Is there systematic bias in exit polling with respect to Senator Barack Obama? Allegations and theories have emerged in the “blogosphere” and elsewhere that Senator Obama’s results in exit poll data are exaggerated, and don’t accurately reflect final, “real”, election returns. I compare election returns and election poll data in 27 states that have had Democratic primaries as of April 23, 2008, examining data for both Senator Barack Obama and Senator Hillary Clinton. In neither case was there evidence of large discrepancies between exit poll results and final primary vote counts. There was, however, some variance across states, and from those variances a pattern may have emerged, which I will discuss later.
The argument
Critics of exit polling doubt the accuracy of respondences when candidates who are African-American or other racial or ethnic minorities are in the race. Poll respondents are said to not wish to declare their vote against these candidates, fearing the appearance of bigotry to exit poll workers. This in turn leads to inflated results for minority candidates, making exit polls less reliable for analytical and predictive purposes. There is also some evidence to indicate that this effect is slowly disappearing.
The pollster John Zogby sees some signs that white voters have grown more comfortable with black candidates. He offers the example of Harold Ford, the young, black Democratic congressman who narrowly lost his bid for one of Tennessee's US Senate seats in 2006. Traditionally, Zogby points out, black candidates do worse on Election Day than in pre-election polling because people tell pollsters they're more comfortable with black candidates than they actually are -- this phenomenon, the so-called Bradley Effect, is what some analysts thought helped Clinton last month in New Hampshire. But, Zogby points out, Ford actually did better in the final vote than in pre-election polling, suggesting a dissipation of the Bradley Effect.
Granted, the above quote refers to pre-election polling rather than exit polls, but the effect is similar.
The raw data
In general, exit polls have been remarkably accurate in predicting and describing voter behavior. For both candidates, the average difference between exit poll projections and eventual vote tallies has been less than one-half of one percent.
...
