect effect.

Discussion
This paper makes three contributions to sociological research on health. First, we reinforce the view that precarious work is bad for health with a broad systematic assessment in a large, heterogeneous sample with an improved measurement strategy. Although it is generally accepted that precarious work undermines health, empirical work has been more equivocal. One explanation is that prior research often operationalizes precarious work with one indicator. In contrast, the present research uses a multi-indicator approach that better reflects theoretical discussions of precarious work and thus improves upon content validity. Here, we draw upon Standing's (2011) and Kalleberg, Reskin, and Hudson's (2000) argument that precarious work may be triggered by non-indefinite contracts but ends up producing multifaceted insecurity-of tenure, of task, of control, of compensation, and of benefits. In this respect, our approach is similar to more recent innovations in the measurement of precarious work that focus on its multi-dimensional character (e.g., Lewchuk 2017; Van Aerden et al. 2016;Vives et al. 2010Vives et al. , 2013. Adopting such an approach appears to produce much more consistent associations between precarious work and health across countries and contexts than is seen in previous work (cf, László et al. 2010;Sverke, Hellgren, and Naswall 2002).
Our second contribution develops and tests a model of mediating mechanisms that link precarious work with health. Previous research has largely ignored the social consequences of precarious work and how these might be an important vector linking it to health (cf., Benach et al. 2014, figure 1). In contrast, we focus on social marginality indexed by perceived self-efficacy, social integration, and social capital that is derivative of precarious work and mediates its relationship with health. Empirically, the full sample analysis shows that precarious work has a negative effect on health in absence of the mediating variables, undermines selfefficacy, social integration, and social capital and is not statistically significant with the three mediators included and where all three mediators undermine health. Evaluation of statistical significance for the three indirect effects show all to be statistically significant and the overall extent of "statistical explanation" is quite remarkable at sixty-six percent. In the aggregate, this provides strong empirical support for a novel perspective on the social epidemiology of precarious work organized around its implications for social marginality. That stated, calculation of standardized coefficients show that the effects of self-efficacy are somewhat stronger, while the collective effects of social integration and social capital have roughly equivalent influences. The theoretical challenge is that we do not and cannot know direction of causal influence among these factors or how they may influence exposure to precarious work over the life course. Future research could fruitfully expand consideration of social and social psychological mediators, as well as tease out the relations between them.
Our third contribution is to develop and test the cross-national consistency of the model, specifically with respect to different welfare state regimes. While expectations for health have never been fully articulated, the basic logic is that stronger welfare states should both mitigate exposure to precarious work and buffer its negative consequences through a variety of extra-market supports (Bambra and Eikemo 2009;Kim et al. 2012). Given this expectation, one would anticipate the weakest associations in Scandinavian countries followed by Continental and perhaps Mediterranean countries. Associations should be strongest in Anglo-Saxon and Eastern European countries where market principles and market reliance are more entrenched. In contrast to these expectations, our findings fail to support expectations of welfare state variation and instead suggest strong generality to our conceptual model. The harmful effects of precarious work are largely consistent across Scandinavian, Continental, Anglo-Saxon, and Mediterranean regimes and there is also remarkable consistency for each aspect of the empirical model. For the most part, coefficients are not significantly different across samples. All in all, this very strong evidence that the conceptual model generalizes across most welfare state regimes.
