The Bradley effect is the theory that polls overstate support for a Black candidate running against a white one, because some voters tell pollsters what sounds socially acceptable and then vote differently in private. It’s named after Los Angeles Mayor Tom Bradley, who led in the polls for California governor in 1982 and narrowly lost.
The best evidence says the effect was real in some races through the early 1990s and then disappeared. A 2009 study by political scientist Daniel Hopkins found a median gap of about 3.1 points in races before 1996 and none afterward. Polling errors still happen, but they have other causes.
Where did the Bradley effect get its name? #
In 1982, Tom Bradley, the Democratic mayor of Los Angeles, ran for governor against Republican George Deukmejian. Pre-election polls showed Bradley ahead, and some early election-night projections called the race for him. Deukmejian won by about 1.2 points, 49.3% to 48.1%, a margin of roughly 93,000 votes.
The theory that followed: some white voters had told pollsters they backed Bradley, or were undecided, to avoid seeming prejudiced, then voted for Deukmejian.
The 1982 race has a complication that often gets left out. Bradley won the votes cast on Election Day. He lost once absentee ballots were counted, and pre-election polls weren’t designed to capture absentee voters well. That’s a reminder that a poll miss can have more than one explanation.
Which other races are cited as examples? #
Three races from 1989 and 1990 are usually offered as further evidence.
| Race | What the polls showed | Result |
|---|---|---|
| 1989 Virginia governor: Douglas Wilder (D) vs. Marshall Coleman (R) | Wilder ahead by about 9 points on average | Wilder won by less than half a point |
| 1989 New York City mayor: David Dinkins (D) vs. Rudy Giuliani (R) | Dinkins ahead by 14 to 18 points a week or so out | Dinkins won by about 2 points |
| 1990 North Carolina Senate: Harvey Gantt (D) vs. Jesse Helms (R) | Two late polls had Gantt up 4 to 6; one had Helms up 4 | Helms won by about 6 points |
Wilder’s win made him the first elected Black governor in US history, even as the polls overstated his margin. Sources: Wikipedia’s Bradley effect summary and the underlying race reporting it cites.
What causes it? Social desirability bias #
The mechanism behind the theory is well documented in survey research. Social desirability bias is the tendency to give the answer that sounds acceptable rather than the true one. It’s strongest when:
- A live interviewer asks the question. Telling a real person something you think they’ll disapprove of is harder than clicking a box on a screen.
- The issue is sensitive. Race, and other topics tied to social norms, raise the pressure to self-censor.
- “Undecided” is available. A respondent doesn’t have to lie outright. Saying “I haven’t decided” avoids the uncomfortable answer.
Does the Bradley effect still exist? #
The research says no, not in any measurable way since the mid-1990s.
Hopkins compared final pre-election polls with results in races that included Black candidates. He found a real gap of about 3 points before 1996, then nothing after. He linked the earlier gap to the racial rhetoric in those campaigns, and its disappearance to changing norms.
The biggest test came in 2008. Many commentators worried that Barack Obama’s polling lead would evaporate on Election Day. Instead the final polling averages were close to the result, and Obama matched or beat his polls in many states. The same was true in 2012.
Researchers also point out that the early misses had other plausible causes: late-deciding voters breaking against the frontrunner, turnout that differed from pollsters’ models, and, in 1982, absentee ballots. In any one race, those can be hard to separate from a Bradley effect.
Is the “shy Trump voter” the same thing? #
It’s a close cousin. The shy Trump voter theory held that some voters hid their support for Donald Trump from pollsters because backing him felt socially risky. It was offered to explain the state polling misses of 2016 and 2020.
The American Association for Public Opinion Research’s review of 2016 found little evidence of voters lying. It pointed instead to polls that didn’t weight by education and to late deciders who broke toward Trump. The 2020 miss is better explained by non-response: Trump supporters were less likely to take surveys at all, which pollsters couldn’t fully correct. We explain that problem in what non-response bias is.
The difference matters. Lying to a pollster and declining to take the poll leave different fingerprints, and they call for different fixes.
How do pollsters reduce social desirability bias today? #
- Self-administered surveys. Online panels, text-to-web surveys and automated phone polls remove the live interviewer. Comparing them with live-caller polls in the same race is one way to test for a hidden vote. See what a polling mode effect is.
- Better weighting. Weighting by education, and sometimes by past vote, corrects many errors once blamed on “shy” voters.
- Indirect questions. Some surveys ask who respondents think their neighbors will vote for, which can surface views people won’t state as their own.
When you look at a single poll, the mode matters, and it’s usually in the pollster’s release. Election Tracker shows each poll’s pollster, sample size, population and field dates on its card, and opens the pollster’s release in an in-app browser so you can check the mode and weighting before deciding how much weight to give it.
Frequently asked questions #
What’s the difference between the Bradley effect and the shy Trump voter? #
Both rest on social desirability bias. The Bradley effect refers to voters hiding opposition to a Black candidate to avoid appearing prejudiced. The shy Trump voter theory refers to voters hiding support for a controversial candidate to avoid judgment. Research found stronger evidence for the first, in its era, than for the second.
Has the Bradley effect been proven? #
Partly. Hopkins’ 2009 study found a measurable gap in races before 1996, which supports the theory for that period. It found no gap in later races, and specific cases like 1982 have competing explanations, so no single race proves it.
Why do polls still miss if the Bradley effect is gone? #
Mostly because of who answers polls and who turns out. Non-response bias, errors in likely-voter models and late-deciding voters account for most modern misses. Our guide on how to tell if a political poll is reliable covers the warning signs.
Do online polls avoid social desirability bias? #
They reduce it, because there’s no interviewer to impress. But online polls have their own problems, especially opt-in panels where respondents choose themselves. See whether online opt-in polls are reliable.