At The Upshot today, Nate Cohn and Josh Katz fire another round in the what-feels-ancient battle between political observers who pay significant attention to polls and other ongoing calculations of political contests (among whom they obviously count themselves), and political science “fundamentalists” who believe to one or extent or another that “objective” factors like incumbency, party ID, fundraising and economic conditions are perfectly adequate predictors of elections. They make the case that in recent years polls–including those early polls we’re always being told to ignore–actually do a better job of projecting Senate results than do the usual “fundamentals” calculations. And that’s what you should expect, since “fundamentals” that presumably shape public opinion are reflected in polling data, but not the other way around.
I’ll leave you to read the details, and the inevitable responses from academia. But I’d make the the observation that the argument here is or should be how to weigh factors like polling data, not whether to lash oneself to the mast like Odysseus and refuse to react to the siren song of the pollsters, as some academics and laymen alike tend to do, lest they become unmoored from the “fundamentals” and wind up obsessively reading Politico. Data is data, and rejecting any of it out of hand is questionable.
Having said that, it’s also helpful, particularly when there’s not enough polling data to create the more reliable averages, to look at polling methodologies that obviously affect the numbers. The biggie, of course, is the sampling methodology: whether the sample is of adults, registered voters, or likely voters, and also whether the pollster weights the results to fit some predetermined turnout pattern.
Occasionally you get a really asinine sampling method, as noted by Daily Kos Elections today in reporting a new Iowa poll from Loras College that arrived with this note down in the weeds:
“Likely voter sample included only those who voted in the 2010 general election and who indicated likeliness to vote in 2014 or who have registered to vote since 2010 and indicated likeliness to vote in 2014.”
If you start off by limiting yourself largely to people who voted in 2010, an epic GOP wave year, you’re going to have a hard time properly approximating the 2014 electorate.