More Right More Often: An Argument for Statistical Sophistication in Political Polling Open Access
Downloadable ContentDownload PDF
More political polls are being conducted today than ever before. Not quite a century old, political polling has developed slowly, tending to lag behind the best practices of the survey research world at large. It took decades, but political pollsters have caught up to those practices, at least with respect to survey design. Survey analysis is another matter.When it comes to predicting election outcomes, the state-of-the-art in political polling analysis today is the same as it was when the Literary Digest conducted its first poll in 1916, the ubiquitous crosstab. For a variety of reasons, this method should not be regarded as sufficient or satisfactory, at least on its own, but one reason alone is sufficient to attempt more sophisticated statistical techniques: the current method is not perfect. That is, it does not perfectly predict the voting behavior of the electorate. The objective of this thesis is to evaluate an alternative method, logistic regression, to determine whether it can yield more accurate predictions and, if so, to assess its usefulness given the jobs pollsters perform in political campaigns, research firms, and the media.