Credit to Gary Winslett for noticing this pattern: I saw this tweet from him earlier today, and decided to do a quick analysis to put some numbers on this observation and visualize the relationship. If you want to read more, here’s a longer article
An alternative hypothesis that I think should be explored is that, regardless of whether it's a red state or a blue state, the places with the highest homelessness AND the highest land use restrictions are places that people want to move to, perhaps due to favorable climate and natural beauty: California and Oregon on the blue state side, Montana on the red. The places with low land use regs and low homelessness are places nobody wants to move to if they can help it: Georgia (too hot) and Michigan (too cold) on the blue side, Alabama and Oklahoma on the red side. (That would make New York quite the outlier, though, since I grew up there and I can confirm the weather and scenery are uniformly terrible outside of fall upstate.) Anyway, it would be interesting to see what the graphs would look like if you tried to find correlations between either homelessness or land use freedom scores and net population gain/loss in those states. My general thought is that you can only get away with restrictive land use regs when lots of people want to move there.
I appreciate that you are showing the full plots rather than just the outcome of the regression. It’s pretty clear that there is strong heteroskedasticity here (variance in the residuals changes with the value of the regressor). A handful of outliers seems to drive the correlation at least in some cases. It would be interesting to throw something more sophisticated at this dataset: robust regression maybe? But even before doing that I’d look into why we have those outliers at all.
Edit: I see that you are sharing the data, so I can actually try to do some more analysis myself.
An alternative hypothesis that I think should be explored is that, regardless of whether it's a red state or a blue state, the places with the highest homelessness AND the highest land use restrictions are places that people want to move to, perhaps due to favorable climate and natural beauty: California and Oregon on the blue state side, Montana on the red. The places with low land use regs and low homelessness are places nobody wants to move to if they can help it: Georgia (too hot) and Michigan (too cold) on the blue side, Alabama and Oklahoma on the red side. (That would make New York quite the outlier, though, since I grew up there and I can confirm the weather and scenery are uniformly terrible outside of fall upstate.) Anyway, it would be interesting to see what the graphs would look like if you tried to find correlations between either homelessness or land use freedom scores and net population gain/loss in those states. My general thought is that you can only get away with restrictive land use regs when lots of people want to move there.
I appreciate that you are showing the full plots rather than just the outcome of the regression. It’s pretty clear that there is strong heteroskedasticity here (variance in the residuals changes with the value of the regressor). A handful of outliers seems to drive the correlation at least in some cases. It would be interesting to throw something more sophisticated at this dataset: robust regression maybe? But even before doing that I’d look into why we have those outliers at all.
Edit: I see that you are sharing the data, so I can actually try to do some more analysis myself.