The Bradley Effect

I've discussed the Bradley Effect with lots of you individually and, since Boris Ricks suggested it, I'll post a couple of my thoughts here along with a link to Nate Silver's post on it this afternoon. That piece includes a number of links to other pieces on the Bradley Effect that are all worth reading.

Here's my 2 cents (though the devaluation of the currency may make my thoughts worth only a cent or so): I have heard many, many smart people argue that, if Obama is winning by 5 or 6 points in polls just before Election Day, it will mean the race will be basically even because of the Bradley Effect. I strongly disagree. I do not see any evidence to support this worry (and I search for evidence to support my worries ALL the time).

First, we have to be clear about what the Bradley Effect is and is not. The Bradley Effect refers to the idea that poll numbers over-estimate Obama's (or any African-American candidate's) actual standing among voters because some small number of voters tell pollsters they are planning to vote for Obama in an effort to appear more tolerant to the interviewer than they really are. The Bradley Effect does NOT have anything to do with people who choose not to vote for Obama because of his race and tell pollsters they are voting for McCain or someone else. This is a very important distinction as my claim that the Bradley Effect is negligible or non-existent is not to deny that Obama's race is an issue for some voters. I am simply saying that whatever racial effect exists is picked up by good pollsters and is already a part of their results.

Now that we're clear on what the Bradley Effect is, we can move on to talking about whether it exists. There just isn't any persuasive evidence that the Bradley Effect has appeared in any recent campaign. On the contrary, as Nate Silver points out, there is some evidence that a "Reverse Bradley Effect" was detected in some states (like South Carolina) during the primaries where Obama significantly outperformed his poll numbers. Two of the more recent high-profile African-American candidates to run for office were Harold Ford, Jr. and Deval Patrick. Both ran statewide campaigns in different states and there was no indication of a Bradley Effect in either case. Ford lost but he did not underperform the late polls in his campaign. Patrick won and also performed as polls predicted. Not only is there no recent evidence of a Bradley Effect, Tom Bradley's pollster argues there was no Bradley Effect ... in Tom Bradley's 1982 race - the race for which the Bradley Effect is named.

There's also a methodological problem going on within this discussion. How would we know if the Bradley Effect existed or not? Let's say the election is tomorrow and Barack Obama is leading in polls by 4 points. And let's further assume that Obama loses the next day by 2 points. What happened? Several things could explain the discrepancy. Late-deciders may have broken for McCain. McCain's turnout operation may have over-performed and Obama's turnout operation may have under-performed. Young voters and African-American voters who are both key to Obama's success may not have turned out as expected or hoped. Or there may be some other problem with likely voter models that led most or all pollsters to an incorrect result. The Bradley Effect would be another possible explanation. In fact, the answer might be many of these things in combination. So how would we know which of these explanations is correct? I would argue that, even with a lot of good exit poll data and a lot of post-election surveys, it would be hard to determine why the polls were off the mark. If the Bradley Effect is estimated to be as large as a 6% shift, you're still only talking about a few dozen respondents in a good poll sample of about 1,000 respondents.

So what do I think is really going on here? Let me suggest a possibility. Democrats are a paranoid lot who assume this election is going to be snatched from them in some horrific way and Republicans are searching for a glimmer of hope amidst a lot of polls that show Obama with a nice lead. Both have a reason, therefore, to latch onto the Bradley Effect as the fuel to feed their respective fears and hopes. It also makes for a good media story. The media wants this thing to be a nail-biter, a cliff-hanger, etc. So they run a lot of stories on the Bradley Effect to keep you tuning in thinking that Obama being up by 7 points really means this race is neck-and-neck.

But fearing it, wishing for it, suggesting it, etc., doesn't make it real. Evidence of its existence would. And that evidence just isn't there. If it does, I'd be happy for someone to show it to me.

Discuss ...

By the way, Mark Blumenthal of pollster.com just posted a comprehensive piece on the so-called "interviewer effect" as contributing to the Bradley Effect. This is a lot of names for something he (and I) conclude doesn't really exist.