Statistician Nate Silver gave Donald Trump a 57.5 percent chance of winning the election if it had been held on Monday after a post-convention bounce in polls, according to fivethirtyeight.com.
Yet, just a month ago, he had Hillary Clinton winning the election with an 80 percent probability. Then, Trump had only a 20 percent chance of winning.
Well, apparently a lot can change in a month.
Like, the polls, which have dramatically turned around in the wake of Trump’s successful nominating convention in Cleveland, Ohio.
Silver’s model is based pretty much on the latest polls. So, if the polls still have Trump ahead by, say, Labor Day, one expects he’ll have to call the election for Trump.
Which, would be unbelievable.
Last fall, Silver only gave Trump only a 5 percent chance of winning the Republican nomination. What went wrong?
By May, as Trump was locking up the nomination in Indiana, Silver admitted that “our early estimates of Trump’s chances weren’t based on a statistical model.”
Instead, “they were what we [call] ‘subjective odds’ — which is to say, educated guesses. In other words, we were basically acting like pundits, but attaching numbers to our estimates.”
Was Silver just making up probabilities against Trump, passing it off as real statistical research on television and in national media outlets — with the intent of engaging in advocacy?
Either way, Silver gave him low odds to win the nomination — and it was used against Trump in the campaign by #NeverTrumpers in an attempt to deny him the nomination. Silver’s admission took him from the role of neutral statistician reporting what he sees to something else.
Ironically, Silver’s bad Trump prediction ultimately had the effect of discrediting #NeverTrump and gave Trump’s campaign an underdog, do-the-impossible narrative to run with. Did that help Trump?
Once again, that is coming into play, where everybody has been told a million times over that Trump has no chance, again quoting Silver. Again, just a month ago, Silver had the odds of a Clinton win in November up to 80 percent — before a single national convention had even introduced the candidates to the vast majority of Americans who are just tuning in.
Perhaps if there is a lesson to be learned about forecasting elections, it is that one should just make a single prediction — and stick to it.
Like Professor Helmut Norpoth of Stony Brook University. He uses the strength of victories in the early primary states, plus where we are in the election cycle, to forecast the general election outcome.
First, according to Norpoth, the American people have tended to award the out party the White House after they have been out of power for 8 years, with certain exceptions, like 1988. That gives the odds to Republicans right off the bat. Next, this year, Norpoth used New Hampshire and South Carolina, both of which Trump won.
Using both, Norpoth confidently forecast Trump would defeat Clinton with 52.5 percent of the popular vote in a two-way race — with an 87 percent probability certainty.
“As a rule, the candidate with the stronger primary performance wins against the candidate with the weaker primary performance. For elections from 1912 to 2012 the primary model picks the [popular vote] winner, albeit retroactively, every time except in 1960,” Norpoth wrote on his website. Caveat: In 2000, he accurately forecast the popular vote winner but not the electoral college winner.
There won’t be any updates to Norpoth’s model. The primaries are over. Norpoth made his prediction and he’s sticking with it. It will either be right or wrong.
Still, in the Austin-American Statesman on July 8, Norpoth appeared to be hedging, stating, “If the election were tomorrow I certainly would have to say Hillary is going to win and Trump is going to lose,” but he stood by his prediction, adding, “the election is four months away. That’s quite a while.” Neither conventions nor inter-party debates had taken place yet, likely determinants for undecided voters.
In other words, a lot can happen, so what’s the use of a constantly changing forecast throughout the year?
On Reddit, Norpoth commented on Silver and other polls-based models in March, noting, “I wouldn’t trust any media, pundits or current horse-race polls to predict this election right now. Polls are of little use until about September. Even then, some get it wrong. As in 2012 or 2004.”
To be certain, polls did understate Trump’s actual voter turnout in Indiana, Pennsylvania, Maryland, Connecticut, Rhode Island, Delaware and New York during the primaries, the last competitive races that Trump was a part of.
And now, the polls have given Trump a bump after the Cleveland convention.
So, right now Silver has Trump ahead, if the election were held today. But if as Silver contends, science consists of “observing the world, formulating hypotheses about it, and making those hypotheses falsifiable,” he must take down the probability indices for impossible outcomes such as hypothetical elections taking place in July that nobody can possibly contest let alone win. How is anyone to prove he’s wrong?
The election is not being held today. Plus, the polls will shift. What if the polls show after Labor Day the election is too close to call? What prediction will be made then? Next week, Democrats might get a bounce, and then suddenly Clinton will be likely to win. Making a new prediction every week is silliness.
The only poll that matters is in November.
Ask Eurocrats in the United Kingdom how reliable the polls were showing that Britons would vote to stay in the European Union. 2016 may be the year that the polls broke.