Great post Elliott! One question. I see in Alexander's graphs that the "pro-police attitude index" variable for hispanics has the highest regression coefficient of any variable on any of the graphs. Yet, your piece (and Alexander) seems to be suggesting that police attitudes were not a strong predictor of vote switching among hispanics in 2020. What am I missing?
(It's been a minute since I've read regression outputs)
Thanks for your comment. I think you're focusing too much on comparing point estimates, whereas Alexander and I are calling attention (a) to the confidence intervals overlapping with different terms and (b) the rather small magnitude of the effect when compared to the sum of all the other ones.
It's also worth noting (c) that the effects are roughly equal by race; if the thesis that Hispanics were particularly influenced by police attitudes was true, we wouldn't expect that. How are police attitudes explaining the 20-point swing among Hispanics but a reverse swing among whites?
But don't get confused: We're not saying that police attitudes aren't important, we're just saying that it's probably a smaller (and more complex!) part of the story than the conventional wisdom has seemed to accept.
That's great clarification. The bit about overlapping confidence intervals is the piece I was missing. As always, the story seems to be more complicated than one data point would suggest. Thank you for doing this btw. I have really enjoyed all of David Shor's stuff recently and have learned a lot, but (at least online) it seems like many of his takes are being widely accepted without any contextualization or pushback.
Great post Elliott! One question. I see in Alexander's graphs that the "pro-police attitude index" variable for hispanics has the highest regression coefficient of any variable on any of the graphs. Yet, your piece (and Alexander) seems to be suggesting that police attitudes were not a strong predictor of vote switching among hispanics in 2020. What am I missing?
(It's been a minute since I've read regression outputs)
Aaron,
Thanks for your comment. I think you're focusing too much on comparing point estimates, whereas Alexander and I are calling attention (a) to the confidence intervals overlapping with different terms and (b) the rather small magnitude of the effect when compared to the sum of all the other ones.
It's also worth noting (c) that the effects are roughly equal by race; if the thesis that Hispanics were particularly influenced by police attitudes was true, we wouldn't expect that. How are police attitudes explaining the 20-point swing among Hispanics but a reverse swing among whites?
But don't get confused: We're not saying that police attitudes aren't important, we're just saying that it's probably a smaller (and more complex!) part of the story than the conventional wisdom has seemed to accept.
That's great clarification. The bit about overlapping confidence intervals is the piece I was missing. As always, the story seems to be more complicated than one data point would suggest. Thank you for doing this btw. I have really enjoyed all of David Shor's stuff recently and have learned a lot, but (at least online) it seems like many of his takes are being widely accepted without any contextualization or pushback.