CCES/YouGov: Biden +8

[Fmr. Gov. NickG]

CCES is a study designed for academic use, and it has been the gold standard of American election data for political scientists since I believe 2006.  The survey is extremely long and they collect a very large sample so that you can get meaningful results at the congressional district level.  CCES stands for “Cooperative Congressional Election Study”, and they ask a lot of questions about congressional candidates and issues/bills voted on the past congressional term.   There is a pre-election survey and post-election survey given to the same respondents.

I used it’s data in my dissertation to generate common-space ideology scores for legislators and congressional constituencies.  It’s a massive and terrific project, but it’s not really designed for day-to-day horserace polling.

[Fmr. Gov. NickG]

So this will end up being off by 3-4 points.  I’m very curious to see dig through the post election panel data to see if they find evidence of a last week shift toward Trump (and if so, who flipped), or if the sample was indeed biased slightly toward Biden.

[Fmr. Gov. NickG]

It's particularly shocking to see the CCES screw up. That will have knock-on effects in most of the post-election research.

Except it’s not clear yet if the CCES actually did screw up.  They could very well find evidence that 2% of voters changed their mind from Biden to Trump in the last month, or that a dispropotionate number of undecideds went for Trump.  And with a >50k sample, you can actually get a good sense of who that 2% are.  

Additionally, it is not clear to me how the topline of this pre-election sample was weighted.  We need to make sure the correct weights are being used before drawing conclusions.  Fortunately, the CCES is data is very transparent, and you can see in the exact weight assigned to each respondent in the final release (though this might not be for several months).