On Republican Math
By now, many of you may have seen the stories indicating that Republican activists—like all partisans when their candidate is behind—are urging people not to believe the polls. Instead, we are to believe that the candidate’s internal polls show him to be tied or ahead.
The particular twist that the folks at FOX News and the Romney campaign are putting on this claim is that the “likely voter” models being used by mainstream polling firms are skewed towards the Democrats.
This, as it happens, is quite possible: many people don’t understand that the reports of “who’s ahead” and “who’s behind” that so dominate the news are not, in fact, simple reports of percentages in a sample. Instead, they are estimates massaged from the data. Pollsters look at the data and then add “weight” for the opinions of those deemed “likely to vote” while discounting the opinions of those deemed “not likely to vote.” Such weighting is particularly common if the sample was overrepresented or underrepresented by any important group: if the sample has too many college students by overall percentage of the population, for example, pollsters will unweight their opinion in the likely voter model derived from that sample because college students vote at lower rates than does the general population.
The thing is, almost every likely voter model ever built is weighted in a way that overestimates votes for Republican candidates rather than underestimates it. This is because the kinds of people who tend to vote in higher numbers — white people, college educated, middle class, etc, — tend to be people who vote Republican. By contrast, groups that tend to vote Democratic — minorities, less educated, less well off — tend to vote at lower rates. Thus pollsters, trying to predict the final vote, overrepresent white, college educated middle class people in their estimates while underrepresenting others.
(As an aside, it was just this issue that stopped me watching CNN in 2000: their likely voter model was so skewed towards Bush I couldn’t stand it. I only watch CNN for coverage of emergencies now … although their election night data is really good.)
So is it possible that all the mainstream pollsters have built likely voter models that advantage Democrats? Sure it is is. Have they?
It’s not very likely.