The defeat of Rep. Allen West (R., Fla.) means that, come January, the United States Congress will have just one black Republican member, Rep.
Tim Scott of South Carolina. West's loss came as a surprise to a number of people, especially considering that
every public poll of his race had shown the freshman congressman leading his Democratic opponent, Patrick Murphy. Yet, on Tuesday morning,
it was West who conceded to Murphy after a two-week recount confirmed that the incumbent had fallen more than 2,000 votes short of winning a second term.
The race in West’s congressional district wasn't the only one where the outcome defied multiple polls: In Utah's newly-created 4th District, Rep.
Jim Matheson, the lone Democrat in the Beehive State's congressional delegation, eked out a victory over Saratoga Springs Mayor
Mia Love (R). Less than a week before Election Day,
a Mason-Dixon poll showed the six-term incumbent trailing Love by an eye-popping twelve percentage points. About a month earlier,
the Deseret News/KSL poll pegged Love's lead at six points, 49% to 43%, but that was a huge shift from
their previous poll of the race, conducted in June, which had Matheson up by a healthy fifteen-point margin, 53% to 38%. Despite this apparent momentum toward Love and a huge voter turnout in a state that went for Mitt Romney by a wider margin than any other, Matheson managed to win with 49% of the vote to 48% for his Republican challenger.
The surprising losses of these two prominent black candidates--and the poll numbers that turned out to be so wrong--bring to mind the so-called "Bradley Effect," a phenomenon in which polls of races with one African-American candidate tend to overestimate his/her support or underestimate the support for his/her opponent, or both. The term arose out of another surprising election result, in the 1982 race for governor of California. Los Angeles Mayor Tom Bradley (D) was poised to become the Golden State's first African-American governor. He went into Election Day with a healthy lead over his Republican opponent, then-Attorney General George Deukmejian, in all the polls. Even the exit poll results indicated a big Democratic win. Much to the Bradley campaign's chagrin, however, Deukmejian triumphed, winning the election by less than 100,000 votes out of nearly 8 million cast.
The following year, in the race for mayor of Chicago,
Harold Washington, an African-American congressman from Illinois's 1st congressional district, defeated his caucasian opponent, former State Rep. Bernard Epton (R), by about 47,500 votes (less than four percentage points). Multiple polls conducted in the run-up to Election Day indicated it would be a rout for Washington. Three separate polls taken in the final weeks of the campaign showed Washington with a fourteen-point lead.
The obvious discrepancies between the pre-election polling and the actual results of these two high-profile races seemed to cry out for an explanation. Arguably, racial tensions between blacks and whites were stronger, or at least more overt, 30 years ago than they are today, so naturally race was floated as a possible factor in Bradley's loss and Washington's unexpectedly narrow win. (
Veteran California pollster Mervin Field was quoted two days after the 1982 election as saying that "race was a factor in the Bradley loss.") While it was clear that these elections had turned out much differently than the polls had foretold, the notion that races between a black candidate and a white candidate presented a special problem for pollsters did not gain widespread acceptance until after a couple of elections with history-making black candidates at the end of the decade.
In 1989, a U.S. Attorney named Rudy Giuliani made his first run for mayor of New York City. Just before the election, two different polls showed him trailing his African-American opponent, Manhattan Borough President
David Dinkins, by double digits:
Dinkins led Giuliani by fourteen points in the ABC/New York Daily News tracking survey of the race,
fifteen points according to Gallup, and
by eighteen points in a poll conducted by the New York Observer. Even the exit polls overestimated Dinkins's eventual two-point margin of victory.
The more infamous race that year took place in Virginia. I'll let Scott Keeter and Nilanthi Samaranayake of
the Pew Research Center explain:
Virginia Democrat and African American Douglas Wilder edged white Republican Marshall Coleman by less than one percentage point to become the nation's first elected black governor. But two of three polls conducted just days before the election showed Wilder leading by double-digits; a third poll had him 4 points ahead.
The exit poll results from the Virginia governor's race were particularly jarring; according to Pew, "an exit poll conducted on Election Day showed Wilder winning by 10 points, while accurately tallying the vote in the other two statewide races." This indicated that some voters must not have been honest with poll-takers. Keeter and Samaranayake noted that, unlike other exit polls that used an anonymous written ballot to collect voters' responses, "this one had interviewers asking voters face-to-face how they voted, a situation that might increase the pressure to provide a socially desirable response."
