It is less common than it should be for pundits to lay out predictions in a falsifiable way. Many are simply overconfident about what can be guessed about the future. Others couch their predictions in such slippery language that they are hardly predictions at all.
Almost four years ago I gave Donald Trump an 83% chance of completing his first term as president of the United States. As of a few hours ago he has achieved this, albeit with plenty of controversy along the way. While it is tempting to bask in this apparent success, this seems a good moment to assess the value of such forecasts.
Trump’s victory in the last presidential election shocked many Americans. Pundits had been writing him off throughout the campaign, and on the eve of the election many thought Trump’s chances were slim to none.
Perhaps the most infamous example of this was HuffPost, which gave Hillary Clinton a 98.2% chance of taking the presidency in 2016. The most guarded forecast came from FiveThirtyEight, which thought that Trump had a 28.6% chance of winning the electoral college, which ultimately decides who becomes president.
To be fair to HuffPost, a 1.8% chance is well above zero. It is often right to give an event that happens a low probability of happening. Aeroplanes almost never crash during a flight, so any given plane you get on is likely to land safely. But every so often a flight will be the outside chance.
Even so, in the case of the American presidential election in 2016, most pundits were overconfident of a Clinton win. Many felt that Trump was so disgraceful that he couldn’t win.
Such logic is intuitive, but flawed. Anybody who gets nominated as a presidential candidate for a major American party has a decent chance of winning the election, much as anybody who gets nominated for a third party has a negligible chance. These general rules are much more significant than any individual’s character.
Similarly, the odds of any president finishing their first term are more general than specific. Most presidents do, and this is true however much you like or dislike them, or what political weapons you believe are being arrayed against them.
As such, I thought it likely Trump would finish his first term. But I also thought he had a decent chance (17%) of not finishing.
Partly this was because he was old, facing legal complaints, and serving as president during a fraught time in American politics. Mostly it was because a decent portion of American presidents have failed to finish their first term.
As I said in March 2017:
With five previous presidents failing to complete their first terms out of 43, we can calculate that an elected president has an 11.6 percent chance of joining their ranks.
Joe Biden has become the 46th president of the US in official terms, although this is because Grover Cleveland is counted twice for serving non-consecutive terms. Of the 44 presidents that now proceed Biden, still only five failed to complete their first terms. Using the base rate, Biden has a 11.4% chance of not completing his first term, much like Trump’s equivalent base rate.
Looking at local factors, many of those that led me to revise the likelihood of failure up for Trump apply to Biden. The new president is 78 years old, roughly the life expectancy of somebody born in the US today. As a rich guy with great access to medical care he’s likely to live longer than the average person, but the presidency is a stressful job and at Biden’s age people often die.
America still looks fractious even if some establishments got their preferred candidate. The events in Washington, DC, earlier this month and widespread scepticism among Republican voters about the integrity of the election can only raise the chances of an assassination attempt against Biden.
In his favour, the new president looks unlikely to suffer impeachment or other legal difficulties. It’s striking that Trump leaves office as he entered it, with the threat of condemnation by the Senate. Biden faces no such risk for now.
Considering the above, I give Biden a 15% chance of not completing his first term. That it’s not that different from Trump’s odds should prove an illustration of how forecasting should work, at least if you believe Philip Tetlock.
Image credit: Joe Biden, October 2015 by DonkeyHotey