01 electoral college protest 2016
CNN  — 

If the 2020 election happened today, President Donald Trump would have, at best, a 1 in 5 chance of beating former Vice President Joe Biden, according to the three major models predicting the outcome.

That 21% chance of a Trump victory comes from 538’s election model, which simulates the election 40,000 times and produces an aggregate of likeliest outcomes based on a variety of factors including historic results, polling and a variety of other variables.

The other two election models most politicos keep tabs on suggest the incumbent has an even smaller chance of winning. The Decision Desk HQ model gives Trump a 17.8% chance of securing a second term in 34 days’ time. The Economist model gives Trump a meager 13% chance of winning.

Which is pretty stunning stuff! And honestly, it would be getting lots more attention if 2016 hadn’t happened. After all, modelers – especially 538’s Nate Silver – became hot commodities after the 2008 election, in which they accurately predicted Barack Obama’s convincing win over John McCain. And Silver proved himself again by nailing the results in all 50 states in 2012.

But 2016 did happen. And the models were wrong.

On November 8, 2016 (Election Day), the 538 model gave Trump a 28.4% chance of winning. The New York Times “Upshot” model pegged his chances at 15%. The Huffington Post model gave the billionaire businessman a minuscule 1.7% chance of victory.

What that failure proved is that a model is only as good as the assumptions (and the data behind those assumptions) that you put into it. Trump was able to find voters in three key states – Pennsylvania, Michigan and Wisconsin – who the models missed. So he won.

But simply because election models missed the Trump phenomenon the first time around doesn’t mean we should ignore them entirely. The models are updated to reflect the realignment Trump set in motion in 2016. The likelihood, then, of missing some sort of hidden pro-Trump factors in the electorate is much less.

The Point: In politics, we have a tendency to throw the baby out with the bathwater: The election models missed Trump’s 2016 victory and so, therefore, we should totally ignore what they are saying about 2020. No!  Of course the models could be wrong again. But there’s a much higher likelihood that they’re right.