General Discussion
In reply to the discussion: Are POLLS reliable these days? [View all]Yo_Mama
(8,303 posts)And then rebalance the raw results by counting the results in proportion to the overall population, not just the people they reached. And polls now do sample cell phone only users.
So, they will try to pick a geographically and socially balanced sample, but then after they get the results, if they find the over-sampled oldsters in relation to the younger groups, they will up the weight of the younger groups they did sample. If they find they have too many men, they'll up the weight of the women they do have.
So the actual output of a poll nowadays is not the results of the poll, but the results of the poll reworked to become more representative.
But in reality, the margin of error is always at least 2 or 3 percent, so it can't predict the outcome of a close election. End results of 52-48 might seem like a nice solid win, but the best poll before hand might have had 50-49 or even 49-50. It's not an exact science and never will be, and even getting the exit polls right is difficult, because you can have sample error there just through adverse selection (early voter turnout different than late voter turnout; willingness to participate lower among pressured workers just zipping by, etc).
The other funny thing about polling is that the longer you make the poll (information and double-checking for enhanced accuracy), the more you skew the results because not everyone will take the time. It's kind of the Heisenberg uncertainty principle of polling. The exit polls are also less valid than they used to be because of early voting.