Something bizarre is happening underneath the national polling, and I think it deserves a closer methodological look.
Trump’s net approval ranges from RMG’s -10 to CNN’s -34 for an extraordinary 24-point spread.
Yet, take a look at the generic ballot:
CNN: Trump –34 | D+8
Ipsos: –30 | D+8
Echelon: –28 | D+8
Fox: –22 | D+7
NYT/Siena: –21 | D+8
Quantus: –19 | D+8
Rasmussen: –13 | D+8
RMG: –10 | D+6
Across 14 recent polls, Trump net disapproval and the Democratic generic-ballot margin are positively correlated, as we should expect, but only modestly: r ≈ .32.
That deserves more attention.
CNN (and Ipsos) can measure Trump 24 points worse than RMG and 10 to 15 points worse than the aggregate, while somehow arriving neatly right at the same congressional ballot average. Remarkable.
Maybe there is a large and growing bloc of Trump-disapproving Republicans who are still voting the GOP ticket. Maybe differences in mode, weighting, question construction or response behavior or something else, anything, that is producing these unusually large house effects in presidential approval.
But when one political measure varies by 24 points across pollsters while a closely related measure remains exactly the same, that should trigger some serious QC work.
I simply can’t accept all of these outlier approval numbers at face value without understanding why the generic ballot isn’t moving with them.