[DenisM]

Surely one could control for wealth to tease out the effect? There has to be enough overlap between two large groups.

Oh wait, here we go: https://pubmed.ncbi.nlm.nih.gov/20560836/

[...] The likelihood of violence associated with crack cocaine users was greater compared to powdered cocaine users at the bivariate level. However, these differences were almost uniformly statistically nonsignificant when demographic, mood and non-cocaine substance use disorders were controlled for. [...]