It's never been accurate as far as I can tell. Always over represents or under represents. If Nate / 538 was honest with themselves they'd drop it.

What was arbitrary in their methodology?

They seem to have been rather accurate given the data input.

Let the votes come in.

Because we now know there's something more important than math... votes!

but everyone craves the polls...their supposed omniscient view gives summary weight to what otherwise is encountered as a jumble of independently encountered personal opinion.

The Mitchell Research poll said Clinton would win Michigan by 37% (66% to 29%) - it was a robo-call poll.

The Target-Insyght poll said Clinton would win Michigan by 32% (62% to 30%) - it was a robo-call poll.

If you graph these robo-call polls, you would think this is how the race would play out:



If you believed the robo-call polls, here is what you would have expected in Iowa:



and New Hampshire:



and Nevada:



and Oklahoma:



and Massachusetts:



These robo-call polls are horribly inaccurate (and they always seem to err in favor of Clinton).

The Hillary crowd is going to continue to post robo-call polls. Just remember how inaccurate they have been in Hillary's favor and don't give those polls any attention.

Likewise, when you see the Hillary crowd blathering about Nate Silver's forecasts, remember that 358 is basing its forecasts on robo-call polls mixed in with the other polling so they have this bias built into the forecast model. Remember, Silver's 538 gave Hillary a 99% chance of winning Michigan. Garbage in, garbage out.

That's exactly what delusional Republicans said about the polls during the Romney - Obama election.