Abstract
Little is known about how workplace problems may influence diagnosable mental health and substance use (MHSU) disorders. We examine the associations between three common workplace problems (experiencing problems with co-workers, job changes and perceived financial strain) and three MHSU disorders (mood, anxiety and substance abuse/dependence). The analysis utilizes longitudinal data on a sample of working-age adults from the National Epidemiological Survey on Alcohol and Related Conditions. These data are well suited for our research objective as the survey was specifically designed to study MHSU disorders. Results show that experiencing these workplace problems is associated with an increased risk for mental health disorders, but not substance use disorders. Importantly, various robustness checks and sensitivity analyses demonstrate that our findings cannot be not fully explained by omitted variables, reverse causality or sample attrition.
Notes
1 The workplace problems we study here are captured by business readjustment (15), change in financial state (16), change to different line of work (18), change in responsibilities at work (22), trouble with boss (30) and change in working hours or conditions (31).
2 Sociological work by Catalano et al. (Citation1993) and others using the Epidemiologic Catchment-Area Project, which collects information on MHSU disorders, is an exception.
3 In particular, involuntary job loss only reduces alcohol dependence symptoms among single respondents only.
4 In unreported analyses, we relax this restriction and all findings are consistent with our core results.
5 Ideally, we would like to consider alcohol and illicit drug use disorders separately. However, the small number of respondents who are classified with an illicit drug use disorder in our sample (n = 214) prevents separate analysis.
6 The NESARC includes diagnoses of panic disorder without agoraphobia, panic disorder with agoraphobia and agoraphobia without panic disorder.
7 Recall that we rely on definitions developed within the psychiatric literature and measured in the AUDADIS instrument, which has been validated in numerous studies. However, one limitation of our MHSU diagnoses is that for some respondents these definitions are based on MHSU problems that negatively impact employment. This feature of the AUDADIS may lead to a tautological relationship between the workplace problems we study and our MHSU disorder outcomes for some respondents. Moreover, removing components of the diagnoses may invalidate the instrument. Our approach is consistent with previous economic studies that utilize the NESARC to study the impact of employment on MHSU outcomes (Dávalos et al., Citation2012; Popovici and French, Citation2013).
8 This variable also includes problems with supervisors.
9 These variables may be endogenous to the workplace problems in our regression models. In unreported analyses, we re-estimate our models without these variables and results are highly robust.
10 To preserve sample size, we include indicator variables whenever an item-specific value is missing in our regression models. All data are self-reported by the respondent and thus vulnerable to recall bias.
11 A randomized control trial (RCT) is generally viewed as the gold standard for estimating causal effects, but an RCT is both infeasible and unethical in this setting.
12 Results are highly robust to utilizing alternative covariate sets.
13 Results are consistent if we utilize a probit model instead of a linear probability model and calculate average marginal effects. We chose to estimate a linear probability model for consistency with core models.