Indiana reminds us: Polling remains dead

So before the results came in from Indiana, I pointed out that Indiana looked winnable for Cruz, and looked at the ways the polls had gotten it badly wrong in neighboring states, ways that grossly underestimated one candidate or another.

Turns out they grossly underestimated Donald Trump.


Dewey Defeats Truman. Via Dave Winer on Flickr.
Dewey Defeats Truman. Via Dave Winer on Flickr.

With 96.8% of precincts in, Decision Desk has it at Donald Trump 54, Ted Cruz 36, John Kasich 8. The poll average had it at Trump 43 (-11), Cruz 32 (-4), Kasich 15 (+7). That’s just terrible.

Every single poll taken over the last week overshot Kasich’s result. Eery single poll except the laughable IPFW/Downs Center’s Cruz +16, underestimated Cruz. And not one poll had Trump taking the majority he did. Some will cherrypick something Gravis got, or something NBC/WSJ/Marist got, and cobble together a justification for the polling. But the polls failed.

Trump was outside the margin of error for Gravis and NBC/WSJ/Marist. Cruz was outside the margin of error for Gravis. Kasich was outside the margin of error for NBC/WSJ/Marist. Such outliers are supposed to be 1 in 20 statistical aberrations, not something we see every single time.

I called it in 2011 that polling was falling apart. It turns out, nobody is adapting well, and all the public polling firms are doing badly. They say that private, professional polling is different, but we on the outside have no way of independently verifying that.

If you want to know who’s going to win the next election, asking a poll is no better than going with your gut.