The level of education achieved predicted the adoption of not just individual health risk behaviours, but multiple healthy behaviours in this cohort of young Australian adults. The effect was stronger for the participants' levels of education than that of their parents. A change in the level of education from one generation to the next was also significantly associated with a high healthy lifestyle score, with these young adults adopting the lifestyles associated with the level of education that they attained.
At the level of specific health risk behaviours, education was associated with BMI, LTPA, smoking, and some aspects of diet, which is consistent with other studies [12, 28–31]. It is difficult to compare our findings regarding overall lifestyle with other studies because few have reported on multiple risk factors and have generally used measures of occupation rather than education [12, 28, 29, 31]. Parental education was not independently associated with having a high healthy lifestyle score. Although others have reported an independent effect of childhood SES with mortality , the results regarding health behaviours are more mixed. Parental occupation has been reported to be independently associated with smoking [12, 32–34], obesity [2, 34], alcohol consumption [12, 32, 35, 36], and having multiple unhealthy risk factors . However, others have failed to find such an association [2, 12, 30, 32], including our recent analyses concerning changes in physical activity and fitness over time . Our null finding is supported by analyses where adjustment for the current level of education has removed the independent effect of parental occupation on risk behaviours [9, 35]. These results may differ to those of others for several reasons. First, our cohort comprises individuals aged between 26 and 36 years in 2004 to 2006. Most studies have included people born several decades before ours from a different generation, with some exceptions [36, 38, 39]. Changes in socioeconomic conditions between generations may account for the null association between parental education and lifestyle in our study. Second, we were looking at a summary measure of a healthy lifestyle. Parental education was associated with some items from the healthy lifestyle score in the univariable analyses. We suggest that because the effect is only present for some aspects of lifestyle, the contribution of parental education to overall lifestyle declines throughout the life course as children's and parent's education levels differ. This does not appear to be the case for BMI, however, which was associated with parental education in our study and also in other studies [2, 34]. We suggest this is because of a confluence of two factors. First, that parental education is associated with childhood BMI . Second, that obesity tracks strongly between childhood and adulthood in this cohort , thus linking parental education at the age of 12 to adult BMI.
The change in the level of education between parents and offspring was significantly associated with having a high healthy lifestyle score. This association was not the result of an interaction between parental and offspring education levels, i.e. the association between the education level of participants and their healthy lifestyle scores was not modified by the level of their parents' education. Other authors have also reported similar results, albeit mostly in relation to mortality [6, 12, 13, 41]. These analyses show the strong and important role of achieved education in determining lifestyle. The weaker contribution of childhood SES, as indicated by parental education level at age 12 years, would not have been apparent without undertaking the social mobility analyses. The exception to this pattern was in males, where those moving downward were significantly more likely to have a high healthy lifestyle score than those with a stable low education. This may suggest a protective effect in males of higher parental education. However, given the large number of comparisons made we cannot rule out that this was a spurious finding. Together, these results suggest that increasing participation and achievement in education could help reduce socioeconomic inequalities in health.
The greatest absolute differences between low and high participant education or, for mobility, stable high or low groups, were for smoking, meat consumption and, less consistently, BMI. This has implications for the diseases that may show socioeconomic disparities in the future in our cohort. In relation to diseases, the differences for BMI, smoking and meat consumption suggest that cardiovascular disease, diabetes and cancers may occur more often in those with lower educational attainment. Our data suggest that interventions targeted at smoking, weight and meat consumption could have an impact on socioeconomic inequalities in health, at least among the current generation of young Australian adults. The influence of such changes on socioeconomic inequalities in health would vary depending on the strength of the association between a behaviour and disease. Furthermore, it remains uncertain whether campaigns targeted at socioeconomically disadvantaged groups are more effective than population-based campaigns to change health behaviours .
This study has several limitations. The loss to follow-up was considerable but we found that our sample was similar to the general population for several key health behaviours. Furthermore, the inverse probability weighting analyses demonstrated that we most likely underestimated the magnitude of effect by having a cohort with higher SES and more favourable health behaviours in 1985. As noted by others , the applicability of education categories over time should also be considered. For instance, between our parental and participant generations, there was a 130% increase in the completion of secondary school in Australia . Even within the participants' generation, the completion of secondary school rose from 53% in 1987 to 72% in 1997, when the youngest participants finished school . We accept that our groupings may not capture the true extent of educational mobility in Australia, but have used these groupings to align with other studies. We used the highest level of parental education rather than paternal education because other researchers have shown that both paternal and maternal SES are important for the health of offspring . Participants retrospectively recalled parental education. Such recall is moderately accurate over a much longer period than the 14 to 24 years in our study , particularly for parental education . Education is more straightforward to describe and stable over time than occupation , therefore increasing its accurate recall. Although using a younger cohort has probably limited reverse causation between lifestyle and education, we cannot discount that this may have occurred. Some researchers have shown that adolescents who take up smoking have worse educational outcomes suggesting that even in younger people lifestyle may affect SES .
While we believe the healthy lifestyle score has strengths, it also has limitations. We acknowledge that each item in the score does not contribute equally to the burden of disease. While it would be possible to derive weights for items based on their contribution to disease, this would make the score less accessible to the general population. Further, the association between the healthy lifestyle score and cardiovascular risk factors in this cohort  and its prediction of mortality [10, 49] demonstrate the validity of the current scoring method. We and others have dichotomised the score for analysis purposes. This is unlikely to be suitable for clinical or public health settings where a focus on improving all health behaviours, and achieving a high healthy lifestyle score, is desirable. Finally, our analyses of the healthy lifestyle score did not specifically take into account that some of the component behaviours more often co-occur than others. Such analyses were beyond the scope of this paper given that over 400 different combinations of the 10 healthy items were present in this cohort.
The study also has several strengths. Despite the loss to follow-up the sample was large and had considerable heterogeneity of exposures and outcomes. The use of the healthy lifestyle score adds strength because this is the only study of this kind to have used a composite measure of lifestyles that predicts mortality. The examination of individual health behaviours and overall lifestyle within a single cohort also makes this study novel. Few studies of socioeconomic inequalities in health have been conducted in Australia, particularly among men.