The Sociological Perspective (or, Imagination) emphasizes that our social landscapes have an impact on how we think about th

Anxious? Depressed? You might be suffering from capitalism: contradictory class locations and the prevalence of depression and anxiety in the USA Seth J. Prins1, Lisa M. Bates1, Katherine M. Keyes1

and Carles Muntaner2

1Department of Epidemiology, Columbia University, USA 2Bloomberg Faculty of Nursing, Dalla Lana School of Public Health, and Department of Psychiatry, University of Toronto

Abstract Despite a well-established social gradient for many mental disorders, there is evidence that individuals near the middle of the social hierarchy suffer higher rates of depression and anxiety than those at the top or bottom. Although prevailing indicators of socioeconomic status (SES) cannot detect or easily explain such patterns, relational theories of social class, which emphasise political-economic processes and dimensions of power, might. We test whether the relational construct of contradictory class location, which embodies aspects of both ownership and labour, can explain this nonlinear pattern. Data on full-time workers from the National Epidemiologic Survey on Alcohol and Related Conditions (n = 21859) show that occupants of contradictory class locations have higher prevalence and odds of depression and anxiety than occupants of non- contradictory class locations. These findings suggest that the effects of class relations on depression and anxiety extend beyond those of SES, pointing to under-studied mechanisms in social epidemiology, for example, domination and exploitation.

Keywords: social class, epidemiology, mental health and illness, social determinants of health

Introduction

Social disadvantage is associated with a higher risk of most adverse mental health outcomes (Dohrenwend 1990, Dohrenwend and Dohrenwend 1969, Faris and Dunham 1988, Hollings- head and Redlich 1953, Muntaner et al. 2013). Indeed, it has been firmly established that there are disparities in both mental and physical health across traditional measures of socioeconomic status (SES) such as income, educational attainment and other indicators of social rank (Lynch and Kaplan 2000, Muntaner et al. 2013). These measures capture – and their theoretical under- pinnings predict – an inverse linear, gradational relationship between SES and physical and mental illness (Mackenbach et al. 1997, Marmot and Smith et al. 1991). There is evidence, however, that individuals towards the middle of social hierarchies may actually suffer higher rates of internalising affective disorders, such as depression and anxiety, than those at either © 2015 Foundation for the Sociology of Health & Illness. Published by John Wiley & Sons Ltd., 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA

Sociology of Health & Illness Vol. 37 No. 8 2015 ISSN 0141-9889, pp. 1352–1372 doi: 10.1111/1467-9566.12315

the top or the bottom (Muntaner et al. 1998, 2003, Wohlfarth 1997). Such patterns are neither detected nor easily explained using standard approaches.

Using traditional measures of SES, research since at least the 1930s has consistently docu- mented the finding that mental illnesses are more common among those with lower levels of income, education and occupational prestige than in those with higher levels of these indica- tors (Dohrenwend and Dohrenwend 1969, Faris and Dunham 1988, Hollingshead and Redlich 1953, Kessler and Cleary 1980, Link et al. 1993). Studies such as Whitehall I and II have generated important insights about the social production and distribution of health disparities by establishing social gradients in physical and mental health (Marmot et al. 1991, Marmot and Brunner 2005, Stansfeld et al. 1995, 1998). Early on, these studies iden- tified an unexplained residual social gradient after accounting for standard risk factors (Mar- mot et al. 1978) and this motivated considerable scholarship and debate on potential additional causes and mechanisms that do not operate through the pathway of behaviourally mediated proximal risk factors. These generative insights prompted substantial research efforts focused on relative versus absolute deprivation (Lynch et al. 2000) and workplace stress and its social construction (Wainwright and Calnan 2002). The granular occupational status hierarchies and longevity of the Whitehall cohorts have also facilitated productive investigations of social causation versus social selection mechanisms (Elovainio et al. 2011), complementing earlier tests of the causation versus selection debate as it pertained to psychi- atric disorders (Dohrenwend et al. 1992).

