By its very nature, election fraud—which we define as efforts to use illegal means to alter election outcomes—should be difficult to detect. After all, given that documented evidence of election fraud is highly likely to result in legal action and criminal penalties, agents committing fraud have very strong incentives to cover their tracks. Perhaps the reason we do not observe widespread empirical evidence of election fraud in the United States is that the perpetrators cover their tracks sufficiently to make it difficult to observe and prosecute.
\nBut decades of theoretical and empirical social science research have shown that elections in the United States are highly predictable.1 Based on a handful of variables, including the state of the national economy, incumbency, and partisanship, we can predict most election outcomes with a high degree of accuracy well before the election occurs. This large body of social science research has two important implications for the study of election fraud. One implication is that because election outcomes are so highly predictable, perhaps election fraud is not widespread in the United States. But the other implication is that unpredictable election outcomes might signal the occurrence of election fraud.
\nAn excellent opportunity to use the theoretical and statistical tools of social science to study election fraud arose in the 2002 general election in Georgia. There, two incumbent Democrats lost their seats in closely contested races: U.S. Senator Max Cleland lost to Republican Saxby Chambliss (45.9 percent to 52.8 percent), while Governor Roy Barnes lost to Republican Sonny Perdue (46.3 percent to 51.4 percent). Not only were the two races close and won by Republicans, but this election was the first large-scale trial of electronic “touch-screen” voting machines in Georgia, a state that in the wake of the 2000 election dispute replaced old punch-card and optical scanning voting devices with the same touch-screen voting system in every precinct.
\nAllegations that these two races might have been marred by manipulation of the new touch-screen voting system have swirled in the wake of a study of what is reputed to be some of the source code that is used to run certain operations of Georgia’s touch-screen voting devices.2 And the controversy only intensified when the chairman and chief executive of Diebold, the corporation that produced the voting machines used in Georgia, was quoted in a Republican fundraising letter as saying that he was “committed to helping Ohio deliver its electoral votes to the president next year.”3 But was there actually a priori evidence of election fraud in Georgia? In this chapter, we seek to answer this question by combining the statistical approach of outlier detection with the wealth of theoretical and empirical research in political science on elections.
", "isbn": "9780815701385", "publisher": "The Brookings Institution", "place_of_publication": "Washington D.C.", "publication_date": "2008", "pages": "149-161" } ]