That’s because the reason polling overestimated Republican margins in the primaries and special elections is unlikely to be duplicated in the general election: It was hard for pollsters to know which voters were going to show up.
Using polling to estimate election results requires making informed choices about the composition of the electorate.
Ding Ding Ding, it's amazing how Shepherd makes this point but doesn't realize that this is also applicable to a general? A lot of us have been ringing the alarm that some of these composites in GE polls are also pretty whack; these changes can make meaningful differences in polls! (i.e. Fox having an D+8 party ID in their PA poll vs. a tie now, etc.) It's not unskewing to simply question why many polls recently are showing more of a GOP-leaning electorate than they did even 6 months ago. Are you simply getting more Republicans to answer the phone? Are you just making assumptions we're gonna have an R turnout edge?
