BUILDING A PROBABLE ELECTORATE FROM PREELECTION POLLS

Prominent among sources of nonsampling error in preelection polls are different methods of screening likely voters. Building on past research, we use National Election Studies data to construct a simple, practical, two-stage model of voter turnout. We compare the performance of this model to alternative estimators, using data from the 1988 and 1992 presidential elections. By allowing for truncated interviews with up to 20 percent of respondents, our two-stage model provides turnout estimates that are slightly more accurate and significantly less expensive than traditional methods for screening likely voters.