Abstract
There is little theoretical understanding of why educational inequalities in depression are larger in some countries than in others. The current research tries to fill this gap by focusing on the way in which important labor market processes, specifically upgrading and polarization, affect the relationship between education and depression. Analyses are based on a subsample, aged between 20 and 65, in 26 countries participating in the European Social Survey (N = 56,881) in 2006, 2012, and 2014. The results indicate that educational inequalities are lower the more a country’s labor market is upgraded, and we suggest that this is a consequence of an amelioration of the labor market position of the lower educated. Analysis shows that polarization is related to more educational inequalities in depression for women. However, central mediating processes are not confirmed, and new perspectives on these mechanisms will be discussed.
Introduction
Given the importance of education as a stratification mechanism (Gesthuizen, Huijts, and Kraaykamp 2012; Miech and Hauser 2001), social scientists have paid a great deal of attention to educational inequalities in depression. Not surprisingly, the positive relationship between education and depression is a consistent finding in social epidemiology. Research within this domain shows that lower-educated people are more likely to have higher rates of depressive symptoms compared with the higher educated (Dudal and Bracke 2016; Lorant etal. 2003). These disparities are often explained by referring to a direct socialization effect (e.g., learned effectiveness) and an indirect allocation effect on the labor market (e.g., higher income, better working conditions) (Mirowsky and Ross 2005; Ross and Mirowsky 1999).
In recent years, increasing interest has been shown in the cross-national variability of the relationship between education and depression, and European comparative research confirms that the size of these inequalities varies between countries (Dudal and Bracke 2016; von dem Knesebeck, Pattyn, and Bracke 2011). This opens up the question of which macro-societal factors are able to explain why, in this regard, the higher educated are relatively better off in some countries than in others. Despite this growing interest, empirically tested theoretical explanations are scarce as the majority of comparative research focuses on describing the size of the educational inequalities rather than explaining the disparities (Gesthuizen etal. 2012).
Nevertheless, Bracke, van de Straat, and Missinne (2014) were the first to contextualize educational inequalities in depression and showed that in economies with lower returns on education—in terms of income and unemployment risk—the higher educated are less able to translate their additional education into better mental health outcomes. These findings offer a good direction for further theoretical research for two reasons. First, if labor market returns on education are related to educational inequalities in depression, the next question is which labor market processes drive these returns and consequently, whether these processes can be linked to the mental health of the different educational groups. Second, a focus on labor market conditions has also been fruitful in research into inequalities in other health outcomes (Gesthuizen etal. 2012). Hence, these insights could be used to gain more knowledge of the cross-national variation in educational inequalities in depression.
This study focuses on how ongoing processes of the upgrading and polarization of European labor markets are related to depression and educational inequalities in depression. We try to link literature on important structural labor market changes to an increasing body of literature that tries to contextualize educational inequalities in depression.
Expansion of Tertiary Education and Occupational Change
After the second world war, scholars observed an acceleration of the expansion of tertiary education in almost all European countries (Schofer and Meyer 2005). Tertiary education can now be seen as one of the most central institutions in modern societies, and the supply of higher-educated people has—viewed historically—never been so high (Schofer and Meyer 2005) and is still increasing (OECD 2016). Not surprisingly, considerable scientific attention has been paid to the causes and consequences of this ongoing process (Powell and Solga 2011; Schofer and Meyer 2005).
