The internet has been all abuzz about the results of a statistical model created by political scientist Helmut Norpoth that says that Trump has a 98%* chance of winning the general election against Hillary Clinton and a 99% chance of winning against Bernie Sanders. According to the New York Post, Norpoth’s model is “scary accurate” and has correctly predicted every presidential election except one in the past 104 years. We know this all must be true because this story broke via the SUNY Stony Brook’s student newspaper. Would students lie to us? I don’t think so. So start packing your apocalypse survival kit because in less than a year President Trump is going to be running the United States.

So how does this “scary accurate” model work exactly? Way back in the pre-Trumpocalypse year 2008 Kevin Drum at Mother Jones gave an explanation of how Norpoth’s model works:

Unlike every other forecaster on the planet, they don’t include any economic variables at all. Maybe the economy is great, maybe it sucks. They don’t care. Their model has two inputs: (1) how well you did in the New Hampshire primary, and (2) whether you’re an incumbent running for reelection.

Since this year has no incumbent, the only input is how well candidates do in the New Hampshire primary. I checked the website for the Primary Model to make sure they haven’t changed it since 2008. They do say that they’ve created another version of the model that also uses the South Carolina primary, but the news reports about the 98% Trump prediction were from before the South Carolina primary took place, so they got to that number using the old version of the model whose only input is the New Hampshire primary.

A model built only on the New Hampshire primary sounds like it probably wouldn’t work, but it does predict every election except one since 1912, so laugh all you want but it’s still “scary accurate” and still predicts a 98% chance that Trump will soon be President Trump.

Well, ok not exactly, but to understand why it may not be as accurate as it seems, you first have to understand how models work. A model is at its core a math equation with some inputs. When you are building a model, you create the basic structure of the model but most of it remains mysterious. You then train the model on existing data. In this case, Norpoth fed the model information on how all party nominees did in the New Hampshire primary going back to 1912 as well as who the incumbent was and then told it which candidate won each election. In other words, they gave it both the inputs and the answers. The model took all this data, processed it, then spit out an equation that predicts the likelihood of a candidate winning an election based on data from past elections. You can then plug in data about future or even hypothetical elections where it doesn’t know who the winner is and it will give you a prediction.

This primary model was created to predict the 1996 election, so we can assume that all elections from 1912 to 1992 were used for training the model. You can plug in information about all those elections back into the model and it will give you a prediction that is right in all but one case, but those elections were used to train the model so the model itself has a huge advantage in prediction of those elections and we can’t really use those to determine the true model accuracy. On their website they claim that in the five elections since the model was first made it correctly predicted the winner in all five. However, the only predictions listed on their website are those for the 2008 and 2012 elections.

According to the Primary Model website, the model predicted that in a 2008 a general election race between McCain and Obama would end in a statistical tie. In fact, Obama won the popular vote by 7 points and McCain didn’t even come close to Obama’s number of electoral votes, so the model failed to predict Obama’s win. You can be generous and say that the model predicted either party winning and therefore was still right since one of the two candidates did in fact win the election. The Primary Model website claims this as a win.

In 2012 the Primary Model predicted an Obama win against Romney by 6 points. Obama did in fact win the 2012 Presidential election so we’ll call this a win for the Primary Model.

Based on these two elections, we can say that the Primary Model is 50% accurate. Even if we trust that the other three elections were a win for the model, we’re looking at 80% model accuracy. Four out of five isn’t terrible, but hardly “scary accurate.”

The other problem with a model such as this one is that it’s merely looking for patterns using past elections. The model can only accurately predict future elections if they are similar to the ones that were used to train the model. Literally every election used to train the model was between two white, Christian male politicians. It’s possible they updated the model by adding the 2008 and 2012 elections, which means there could be two examples of an election between just two Christian male politicians in the mix. Regardless, the election this year is like nothing this country has ever seen. None of the five top contenders fit the mold of “white Christian male politician.” There has certainly never been a candidate like Trump,  a non-politician party outsider with fascist leanings who brags about the size of his dick during primary debates and who nonetheless is the likely party nominee.

In other words, this election is not like anything we’ve seen before and may not follow the same patterns we’ve seen before. For example, if voters in some states are more misogynistic or antisemitic than voters in New Hampshire, that could disrupt the pattern. If Trump is able to change his image over time by changing all his extremist positions that he’s used to win the primary into moderate ones, something which he’s recently started to do, that could change the pattern. If Sanders wins the popular votes but the superdelegates stick with a Clinton vote (something that is unlikely to happen but definitely not out of the realm of possibility, especially if the popular vote delegates come out to a toss-up), the patterns may no longer hold. If Trump wins the most delegates but not so much that he has a mandate, the GOP may refuse to nominate him, which could result in a breaking of the pattern. In other words, just because in the past when candidates got a certain number of votes in New Hampshire it could be translated into the number of votes they would get in the general election doesn’t mean that that pattern will always hold, particularly if something weird is going on with the election and this year definitely has an abundance of weirdness.

I’m not saying Trump is not going to get the Republican nomination or become president, just that it’s probably much lower than a 98% probability. Even so, it’s not that low. Head-to-head polls against Clinton or Sanders mostly show him losing but not always and not by that much, but head-to-head polls this early on just aren’t very accurate. We have no idea what will happen between now and election day. If someone like Bloomberg enters the race, which is starting to seem likely to happen, he could split the anti-Trump vote allowing Trump an easy win in November.

My point is, we can laugh at the 98% number, but don’t get complacent. Even if your candidate does not win the democratic nomination, whoever does is going to need your support. If even a small number of supporters of the losing candidate refuse to support the winner, it could swing the race in Trump’s favor. A future Trump presidency is not a joke but an honest to god possibility and if you aren’t yet scared you should be, but there are things you can do to help prevent this outcome.

