Looking back to the mid-1970s, Rothschild says his model accurately predicts the winner of more than four in five statewide Senate or governor elections based on data available by late May or early June. Because very little polling in these kinds of races exists that far before the election, "it fills a bigger hole when it comes to providing some insight" at this early state, Rothschild said.
The model also yields a few interesting findings. Forget the idea that voters cast ballots differently for federal office than they do for statewide office; the midterm effect -- that is, whether being a member of the president's party affects your performance in a midterm election -- is identical for both types of elections.
Rothschild's model eschews all horse-race polling entirely. Indeed, only after Labor Day, he says, does an analysis of publicly-available state polling eclipse his model in predicting the winner.
The model isn't perfect, but it's an important reminder that the fundamentals of the race early on are better indicators of one side's likely performance than the latest polls in the news today.