Sample and study design
The Generation 1 study is a longitudinal cohort study of women and their children (n = 557) established between 1998 and 2000 in Adelaide, South Australia. The study is described in detail elsewhere [17–19]. In brief, women were recruited prior to 16 weeks gestation at a public hospital or through the private practices of three obstetricians. Eligible women were aged ≥18 years, Caucasian, and free from certain conditions known to affect fetal growth . The sample of mothers was broadly representative of all women who gave birth in South Australia during the cohort’s establishment, as described previously . The University of Adelaide Human Research Ethics Committee approved the study. All participants gave written informed consent, and the study procedures conformed to the principles of the Declaration of Helsinki.
Mothers and children were followed up during pregnancy (at 16 and 32 weeks), at birth and on eight occasions in early childhood (at 3, 6, 9 and 12 months, and at 2, 3.5, 5 and 9 years). Data pertaining to the children and their wider family circumstances have been collected at each study wave.
The outcomes considered were internalizing behaviour problems (withdrawn/depressed) and externalizing behaviour problems (aggressive/destructive), derived from Achenbach’s Child Behaviour Checklist (CBCL) age 6–18 years parent report form . The CBCL is a well-validated 113-item instrument designed to record children’s competencies and problems as reported by their parents. In the present study, mothers rated each CBCL item as not true (0), somewhat true (1) or very true (2), as part of the interview schedule when study children were aged 9 years. Internalizing and externalizing behaviour problem scores were derived according to standard methods . Analyses were based on raw scores, which are recommended for use in community based samples .
Three aspects of residential mobility were assessed. The effects of the number of house moves in the periods birth to <2, 2 to <5 and 5 to 9 years, were separately considered. In each of these periods, the number of house moves was classified as 0, 1, or ≥2. The effect of the lifetime (total history of) number of house moves from birth until 9 years was also considered (0, 1, or ≥2 moves).
In order to assess moves associated with different housing trajectories, a variable describing change in housing tenure between 2, 3.5 and 9 years of age was constructed. At each of these time points, housing tenure was classified as private rental, public rental, mortgage/own (including purchase of public housing), living with extended family or living in a residence rent free. Families were classified as having an upwardly mobile housing trajectory if at any time the housing tenure changed from private rental, public rental or living with extended family to mortgage/own or living in a residence rent free, and the change was sustained (i.e. no subsequent tenure changes). Families were classified as having a downwardly mobile housing trajectory if at any time the housing tenure changed from mortgage/own or living in a residence rent free to private rental, public rental, or living with extended family, and this change was sustained. The trajectory was classified as ‘mixed’ when both upward and downward housing tenure changes occurred. Families with no upward or downward housing tenure changes were classified as having a trajectory of either continuous home ownership (including families who never moved or moved between purchased properties) or continuous rental occupancy (including families who never moved or moved between rental properties). The number of house moves in each period was used as an additional check in the derivation of the measure of housing trajectories.
Information on potentially relevant covariates was taken from the pregnancy questionnaires and the assessments at previous study waves. Covariates considered were maternal age at birth of the study child, child sex, highest level of maternal education prior to the birth of the study child (High School (HS) not completed, HS not completed but Technical and Further Education College (TAFE) or University completed, HS completed only, HS and TAFE completed, or HS and University completed), average annual household income prior to the birth of the study child (≤$31,199, $31,200-$51,999, ≥$52,000), change in parental relationship status (separation at any time from birth to 9 years), changes in the number of children in the household (≤1, 2, 3 or ≥4 children in the household at 2 years; ≥1 additional child in the house from 2 to 3.5 years), stressful events in the family between birth and 2 years (any of the following: family court matters, restraining orders, criminal charges, deaths in the immediate family, or any other stressful events nominated by the family), and change in school for the study child between reception (the first year of school) and 9 years (0, 1, ≥2 school moves).
The bivariate associations between housing variables, child behaviour problem scores and covariates were investigated through chi-square tests of association for categorical variables and Mann–Whitney U-tests for variables with a continuous distribution.
Linear regression analyses were used to investigate the effect of the housing variables on internalizing and externalizing behaviour problem scores at 9 years. Models were built with the following: (1) number of house moves in each time period; (2) lifetime number of house moves at 9 years; and (3) moves associated with different housing trajectories characterized by housing tenure change. To assess the effects of sociodemographic and household covariates on the fit of each model, models that adjusted for the following factors were also included in the suite of analyses: (a) maternal age, child sex and maternal education; (b) = (a) plus household income, parental relationship status, household composition, stressful events between birth and 2 years, and change in school between reception and 9 years. A potential interaction between house moves from birth to 9 years and school moves from reception to 9 years was tested in each model.
Due to the positive skew in the distribution of the CBCL responses, a square root transformation was applied to the externalizing and internalizing problem scores, and the transformed scores were used in subsequent analyses. Results from the regression analyses are presented as regression coefficients (β) and their 95% Confidence Intervals (CI). A P-value <0.05 was considered statistically significant. All regression analyses were conducted using the REG procedure in SAS version 9.2 (SAS Inc, Cary, NC).