DISCRIMINANT-ANALYSIS AS A PREDICTION TOOL FOR UNCOMMITTED VOTERS IN PREELECTION POLLS

Uncommitted voters in pre-election polls include both those who have n ot yet made their choices (undecided voters) and those who have alread y made their choices but do not disclose them, for any reason (undecla red voters). The voting patterns of these uncommitted voters are criti cal, especially in cases in which the size of the group is greater tha n the gap between the two leading runners in an electoral race. In the present study, discriminant analysis was applied to predict the votin g patterns of the uncommitted voters in two different election situati ons: the 1999 presidential election in the state of North Carolina, US A, and the 1992 presidential election in South Korea. The adjusted per centage breakdown, which was created after discriminant analysis assig ned the uncommitted voters for each candidate, appeared noticeably clo ser to the actual results than the simple percentage breakdown with th e uncommitted voters excluded from the analysis. Discriminant analysis was also applied to predict the effect of one candidate's possible wi thdrawal from elections with several candidates. By treating the forme r supporters of the withdrawing candidate as another uncommitted group , discriminant analysis provided a new percentage breakdown among the remaining candidates; the result was consistent with experts' intuitiv e prediction.