This is not 2016, it is not a difficult election to forecast. I am predicting that Trump is re-elected for a second term with 95% confidence.

In 2016 I did say, repeatedly, that Trump would win. However, truth be told 2016 was a difficult election to forecast. We had two candidates that were profoundly unpopular with their bases. While I predicted Trump would win, honestly a lot of good fortune went his way. Stuff that no one could have foreseen.

This time around it is much simpler, and whilst the pollsters will say they think Biden is winning, the opinion polls are not an accurate proxy for November 3.

So the first thing I’ll note is that, no, Trump is not a “terrible” president. He’s actually a conventional one. This is the single point that most forecasters have overlooked. Yes the US is more partisan than ever – but that isn’t because of Trump. He has also handled himself well in office overall – within 18 months in the job Obama had visibly aged, whereas you’d probably have to go back to Johnson to find a president that aged less than Trump over his first term.

What the models predict

There are a few different models to look at. I’ll start with the opinion polls and then run through some of the others that I think are better indicators of who will win office on Nov 3.

Opinion Polls

The opinion polls all point to a Biden victory, most with around 80-90% confidence. The opinion polls however are problematic – it is more difficult than it has ever been to get reliable random samples that are truly representative of the voter base. The response rate to telephone polling in the US as of 2016 was just ten percent. Let that sink in for a moment. 90% of people called are not recording responses. The non-participants are not properly accounted for, yet there is evidence for the shy Tory factor, meaning it can no longer be credibly asserted that all political views have an equal chance of non-response. Getting samples from representative locations is difficult because at least 50% of the population now has mobiles only, no landlines, and working people often have a second work mobile as well meaning they will be over-sampled. Getting accurate location data for mobiles is difficult, contains more errors, and people often lie about their location when asked in surveys (I do). We don’t understand the true error margin of political polls. Pollsters don’t really understand turnout. If the turnout is only going to be 53% you need a pretty good method to work out who the 47% of non-voters will be, and what their political views are, otherwise your whole poll is worthless. And finally, the polls were wrong in 2016.

One of the most honest responses from a pollster following the 2016 presidential election was by Natalie Jackson (emphasis added):

Our model predicted that Democrat Hillary Clinton had a 98 percent chance of being elected. … The model relied completely on polls. No adjustments, no ‘fundamentals’ (which include other things, like presidential approval or economic indicators, that help predict elections). Just polls. … The model structure wasn’t the problem. The problem was that the data going into the model turned out to be wrong in several key places. … It gutted me to realize I had been wrong. I truly did my best and read the data the way I saw it. No one wants to feel like this ― to be so utterly and publicly mistaken. People on Twitter have been calling for me to be fired. But what happened is done. What I can do now is learn from it, and be open and honest about the process. … Polls alone probably don’t make a reliable forecast model.

Also see this follow-up.

The Primary Model

Prof. Helmut Norpoth’s Primary Model correctly predicted the 2016 election, with an incorrect prediction for the popular vote. The premise is really simple, Prof. Norpoth only looks at the candidate’s performance in the early primaries, nothing else, and finds that their performance is a strong indicator of who will go on to win the presidential election. Through 1912 to 2016 the model correctly picks the winner 25 out of 27 times. Norpoth’s forecast was released as of 2 March 2020 and gives Trump a 91% chance of being re-elected for a second term. Trump has smashed the record for the number of votes an effectively unopposed incumbent Republican president got in the primaries. He got 18 million votes, compare that to the less than 8 million that GW Bush got. Biden did not do well in the early primaries at all.

The Fair Model

Prof. Ray Fair’s Fair Model also correctly predicted the result of the 2016 election. The Fair Model says incumbent presidents running have an advantage, that voters like change after 2 terms, there’s a slight persistent bias favouring Republicans, and that the state of the economy affects the incumbent party’s vote. As of 30 Jan 2020 it was forecasting a win for Trump. However Prof. Fair declined to continue as of April with his model this election, saying that it has not been tested for the effects of a pandemic and adjusting it would require guesswork. Calculated by PolyVote, as of August it shows a narrow win for Biden.