Critics, however, have long admonished that traditional indicators of SES such as income level, educational attainment and occupational grade are incomplete explanatory and control variables in population research (Krieger et al. 1997, Muntaner and O’Campo 1993, Muntaner et al. 1991, 2000, Navarro et al. 2006, Wright 2009). This is in part because such indicators arise from numerous political and economic processes, which may have significant and direct effects on health in addition to those mediated by typical measures of SES. Such processes traditionally have been understood using relational theories of class, which have had less trac- tion than traditional stratification approaches in population health research (Galobardes et al. 2006a, 2006b, Krieger et al. 1997, Muntaner et al. 2000). Instead, traditional measures of SES remain predominant, as they are easier to measure than relational constructs and because research has not tended to emphasize the causes of socioeconomic inequality, but rather the effects of socioeconomic position on health (Lynch and Kaplan 2000, Muntaner et al. 2010). Focusing only on the latter, however, without understanding the processes of power, context and meaning through which socioeconomic resources are obtained and experienced may ignore other aetiological pathways between political-economic structures and mental illnesses (Brenner 1973, Brown and Harris 1978, Faris and Dunham 1988, Hollingshead and Redlich 1953, Liem and Liem 1978, Srole and Langner et al. 1962). Furthermore, these measurement issues matter because standard approaches to SES may implicitly valorise extant social struc- tures and therefore constrain the range of social and policy interventions deemed feasible and valid.

The primary aim of the present study is to explore how social class may influence depres- sion and anxiety in ways that may be masked or incompletely explained by standard SES mea- sures. A secondary aim is to extend earlier efforts to introduce explicitly theory-driven operationalisations of social class to social and psychiatric epidemiology. In the remainder of this section we briefly review the theoretical framework that informs traditional measures of SES and contrast it with the class theory that motivates our analysis. Next, we explore how class relations have been implicated historically in the aetiology of depression and anxiety. We then identify a potential mechanism for this relationship by connecting contemporary class theory to the body of literature on job strain and job control.

© 2015 Foundation for the Sociology of Health & Illness

Anxious? Depressed? You might be suffering from capitalism 1353

Stratificationist versus relational theories of class

Stratificationist conceptualisations of socioeconomic disadvantage provide the theoretical foundations for traditional measures of SES. However, critics (Krieger et al. 1997, Lynch and Kaplan 2000, Muntaner et al. 2000) have observed that studies that include simple stratifica- tion indicators such as income, education, occupational grade and so on rarely make an expli- cit reference to their theoretical underpinnings. A stratificationist perspective draws on the sociological tradition of structural functionalism, which contends that social stratification is universal and natural, and therefore must serve a purpose. For example, Davis and Moore (1945) suggested that inequality is functionally necessary to ensure that social positions of the greatest functional importance to society are conscientiously occupied by the most motivated and qualified individuals. This is achieved by differential remuneration. Contemporary stratifi- cation indicators, however, are only de facto functionalist, because they do not necessarily cap- ture constructs that account for how individuals arrive in different social strata or address inequalities and interdependencies in the positions people occupy. Instead, they consist of attri- butes and conditions that are associated with people who are already situated in classes (Wright 2009). As critics (Muntaner and Lynch 1999, Muntaner and O’Campo 1993, Wohlf- arth 1997) have argued, it may be more apt to consider stratification indicators (like SES) as ‘outcomes’, or proxies for processes that have already occurred, rather than ‘exposures’. Doing so directs attention to more fundamental (Link and Phelan 1995) upstream antecedents respon- sible for numerous political-economic pathways to health outcomes in addition to SES.

