[jihadjihad]

It's a pretty typical result from social science. 1500 respondents is a fairly large sample size--I've seen papers where you're talking closer to ~50.

Going from 22% to 32% is a 10 percentage point difference, but a 46% increase. I agree that the conclusion at the end is flimsy, and there are undoubtedly confounding factors--but the statistics are pretty well in line with what you'd expect in social sciences.

[mistercow]

Yeah, but think about the error bars there. You’re looking at an MoE of around 3% for the entire sample, and more for these subgroups. The real difference could be just a couple percentage points and still be within the confidence interval.

[jihadjihad]

I completely agree--many times, papers from the social sciences appear as exercises in "wringing signal from a sea of noise." They're not always successful or convincing! But often there is wide latitude because of the inherent noise...don't ask me why! I got out of economics for a reason!