What's wrong with social epidemiology, and how can we make it better?