1) Economic instability and health outcomes

Evidence from prior research suggests several mechanisms that may underlie the relationship between employment type and health outcomes. First of all, limited economic resources and the financial insecurity of nonstandard workers can affect workers’ health adversely since economic resources from employment are the major source for workers to purchase goods and services critical for maintaining and enhancing good health (Burgard & Lin, 2013). The importance of economic resources is well documented: for instance, unemployment and associated economic strain have a detrimental impact on both physical and mental health outcomes (Price, Choi, & Vinokur, 2002;). Among South Korean workers, low income was found to be among the main factors accounting for high depressive symptoms for temporary workers as well as self-employed individuals (e.g., Song & Kim, 2012).

Beyond objective economic conditions, economic insecurity and unstable labor market status may harm workers’ health. Studies increasingly document that the fear or perception of economic insecurity itself is a risk factor for health. The fear of losing one’s job in the near future is negatively associated with both mental and physical health outcomes (Sverke, Hellgran, & Näswall, 2002). In addition, the direct experience of employment instability (e.g., firm restructuring or layoff) led to poorer physical health (self-rated) and biomedical indicators (e.g., Ferrie, Shipley, Stansfeld, & Marmot, 2002; Geuskens, Koppes, van den Bossche, & Joling, 2012). Various theories, including work stress theories and the job demand control model, also indicate that job insecurity has a negative effect on workers’ health due to lack of control and an imbalance in expectations between employers and employees (e.g., See De Witte et al., 2016 for a review). Under such circumstances, those with an unstable employment status (e.g., part-time or temporary jobs) are thus exposed to immediate or prolonged stress. Evidence shows that stress can cause a series of health problems, including the malfunction of the immune system, which may trigger negative health consequences (e.g., McEwen, 2002).

In fact, the negative effects of perceived job insecurity on both physical (e.g., incidents of coronary heart disease) and mental health are widely found as confirmed by several meta-analyses (e,g., Sverke, Hellgren, & Näswall, 2002; Virtanen et al., 2013). Furthermore, there is evidence that job instability may link nonstandard work to bad health outcomes: according to a meta-analysis of 27 studies, increased job instability among temporary workers was associated with increased psychiatric morbidity and occupational injuries (2006). Among South Korean workers, job insecurity resulting from unemployment and changes from regular to irregular jobs are also associated with increased stress and a higher level of depression (Byun & Lee, 2018; Koh et al., 2004).

2) Psychosocial stressors and spillover effects

Theories on the link between work and health have also focused on the interplay of work with other domains of life. For example, responsibilities in work can intersect with responsibilities in other domains, such as a caregiver role for the elderly or children (Bianchi & Milkie, 2010). When conflicts or spillover in responsibilities in work and family occur, workers tend to develop various health problems such as poor self-rated health, hypertension, obesity, and musculoskeletal problems (e.g., Frone et al., 1996; Grzywacz, 2000).

Besides, research documents that the perception of social position is an important determinant of health. According to the Whitehall study of British civil servants, there is a gradient in health (reduced mortality risks) along with the hierarchy of occupational status and rank (Marmot et al., 1998). The effect of relative positions in occupation and employment on health is also found in other studies on hypertension and heart attacks (e.g., Colhoun, Hemingway, & Poulter, 1998; Moller, Theorell, deFaire, Ahlnom, & Hallqvist, 2005). All this evidence implies that nonstandard work (and self-employment) may harm workers’ health through increased psychosocial stressors from work and negative spillover effects on other domains of life. The importance of employment in determining one’s health may be more salient in South Korea in light of rigid labor market segmentation and limited mobility from nonstandard/self-employment to standard employment. In this context, one’s employment status not only signals one’s current economic condition and labor market status, but it also indicates future career trajectories and socioeconomic status.

In fact, prior research on US workers finds that nonstandard workers, both men and women, report lower job satisfaction due to the lack of control over the work process, high level of work-related stress, and an undesirable work environment (Benach, 2014; Price & Burgard, 2008). Research further documents that those with lower job satisfaction tend to change jobs more frequently than those satisfied with their jobs (e.g., Carmeli & Weisberg, 2006). Recent studies in South Korea find that nonstandard workers are also less likely than standard workers to be satisfied with their jobs and that a job change from standard to nonstandard work results in declining job satisfaction (e.g., Park & Kang, 2016; Sohn, 2011). More important, evidence shows that differences in job satisfaction are found to help explain health disparity between standard and nonstandard workers (e.g., Lee et al., 2017).

