(THREAD) I see quite a few people (mostly Trumpsters) claiming that the polls were horrible in 2016, even going so far as to claim that 2016 proves polls don't work.

First of all, polls that poll the national vote margin are not predicting the EC result.

1/ It is rather strange to say a poll that was very close for the popular vote was way off because it didn't predict something it didn't poll for, but this is some of the thinking I see.

2/ The final popular vote margin for Hillary vs Trump in 2016 was 2.1%. I see this get compared to the 3.2% margin for polls the day before the election, but it seems a little bit more complicated to me, since a lot of people vote early.

3/ According to this article, about 17.2% of people voted early in 2016, while close to 40% voted early, absentee, or used vote by mail:
https://www.eac.gov/documents/2017/10/17/eavs-deep-dive-early-absentee-and-mail-voting-data-statutory-overview

4/ Of course, some vote by mail people fill out their ballot early, and some fill it out right at the end. I don't know the distribution of when people voted, but as far as what the polls showed, it changed by the day. Here are the last 30 days of polling:

5/ Some notable gaps in the last 30 days from polls in 2016:

10/7/16: 4.6%
10/17/16: 7.1%
10/28/16: 4.6%
10/31/16: 2.2%
11/1/16: 1.7%
11/2/16: 1.3%
11/3/16: 1.6%
11/4/16: 1.5%
11/5/16: 1.8%
11/6/16: 3.2%
11/7/16: 3.2%

8/ As I said, I don't know when people voted. However, I do find it interesting that in the last 8 days of polling the average would have brought things closer to the actual 2.1% margin. Of course, for those who voted 30 days early, the polls had a bigger gap.

9/ If the polls were off by 1%, that is a cumulative error of 1 out of 200 voters flipping from Hillary to Trump. For a 1000 person poll, estimating Hillary's margin for the popular vote by 1% too high is the equivalent of the poll getting 5 people wrong.

10/ For example, if a poll of 1000 people ended up with 51.5% for Hillary and 48.5% for Trump, that is 515 people for Hillary and 485 people for Trump. If the final vote result was 51% for Hillary and 49% for Trump, that is 510 vs 490, or 5 people flipping from Hillary to Trump.

11/ I consider getting the equivalent of 5 flippers wrong in a 1000 person poll to be really close, but of course not everybody sees things the same way.

For the above I only considered 2 candidates to simplify things.

12/ As far as people who used polls to predict the chances of each person winning, on election day the 538 site said Trump had a 29% chance of winning. That is higher than the odds of getting heads twice out of 2 coin flips.

13/ If every presidential election had the underdog at 29%, then for each century of 25 elections, the underdog should win an average of just over 7 times. For 15%, they should win an average of just under 4 times. Obviously, an underdog winning once doesn't disprove those odds.

14/ If pollsters had the underdog at 15% to 29% chance for every presidential election and the underdog never won, that would be damning for pollster predictions, unlike the underdog sometimes winning.