Article Text

Download PDFPDF

Early life socioeconomic determinants of dietary score and pattern trajectories across six waves of the Longitudinal Study of Australian Children
  1. Constantine E Gasser1,2,
  2. Fiona K Mensah2,3,
  3. Jessica A Kerr1,2,
  4. Melissa Wake1,2,4
  1. 1 Centre for Community Child Health, Murdoch Children’s Research Institute, Royal Children’s Hospital, Parkville, Victoria, Australia
  2. 2 Department of Paediatrics, University of Melbourne, Royal Children’s Hospital, Parkville, Victoria, Australia
  3. 3 Clinical Epidemiology and Biostatistics Unit, Murdoch Children’s Research Institute, Royal Children’s Hospital, Parkville, Victoria, Australia
  4. 4 Department of Paediatrics and The Liggins Institute, University of Auckland, Auckland, New Zealand
  1. Correspondence to Constantine E Gasser, Centre for Community Child Health, Murdoch Children’s Research Institute, Royal Children’s Hospital, Parkville 3052, VIC, Australia; constantine.gasser{at}mcri.edu.au

Abstract

Background Social patterning of dietary-related diseases may partly be explained by population disparities in children’s diets. This study aimed to determine which early life socioeconomic factors best predict dietary trajectories across childhood.

Methods For waves 2–6 of the Baby (B) Cohort (ages 2–3 to 10–11 years) and waves 1–6 of the Kindergarten (K) Cohort (ages 4–5 to 14–15 years) of the Longitudinal Study of Australian Children, we constructed trajectories of dietary scores and of empirically derived dietary patterns. Dietary scores, based on the Australian Dietary Guidelines, summed children’s consumption frequencies of seven groups of foods or drinks over the last 24 hours. Dietary patterns at each wave were derived using factor analyses of 12–16 food or drink items. Using multinomial logistic regression analyses, we examined associations of baseline single (parental education, remoteness area, parental employment, income, food security and home ownership) and composite (socioeconomic position and neighbourhood disadvantage) factors with adherence to dietary trajectories.

Results All dietary trajectory outcomes across both cohorts showed profound gradients by composite socioeconomic position but not by neighbourhood disadvantage. For example, odds for children in the lowest relative to highest socioeconomic position quintile being in the ‘never healthy’ relative to the ‘always healthy’ score trajectory were OR=16.40, 95% CI 9.40 to 28.61 (B Cohort). Among the single variables, only parental education consistently predicted dietary trajectories.

Conclusion Child dietary trajectories vary profoundly by family socioeconomic position. If causal, reducing dietary inequities may require researching underlying pathways, tackling socioeconomic inequities and targeting health promoting interventions to less educated families.

  • diet
  • education
  • socio-economic

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Introduction

Differences in children’s diets between population subgroups1–3 could partly underlie the profound social patterning4 of diseases such as obesity5 and dental caries.6 Understanding how, and how much, socioeconomic determinants shape dietary trajectories throughout the first 15 years seems an essential prerequisite to understanding the early life determinants of dietary-related diseases.

Myriad determinants may plausibly underlie social disparities in diets.7–10 First, higher adult education is associated with better nutritional knowledge9 which, for parents, may translate to better dietary habits in their children.9 Additionally, Australian research shows that compared with high-income families, low-income families must spend a greater proportion of their household income to maintain a healthy diet.10 Remoteness may be another important driver of children’s diets.8 Research conducted in Queensland, Australia, shows that the cost of healthier foods is higher in very remote areas, whereas the availability of fruits and vegetables is greater in highly accessible areas.8 Finally, parental employment is associated with less time preparing food, grocery shopping and eating with children.7 These time pressures might result in parents purchasing more processed food products for their children.7

Several longitudinal studies have considered socioeconomic determinants of diet in infancy, childhood and/or adolescence.1–3 11–15 Higher socioeconomic status (socioeconomic status considered parental education, family eligibility for public assistance, eligibility for free or reduced cost school meals and parental employment status12) and maternal education predict healthier dietary patterns or clusters.1–3 11 12 Conversely, lower socioeconomic status, lower maternal education and a higher household disadvantage composite index predict unhealthier dietary patterns or clusters during childhood or adolescence.1–3 11 12 15 Household/family income and parental employment have generally not predicted dietary patterns, clusters or indices during early to middle childhood.2 11 13 14

