Two mistakes, not just one. The first is the one mentioned, introducing his own biases into the mix.
The second, more important one is basing his predictions on technical analysis (in financial terms) rather than fundamentals. Lots of people bet themselves into the poorhouse when they predicted that the financially sound status of the vacuum tube couldn’t possibly be brought down by a tiny little transistor marketed by a lowly Japanese company. The transistor, however, had the fundamentals on its side, as demonstrated by the decades of dominance of Moore’s Law.
In this election, the fundamentals outweigh the technical analysis. People want change, both in policy and in tone. They’ve had enough of the same old promises, the same old results, and the same old lies from the same old liars.
It took a marketing expert by the name of Donald J. Trump to see the time was right for a brand-new — and hugely improved — presidential product, and a brave hero — in the same person — to take it to market.
Overly technical commentary:
Nate Silver's analysis is based on the assumption that the polls are unbiased. His probabilities essentially come from a Monte Carlo simulation drawn from a population that would produce the observed poll numbers. If the polls are biased, then Nate Silver's probabilities are useless. In this case, we all know the polls are biased. Even if pollsters wanted accurate results, polling response rates have dropped dramatically, and we don't know the extent of the drops among those who support Trump or among those who do not.