ABSTRACT
This chapter discusses ways — conceptional and pragmatic — to apportion undecided respondents in pre-election polls. Pre-election polls are used to measure and analyze strength of support for a candidate, the depth of that support, the differences among candidates and the mood of the electorate in general. National poll respondents were screened for registered voters, then a single, self-professed likelihood-to-vote question was used to determine likely voters. The statewide poll in both the 1988 and 1992 election had more percentage points of error for all allocation techniques than the nationwide poll. The development of discriminant functions using more than just demographic variables seems quite promising. In the 1992 election all allocation methods resulted in a larger estimate of error for both state and national polls and this election was characterized by a high degree of voter mistrust in government.
