The 2016 U.S. Presidential Election Predictions

I don’t really follow politics much, but I am deeply fascinated by predictions. I can’t ever recall a more extreme set of incorrect predictions as the ones I saw for the 2016 U.S. presidential election.

I read poll after poll that predicted one of the largest voting landslides in history, with Hillary Clinton crushing Donald Trump. It didn’t happen that way. It was a convincing win by Trump.

But months before the November election, I remember briefly glancing at a news story about a college professor who predicted a Trump win. I more or less discounted the story as click-bait and didn’t really read it.

But after the election I went back and looked over the prediction model of Professor Allan Lichtman who works at American University. Lichtman’s prediction model has 13 predictor features. Each feature is a Boolean true/false value where a true value favors the incumbent party (Democrats in 2016). If 6 or more of the predictor features are false (favoring the challenging party), the challenging party will win.

In 2016, exactly six of the predictors had false values, barely indicating the incumbent Democratic party would lose. Those six features were:

1. Mid-term elections success - false (Democrats got crushed)
2. Sitting president is running - false (Obama is done)
3. A major policy change enacted - false (Nothing happened)
4. A foreign policy success - false (Nothing happened)
5. Candidate charisma - false (Hillary is shrill and frumpy) 
6. Third-party less than 5% vote - false (Johnson 6.5%)

On the one hand you can argue that this is just historical correlation. But you can also argue that these features (and the seven I didn’t show) are reasonable.


The two morals of the story are that predictions based on polls are very dicey, and that custom prediction models are often very powerful. (My Zoltar NFL football prediction system is an example of both of these).

Anyway, the 2016 elections were fascinating. I had no skin in the game — I felt both candidates were equally objectionable. But I must admit I got a certain amount of satisfaction in seeing all the post-election angst among Hollywood stars who overwhelmingly supported Clinton. Hollywood stars are a singularly uneducated group with a grotesque over-estimation of their limited brain power.

This entry was posted in Machine Learning, Miscellaneous. Bookmark the permalink.

2 Responses to The 2016 U.S. Presidential Election Predictions

  1. Steve Svelmoe says:

    I can take the predictor down to 1.

    Since television has taken over the election process (Nixon v Kennedy) American has elected the most charismatic candidate. Only one exception to that rule was Ford v Carter because Ford pardoned Nixon the country hated him for it.

    Steve Svelmoe
    Senior Software Engineer
    TriTech Software Systems |
    Direct: (858) 799-7381

    • Interesting idea and I think you might be correct that the more charismatic candidate has won, at least going back to the 1960 when TV took hold big time.

      1960 Kennedy (vs. Nixon)
      1964 Johnson (vs. Goldwater)
      1968 Nixon (vs. Humphrey)
      1972 Nixon (vs. McGovern)
      1976 Carter (vs. Ford)
      1980 Reagan (vs. Carter)
      1984 Reagan (vs. Mondale)
      1988 Bush (vs. Dukakis)
      1992 Clinton (vs. Bush)
      1996 Clinton (vs. Dole)
      2000 Bush (vs. Gore)
      2004 Bush (vs. Kerry)
      2008 Obama (vs. McCain)
      2012 Obama (vs. Romney)
      2016 Trump (vs. H Clinton)

Comments are closed.