Hacker News Reader

Comment by tdb7893 | Hacker News Reader

The polls weren't that off actually. They missed the mark overall by about 2%, which is actually pretty similar to the error in past elections. That's not very surprising because without being able to sample the actual distribution you need to make some assumptions for the statistics to work.

View on HN

I presume that pollsters did their job well. There is a pattern here, they got it wrong both with Brexit and with Trump. I think bad polling was not the reason. The media, the government and academia have created an atmosphere of militant orhodox liberalism, where dissenting voices are not only silenced, but often ridiculed, doxxed, forced to resign etc. These people will choose to disappear from the polls. They will "Think as they want, but behave like others", until they are given the chance to fight back.

Thankfully, in democracies we have elections to release those tensions. And setbacks in democracies are inevitable.

He is in charge of the EPA now but he tweeted that global warming is a conspiracy created by the Chinese. He is in charge of the State Department but he thinks that Mexico will pay for a border wall. He is charge of the FDA and he thinks that vaccines cause autism. Some of his views are deeply disturbing on a level much deeper than just "orhodox liberalism"

The response to people who are wrong should be to engage and argue with them. By tossing them as deplorables they will keep to themselves until they can get back at you.

The use of "deplorables" is presumably a reference to Clinton's remarks and it's really interesting you reference that here because her point of that comment is similar to yours in that you can't just lump all Trump supporters into the racist and xenophobic basket that many people did. I agree with you and I will gladly argue politely with people who are willing but I don't think it's unfair to criticize such damaging views in a presidential candidate.

Regardless of how you feel about him, you can't deny that he knows his audience. He's a salesman par excellence.

> They missed the mark overall by about 2%

This is misleading. Only on a national level were the polls that close. In the actual swing states that gave Trump the presidency, the polls were off well outside the margin of error.

For instance, in Wisconsin no poll ever showed Trump winning that state, and the closest IIRC showed Client winning by 3%. The average was significantly higher than 3%.

I will give you that they were wrong in Wisconsin in a pretty spectacular way.

When you take wildly changing vote count estimates (from +10 to +2 within two weeks) and turn that into more than 80% of probability of a specific candidate winning, that's a completely special level of stupidity.

Nassim Taleb also wrote about this https://twitter.com/nntaleb/status/796357059813974017

Nate Silver had an excellent response to this on one of the "Model Talk" editions of the FiveThirtyEight Elections Podcast[1]. TL;DR of the problems with Taleb's critique:

1. "Volatility" in the model is not random, as Talib asserts -- it happens at predefined points in the time. For example, the accuracy of polls jumps greatly after the conventions complete.

2. Volatility at a previous point in time does NOT mean there must be similar volatility going forward, as Talib implies.

3. Taleb's quickie models that he posted for explanation don't remotely match empirical data.

I respect Taleb, but he seems to have fallen into a trap really smart people often fall into: being right with limited data, but wrong with complete data (in this case, he creates assumptions from his knowledge in other areas like options pricing, which are wrong in this domain).

[1] Skip to 13:30 remaining http://fivethirtyeight.com/features/what-makes-a-tipping-poi...

EDIT: formatting

Reply on news.ycombinator.com