Upgrading of the Labor Market
Given the context of the rapid expansion of tertiary education, we might wonder if the higher educated are still able to use their educational level to obtain access to the health-promoting jobs that have been typically assigned to them. To a large extent, the answer is dependent on how the demand for higher-educated labor has changed during recent decades. Some scholars claim that technological development has led to an increase in the demand for higher-educated labor (Aberg 2003; Berman, Bound, and Machin 1998). The suggestion is that technological changes are selective and favor the position of the higher educated. These skill-biased technological changes can be seen as the main factor in explaining occupational changes during recent decades (Oesch and Menes 2011). Additionally, it is suggested that supply can create its own demand (Machin and Manning 1997; Verhaest and Van der Velden 2013) as it is theoretically argued that a substantial group of higher-educated people lowers the investment costs for employers to create appropriate jobs (Machin and Manning 1997). In general, scholars agree that the expansion of tertiary education has gone hand-in-hand with an upgrading of the labor market and has enhanced the employment opportunities for the higher educated (Aberg 2003; Oesch and Menes 2011). Recent research on occupational change shows that occupational upgrading has occurred in almost all European countries, but the extent seems to vary between them (Fernandez-Macias 2012; Goos, Manning, and Salomons 2009).
Link with health
Since it is largely acknowledged that a person’s individual labor market position plays an important mediating role in the translation from education to mental health (Mirowsky and Ross 2003; Ross and Mirowsky 1999), it is likely that upgraded labor markets moderate the relationship between education and depression. For the higher educated, it seems that as the increasing supply of higher-educated people has been hand-in-hand with an increasing demand for higher-educated labor, it is likely that they have still ended up in jobs with health-promoting characteristics. Hence, it is arguable that an upgraded labor market acts as a facilitator in the translation from education into better mental health for the higher educated.
Furthermore, there are some reasons to suggest that the upgrading of the labor market has not onlybeen ameliorative for the higher educated but also had some beneficial results for the lower and intermediately educated. First, the shift toward a modernized labor marketalso involves an improvement of job quality and conditions at the lower levels of the labor market structure (Gesthuizen etal. 2012; Gesthuizen and Scheepers 2010). This process has had profound implications for the lower educated as it has been observed, for example, that the upgrading of the labor market has reduced their economic vulnerability (Gesthuizen and Scheepers 2010), which is likely to ameliorate the mental health status of this group (Mirowsky and Ross 2003). This implies that lower- and intermediately educated people do not necessarily end up in jobs with more detrimental health consequences but also take labor market positions with health-promoting features (Gesthuizen and Scheepers 2010). In addition, these processes arefurther enhanced as employers started—however initiated by an aim to increase workers’ productivity—to pay more attention to psycho-social factors and well-being at work (Guest 2002; Sparks, Faragher, and Cooper 2001).
A second argument can be found in the literature on diffusion of capital. It is likely that the upgrading of the labor market led to a higher-educated composition in the workplace and created a context that allows the diffusion of different forms of capital. In this vein, it is suggested that higher-educated people have specific forms of cultural health capital that enable them to achieve a better mental health status (Pinxten and Lievens 2014; Shim 2010). It is possible that the labor market, or the workspace, to be more specific, acts as a facilitator in the process of the accumulation and interchangeability of cultural health capital and can elevate the mental health status of the lower and intermediately educated.
It is clear that upgraded labor markets can have important advantages for the entire labor market spectrum. However, we expect that upgraded labor markets can alter the negative mental health consequences that are usually related to the labor market position of the lower and intermediately educated, and accordingly we hypothesize that:
Hypothesis 1: Educational inequalities in depression will be lower in more upgraded labor markets.
Polarization of the Labor Market
Despite the general trend of upgrading, it is argued that this new knowledge economy has also led to the creation of new low-skilled jobs in the lower segments of the labor market (Gesthuizen and Wolbers 2010; Oesch and Menes 2011). These jobs are mostly in the personal service sector and can be seen as complementary to the high-skilled jobs (Goos etal. 2009; Oesch and Menes 2011). This shift is in line with the routinization hypothesis, which argues that technological development has replaced routine labor in the middle segments of the labor market structure (Goos etal. 2009). Hence, in some countries, there has been a shift toward a more polarized employment structure, characterized by more jobs at both the lower and the higher ends of the labor market and fewer jobs in the middle segment (Fernandez-Macias 2012; Goos etal. 2009). The extent to which this occurs seems to vary between countries—and is more profound in some countries, including Belgium, the Netherlands, and the United Kingdom (Fernandez-Macias 2012, Goos etal. 2009)—but it is likely that this process will develop in other countries (Sparreboom and Tarvid 2016).
