That doesn’t mean that if we somehow re-ran the 2016 presidential election 10 times, Clinton would win nine of them. Instead, it is a measure of how deep the strength of the evidence is in favor of a Clinton victory. Based on everything we currently know about this election—by comparing it to past campaigns, and interpreting the latest polls—we can be relatively certain that Hillary Clinton will win: 90 on a scale from 0 to 100.
It wasn’t always this way. If we start from the “fundamentals” of the election—that is, the systematic, long-term economic and political factors that correlate with presidential election outcomes—it’s Donald Trump (not Hillary Clinton) who would be considered the favorite. Our basis is the highly regarded Time for Change model, created by political scientist Alan Abramowitz. That model generates a forecast of the national election based on just three variables: second quarter GDP growth, presidential approval, and the number of terms that the incumbent party has held the presidency. Between a slowly growing economy and the fact that presidential candidates generally perform worse after their party has been in office for two terms, the Time for Change model would expect Clinton to have only a 35 percent chance of winning.
Current polling, however, gives us a basis to update that initial estimate. Clinton came out of the Democratic convention with substantial leads in the polls across many states. Previous presidential candidates with advantages this large at this stage of the campaign have all gone on to win. As of today, Clinton is ahead of Trump in every state that Barack Obama won in 2012. That alone would be good for 332 electoral votes out of the 270 needed for victory. But polls also indicate that Clinton has a reasonable chance to add victories in North Carolina, Arizona, Georgia—and even Missouri.
Either the polls are badly wrong, or Trump would have to stage a major comeback, in a large number of states, to win the presidency. These aren’t impossibilities, but they’re not likely, either. We think there’s around a 10 percent chance that either (or both) of those will come to pass.
What would it mean for the polls to be “wrong”? Our model is based on state-level public opinion data from every source we can find. Mostly these are news organizations or polling firms who publish survey results as a way to make news, or generate publicity. Individual campaigns, advocacy organizations, non-profits, or political party groups all occasionally sponsor and release poll results as well.
To argue that the polls are wrong is to believe that all of these different firms and organizations, with all of their different questionnaires and methodologies, in all of these different states, are all generating results that are, on average, too favorable to Clinton or Trump. That is to say, they’re all making the same mistakes (at least, on average). In 2012, public polls showed Mitt Romney performing a bit better, on average, than he actually did, so it is possible. But the larger the Clinton lead, the less likely it is to be “reversed” by relatively smaller polling inaccuracies. In addition, there’s always the possibility that the polls are biased in the other direction: Perhaps Clinton will outperform her forecasts on Election Day. After all, older white voters—who are more likely to support Trump—are exactly the types of people who are more likely to respond to polls.
As described in our methodology, we do adjust polls in one way to minimize potential biases. Poll results that are released by explicitly partisan firms or political organizations tend to be slightly more favorable to candidates of the same party. To account for this effect, we subtract 1.5 percent from those candidates, and add 1.5 percent to candidates from the opposing party. We don’t try to perform any other “corrections” to the polling data that goes into our model, because in our view, that would only add unnecessary complexity without systematically reducing bias. Instead, we assume that all of the other types of errors that can go into poll results (sampling, wording, design, etc.) will cancel out as random noise.
Interestingly, the divergence between the fundamentals and the polls makes 2016 a very different presidential election than either 2008 or 2012—and one that is more challenging to predict. In both 2008 and 2012, the election fundamentals and the polls never really disagreed: Barack Obama began as the favorite, and evidence from polls only reinforced that position. As a result, forecasts in those two elections were much more stable and less prone to revisions as new polling became available.
This year, for our forecast to continue showing a high probability of a Clinton victory, she will need to sustain a lead in the polls that is large enough to counteract the drag of the political climate. Indeed, there is no guarantee that our high level of certainty in the outcome will persist. There has not even been a presidential debate yet. In 2012, President Obama’s lackluster performance in the first debate was arguably the most damaging event of his campaign. If Clinton’s poll numbers start to fall more in line with her fundamentals, our forecast of her chances of winning will also decline. Likewise, any day without new polling data will cause our model to "revert" toward what the historical factors would predict, which will drive Clinton's chances of winning slightly downward.
We hope you enjoy the new Daily Kos Elections site and find it interesting and informative. And come back soon: Whether or not any new polls are released, our forecasts for every race will be updated daily from now through Election Day.
Comments are closed on this story.