> Where they eliminated that one 19 year old, and saw very little effect on the poll.

How is that a compelling answer?

It's like if a news investigation found that a company had poor quality control and features a photo of a defective product in their writeup and the company's response was "we issued a refund for the product the news team purchased."

All I can say is that you sound an awful lot like the 2012 talking heads who were convinced Romney would win despite all the polling pointing to the contrary. For the record, Nate Silver's biggest mistake in the primaries was not listening to the polls enough. For him (and every other data-driven forecaster) to be getting the general wrong they'd have to be making the exact opposite mistake as before.

I'd be happy to give you 10:1 odds on Trump losing. And I'm biased here: a few months ago, I was convinced Trump would win.

I haven't seen the crosstabs on the USC polls, but my gut feeling is that you're right; the kind of very fine demographic reweighting they're doing isn't common, because (A) it's a pain in the ass and of dubious real-world value; and (B) it's vulnerable to magnifying sample-size and selection problems, of which the skewed Trump support amongst young African-Americans is one example. So while I'll be interested to do a retrospective on the poll's accuracy and precision after the election, I'm personally not putting much faith in it for now.

That's not true, they saw the same affect as upshot, which is to say nearly 1 percentage point movement by removing him.

1 responder should not move a 2500 person poll by 1%

> Of the poll’s 14 weeks so far, there were three times when removing that weight switched the result from a Trump lead to a Clinton lead. Generally, the shift was between one and two percentage points...Some people have ignored the margin of error in order to make a political point, but there’s not much we can do about that.

That seems highly disingenuous a response.