Some brief observations after reviewing the crosstabs for the NYT/Sienna College national poll for October.
Every time I go through the crosstabs for this poll something seems off. This is not the first time I’ve seen the south over-represented, but the rest was generally more accurate than what I’ve seen before. The n for the Midwest: 429. For the Northeast: 504. For the West: 612. For the South: 1800. That’s more than the other three combined! Unweighted they have the South as 53% of all registered voters. When weighted they bring it down to 37% which is much more accurate, but why so high before weighting? Also, how does such a large n for the South impact all the sub-questions?
Seems strange to me, but IANAP (I am not a pollster). Unlike the last time I went through their tabs, when those without a college degree were over-represented, education levels are more representative of census data. The age group 45-64 is 32% of RVs--more than any other age cohort and 10% higher than 30-44. This is also the Trumpiest cohort. I did not see any data on rural/urban/suburban, but last time the rural percentage was substantially higher compared to the 2020 election.
I’m not clear on how “weighting” is different from changing the results to fit a preconceived conclusion. I know they use models and there is a methodology, but what about randomness? Why not get the correct percentage of respondents from the respective regions? I’m skeptical of all polls and they do not freak me out or make me complacent, but I do think they are exaggerating support for tfg.