NYT admits partisan non-response is source for Trump's lead in poll

[Ms. Toad]

53. Who knows.

I was simply addressing the contention that they didn't adjust at all for oversampling Republicans (so of course Trump would poll higher).

How good their questions were (on the front end - to provide enough information to adjust their data to the expected electorate) and how good the algorithm (on the back end to actually make the adjustment) is a different question.

If you ask the right questions to be able to sort out who is likely to show up at the polls, the adjustment to account for imbalanced sampling is just a bunch of math (and I've got two math degrees - the second in applied math). They did a bunch of math in this poll, but whether they did the right math is part crystal ball, part asking the right questions to be able to sort who is actually likely to vote. One of the polling entities I respond to frequently asks a lot of unrelated questions (some political, some not overtly). Early in the poll they usually ask if I voted in the last election - and they usually ask who I voted for in a race that someone who didn't actually vote probably wouldn't even recognize. That gives them some ability to sort out likely voters based on semi-verified past behavior. Later in the poll - enough later that someone trying to fool the pollster wouldn't necessarily remember how they answered the first questions - they ask about likelihood of voting in the next election (and who I might vote for). And at the end they ask demographic information - including party information (and which way you lean if you indicate you don't belong to a party). With that information, and a bit of a crystal ball, there is a reasonable chance they can simulate the predicted electorate (balanced for age, gender, race, political affiliation - and historical voting pattern). I didn't participate in the poll in question, so I don't know what kind of questions they asked, or what (if anything) they adjusted for beyond party affiliation.