Measurement and Mismeasurement of the Validity of the Self-Reported Vote

External validity checks of vote self-reports are of particular interest to political scientists. However, most validity checks have used statistics that poorly reflect the behavior they are trying to verify. Earlier researchers have used two measures of the level of agreement between self-reported and validated vote: (1) the proportion of those who claim to have voted who, according to the vote validation, actually did not vote; and (2) the proportion of all respondents who misreported their voting behavior. These measures are substantially affected by the marginal distribution of actual voters and nonvoters in the survey. The measures often will be misleading indicators not only of the propensity of respondents in a given survey to overreport voting but also of differences in this propensity among subgroups of the electorate or across different elections. We examine a measure that does not suffer from this problem. This measure reflects the propensity among members of the appropriate population at risk to overreport voting; it is the proportion of actual nonvoters who claim that they voted. This measure takes into account the fact that virtually all respondents who misreport whether they voted are actually nonvoters. The substantive interpretation can be greatly affected by the measure used. An analysis of SRC 1980 National Election Study data shows that with the new measure, the relation of education to vote misreporting is very different than that shown by the other two measures. Also, contrary to previous research, blacks are only slightly more likely to misreport voting than whites.