Straining under contradiction? Contradictory class locations and job strain in the United States

Abstract

Research suggests lower-level supervisors may suffer greater mental-illness and substance-use burdens than workers and upper management, a pattern that may arise from their contradictory class location. For example, lower-level supervisors may be compelled to discipline subordinates and enforce policies over which they have little say, and contend with antagonism from workers, exposing them to stressors like job strain. However, to our knowledge, no U.S. studies have investigated whether job strain is elevated among lower-level supervisors. We addressed this gap using 2002–2022 General Social Survey data (n=9,261). We classified respondents as workers, lower-level supervisors, higher-level supervisors, top executives, petit bourgeoisie, or capitalists using self-employment and supervisory-authority items, and created a continuous job-strain score using five job-control and six job-demands items. Age- and gender-adjusted Poisson models suggested job strain decreased approximately linearly across classes, as mean scores were 6% (95% CI: 3%-8%) lower among lower-level supervisors and 23% (95% CI: 19%-27%) lower among capitalists than among workers. Patterns were similar for job control alone. However, lower-level supervisors did report elevated job demands, including 14% (95% CI: 10%-17%) greater mean scores than workers. Patterns persisted after thorough confounder adjustment. Our findings suggest job strain may not explain lower-level supervisors’ mental-illness and substance-use burdens.

1. Introduction

Why might lower-level supervisors suffer greater burdens of mental illness and substance use than workers and upper management in capitalist societies?^1–3 Such patterns cannot be explained by stratificationist theories, like those based on the “socioeconomic gradient”, which predict linear improvements in health with rising social position.³ However, relational theories, which analyze health inequities in terms of social relations and processes, like conflicts between social classes struggling for control over labor and capital (i.e., productive property), may be more explanatory.^3–8

Wright’s relational theory is among the most influential in quantitative social sciences, including his explication of the concept of contradictory class locations.^3–8 According to Wright, supervisors and managers occupy such contradictory locations by embodying features of capitalists (they control others’ labor) and workers (they are controlled by capitalists and lack capital).^7,8 This exposes them to certain occupational hazards.^1–3,9 For example, lower-level supervisors may be compelled by upper management to discipline subordinates and enforce policies over which they have little say (e.g., hirings, firings, promotions, and demotions), while simultaneously contend with antagonism from workers below them.^1–3,9 Moreover, they may be excluded from labor unions, which can provide workers with some control over the terms and conditions of their employment.^10,11 Thus, lower-level supervisors may be disproportionately exposed to psychosocial stressors like job strain, characterized by high demands and low control at work, a risk factor for mental illness, substance use, and other adverse outcomes.^12–18 In contrast, higher-level supervisors and managers may have substantial authority and autonomy at work, protecting them from such stressors.^1–3,8

However, while psychosocial stressors are a plausible mediating mechanism linking certain contradictory class locations to adverse outcomes, to our knowledge, no US studies have investigated whether the stressors are indeed elevated among those in such locations. We addressed this gap. Specifically, using General Social Survey (GSS) data, we examined whether job strain followed patterns predicted by the contradictory class location concept, hypothesizing strain, and especially high demands therein, would be greatest among lower-level supervisors.

2. Methods

2.1. Data and analysis overview

The GSS is a nationally representative survey of non-institutionalized US adults.¹⁹ Every four years from 2002–2022, the GSS included the Quality of Worklife (QWL) module, which contained detailed questions on working conditions.¹⁹ The GSS conducts most interviews in-person, but conducted many 2022 QWL interviews online.¹⁹ Our sample included QWL respondents working fulltime/parttime and those temporarily not working.

We conducted our analyses in R version 4.3.2.²⁰ We incorporated GSS’s sampling weights and design parameters to produce nationally representative estimates and accurate standard errors.^19,21 Our code is on Open Science Framework (https://osf.io/pt2u7/?view_only=8036b9e78b794bd4936b5da21fcf3d49), which contains information about accessing GSS data.

2.2. Measures

2.2.1. Social class

Drawing from Wright and prior GSS analyses,^7,8,22–25 we measured social class relationally using data on respondents’ self-employment status and supervisory authority; GSS does not contain consistent data on respondents’ control over company or organizational policy, another variable Wright used to distinguish social classes. Workers were those who were not self-employed and who did not supervise others. Lower-level supervisors were those who were not self-employed, who supervised others, who had a supervisor, and whose supervisors had a supervisor, while higher-level supervisors were those who were not self-employed, who supervised others, who had a supervisor, and whose supervisors did not have a supervisor. Top executives were those who were not self-employed, who supervised others, and who did not have a supervisor. Finally, the petit bourgeoisie were those who were self-employed and who did not supervise others, while capitalists were those who were self-employed and who supervised others. eAppendix 1 contains a decision tree and questionnaire wording. In sensitivity analyses, we distinguished the petit bourgeoisie and capitalists using data on the number of employees that self-employed respondents had; eAppendix 8 contains details.