In contrast, emphasis on processes and relations, as opposed to mere position, follows a rich tradition of Marxian and Weberian class analysis, in which class is properly understood not as an individual attribute but as individuals’ relation to productive assets and their access to and exclusion from certain economic opportunities (Sørensen 2000, Wright 1997, 2009). In this relational perspective, classes are defined by mutually antagonistic self-interest, that is, the material welfare of one group depends causally on the material deprivations of another (Søren- sen 2000, Wright 1997). In Wright’s (1997) elaboration, social position is not simply a func- tion of the inherited or achieved attributes of individuals but arises from the processes by which certain groups control productive resources by (i) excluding other groups from access to those resources and controlling their labour activities (domination), and by (ii) appropriating the fruits of that labour (exploitation). Thus, beyond a sorting mechanism, class relations are ongoing, dynamic interactions, and it is within these relations that we seek to explore determi- nants of depression and anxiety.

Depression and anxiety: sequelae of class relations?

We chose to focus this analysis on depression and anxiety because of historical attention to these outcomes in social theory, their prominence in social stress and social constructionist models of psychiatric illness, and prior evidence of their onlinear relationship with social class. They are the most common mental disorders in the general population (Kessler et al. 2011) and have been invoked in various forms by social theorists from Marx (2007) to Durkheim (2014), Sartre (2004) and Sennett and Cobb (1972), reflecting on the impact of capitalism on the psyche.

The effect of class relations on the development of depression and anxiety can be antici- pated at the conjunction of social theory and the social stress paradigm (Kohn and Schooler 1983). The alienation of workers from production and the products of their labour is thought to diminish their self-efficacy and result in a sense of powerlessness (Gecas 1989, Seeman © 2015 Foundation for the Sociology of Health & Illness

1354 Seth J Prins et al.

1959) and self-estrangement, that is, engagement in activities that are not intrinsically reward- ing. (Roberts 1987, Seeman 1975). Weber extended the issue of alienation qua of powerless- ness and took industrial workers as a special example of those afflicted by universal trends towards bureaucratisation, mechanisation, depersonalisation and ‘oppressive routine’ (Gerth and Mills 1946: 50). Regarding self-estrangement, Marcuse (1991: xlvii) described the produc- tive apparatus in advanced capitalism as shaping ‘not only socially needed occupations, skills, and attitudes, but also individual needs and aspirations’.

Descriptions such as these are concordant with diathesis-stress models of psychopathology (Abramson et al. 1978, Beck and Alford 2009, Monroe and Simons 1991) that emphasise the interaction between individuals’ predispositions and stressful life experiences, both of which are themselves socially patterned and produced (Aneshensel 1992). In other words, relations to production and labour conditions may both shape and interact with the stressors to which workers are exposed and their reaction to those stressors (that is, their psychosocial resources). Such resources may include a variety of psychosocial constructs such as self-efficacy, attribu- tional dispositions, internalising versus externalising locus of control, and workplace demand/ control (that is, job strain), relevant to the development of psychopathology.

In particular, locus of control and job strain have been studied extensively with respect to work and affective disorders (Griffin et al. 2007, Landsbergis et al. 2012, Stansfeld and Candy 2006), although these constructs avoid explicit engagement with social class (Munta- ner and O’Campo 1993). For example, research using the National Longitudinal Survey, of the labour market experiences of several large age cohorts has examined the effect of pow- erlessness (operationalised as external locus of control) on physical and mental distress. Powerlessness exacerbates the effect of job-related and economic-related stressful life events on psycho-physiological distress (Krause and Stryker 1984) and prospectively predicts greater limits to individuals’ activity, psychosocial symptoms and deteriorating health condi- tions (Seeman and Lewis 1995), controlling for demographic characteristics and baseline health.

Similarly, extensive research has shown that psychosocial work stress and, in particular, job strain, are important risk factors in the development of depression (Eaton et al. 2001, Plaisier et al. 2007, Stansfeld et al. 1999, Wang et al. 2009). Workers with jobs low in decision lati- tude and high in demands show higher depressive symptoms than workers with jobs high in decision latitude and low in demands (Karasek 1979) and those with jobs low in direction, control and planning, as defined by the US Department of Labor, have higher psychological distress and major depression than those with jobs high in direction, control and planning (Link et al. 1993). Occupations in which individuals have high degrees of direction, control and planning over their own and others’ work foster a sense of mastery and personal control, which in turn have been shown to be protective against depressive symptoms (Link et al. 1993). Viewed from the perspective of relational class theory, these situational workplace experiences may act as mediators between class relations and mental health outcomes (Munta- ner and O’Campo 1993). In other words, as we discuss below, class relations may structure access to occupations with varying degrees of direction, control, and planning, which then determines individuals’ risk of depression and anxiety.

