If Vox wants to make a case that bivariate analysis is not appropriate, and multivariate analysis is the only thing with any meaning, then that’s fine. That could be a reasonable case. They should make that case instead of manipulating the data in bivariate analyses and using that data manipulation and misrepresentation to push a narrative that a correlation is clear, when it’s not clear. Any mistake I’ve made in data presentation is not only a mistake they make, but compounded by their dishonesty in how they’re making it.
If they want to default to multivariate analyses on homicide drivers, then they need to be honest (A) about correlation/causality, and (B) all the other correlations which are far more correlated than gun ownership. And (C) they need to quit peddling their bivariate graphs for clicks, because they’re not helping.
I talk about accident numbers a small bit in the second piece, and hope to return back to them in the 5th, which is still a ways out. The main problem is suicide, any way you slice it.
What you find when you dig through these studies, is they’re all using different ownership rate data. Some are using suicide proxy, some are using suicide proxy blended with other factors. I found one that blended suicide proxy with subscriptions to Guns and Ammo magazine. But all these end up with different rates, and none of the rates line up with the rates shown in the direct polling study, which was widely circulated. One or more of these must be wrong, because they have different data. In the second article, I cede that their methodology may be pretty good, and discuss it.
But there is absolutely a self-fulfilling prophesy thing going on if you’re plotting gun deaths vs gun ownership and the gun ownership is derived from suicide numbers, because suicide is two thirds of gun death.