New York District 12 Polls
NY-12 is an interesting race. It's an open seat because a very senior Democrat is retiring, and there are four candidates running for the seat with over $1,000,000 in donations. If you want a full candidate list, it's in this post.
Out of curiosity, I poked at the polls for this race, and saw a good opportunity to show some stuff off. I've talked before about sponsored vs. unsponsored polls, but with this race, there were a pair of sponsored and unsponsored polls taken at almost the same time two months apart. So I put them side-by-side.
The thing I want to draw attention to is how the differences between the sponsored and non-sponsored polls change. Compared to the unsponsored polls, in the sponsored polls:
Alex Bores does worse in the March Poll (9%) but better in the May Poll (7%)
George Conway does better in both (3%, 8%)
Jack Schlossberg does better in March (7%) and worse in May (6%)
Micah Lasher does worse in both (8%, 5%)
As I noted in the Illinois breakdown, the thing that made me think better of those sponsored polls was that the order was basically consistent between the polls, even if the numbers were different. Here, there's not really a consistent direction, which makes me ask a lot more questions about the methodologies used.
There’s a possible explanations for why the results are so different: the unsponsored May poll only asked half as many people as the unsponsored March poll. Generally, when you take a poll, you are trying to ask a small group of people for their opinions and then use those as a general pattern for a much larger group of people. Logically, having more people in your smaller group makes it easier to capture more opinions. Usually, the magic number for getting enough people to get a reasonable reflection of a larger population without making too much extra work is around 1,200 people. All the polls except the May poll use between 500-600. This isn’t an ideal size, but it’s still workable within the accuracy math. The smaller you get, the steeper the dropoff is, and 300 respondents is much less reliable.
With all that in mind, what’s useful to take away from this?
Sponsored vs. unsponsored polls are cool to look at, especially if the sponsoring candidate didn’t get the data they wanted to use.
For all of the issues with the May unsponsored poll, it’s the only one that on the face of it, got the correct winner. Poll nerds can argue about the margin of error here, and that’ll probably get its own post, but if you want the simplest possible visual answer, weirdly enough, the small poll got it right.