Work can also have spillover effects on other domains of life, such as family relationships and social network/resources, which may in turn affect health. For instance, differences in social and family relationships account for health differentials between standard and nonstandard workers, and how stressors in each domain affect health also varies by gender in South Korea (Kim, Lim, Song, & Han, 2012). These study findings suggest that holding a nonstandard job or being self-employed, which is likely involved with undesirable job conditions with limited prospects for career advancement in the South Korean labor market, may have a negative impact on a worker’s overall life satisfaction and well-being.2)

3) Our study

As noted above, one important imitation of prior research is that most studies examining linkages between employment type and health have often relied on a cross-sectional research design (e.g., using a single wave from longitudinal data) and the role of selection and endogeneity was not adequately addressed. It is therefore not clear to which extent the observed effects of mediating factors reflect selection or causality. This is unfortunate in light of growing evidence that unobserved selection may explain substantial health disparity across different employment types (e.g., Byun & Lee, 2018; Kim et al., 2012; Lim et al., 2018). Therefore, evidence is very limited about exact pathways linking specific employment types to health outcomes when the potential role of selection is controlled for.

To address these limitations, in this study we attempt to account for endogeneity using multiple waves of longitudinal data and fixed-effects models. Fixed-effects models control for time-invariant, unobserved individual heterogeneity such as personality, unmeasured health conditions, and motivations/preferences that may select individuals into different employment types and affect the association between work and health (Wooldridge, 2002). By doing so, we evaluate the role of mediators in linking employment status to health outcomes while taking into account endogeneity. In addition, we broaden the literature by evaluating several posited mechanisms that may underlie the associations between employment types and multiple health outcomes, including economic insecurity and psychosocial stressors (see Data and Methods section for details). Employing multiple health outcomes and various mediators will enhance our understanding regarding the health impact of nonstandard work since pathways linking work to health might differ by the dimension of health considered (e.g., Burgard & Lin, 2013).

1) Health outcomes

In this study, we consider both mental and physical health outcomes using depressive symptoms, self-esteem, and self-rated health. First, depressive symptoms are measured based on the CES-D scale (11 items) with the question about how many days a respondent has experienced specific symptoms each week. Four response options range from 0-1, 2-3, 4-5, to 6-7 days. After reverse coding two items with positive wording, the level of depressive symptoms was calculated as a mean response to the 11 items (an alpha reliability is 0.87) in order to address missing responses for some of the items (e.g., Lim & Raymo, 2016). Second, self-esteem is based on 10 times from Rosenberg’s global self-esteem. Respondents are asked on a 4-point scale for 10 items (e.g., I have a positive attitude toward me), ranging from strongly disagree, average/so- so, agree somewhat, to strongly agree. Similar to self-esteem, items with negative wording (e.g., I feel I do not have much to be proud of) were reverse-coded and all items are calculated as a mean response to 10 items (an alpha reliability is 0.80). Third, self-rated health is based on a question about a respondent’s overall physical health conditions. Five response options include very dissatisfied, dissatisfied, neither dissatisfied nor satisfied, satisfied, and very satisfied.3) For these health outcomes, high scores thus refer to higher depressive symptoms/self-esteem, or better physical health conditions.

2) Employment type

Using information on a respondent’s labor force participation and employment status, we categorize a respondent’s employment type into standard employment (reference), nonstandard employment, self-employment, and non-employment. Specifically, wage workers are divided into standard and nonstandard workers: regular wage workers are coded as standard workers whereas temporary wage workers, daily laborers, and those with other (temporary) public jobs are classified as nonstandard workers. Individuals with self-employment include business owners, self-employees, and family workers. Following the classification of the Korean Statistics Bureau and in consideration of a large proportion of self-employment in Korean labor market, we separate self-employment from nonstandard employment in the analyses.4) Non-employment refers to those who are unemployed or out of the labor force.5)

3) Posited Mediators

First, we examine the mediating role of economic insecurity using three measures: (1) annual income (logged), (2) the experience of job instability, and (3) the experience of income instability. For the measure of the experience of job instability, those who experienced a job loss due to lay off, plant closure, or termination of contract over the past year are coded 1. If a respondent is not employed, we coded those who experienced difficulty finding a job due to unstable employment as 1. In addition, those who report “low wages/income or slow business” as a reason for leaving a job (or for those not in the labor force, if they experienced difficulty finding a job since jobs available offered too low wages) was considered to have experienced income instability. Second, as a proxy for psychosocial stressors and spillover effects of work, we include the levels of satisfaction in job and overall life conditions. These measures are based on the question, “How much are you satisfied in this area?” with five response options from very dissatisfactory to very satisfactory. Considering the distribution of responses, we collapse them into three categories, consisting of dissatisfactory, neither satisfactory nor dissatisfactory, and satisfactory.

