If I picked a sample of 5 cherries and I made sure they were all rotten, it sort of reduces the amount of statistical inference you can do about the true distribution of the cherries to zero. Your use of "even with selection bias" doesn't really make sense here.