Interesting paper. My first thought was that it would be impossible to distinguish causation from correlation, but it looks like they definitely tried. That is, how do you account for the fact that "people want to kill you" increases both the likelihood carrying a gun and of being shot. In Pearl's terminology, have they really calculated the effect of do(gun) instead of just (gun)?

Skimming the paper, it looks like they paired shooting victims with demographically similar control subjects. While the overall conclusion seems plausible (gun possession causes an increase rather than decrease in risk of being shot) this approach seems really hard to do correctly. Are you familiar enough with their methodology (and with the causative pitfalls involved) to discuss it?