It didn't take long for major media outfits to trumpet explanations for why the polls in these two particular elections were so far off. Two days after Election Day, the New York Times ran an article about the "broad disparities" between the polls and election results in New York and Virginia:
In Virginia, Mason-Dixon Opinion Research Inc. acknowledged making a fundamental error in the way it conducted its Election Day polling. Its workers stopped voters outside polling places and asked them face to face how they had voted, rather than following the more widely accepted practice of having them fill out ''secret ballots'' and drop them in a box. All the New York polls used this method.
Brad Coker, Mason-Dixon's president, said that in hindsight he thought some voters might have been reluctant to admit to a poll taker that they had not voted for the black candidate, L. Douglas Wilder. The organization's Election Day polls gave a 10-point lead to Mr. Wilder, the Democrat, and at least three television stations reported the figures right after the polls closed and declared Mr. Wilder the winner. Today it appears that Mr. Wilder did defeat J. Marshall Coleman, the Republican, but by only one quarter of a percentage point.
''I'm a big supporter of pre-election polls and exit polls,'' said Peter D. Hart, a leading Democratic poll taker here, ''but only for figuring out the dynamics of an election and not who is going to win. What bothers me about yesterday is that the exit polls are obscuring the news, setting up incorrect expectations and leaving everyone talking about 'Oops, what happened?' rather than the historic story of the elections of Wilder and Dinkins.''
Virginia political scientist Larry Sabato was even more blatant in his analysis. ''Whites tell pollsters ahead of time that they are voting for the black candidate,'' he said, ''and then they go into the voting booth and can't quite pull down that lever.'' And, a Richmond Times-Dispatch article that ran the same day quoted Coker as saying, "Virginians were put on the spot: If we don't vote for Wilder, [then] we're going to be perceived of as racist." Whatever the reason was, the gap between poll results and election results in races between a black candidate and a white candidate was becoming a recurring problem.
For Wilder, it wasn't the first time his poll numbers had foretold a different election result. In 1985, he was elected lieutenant governor by three precentage points, but pre-election polls had pegged his lead at anywhere from nine to 22 points. The same polls were much more accurate with regard to the governor's race that year. In part because Wilder's experiences provided stronger evidence (and more of it) that voters lied to pollsters about voting for the black candidate than did Tom Bradley's 1982 campaign, some individuals prefer the term "Wilder Effect" to "Bradley Effect".
Whatever you call it, the concept that the effect was attributable to white voters who were reluctant to tell poll takers they were not voting for a black candidate gained steam following the '89 elections. In 1993, ahead of Mayor Dinkins's rematch wth Giuliani, the New York Times reported on a study by Larry Hugick, the director of political and media surveys at Princeton Survey Research Associates, who concluded that, in general elections that pitted a black Democrat against a white Republican, polls consistently overestimated the black candidate's strength. Said the Times:
[Mr. Hugick] concluded that white Democrats and Democratic-leaning independent voters were more likely than blacks to be undecided, torn between their allegiance to political party and to race. These ambivalent voters told interviewers conducting polls that they were undecided, but overwhelmingly supported the white candidate in the voting booth.
Mr. Hugick speculated that some of these voters were genuinely undecided, while others were reluctant to admit they were voting against a black Democrat, perhaps worried they would be perceived as racially biased.
"To some extent, these voters are torn between party and doubts about the ability of black candidates to do a good job," he said. "It's racial stereotyping. It's a way race affects us in this country. We've understood racial intolerance is highly unacceptable, but that doesn't mean that people have internalized tolerance."
After Giuliani defeated Dinkins in another close race, Hugick sent out a memo to his clients and friends, detailing his findings and asserting that the 1993 New York mayoral election "again demonstrated that late polls in biracial general elections overstate the black candidate's vote-getting ability." Following that election, however, the Wilder/Bradley effect appeared to be waning. African-American candidates who waged multiple campaigns provided a good "control" group with which to examine whether the Bradley/Wilder effect still occurred. Polls showed Mayor Dinkins's 1993 rematch with Giuliani, for example, to be a real dead heat. Giuliani, of course, won the race by a margin comparable to Dinkins's narrow 1989 victory.
In 1990, former Charlotte Mayor Harvey Gantt, an African-American Democrat, challenged longtime Sen. Jesse Helms (R-NC). Two of three independent polls conducted just before the election showed Gantt leading Helms, but none showed the controversial senator with a lead of six percentage points, his eventual margin of victory. Six years later, when Helms again fended off a challenge from Gantt, the polls were much more accurate; the incumbent's 53%-46% win was consistent with polling that gave him a modest lead.