Contradictory class locations and occupational control

Presumably, as one moves down an organisational hierarchy relative to its owners, one encounters more stress and adversity due to exploitation, alienation and exposure to poor working conditions. It would therefore be reasonable to assume that the relationship between

© 2015 Foundation for the Sociology of Health & Illness

Anxious? Depressed? You might be suffering from capitalism 1355

workplace status (and therefore SES) and depression and anxiety would be more or less inverse and linear, with lower status consistently translating into increased disorder. Yet insights into the contemporary class structure have led to hypotheses about the distribution of these disorders that are not readily predicted by such stratificationist theories and measures. Specifically, Wright’s (1985) notion of contradictory class locations has been hypothesised to lead to psychosocial stressors that are known risk factors for depression and anxiety, and may explain why those occupying intermediate locations in class hierarchies appear to suffer higher rates of depression and anxiety than those at either the top or bottom (Muntaner et al. 1998, 2003, Wohlfarth 1997).

The concept of contradictory class locations emerges from Wright’s (1985, 1997) effort to accommodate, in modern class analysis, the heterogeneous relations to production evident in post-industrial economies. In contemporary capitalism 85–90 per cent of the labour force does not own the means of production and must sell its labour on the market, but much of that group does not perform the sort of manual labour commonly associated with the blue collar working class, nor is it exploited and dominated in the same way (Wright 1997). Wright clas- sifies this group along two dimensions: the possession of skills and expertise and the degree of formal authority within organisational hierarchies in relation to production, both of which con- fer privilege and strategic advantage. For example, someone with valued and uncommon skills can obtain higher wages (that is, endure less exploitation) and more autonomy (less domina- tion) than unskilled workers. Likewise, upper management (employees closer to the top of an organisational hierarchy) may receive delegated ownership authority to participate in develop- ing company policy, whereas supervisors (employees in the middle and lower ranks of the or- ganisational hierarchy) may be expected to implement company policy but not develop it, entitling them to higher wages and autonomy than workers but lower wages and autonomy than managers. Such locations within class relations are contradictory because they embody aspects of both ownership and labour.

As developed by Muntaner and O’Campo et al. (1993), Muntaner et al. 1998, 2003) and Wohlfarth (1997), the constructs of skills, expertise and authority (vis-!a-vis contradictory class locations) are conceptually related to Karasek’s (1979) job strain model and research on occu- pational direction, control and planning (Link et al. 1993). The intersection of these two mod- els suggests why contradictory class locations may predict adverse outcomes such as depression and anxiety relative to lower class positions: broadly speaking, all but the highest level of managers may be expected to enforce policies in which they have little say, while simultaneously facing the antagonism of subordinates (Muntaner et al. 1998).

There is some empirical evidence that contradictory class locations confer an elevated risk of depression and anxiety. Muntaner et al. (1998) observed in community-based longitudinal data from the Epidemiologic Catchment Area Study (Robins and Reiger 1991) that higher level managers displayed lower rates of major depression, anxiety disorder and alcohol disor- ders than either supervisors or workers, while supervisors displayed higher rates of major depression and alcohol disorders than either managers or workers. Subsequently, in the 2000– 2001 Barcelona Health Interview Survey of 4219 city residents, Muntaner et al. (2003) found evidence that the more contradictory the class location, the poorer the mental health of respon- dents.

Thus, population-based studies suggest that contradictory class locations are important for mental health. However, these findings are from selected cities (Baltimore, Maryland and Bar- celona, Catalonia) and do not represent the national class structure in the US. Furthermore, the study in Barcelona used a non-diagnostic mental health measure, which has been found to be not associated with social stratification or social class in European samples (Muntaner et al. 2003). Finally, a stronger test of the contradictory class locations hypothesis would allow © 2015 Foundation for the Sociology of Health & Illness

1356 Seth J Prins et al.

distinctions between the private sector, where the concept and process of ownership is more straightforward, and the public sector.