Link with health
It is arguable that the processes of polarization negatively affect the mental health of the lower educated for several reasons. First, these new jobs are located in the lowest segments of the labor market and are ascribed a very low status (Oesch and Menes 2011). This can negatively influence the status attainment of the lower educated and hence their mental health (Adler etal. 2000). An additional argument can be found in recent literature on the flexibilization of labor markets. This process introduced, among other things, a shift toward new forms of employment (part-time work, short-term contracts, etc.) and has led to increased job insecurity among the workers affected (Scherer 2009). This flexibilization process affects the workforce unequally as it has been shown that the trend is more profound for workers in low-skilled service jobs, which are typical for polarized labor markets and the lower educated (Oesch and Menes 2011). Increasing evidence points to the detrimental mental health effects of this shift (De Witte 1999; Ferrie etal. 2002).
In addition to mental health effects for the lower educated, we expect that polarization also affects the mental health status of the intermediately educated as it is likely that this process will also influence their labor market position. If it is becoming more difficult for the intermediately educated to find jobs in the middle segment of the labor market, we should consider where these people end up in the labor market structure. According to labor market theories, it is likely that if they cannot find a job that fits their educational attainment, they will move down the structure and accept a lower-level job (Oesch and Menes 2011). It is expected that this will lead to a deterioration of their mental health as it has been shown that job-education mismatch is related to a higher likelihood of having enhanced depressive symptoms (Bracke, Pattyn, and von dem Knesebeck 2013) and reduced job satisfaction (Verhaest and Omey 2009).
These mechanisms could also affect the lower educated as they in turn will have to move further down the hierarchy or might even be totally excluded from the labor market (Goos etal. 2009), which is negatively linked to various mental health outcomes (Mathers and Schofield 1998; Paul and Moser 2009). This process can be explained by job competition theory, which states that if the lower educated have to compete for the same jobs with people who have better educational credentials—which is likely in a context of polarization—the latter will win this competition because of their level of education (Gesthuizen and Wolbers 2010). It is argued, for example, that higher educational levels reduce training costs for employers. Hence, higher-educated people can be assigned a more varied range of tasks, which gives them a competitive advantage relative to the lower educated. In addition, more favorable characteristics (better work attitude, better health, etc.) are ascribed to higher-educated people (Sanders etal. 2011).
With regard to the higher educated, if there is an increasing proportion of jobs at the upper end of the labor market, it is likely that this group will still be employed in accordance with their educational level. Hence, even in polarized labor markets, it is arguable that people with a higher level of education can translate this into better mental health.
Based on these theoretical propositions, we argue that more polarized labor markets can strengthen the negative mental health consequences that are usually linked to the labor market position of the lower and intermediately educated. Hence, it is expected that this further widens the educational gap in depression, which leads to the hypothesis that:
Hypothesis 2: Educational inequalities are larger the more a labor market is polarized.
Methods
Data
Our analyses are based on the European Social Survey (ESS), rounds 3, 6, and 7 (2006, 2012, and 2014). Respondents were selected using strict probability samples of the resident national population aged 15 or above living in private households. Data were gathered via face-to-face interviews. Response rates range from 33.76 percent (Germany, ESS 6) to 77.12 percent (Portugal, ESS 6). The sample we use is restricted to respondents aged 20 to 65. To retain only those who were in the labor market, respondents who were still studying, homemakers, permanently ill, or disabled were removed from the sample. Additionally, respondents with information lacking for the variables used (2,469; 4.2 percent) were also excluded. The final sample comprises 56,881 respondents from 26 countries.