In a life course framework, the accumulation of diet with interacting risk factors over time is likely to protect against or lead to dietary-related diseases.1 In order to understand diet’s impact on evolving health and disease, it is therefore vital to measure diet at multiple time points because dietary patterns or scores accumulate and evolve over childhood. Yet to be investigated are socioeconomic determinants of childhood diet over long follow-up periods (more than 6 years), including several consecutive dietary measurements (at more than three time points) and in more than one cohort allowing generalisation of findings.16

The Longitudinal Study of Australian Children (LSAC) provides an ideal opportunity to investigate these issues. Dietary scores add up the frequency of foods eaten that are believed, on the basis of a priori knowledge, to either lead to or protect against disease.17 Dietary patterns, however, describe empirically how foods actually group together within typical diets with no preconception as to their healthiness.17 We aimed to quantify, in parallel population representative cohorts of Australian children, which of the major early life socioeconomic factors most strongly predict dietary (1) score and (2) healthy and unhealthy pattern trajectories from 2–3 to 10–11 years of age and from 4–5 to 14–15 years of age.

Methods

Recruitment and sampling

This study used data obtained from LSAC between 2004 and 2014.18 Using the universal Medicare database in which 98% of Australians are enrolled by 12 months, LSAC recruited a sample designed to represent Australia’s states/territories and urban/rural mix, excluding only its most remote areas.18 The study involved a two-stage clustered sampling design, comprising random selection of 311/3325 (9.4%) of Australian postcodes, and then of an average of 40 children in the larger states (20 children in the smaller states and territories) from each of these postcodes.18 LSAC recruited 4983 and 5107 children into its Kindergarten (K) and Baby (B) cohorts, aged 4–5 and 0–1 years at baseline, respectively.18 19 LSAC interviewers have collected data in participants’ homes every 2 years since 2004.19–21 Online supplementary figure 1 shows the flow and retention through LSAC and the percentages of children remaining in the study after each wave. Retention methods included birthday cards, annual calendars, newsletters and between wave mail-out and online questionnaires to keep contact with study families.20 Families gave written consent to participate, and the Australian Institute of Family Studies Ethics Committee approved each wave.

Supplementary file 1

Procedures and measures

LSAC’s data collection methods, relevant to this publication, included online questionnaires, completed by children from age 10 onwards for outcome data (relating to diet), and face-to-face interviews with the primary caregiver (parent 1) for all exposures, all covariates and outcome data for children under 10 years of age.22 Parent 1 was usually the biological mother, and parent 2, usually the biological father, was either the partner of parent 1 or another adult with a parental relationship to and in the same home as the study child.20

The six outcome variables, three in each cohort, were our published trajectories of the study child’s overall dietary scores, ‘healthy’ dietary patterns and ‘unhealthy’ dietary patterns,23 as shown in online supplementary figures 2–4. These trajectories were derived from the short food diary completed at each wave. Briefly, children (aged 10 years or older) or parents (for children aged 2–9 years) completed 12–16 questions regarding the study child’s consumption of groups of or individual food or drink items within the last 24 hours (or yesterday) (online supplementary table 1).23

Supplementary file 5

Briefly, we first derived an a priori dietary score for each individual and at each wave. This score, which was based on the Australian Dietary Guidelines,24 comprised the sum of a child’s consumption frequency of seven categories of foods or drinks over the previous 24 hours: fruit, vegetables, water, fatty foods, sugary foods, sweetened drinks and milk products or alternatives. Fruit, vegetables, water and milk products or alternatives were positively coded, whereas fatty foods, sugary foods and sweetened drinks were negatively coded. Overall scores ranged from 0 to 14 (0–2 for each category), with 14 being healthiest. In order to derive overall score trajectories, we employed group-based trajectory modelling, using the ‘traj’ plug-in in Stata/IC V.14.1.25 These trajectories provide a summary of a how children’s dietary scores change from 2–3 to 10–11 years (B Cohort) or from 4–5 to 14–15 years (K Cohort) (online supplementary figure 2).

