4 ms·

There's a variable missing from these models which probably wasn't all that relevant until Donald Trump showed up: How many people are going to lie about or re

There's a variable missing from these models which probably wasn't all that relevant until Donald Trump showed up: How many people are going to lie about or refuse to admit voting for a candidate like Donald Trump? I would expect this bias to be greater today than in 2016.

kachnuv_ocasek 6y ago
Andrew Gelman, one of the authors of the model, has some thoughts about that. I can't find the earlier article right now, but the Economist model counts with that as well. See the blog post (down in the first P.S.): https://statmodeling.stat.columbia.edu/2020/08/14/new-fivethirtyeight-com-election-forecast/ https://statmodeling.stat.columbia.edu/2020/08/14/new-fiveth...
evgen 6y ago
The so-called 'shy Tory' effect has been fairly conclusively proven to not actually exist; people will claim to be 'undecided' but will not claim to support X and then vote for Y. Look to size of the undecided group and then see where they break if the pollster forces a decision -- in 2020 the undecideds are smaller, and unlike 2016 they are breaking hard against Trump.
IfOnlyYouKnew 6y ago
I see no reason to believe this to br greater today than in 2016. And I believe the 2016 data did not show any of it. More polls have also moved from live interviews to automated system, where it's less likely that anyone will be "shy".
- gridlockd 6y ago
  In 2016 Trump polled way worse than he performed in key states. People are also more paranoid today than in 2016. If the communists take over, you do not want to be on some voted—nazi list.
  - jcranmer 6y ago
    > In 2016 Trump polled way worse than he performed in key states. That's not really true. The state-level polls were generally within normal polling error, although there was a systematic error due to educational attainment. In 2020, more pollsters have incorporated educational attainment into their weighting, so this source of error should be reduced.
- syshum 6y ago
  Really? Since 2016 the far left has gone insane with conspiracy theory and cancel culture There are all kinds of reasons to believe people would lie to pollster calling them up That is with out getting into the real sampling problems they have as when you look into the data they over-sample Democrats a good deal and I do not believe they properly account for this over sample largely because they do so on a base ratio, the problem with that is I believe they are also massively oversampling never trump republicans so they have an inaccurate republican sample for which to re-balance the polling They are not polling Trump's base, or Trump supporters or really the moderate middle They are getting Democrats and Never Trump republicans in their polls
jcranmer 6y ago
I have seen this argument most often from vocal, proud Trump supporters trying to explain away Trump's consistently low approval ratings. Why would someone so enthusiastically for Trump that they're willing to shout it to the world suddenly be circumspect about indicating their support in an anonymous poll with no human on the other side? Of course, you can argue that those people are not the same groups of people. But there's no evidence that this thing exists, even when people have looked for it in the past (see "shy Tory effect").
- gridlockd 6y ago
  I will indeed argue that these are dissimilar groups. Being a Trump supporter and reluctantly voting Republican are two entirely different positions that nevertheless lead to the same outcome. I do not think one should equate Tory and Trump either. Like I said, it is a new variable.

kachnuv_ocasek 6y ago

Andrew Gelman, one of the authors of the model, has some thoughts about that. I can't find the earlier article right now, but the Economist model counts with that as well. See the blog post (down in the first P.S.): https://statmodeling.stat.columbia.edu/2020/08/14/new-fivethirtyeight-com-election-forecast/ https://statmodeling.stat.columbia.edu/2020/08/14/new-fiveth...

evgen 6y ago

The so-called 'shy Tory' effect has been fairly conclusively proven to not actually exist; people will claim to be 'undecided' but will not claim to support X and then vote for Y. Look to size of the undecided group and then see where they break if the pollster forces a decision -- in 2020 the undecideds are smaller, and unlike 2016 they are breaking hard against Trump.

IfOnlyYouKnew 6y ago

I see no reason to believe this to br greater today than in 2016. And I believe the 2016 data did not show any of it. More polls have also moved from live interviews to automated system, where it's less likely that anyone will be "shy".

gridlockd 6y ago
In 2016 Trump polled way worse than he performed in key states. People are also more paranoid today than in 2016. If the communists take over, you do not want to be on some voted—nazi list.
- jcranmer 6y ago
  > In 2016 Trump polled way worse than he performed in key states. That's not really true. The state-level polls were generally within normal polling error, although there was a systematic error due to educational attainment. In 2020, more pollsters have incorporated educational attainment into their weighting, so this source of error should be reduced.
syshum 6y ago
Really? Since 2016 the far left has gone insane with conspiracy theory and cancel culture There are all kinds of reasons to believe people would lie to pollster calling them up That is with out getting into the real sampling problems they have as when you look into the data they over-sample Democrats a good deal and I do not believe they properly account for this over sample largely because they do so on a base ratio, the problem with that is I believe they are also massively oversampling never trump republicans so they have an inaccurate republican sample for which to re-balance the polling They are not polling Trump's base, or Trump supporters or really the moderate middle They are getting Democrats and Never Trump republicans in their polls

jcranmer 6y ago

I have seen this argument most often from vocal, proud Trump supporters trying to explain away Trump's consistently low approval ratings. Why would someone so enthusiastically for Trump that they're willing to shout it to the world suddenly be circumspect about indicating their support in an anonymous poll with no human on the other side? Of course, you can argue that those people are not the same groups of people. But there's no evidence that this thing exists, even when people have looked for it in the past (see "shy Tory effect").

gridlockd 6y ago
I will indeed argue that these are dissimilar groups. Being a Trump supporter and reluctantly voting Republican are two entirely different positions that nevertheless lead to the same outcome. I do not think one should equate Tory and Trump either. Like I said, it is a new variable.