[gorgonical]

Echoing a sibling comment, it seems strange it's so common that people look at probabilities like this and say that because the thing that happened was less likely that the model was wrong.

They say explicitly how this works: they run simulations and count winners. That means among their simulations is very likely to be an instance that is similar to the real outcome.

[iconjack]

Yes, I agree. If you are tasked for example with analyzing a coin to assess its bias, you could do a very thorough job, measuring precisely its mass distribution, its contours, etc and come up with an extremely accurate estimate that say, the probability of it landing Heads is 73.22%.

If the coin is subsequently flipped and comes up Tails, it doesn't mean your analysis was wrong.

On the other hand, 538's simulation was based on probabilities of winning each state, and as I think most will agree, before and after the election, the most important states were Wisconsin, Pennsylvania, and Michigan. On election day, 538 gave the following probabilities:

Wisconsin 83.5% Clinton [1]

Pennsylvania 77.0% Clinton [2]

Michigan 78.9% Clinton [3]

I think Nate really screwed the pooch here. The vast bulk of his effort should have been spent on these states. If that is in fact the case, then I retract my criticism.

[1] https://projects.fivethirtyeight.com/2016-election-forecast/... [2] https://projects.fivethirtyeight.com/2016-election-forecast/... [3] https://projects.fivethirtyeight.com/2016-election-forecast/...

[GameOfFrowns]

>That means among their simulations is very likely to be an instance that is similar to the real outcome.

So what? If you only make enough enough predictions then some of those will turn out predict the future correctly. It is how "The Simpsons did it first" meme came into being.

[beardedetim]

> If you only make enough enough predictions then some of those will turn out predict the future correctly.

If we have a coin and we say "It's 50% chance it lands heads" and it lands tails, were we wrong?

If we have a 20-sided die and we say "It has a 5% chance of landing on 20" and it lands on 20, were we wrong?

[GameOfFrowns]

Depends, is the coin a Democratic party establishment candidate and the predictor part of the liberal politico-media complex trying to prime an audience?