How To Replace Early Polls
By Steven Shepard // May 18, 2012 | 12:15 p.m.
ORLANDO, Fla. -- Inundated by polling? Don't be. In fact, until Labor Day, you don't even need to see polling data in Senate and governors races to accurately predict the outcome, according to a new model a top public opinion expert has devised.
In a paper presented to polling experts at the annual gathering of the American Association of Public Opinion Researchers, Microsoft Research's David Rothschild suggests a fundamental model for predicting the winners of these state races months well before Election Day.
Rothschild uses weights based on incumbency, past election results in the state, economic indicators (change in income during the first quarter of the year of the election is most instructive, he says), ideological indicators (the American Conservative Union's grades help out) and biographical information (using the candidate's most recent elected position).
Rothschild also factors in the president's approval rating in the given state. For Senate and governor elections, Rothschild says, "Coattails do exist, and they [are] reasonably strong."
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.
Leave A Comment