> generating a truly random dataset is really hard
IMO it's not just hard but impossible for polls to be unbiased, because people that don't like answering polls are never represented. If the subset of people that don't answer polls isn't itself following the same distribution over all possible opinions as the rest of the population the results are biased, and by definition we can't test for these correlations.
IMO it's not just hard but impossible for polls to be unbiased, because people that don't like answering polls are never represented. If the subset of people that don't answer polls isn't itself following the same distribution over all possible opinions as the rest of the population the results are biased, and by definition we can't test for these correlations.