To do proper statistical analysis, you'd need to compare miles driven with FSD to miles driven by humans in similar conditions. That's not the easiest thing to do. Tesla had occasionaly released statistics with their earlier driver assistance technology, but without enough detail to form informative comparisons. Based on past experience, I don't expect Tesla to provide meaningful statistical data or to allow outside access to their raw data (which fwiw, seems like a huge privacy risk to collect), so I don't think we'll have an empirical result unless there's discovery for a lawsuit.