First of all, register to vote if you are eligible. Check the voting laws in your state to make sure you have the required IDs. In some states even a drivers license or state ID may not be enough if it’s from another state or the address doesn’t match the one where you are registered to vote, so if you’re an out of state student or someone who moved recently, make sure you check on this. In fact, out of state students can choose whether to vote in their home state or the state in which they are attending school, so if you have a choice then you should choose the state most likely to have a close election. If you’re able to, donate money to the candidate of your choice. After primary is over, donate money to the Democratic winner whether it be Clinton or Sanders, even if it wasn’t who you wanted to win. Volunteer to knock on doors or make phone calls for the campaign, particularly if you’re living in a swing state. This stuff matters. This is the stuff that will win the election and prevent a President Trump from ever becoming a reality.

*For some reason a lot of websites are reporting Trump’s chances against Clinton as 97%, but since the actual number is 97.6% if you’re going to round to the nearest integer it should be 98% not 97%.

Featured photo by Gage Skidmore

Jamie Bernstein

Jamie Bernstein

Jamie is a data, stats, policy and economics nerd who sometimes pretends she is a photographer. You can usually find her at skeptic events in Chicago or on Twitter or Flickr. She also blogs about music at Notes From Chicago Music Underground.

Previous post

CNN #DemDebate Liveblog: Election Boogaloo

Next post

Peeple: Harassment is Destiny?

10 Comments

  1. March 7, 2016 at 1:28 pm —

    Well considering that the turnout for the Democratic Primaries is lower that the GOP Primaries; that the Sanders draws in more independents (myself included/ Lean way towards Jill Stein). I would argue that Clinton will not prevail against any GOP candidate she will face.

    We either continue are slow (DNC third way) slide into despair, or one quick (GOP) drop.

    Sanders, I’ll argue again, is one that could turn that around.

    • March 7, 2016 at 6:36 pm —

      Well, my solution is to recruit like-minded individuals to the Democratic Party. Then they’ll recruit like-minded individuals. And so on. And so on. And eventually, we will outnumber the Joe Lieberman wing of our party, and we can make them very uncomfortable until they follow Joe’s cue, and we can wave and say “Bye, Felicia!”

      (Kossacks will get the reference.)

    • March 9, 2016 at 9:10 am —

      “Well considering that the turnout for the Democratic Primaries is lower that the GOP Primaries…”

      Is primary turnout a good indicator of general election turnout? Seems to me that in a year when the GOP is apparently fracturing in a divisive and hotly contested primary, we’d expect to see higher turnout, as the base are motivated to vote by the internal conflict. Conversely, a lot of Democrat voters, while they may have a bit of a preference between Sanders and Clinton, generally seem to feel pretty good about both candidates, suggesting there would be less motivation to turn out.

      If this is the source of the discrepancy, it doesn’t seem like it would translate to general election turnout: democrats who were more or less indifferent between their primary options will be highly motivated to vote against Trump (or Cruz), regardless of who their nominee is; whereas, with all of the division and divisiveness in the GOP, voters who are opposed to their eventual nominee (there are factions both strongly opposed to and strongly in favor of Trump) may be tempted to stay home (whether out of apathy or protest) or vote for a 3rd party candidate, or a write-in.

      • March 9, 2016 at 6:54 pm —

        Primary turnout has no consistent bearing on general election turnout. If we look historically, the 1988 Democratic primary had the highest turnout of any primary ever at the time (it was later overtaken by Obama’s run in ’08). But that ’88 primary turnout didn’t translate into a win for Dems come November – as evidenced by the fact that we never had a President Dukakis.

        Now of course, you can also point to high Dem primary turnout in ’08 that led to an Obama win, and low GOP turnout in ’12 that led to a Romney loss… but nothing indicates that’s anything more than a correlation. And as we all know correlations =! causation.

        There have also been studies that suggest more polarizing elections lead to higher voter turnout (this one springs to mind). That would certainly help explain while GOP turnout is so high right now. And while Bernie certainly has a passionate fanbase, it’s not quite as inspiring as the ’08 election where Obama/Clinton were completely neck in neck.

        Of course, as the article points out, this is an election quite unlike anything we’ve ever seen before. Everything is up in the air, because this cycle is SO batshit.

  2. March 7, 2016 at 2:29 pm —

    So 1996 was 104 years ago huh?

    Our failing education system everybody.

    • March 7, 2016 at 6:34 pm —

      I’ve heard people say “36 hours, a full two and a half days” before. And there was this one pundit who said that his previous prediction of a 30% ceiling meant the tens (or tenths, depending on how you think about it) digit would never rise above 3. And then it did.

      • March 8, 2016 at 10:20 am —

        Well, the Big Book of Multiple Choice says that Sunday is three days after Friday so you can see that humans have never been too good with the numbers.

    • March 9, 2016 at 8:50 am —

      “Norpoth’s model is “scary accurate” and has correctly predicted every presidential election except one in the past 104 years.”

      That’s the line that got me too. Truthfully, the first thing I thought when I read that was “how old is Helmut Norpoth?”

      We don’t predict past events, but we can look for patterns in them. However, if you assume those patterns predict future events you are assuming correlation/causation.

  3. April 5, 2016 at 5:08 am —

    They don’t even have meaningful primary results for every election. An incumbent President usually doesn’t have a seriously contested primary in New Hampshire (the only recent exception I can think of is Gerald Ford). And prior to 1968, the modern primary system didn’t exist. The “model” is little more a voodoo election heuristic, of the type skewered by XKCD: https://xkcd.com/1122/

Leave a reply