National Journal: “The entire concept of polling depends on having a set population from which one can take a random sample and get a generally representative snapshot. Pre-election polls have no existing population—the election hasn’t happened yet, and voting isn’t compulsory in the U.S., so we simply don’t have a population of who voted until all the polls have closed on Election Day.”
“We can’t remedy that. The population of voters will never exist prior to the election. Expecting polls to be able to consistently, accurately predict an election is asking more than is statistically and theoretically possible. Yes, we (pollsters) have a lot of information from past elections to help figure it out. But it’s gotten harder to poll a representative sample of the entire American adult population over the last few decades. Just in the past couple of election cycles, we have seen candidates activate people who typically don’t vote, so is it really surprising that the error rates of horse-race polls have increased?”