Variables
Dependent
Depression is measured using an eight-item version of the Center for Epidemiologic Studies Depression Scale (CES-D) (Radloff 1977). Respondents were asked to indicate how often in the week before the survey they had felt or behaved in a particular way (felt depressed, felt that everything was an effort, slept badly, felt lonely, felt sad, could not get going, enjoyed life, felt happy). Response categories range from none or almost none of the time (0) to all or almost all of the time (3). The total CES-D8 score ranges from 0 to 24, with higher values indicating a greater amount of depressive symptoms. Missing values were handled by respondent mean substitution on the condition that at least five items of the scale had been answered. Previous research supports the measurement equivalence of the shortened eight-item CES-D scale among European respondents, which allows us to use this variable in a cross-national design (Missinne etal. 2014; Van de Velde etal. 2010). Cronbach’s alpha for the CES-D8 scale ranges from .7 (Denmark) to .86 (Bulgaria) for men and from .76 (Denmark) to .87 (Poland) for women.
Independent
Individual-level variables
From a comparative point of view, Schneider (2007, 2010) consistently argued that the use of years of education is not optimal for comparative research within the European context. Therefore, to enhance the comparability of our results, we make use of the International Standard Classification of Education 1997 (ISCED-97) to measure educational attainment. Additionally, this operationalization allows us to investigate how specific types of education are related to each other in terms of mental health benefits. This classification differentiates between seven levels of education, ranging from pre-primary education (Level 0) to second stage of tertiary education (Level 6). For our purposes, we recoded this variable into the following categories: (1) lower educated (ISCED 0–2), (2) intermediately educated (ISCED 3–4), and (3) higher educated (ISCED 5–6).
Macro-level variables
The extent to which a labor market is upgraded is assessed by the proportion of jobs for the higher educated in the total labor force. In accordance with the definitions and categorizations of the International Labor Organization (2012), we conceptualize such jobs as those that fit into ISCO-08 Groups 1 to 3 (Sparreboom and Tarvid 2016). The process of polarization deals with the proportion of labor market positions in the lowest and middle segments of the labor market structure, which implies that we could either use the share of jobs for the lower or the intermediately educated to measure polarization. However, we decided to use the proportion of jobs for the lower educated (ISCO-08 Group 9) for two reasons. First, multicollinearity tests with the alternative operationalization (i.e., the proportion of jobs for the intermediately educated) showed that upgrading and polarization could not be tested in one statistical model and in fact relate to the same concept. Operationalizing polarization based on the proportion of jobs for the lower educated allows us to simultaneously assess the effect of both variables. Second, it has been shown that upgrading has occurred in the large majority of European countries. For some countries, this has involved a decrease in the share of jobs for both the lower and the intermediately educated (Fernandez-Macias 2012). However, for some countries, this process has gone hand-in-hand with an increasing proportion of jobs in the lower segments. Hence, polarized labor market structures can be seen as a specific form of upgraded labor markets, which implies that a focus on the proportion of jobs for the lower educated is more valid in substantial terms. All macro-level data were retrieved from the Key Indicators of the Labour Market (KILM) database from the International Labour Organization.
Relevant control variables
Age, marital status, equivalent household income, work status, parental education, and over-education are included as individual-level control variables. Age is measured in years, and additionally, its square term (age2) is included to control for the nonlinearity in the relationship between age and depression. Marital status indicates whether a respondent was married or cohabiting (reference category), divorced or separated, widowed, or single. Based on the country-specific median income, we capture the equivalent household income as a categorical variable: lower than 50 percent of the median income, 50 percent to 80 percent, 80 percent to 120 percent, or more than 120 percent. An additional “missing” category is included as 23.7 percent of the respondents did not answer the question on household income. Work status indicates whether respondents were employed (reference category), unemployed looking for work, or unemployed not looking for work. To measure parental education, we make use of the ISCED-97 classification: primary or lower (reference category, ISCED 0–1), lower-secondary completed (ISCED 2), upper-secondary completed (ISCED 3), postsecondary nontertiary completed (ISCED 4), and tertiary education completed (ISCED 5–6). We also include an extra category if the level of education is unknown. The highest educational level was chosen if the two parents had different levels. Over-education is measured in accordance with the “realized matches” method (Bracke etal. 2013) and is applicable if someone’s educational level is more than one standard deviation above the mean level within their occupational category.
