What We Say Versus What We Mean

I'm cribbing from myself on this post. I wrote a comment at Article 19 about Don's response to a post by Kevin Drum, which in turn was commenting on a piece in today's Wall Street Journal. The Journal was taking a look at the racial overtones of Tennessee's senatorial race for Majority Leader Bill Frist's seat. If Dem Harold Ford, Jr., wins, he'll be the first African-American senator from the South since Reconstruction.

The specific point that spurred all this discussion was the phenomenon of African-Americans doing better in pre-election polls than they do in the final vote count. The first time that I'm aware this situation was identified was in the 1982 California governor's race. In pre-election polls, Tom Bradley, the African-American mayor of LA, had a nice lead over state Attorney General George Deukmejian (numbers are hard to come by 24 years later, but the final Field poll gave him an 8-point lead, 54-46), but somewhere between the poll and the ballot box, enough people changed their minds that Bradley lost 51-49. (Interestingly, that same year the final Field poll for the California senate race between two white guys nailed the final numbers exactly.) No one had an explanation for Bradley's loss except that voters were claiming to support him while voting against him.

In Douglas Wilder's 1989 Virginia gubernatorial race, pre-election polls put him up by an average of 7 points, yet the final vote was close enough that, although he won, it was only confirmed after a recount. That same year, David Dinkins, predicted to beat Rudy Giuliani for New York mayor by 14 points ended up winning by only 2. Harvey Gantt, who ran against Jesse Helms for a North Carolina senate seat in 1990, is sometimes cited as another example of the phenomenon, but I'd like to see the specific numbers on that one. Gantt was leading at some points of the campaign, but Helms pulled out a raft of racist ads and clawed his way back into the race. I'd be curious to see whether the final polls reflected his comeback and predicted a close race or if they still gave it to Gantt.

One potential explanation for the discrepancy was offered by Time magazine in the aftermath of the Wilder and Dinkins contests. One pollster looked at some of his numbers from the Wilder race in an unexpected way and discovered that when white respondents were questioned by what they believed were white interviewers, they would break toward Wilder's white opponent by 16 points. If the interviewers were perceived to be African-American, they supported Wilder by 10.

Of course, this was seventeen years ago. Have we changed during the interim, or does the same situation still apply? The Ford-Corker race might give us some clue, but Corker and the Repubs are digging deep into their past and pulling out some racist tricks so that they win the old-fashioned Southern strategy way. Time ended its article by suggesting, "Pollsters in black-white contests should learn to take the discrepancy into account -- at least until such racial match-ups cease to be novelties." Maybe the novelty hasn't warn off yet.