Nonetheless, the findings summarised above are inconsistent with stratificationist theories of SES, which predict an inverse linear relationship between SES and mental health, but not with relational theories of class, which instead predict such nonlinear findings through numerous psychosocial mechanisms. The present study, then, builds on evidence for the role of contra- dictory class locations in explaining nonlinear patterns of mood and anxiety disorders by using a large, nationally representative survey of the general US population, the National Epidemio- logical Survey on Alcohol and Related Conditions (NESARC), which includes a fully struc- tured diagnostic interview for the assessment of psychiatric disorders as well as extensive measures of SES.

Methods

Sample This sample consists of participants in the 2001–2002 NESARC, a nationally representative US survey of civilian non-institutionalised participants aged 18 and older, interviewed in per- son. The National Institute on Alcohol Abuse and Alcoholism (NIAAA) sponsored the study and supervised the fieldwork, conducted by the US Bureau of the Census. The research proto- col received full ethical review and approval from the US Census Bureau and US Office of Management and Budget. Young adults, Blacks and Latinos were oversampled; the overall response rate was 81 per cent. Further details of the sampling frame, demographics of the sam- ple, and details about the interviewers, training and field quality control are described else- where (Grant et al. 2003a, 2003b, 2004a, 2007).

We restricted NESARC data to respondents who reported currently working full time (35+ hours per week) and who were not full-time homemakers (n = 21,859), as the role of domes- tic labor in individual versus household relation to production, while an important area for investigation, is beyond the scope of the present analysis. Separate analyses were conducted for the private sector and all sectors. The private sector includes employment by a private for-profit company, business or individual. All sectors include the private sector in addition to private not-for-profit, tax exempt or charitable organisation and federal, state, and local govern- ment (excluding armed forces).

Dependent variables The NIAAA Alcohol Use Disorder and Associated Disabilities Interview Schedule DSM-IV (AUDADIS-IV) (Grant et al. 2001) was used to assess DSM-IV psychiatric disorders. This instrument was specifically designed for experienced lay interviewers and was developed to advance the measurement of substance use disorders and other mental disorders in large-scale surveys. Although evidence supports the contention that psychiatric nosology is shaped by sociocultural and political-economic context, for example, shifting moral narratives and social norms, political struggles and economic interests (see Branaman 2007, Conrad and Slodden 2013, Horwitz 2011, Wakefield 1992), the focus of the present study is not on the role of class relations in the social construction of depression and anxiety but rather on their prevalence and determinants as currently constructed.

DSM-IV-diagnosed disorders assessed by the AUDADIS-IV included major depression (‘depression’) as well as generalised anxiety disorder and panic disorder (‘anxiety’). We chose these disorders because the average age of onset relative to other mood and anxiety disorders is later (after age 18: Kessler et al. 2005), and we sought to focus on disorders that might arise

© 2015 Foundation for the Sociology of Health & Illness

Anxious? Depressed? You might be suffering from capitalism 1357

after entry in the workforce. The reliability and validity of mood and anxiety disorder diagno- sis range from fair (kappa for panic disorder diagnosis = 0.42) to good (kappa for major depressive disorder diagnosis = 0.65) (Canino and Bravo et al. 1999, Grant et al. 1995, 2003a), including test–retest and clinical re-appraisal studies. The reliability of anxiety disor- ders diagnosed in the AUDADIS-IV is similar to those found for other instruments designed for national surveys such as the Composite International Diagnostic Interview and the Diag- nostic Interview Schedule (Haro et al. 2006, Semler et al. 1987). Diagnoses were further vali- dated using the Short-Form 12-Item Health Survey, Version 2, a mental disability score, in controlled linear regressions (Grant et al. 2004a, 2004b, 2005, Hasin et al. 2005).