4) Controls

All models control for various demographic and socioeconomic characteristics that might be related to both one’s employment status and health (e.g., Price & Burgard, 2008). Specifically, we consider age, a place of residence (Seoul, metro cities, other cities, town/villages), marital status (single, married, divorced, widowed), and the number of children (0, 1, 2 or more). In addition, we control for the level of education, which includes less than high school, high school, junior college, and university. Finally, drinking behavior is also considered since alcohol consumption is associated with health outcomes and the pattern of drinking might differ across employment types (Chon, Kim, Cho, & Ryoo, 2010). We measure drinking behavior using the frequency of weekly alcohol consumption, ranging from never, less than 1 time, 2-3 times, to more than 4 times. Except for gender, all covariates are time-varying variables, measured at each interview.

1) Sample characteristics by gender

<Table 1> presents sample characteristics (means/percentages and standard deviations) for the entire sample and separately by gender. The percentage of standard employment is 28.9%, followed by nonstandard (22.2%) and self-employment (16.6%). When broken down by gender, men are much more likely to be employed in standard employment (39.8%) than women (19.1%). It is also clear that women are concentrated in nonstandard employment: about one fourth hold nonstandard jobs while 45% of women are out of the labor force altogether. The proportion of self-employment is much higher for men (21.2%) compared to women (12.5%), which reflects that many displaced workers (due to layoff or involuntary early retirement) are absorbed into self-employment (Keum, 2012). As for health outcomes by gender, women report more depressive symptoms than men (p<0.001), consistent with prior research (Salk, Hyde, & Abramson, 2017). On the contrary, men rate their health better than women (p<0.001). There are no notable gender differences in self-esteem.

2) Health outcomes and mediators by gender and employment status

Due to substantial differences in employment type by gender, we further present mean differences in health outcomes and hypothesized mediators by employment status and gender (Table 2). According to Table 2, standard workers, both men and women, have better health outcomes compared to those in other employment types. But there are some nuanced differences too. Nonstandard workers, for instance, report the highest level of depressive symptoms among employed men, but the levels of depressive symptoms are similar between women in nonstandard employment and self-employment. Besides, the finding that self-employed men and women rate their health worse than employees with nonstandard work suggests that working conditions of self-employment, including long hours and poor financial rewards, might have a negative impact on their health (Ban, 2012; Kim, 2014).

With regard to posited mechanisms, those in nonstandard employment are exposed most to both employment instability and income insecurity, in particular among men. It is also worth noting that nonstandard workers, regardless of gender, are least likely to be satisfied with their jobs and overall life conditions among those in the labor force. Overall, results from Tables 1-2 show that those in nonstandard employment and self-employment fare worse than standard workers in terms of health outcomes and posited mediators considered in this study.

1) Comparing results from OLS models and fixed-effects models

While findings from Tables 1-2 are informative, they do not account for potential confounders and the role of selection into different employment types. Therefore, in Table 3 we present results from simple cross-sectional OLS models and fixed-effects models to illustrate the role of selection in the relationship between employment status and health while controlling for various compositional characteristics (that may be related to both labor force participation and health) and robust standard errors in all models.

Results from OLS models reveal stark health disparity between standard employment (reference category) and other employment types. More important, those in nonstandard employment show clear health disadvantages, regardless of gender across all health measures. In addition, self-employed men and women fare worse than standard workers in all health outcomes: one exception is that self-employed women do not differ from those in standard employment in terms of self-rated health.

Next, results from fixed-effects models show that, compared to simple cross-sectional OLS models, many of the negative associations between nonstandard/self-employment and health disappear in fixed-effects models. For example, the levels of depressive symptoms are no longer significantly higher for those in nonstandard employment and self-employment, regardless of gender (reference is standard employment). It implies that higher depressive symptoms of nonstandard workers, often observed in prior research, might be a reflection of selection of individuals, who have unobserved characteristics that may have a negative impact on mental health, into nonstandard employment (Byun & Lee, 2018; Lim et al., 2018). It, therefore, indicates that the role of selection needs to be taken into account when evaluating health outcomes of nonstandard workers as well as self-employees. At the same time, it is worth noting that nonstandard workers, both men and women, still report lower self-esteem than standard workers even after compositional characteristics—both observed and unobserved—are controlled for.

