Multidimensional Profiles of Smokers Social Networks Utilizing Mixtures of Multivariate t-Distributions

Smoking remains the leading preventable cause of death in the United States. An increasing body of literature has highlighted the role that social relationships play in our health (Cohen, 2004). In particular, a wide range of health behaviors such as tobacco use are developed and reinforced in the context of a person’s social network, and social influence has been found to be an important factor in determining a person’s ability to quit smoking (Christakis & Fowler, 2008). However, little research has been devoted to defining the social network and social support systems of smokers who have successfully quit and maintained abstinence from smoking. In addition, an individual’s social networks influence on smoking behavior has largely been examined utilizing unidimensional features of the social network. The social environment is complex; therefore, it may be more informative to examine social influences utilizing quantitative methods that capture this multidimensional interplay.

Using a large sample of smokers (N = 1592, 53% female, 84% Caucasian) who volunteered to take part in a smoking cessation intervention, responses to 12 “indicator” items collected from a social network interview were analyzed utilizing finite mixture modeling (FMM). In order to prevent biased parameter estimates and the overextraction of mixture components from non-normal data, mixtures of multivariate t-distributions were used as a robust alternative to the traditional Gaussian mixture model (e.g. Model-Based Clustering, Latent Class Cluster Analysis, MCLUST). Six latent clusters were identified and revealed an underlying structure composed of significantly different social network sub-groups within the sample of smokers. Latent clusters were found to vary systematically with individual difference factors relevant to smoking behavior and dependance (e.g., gender, ethnicity, education).   Moreover, specific clusters were found to be predictive of smoking cessation success 6 months and 1 year post-treatment, highlighing the importance of social networks for understanding the dynamics of smoking behavior. FMM utilizing mixtures of multivariate t-distributions is a unique and robust alternative to traditional mixture modeling techniques commonly used in the behavioral sciences. The novel use of distributions robust to non-normal (t-distributions) data is a powerful way to examine the complex interactional nature of an individual’s social network and may serve as a useful tool for tobacco interventionists in order to target the social context of the smoker.