The crosstabs Nate Silver is linking to seem to have been taken down. I did find this previous Trafalgar Michigan poll
https://drive.google.com/file/d/1bfV...9ZIThV_7x/view
Response Rate: 1.37% - in other words "We called 75000 people and here's what the 1025 people who picked up and completed the survey had to say".
This might explain some of the wilder results posted earlier.
Also I checked out their numbers for Pennsylvania 2016:
https://projects.fivethirtyeight.com.../pennsylvania/
Their survey was done from 3-5 Nov. Similar dates were
3-5 Nov You Gov Clinton +2
3-6 Nov Gravis Clinton +6
1-7 Nov SurveyMonkey Clinton +3
3-5 Nov Trafalgar Trump +2
2-6 Nov AA/ESA/POR Trump +2
1-7 Nov Google Tie
Actual Result: Trump +0.7%
So in fact google were closer to the result.
For Michigan they polled Nov 6 and said Trump +2 when it was Trump+ 0.2% (i.e. they were as far off as the national "polling miss" that is often criticised) but the only other poll taken that late was the D-rated Mitchell Research one which gave Clinton +6 so they win the direct comparison. (although Google did 1-7 Nov and nailed it as a tie).
Quote:
Originally Posted by Mason Malmuth
But a good product can also have something to do with the need that is out there. If, for instance, as you point out, that polls which consistently lean certain ways have customers because these customers might want polls that lean certain ways, then in a sense they're putting out a good product for that customer.
Yes, and isn't it the case that for these polling companies, politics polling is mostly about marketing themselves for their actual business of selling polling services to companies for non-political polling. It would seem to make sense that they would want to stand out a bit in some way.
In non-political polling, of course if I want to know if the public prefers celebrity A or celebrity B so i can decide who to get to endorse my product, it's important I find out whether the public prefers someone 75-25, or 60-40. If it's anywhere near 50-50 it means that it really doesn't matter which one I pick. That's very different to politics polling where the difference between 51-49 and 49-51 is massive.