Talk Elections

The polls were conducted using automated telephone calls, internet panels of cell phone respondents, and weighted by voting patterns.

The polls were conducted using automated telephone calls, internet panels of cell phone respondents, and weighted by voting patterns.

Don’t get lost in the crosstabs.
Don’t get lost in the crosstabs. Trust us — you don’t want to take the route of scrutinizing the poll’s crosstabs for demographic anomalies, hoping to “prove” that it can’t possibly be right. Before long, you’ll wind up in the Valley Of Unskewed Polls. Sample sizes are one issue. If a 600-person poll breaks out the results for men, women, Hispanics, blacks, Democrats, Republicans, older voters, younger voters and so forth, those subsamples will have extremely high margins of error, pretty much guaranteeing there will be some strange-looking results. Also, these comparisons are often circular. It might be asserted that a poll must be wrong because its demographics don’t match other polls. But no one poll is a gold standard — exit polls certainly aren’t. There are also legitimate disagreements over methodology — some polls weight by partisan identification and some don’t, for example. Although some of these debates may be important in the abstract, our experience has been that they involve a lot of motivated reasoning when raised in the middle of the horse race.