This study has shown that social relationships were independently associated with a range of health behaviors in a national sample of older US adults, even after adjusting for the effect of broad demographic and socioeconomic determinants. In particular, being widowed or divorced/separated was significantly associated with being a current smoker, and amongst older men, with being a heavy drinker. Single older women were also more likely to be heavy drinkers. Widowed older people were less likely to attend a dentist. In addition, older people with more close friends more commonly engaged in physical activity. Furthermore, the availability of emotional support did not mediate any of the above associations, but emotional support was associated with physical activity but none of the other behaviors assessed. Access to financial support through social relationships was not related to any of the examined health behaviors.
An extensive body of research has demonstrated the effect of social relationships on health and mortality [43, 44]. One of the potential pathways linking social relationships to health is via behavioral factors. However, very few studies have investigated the association between social relationships and health behaviors amongst older people. Frongillo et al.  showed that older people with strong social ties had better dietary behavior . In a population of older black Americans, higher levels of social support were associated with attendance at cancer screening services . We are not aware of any other studies that have assessed a combination of health behaviors and their association with both measures of social support and social networks in a sample of older people.
Our findings on the importance of marital status on health behaviors are in accordance with previous reports amongst younger and middle aged adults linking marital dissolution with health compromising behaviors. Longitudinal studies have shown that marital termination, either due to the death of a spouse or divorce, were associated with tobacco and alcohol consumption [45–47]. It is likely that changes in both social support and stress levels are key factors in explaining this link . Smoking and excessive drinking are both related to high levels of stress [48, 49] and low levels of social support . Indeed support from a partner may act as a buffer against the harmful effects of stress, and thereby lead to reductions in tobacco and alcohol use .
Previous studies have shown that amongst adults, poor social relationships  and with older people, limited social networks  were both associated with low levels of physical activity. Our results support these findings and highlight the importance of social relationships on physical activity. Older individuals with well-developed and supportive social relationships may be encouraged by their peers and family to adopt and maintain physical activity and may have more information to access local services and amenities.
The associations between social relationships and the three behaviors physical exercise, dental visits and smoking were attenuated after adjusting for socio-economic status, and in the case of dental visits the association was no longer statistically significant. This suggests that those who were poorer and less educated also tended to have smaller friendship networks and were less likely to live with a partner. As expected, there was no attenuation of the results for heavy drinking, which showed a reverse social gradient, i.e. was more common among the more advantaged. Although we observed social gradients in relation to physical exercise, dental visits and smoking, the availability of financial support was not related to any of the examined behavioral outcomes. It is possible that financial support from relatives and friends is given only in the event of an emergency, and is not enough to alleviate the effects of poverty and limited income on patterns of health behaviors.
This study used a large nationally representative sample of US older people, included a range of both health promoting and health damaging behaviors and the detailed analysis controlled for a diverse spectrum of potential confounders. However it is important to acknowledge the limitations of this study. Social support and social networks were only partially assessed due to lack of relevant data, i.e. validated instruments, in NHANES. We only assessed perceived emotional support and did not include any measures of informational and appraisal dimensions of social support. The assessment of social networks covered the size of the networks but did not assess the intensity or quality of social contacts . We also acknowledge that the social network and social support variables could have been categorised in various different ways. For example, it is possible that people with no close friends are very different from people with four close friends. However, having zero friends was very rare in our sample (reported by only 3.7% of all participants). As there is no other cut-off point for this variable that could be conceptually justified, dividing the sample into tertiles was an alternative. Sensitivity analyses using “number of friends” as a continuous variable produced consistent results, i.e. a modest statistically significant association with physical activity but not with the other behaviors. Further to this, we also tested whether dividing the variable “lack of emotional support” into three categories (having someone to provide support plus having received enough support; having someone to provide support but could have used more support; and having no one to provide support) would influence the results, which was not the case. As for the binary variable, the only significant association was with physical activity, with very similar prevalence ratios for the categories “not enough support” and “no support”.
The behavioral outcomes were all assessed through self-report measures which are subject to bias and under reporting. Physical activity in particular is notoriously difficult to assess accurately. NHANES uses its own physical activity questionnaire (PAQ), which although containing an extensive array of questions, cannot be considered to be a validated measure. Our study assessed dental visits as an indicator of health-service use but did not examine the use of other health services, for which associations might be different. Both the social relationship measures and behavioral outcomes were recorded cross-sectionally and did not assess the dynamic nature of these constructs . In the analysis although we adjusted for a range of important covariates, residual confounding may still be an issue. Finally, the cross sectional study design limits any consideration of causal relationships. Indeed it is possible that the association between social relationships and behaviors is due to reverse causation – people who engage in health compromising behaviors may not be able to establish long term stable relationships and older people who smoke and drink heavily might become socially isolated because these behaviors are considered socially undesirable .