I *am* a research statistician, with an expertise in experimental design. SlugLife is correct here; as multiple folks have pointed out, there's not a sufficient sample size to draw any reasonable conclusions, and the list of potential confounds is so long that even if you were to find an effect, it could be due to any number of causes - some of which we can measure and (partially) control for, and some of which we can't.as main data keeper for me data- its all rubbish and waste of ressources but i thought maybe one had the spirit and enthusiasm to drag them together and make funny columns or checkered lines
The only real way to know would be to assemble a very very large group (you'd need a huge sample given the low baserate of deaths/incidents) of divers trained in both SM and BM, who would be willing to be randomly assigned by a researcher to do their dives in SM vs BM for a given stretch of time. (You could also randomly assign per dive, which would increase statistical power, but also add enormous logistical complexity). Then see whether more of the divers in the BM condition have accidents/incidents than those in the SM condition. I highly doubt any ethics board would even approve that study (i.e., could end up randomly assigning someone to dive a SM-only cave in BM), and even if they did, there's almost certainly not any way to get the sample size/level of compliance you'd need for causal inference.
Some questions are just not very amenable to being answered empirically; a similar problem exists (for example) for whether any level of diving is safe during pregnancy. The number of women diving while pregnant is so low, and complications rare enough, that the published literature has more or less concluded that it's unlikely we will ever have the sample size to allow us to statistically determine the risk (and whether any safe limits exist). The best we can do is draw on animal studies, and our theoretical knowledge of physiology and how diving could affect that in pregnancy. The same approach is the best that can be taken here; there may be theoretical arguments for SM vs BM, but it's unlikely we'll ever have sufficient data to test those arguments empirically.
And, as I like to tell my students: the plural of anecdote is not "data".