More to come from me on this in a long form I am working on to put out from @coldspark but we're seeing a historic response bias on surveys that is setting the table for a large polling miss this fall.
In the meta data from the call centers college educated Dems are 3-4x more likely to answer than non-college. While weighting can help minimize the bias if done correctly it won't totally eliminate the problem.
For example even if you quota'ed for party (something I have very mixed feelings about) AND for education at the topline you can have the college Dems consume such a big chunk of Democrats that you miss the downscale Dems that are MUCH less partisan loyal.
I am working on a longer form piece on this but I think its time to really raise the alarm on a lot of the publicly released survey work right now that is showing historically liberal etc oriented electorates even when they weight back to party.
I know people will say this is "poll truthing" but the response bias issue is a real issue beyond arguing over a few cross tabs.
Cherry picking cross tabs is almost always fools gold but this is a critique of the methodological approach people are using to try to adjust for this issue.
One possible solution though it is messy as well is past vote history. We're seeing a lot of surveys WAY oversampling the 'always voters' and not getting enough of the infrequent voters that are so important in a Presidential race and Trump does better with low turnout folks.
One more tweet than I swear I am done til the piece is out. This is a hard problem to solve for bc it look as if poll response is tied to Partisanship/Trump support across multiple demos so weighting for the demos won't fix it.
I don't blame the public pollsters for not fixing the response bias issue they are operating on really tight budgets but I do think they are herding results with weighting as well which I do blame them for.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
