A topic that inevitably receives a lot of focus during election season is the partisan spread of the major media polls. Conservatives regularly complain that the polls are tilted against their side, and thus favor the Democrats.
They have a point.
To be sure, we can conclude that without accusing any pollster of malfeasance. In this essay, Mark Blumenthal of the Huffington Post correctly notes that the problem gets down to using registered voter polls. These tend to oversample Democrats. The argument in support of them is that, while a likely voter screen would draw a larger Republican sample, it would create more variability as pollsters would be guessing unduly at the final party turnout.
That is a fair point in some respects, although there is a cost associated with either choice. Using registered voter polls might cut down on variability, but they also create statistical bias. That is, the polls tend to oversample Democrats in a systematic fashion. And because partisan support is so strong – with 90 percent of Republicans supporting GOP candidates and 90 percent of Democrats supporting Democratic candidates – you regularly see the Democratic candidates’ margins overstated in polls of registered voters.
Let’s see if we can quantify this a little bit. To begin, we need to know the historical partisan tendencies of the electorate. For that, we can turn to the exit polls going back to 1972.