The “Bradley Effect” is the supposed polling error that overstates support for black candidates because, the theory goes, some white voters will tell pollsters that they support the black candidate but nonetheless don’t vote that way on election day. This has never made a lot of sense to me. Why would a voter would feel the need to appear “politically correct” to an anonymous (or even robotic) pollster, particularly since the voter usually needn’t disclose the exact reason for his or her decision? Furthermore, if a white voter won’t vote for a black candidate but is embarrassed to admit that, why not simply conjure up a more socially acceptable reason?
In any event, several interesting stories in recent weeks have pretty much debunked the notion that the alleged “Bradley Effect” actually had any effect at all in Tom Bradley’s loss to George Deukmejian in the 1982 CA Governor’s race. And there is a whole lot of evidence that even if this “Effect” did exist in the past, it is gone now. Nate Silver talks about what he thinks happened at this link. And just look at the polls. When you hover over any state on Pollster.com’s election night map, it shows you where the polling trend lines were heading, and what the actual result was. Inevitable conclusion: the polls were right, as they usually are. In many states (e.g., OH, FL, PA, IN, and NV), undecideds broke for Obama; in some (e.g., VA, CO, MO, and IA) they broke for McCain; in a few (e.g., NC, WI) they broke about evenly. But in no state was there an unexpectedly large McCain surge, and on a quick perusal I didn’t see any close-ish state in which Obama underperformed the polls. Just as in Deval Patrick’s victory here two years ago, in which the election eve polls were almost exactly right, last night voters did almost exactly what they told pollsters they would do.
So let’s put this canard to bed once and for all. There is no “Bradley Effect.” And, as I’ve said before in similar circumstances, polling works.
Named after the sheriff in Blazing Saddles. When things are really really bad, people look past color and get behind whoever is willing to come up with solutions to the problem.
<
p> – Dan
There does seem to be an effect of polling bias in that Alaskans being polled were reluctant to say they’d vote for a convicted felon.
<
p>Possibly something similar happened in California regarding Proposition 8.
A recent piece in the WSJ, by someone who worked on the Deukmejian campaign, laid out a pretty good debunking of the so-called Bradley effect.
The so-called “Bradley effect” was definitely, incontrovertibly real in both of Harold Washington’s elections to mayor of Chicago, to a dramatic extent. In other elections in the 80s it was more or less clear, and some of the ones often cited are open to interpretation.
<
p>To whatever extent it was real in the 80s, I don’t think anyone doubted that it had dwindled since then. One topic of debate was whether it was all gone, or still here but just to a fairly small extent (1%? 2%).
<
p>Keep in mind that this effect is not defined as “white voter tells pollster they’ll vote for the black candidate even though they don’t mean it, because they want to appear ‘good'”. The effect is “white voters who say they’ll vote for the black candidate end up voting for white candidate to some extent”. The “politically correct” hypothesis is just that – a hypothesis, an attempt at coming up with an explanation for the (apparently) observed behavior. Personally, it’s not the hypothesis I’d find most compelling. I would much more readily believe “white voters say they’ll vote for the black candidate but some of them, when they get to the voting booth, balk and can’t bring themselves to do it”. Obviously we have no way of ascertaining if either of these is true, though at least the latter has the advantage of explaining why the effect would’ve been stronger in the 80s than today.
<
p>Also, I don’t think anyone familiar with polling who was talking about the Bradley effect ever suggested that it was the only “effect” that might affect polling numbers in systematic ways. Just for some examples:
<
p>So yes, polls were pretty much right on, and I was operating under the assumption that they were fairly accurate (though also that they were not predictive of the election result until the final weeks). However, leaping from that to your statements about the Bradley effect is indeed leaping.
I do not say anywhere in the post that the “Bradley Effect” has never existed anywhere (maybe you interpreted my last paragraph that way, though that’s not exactly what I meant). What I do say, and what you don’t dispute, is that it didn’t exist in Tom Bradley’s race. (Maybe it existed in some local races 20 or so years ago; maybe not; I don’t know.) I also said that it didn’t exist in Deval Patrick’s race, or in Obama’s race. Nate Silver didn’t analyze the Patrick race, but he agrees with me on Obama: “No evidence of a Bradley Effect — none whatsoever.” I can’t tell whether you disagree, but if you do, have you got any evidence?
<
p>Also, you seem to question my definition of the Bradley Effect, but I don’t see any meaningful difference. Your definition:
<
p>
<
p>Mine:
<
p>
<
p>On the other stuff — sure, obviously the cell phone issue could cause a systematic bias. But your speculation — “what if the Bradley effect is real across the entire country, and just happens to be exactly the same as the cellphone effect but in the opposite direction?” — is just that: speculation. And pretty far-fetched speculation, at that. I’m not really sure how much that adds to the discussion, especially since some pollsters did call cell phones.
I am not making any statement about whether the Bradley effect still exists or does not, and you seem to have figured that out but are puzzled by it (and therefore don’t seem to believe it), which is why you missed the point of my “what if”: it’s meant to show you that you can’t conclude that it didn’t happen, any more than you could conclude that it did. There’s no strong evidence either way, from this election. Maybe it’s there and maybe it’s not.
<
p>Your definition, however, remains wrong, because instead of describing the actual observed effect, you are suggesting a reason for the effect. And because the reason you proposed is one that seems like it doesn’t make sense, your definition therefore presents the whole idea of a Bradley effect as if it were something that can’t make sense. If you merely described the effect neutrally, that implication would (properly) no longer be there.
<
p>As for cell phones, I wasn’t speculating, but you seemed to have read my comments about that as if I was. I mean it: it has been reasonably established that landline-only polls understated Obama’s numbers by about 2%, and since some polls did include cell phones, the overall skew was estimated to be around 1%-2%.
<
p>P.S.
Chicago, incidentally, isn’t really “local” – it’s the third-largest city in the US, it dominates Illinois politically (including several congressional districts), and it’s where Deval Patrick is from and where Barack Obama first got elected – in a very tough campaign, incidentally. It’s also the politics Jesse Jackson out of. Traditionally, people from Chicago who got elected to Congress or the US Senate or found other high profile federal government roles always viewed the mayoralty of Chicago as “higher office”, the step up that people strive for after they’ve done “smaller” jobs like, say, US Senate. Harold Washington’s elections were a national political earthquake, and possibly the most-covered political story in the country in the early 80s outside of Reagan. Contested elections for mayor of Chicago have historically been more significant, in national terms, than most elections for Congress and many elections for Governor.