Talk Title: Never Send a Statistician to do a Sociologist's Job
Abstract: Practicing social analysts usually learn their statistics from statisticians who specialize in figuring out how to estimate parameters from correct models. But in practice, we never know the true model, and we don’t get a lot of guidance on how to use statistics to learn true things about the world. Many of the techniques that we imagine are best create more problems than they solve. We need to stop using statistics as if it were a magic spell that could answer impossible questions, and instead use data to rule out wrong interpretations of robust patterns. Instead of concentrating on the niceties of our models and estimation techniques, we need to have a sense of what our data are (what interactions produced them?), where they are in a space of possibilities (are most people near the lowest end? near the highest?), and, most important, how larger structures in which our data might be embedded—such as time, space, and social structure—easily lead us to predictable, false, conclusions.