Similarly, when Carol Moseley Braun became the first (and, to date, only) African-American woman elected to the U.S. Senate in 1992, the election results were much closer than polls had suggested. Braun defeated Republican Richard Williamson by a ten-point margin, but polls taken just days before the election showed her with a lead ranging from 17 to 20 points. (Notably, all of these polls pegged Braun's level of support at less than the 53% of the vote she ultimately received, so it's certainly possible that everyone who claimed to be voting for Braun did actually vote for her, but if these polls were accurate, then it would mean that Williamson won nearly all of the late deciders.) Six years later, Braun lost her bid for re-election by less than 100,000 votes, or three percentage points, to Peter Fitzgerald, a wealthy state senator from the Chicago suburbs. It could have been a lot worse; pre-election polls gave Fitzgerald a lead of three to ten points. A strong turnout among Illinois's black voters may have boosted Braun.
The cases of Gantt and Braun suggest that, to the extent that polls significantly overstate a black candidate's support and/or undertell his/her white opponent's strength in the black candidate's first run for statewide office, this phenomenon does not repeat itself when the same black politician runs again for the same office. Tom Bradley himself got creamed by Deukmejian in their 1986 rematch, but at least the mayor wasn't misled by poll numbers that showed him to be ahead. Prior to the election, all public polls had Bradley trailing the incumbent by double digits. Though Deukmejian won re-election by a larger margin than the polls indicated he would, none greatly overtold Bradley's level of support.
The paucity of competitive general-election match-ups between a white candidate and a black candidate in 2000, 2002 and 2004 made it difficult to gauge whether the Bradley effect was still waning. There was the race for governor of New York in 2002, in which incumbent Gov. George E. Pataki (R) crushed his opponents to win a third term. Pataki won 48%, finishing well ahead of the state's African-American Comptroller, H. Carl McCall (D), who took 32%, and a slew of other candidates who weren't black. Pataki had led McCall by eleven points in a survey taken in late September.
In 2004, two members of the U.S. House of Representatives from the Atlanta area squared off against each other for the Senate seat being vacated by Georgia Democrat Zell Miller. Rep. Johnny Isakson, a white Republican, defeated Rep. Denise Majette, an African-American Democrat, winning 58% of the vote to Majette's 40%. The weekend before the election, the Atlanta Journal-Constitution published the results of a poll that had Isakson leading Majette by thirteen points, 49% to 36%.
Then came 2008. There wasn't a single black candidate, Republican or Democrat, in a highly competitive gubernatorial or senatorial contest that year, but the presidential candidacy of Barack Obama revived talk about the Bradley/Wilder effect. Other than Obama's surprise loss to then-Sen. Hillary Clinton in the New Hampshire primary, however, there were no conspicuous gaps between Obama's poll numbers and share of the vote he actually received in the 2008 Democratic primaries. In the general election, Obama experienced a fate similar to that of Washington, Wilder and Dinkins: he won, but some polls overtold hs eventual margin of victory. A lot of polls underestimated voter support for the Republican nominee, Sen. John McCain of Arizona, both nationally and at the state level, but very few overestimated Obama's support. This could simply have meant that late deciders broke for McCain. Whether that's true or not, there was not a single battleground state in which all the polls were way off. Obama did not experience the Wilder effect in November.
To what, then, should we attribute the salient discrepancy between the poll and election results in Utah and Florida this year? Regardless of the cause(s), this much is clear: The Wilder Effect still exists, but with a twist; the effect seems to only manifest itself in races between a black Republican and a white Democrat.
I first noticed this correlation between the presence/absence of a Bradley/Wilder effect and the political affiliations of the candidates in 2006. That year saw more competitive races for statewide office between a black candidate and a white candidate than any other election cycle I can remember. The U.S. Senate races in Tennessee and Maryland provided data for a good comparative analysis. Both featured a charismatic, energetic black politician vying against an older, duller white politician for an open U.S. Senate seat. (In Mississippi, State Rep. Erik R. Fleming, a black Democrat, challenged Sen. Trent Lott, a white Republican, for his Senate seat, but that race was not considered competitive, and Lott won re-election by a decisive margin.)