At the macro level, we control for GDP per capita as it is linked to several health outcomes (Bracke etal. 2014; Gesthuizen etal. 2012) and educational expansion (Schofer and Meyer 2005). To control for the cyclic component of the economy, we include a country’s unemployment rate in the analysis. The data for these control variables were also retrieved from the KILM database.
Statistical Methods
Since we make use of repeated cross-sectional data from multiple countries, our data are clustered in time and countries. Hence, the use of multilevel modelling with a three-level structure is appropriate (Schmidt-Catran and Fairbrother 2016). The questions on depressive symptoms are only available for three rounds of the ESS. This means that we only have three points in time, which does not allow us to treat this as a separate, higher level (Stegmueller 2013). To solve this problem, we rely on the models of Fairbrother (2014), who clusters individuals (Level 1) within country-periods (Level 2), which are again clustered in countries (Level 3). This should result in 79 Level 2 units (3 points in time × 26 countries), but due to data restrictions (labor market information is not available for some points in time for some countries), we have to limit our model to 59 Level 2 units, which is a sufficient amount to treat this as a higher level. Additionally, a random slope for the educational attainment variable is included to take into account the unobserved heterogeneity in the association between education and depressive symptoms (Albert and Davia 2011).
The analysis process started with estimating a baseline model, which only includes the educational attainment variables and a series of individual control variables (age, age2, marital status, and parental education). Subsequently, we estimated cross-level interactions with the macro labor market variables and educational attainment to test our central hypotheses. In a third model, we added a set of mediating variables (household income, employment status, and over-education) to see if and how the effect sizes were altered.
Results
Descriptives
Tables 1 and 2 show the descriptive analysis both for individual- and country-level characteristics. The mean age of our sample is 42.4 years (SE = 11.4) and comprises 48.2 percent women. The majority are married (55.9 percent), and 8.4 percent have an income below 50 percent of the country median. Over-education is observed for, respectively, 15.6 percent of the sample. The proportion of lower-educated people is 16.3 percent, while those of the intermediately and higher-educated people are, respectively, 52.2 percent and 31.5 percent.
Descriptive Statistics.
Note. European Social Survey rounds 3, 6, and 7 (N = 56,881). Weighted data.
Country Characteristics, Descriptive Statistics.
Note. European Social Survey rounds 3, 6, and 7 (N = 56,881). Weighted data.
The highest depression scores are obtained for Hungary (men mean = 7.21; women mean = 7.64) while the lowest are found for Norway for men(mean = 3.8) and for Iceland for women (mean = 3.9). Furthermore, Table 2 shows that there is large cross-national variation in the proportion of jobs for the lower and higher educated. With regard to the former, the largest proportion is found in Ukraine (24.9 percent), whereas the lowest is in Norway (4.2 percent). At 47.3 percent, the Netherlands has the highest proportion of jobs for the higher educated, whereas Portugal (26.1 percent) has the lowest.
