ABSTRACT
This chapter discusses the methodology of scientific public opinion polling. It discusses how a seemingly small sample of, say, 1,500 respondents, can describe the views of the larger group that it purports to represent. A random sample is the gold standard. But polling is complicated. For example, often it is required to first randomly sample only within randomly determined geographic clusters. The problem of nonresponse requires that pollsters post-stratify—or weight respondents based on their group characteristics so that different demographic categories are properly represented. Two instances of bad polling in presidential campaigns—the notorious Literary Digest poll of 1936 and Gallup Poll of 1948 are discussed as case studies in what not to do. The chapter discusses the transition from in-person to telephone polling in the late twentieth century and the more recent introduction of robotic telephone polls and Internet polling. The chapter shows that late campaign polls generally are good predictors of presidential elections. The 2016 predictions of a Hillary Clinton victory did get the popular vote winner right and were not far off in terms of the winning margin.