2.2.2. Job control, demands, and strain

We used Likert-scale responses to five items regarding job control (e.g., “I am given a lot of freedom to decide how to do my own work”) and six items regarding job demands (e.g., “I have enough time to get the job done”);²⁶ eAppendix 2 contains item details. To measure low control or high demands, we summed respondents’ responses to the relevant items (e.g., 0=very true; 3=not at all true) to create continuous scores (range: 0–15 for control, where 15 means minimum control; range: 0–18 for demands, where 18 means maximum demands). To measure job strain, we multiplied the control score by 6/5, then summed the control and demands scores (range: 0–36, where 36 means maximum strain); multiplying the control score by 6/5 gave it the same weight as the demands score in the summed score.²⁶ Finally, we divided each of respondents’ scores by the maximum possible score for a given stressor and multiplied it by 100, which normalized the scores to a range of 0 to 100.

In secondary analyses, we analyzed the control and demands scores as binary measures. Our primary approach classified respondents as experiencing “low control” or “high demands” if they reported scores greater than the sample median.¹³ Our secondary approach classified respondents as “low control” or “high demands” if they averaged >1 on each of the items that went into a given score (with scores ≤1 corresponding to disagreeing about having an adverse condition). We then used these approaches to create four binary demand/control measures (high demands/low control [job strain], high demands/high control, low demands/low control, low demands/high control).

We used the continuous scores in our primary analyses because: 1) job control, demands, and strain exist on a spectrum of intensity rather than being binary constructs, 2) dichotomization can reduce statistical power,^27,28 3) there are no established cutoffs for dichotomizing our continuous scores, and sample-specific quantiles may not correspond to meaningful population quantiles or quantiles in other samples,²⁸ and 4) our measurement approach has been used in previous job-strain research in the GSS.²⁶

2.2.3. Covariates

Covariates of interest, which we considered confounders of the class-outcome relationships, included respondents’ age, gender, place of birth, race/ethnicity, education, marital status, census region of residence, family income, occupation, and industry, as well as the number of employees at the location where respondents worked (i.e., firm size). Section 2.3. contains further discussion of the covariates.

2.3. Analyses

First, we calculated class-stratified descriptive statistics of our sample, including in the sample with missingness (“response” sample), in the sample without missingness, and in the first imputed sample.²⁹ Second, we calculated the distribution of the four binary demand/control variables by class. Finally, using R’s ‘survey’ package,²¹ we estimated the percentage (“relative”) or percentage-point (“absolute”) difference in mean job control, demands, or strain among each class relative to the mean among workers using Poisson or linear models, respectively. We estimated three sets of models: set one (least adjusted) included only age and year as covariates; set two (more adjusted) included set one’s covariates plus gender, place of birth, race/ethnicity, education, marital status, region of residence, and industry; and set three (most adjusted) included set two’s covariates plus occupation and firm size. Models incorporating set one captured the total magnitude of class differences in the outcomes, net of class differences in age and year, while sets two and three assessed whether class differences persisted after more thorough adjustment for confounding by social stratification. We excluded occupation and firm size from sets one and two because of their collinearity with social class. For example, managerial occupations frequently include supervisory responsibilities, while only firms with multiple employees can have multiple levels of supervisory authority. Our primary regression analyses used the continuous scores as outcomes; we used the binary job-strain measure in secondary analyses.

2.4. Missing data

To address unplanned missingness (sample: 87% complete cases), we used multiple imputation by chained equations with 24 replications and 25 iterations via R’s ‘mice’ package,³⁰ assuming missing values were missing at random conditional on observed values of the variables of interest.³¹ We used Rubin’s Rules to combine regression estimates from the multiply imputed datasets,^30,31 excluding those with imputed outcome values.³²

3. Results

3.1. Descriptive statistics

Our sample included 9,261 respondents, who were 56% workers, 22% lower-level supervisors, 6% higher-level supervisors, 3% top executives, 8% petit bourgeoisie, and 5% capitalists (Table 1).

Table 1.

Descriptive statistics of response (sample with missingness), complete-case, and imputed samples, overall and stratified by social class.