The Knox Model

Developed just for the election by Australian economist, chief economist of Morgans, Michael Knox. The Knox Model is an adaptation of the Fair Model. Knox notes that he has been interested in the Fair Model for some time (he also shows familiarity with the Primary Model, which is interesting because it has nothing to do with economics), and he says he’s adapted it in the past for use with Australian elections. He says that Ray Fair has been unable to produce a forecast because of the variability in GDP per capita this year. So he set about looking at a broad range of economic variables to identify what may provide a better estimate. The Knox Model says that an incumbent president gains a swing of 5.15% of the popular vote when he stands for re-election, and that a 1 point decline in the federal fund rate generates a decline of 62 basis points (0.62%) of the popular vote. The model therefore predicts Trump should win with margin of 5.76% of the popular vote. Factoring in the error margin, Knox finds that Trump has an 89.6% chance of being re-elected.

The 13 Keys to the White House

Prof. Allan Lichtman developed this system, which as of 2016 predicted only the winner of the public vote: “The keys to the White House focus on national concerns such as economic performance, policy initiatives, social unrest, presidential scandal, and successes and failures in foreign affairs. Thus, they predict only the national popular vote and not the vote within individual states.” (Lichtman 2016 & 2020 p. xi). The system therefore incorrectly predicted Trump would win the popular vote in 2016, despite Prof. Lichtman claiming for the past 4 years that he correctly predicted the result. In retrospect, one of the keys, key 4, turned true and therefore did predict Trump to lose the popular vote. I digress. For this election, the 13 keys predict a Biden victory. How do they work? Well each “key” is a predictor of whether the incumbent party holds office or not, and when six or more keys are false the incumbent party loses the popular vote. It’s a good forecasting system that has shown remarkable reliability.

Stantic’s Big Data Analytics

Prof. Bela Stantic correctly predicted the results of the 2016 Brexit referendum, the 2016 US Presidential election (49 out of 50 States), and the 2019 federal election (Aust.). In all of these cases the opinion polls predicted the opposite result. He correctly predicted the 2017 Queensland State election result as well, and he was famously incorrect about the 2017 same sex marriage survey (and then did a post-mortem on why). As of 21 September 2020 his system is predicting Trump will be re-elected, but narrowly and by an even smaller margin than in 2016 (meaning Trump is forecast to again lose the popular vote). Prof. Stantic’s system works by scraping millions of publicly available social media posts, such as Twitter, and then analysing the content and assigning value to the posts. Prof. Stantic’s final prediction won’t be published until Election Day at earliest, he shares his final predictions privately with journalists who he trusts not to publish until Election Day because he doesn’t want to unduly influence the results. The premise of the system is that people are more honest when talking to their friends on social media, and that you can more accurately gauge enthusiasm and opinions from this content. It’s highly scientific, Prof. Stantic just scrapes the data and lets his maths equations assign value to the Tweets and other social media posts he feeds it.

My take

There’s actually value in all of the above viewpoints. The polls – though they undoubtedly have a bias due to sampling error – are helpful. The identification of historical indicators is helpful. Analysis of how the economy affects the presidential election is also helpful. I agree with Lichtman that a presidential election is primarily referenda on the incumbent party. I think Helmut’s model is really insightful because it helps to identify enthusiasm for the candidate without opinion polls. The incumbency effect calculated by Fair and Knox is also very useful. Looking at this from a broad range of views I think is advantageous and avoids the potential pitfalls of narrow focus on a single set of data.

