Correcting Polling Error in Over-Engaged Samples Using Representative Turnout Weighting

Abstract

In previous research (Mellon & Prosser, Missing Non-Voters and Misweighted Samples: Explaining the 2015 Great British Polling Miss) we show that the primary cause of the polling miss at the 2015 UK General Election was the undersampling of less politically engaged respondents, particularly non-voters. We show that this problem can partially be corrected after the fact by including turnout as a weighting factor. In this paper we elaborate and extend this method and show that it could be applied prior to an election as well. We do so by estimating the probability of respondents will vote in the election and calibrate the distribution of turnout probabilities so that the expected level of turnout in the sample matches is pegged to that expected in the election. Using data that was available prior to the 2015 election we demonstrate that had this method been applied to the BES Internet Panel survey it would have substantially reduced the level of error in the survey.