Estimates of social influence - the "unfriending problem"

In the latest issue (Aug 26) of Science Magazine Barbara Jasny does a nice summary of recent work by Noel and Nyhan:

Studies of social influences on behavior have led to the idea that a range of characteristics from loneliness to obesity might be contagious. A significant problem for the field has been to distinguish effects due to similarities between people (homophily) from social influence. One strategy for doing this has been to look at changes that occur over time. However, such studies have been the subject of considerable debate, and Noel and Nyhan now add a cautionary note. Their analyses of a model used in past social contagion studies suggest that previous investigations have not fully controlled for the possibility that friendship formation and termination are dynamic processes, and friendships between people who are more similar may tend to be more stable over time. Or to put it in Facebook terms, friendships that are between people who are less similar may be less stable, and therefore may result in “unfriending.” Homophily might thus be having a larger effect than appreciated, and under certain conditions could account for most of the contagion effects observed. They conclude that this unfriending problem renders a determination of causality much more complicated in longitudinal social network data.