I look at US presidential politics and see a highly partisan and divisive system. I don’t see Trump as an aberration at all, and I think it’s a serious mistake to treat him that way. I see Biden as a lame duck candidate. I think for the Democrats to have a chance they needed to pick a candidate that did not have 47 years of political baggage, they needed to pick someone not involved in the Obama administration, and they needed to do it without the messy and politically damaging primary process. As the Democrats have been pulled further left, they have left behind the working class. They left them behind decades ago in the opinion of Noam Chomsky. These are the voters that Obama ignored and sidelined, that Hillary Clinton called a “basket of deplorables” who are irredeemable. I see the Biden campaign trying to win over Trump voters instead of the pool of Democratic voters who stayed home in 2016. I honestly don’t think the Democrats have understood 2016, they’ve certainly shown no contrition for their errors, and I see the campaign making the same errors all over again.

Ultimately I see a system that is won and lost on turnout. Which side is more excited to vote for their candidate? I think this is an extremely important question because I believe it directly affects turnout. I think Trump will ultimately be judged on his record, not his personality, and I see a record that is positive economically and positive in terms of foreign policy. I see a presidency that has very clearly been stronger on foreign policy than the previous one. I also think the first term Trump presidency expanded and strengthened both manufacturing and mining. Yet we had Biden promising to ban fracking in the primaries, and still promising to ban it on all US federal government land. He doesn’t seem to be running on a positive vision.

The worst US President in my lifetime has been GW Bush. As others have noted, nothing that Trump has done has even come close to the five worst atrocities of the first term of the GW Bush presidency. (1.) He started a war in Iraq based on an outright lie, that killed at least 500,000 innocent civilians, that destabilised the entire region and created a global refugee crisis. (2.) He opened and maintained illegal prisons in Guantanamo Bay and Abu Ghraib in Iraq where his administration held prisoners indefinitely and sanctioned torture contravening the Geneva Conventions. (3.) He sanctioned illegal NSA spying. (4.) He created the 2008 global financial crisis by deregulating the housing and financial markets (the deregulation happened in his first term, the consequences happened in 2008). (5.) Bush v. Gore weakened the US presidential electoral system. Despite these horrific sins in his first term, which okay I’ll grant you the housing market collapse hadn’t happened yet but the other four certainly had, Bush was re-elected for a second term.

This is not to say that Trump is a saint, it’s merely to put his record into context. If GW Bush can be re-elected with his first term record, there is absolutely no reason that Trump can’t be.

All of the “noise” aside, I think the most important issues to voters is, as the opinion polls find, the economy. Other issues are important, but less so. I am not seeing strong economic leadership from Biden, or a strong belief in his economic credentials. Foreign policy also matters, and I give it easily to Trump while noting his extreme disrespect of veterans will cost him. For the other domestic policy issues, I while the polls find that Biden has a slight edge, I don’t think they clearly favour either candidate. And from there I rely on the historical indicators: Biden is too old, and he has too much political experience (36 years in the Senate and two terms as VP in the White House).

In 2016 while I forecast that Trump would win, the mistake that cost Clinton the election in 2016 was to label Trump voters as irredeemable deplorables. Had she not done that, there was a good chance she would have won. I viewed her as a toxic candidate, but nobody could have predicted exactly how her toxicity would manifest. However note that damaging stuff came out about Trump as well, and that this time around even with a more serious and more credible allegation against Biden (Tara Reade) than against Trump in 2016, it appears to have had almost no effect whatsoever on voters. While Stantic’s forecast does have me concerned, this time around I think the circumstances are clearer and it’s a much easier election to forecast. It’s true that no one can predict the pandemic effect on the election, however I’ve made the assumption that it’s a wash. I could end up being wrong. We also have a massive increase in postal voting, and I think that also favours GOP because I see there as being less opportunity for motivated voters to bring along extra voters to the poll who would otherwise sit out the election, increasing turnout more for GOP.

In conclusion I’m confident of a Trump victory, I think he will probably win the popular vote and win the presidency in a landslide this November. I give Biden a 5% chance of victory, with effectively no chance of a landslide. If I’m wrong, and Biden wins, it means that there is something seriously wrong with my methodology. Short of something really disruptive and obvious to everyone, like Trump becoming incapacitated between now and Nov 3, I’ll have to come back and explain why I was wrong.