![Sample our Economics, Finance,Business & Industry journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5931/TOC-Subject-Sample-2021-Economics-Finance-Business-Industry.jpg"})
ABSTRACT
Using six cycles of Canada’s longitudinal National Population Health Survey data (2000–2001 to 2010–2011), this article examines the relationship between job insecurity and mental health. Job insecurity is evaluated in both subjective (perception of job insecurity) and objective (probability of joblessness) terms while mental health is measured using a standardized psychological distress index. Applying a person-specific fixed-effects estimator, results indicate that for males and females age 25–64, job insecurity, regardless of how it is measured, is associated with an increase in psychological distress. Results regarding unemployment are not as conclusive, suggesting that it is not so much the actual occurrence of job loss but the threat of unemployment that is associated with higher psychological distress. Estimates of the relationship between job insecurity and psychological distress using pooled ordinary least squares are much larger, implying that much of the psychological distress/job insecurity correlation may be due to unobservable fixed characteristics. All results are robust to the inclusion and exclusion of a host of other potential determinants including income-related variables, education, and various health measures.
Acknowledgements
We are very grateful for the helpful comments received from Shelley Phipps, Courtney Ward, and Lori Curtis.
Disclosure statement
No potential conflict of interest was reported by the authors.
Supplemental data
Supplemental data for this article can be accessed here.
Notes
1 A set of time-invariant variables are also swept out of the equation after the fixed-effects application. To test that the difference in coefficients is not at least partially attributable to these time-invariant variables, pooled OLS regressions were run omitting these variables. Results for the job insecurity variable remained virtually unchanged.
2 The 1994–1995 (cycle 1) and 1996–1997 (cycle 2) NPHS surveys did not ask individuals to report their actual annual household income, only the income interval within which their income, after transfers but before taxes, fell. In cycles 2 and 3, NPHS did not ask about perceived job insecurity. Inclusion of cycle 1 data requires exclusion of the marital transition variables (see below) from the analysis. Regressions excluding the marital transition variables were run with and without the inclusion of cycle 1. Key results were little changed. The usable data for this article is thus restricted to cycles 4–9 inclusive.
3 A test akin to that performed by Burton, Phipps, and Zhang (Citation2014) and Ding and Lehrer (Citation2010) was evaluated. For this test, our regression models were run for only cycle 4 observations using OLS. These regression models also included an attrition variable equal to unity if the respondent attrits at some point in the future (i.e. possesses some missing data relevant to the study during future cycles of data); zero otherwise. The attrition variable was then interacted with all explanatory variables. Results for working age males and females found the attrition variable and all interaction terms to be statistically insignificant. Thus, based on this test it appears attritors do not statistically differ from those who remain. These regression results are available upon request.
4 Recent papers published using this questionnaire as a measure of mental health include: Hilton et al. (Citation2008), Marchand and Blanc (Citation2011), Marchand, Demers, and Durand (Citation2005), Orpana, Lemyre, and Gravel (Citation2009), and Stephens and Joubert (Citation2001).
5 The distribution of the distress variable has a positive skew. To test for robustness, analyses were also carried out measuring distress using the 0–24 point scale – results remained virtually unchanged.
6 For working age males, the correlation coefficient regarding these two variables is 0.12; for working age females, 0.16.
7 Those strongly disagreeing with this statement represent less than 1.5% of the sample.
8 This derivation is akin to past research by Marchand, Demers, and Durand (Citation2005) and Meltzer et al. (Citation2010).
9 The index is based on self-reported assessments of eight attributes – vision, hearing, speech, mobility, dexterity, cognitive ability, and pain. The index ranges from −0.36 to 1 with higher values associated with better health. Based on societal views concerning health status, perfect health is rated 1 and death is rated 0. Values below 0 are perceived to be worse than death.
10 Longitudinal weights provided by NPHS are used to account for the stratified sampling design and attrition.
11 For both males and females, a Hausman test rejects the null hypothesis that the preferred model employs random effects in lieu of fixed effects – i.e. the idiosyncratic errors appear to be correlated with the explanatory variables.
12 Real equivalent income is measured as the respondent’s best estimate of total household income after transfers but before taxes and deductions of all household members from all sources in the 12 months preceding the survey, adjusted for inflation and divided by the square root of the household size.
13 Measured as a dummy variable equal to unity if the respondent believed the statement ‘You don’t have enough money to buy the things you need’ to be true; zero otherwise.
14 Measured as a dummy variable equal to unity if the respondent believed they were having relationship problems; zero otherwise.
15 This index is based on research by Sherbourne and Stewart (Citation1991) with higher values on the scale representing higher levels of social support ranging from 0 to 32. The scale is based on the summation of eight questions, each with a five point scale (0 for ‘none of the time’ to 4 indicating ‘all the time’). The questions relate to the respondent having someone to: (i) listen to (ii) help them in a crisis, (iii) provide information, (iv) provide advice, (v) provide suggestions, (vi) share their most private worries and fears, (vii) confide in and (viii) understand their problems.
16 This index measures the daily energy expenditure of a respondent during leisure time activities over the 3 months preceding the survey using the frequency and duration of physical activity.
17 NPHS classified the respondent as having a chronic illness if they reported any of the following issues: Allergies, asthma, fibromyalgia, arthritis, back problems, high blood pressure, migraines, bronchitis/emphysema, diabetes, epilepsy, heart disease, intestinal/stomach ulcers, urinary incontinence, bowel disorder, dementia, cataracts, glaucoma, or thyroid problems.
18 NPHS asked respondents to report on their general health using a five point Likert scale ranging between ‘poor’ and ‘excellent’. If the respondent reported their general health as ‘fair’ or ‘poor’, they are classified as having poor self-reported health for that time period.
19 Based on the groupings of census divisions, Statistics Canada derives a set of 72 economic regions.
20 We recognize that any remaining variation over time in unobserved person-specific characteristics can still contaminate assertions of causality. Such concerns are discussed in Section VI.
21 A fixed-effects model evaluates the impacts of change. Unlike the pooled OLS model, which uses both between and within-person variation of each variable in determining the parameter estimate, the fixed-effects model uses only within-person variation – e.g. the parameter estimate for unemployment is based on individuals transitioning into and out of the state of unemployment.
22 For example, we instrumented for job insecurity using provincial level insecurity average and median values.