In Tennessee, Democrats nominated Harold Ford, Jr., a five-term congressman from Memphis, to run for the seat of then-Senate Majority Leader Bill Frist (R), who was retiring. Ford and his Republican opponent, former Chattanooga Mayor Bob Corker, traded the lead in the polls back and forth until October, when Corker appeared to be pulling away. Heading into Election Day, independent polls put Corker's lead at anywhere from three to twelve points. The only public polls that showed Ford leading were conducted by partisan Democratic pollsters. Corker did end up winning, but only by about 50,000 votes (less than three percentage points). There had been no Wilder effect in the Tennessee Senate race.
In Maryland, it was a different story. Perhaps no U.S. Senate candidate in 2006 garnered as much enthusiasm as Michael Steele. Four years after becoming Maryland's first African-American lieutenant governor, Steele appeared to be on the verge of beating the odds to win a U.S. Senate seat in a state where 55% of voters were registered Democrats. He had fought the Democratic candidate, ten-term Rep. Ben Cardin, to a draw in the polls. There was even talk of a "Reverse Wilder Effect" in the race.
“With Steele sporting a 7-point lead among white voters,” pondered RealClearPolitics’s John McIntyre just days before the election, “what happens to Cardin's lead if Steele's 12% of the black vote goes to 20% or 25%?” We found out: Steele managed to take 25% of the black vote, according to exit polls, but lost to Cardin by a 54%-44% margin.
Having followed the 2004 elections as closely as I did, I knew how pollsters could sometimes get it wrong (This was before I looked into the history of the "Bradley Effect" and discovered just how wrong.), but the results of the Maryland Senate race had left a particularly bad taste in my mouth because (1) I had gotten my hopes up, based on what appeared to be momentum towards Steele in the final weeks of the campaign, and (2) I couldn't find another comparably high-profile race that cycle in which so many polls had been so far off.
I should mention that, in the three 2006 gubernatorial contests between a black candidate and a white candidate, there was no such disparity between the polls and the outcome. In Ohio, Republican
Ken Blackwell, an African-American who had served as mayor of Cincinnati, state Treasurer and Ohio Secretary of State, lost in a rout to
Ted Strickland, a white Democratic congressman from Lucasville. Strickland's 900,000-vote margin of victory accorded with
polls that put him ahead of Blackwell by about 20 points. Next door, in Pennsylvania, Gov.
Ed Rendell (D) held off a challenge from businessman and former Pittsburgh Steelers Wide Receiver
Lynn Swann (R) to win a second term. Swann actually
outperformed his standing in most polls, winning 40% of the vote to Rendell's 60%.
In Massachusetts, Deval Patrick (D) was elected the Bay State's first black governor by a 56%-35% margin.
Some polls had shown him leading his opponent, Lt. Gov. Kerry Healey (R) by more, but every poll of the race had understated Healey's support. Thus, much like the 2008 presidential election, it appears that most undecided voters in the Massachusetts and Pennsylvania gubernatorial races ended up supporting the Republican. Race made no difference.
In light of this data, I was inclined to write off Steele's discouraging loss as a fluke. Then, in 2010, I was again bemused by the results of an election that did not comport with the polling. This time, it was in Colorado. Rep.
Ed Perlmutter (D), an aging, white pol from Golden,
faced a tough challenge from Aurora City Councilman Ryan Frazier (R), an ebullient young African-American. Things seemed to be going Frazier's way:
the left leaning Denver Post endorsed him, and
a desperate-looking Perlmutter actually smacked him at a debate in October. Even more encouraging,
the only public poll of the race showed Frazier leading Perlmutter, 40%-39%. While a one-point lead is no cause for ecstasy, an incumbent polling at 39% is generally not considered to be in good shape. So you can imagine how disappointed I was to learn that
Perlmutter won re-election. It wasn't even close; the incumbent won by nearly 25,000 votes, taking 53% of the vote to Frazier's 42%. Ouch.
Now, the losses of Love and West, belied by multiple polls, add to this pattern of black Republicans who lost elections despite poll results that showed them tied or leading their white Democratic opponents. What could explain this 21st-century version of the Wilder Effect? Are large numbers of white voters reluctant to unequivocally declare their support for a black candidate's opponent? If so, then why did Ford, Blackwell, Swann, Patrick and Obama not experience the Wilder effect? It's certainly possible that the polls in Utah and Florida, like many other polls this cycle, underestimated Democratic turnout and/or overestimated Republican turnout, but then what accounts for the unexpectedly lopsided defeats of Frazier and Steele? I will not endeavor to answer these questions in this post; my only point is that the Bradley/Wilder effect still occurs but (for whatever reason) only in races between a white Democrat and a black Republican. (Take note, Herman Cain.)