Supplementary file 2

Second, we empirically derived dietary patterns at each wave by conducting exploratory factor analyses26 using 12–16 healthy and unhealthy food or drink items, depending on wave, to derive the factors. For factor analyses, we used polychoric correlation matrices26 because all dietary variables were ordinal. At each wave of both cohorts, we obtained similar ‘healthy’ and ‘unhealthy’ factors. The ‘healthy’ pattern was characterised by frequent consumption of cooked vegetables, raw vegetables or salad and fresh fruit in all waves, and water in most waves, each with high factor loadings of 0.3 or higher, reflecting these as foods or drinks that stand out most in the ‘healthy’ factor. The ‘unhealthy’ pattern was characterised by frequent consumption of savoury snacks and sweetened drinks in all waves; hamburgers, sausages or sausage rolls, meat pies, hot dogs, hot chips and fruit juice in most waves (each with high factor loadings of 0.3 or higher); and a high negative factor loading (below −0.3) for water consumption in 6 out of 11 waves. We then calculated dietary pattern scores for the healthy and unhealthy factors separately, for each participant and each cohort, using regression scoring. In order to derive the separate sets of ‘healthy’ and ‘unhealthy’ pattern trajectories, we similarly used group-based trajectory modelling.

All dietary trajectories were derived from dietary measures over five biennial waves for the B Cohort (from age 2–3 years) and six biennial waves for the K Cohort (from age 4–5 years). Individuals needed to have overall or pattern scores from two or more waves to be included in the relevant trajectories (online supplementary figure 1). The six trajectory outcomes (scores, healthy patterns and unhealthy patterns for each cohort) each comprised four categorical trajectories. The overall score and ‘healthy’ pattern dietary trajectories consisted of: ‘always healthy,’ ‘becoming less healthy,’ ‘moderately healthy’ and ‘never healthy’. The ‘unhealthy’ pattern dietary trajectories consisted of: ‘never unhealthy,’ ‘becoming unhealthy,’ ‘moderately unhealthy’ and ‘always unhealthy’. It was possible for the same child to follow both the ‘always healthy’ (with a high intake of healthy foods, such as fruit) and ‘always unhealthy’ (with a high intake of unhealthy foods, such as savoury snacks) dietary trajectories.

Exposures and covariates were measured at the baseline wave (wave 1) of each cohort and related to the study child, parent 1, the partner of parent 1 and/or parent 2. Table 1 summarises the individual and composite socioeconomic exposures and covariates for this study. The census-based Australian Bureau of Statistics Index of Relative Socio-economic Disadvantage summarises information regarding economic and social resources of households and people in an area,27 in this case postcode. The index includes information on 17 different measures of relative disadvantage, including low education, low income, unskilled occupations and high unemployment.27 Socioeconomic position combines and averages information on parental occupational status, parental educational attainment and annual family income (table 1).28

Table 1

LSAC exposures and covariates at wave 1 and their handling for this study

Statistical analysis

We performed all statistical analyses using Stata/IC V.14.2. Analyses used survey methods with the cross-sectional sampling weights from wave 1 of the B and K cohorts. These methods account for non-response and the multistage and clustered sampling design.20 We investigated associations between socioeconomic factors in wave 1 of both cohorts and children following score and pattern trajectories using univariable and multivariable multinomial logistic regression analyses. For multivariable analyses, we identified potential confounders or covariates a priori from previous studies1–3 11 13 29–32 and directed acyclic graphs. Analyses included neighbourhood disadvantage and various single socioeconomic measures in the same model, along with the covariates ethnicity (measured by study child Indigenous status and study child language other than English spoken at home), parent 1 age and parent 2 age. Further analyses included neighbourhood disadvantage and the composite measure socioeconomic position in the same model and with the same covariates. Drawing on previous literature,2 11 29–31 we conducted analyses whereby models were additionally adjusted for the study child’s number of older siblings and whether the study child was a singleton or from a multiple birth. For exposure variables with more than two categories, we calculated overall p values, using the Wald test, for all non-reference exposure categories combined (compared with the reference category) and for each non-reference dietary trajectory.

Results

Baseline characteristics

Online supplementary figure 1 shows the numbers of participants at each stage of LSAC; 3764 (73.7%) and 3537 (71.0%) children from the B and K cohorts, respectively, remained in the study after wave 6. Table 2 shows the baseline characteristics of the sample. The mean ages of children included in and excluded from the trajectories (0.7 and 4.8 years for the B and K cohorts at wave 1, respectively) were very similar, as were proportions of boys and girls. In both cohorts, compared with excluded children, higher proportions of those included in all three outcomes belonged to the least disadvantaged quintile of neighbourhood disadvantage and highest socioeconomic position quintile and had the highest levels of parental education. Dietary data for the B Cohort were collected at five waves for 8 years, and dietary data for the K Cohort were collected at six waves for 10 years. Individuals with data from at least two waves were included in the dietary trajectories (87.0% and 92.7% from the B and K cohorts, respectively) (online supplementary figure 1).

