Do we then need to discount for “the particular failure modes of the most popular models are more likely to appear in the overall accident stats, and thus more

Do we then need to discount for “the particular failure modes of the most popular models are more likely to appear in the overall accident stats, and thus more likely to be used for the next set of test protocols”? :)