Vote Miscount or Poll Response Bias: What Causes Discrepancies between Exit Poll and Reported Vote Shares?

Abstract

Is vote miscount detectable in the United States? United States' citizens lack any federal right to publicly access records necessary to judge the accuracy of electoral outcomes. Most U.S. states also lack such a right; and tally votes using trade secret software without conducting publicly verifiable manual post-election audits of election outcomes. In addition to this lack of public verifiability of United States' election results, there is a dearth of detailed election data reporting in the United States. Thus, a limited number of statistical methods are available for detecting patterns consistent with vote miscount. This paper develops a statistical method using maximum likelihood estimation to determine the approximate proportions of vote miscount and exit poll response bias consistent with precinct-level patterns of discrepancies between exit poll and election results. Simulated data shows the method works well. This statistical method can be used to detect possible vote miscount by anyone, such as exit pollsters, having access to both precinct-level exit poll and election results data. Applying this within precinct discrepancy analysis to the pattern of discrepancies in the 2004 Ohio U.S. Presidential election show a consistency with vote miscount sufficient to have altered the outcome of the 2004 presidential election contest.