This is very much like vorp, where we try to figure out how much marginal impact (i.e. counterfactual impact) a person has: we say what was the likelihood of x happening given person a was there, vs what would the likelihood have been of x happening if person a wasn’t there