Educational Attainment and Depression
The results of the multilevel analysis, presented in Tables 3 and 4, clearly show the presence of educational inequalities in depression. As can be seen from Model 1, for men, both the intermediately (Bintermediate = −.787; SE = .12; p < .001) and higher educated (Bhigher = −1.17; SE = .17; p < .001) have a lower likelihood of greater depressive complaints in comparison with the lower educated. As expected, the size of these inequalities is higher among women (Bintermediate = −.894; SE = .09; p < .001; Bhigher = −1.433; SE = .12; p < .001). Moreover, based on the variations in the slopes for the intermediately (men γ = .295, SE = .099, p < .05; women γ = .109, SE = .066, p > .05) and higher educated (men γ = .75, SE= .27, p < .05; women γ = .488, SE = .194, p < .05), we can confirm the cross-national variation in the size of educational inequalities in depression. The estimated 95 percent plausible value ranges for the intermediately educated slopes are −1.85 < B < .27 (= −.79 ± 1.96 × [.3]1/2 = −1.01 ± 1.06) for men and −1.541 < B < −.247 for women. The slope for the higher-educated variable has a range of −2.87 < B < .53 for men and −2.803 < B < .063 for women. Moreover, there is substantial variation in the size of educational disparities if we take into account the extreme ranges of the labor market variables (i.e., the least or most upgraded/polarized labor market). As shown in Table 5, we see that, for example, the coefficients for the intermediately educated variables are −.38 (men) and −.13 (women) in the most upgraded labor markets, while this rises to −1.237 (men) and −1.399 (women) in the least upgraded labor markets.
Depressive Symptoms Regressed on Individual and Labor Market Variables for Men.
Note. European Social Survey (ESS) rounds 3, 6, and 7 (N = 29,437); weighted data. Models 2 and 4 controlled for age, age2, marital status, parental education, ESS round, GDP per capita, and unemployment rate. Models 3 and 5 controlled for age, age2, marital status, parental education, ESS round, GDP per capita, unemployment rate, household income, work status, and over-education.
p < .05. ***p < .001.
Depressive Symptoms Regressed on Individual and Labor Market Variables for Women.
Note. European Social Survey (ESS) rounds 3, 6, and 7 (N = 27,444); weighted data. Models 2 and 4 controlled for age, age2, marital status, parental education, ESS round, GDP per capita, and unemployment rate. Models 3 and 5 controlled for age, age2, marital status, parental education, ESS round, GDP per capita, unemployment rate, household income, work status, and over-education.
p < .05. **p < .01. ***p < .001.
Educational Attainment Coefficients Range by Degree of Upgrading and Polarization.
Minimum = value for the most upgraded labor market; maximum = value for the least upgraded labor market.
Minimum = value for the least polarized labor market; maximum = value for the most polarized labor market.
Upgrading and Polarization
Model 2 shows a significant negative effect for the upgrading variable for both men (Bupgrading = −.074; p < .05) and women (Bupgrading = −.069; p < .05) and indicates that the higher the proportion of jobs for the higher educated, or the more a labor market is upgraded, the fewer depressive complaints the lower educated will have. As revealed by the cross-level interactions, the effect for the intermediately educated is slightly higher but remains negative (men B = −.074 + .036 = −.038, p > .05; Women B = −.027, p < .05). For men, the effect for the higher educated increases even more but is still negative (B = −.012, p < .05), while for women the effect becomes zero (B = 0, p < .01). Hence, these cross-level coefficients point to the observation that educational inequalities in depression are lower in labor markets that are more upgraded. Additional analyses (results available on request), however, make it clear that the effect of upgrading is only significant for the lower educated, while the coefficients for the intermediately and higher educated remain nonsignificant after changing the reference category. In a subsequent model (Model 3), we tried to explain these effects by adding a set of mediating variables (household income, employment status, and over-education). We see that the effects of upgrading are partly explained by these variables as the effect for the lower educated increases or disappears (men Bupgrading = −.057, p < .05; women Bupgrading = −058, p > .05). The variance components from Model 2 show a reduction in the aforementioned between-country variation for the intermediate and higher-educated slopes and indicate that upgrading variables are able to explain a part of the cross-national variation in the size of educational inequalities (men γintermediate = 3.7 percent [(.295 − .284)/.295) × 100], γhigher = 5.1 percent; women γintermediate = 35.8 percent, γhigher = 29.5 percent).