Table 2

Baseline characteristics* of the sample at wave 1 by cohort

Associations between socioeconomic factors and dietary trajectory outcomes

In univariable analyses, low parental education, low combined parental yearly income, family food insecurity, living in rented accommodation and belonging to the most disadvantaged quintile of neighbourhood disadvantage and lowest socioeconomic position quintile were strongly associated with children following all three less healthy dietary score trajectories (‘never healthy,’ ‘moderately healthy’ and ‘becoming less healthy’) rather than the ‘always healthy’ trajectory in both cohorts (online supplementary table 2).

Figures 1–3 graphically summarise the relationships (expressed as ORs with 95% CIs) between socioeconomic position, neighbourhood disadvantage and parental education, and the dietary score, ‘healthy’ pattern, and ‘unhealthy’ pattern trajectories, respectively, in multivariable analyses. In multivariable analyses, adjusting for covariates, low parental education and belonging to the lowest socioeconomic position quintile remained strongly associated with children following all three less healthy dietary score trajectories in both cohorts (figure 1; online supplementary table 3). Thus, compared with the highest socioeconomic position quintile, children in the lowest socioeconomic position quintile experienced over 16 times the odds of following the ‘never healthy’ rather than the ‘always healthy’ score trajectory (OR 16.40; 95% CI 9.40 to 28.61; p<0.001) in the B Cohort. Belonging to the most disadvantaged quintile of neighbourhood disadvantage greatly increased, but to a much lesser degree, the odds of children following the ‘never healthy’ score trajectory in both cohorts (OR 2.14; 95% CI 1.25 to 3.65; p=0.006 for the B Cohort and OR 2.44; 95% CI 1.53 to 3.89; p<0.001 for the K Cohort) and the ‘moderately healthy’ score trajectory in the B Cohort. We observed some associations between single socioeconomic measures other than education (combined parental yearly income, remoteness area, home ownership and family food insecurity) and less healthy dietary score trajectories. However, these associations were inconsistent between the three trajectories and two cohorts (online supplementary table 3).

Supplementary file 3

Figure 1

Results of multinomial adjusted multivariable logistic regression analyses, showing associations between SEP, SEIFA neighbourhood disadvantage and either parent highest education, and overall dietary score trajectories for the B Cohort (A) and the K Cohort (B). Symbols correspond to ORs and short black horizontal lines correspond to lower and upper limits of 95% CIs. The reference category for SEP and SEIFA neighbourhood disadvantage quintiles is quintile 1, and the reference category for either parent highest education is ‘degree’. Quintile 5 is the most disadvantaged quintile for SEP and SEIFA neighbourhood disadvantage. ORs are relative to the ‘always healthy’ trajectory. Associations presented for SEIFA neighbourhood disadvantage are those for the model where SEIFA neighbourhood disadvantage was included with single socioeconomic measures. B Cohort, Baby Cohort; BLH, becoming less healthy; K Cohort, Kindergarten Cohort; MH, moderately healthy; NH, never healthy; SEIFA, Socio-Economic Indexes for Areas; SEP, socioeconomic position.

Figure 2

Results of multinomial adjusted multivariable logistic regression analyses, showing associations between SEP, SEIFA neighbourhood disadvantage and either parent highest education, and ‘healthy’ pattern trajectories for the B Cohort (A) and the K Cohort (B). Symbols correspond to ORs and short black horizontal lines correspond to lower and upper limits of 95% CIs. The reference category for SEP and SEIFA neighbourhood disadvantage quintiles is quintile 1, and the reference category for either parent highest education is ‘degree’. Quintile 5 is the most disadvantaged quintile for SEP and SEIFA neighbourhood disadvantage. ORs are relative to the ‘always healthy’ trajectory. Associations presented for SEIFA neighbourhood disadvantage are those for the model where SEIFA neighbourhood disadvantage was included with single socioeconomic measures. B Cohort, Baby Cohort; BLH, becoming less healthy; K Cohort, Kindergarten Cohort; MH, moderately healthy; NH, never healthy; SEIFA, Socio-Economic Indexes for Areas; SEP, socioeconomic position.

