MAKING INFERENCES ABOUT ELECTIONS AND PUBLIC OPINION USING INCIDENTALLY COLLECTED DATA

Abstract

This chapter discusses the use of large quantities of incidentally collected data (ICD) to make inferences about elections and public opinion. ICD is data that was created or collected primarily for a purpose other than analysis. Several uses of ICD are especially relevant to the study of elections and public opinion: making point estimates, election forecasting, and estimating causal relationships. Another major use of ICD is election forecasting. Many papers have used data on the number of times different candidates are mentioned on Twitter to forecast elections, with the assumption that more Twitter mentions is associated with a higher vote share. Another form of inference that is sometimes used with ICD is to argue that even though the sample is unrepresentative, the social mechanisms that the authors are testing are not likely to be affected by the sample’s un-representativeness. This is essentially the same logic that governs external validity in laboratory experiments.