Elliot Morris, the new boss at 538, has up a helpful discussion on the question of just what polls in March mean about the outcome of the November election. As you’d expect, they’re not terribly predictive. In fact, when Morris goes back through 538’s database, which goes back to the 1940s, they’re not really predictive at all and are frequently wildly off. You can read the piece for the examples. Of course that doesn’t mean they’re “wrong” necessarily. It just means they’re not predictive.
The big qualifier is that the big swings from earlier polls to final results have gone down over time. And the big driver of that is partisan polarization. The biggest example of a huge swing from March to November was Jimmy Carter being up by 14 points and then losing to Ronald Reagan by 10 points. Neither of those margins are remotely plausible today in a presidential general election. It just shows how many fewer voters are really up for grabs these days. The other factor which likely constrains movement in the polls, unique to this race, is the fact that the race is between, in effect, two incumbents. We literally know in advance what each man would be like as President.
