This study used data from the second (2009) wave of HABITAT (How Areas in Brisbane Influence healTh and ActTvity), a multilevel longitudinal study (2007–2018) of mid-age adults living in Brisbane, Australia. Consent to participate was obtained via return of the participants’ completed survey. Data for this study were analysed from June 2020 to January 2021. Methods were carried out in accordance with relevant guidelines and regulations; HABITAT received ethics approval from the Queensland University of Technology Human Research Ethics Committee (Ref. No. 3967H & 1,300,000,161).
Study sample
HABITAT’s sampling design has been published elsewhere [30]. Briefly, a multi-stage probability sampling design was used to select a stratified random sample (n = 200) of Census Collector’s Districts (CCD) in 2007 [31]. CCDs were the smallest administrative units used by the Australian Bureau of Statistics (ABS) in 2007, containing an average of 200 private dwellings. A random sample of people aged 40–65 years from each neighbourhood were invited to participate (approximately 85 people per CCD).
Eligible study participants were mailed a self-administered survey between May and July in the years 2007, 2009, 2011, 2013 and 2016 using the method by Dillman [32]. Of 16,128 surveys mailed in 2007, valid responses were received from 11,035 (68.4% response rate). Respondents were representative of the 2006 Brisbane population, although residents from disadvantaged areas, blue-collar employees, and persons who did not attain a post-school educational qualification were underrepresented [33]. In 2009, 7866 (72.3%) eligible and contactable participants responded.
Measures
Mental well-being, the outcome, was measured using the Warwick-Edinburgh Mental Wellbeing Scale (WEMWBS), which comprises items on subjective well-being, psychological functioning, and relationships [7]. The 2009 wave of HABITAT asked all 14 WEMWBS items. The 14-item WEMWBS scale has been well validated [7, 34] and used in the UK to monitor population-level well-being and evaluate interventions, policies, and programs aimed at improving mental wellbeing [7]. Responses to each item are scored on a 5-point Likert scale, ranging from ‘none of the time’ (1) to ‘all of the time’ (5), then summed to give a total score. The potential minimum and maximum scores are 14 and 70, respectively, with scores of 45–59 indicating average MWB and scores of 60 or more indicating high MWB [7]. In the current study, Cronbach’s alpha of the scale items was high at 0.96.
Socioeconomic predictor variables included education, occupation, household income and neighbourhood disadvantage.
Education
Participants selected their highest level of education attainment from nine response categories. These were recoded to bachelor’s degree or higher (including graduate certificate or diploma, Masters’ degree or doctorate), diploma or associate degree, certificate (trade or business) and no post-secondary school qualification.
Occupation
Employed respondents provided the full title of their occupation. This information was subsequently coded to the Australian and New Zealand Classification of Occupations (ANZCO) [35]. The original nine-level classification was recoded into three categories: managers/professionals (managers and administrators, professionals and para-professionals), white collar employees (clerks, sales-persons and personal service workers), and blue-collar employees (tradespersons, plant and machine operators and drivers, and labourers and related workers). Non-employed respondents were classified as home duties, retired, permanently unable to work, unemployed, or not easily classifiable (student, other, or missing).
Annual household income
Participants selected their pre-tax household income from 13 categories. These were recoded into six categories: ≥A$130,000, A$129,999- A$72,800, A$72,799-A$52,000, A$51,999-A$26,000, ≤A$25,999, and ‘don’t know’/ ‘don’t want to answer’.
Neighbourhood disadvantage
A neighbourhood socioeconomic disadvantage measure was derived using scores from the Australian Bureau of Statistics (ABS) Index of Relative Socioeconomic Disadvantage (IRSD) [35]. IRSD scores were calculated using 2006 census data and derived by the ABS using principal component analysis [36]. A neighbourhood’s IRSD score is a measure of an area’s overall level of disadvantage. It was calculated using 17 variables that captured a wide range of socioeconomic attributes, including education, occupation, income, unemployment, household structure and household tenure. For analysis, the 200 sampled neighbourhoods were grouped into quintiles, with Q1 denoting the 20% (n = 40) least disadvantaged and Q5 the most disadvantaged 20% (n = 40) areas, relative to the whole of Brisbane.
Covariates were age and sex. Age was derived from self-reported date of birth and categorised into five groups: 42–46, 47–51, 52–56, 57–61, 62–67 years.
Statistical analysis
Of 7866 residents who completed the 2009 survey, the 568 (15%) who changed their residential address after the 2007 data collection were excluded to reduce potential selection bias due to movers being influenced by unmeasured preferences related to both residential choice and MWB [37]. Another 162 were excluded because they were not the same household respondent as in Wave 1, which resulted in their education data, collected in 2007, not being relevant to the data collected in 2009, and 277 were excluded because they had not completed all WEMWBS items. Respondents with missing data on any individual-level predictor variable, except occupation, were excluded from analysis (n = 138). After excluding these respondents, 6721 individuals were available for analyses.
Two multilevel linear regression models were used to examine associations between individual-level SEP, neighbourhood disadvantage and MWB. MWB score was the outcome variable. The independent variables of interest in Model 1 were education, household income and occupation as measures of individual-level SEP, with adjustment for age and sex. Model 2 added to Model 1 neighbourhood disadvantage as an independent variable of interest. Preliminary analysis showed that results for Models 1 and 2 were similar for men and women; therefore, they were analysed together. A sensitivity analysis was conducted to determine if the results for Model 2 changed when a variable representing the years lived at the current address was added. The results did not change, and hence, the results of the sensitivity analysis are not presented. Cross-level interactions between individual SEP and neighbourhood disadvantage on MWB scores were also modelled (adjusted for the other SEP variables) to examine variation in the mean MWB score for education, occupation and household income, by level of neighbourhood disadvantage. Data were analysed using Stata 15.1 (StataCorp, College Station, Texas).