Exit polls are excellent sources of information to understand voter behavior and to predict election outcomes. This paper presents the current methods of exit polls, which is conducted primarily by the Mainichi Newspapers, and discusses issues related to exit polls.
Exit polls are based on a two-stage sampling design, with a polling precinct as the primary sampling unit and a voter as the final sampling unit. Precincts are stratified and selected systematically with probability proportional to the precincts' forecasted number of voters. The voters of a precinct are first selected systematically, with the interval predetermined based on the forecasted numbers of voters, and subsequently, asked for interviews.
The datasets of exit polls that have been collected during 23 governor elections held since April, 2003 are analyzed. Standard errors of the estimated proportions of votes for 90 candidates are, on average, 1.5 times larger than the respective standard errors based on the simple random sampling.
The analysis also reveals that the estimated proportions of votes for 20% of the candidates are outside the 95% confidence intervals, which suggests that there exist some sources of errors apart from the sampling error. The issues related to exit polls, including those concerning errors caused by an increasing number of early voters and non-responses, are discussed.