Reply #53: How can you know?

[Salomonity]

The NEP people asked a few thousand people who they voted for, and various other questions.

If they asked a true random sample, we'd have proof of a stolen election.

They didn't. First, because of clustering--the various voters sampled tended to come from the same precincts, so that radically enlarges the margin of error. If one of the polled precincts happens not to be representative, the poll is no good.

Suppose you've polled 15 precints in a given state. That just tells you what happens in each of them--the problem is, there's no particular reason to think that those precints are a representative sample of precincts in the state. I'm unaware of them having been chosen randomly; my understanding is that Mitofsky picks them for his own reasons.

They don't have to be representative, because the raw results aren't actually used by anyone. As election returns come in, they weight some precincts more than others, to control for increased or decreased turnout in some regions.

The reason exit polls are accurate is that after the fact, you can adjust the poll to make the presidential vote numbers coincide with the real data, and then hopefully the other columns also line up.