Figure 3

Results of multinomial adjusted multivariable logistic regression analyses, showing associations between SEP, SEIFA neighbourhood disadvantage and either parent highest education, and ‘unhealthy’ pattern trajectories for the B Cohort (A) and the K Cohort (B). Symbols correspond to ORs and short black horizontal lines correspond to lower and upper limits of 95% CIs. The reference category for SEP and SEIFA neighbourhood disadvantage quintiles is quintile 1, and the reference category for either parent highest education is ‘degree’. Quintile 5 is the most disadvantaged quintile for SEP and SEIFA neighbourhood disadvantage. ORs are relative to the ‘never unhealthy’ trajectory. Associations presented for SEIFA neighbourhood disadvantage are those for the model where SEIFA neighbourhood disadvantage was included with single socioeconomic measures. AU, always unhealthy; B Cohort, Baby Cohort; BU, becoming unhealthy; K Cohort, Kindergarten Cohort; MU, moderately unhealthy; SEIFA, SocioEconomic Indexes for Areas; SEP, socioeconomic position.

In both cohorts, we saw similarly consistent and marked univariable and multivariable associations of both low parental education and being in the lowest socioeconomic position quintile with children following all three of the least healthy and most unhealthy dietary pattern trajectories (online supplementary tables 4–7; figures 2–3). For example, compared with the highest socioeconomic position quintile, children in the lowest socioeconomic position quintile experienced approximately 22 times the odds of following the ‘always unhealthy’ rather than the ‘never unhealthy’ pattern trajectory (OR 22.02; 95% CI 7.66 to 63.29; p<0.001) in the K Cohort.

Discussion

Principal findings

This study shows profound relationships between family socioeconomic status and parental education in infancy or early childhood and dietary trajectories across the entirety of childhood to at least 15 years of age. These findings were consistent across trajectories derived from both a priori and empirical summary dietary measures and across two distinct cohorts. Our composite measure of neighbourhood disadvantage and single measures of remoteness area, parental employment, income, food security and home ownership showed much less robust relationships.

Strengths and weaknesses

In this study, we conducted a secondary analysis on dietary data obtained at multiple waves over a 10-year period of childhood and early adolescence, giving a comprehensive picture of the accumulation and evolution of children’s diets over time. Considering two parallel cohorts of children provides confidence in generalising findings to the broader population.16 We also incorporated a wide range of single and composite socioeconomic measures to help pinpoint potential targets for prediction, prevention and intervention strategies, that is, education and socioeconomic position.

Limitations include the self-report nature of the dietary diary, which is open to inaccurate dietary reporting33 and recall.34 The small number (12–16) of dietary questions at each wave was not previously validated. Moreover, the dietary questions do not cover all of the Australian Dietary Guidelines.24 For example, they do not consider grain (cereal) foods, meat, poultry or fish. Additionally, we did not consider serving sizes but instead relied only on frequency of consumption. Frequency of dietary intake was limited to the previous 24 hours, which may not represent a child’s habitual intake. Inaccuracies in reporting socieoconomic variables could include an unwillingness to correctly report income, possibly explaining the null associations between income and dietary trajectories. This potential limitation will be addressed in future waves once LSAC is linked to lifetime administrative Australian benefits and welfare data. Occupation may be an important codriver of the strong associations of dietary trajectories with socioeconomic position, over and above parental education alone, but we could not assess this. While occupational scores, ranked from lowest to highest status, were used to derive socioeconomic position, these scores were not available individually in the LSAC dataset.28 Despite applying survey weights and achieving a wide range of social circumstances, unequal loss to follow-up confounding baseline under-representation of the most disadvantaged means we have most likely underestimated the effects.

Strengths and weaknesses in relation to other studies

Our comprehensive child and adolescent life course findings are generally consistent with findings in past studies that have considered diet over a maximum of 6 years and three time points. These findings include associations of both higher socioeconomic status and maternal education with healthier diets, of both lower socioeconomic status and maternal education with unhealthier diets1–3 11 12 15 and limited/null associations of household/family income, parental employment and food security with diet,2 11 14 29 30 in studies conducted in Europe, the USA and Brazil. However, some of our null findings contrast with publications arising from Europe regarding home ownership,1 rurality and area deprivation.29 35 36 Possible reasons include previous studies1 29 35 36 using different categories for area of residence or home ownership, examining cross-sectional associations between area of residence and diet, and including a larger number of food groups and greater subset of covariates, for example, maternal smoking, in their analyses.1 29 35 36 Because of very small numbers, we combined ‘remote’ and ‘very remote’ areas with ‘outer regional’ into one category. Combining these categories may have obscured real discrepancies for a very small proportion of Australian children.

