An online survey of Canadians living with and without a disability was conducted in August 2018. Eligible participants were 18 years of age and fluent in English and were employed for pay for at least 15 h/week. Given the challenges associated with recruiting large community-based samples of employed people with disabilities [32], potential survey participants were purposively recruited from an existing panel that is maintained by a research firm. A targeted recruitment approach was used to identify participants from the panel who were living with or without disabilities and across different age groups. The panel consisted of over one million Canadians and is nationally representative according to region and income [33]. Potential participants identified from the panel were contacted, provided study information and asked to complete a short screening questionnaire to determine eligibility. For those who chose to participate, informed consent was obtained, and the full online survey was administered. All study procedures were approved by the University of Toronto research ethics board (REB# 36184).
Survey measures
The survey was developed by the research team. Items and measures were selected based on their feasibility and evidence of precision, validity and reliability in previous studies of people with and without disabilities. When no existing item or measure existed, new ones were developed.
Outcome: precarious work
At the time of survey development, the research team were unable to identify any established measurement tools to examine precarious work that had been validated among people with disabilities. For the purpose of this study, we developed a specific measure of precarious work that draws directly from a synthesis of existing conceptual models and is relevant to people with and without disabilities [1, 3, 4]. Our measure included the following four items (see Appendix, Table S1):
Work hours
Using an open-ended question, participants were asked about the number of hours they worked per week. Respondents working < 30 h/week were categorized as working part-time and those working ≥ 30 h/week were categorized as working full-time [34].
Employment contract
Using one item developed for the survey, participants were asked if they were currently employed in a permanent position (i.e., no limit to duration) or a non-permanent position (i.e., limited duration in contract).
Job control
Participants were asked the following item “To what extent do you have control over your work schedule and how you do your work?”. Item response was on a five-point Likert scale (1 = not at all; 5 = a great deal). Those reporting ‘not at all’ or ‘a little’ job control were categorized as having low job control.
Union membership
One item, developed for this study, was used to assess whether participants belonged to a union and receive regulatory protection (1 = yes; 0 = no).
Using each item, participants were categorized as working precariously when they were employed in part-time hours and/or were working in a non-permanent contract and/or had no union representation and reported low job control.
Predictor variables
Disability
We utilize an adapted version of the Disability Screening Questionnaire (DSQ), which was designed by Statistics Canada to identify people living with a disability within population health surveys [35]. The DSQ is based on the WHO biopsychosocial model of disability to identify individuals who face activity limitations related to five disability categories [35]. In our survey, five items were used to ask participants about the extent to which they face difficulties at work that have lasted or are expected to last for 6 months or more and are related to a physical, cognitive, mental/emotional, sensory, or other disability. Item response occurred on a four-point scale (0 = no; 1 = some; 2 = often; 3 = always). Participants who reported at least ‘some’ difficulty on at least one item were categorized having a disability [35]. The DSQ has been extensively psychometrically tested and has exhibited reliability and validity [35].
Drawing from an industrial gerontological framework, both age and job tenure are examined as separate predictor variables [25, 26].
Age
Based on their age, participants were divided into: young (18–35 years), middle-aged (36–50 years) and older adult (> 50 years) groups.
Job tenure
Number of years employed in current job.
Covariates
Drawing from the WHO’s biopsychosocial model of disability and a large body of previous research, our analytical models adjusted for sociodemographic, health and work context factors [18, 36]. Specific covariates were selected when they were relevant to participants with and without disabilities.
Sociodemographic
Gender, educational attainment, marital status and personal income was collected.
Health factors
Participants were asked about their perceptions of their health using the widely utilized one-item self-rated health (1 = poor health; 5 = excellent health) [37]. Of note, self-rated health is a commonly used measure that can be applied to people with and without disability to capture global ratings of health and is seen as a powerful predictor of mortality and healthcare utilization [37, 38]. Additionally, pain and fatigue were also examined using visual analog scales (0 = no pain/fatigue; 10 = worst possible pain/fatigue) [39].
Work characteristics
Work characteristics were examined as covariates to account for different occupations and job roles of study participants.
Job sector (e.g., business/administration, health/science/teaching, sales/service, and trades/ transportation sectors) and organizational size were collected (e.g., small [1–50 people], medium [51–150 people] and large [> 150 people]). Two questions asked about the extent of physical and mental work demands (1 = not at all; 5 = a great deal). To assess productivity loss, one item from the Work Productivity and Activity Impairment instrument was utilized. Participants were asked the extent to which their health affected their job in the last month (0=“health had no effect on my work”; 10=“health completely prevented me from working”). The item is a valid and reliable tool to examine lost productivity attributed to disability [40].
Analyses
Descriptive statistics (i.e., frequencies and means) were used to build a profile of the study sample and to examine variable distributions. Bivariate analyses (chi-square and t-tests) were conducted to examine how study variables differed between those with and without a disability.
Univariable logistic regression models were conducted for the total sample and for those with and without a disability to examine the association between predictor variables and study covariates and the likelihood of reporting precarious work. For multivariable modelling, covariates that were significantly associated with employment in the univariable model and did not exhibit multicollinearity with other covariates or the outcome variable were carried forward. To test study hypothesis one, a multivariable logistic regression model was conducted to examine the relationship between disability and precarious work when adjusting for study covariates.
To test study hypotheses two and three, a multigroup probit model using weighted least square mean and variance adjusted estimation (WLSMV) was conducted to examine the relationships between age and job tenure and precarious work, as well as to examine the modifying effect of disability. Multigroup modeling tests similarities and differences in coefficients of interest across a grouping variable (i.e., disability). To develop a multigroup model, a partially constrained model (i.e., parameter of interest differs between those with and without a disability) is compared to a fully constrained model (i.e., parameters are fixed across those with and without a disability). Through this approach, the modifying effect of disability can be determined. In our study, separate partially constrained models were conducted where all coefficients were constrained except for age (model a) and job tenure (model b). Also, separate models were also conducted for gender (model c) and physical job demands (model d) and mental job demands (model e) to account for their theoretical importance to precarious work. Each of the partially constrained models were then compared to a fully constrained model. Equality of coefficients across those with and without a disability were tested using the Satorra-Bentler scaled chi-square difference test for WLSMV implemented in Mplus [41].
Drawing from the findings from the multigroup probit model using WLSMV estimation, a final partially constrained multigroup probit model enabled the estimation of odds ratios. The multigroup probit model was stratified for those with and without disabilities. Also, age and job tenure, as well as gender and physical and mental job demands, were unconstrained in the final model; all other parameters were constrained to be equal. The multigroup model was also estimated using maximum likelihood parameter estimates with standard errors and a chi-square test statistic that was robust to non-normality. Analyses were conducted using SAS version 9.3 [42] and Mplus software [41].