Purpose Most previous research place great importance on the influence of family and maternal background on child and adolescents’ mental health. However, age of onset studies indicates that the majority of the mental health disease prevalence occurs during the youth years. This study investigates the relationship of family and maternal background, as well as individual circumstance on youth mental health status. Method Data from 975 participants and 4632 observations of aged cohort 15 to 19 years in the Household, Income and Labour Dynamics in Australia (HILDA) longitudinal study were followed for 10 years (2007–2017). Multilevel logistic regression models were used to analyse the impact of youth circumstances on mental health status. Results The findings suggests that not all dimensions of family and maternal background (especially maternal education) have impacts on youth mental health. We found low household income (AOR: 1.572, 95% CI: 1.017–2.43) and adverse living arrangement (AOR: 1.586, 95% CI: 1.097–2.294) significantly increases mental disorder odds whereas maternal education or occupation fixed effects were not significant. Individual level circumstances have much stronger impact on youth mental health. We found financial shock (AOR: 1.412, 95% CI: 1.277–1.561), life event shock (AOR: 1.157, 95% CI: 1.01–1.326), long term health conditions (AOR: 2.855, 95% CI: 2.042–3.99), smoking (AOR: 1.676, 95% CI: 1.162–2.416), drinking (AOR: 1.649, 95% CI: 1.286–2.114) and being female (AOR: 2.021, 95% CI: 1.431–2.851) have significant deteriorating effects on youth mental health. Conclusions Our finding is in contrast to the majority of studies in the literature which give a preeminent role to maternal characteristics in child and youth mental health status. Mental health interventions should consider heterogeneity of adverse youth circumstances and health-related behaviours.
All our analyses are based on sample data from the Household, Income and Labour Dynamics in Australia (HILDA) panel survey [16]. This nationally representative household survey has been carried out annually from 2001 through 2018 (waves 1–18). It interviews and subsequently reinterviews all members aged 15 years and over of the same selected household every year. More than 30,000 individuals (40,000+ enumerated) have participated in the survey over the years and on average 15,000 individuals have been interviewed every year. A 90% wave on wave response rates of HILDA survey are comparable with other large longitudinal surveys like the British Household Panel Study (BHPS) or Panel Study of Income Dynamics (PSID) [17]. Details of HILDA sample design, survey response rates and attrition rates can be found elsewhere [17]. For the purpose of this study, we limit the sample to young Australians aged 15–19 years (late adolescent period) at the baseline wave (wave 7) and then followed the participants for 10 years (up to six measurement points) which covers youth (20–24 years) and transition to adulthood phase (25–29 years) in the follow up. We chose to start from wave 7, because HILDA survey did not start to collect Kessler Psychological Distress Scale (K10) scores (our main outcome of interest) in earlier waves and it provides the score subsequently in every odd wave (every two years) thereafter. Thus, we constructed an unbalanced panel data using wave 7, 9, 11, 13, 15 and 17. To be included in the analyses, the participants had to be interviewed in the baseline wave 7 and has to appear in at least one of the follow-up waves. Our final sample contains 975 participants across the six waves with a total of 4,632 observations. The 15–19 age cohort was thus followed up to 25–29 years with an average of 5.18 observations per person. The participant flow into the sample is shown in Fig 1. This study uses the Kessler Psychological Distress Scale (K10) as the measure of mental health outcomes and is the main dependent variable for analyses [18]. In clinical practice, the scale is used to assess the likelihood of having a mental disorder; for example, a person with a score of 10–15 has a low risk of having a mental disorder whereas a person with a score of 20–24 is likely to have a mild mental disorder, a score of 25–30 would indicate a likely moderate mental disorder and a person with a score of 30–50 is likely to have a severe mental disorder [19]. In the analyses, we use a dichotomous K10 variable (where a score of greater than 20 depict the likelihood of a mental disorder) as measures of our dependent variable for mental health performance [20]. Following Roemer’s equality of opportunity theory [21, 22] we classify all our exposure variables into two types: i) circumstances category and ii) effort category. The theory of equality of opportunity revolves around the goal of compensating for ‘negative’ circumstances (such as parental background) on health outcomes while controlling the health inequalities generated by effort category variables (such as lifestyle or health habits) that can be attributed to the behaviour of an individual. We use the biological mothers’ education level and occupational status, household income and family