This Post-Kaiser-Harvard survey was conducted among a random sample of Virginia residents, age 18 and older. The sample was selected through Random Digit Dialing. (For a more complete explanation of methodology, see www.washingtonpost.com/polls.)
The sample was weighted to match Census estimates of the adult population in terms of age, race, sex, education and region.
As a group those surveyed who said they voted in 2006 closely resemble the National Election Pool exit poll estimates of the Virginia electorate, except in one key respect. The two polls are within sampling error of one another in terms of the racial composition of voters, their relative ideological conservativeness, their party identification and their levels of church attendance.
One glaring difference, however, is the reported vote in the Senate race between Republican incumbent George Allen and his Democratic challenger Jim Webb. Famously, the two evenly split the vote in 2006. But in this poll, 52 percent said they voted for Webb, 41 percent for Allen.
In part, this is yet another shining example of "recall bias;" post-election surveys frequently underreport votes for losing candidates. One example of this was in the 2004 National Election Pool national exit poll, in which Bush "beat" Gore by an 8-point margin. We'll posit a hypothesis that recall bias may be particularly true in instances where clear frontrunners suddenly and spectacularly ("Macaca") lose their leads.
Moreover, we did not set out to recreate the exit poll in Virginia, six months after the fact. We asked the 2006 Senate question not in the expectation to get that number "right," but to gauge how voters who said they voted for Webb differ in terms of their 2008 outlook from those who said they voted for Allen.
View the full report.