There is, however, one context where the overall model has little salience: Eastern European countries. Here, there is no initial association between precarious work and health and even the simple bivariate relationship is not statistically significant. Thus, there is no mediation to assess, even although selfefficacy and social integration are undermined by precarious work and both selfefficacy and social capital influence SRH. From the standpoint of welfare state epidemiology, one might have expected particularly severe effects of precarious work given the rapid transition to market logics and the overall pervasiveness of marketization in Eastern European social and political life (Fischer and Gelb 1991). The opposite pattern appears to be the case, however. Further analysis (not shown) shows that exposure to precarious work is much, much higher in Eastern European countries and may even be normative and not just for the younger generations who have higher risk, particularly for the multifaceted precarity that combines employment precarity with work precarity. 6 At the same time, heterogeneity within the subset of eastern European countries is substantial and dynamic over time (Hemerijck 2013). Country-specific analyses (not shown) reveal that the null effects are particularly pronounced among former USSR countries (i.e., Lithuania, Russia, and the Ukraine). Given greater exposure to "shock therapy" (Sachs 1994) that were particularly corrosive of local labor conditions in such countries (Stuckler, King, and McKee 2009), precarious work may be less detrimental for perceptions of self, society, and social relations because it is more prevalent and may be even expected. Moreover, if precarious work is anticipated, it may foster strategies for survival, subsistence, and sociality (Clarke 2002). In contrast, precarious work may be particularly stressful in Scandinavian, Continental, Anglo-Saxon, and Mediterranean contexts given that it is both unexpected and does not have inter-generational precursors (i.e., older generations who also have experienced precarious work). Our thesis, however, is speculative and the issue warrants further research.
Several limitations should be noted. First, the ESS data for Round 6 shows evidence of deviations from population characteristics (Koch 2018). Although the extent of this appears small, there is underrepresentation of younger age groups, of non-nationals, and of persons living alone. We also suspect that those respondents who are socioeconomically marginalized may also be underrepresented. As this will be variable across countries given differences in sampling strategies, we cannot formally assess the implications of sampling deviations. Yet, this may actually underestimate the true size of effects across European countries if it truncates the range of the focal independent variable. Second, measures like self-rated health ultimately tells us little about the disease processes that link precarious work to health. Future research could explore how the social and social psychological mechanisms we identify connect to norms and values around diet, exercise, and related health behavior that then connect to body mass and obesity, heart disease, and other cardiovascular risks. 7 There is also an emerging body of work on "diseases of despair" that links chronic exposure to toxic situations to myriad health problems, particularly psychological morbidity and increased suicide risk (Shanahan et al. 2019). Ultimately, further work could consider psychological distress directly or think longitudinally about how exposure to precarious work produces variation in mortality over the life course.
Third, the models necessarily make strong assumptions about causality, including that precarious work and the mediators are exogenous to self-rated health and that the full model's covariates include all plausible predictors of lefthand variables. The models are, however, theoretically specified and future work could usefully explore alternative identification strategies. We also take some comfort given the extensive set of control variables that capture a wide range of possible sources of spuriousness. Fourth, there is likely national differences in socioeconomic policies that are masked by the welfare state classification. As such, our measure of welfare regimes is a proxy for more complex phenomena that cannot be directly assessed, including the diverse range of social policies that characterize the "varieties of capitalism" in Europe. In the end, our effort to understand variation across welfare states in the links between precarious work and health lack depth. In particular, the lack of an association in Eastern European contexts remains a black box. Examination of anomalous cases can be theoretically rewarding and a focus on the Eastern European context that explores not just further structural features of countries but issues of culture, institutions, and history may be particularly illuminating.
In the end, socioeconomic inequality in health is a central pillar of research in social epidemiology and has fostered one of the most influential perspectivesthe "fundamental causes" perspective (Link and Phelan 1995). Key to this perspective is the idea that socioeconomic causes and specific consequences morph and change in symbiotic ways such that deprivations, however measured, are associated with poor health, however measured. Such perspectives are highly useful heuristics, but some of the most interesting work shows how variation in the meaning and dimensionality of socioeconomic status reveals novel associations with health. Our work extends such perspectives by highlighting not just if precarious work undermines health but why it undermines health. In doing so, it ties new aspects of socioeconomic inequalities to important social mechanisms that link them to health (e.g., Palloni 2006). That these consequences are largely divorced from political and economic context suggests both the pervasive and enduring role of socioeconomic differences in the generation of health