Vote Miscount or Poll Response Bias? Estimating the Causes of Discrepancies between Polls and Reported Vote Shares
50 Pages Posted: 4 Nov 2012 Last revised: 11 Jan 2017
Date Written: October 12, 2013
Is vote miscount detectable in the United States? In the United States, there is currently, no federal right to publicly access records needed to evaluate electoral outcome accuracy. Most U.S. states also individually, lack such a right and tally votes using trade secret software without publicly reporting detailed election data by vote type and location and without conducting publicly verifiable post-election manual audits of election outcome accuracy. Thus, only statistical methods are available for detecting patterns consistent with outcome-altering vote miscount. This paper develops and tests a statistical method for estimating systemic vote miscount and partisan exit poll response bias, which produces distinct patterns of discrepancies between exit poll and election results. Simulations show the method recovers direction and magnitude of systemic vote shift and relative response bias parameters; and could be used by pollsters with access to unadjusted exit poll and election results data to detect systemic vote miscount. Statistical analyses of the 2016 U.S. presidential election vote--poll discrepancies are not consistent with systemic vote miscount occurring in a majority of polled states, but reveal significant discrepancies in Pennsylvania and North Carolina for vote miscount to have potentially altered the 2016 U.S. presidential election outcome, even after adjusting for estimated systematic partisan exit poll response bias and false discovery rate.
Keywords: 2004 Ohio presidential election results, exit poll response bias, election results, vote miscount, response bias, discrepancies between polls and election results
Suggested Citation: Suggested Citation