	Response	Complete	First imputed dataset, survey-weighted except ‘n’
	Overall	Overall	Overall	Worker	Low sup	High sup	Exec	PB	Cap
n	9261	8078	9261	5219	2027	513	288	746	468
Class (%)a
Worker	56.3	56.1	55.8	100.0	0.0	0.0	0.0	0.0	0.0
Low sup	21.9	22.6	21.5	0.0	100.0	0.0	0.0	0.0	0.0
High sup	5.5	5.7	5.8	0.0	0.0	100.0	0.0	0.0	0.0
Exec	3.1	3.2	3.2	0.0	0.0	0.0	100.0	0.0	0.0
PB	8.0	7.4	8.2	0.0	0.0	0.0	0.0	100.0	0.0
Cap	5.1	5.0	5.4	0.0	0.0	0.0	0.0	0.0	100.0
Male (%)	48.6	48.4	50.1	45.3	51.4	57.7	60.6	52.2	76.1
Born in the US (%)	86.7	88.5	84.8	85.1	84.0	90.3	87.4	79.9	85.2
Race/ethnicity (%)b
NH white	66.5	68.3	67.1	64.2	67.1	79.5	72.9	70.4	75.7
NH Black	14.6	14.4	12.9	15.1	13.2	5.4	11.8	7.8	5.7
NH other	4.4	4.3	4.9	4.4	5.5	3.0	4.4	6.4	7.9
Hispanic	14.5	13.0	15.1	16.3	14.3	12.1	11.0	15.4	10.6
Education (%)
< high school	8.9	8.1	9.5	10.2	6.1	11.6	9.6	12.3	9.2
High school	48.0	48.2	49.1	52.8	40.8	49.5	45.9	51.8	41.5
Junior college	9.2	9.4	9.2	8.6	11.1	8.6	10.9	8.5	8.7
≥ College	33.9	34.2	32.2	28.4	42.0	30.3	33.6	27.4	40.7
Marital status (%)c
Married	46.9	46.5	54.8	50.8	58.6	56.8	58.0	58.1	71.7
Never married	30.0	30.1	27.8	31.8	26.6	26.1	19.4	20.3	9.9
Wid/div/sep	23.1	23.4	17.4	17.4	14.8	17.2	22.7	21.6	18.4
Region (%)
Midwest	23.2	23.9	22.3	23.5	21.7	24.9	15.5	19.7	18.8
Northeast	16.3	16.1	17.0	16.9	18.8	18.4	17.9	13.3	14.7
South	32.3	31.5	31.8	32.4	31.4	28.6	32.7	32.5	29.8
West	28.3	28.4	28.8	27.2	28.1	28.1	34.0	34.5	36.7
Firm size (%)d
<10	26.2	25.4	27.7	19.7	11.2	31.1	40.6	87.4	74.3
10–99	35.7	36.0	35.4	38.5	36.4	51.6	38.1	10.1	19.7
100+	38.1	38.7	36.9	41.8	52.4	17.3	21.3	2.5	6.1
Occupation (%)
Managers	9.6	9.7	9.9	3.5	18.7	22.1	24.5	8.9	20.8
Professionals	30.2	30.2	29.1	24.9	37.1	30.4	35.5	29.2	34.6
Blue collar	24.9	24.5	26.2	29.4	17.4	23.8	21.4	29.9	27.7
Services/other	35.3	35.6	34.8	42.2	26.8	23.8	18.6	31.9	16.9
Industry (%)e
Ag, const, manuf	19.0	18.8	19.7	17.9	17.4	29.4	28.8	18.6	32.9
FIRE	7.0	7.0	7.2	6.9	6.6	7.1	9.3	9.7	7.4
Other services	47.7	47.6	47.1	47.4	44.6	44.6	40.3	58.0	43.6
Pub ad/military	6.0	6.1	5.6	5.5	10.2	3.2	2.5	0.3	0.6
Trans/wareh/util/info	8.2	8.4	8.2	9.0	8.1	4.7	6.0	7.6	7.5
Trade	12.1	12.2	12.2	13.3	13.0	11.1	13.1	5.8	8.0
Age (Q2 [Q1, Q3])f	42 [32, 53]	42 [32, 53]	42 [31, 53]	40 [29, 52]	41 [32, 51]	42 [30, 52]	45 [35, 56]	48 [37, 57]	50 [40, 59]
Control (Q2 [Q1, Q3])f,g	27 [13, 33]	27 [13, 33]	27 [13, 33]	27 [20, 40]	20 [13, 27]	20 [7, 27]	20 [7, 33]	20 [13, 33]	13 [7, 27]
Demand (Q2 [Q1, Q3])f,h	39 [28, 50]	39 [28, 50]	39 [28, 50]	39 [28, 50]	44 [33, 56]	39 [28, 50]	39 [28, 50]	28 [17, 39]	33 [28, 44]
Strain (Q2 [Q1, Q3])f,i	31 [23, 39]	31 [23, 39]	31 [23, 39]	33 [24, 42]	31 [23, 39]	28 [21, 36]	29 [21, 36]	25 [18, 32]	24 [19, 32]

