After the election, Justin Wolfers wrote a column saying that questions about who people expected to win were better predictors of election outcomes than questions about who people intended to vote for. He offered this explanation: "Asking voters about their expectations allows them to reflect on everything they know about the race — which way they currently intend to vote, how likely they are to vote, whether they’re persuadable, the voting intentions of their friends and neighbors, and their observations about bumper stickers, yard signs, the resonance of a candidate’s message and the momentum they sense in their communities."
You can see how this explanation would appeal to an economist, because it's a parallel to the way that markets work: combining scattered information in an optimal (or more realistically, pretty good) fashion. But there's another possibility: that voters are reflecting what the "experts" are saying, rather than information they have from their own lives. Even if they're not paying close attention to the campaign, voters are likely to get a sense of what "everybody thinks" will happen. In 2014, this would mean good predictions, because all the experts were saying that the Republicans would win big, while the polls left more doubt. But there have been other campaigns in which the the experts were wrong, notably 1948. Wolfers's explanation says that voters would have called that one correctly, or at least come close.
Did they? In late September 1948, a Gallup poll asked "regardless of how you, yourself, plan to vote, which candidate do you think will carry this state: Truman, Dewey, or Wallace?" 25% said Truman, 56% said Dewey, 15% don't know, and the rest Wallace or someone else (presumably Thurmond, who did carry several states in the South). Of course, the polls were famously wrong in that year, but the survey also asked who they would vote for: it was 46% for Dewey, 40% for Truman, 4% Wallace, 2% Thurmond. Given that it was not a large margin for Dewey, it seems like voters' own stated preferences might have been better predictors of how their state would go. Gallup did record state, so that could be checked, and I will do that in a later post.
PS: After the election, only 19% said they had expected Truman to win.