by Pete Brodnitz, a principal at the Democratic polling firm Benenson Strategy Group, was pollster for the Ford campaign.
In a year when voters were looking for change, no one offered it, articulated it, and even embodied it better than Harold Ford Jr. Nevertheless, Ford’s campaign for Senate in Tennessee faced several significant challenges, not the least of which was the conventional wisdom that whatever support he had in the polls was overstated.
A widely accepted hypothesis holds that polls routinely overstate support for African American candidates. Proponents of this theory cite four examples, all involving African American Democrats: Bradley’s race for governor of California, Gantt for Senate in North Carolina, Kirk for Senate in Texas, and Wilder for governor in Virginia.
There was no solid rule of thumb we were supposed to apply to account for the so-called “Bradley Effect,” but, in practice, it meant that even if polls showed Ford was even or slightly ahead, many observers would conclude we were actually trailing.
I was not involved in any of those other races, but as a team we attempted to ascertain why this might occur so we could determine if our own polls were accurate or not. One simple question we asked ourselves was whether it was caused by an interaction between the race of the interviewer and the race of the survey respondent. We looked, and in our own polls that was not a factor.
It is also possible that these campaigns for some reason did not appeal to the types of voters who tend to make up their minds late in a campaign. For instance, the Doug Wilder race had a major focus on the issue of abortion rights. This year in both Virginia and Tennessee the final undecided voters tended to be middle-aged to slightly older white women who were married, likely to go to church regularly and were fairly moderate to conservative. Perhaps the late-breaking voters in the Wilder race had a similar composition. If that was the case, the emphasis on the issue of abortion rights could have led many undecided voters to break away from the Democratic candidate.
Solidly Middle Class
Finally, in some focus groups conducted earlier in the year in an affluent northeastern community, we noticed voters had a negative reaction to an African American candidate whose message emphasized his Horatio Alger-like rise from poverty and how the candidate broke down doors by becoming the first African American to achieve various things. It appeared that the reaction resulted from the impression among white voters that the candidate was using race as a qualification for office.
While Harold Ford has a different type of background, it led us to believe that in other contests where this kind of message was used it might have created some distance between the candidates and their largely white, middle class electorates, particularly since this type of story emphasizes the fact that the candidate has a background different from that of most middle class voters.
Based on this experience we looked out to see if we might find similar things in Tennessee. Our focus groups suggested it was also the case in that state. Based on this we concluded that it was important for voters to know about Harold Ford’s solidly middle class background and how his issue positions would impact middle class families. Ford’s message, consistent with his career, was aimed squarely at addressing the concerns of middle class Tennessee residents. And even though others introduced the issue of race into the campaign, Ford did not. As a result, we did not expect that we would see support for Ford disappear between the final poll and the actually balloting, and it did not.
In the final weeks the Ford campaign had to fight against two tides. One was an intense focus on a perceived racial angle to Republican advertising that was at odds with how the public viewed the ads and with Ford’s own focus on other issues. The other was a spate of bad public polls that indicated that Ford was running 10 or 12 points behind his Republican opponent, Bob Corker, when in fact our internal polls indicated that the race was a statistical tie with Ford slightly behind.
In mid-October, the Republican Party ran an ad which, among other things, included a blond woman who said she met Ford at a Playboy party. The ad contained a number of lies and was despicable, but the controversy over the ad was a problem for the Ford campaign. Because it received intense media exposure, awareness of the ad in our late October tracking reached 73%.
While the negative and personal tone of the ad had the potential to harm Corker, when Corker denounced the ad and claimed he was powerless to stop it from airing, the press generally (not totally) appeared to accept his statement. Despite this, almost four in ten voters who knew about the ad did believe he supported its airing.
Corker’s Support for the Attack Ad
Among those aware of the ad:
“As far as you know, did Bob Corker support putting this ad on the air or not?”
|Undecided %||All Voters %|
More problematic was the fact that, while Ford characterized the ad as “trash and smut” because it included a reference to porn movies—and this ad was airing during family viewing hours—the media focused on what analysts called the racist tone of the ad and did not appear to realize that Ford was referring to another element of the ad.
However, unlike the press, the public had not concluded that it was racist. In fact, most were in step with Ford’s conclusion that the problem with the ad was that it was smut.
Assessing the Attack Ad
Among those aware of the ad:
“Which is closest to your view of this ad? . . .” Read and rotate choices
|Undecided %||All Voters %|
|It is a fair criticism of Ford||19||11|
|It is racist||16||5|
|It is not proper for viewing by children||38||48|
Internal Polls vs. External Polls
Once the controversy over the ad died down, the Ford campaign had to deal with an additional challenge: bad public polls.
Our internal polls in Tennessee used samples that were drawn from voter lists. We identified respondents by name from the voter list for each interview. Because we used these voter lists we knew the vote history of our respondents and almost all of the respondents were known to have voted in prior elections. In addition they were screened to ensure they would vote in the upcoming election.
Unfortunately, none of the public polls conducted in Tennessee in the final weeks used voter lists, therefore, these public polls were wildly inaccurate.
Another factor was that, in Tennessee, early voting is widely used. Approximately half the electorate cast their ballot before election day, but as far as we can tell, none of the public polls accounted for this. In fact, even the exit polls conducted on election day in Tennessee failed to account for this and are therefore not a true reflection of the Tennessee electorate.
In some states the difference between using voter lists with vote history and the impact of early voting are minor. Our internal and the external polls were largely in sync in Virginia. In Tennessee they were not and, as a result, Ford spent the last week of the campaign fielding press inquiries about flawed public polls when he could have been talking about the new generation of leadership he intended to bring to the U.S. Senate.
The first problem poll was released by CNN. It was conducted between the 26th and 29th of October. It apparently screened respondents from a random sample. It suggested that Corker had an 8 point lead. Just prior to this poll, both Zogby Interactive and SurveyUSA released polls that indicated that the race was a tie, as did a Rasmussen poll conducted on October 30th.
Right after the CNN survey, however, Zogby released a new poll that claimed Corker had a 10 point lead, even though the other Zogby poll that had reported a tie was conducted over almost exactly the same time frame. Rasmussen followed up with a poll that showed that Ford was down by 8 points—a shift of 6 points in three days that was so inexplicable that Rasmussen went back into the field, which produced a poll that had a 4 point Corker lead. (All three of these Rasmussen polls were fielded in the span of five days.)
During this period the poll that was closest to the final results was a USA Today/Gallup survey that showed a 3 point Corker lead. Of all the polls it was the survey that appeared to use the most care to properly screen respondents to ensure that they were likely to actually vote. However, during the same fielding period, Mason-Dixon conducted a survey that showed Corker with a 50% to 38% lead, leading to stories that drew conclusions like those reported in the last week, which claimed that polls showed “Bob Corker’s lead over Democrat Harold Ford Jr. surging to double digits days before Tuesday’s election.”
In the end, Corker won by a 51% to 48% margin, which meant that, contrary to conventional wisdom, the polls actually had underreported Ford’s support and not overstated it.