Skip to main content
ABC News
A Guide to Forecast Model Updates

Now that our presidential forecast model has been released, we’re looking forward to obsessing over every detail and seeing how the forecast changes as new polls and economic data are released. I hope our readers will feel the same way, so here’s a quick guide on what to look for in our model updates.

In 2008 and 2010, we published an article to accompany every new update to the forecast. We moved away from that rule in the primaries this year, and I think it worked better. (Among other things, having to publish an article could sometimes be a deterrent to updating the model at all.)

However, we do plan to run model updates frequently — perhaps four or five times a week for the time being, then almost every day before long. So how will you know when there are new numbers on the site?

One way is to look at the top of the right-hand column. The red text under “FiveThirtyEight Forecast” will tell you when we’ve most recently updated the model.

The other way is to follow the FiveThirtyEight Twitter feed. I’ll almost always post on Twitter when there’s a new forecast on the site.

The sweet spot for model updates will be in the late afternoon or early evening. Polls generally come out in two batches: some in the early morning to catch the top of the news cycle, and some in the late afternoon to catch the evening news. Also, the economic index in our model uses the stock market along with other factors, and the market closes at 4 p.m. So you’ll more often see our updates in the p.m. hours, but there will be exceptions, especially when there is a large volume of new polling.

For the time being, you should generally not expect to see huge changes in the numbers unless there is major economic news. The model is designed to be quite conservative about how it weighs new polls at this early stage, and a couple of new surveys will rarely make more than a marginal difference.

For this reason, our discussions over the next month or so are likely to take more of a macroscopic view toward the election. The danger in focusing too much on individual polls is that you’ll pay more attention to the “newsworthy” outliers than to the well-behaved and well-constructed polls that are in line with the consensus.

At this point, instead, new economic news will often have a more discernible effect on the forecast than new polls, since they affect the numbers in every state. Over time, this will change, since the model weights the polls more and the economic index less as we get closer to the election.

Since the stock market — as measured by the change in the S&P 500 over a rolling interval — is the one economic component that gets updated daily, it may have an especially evident effect on the daily forecasts. If the stock market moves substantially over the day (say, a gain or loss of greater than 1 percent), this will usually outweigh the polling news for the day, at least at this stage when polling is sparse and the model weights the economic component fairly heavily.

We did some further testing on the stock market component of the model over the weekend after it produced some especially large changes in the forecast last week, and dampened its effect slightly — but our analysis suggests that accounting for the stock market should improve model performance. The reason is not that the stock market is important to voters per se, but it provides near-instantaneous data on consensus views on the economy. For instance, if the stock market declines by 3 percent in a day because of unfavorable developments in Europe, that will be reflected in the forecast almost immediately, rather than our having to wait for the effects of the worsened economic outlook to show up in polls or other economic indicators.

If you want to get a sense for the flow of the polls and polls alone, you can look at the “now-cast” — our estimate of what would happen if the election were held today. The now-cast removes the economic component from the model, and it is somewhat deliberately designed to be volatile, just as the forecast will become much more aggressive in the few days leading up to the election. In the now-cast, a candidate’s Electoral College winning chances might easily change by 5 percent if he has a day of strong or weak polls.

Finally, we frequently get questions about which polls are not included in the model. To make everything a little more transparent this year, we have decided not to list a poll until such time as we have run a forecast that includes it. So, you should have a sense for exactly what inputs are producing the model output’s at any given time.

We do appreciate tips, however, on polls that you think we may have missed, especially if they come from a smaller polling organization that may not do as good a job of publicizing its results. We catch perhaps 95 percent or 98 percent of polls as they come in, but we do miss some.

Nate Silver founded and was the editor in chief of FiveThirtyEight.


Filed under