Did they really get it wrong? Surely after Brexit the pundits wouldn’t be that stupid. This week, we look back at the forecasting from the last 6 months on who would be the 45th President.
The hot
Helmut Norpoth of Stony Brook University predicted Trump as President. His statistical model uses each candidates’ primary results (as well as other data) to predict the winner. Because Trump beat Cruz by a bigger margin than Clinton beat Sanders in the primaries, Norpoth’s system predicted Trump to win. Back in February, Norpoth said there was a 97% chance of Trump beating Clinton and a 99% chance of beating Sanders. Everyone thought he was mad. His model, which he created in 1996, has correctly predicted all but one US election since 1912 (the exception was Nixon’s marginal win in 1960).
… and the not so hot
With similar confidence, but just before the election, HuffPost claimed that “Clinton has 98% chance” of winning. The US’s most respected forecaster, Moody’s Analytics, also said Clinton would win, predicting a clear win with 332 electoral votes. Their system looks at various real-life factors (e.g. household personal income, house price growth, fuel prices) and deduces that, if people are happy with their lot, then they will re-elect the incumbent party. They have had been correct every election since they built their model in 1980.
Those on the fence …
By the time election day arrived, there were only four points between the two candidates. It was hard to call. The wise acknowledged this without trying to be heroes. For example, Allan Lichtman’s system, which uses 13 true-false questions such as “has the party in power done well on foreign policy” (where fewer than six false answers indicates incumbent party will remain in power; six or more false answers predicts a win for the challenging party), said that his model was “still uncertain … but it’s looking grim for the Democrats.”
… and the downright ridiculous
Let’s not forget superstition and non-causal outcomes. Back in June, Olivia Lace-Evans reported some of the more humorous theories:
If your surname ends with the letter ‘N’, then you have a better chance of becoming president – so Hilary Clinton would beat Trump. At least Trump’s win has put that theory in the trash can!
But one superstition still stands undefeated: if the country hosting the Olympic games in the year of election has not hosted before, then the incumbent party will lose. Brazil had not hosted the Olympics before so it predicted a Trump win.
The reality
So, did the pundits get it wrong?
Let’s remind ourselves what probability means: “the relative possibility that an event will occur, as expressed by the ratio of the number of actual occurrences to the total number of possible occurrences.”
A European roulette wheel goes from 0 to 36, which means there is a 97.3% chance of it not coming up on zero (36 times out of 37). If you spin a roulette wheel 1,000 times, you will (on average) get a zero on 37 occasions. That doesn’t make the 97.3% probability of it not being zero wrong when it happens.
So, apart from those who gave 100% confidence of a Clinton win, we can’t say any of the predictions were ‘wrong’. For example, Nate Silver’s FiveThirtyEight said the probability of a Clinton win was 71.4%. It ran 20,000 simulations per day and Clinton won 14,280 of them (71.4%) while Trump won 5,720 (28.6%).
It just happens that in the spin of the wheel that took place in the US on 8 November 2016, the ball fell on Trump. Unfortunately for Clinton, there’s no second spin of the wheel.