The results of this study show that the probability of engaging in regular exercise or eating a healthy diet is higher when individuals have friends who also engage in these behaviours. Furthermore, the stronger the relationship between the ego and the alter is, the greater the probability that the ego will behave in a similar way to the alter. These findings are consistent with the broad pattern observed in the social network literature, by which the behaviours observed for connected individuals are persistently correlated. However, the analyses presented above show that the degree to which an ego’s alters are similar to the ego, in terms of being the same gender, or from the same migration background, does not affect the association between the health behaviours of the ego and alters. It was also shown that being a relative of the ego did not affect the association between the ego and alter health behaviours under focus. Nevertheless, when a pooled analysis for males and females was conducted, the results observed were in line with previous findings. These analyses showed that gender homogeneity in the ego-alter relationship showed a statistically significant increase in the probability of the ego and alter behaving similarly in terms of exercising regularly. Given that past studies have also observed that the ego-alter relationship type does affect the degree of association between the health behaviours of alters and the ego, such as smoking and alcohol consumption
[10, 37], it is possible that the lack of statistical significance in the results for migration status homogeneity and relative-status could therefore be due to a relative lack of power in the statistical analyses.
This study has not been able to identify causality both due to the cross-sectional and observational nature of the data employed, as well as more fundamental problems concerning the identification of peer influence in social network research. Previous research has shown that this causal relationship is not identifiable in the absence of full information concerning both the choice of friends and the proclivity to engage in a given health behaviour
[20, 22]. While the data used for this study is cross-sectional, these problems of causal identifiability are equally prevalent when analyzing longitudinal datasets, and the use of lagged measures is not an adequate solution
. However, while the challenges of identifying causal relationships in social networks from observational data are still being addressed, experiments strongly support the conclusion that peer influence is a real phenomenon
[17–19]. Furthermore, sensitivity analyses indicate that the results of studies showing induction of obesity and smoking behaviour (
[10, 15], for example) would have to be very seriously confounded to explain away the strongest of the findings
While we are unable to identify causality, we do assume that the exercise and eating behaviour that we investigate in this study have a strong potential for peer transmission because they are visible, and because both exercising and eating are common social activities
. Furthermore, the results that we have observed are consistent with the large and growing body of research showing that the behaviours of individuals are associated with the health behaviours of their peers. We have attempted to adjust our estimates for a number of different factors that are likely to influence both the propensity to form friendships as well as to influence the likelihood of engaging in exercise and healthy eating, but there are undoubtedly unobserved factors that we have not been able to adjust for. To be very specific, we have not been able to distinguish between peer influence and assortativity in this study. Even when considered outside of more formal problems of identification
, it is very difficult to investigate how, for example, exercise behaviour patterns are influenced by peer behaviour, as organized sporting activity, and particularly team sport, serves as a focal point for socialization for many individuals in this age group
. Individuals who play team sports engage in exercise themselves, and are more likely than not to also be friends with their teammates. Although it is not possible for us to distinguish between the extent to which the associations that we have observed are due to homophily or an induction effect, they do nevertheless indicate that patterns of eating and exercising are socially clustered. Since these behaviours are socially clustered, individuals who eat a poor diet or exercise infrequently are less likely to adopt a new behaviour, because they are less likely to be connected to individuals who do live more healthy lifestyles
. These results points towards potential policy-interventions being targeted towards groups of individuals rather than trying to shape individual behaviour through individualized incentives.
By using egocentric data for this study, we have relied upon the accuracy of the ego in terms of reporting both their own behaviours as well as the behaviours of the alters that they have named. This will obviously lead to a certain degree of measurement error, and this measurement error is likely to be greater when the reported strength of the relationship between the ego and alter is weaker; a weaker relationship is likely to be characterized by, relatively speaking, less intimacy, spending less time together, and having less knowledge about one another, meaning a greater likelihood of lower precision in terms of reporting alter behaviour. Furthermore, by limiting the number of alters that the ego could potentially report to five, the survey procedure will necessarily have restricted the scope of the ego’s full social network that could be captured in the data. As a counterbalance to this limitation, it is probably more likely that the reports on alter behaviour are relatively more accurate than if the ego had had the opportunity to name a greater number of alters, as it is more likely that they would name individuals with whom they had a closer relationship. In addition, past findings suggest that what is important for predicting the association between the ego and alters behaviours is the ego’s perception of the strength of the ego-alter relationship
However, despite these factors, this study contributes strongly to what remains a small, though growing, literature concerning the clustering of health behaviours in social networks. In particular, despite research existing on the relationship between social support and exercise adherence, there is only a small body of research concerning the association between peer exercise behaviour and ego exercise behaviour using actual social network data
[26–28]. The situation is similar for research concerning eating behaviour, with only a small number of studies addressing this question
[26, 27, 32, 42]. Most previous research has been conducted using data from the United States, though a number of these studies have also used data collected in Australia. This study, using a cohort of nineteen year-olds in Sweden, indicates that the patterns of association between ego and alter behaviours that have been observed in samples of adults
[10, 15, 32], as well as samples of young adolescents
[26, 28], are also present amongst young adults. Furthermore, this study is the first to address this question using data from the Nordic region. It is important to note that this pattern prevails even in a country where the obesity crisis is at a relatively less severe stage of development.
As has been pointed out before, it is very possible that changes in attitudes towards obesity that are influenced by peers could translate into actual changes in weight through different behaviours
[15, 43]. For example, if a peer abandons smoking, it might not only influence the ego to abandon smoking his or herself, but may also induce a change in attitudes that leads them to adopt a more healthy diet. Future research that addresses parallel changes in a number of different health behaviours in a consistent direction over time would begin to clarify this question. It is important to note that observational data fundamentally restricts the degree to which it is possible to parse out the relative importance of selection effects, exogenous factors, and induction for the transmission of social behaviours
[43, 44], even with longitudinal data
. However, carefully designed experiments have shown that this task is still possible
[18, 19]. Almost all modern societies face a growing public health dilemma as individuals exercise less and consume a less healthy diet, leading to an increase in the incidence rates of diseases of the circulatory system and cancers. With just cause, concern about the growing number of individuals who are overweight or obese has increased dramatically. While many of these changes are related to shifts in the type of labour in demand due to broader changes to the macroeconomy, and changes to patterns of regular consumption, social influence on health behaviours and changing weight norms are clearly highly salient factors. To develop a greater understanding of the mechanisms by which obesity is spread through a social network, it will be important for future research to continue to go beyond documenting correlations between peers in terms of changes to BMI, and to attempt to elucidate this pattern by investigating the diffusion of actual health behaviours.