The 2004 Pre-Election and Exit Polls: A Total Survey Error Analysis
Herbert F. Weisberg
The Ohio State University
"The pre-election polls for this election were off more than ever." "This year's
pre-election polls were on target." "Our pre-election poll was the most accurate." These
are the claims that we hear after each presidential election, some people disparaging all
the polls, others defending them, and some pollsters proclaiming theirs the victor. The
problem is that pre-election polls are evaluated on an ad hoc basis, permitting conflicting
evaluations of this sort. Election exit polls pose similar problems. They have generally
been accurate, except for an occasional wrong call of a Senate election, but they were
wrong at the presidential level in 2004. The intention of this paper is to develop
standards for discussing the sources of error in pre-election and exit polls.
The Total Survey Error approach has become a paradigm for understanding the
sources of error in surveys (see Groves 1989; Groves et al. 2004; Weisberg 2005). This
approach has become common in the survey field, though it has received little notice in
political science to date. Figure 1 shows the different types of error, each of which is a
potential problem for pre-election and exit polls.