Let’s be fair here and try to understand what Silver actually does. He technically does not make any predictions, he merely calculates probabilities based on currently available data and past applications of such data.
As such, it is incorrect to call Silver wrong if the most probable outcome fails to occur. For example, if you flip a coun that to the best of my knowledge is a fair one three times, I would tell you that there’s an 87.5% chance it will come up heads at least once. If you go ahead and flip the coin three times and get three tails, I was not wrong. I didn’t predict anything, I merely gave a probability.
I am really only wrong if I miscalculated the probability. In fact, if I told you that you had a 75% chance of gettting at least one head and you did, my statement would be wrong - the true probability is 87.5%. The correctness of a statement of probability is independent of the actual outcome (except if the stated probability is either 0 or 100% of course).
In the specific cases discussed here, it is logical to credit Silver with at least a reasonable accuracy in his probabilities. Be honest, when the Falcon’s took a 28-3 lead in the third quarter, how many of you would have agreed to bet a significant sum of money that NE would come back and win the game? I suspect that the grand total of people reading this who would have bet say $1000 on NE at evev up odds at that point is zero. That comports rather well with Silver’s probability estimate, doesn’t it? The reality is that Silver’s low estimate of NE’s chance of winning was spot on; that WAS a very improbable comeback.
As for the election prediction, he used his only available data, poll results, to come up with an estimate of Trump’s chances. That poll data showed Hilary with a 4-5 percent lead over Trump. Assuming that data to be correct, it was perfectly reasonable to believe Hilary was more likely to win than Trump. Whether those chances were 70%, 60% or some other value is a debatable question. What is not debatable is that Hilary was certainly a favorite. The pre-election betting markets reflected that - most of the money went toward Hilary.
It should also be noted that Silver did not use bad information in his model. Despite the outcome, the polls were not wrong. The results were within the margin of error of the polls. Trump lost the popular vote by about 2%, which is within the margin of error of most polls.
What was wrong was the interpretaion of polling data by most media outlets, not the data itself. The media assumed that the polls reflected voter preference PERFECTLY rather than approximately. If Hilary actually had won by 4-5 points in the popular vote, she likely would have won. They also forgot the fact that while the popular vote usually is a good predictor, the electoral vote is determinative. The polling companies share some of the blame here as well; they assumed that states like MI and WI were not in play and therefore polling data was limited in these states.
Of course Silver’s model is not beyond criticsm, but to just write him off as wrong is overly simplistic. For what he does, he is pretty good at it. The NE comeback was a VERY unlikely event. Trump’s win was far less improbable, but Silver’s model told us that.
To knock Silver for his prediction is to take away from the greatness of the Patriots victory.
Which I basically agree with by the way.
The most accurate probability data I saw on Election Night came from the New York Times, of all places. They basically had dials of each battleground state and the race overall that changed as the night went on based on how many votes came in from each state and from those results, calculated how the remaining votes (from each precinct and county) would come out. From what I could tell, those dials were fully attuned to actual incoming data with no human intervention or bias.
With uncanny accuracy, those NY Times dials successfully predicted the outcome hours before the networks called the race for Trump. Matt Drudge started linking to them early that evening on his main page. It was there that I saw that the needles were moving in Trump's favor in states like OH, FL, PA, MI, WI and NC quite a bit of time before the networks started calling them. By 9:30PM, I was convinced of a Trump win even though the networks and their overpaid pundits continued for hours droning on and on about possible paths for Hillary to still win.
I also saw that the NYT had Trump losing NH, MN, CO and NV early on and they ended up being right about that as well.
My point is, I think the era of election night punditry by so-called experts like Karl Rove and Nate Silver are over. Incoming voting data will be processed and accurate predictions will be made on the remaining votes based on statistical data with no human intervention or bias.