This! This! This all day long. Without comment

[BzaDem]

96. This is absurd, and shows a basic lack of understanding of high school probability and statistics.

For those who think the results were tampered with, were you making the same argument when Obama overperformed his poll results in 2012 by more than Trump overperformed his in 2016?

This math teacher (really?) is assuming that the polls of the various swing state accurately sampled the population, and were only subject to sampling error (margin of error). In such an imaginary universe, the probability of the errors all going in one direction is very low.

But in the actual universe we live in, there is no easy way to get an accurate random sample of voters. 90% of people polled do not answer their phones. (Think about it -- when was the last time you picked up the phone to an unknown number?) Many people who do answer the phones say they are likely to vote, but don't end up voting. Pollsters make all sorts of assumptions when deciding how to weight the resulting data, and these assumptions are often very wrong. Yet making assumptions is the best they can do, because 90% of the population doesn't respond, and people do not accurately predict their likelihood to vote.

The resulting modeling error (different than sampling error) is that the polls are often off by several points. It is impossible to know before the election what direction the polls will be off, or by how much. But unlike with sampling error (where being off in either direction is equally likely, allowing the error to be reduced by averaging many polls), it is very likely that modeling error will cause most polls to be off in the SAME direction. After all, if (say) non-college educated white voters were marginally less likely to respond to pollsters in Ohio, why in the world would this not also be the case in Wisconsin (or other surrounding states)?

This is precisely why people on election night knew Clinton was in profound trouble once it became clear how she was doing in Florida. Florida was not remotely a must-win state for Clinton. But if she was underperforming her polls in Florida by several points, it was very likely she would do so in the midwest as well. Sure enough, that's exactly what happened.

In this case, the modeling error was a slight undersampling of non-college-educated white voters. This had a decisive outcome on the election, because the swing states in question were primarily in the midwest (with a higher-than-average percentage of non-college-educated white voters). But there were also many non-swing states where Trump over-performed his polls, for the same reason. The idea that Putin or anyone else hacked large numbers of votes in totally irrelevant states is ludicrous.