Quote:
Originally Posted by burn_this_city
Most of the polls in 2016 were within the margin of error. Certain media companies interpreted them as having no margin of error and put ridiculous graphics that had Clinton at 99% ect.
|
That wasn't actually the problem. The problem was the assumption that polling errors are independent. I.e., "there's a 30% chance that each of these three polls are wrong. Therefore, theres a 30% x 30% x 30% chance that all three are wrong, which is very unlikely". That's not how polling errors work. Sure, there's a 30% chance that each is "wrong" (i.e. off by a few points), but if it turns out that the 30% chance materializes in the case of poll "A", the chances of polls "B" and "C" being wrong in a similar fashion is no longer 30%, but
much higher.