Nate Silver, who runs the popular polling / prediction site FiveThirtyEight.com, developed a forecasting model that correctly picked the outcome of the 2012 presidential election. Now, he is out with the 2016 version, which is based on the one used in 2012, but has been improved, notably by splitting the model into three separate forecasts, using slightly different variables.
The easiest to understand is the polls only model, which determines the odds of a candidate winning a given state based on how that candidate did in recent polling. That model also accounts for how much time is left until the election, and looks at polling in nearby or politically similar states for guidance. The second model is called now-cast, and it uses the same inputs as the polling only model, except it assumes that the election is held today, not four months from now. As a result, the now-cast model’s assumptions are more liberal than the polls only model. The final model, polls plus, uses economic data and past voter behavior to make assumptions about how the election will turn out.
Unlike other models, FiveThirtyEight doesn’t assign categories such as solid, likely or lean to a state. Instead, it determines the chances for each candidate to win each state. By simulating the election thousands of times using the state data, the model comes up with a national result.
What does all that mean? All three forecasts agree that in January, Chief Justice Roberts will swear in President Hillary Clinton, although there’s a good chance that Donald Trump will win in Georgia. Using the polls only model, there is an 80.3% chance that Clinton wins nationally, but a 56.8% chance the Trump wins in Georgia. Clinton would have 354 electoral college votes to Trump’s 183. The now-cast model increases the chances of a Clinton romp nationally, with an 85.5% chance of winning, compared to 14.5% for Trump. There’s not much change in the Peach State results, though. There’s a 56.4 chance Trump would win, compared to a 43.6% chance for Clinton. Hillary picks up another four electoral college votes in this model to 358.
Perhaps because it uses historical data, the polls plus model has Georgia solidly in red territory, with Trump having a 73.2 chance of winning here, compared to a 26.7% chance for Clinton. Nationally, however, the race tightens up, with a 73.5% chance of a Clinton victory, 26.5 chance for Trump. In this scenario, the electoral college elects Clinton, 319 votes to 219 votes.
One more thing of note: It appears that these models began tracking data in early June. Since then, the odds of a Clinton win have greatly increased. In the national polls-plus model the odds of a Clinton win have increased from 62.9% to 73.5% since June 8th. The Georgia polls-plus model has the chance of a Trump win here declining from 80.5% to 73.2%. The polls-only model is even worse, with Trump having a 70.3% chance of winning back on June 8th to a 56.8% chance today, a drop of 13.5 points.
The two parties have their conventions next month. It will be interesting to see what, if anything, has changed in the forecast at the end of July.