Presidential election forecasts a month before Election Day can be way off

Clinton is starting to pull away from Trump in the various election forecasting models, but it’s clear from the forecasting models that Trump still has a chance:

• FiveThirtyEight has Clinton winning at 70% – 73% (and 80% if the election were held today)
• The Upshot from the New York Times has Clinton winning at 79%
• Princeton Election Consortium has Clinton winning at 86% – 91%
• Election Analytics at Illinois has Clinton winning at 99% (if the election were held today)
• Daily Kos (Drew Linzer model) has Clinton winning at 72%

In 2012, FiveThirtyEight forecast had Obama’s win probability at only 61.1% on October 11, 2012 with less than a month before the election (November 6, 2012). I showed the time series of the 2012 forecasts below, to show how the forecast firmed up and turned toward Obama before the election. Obama won in a landslide with 332 votes. The last FiveThirtyEight forecast right before the election had the actual election results as the most frequent outcome in the simulation (Nate Silver nailed it). Most of the other forecasting models picked at least 49 states correctly within a week of the election, but they were not nearly so accurate a month before the election. I’m showing FiveThirtyEight’s 2012 forecasts here to illustrate that it’s hard to forecast a landslide victory.

FiveThirtyEight’s 2012 Presidential Election Forecast

There is no guarantee we will see a similar pattern in 2016, but polls lag public opinion, and the polls taken since the debate last week show a huge bounce for Clinton. There is some information on Vox below about what the polls mean and where the bounce is the biggest for the 2016 election.

Election Analaytics @ Illinois has Clinton with the win at 99% if the election were held today, which is up from 77% last week. A lot has changed in a week. They are showing Clinton as a heavy  favorite, even if the election were to lean heavily toward Trump.

I’ll try to blog some more about the election in the next month. I’m pretty interested in methodology used in the forecasting models. If you have any requests, let me know.