The Bradley Effect is Dead; Pollsters Are Actually Good At Their Jobs

November 5, 2008

Mac Zilber

One interesting facet of this election that is (understandably) not going to get much coverage in the media but is useful for future reference is that the pollsters have proven that they are pretty good at their job. There was no evidence of a Bradley effect, youth effect, undecided effect, Barr effect, GOTV effect, cell-phone effect, or any other inaccuracy in the polling. To the contrary, pollsters essentially were spot-on in calling the election. You can fool all of the pollsters some of the time and some of the pollsters all of the time, but you can’t fool all of the pollsters all of the time.

Nationally, the polling average projected Barack Obama would win 52.0 to 44.4, according to pollster.com. In actuality, he is currently winning 52.4 to 46.3 (that number may go up a tenth of a point or two), not a bad result for the pollsters. Furthermore, two pollsters, Pew and Rasmussen, called the election exactly on the nose (52-46), and, if you extract the outliers, the final polling average showed Obama leading by 6.6%. Being within one half of one percent of being correct is pretty darn impressive.

On the state level, the polls had a similar level of accuracy. In Colorado and Florida, the polling average was off by only eight tenths of a percent. In Ohio and Georgia, the polling averages were within nine tenths of a percent of being correct. The polls were off by only 1.3 points in Missouri, 1.1 points in Virginia, and 2.1 points in Indiana. Most impressively, in North Carolina, the polling average was within two tenths of a percent of being accurate. That is one heck of a batting average.

There were a few notable polling inaccuracies on election day. In North Dakota, the polls were essentially tied, and John McCain ended up winning by 8.6%. However, I should mention that North Dakota is such a small state that a 7.9% polling error only accounts for about 24,000 votes. Nevada was the biggest polling failure among battleground states, as it was 5.3% off, but that was for good reason: black turnout in Nevada jumped 43% from 2004 and Latino turnout jumped 50% from 2004, with both groups voting overwhelmingly for Obama. This appears to be the single solitary state in which pollsters underestimated minority turnout.

On the senate level, the polling was, by and large, good, but had more of an error rate than the presidential polls. Louisiana senator Mary Landrieu won re-election, but by a tiny margin, when she was expected to win it by about ten points. In Oregon, senator Gordon Smith is in the middle of an incredibly close contest in which the votes are still being counted, despite trailing by an average of 5.3 points going into the election. The biggest surprise in the senate, however, was that Ted Stevens appears as if he is going to win in Alaska (I use the present tense, rather than the past tense, because the election has not quite been called yet). Stevens trailed by 8 and 22 points in the two polls that were taken after he was convicted of seven felonies, and it appeared he was dead in the water. I don’t believe this was a polling error, though, as much as it was a byproduct of two things:
1. Alaskans didn’t want to admit to pollsters that they would vote for a convicted felon just so he could continue making Alaska a welfare state.
2. Ohio was called for Obama at 6:23 P.M., Alaskan time. A lot of people who would have otherwise voted probably decided against voting because they figured the election was already over. This probably favored Republicans, as Alaska is one of the few states where McCain voters were more enthusiastic and motivated than Obama supporters, because of Palin.

Whatever the reason for the error, polls in Alaska were wrong across the board, as equally shady Republican representative Don Young outperformed polling by 17 points, and John McCain outperformed his polling numbers in Alaska by double-digits as well. Clearly the state simply ended up voting way more Republican than previously expected.

Despite these few blips, however, I think that what we can deduce from yesterday’s election results is that polls are generally a very accurate scientific measure of a candidate’s support and should not be dismissed. That is not to say that there aren’t bad polls, or even bad pollsters (I’m looking at you, Zogby), but the general average among polls is generally almost exactly right.

Questions or Comments? Email me at themuse [at] albanyhighcougar [dot] com and I will be sure to respond.