There is no Bradley Effect!

After Hillary Clinton's surprise win in the New Hampshire primary, speculation abounded that Obama was the victim of the so-called "Bradley Effect". After her win in California, such speculation resurfaced. The theory is that closet racists would lie to pollsters about their willingness to vote for a black candidate. This would cause the black candidate's vote tally to lag behind the polls' predictions.

But is this really true? critics of the Bradley Effect hypothesis have pointed to other possible explanations for New Hampshire and California, such as Clinton's display of emotion before the NH voting, and California's early voting.

To put the argument on a solid footing, I reviewed the results of all the primaries to date, to see if there was any evidence of a systemic Bradley Effect. Here was my methodology:

First, I identified states that met these three criteria:

• Used primaries instead of caucuses. This is because proponents of the Bradley Effect explanation have claimed that the Bradley Effect is only evident in the privacy of the voting booth. Supposedly the public nature of the caucus is enough to keep closet racists in the closet.
• Had polling data available from the week before the primary date. Any discrepancy between older polls and the election results are too easily explained by late surges.
• Have already held their primaries. For obvious reasons.

As of today (Februrary 14), seventeen states meet these criteria. For each of these states, I compared the vote spread between the two candidates to the average poll spread as reported by RealClearPolitics.com. (For some reason, RCP didn't average the poll spreads for Oklahoma, so I just took the average of the spreads from the last four polls.) I then sorted the results by the difference between the actual spread and the RCP spread. Here are the results:

State	Actual Spread	RCP Spread	Difference
GA	35.3% Obama	18.0% Obama	17.3% Obama
SC	28.9% Obama	11.6% Obama	17.3% Obama
AL	14.1% Obama	1.2% Clinton	15.3% Obama
VA	28.2% Obama	17.7% Obama	10.5% Obama
CT	3.1% Obama	4.0% Clinton	7.1% Obama
MO	1.2% Obama	5.7% Clinton	6.9% Obama
FL	16.7% Clinton	19.6% Clinton	2.9% Obama
OK	23.6% Clinton	25.5% Clinton	1.9% Obama
MD	23.5% Obama	22.3% Obama	1.2% Obama
NY	17.5% Clinton	17.2% Clinton	0.3% Clinton
TN	13.3% Clinton	13.0% Clinton	0.3% Clinton
IL	31.5% Obama	33.0% Obama	1.5% Clinton
NJ	9.8% Clinton	7.7% Clinton	2.1% Clinton
AZ	8.8% Clinton	6.0% Clinton	2.8% Clinton
MA	15.0% Clinton	7.0% Clinton	8.0% Clinton
CA	9.6% Clinton	1.2% Obama	10.8% Clinton
NH	2.6% Clinton	8.3% Obama	10.9% Clinton

These results yield some interesting observations:

• Obama beat the polls' expectations in nine races, whereas Clinton beat the expectations in eight.
• In six races, Obama beat the expectations by more than the 3-4% margin of error. Clinton only did so in three races.
• The two supposed Bradley Effect races were the only two in which Clinton beat expectations by more than 10%. Obama accomplished this in four races.

If the Bradley Effect were real, we would expect to see a trend of Clinton outperforming expectations. In fact, we see no such trend, and perhaps even a small trend in the opposite direction.

This is not to say that racism is not playing a role in the primaries. Remember that the Bradley Effect is a particular kind of racism - one in which voters lie to pollsters, distorting the poll results. Racists who are honest about their candidate preference are a different matter entirely.

Rather than focusing on the non-existant Bradley Effect, perhaps we should instead be questioning the dismal track record of the horserace polls this season. With over half of the spreads outside of the polls' statistical margin of error, there would appear to be some significant problems with the pollsters' methodology.