Insightology

The Bradley Effect is quite often mentioned in the media as we approach the Presidential election. It refers to the under-performance of a black candidate as compared to poll numbers. A good summary of various issues can be found here. A lot of the information about this effect is speculative or based on sparse polling numbers from other races which have led to questions of whether the effect really existed, whether it was seen in other races and whether in 2008 it is still likely to be seen. Perhaps the best study on this issue was conducted by Anthony Greenwald and Bethany Albertson the University of Washington. They used data from the Clinton-Obama primary, the most comprehensive source available and studied it using regression analysis to explore the existence of the Bradley effect.

The researchers used data from 32 primaries where Clinton and Obama competed against each other and where pre-election polls were conducted in the week before the primary. They compared (regressed) the inaccuracy in prediction between Clinton and Obama against the proportion of blacks in the state population. Inaccuracy in prediction is derived as the difference between the poll prediction and actual voting patterns as obtained through exit polls. Results of the analysis are shown in this interesting chart. Focus on the states above and below the dashed lines as they are outside the margin of error.

Of the 32 states studied, three (New Hampshire, California and Rhode Island) showed a Bradley effect (i.e.) where Obama's support was over estimated by 7% points or more. A reverse Bradley effect was seen in 12 states whereby Obama's actual performance was underestimated by 7% or more, and the remaining 17 states performed within the margin of error of the polls. Apart from the fact that a reverse Bradley effect was unearthed, the reason these results are interesting is that there is a strong relationship between the prediction gap and the proportion of black population in the state. Overestimation of Obama's performance tends to happen in states where the proportion of the black population is about 8% or lower, while underestimation is more likely in states where the black population is about 20% or more.

So, it appears the Bradley effect existed in the Democratic primary, but an even stronger reverse Bradley effect also existed and in some states (Georgia, Alabama and South Carolina) it underestimates Obama's actual vote by at least 15% points. Will this have an effect on the election between Obama and McCain? Based on this research Greenwald and Albertson say that (http://uwnews.org/article.asp?articleID=44314 ) current polls may be underestimating Obama's support by 3 to 4% points, pointing out that both blacks and whites were found to be understating their support for Obama even in Republican states. But several caveats need to be considered.

While these are better numbers than any that have been used to study this issue, they still have potential problems. The numbers come from the Democratic primary and it is unclear what will happen in a general election when Republican and Independent voters are mixed in. The impact of gender is also not clear because the Democratic primary race between Clinton and Obama did not feature a white male. Finally, a different study found that race could be an issue and could be costing Obama as much as 6% points in support.   

We should know the answer in about three weeks.