The analyses for the polarization variable are presented in Model 4 and show no main or interaction effects for men. On the other hand, we see a positive effect for women (Bpolarization = .093; p < .05), which indicates that a higher proportion of jobs for the lower educated is associated with increased depressive complaints for the lower educated. The cross-level interactions show that this effect is smaller for the intermediately educated (Bpolarization×intermediate = −.096; p < .05) but does not differ for the higher educated (Bpolarization×higher = −.058; p > .05). In addition, we find that polarization only has an effect for the lower educated (results available on request). Adding a set of mediating variables did not alter the size of the coefficient (Bpolarization = .095; p< .05), which means that these variables do not explain the detrimental effect of polarization for the lower educated. Lastly, a reduction of the between-country variation in the slope for intermediately educated women is observed and indicates that the polarization variables explain a part of the cross-national variation in educational inequalities in depression between these groups (γintermediate = 28.4 percent, women).
In the vein of these main results, additional analyses (results presented in the appendix available in the online journal) produce some interesting findings. First, we find that educational inequalities in the risk of being poor (measured by household income) between the lower and intermediately educated are smaller the more a labor market is upgraded. The same observation is found for the lower and higher educated but is only statistically significant for men. Second, the additional analyses show that polarization is not related to an elevated risk of being over-educated for the intermediately educated and not related to an enhanced unemployment risk for the lower educated. These findings counteract the central processes that are usually linked to polarization but could help us understand why we only find limited evidence for the polarization hypotheses. This issue will be further addressed in the following section.
Conclusion and Discussion
This research started out from the observation that education is a central element of social stratification processes in contemporary societies (Miech and Hauser 2001). Hence, a focus on education seems inevitable when seeking to explain social disparities in health. Despite the fact that social-epidemiological research consistently stresses the importance of education for health—even more than income and social class—there is little research that contextualizes these inequalities in order to explain them (Bracke etal. 2014; Gesthuizen etal. 2012; Goesling 2007; Mirowsky and Ross 2003). The current research explicitly addresses this issue and tries to contextualize educational inequalities in depression by focusing on structural labor market characteristics as these influence the way different educational groups are allocated in the labor market.
The main finding is that two important structural labor market changes—namely, upgrading and polarization—do play a role in the translation from education to mental health and hence in the magnitude of educational disparities in mental health. With regard to upgrading, we see that even after controlling for GDP per capita, the educational gap in depression becomes smaller the more a labor market is upgraded, which leads to confirmation of our hypothesis. It is likely that the decreasing inequalities are the result of an amelioration of the situation for the lower educated as we only find an effect of upgrading for this group. Hence, it seems that upgraded labor markets can have important advantages for the lower educated and as a consequence can reduce the educational gap in depression. As stipulated in the theoretical outline, labor markets alter the relationship between education and mental health in multiple ways by intervening in a complex web of mediating relationships linking education to mental health. In this regard, the additional analyses make it clear that, for example, income inequalities between different educational groups are lower the more a labor market is upgraded. Hence, it is likely that upgraded labor markets improve the mental health status of the lower educated by reducing the income gap. These results add some new perspectives to empirical research on the consequences of skill-biased technological change. Based on data from the 1970s and 1980s, scholars have warned about the detrimental consequences of skill-biased technological change for the lower educated as, among other things, rising income inequalities have been observed for a substantial number of countries (Berman etal. 1998; Card and DiNardo 2002). Based on more recent data, our research and that of Gesthuizen and colleagues (2012) points to contrary findings as an amelioration of the conditions and (mental) health outcomes for the lower educated is observed.
