Today’s D/R/I is 32/22/38, which means this model would only be predictive for a turnout model where only 22% of voters are Republican. Just to remind readers, the 2008 turnout split from exit polls showed a 39/32/29 split, and that was considered a nadir for Republican turnout. In the 2010 midterms, the split was 35/35/30.
Take a close look at the Republican representation in WaPo/ABC polls this year. Starting in January, that has been 25%, 23%, 27%, 23%, and now 22%. The pollster seems incapable of finding a representative number of Republicans for this poll series. Perhaps that should give the two news organizations involved a hint about finding a new pollster.
Why should they? The pollster is fulfilling said news organizations’ needs perfectly.
PS: Note that the needs of the WaPo and ABC are not the needs of their Democratic readers/watchers. Not to be cynical about this, but I remember that similar fiddling with the numbers took place during the run-up for 2010; that resulted in things looking… reasonably OK… for the Democrats, up to the moment when they suddenly didn’t. I imagine that the resulting panic was good for selling papers and getting viewers, but it’s a heck of a way to run a news organization.