Notes:

^alow sup = lower-level supervisor; high sup = higher-level supervisor; exec = top executive; PB = petit bourgeoisie; cap = capitalists

^bNH = non-Hispanic

^cWid/div/sep = widowed, divorced, or separated

^dNumber of employees at the location where respondents worked

^eAg, const, manuf = agriculture, construction, and manufacturing; FIRE = finance, insurance, and real estate; pub ad/military = public administration and military; trans/wareh/util/info = transportation, warehousing, utilities, information

^fQ2 = quartile 2; Q1 = quartile 1; Q3 = quartile 3

^gJob-control score, as defined in the main text.

^hJob-demands score, as defined in the main text.

ⁱJob-strain score, as defined in the main text.

3.2. Distribution of binary demand/control measures by class

Workers reported the greatest prevalence of job strain (i.e., high demands/low control) (20%), while the petit bourgeoisie (5%) and capitalists (6%) reported the lowest (Figure 1). Patterns flipped for low demands/high control, with capitalists reporting the greatest prevalence (61%) and workers the lowest (37%). Nonetheless, control aside, lower-level supervisors reported a greater prevalence of high demands (51%) than other classes.

Figure 1.

Mosaic plot depicting the distribution of the binary demand/control measures within each class (vertical) and the distribution of the classes within each binary demand/control measure (horizontal), with the area of each rectangle proportional to each group’s sample size.

Notes: Estimates calculated using survey-weighted, multiply imputed data from the 2002, 2006, 2010, 2014, 2018, and 2022 waves of the General Social Survey’s Quality of Worklife (QWL) module, excluding those with imputed values of the outcome variable. Sample restricted to the first imputed dataset (n=8,446). Respondents classified as having job strain (high demands/low control) if they reported job-control and job-demands scores greater than the sample median.

3.3. Regression analyses

Patterns in confounder-adjusted regression analyses mirrored those in descriptives analyses.

Regarding job control, all classes reported better mean scores than workers (Figure 2 and eAppendix 3). For example, in least-adjusted Poisson analyses, lower-level supervisors reported 30% (95% CI: 27%, 33%) lower mean scores than workers, while capitalists reported 49% (95% CI: 43%, 54%) lower mean scores. These differences persisted after more thorough adjustment.

Figure 2.

Relative or absolute difference in mean job-control scores (higher score corresponds to less control) among each class relative to the mean among workers.

Notes:

Relative differences estimated using Poisson models, while absolute differences estimated using linear models, with standard errors estimated via Taylor series linearization. Least-adjusted models included only age and year as covariates; more-adjusted models added gender, place of birth, race/ethnicity, education, marital status, region of residence, and industry as covariates; and most-adjusted models also added occupation and firm size as covariates. All models run on survey-weighted, multiply imputed data from the 2002, 2006, 2010, 2014, 2018, and 2022 waves of the General Social Survey’s Quality of Worklife (QWL) module, excluding those with imputed outcome values (n=8,564).

Regarding job demands, lower-level supervisors reported greater mean scores than workers and other classes (Figure 3 and eAppendix 4). For example, in least-adjusted Poisson analyses, lower-level supervisors reported 14% (95% CI: 10%, 17%) greater mean scores than workers. In contrast, the petit bourgeoisie reported 26% (95% CI: 22%, 29%) lower mean scores. These differences attenuated somewhat after more thorough adjustment.

Figure 3.

Relative or absolute difference in mean job-demand scores (higher score corresponds to more demands) among each class relative to the mean among workers.

Notes:

Relative differences estimated using Poisson models, while absolute differences estimated using linear models, with standard errors estimated via Taylor series linearization. Least-adjusted models included only age and year as covariates; more-adjusted models added gender, place of birth, race/ethnicity, education, marital status, region of residence, and industry as covariates; and most-adjusted models also added occupation and firm size as covariates. All models run on survey-weighted, multiply imputed data from the 2002, 2006, 2010, 2014, 2018, and 2022 waves of the General Social Survey’s Quality of Worklife (QWL) module, excluding those with imputed outcome values (n=8,515).