living arrangements (whether the participant lived with both parents at the age of 14 years old) to determine the family and maternal background status as a group level characteristic of the circumstances category. We define maternal education level as low if the highest qualification level obtained by the mother is secondary level or lower. We use the Australian Socioeconomic Index 2006 (AUSEI06) occupational status scale as the measure of the occupational status of mother [23]. We assign occupational status as low if the value range falls in the lowest quintile. Similarly, we assign household income as low if the equivalised household income range falls in the lowest quintile. Using household income, family living arrangement, maternal education and occupational status we have constructed 16 (2x2x2x2) different types of family and maternal background history groups for the multilevel analyses. We use the number of financial shocks, number of life event shocks and long-term health conditions in the individual level circumstances category [12]. The number of financial shock variable shows the number of adverse financial events the study participant has experienced (for example: went without meals or asked for financial help from friends or family). Similarly, life event shock variable shows the number of life events related to grief, loss or injury the study participant has suffered (for example: death of a family member or serious personal injury). The list of events that constitutes financial and life event shocks are given in the S1 Appendix. We use negative health habits such as being obese (as a proxy of unhealthy eating and lack of exercising), being a daily smoker and regular drinker (drinks more than four standard drink/day), and positive health habits such as being an active member of a sporting/hobby/community-based club or association as an effort type of variables. This study also included gender and rural residency as demographic covariates in the analyses based on past literature [24]. In addition, we construct our time variable by setting zero at the baseline wave 7 and subsequently adding two for each additional measurement point (since between wave time is two years and there are up to six measurement points) to get a ten-year follow-up at wave 17 (t = 0,2,4,6,8, and 10). The authors constructed an unbalanced longitudinal data set of the youth cohort by linking an individual’s record who participated in the baseline (wave 7) at age 15–19 years and in one of the follow-up waves (9, 11, 13, 15 and 17). Descriptive statistics and mental health opportunity profile were summarized to understand the impact of family and maternal background group characteristics on youth mental health. Visual trends of psychological distress scale were analysed for group level characteristics. Traditional single level regression analysis such as logistic regression model only assumes fixed-effect impacts of dependent variables and does not allow for random effects of intercepts and slopes for individual and group level characteristics. However, data structure can be nested or clustered by some observable characteristics that creates similarity between individuals and ignoring these phenomena can violate the independence assumption of regression analysis. Multi-level models allow for a nested data structure (i.e., repeated measures) and make it possible to study sources of variance at different levels of an outcome variable [25]. The nested data structure is illustrated in Fig 2. In our analyses, we used both single level logistic regression and multilevel logistic regression models. we have nested our data structure into three levels: i) time, ii) individual, and iii) family and maternal background history types (a total of 16 different background history types; for example a background history type could be: household income- high; from two types: ‘high’ and ‘low’, mothers education- low; from two types: ‘high’ and ‘low’, mothers occupation- low; from two types: ‘high’ and ‘low’ and family living arrangement—whether not lived with both biological parents- yes; from two types: ‘yes’ and ‘no’. Thus, we have 2x2x2x2 = 16 types. A full combination of 16 types can be seen in Table 2’s opportunity profile). We assigned unique identifiers (From 1 to 16, see Table 2’s opportunity profile’s rank number for identifiers) for each group for the analysis. We control for individual fixed effects characteristics like circumstances and effort covariates in level 2 and group level fixed effects characteristics like various family and maternal background group characteristics in level 3. All statistical analyses were conducted using Stata 15. The HILDA study was approved by the Human Research Ethics Committee of the University of Melbourne. The study used only de-identified existing unit record data from the HILDA survey. The authors completed and signed a confidentiality agreement with NCLD (ua.vog.ssd@hcraeserdlcn) and obtained database access from the Australian Data Archive (ua.ude.una@ada) following application acceptance. Thus, the dataset studied during this work were subject to the signed confidentiality agreement.