Meaning of the study for clinicians and policymakers

This study suggests that children’s diets are much more influenced by family (eg, education and socioeconomic position) than by area level (eg, neighbourhood disadvantage and remoteness area) socioeconomic status. Our findings also suggest that socioeconomic factors related to choices (parental education and occupation) are more important for predicting children’s diets than parental employment and monetary resources. Both parental education and occupation may be important, as associations of dietary trajectories with the composite socioeconomic position measure (which includes occupation) were stronger than those with parent education alone. Specifically, various mechanisms explain how parental education may result in social discrepancies in dietary trajectories. For example, education may enhance nutritional knowledge9 and ability to understand information on and use nutritional labels.37 38 If the associations between parental education and dietary trajectories are causal, we would need to directly tackle educational gradients in order to decrease dietary inequities.

We assume that our extreme effect sizes reflect a cumulative impact of social disadvantage on diet that operates continuously across the first 15 years of life. If steadily poorer health phenotypes simultaneously accrue, then dietary intervention and prevention strategies may need to be continuously implemented during a significant period of childhood and early adolescence to prevent irreversible damage to developing health phenotypes. This need may be urgent for families from lower socioeconomic status backgrounds, in particular those with low levels of parental education.

Unanswered questions and future research

Family socioeconomic and educational status appear to underlie immense discrepancies in cumulative diet across the child and adolescent life course. Immediate unanswered questions relate to causality and mechanisms. Studying the longitudinal associations of environmental factors with dietary trajectories could offer insight into target groups and risk factors for interventions. Most important is to quantify associations between these dietary trajectories and developing health and disease outcomes across the social gradient and determine when in the life course these associations first become evident.

What is already known on this subject

  • Several longitudinal studies have shown that higher socioeconomic status and maternal education are associated with healthier child and adolescent diets, while lower socioeconomic status and maternal education are associated with unhealthier child and adolescent diets.

  • However, these studies are limited by short follow-up, few consecutive dietary measurements and a lack of replication across cohorts.

What this study adds

  • Our study shows that low socioeconomic position and parental education are extremely powerful, consistent and replicable predictors of following the unhealthiest dietary score and pattern trajectories over periods of up to 10 years in two population representative cohorts of children.

  • If causal, reducing dietary inequities may require researching underlying pathways, tackling socioeconomic inequities and targeting health promoting interventions to less educated families.

Supplementary file 4

Acknowledgments

We thank all the parents and children who took part in the Longitudinal Study of Australian Children. This paper uses unit record data from Growing Up in Australia, the Longitudinal Study of Australian Children. The study is conducted in partnership between the Department of Social Services (DSS), the Australian Institute of Family Studies (AIFS) and the Australian Bureau of Statistics (ABS).

References

Footnotes

  • Contributors All authors formulated the research questions. CEG designed the research, with input from FKM, JAK and MW, conducted the research and analysed the data, with supervision from FKM, JAK and MW and wrote the article, with input from FKM, JAK and MW. All authors have read and approved the final manuscript. MW is the guarantor. All authors had full access to all of the data and can take responsibility for the integrity of the data and the accuracy of the data analysis.

  • Funding Authors of this work were supported by the Australian National Health and Medical Research Council (FKM: Early Career Fellowship 1037449 and Career Development Fellowship 1111160; MW: Senior Research Fellowship 1046518); an Australian Government Research Training Program Scholarship (CEG); and Cure Kids New Zealand (MW). Research at the Murdoch Children’s Research Institute is supported by the Victorian Government’s Operational Infrastructure Support Program. The funders had no role in study design, data collection, data analysis and interpretation, writing of the report or the decision to submit the article for publication. LSAC is funded by the Commonwealth Government of Australia.

  • Disclaimer The findings and views reported in this paper are those of the authors and should not be attributed to DSS, AIFS or the ABS.

  • Competing interests All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare: CEG received grants from the Australian Government and Murdoch Children’s Research Institute for the submitted work; FKM receives a Career Development Fellowship from the National Health and Medical Research Council in contribution for her salary; MW receives hourly financial reimbursement from the Australian Government for her advisory role on the Longitudinal Study of Australian Children and has received grants from the Australian Government National Health and Medical Research Council and competitive philanthropic grants; no other relationships or activities that could appear to have influenced the submitted work.

  • Ethics approval Families gave written consent to participate, and the Australian Institute of Family Studies Ethics Committee approved each wave.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement No additional data available.