If you look at aggregator sites that show what gabbers are gabbing about today, there will naturally be lots of “Tax Day” stuff, some of it prefab fluff and partisan spin, some of it actually useful (viz. this set of graphs from WaPo comparing top marginal tax rates since the 1930s with econonomic growth indicators, which pretty much makes a mockery of supply-side economics).
But the other buzzy topic this morning involves general election polls, with many conservatives bellowing in excitement at the first official Gallup Tracking Poll for the general, showing Mitt Romney with a two-point lead, and many liberals responding by pointing at a new CNN/ORC poll showing Obama up by nine points.
So it’s the perfect time for Nate Silver of FiveThirtyEight to issue one of his excellent reminders about how to read general election polls, and the common mistakes to avoid.
Nate offers twelve separate points of advice, but I’d emhasize and amplify five: (1) particularly this far out from the election, polling averages and internal trend lines are the most reliable indicators; (2) it’s helpful to understand “house effects,” the methodogical habits of individual pollsters that tend to produce different results–but that doesn’t mean just ignoring or throwing out data from pollsters whose “house effects” tend to produce top lines you don’t like (e.g., Rasmussen!); again, polling averages will sort all that out; (3) crosstabs providing demographic breakdowns are fascinating and sometimes useful, but be aware they typically involve very small samples and very high margins-of-error; (4) the closer we get to election day, the more it becomes important to distinguish between polls testing “likely voters” and those with less selective samples (e.g., registered voters, all adults, etc.), and the more assumptions about likelihood to vote will be worth arguing about. I’ll directly quote Nate on a fifth point, because I want to return to it later today:
Don’t over-learn the lessons of history. A final and more general point is that there have been only 16 presidential elections since World War II. That simply isn’t a lot of data, and overly specific conclusions from them, like “no recent president has been re-elected with an unemployment rate over 8.0 percent” or “no recent incumbent has lost when he did not face a primary challenge,” are often not very meaningful in practice and will generally not carry much predictive weight.
It’s tempting, of course, to ignore all polls as just so much noise and spin, or even as actively evil influences on U.S. politics. I’m in the camp of those who think the problem with excessive reliance on polls is usually one of exaggerating limited or bad information, and that the answer is more and better information, along with the perspective no public opinion survey can supply.