Opinion polling for the upcoming United States Presidential election consistently puts Democratic nominee Joe Biden far ahead. But does this mean we are safe from a Donald Trump re-election victory on November 3rd? No – a number of key polls are frighteningly similar to how they were in 2016.

It’s easy to wager the charge at polling companies that they have become worse at predicting the outcomes of elections from the second half of the 2010s onwards – almost no pollster predicted Brexit or Trump, Theresa May’s predicted slim majority in 2017 was actually a hung Parliament, and in 2019 Boris Johnson handed Labour their worst defeat since 1935 as predictions for only a moderate Conservative majority circled. However, this problem, while it became apparent in 2016, has been brewing for a while. You only need to look at the front page of the Sun from the 2015 election, which predicted both parties would get 37% and thus a similar number of seats, or any polling from the 1992 election, in which even the exit poll was wide of the mark, to know that pollsters’ jobs of predicting the future have never been easy. Where does this leave us with this year’s election? Well…

The New Statesman have for the first time in history created their own prediction model for the Presidential election. Last updated on 21st September (at the time of writing), it predicts that Biden has an 80.3% chance of winning, compared to Trump’s 19.7%, with the most likely electoral college outcome being 327 to Biden and 211 to Trump. This may make it seem like a done deal for Biden, as his probability absolutely dwarf’s Trump’s. However, this actually represents Trump’s chances of winning as greater than 1 in 6, or greater than the chances of rolling a 1 on a normal die. If you were approached by a person with a die and were told that the fate of the most powerful country in the world rests on you not rolling a 1, would you really not be incredibly distressed about the possible outcomes?

The New Statesman model also gives probabilities for outcomes in each state, which provides greater insight into how the electoral map may look, as well as evidence for their 80.3% figure. At first glance, their state-by-state probabilities make their 80.3% look confusing, as to win the electoral college Biden needs to win (in order of decreasing probability of a Democrat victory) all the certain states as well as Oregon (99%), New Mexico (97%), Virginia (97%), Colorado (93%), Maine (93%), New Hampshire (85%), Nevada (85%), Minnesota (81%), Michigan (80%), Wisconsin (75%) and Pennsylvania (75%). If these probabilities are treated as independent, then to find the probability of winning all of these states, we would multiply all of these probabilities together, which gives only a 21% chance of a Biden victory. However, the reason this is false, and contrary to the model from which the probability data was drawn, is because the winning of states is not independent. A 75% chance of a Democrat victory in Pennsylvania means that there is a 75% chance that the national swing towards the Democrats will be great enough for them to win Pennsylvania – in this outcome they will almost certainly win all of the states for which the model predicts higher than a 75% chance of a Democrat victory. So this suggests that Biden’s chances of winning are actually 75%. This, coupled with the nuances of the national swing having small but significant regional variation, and there being other combinations of states Biden could win to win the election that have a non-zero probability of occurring, links the state probabilities with the 80.3% chance of Biden winning that the model predicts.

The New Statesman is of course not the only organisation to give Biden a high chance of winning. FiveThirtyEight, a poll aggregator, used polling data from a range of sources to simulate the election 40,000 times, and gave Biden a 77% chance of winning overall. YouGov, who are predicting a slightly more pessimistic scenario than the New Statesman as they say that Florida is neck and neck (New Statesman gives a 65% chance of Democrat victory), still give Biden a healthy 8.5% point lead in opinion polling (they do not provide a probability of Biden or Trump winning). Indeed, only one major poll, conducted by Pulse Opinion Research, predicts a Trump win. So why is this a cause for concern? These polls are remarkably similar to those conducted between September and October 2016, in which almost all polling companies agreed Hillary Clinton had a 75%-90% chance of winning, and placed her variously 5% to 14% ahead in opinion polls.

Indeed, of the very few people who did predict a Trump victory in 2016, not many can claim to be more accurate than historian Helmut Norpoth. He not only predicted Trump’s victory in 2016, but also correctly predicted five of the past six Presidential elections. Furthermore, applying his methodology to historical elections delivers the correct result in 25 of the 27 most recent Presidential elections. Norpoth’s model gives Trump a 91% chance of winning.

While polling companies can sample opinion, and even predict the electoral college, what they cannot do is predict the outcome in the event of disputed electoral college votes. This can occur in the event of a tie (which FiveThirtyEight says has a 1% probability), if the Electoral College is won by the candidate that lost the popular vote, or for a multitude of other reasons. For example, in 2016, pollsters correctly predicted that Hillary Clinton would win the popular vote, but she only won it by 2%, which, thanks to America’s Byzantine system to decide the outcome of a Presidential election, handed the Electoral College to the Republicans. This led to a number of faithless electors – members of the Electoral College who refused to vote the way their state voted – voting either for Clinton or for some third party candidate when they should have voted for Trump. Another example is the 2000 Presidential election, where the Republicans were just 537 votes ahead in Florida, and whichever candidate won Florida would win the very precariously balanced Electoral College, and therefore the Presidency. A vote margin that small required an automatic recount by hand under Florida state law, but the Republican Party sued and got the Supreme Court to state that the original count was the only valid count, thus granting George Bush the state and the White House. None of this could possibly have been predicted by opinion polling companies.

We also cannot rule out other novel methods by which the result might be influenced. Rigging the vote in favour of the Republicans could occur, because that’s what Donald Trump keeps saying he is going to do, in particular he has encouraged supporters to vote multiple times and to obstruct the voting of others. But also, entire states could direct their Electoral College votes to the winner of the national popular vote regardless of the outcome of the vote in that specific state under the terms set out in the National Popular Vote Interstate Compact. Furthermore, particularly if there are allegations of election rigging, there could be very large amounts of faithless electors, acting either on the orders of state governments or of their own volition to not implement the result their state officially delivered at the Electoral College. And then it must be considered what will happen if Trump does not accept the outcome of the election in the event of a Biden victory. He has refused when questioned to commit to respecting the result of the election, so this is yet another major issue with predicting what will happen in this election.

To conclude, polling organisations are in the business of predicting the election alone. Maybe they will be right, but many indicators point to a 2016-style upset currently. But, perhaps more importantly than that, polling organisations cannot predict what may happen in the days and weeks following the election that could affect who is actually sworn in on 20th January next year. Questions pertaining to the outcome and legitimacy of this election may well arise if things continue on their current trajectory. These questions are answered in a best case scenario in the courts, and in a worst case scenario on the streets.

*Jack Harrison is a political blogger and student at the University of Cambridge. He was the author of the blog Minority 2017 from 2017 to 2019. He can be found on Twitter **@JackH1010**.*