Despite the good match between model predictions and data, a number of limitations remain. We have made
an implicit mean-eld" approximation in taking social utility to be a function of the overall smoking prevalence x,
rather than the local smoking prevalence among contacts in an individual's social network. Similarly, we have taken
individual utility to be uniform across the population (though not in time), whereas a more detailed model might
allow for individual variation. As a mild justication for these assumptions, we point out that analysis of a similar
model in another context [19] suggests that inclusion of more detail will not change qualitative predictions.