Cohn should be given the benefit of the doubt here. He did miss in Texas in 2018, but given how intelligent and transparent NYT/Siena has been previously about methodology, I'd have to think they would've tried to resolve their SW polling issue.
Either way, it seems ridiculous to constantly criticize an A+ pollster because you don't like their results. And then turn around and use a 25 LV subsample of Collin County to disprove the rest of their poll.
The number of undecideds is still a big problem, and as for the Hispanic thing, I’ll believe it when I see it. Can I believe they’ve improved their methodology to deal with their problems with these voters and this state in particular? Sure. But we won’t know if they succeeded until next week.
Is the number of undecideds a big problem? It's 5% in this poll. 47 Trump, 43 Biden, 3% Jorgensen, 1% Someone Else, 1% No Vote for President. Oftentimes for NYT, the number of undecideds appears larger than it is because they don't include the Lib candidate in their press release but do in the survey. Even so, the relatively larger number of undecideds is a result of probabilistic weighting for likelihood to vote, where someone who says they are not likely to vote still has a 20% to vote. These people still do vote occasionally, and much more likely to be undecided.
Bottom line, this is empirically and factually a more accurate method than a hard cutoff LV screen and just because the larger number of undecideds makes users here uncomfortable doesn't mean the poll is inaccurate.
