This approach should be sensitive to sample size, but I don't see any discussion of that. If he's tried it a lot, there should be some discussion of error on the first 10 hours, 100 hours, 1000 hours...