Finally, regarding job strain, mean scores decreased approximately linearly from workers to capitalists (Figure 4 and eAppendix 5). For example, in least-adjusted Poisson analyses, lower-level supervisors reported 6% (95% CI: 3%, 8%) lower mean scores than workers, while capitalists reported 23% (95% CI: 19%, 27%) lower mean scores. These patterns persisted after more thorough adjustment and when operationalizing strain as a binary measure (eAppendices 6 and 7). For example, in least-adjusted Poisson analyses that used the median-cutoff job-strain measure, lower-level supervisors reported a 33% (95% CI: 21%, 43%) lower prevalence of job-strain than workers, while capitalists reported a 71% (95% CI: 42%, 86%) lower prevalence.

Figure 4.

Relative or absolute difference in mean job-strain scores (higher score corresponds to more strain) among each class relative to the mean among workers.

Notes:

Relative differences estimated using Poisson models, while absolute differences estimated using linear models, with standard errors estimated via Taylor series linearization. Least-adjusted models included only age and year as covariates; more-adjusted models added gender, place of birth, race/ethnicity, education, marital status, region of residence, and industry as covariates; and most-adjusted models also added occupation and firm size as covariates. All models run on survey-weighted, multiply imputed data from the 2002, 2006, 2010, 2014, 2018, and 2022 waves of the General Social Survey’s Quality of Worklife (QWL) module, excluding those with imputed outcome values (n=8,446).

Distinguishing the petit bourgeoisie and capitalists using data on the number of employees that self-employed respondents had did not meaningfully alter any of the estimates (eAppendix 8).

4. Discussion

Using nationally representative data and a relational social class measure reflecting power over capital and labor, we estimated class differences in job strain. However, these differences did not align with the contradictory class location hypothesis, as all classes reported less strain than workers, including lower-level supervisors and particularly capitalists and the petit bourgeoisie. These patterns were even starker for job control. Nonetheless, lower-level supervisors did report greater job demands than workers and other classes. However, class differences in demands were relatively small.

Thus, our study does not implicate job strain as a potential mechanism explaining elevated burdens of mental illness, substance use, and other adverse outcomes among lower-level supervisors.^1–3,9 That said, if exposure to job strain is more harmful among lower-level supervisors than among workers or other classes, the stressor may still contribute to their excess burdens of the outcomes.³³ However, we have little reason to anticipate such exposure-mediator interaction. Likewise, job demands may contribute to lower-level supervisors’ excess burdens, especially if its adverse effects swamp the adverse effects of job strain or control. However, to our knowledge, no research suggests job demands has more potent effects than the other stressors. In contrast, our study does implicate job strain as a potential mechanism contributing to elevated burdens of many other adverse outcomes, including poor/fair self-rated health and mortality, among workers.^4,23,34–37 This aligns with the findings of a prior study, which estimated excess burdens of certain psychosocial and material stressors among workers relative to capitalists and other classes.²²

Although the GSS contains more detailed social-class items than most other nationally representative US datasets, and our social-class measure was based on measures used in prior research,^22–25 our analysis may have suffered from misclassification. For example, our “petit bourgeoisie” category may have included gig workers who identified as self-employed, but who should have been classified as “workers”, given their lack of capital. Likewise, our “top executive” category may have included chief executives who did not identify as self-employed, but who could have been classified as “capitalists”, given their power over capital (e.g., via stock ownership) and corporate decision-making. Nonetheless, we do not believe misclassification undermined identification of our focal social class, lower-level supervisors, nor do we believe misclassification affecting other social classes was substantial enough to qualitatively alter our findings.

Given that our findings regarding patterns of job strain across social classes did not align with the contradictory class location hypothesis, future research should replicate our analyses in additional datasets, including in those with alternative social-class and job-strain measures. Moreover, future research should assess whether excess burdens of mental illness and substance use persist among lower-level supervisors in contemporary samples, and if so, use formal mediation methods^33,38,39 to directly analyze whether job strain or other psychosocial stressors, like effort-reward imbalance,⁴⁰ can explain the observed patterns.

Data availability statement:

The code used in our analyses is available at https://osf.io/pt2u7/?view_only=8036b9e78b794bd4936b5da21fcf3d49, where readers can find information about accessing the underlying data.