Interestingly, results from fixed-effects models for self-rated health are quite different from those from OLS models. Significantly negative association of nonstandard work (for both men and women) and self-employment (for men) with self-rated health disappears in fixed-effects models, again indicating the potential role of negative selection. However, women in self-employment report better self-rated health than those in standard employment. It is plausible that self-employed women have unobserved characteristics (e.g, personality or unmeasured health conditions), which may have negative implications for physical health and have been working as a suppressor, and, once they are controlled for, self-employment appears to be conducive to women’s self-rated health (Gordon, 2015).

2) Evaluating hypothesized mediators using fixed-effects models

After examining the potential role of selection in the associations between employment status and health outcomes, we evaluate posited mechanisms underlying these associations using fixed-effects models (Tables 4-6). Baseline models (Model 1) are equivalent to those presented in <Table 3> (Panel B).

First, <Table 4> provides results for depressive symptoms for men and women. Results from the baseline model show that nonstandard work is not associated with depressive symptoms once time-invariant individual heterogeneity is controlled for. It applies to both men and women. As discussed above, these results suggest the role of selection into nonstandard and self-employment, which has negative implications for mental health (see Table 3). In addition, non-employment is positively associated with depressive symptoms, regardless of gender. In the next two models (Models 2-3), two sets of posited mechanisms linking nonstandard work and health are introduced, i.e., the measures for economic instability and the levels of satisfaction in job and overall life conditions (as psychosocial stressors). One notable change is that non-employed women no longer report more depressive symptoms than standard workers in Model 3. This change indicates that non-employed women’s lower levels of job/life satisfaction (Table 2) account for their significantly higher depressive symptoms than standard workers observed in Model 1. In addition, we find that for men, economic insecurity and lower job/life satisfaction jointly explain a higher level of depressive symptoms among the non-employed (Model 4).

Next, Table 5 presents results for self-esteem. Results from Model 1 show that both men and women in nonstandard employment have significantly lower self-esteem compared to those in standard employment (p<0.01), even after controlling for observed and unobserved compositional characteristics. According to results from the subsequent models (Model 3), for men, dissatisfaction in job/life account for lower self-esteem among nonstandard workers observed in Model 1. As seen in Table 2, the levels of job/life satisfaction are lowest for nonstandard workers among employed men, which appears to fully mediate the negative association between nonstandard employment and self-esteem among men found in Model 1.6) For women, nonstandard work is still negatively associated with self-esteem when the levels of job/life satisfaction are taken into account but the coefficient drops in magnitude (Model 3). This result implies job/life satisfaction partially mediates the relationship between nonstandard work and women’s self-esteem as discussed in the previous section.

Another interesting finding is that self-employed men have higher self-esteem than standard workers with the inclusion of satisfaction measures in Model 3 (p<0.05). It appears that self-employed men’s lower job/life satisfaction than men holding standard jobs (see Table 2) were suppressing the relationship between self-employment and self-esteem. With the presence of suppressor variables, the total effect (i.e., the effect of self-employment on self-esteem) becomes close to zero and a significant association is revealed only when a mediator (job/life satisfaction in this case) is included in the model (Gordon, 2015; Wheaton, 1985).7) Therefore, once the levels of job/life satisfaction, i.e,. mediators that behave as suppressors between the predictor and outcome variable, are controlled for, the positive association between men’s self-employment and self-esteem emerges. It is also worth noting that female nonstandard workers’ lower self-esteem (compared to a reference group) disappears in Model 4: we find that economic insecurity and dissatisfaction in job/life jointly account for their diminished self-esteem.

Lastly, Table 6 presents results for self-rated health. At baseline (Model 1), there are no differences in self-rated health among employed men when unobserved heterogeneity is controlled for. As discussed in Table 3, this finding implies the presence of negative selection (e.g., health and other unmeasured characteristics) into nonstandard and self-employment among men. It is, however, important to note that self-employed men rate their health better than those in standard work when the levels of job/life satisfaction are introduced in Model 3. Similar to results for self-esteem (Model 3, Table 5), self-employed men’s lower job/life satisfaction (relative to those in standard jobs) seems to have been a suppressor for the association between self-employment and self-rated health.

In contrast, self-employed women report better self-rated health than those in standard employment at baseline, and economic insecurity or levels of job/life satisfaction do not account for their better physical health (Models 2 and 3). Interestingly, when life satisfaction is taken into account, female nonstandard workers now have better health than standard workers (Model 3, supplementary results). This change also implies the suppressing effects of female nonstandard workers’ low life satisfaction in self-rated health. According to Table 2, the proportion of nonstandard workers reporting their life to be satisfactory (49.1%) is the lowest among all women, even lower than non-employed women (54.7%).