Christakis and Fowler published an analysis of social network effects on smoking behavior (paper can be found here). It cought my attention as my own interests right now orbit around selection and influence effects in social networks. In the smoking case we see frequently that smokers and non-smokers tend to be socially linked among each other than across, expressing homophily. The question is then whether that happens because of influence (e.g. smokers are linked to smokers because people start to smoke because their friends smoke) or selection (e.g. smokers are linked to smokers because non-smokers cannot stand the smoke at sever the social ties). I was hoping to find answers to these questions in the paper.

The analysis is based on a data from Farmingham Heart Study. The data seem to impressively rich: panel capturing almost everybody in the Farmingham community and monitored roughly every second year from 1973 up to 2003 (!!!). The authors are mostly concerned with quitting to smoke, not so much about starting to smoke, but also it is quitting that they mostly observe. Results show that there seem to be strong influence effects in that respect: spouse quitting to smoke increase the likelihood of respondent’s quitting by 67%, friend by 36%, coworker by 34%, and sibling by 25%. Great analysis, especially because it is based on tracing one cohort for over 30 years. Their analysis even made the news in New York Times .

Some comments:

  1. When I saw this paper I was immediately interested whether the results would be similar to other studies of the effects of social networks on smoking (or other substance use). I was thinking about for example Pearson&Michell (2000) and Pearson et al. (2006) in Drugs: Education, prevention and policy, Pearson&West (2003) in Connections. To my surprise they do not refer to their results.

  2. The analysis seem to focus only on influence through social ties, although the authors mention the selection process as a possible alternative explanation, they call it “induction”. However I do not find it fully addressed in the analysis. Authors write that smokers are pushed to the periphery of the network, but do they find different network alters who are smoking as well? Steglich, Snijders and Pearson in their paper on SIENA models (available on SIENA’s website) analyze the data from school classes and do not find any selection effects, but strong influence effects. It would be extremely interesting whether similar mechanisms would govern quitting to smoke, and also among adults.