On the other hand, evidence for the polarization hypotheses is limited as we only find effects for women. These results indicate that educational disparities in depression are more profound—even after controlling for the process of upgrading—the greater the proportion of jobs for the lower educated in the labor market, which is likely to be the result of a deterioration of the situation for this group. It was theoretically expected that factors such as income, job-education mismatch, and employment status would explain these effects, but this is not the case. Moreover, the additional results that polarization is not related to an elevated risk of being over-educated for the intermediate group and not related to a worsening of the employment opportunities for the lower educated raise some questions about the crowding-out argument, which is seen as an important process in polarization literature. Both observations might suggest that jobs are not as variable as assumed by the crowding-out argument. It could be the case that under some circumstances, for example good unemployment protection and deteriorating job characteristics at the bottom of the labor market, the intermediately educated prefer to accept being unemployed rather than moving further down the labor market structure (Caliendo, Tatsiramos, and Uhlendorff 2013; Katz and Meyer 1990; Marimon and Zilibotti 1999). This line of reasoning is in congruence with the observation of Gesthuizen and Wolbers (2010) that an oversupply of the higher educated, which is theoretically assumed to enhance processes of crowding out and especially in polarized labor markets, does not widen the gap in employment exit risk between the lower and intermediately educated.
The fact that we only find associations for women might be related to the nature of these new jobs as these are located more in the personal service (catering, cleaning, etc.) and care sectors (Dwyer 2013; Goos etal. 2009). Despite decreasing gender inequalities in some segments of the labor market, we still see that women—and especially the lower educated—have a higher likelihood of ending up in these job sectors (England 2005, 2010).
Before we continue with suggestions for additional research, we should point out some limitations of the current research. First, we examine broad categories in the labor force and ignore within-group heterogeneity. For example, it has been shown that there is large variation in the mental health status of managers, with some managerial positions having worse mental health outcomes in comparison with manual workers (Muntaner etal. 2003). However, despite this heterogeneity, our results point to systematic between-group differences according to educational level. Second, we measure upgrading and polarization in the twenty-first century. It is likely that for a substantial proportion of the population under study, the career trajectory is influenced by processes or events at the beginning of their working life. This could be counteracted by applying another methodological approach (e.g., age-period-cohort analysis), which would allow us to expand the analytical timeframe. However, it is likely that data restrictions will occur as these data are not available for every country under study.
While adding theoretical insights to the increasing body of literature on educational inequalities in depression, our results also offer some interesting paths for future research. First, as we have stipulated some new perspectives onthe crowding-out argument, it would be of interest to test empirically under which conditions the intermediately educated are willing to take a lower-skilled job and how these conditions affect their mental health status. A focus on how different institutional contexts interplay with these processes could be fruitful as it has already been shown that, for example, differential wage setting institutions influence the formation of a polarized labor market structure (Oesch and Menes 2011). Second, in the theoretical outline, we mainly focus on how technological changes and the expansion of tertiary education are linked to processes of upgrading and polarization of the labor market. Additional research could pay attention to other processes that are at work. For instance, a focus on migration patterns and how they influence the disparities between and within educational groups could be fruitful as it is suggested that migration is linked to the creation of low-ranking service jobs at the bottom of the labor market and can increase competition for jobs for the lower educated (Oesch and Menes 2011). Third, as our analysis points to some positive effects of occupational upgrading for the lower educated, it could be fruitful to examine whether these findings hold true for some subgroups within this broad category. Future research could investigate whether, for example, immigrants benefit from these labor markets as empirical research consistently points to more negative mental health outcomes for immigrants (Missinne and Bracke 2012) and discrimination on the labor market (Boeri etal. 2015; Carlsson and Rooth 2007). Lastly, despite an upgrading of the labor market, several authors point to more negative mental health outcomes for the higher educated (Levecque et al. 2017). Additional research could test whether and how recent processes of flexibilization of the labor marketalter the consequences of upgraded economies. A focus on increasing pressures on working conditions and how they influence mental health could be fruitful and expand our research as we only focus on the proportion and not the qualitative characteristics of these jobs (Tausig and Fenwick 2016).
Footnotes
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Funding for this research was provided by the Research Foundation Flanders (FWO) (G0B8714N).
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