Independent variables Relation to production We constructed a class measure informed by Wright’s typology and based on available indicators in the AUDADIS-IV. We categorised classes as owners, manag- ers, supervisors and workers. Owners consist of respondents who identified as self-employed and earned more than or equal to $71,500 (the 90th percentile) in annual income. The 90th per- centile was chosen because it clearly separates capitalists from small employers and the petty bourgeoisie, but is still obtainable by workers, managers and supervisors in a variety of occu- pations. A sensitivity analysis using a different income cut-off is described below. Managers consist of respondents who identified their occupation as executive, administrative or manage- rial and had more than or equal to a 4-year bachelor’s degree. Supervisors consist of respon- dents who identified their occupation as executive, administrative or managerial and had less than a 4-year bachelor’s degree. A bachelor’s degree was chosen as a broad proxy for skills and expertise, in order to separate higher level management from lower level supervisors across a variety of occupations in which more specific educational credentials may have differ- ent meanings. A sensitivity analysis with no education proxy is discussed below. Workers con- sist of respondents who identified their occupation as private household; other services; farming, forestry, and fishing; operators, fabricators, and labourers; transportation and material moving; or handlers, equipment cleaners and labourers. We chose the above occupations to represent workers because examples provided under each of these categories on the AUDA- DIS-IV interview flashcard (Appendix A) did not explicitly mention special skills, expertise or managerial or supervisory functions. Sensitivity analyses using additional occupations to repre- sent workers are discussed below.

Socioeconomic status We examined two traditional measures of SES to determine whether the NESARC sample is consistent with known SES prevalence patterns for depression and anxiety and to adjust for traditional SES measures in models presented below. For personal income, we grouped the continuous income variable into seven categories ranging from less than or equal to $20,000 (n = 6142) to more than $120,000 (n = 570). We grouped educational attainment into eight categories, ranging from ‘none to grade 8’ (n = 882) to ‘completed grad- uate or professional degree’ (n = 2183).

Analysis

We tabulated the prevalence of any lifetime and 12-month depression and anxiety by class cat- egories, first restricting the data to the private sector and then including all sectors. We also tabulated depression and anxiety by income and education in the full sample. We constructed bivariate and adjusted logistic regression models (one for each of lifetime and 12-month depression and anxiety as the outcomes) to determine the odds of disorder across classes, in © 2015 Foundation for the Sociology of Health & Illness

1358 Seth J Prins et al.

the private sector and all sectors. Adjusted models include sex, age, ancestry group and metro- politan statistical area. Descriptive statistics were conducted using PROC SURVEYFREQ and regression models using PROC SURVEYLOGISTIC in SAS 9.3. The study design oversam- pled hard to reach groups, thus sample weights were incorporated to generate estimates that are nationally representative of the demographics of the USA based on the 2000 census. Fur- ther, design weights were incorporated to account for the stratified complex sampling strategy. Standard errors were estimated using Taylor series linearisation.

Sensitivity analysis We performed four sensitivity analyses on our class measures. We wanted to ensure that our results were not contingent on our education proxy for managers and supervisors, our income cut-off for owners or our choice of worker occupations. In our first sensitivity analysis we col- lapsed the manager and supervisor categories by removing the education proxy and the income cut-off for owners. Owners consist of respondents who identified as self-employed, and man- agers or supervisors consist of respondents who identified their occupation as executive, administrative or managerial. In our next two sensitivity analyses we utilised our original class categories but systematically added occupations to the worker category. We did this because our original class operationalisation excluded from the worker category any occupations that involve special skills or expertise, or managerial or supervisory functions. Therefore, in our second sensitivity analysis we constructed an eight-occupation worker category that included sales; administrative support including clerical; private household; other services; farming, for- estry, and fishing; operators, fabricators, and labourers; transportation and material moving; and handlers, equipment cleaners, and labourers. In our third sensitivity analysis, we con- structed a 12-occupation worker category that included professional specialty; technical and related support; sales; administrative …