Nate Silver: "PPP is basically just duplicating polling averages".

[Nichlemn]

Source of quote: Twitter
Article

[Mr. Morden]

That article's a good read:

You must be logged in to read this quote.

[Miles]

Silver was AWOL for the better part of this cycle.

[publicunofficial]

Isn't "duplicating polling averages" Nate Silver's job description?

PPP was the most accurate pollster in Colorado this cycle, and in several other states. I have no idea why Silver has this vendetta against them.

[Nichlemn]

Isn't "duplicating polling averages" Nate Silver's job description?

Sort of - he's using the polling average to make predictions. But his job is very different from a pollster. They are providing the raw data that he's interpreting, and if they're fudging it, they're making predictions harder.

You must be logged in to read this quote.

Pollsters are meant to add information. If you're pretty much just copying other pollsters, all you're adding is noise, even if your final results seem accurate from a purely error-based analysis.

[Lief 🗽]

lol, ok nate

[Panda Express]

I agree with the larger point about inliers/herding being kind of a polling problem in general but Silver singling out PPP again just reeks of pettiness and makes him look like a big fat whiny baby

[muon2]

Herding should not be discounted, but it looked like there was at least one other factor influencing the polling after Labor Day. After 2012, many pollsters realized that their likely voter screens were too tight and they underestimated Dem turnout compared to 2008. In reaction to 2012 likely voter screens were relaxed for 2014, but that overestimated turnout, and with it the Dem vote. As it turned out turnout plummeted, and with it the Dem vote in many states. Pollsters couldn't catch that because they were trying to avoid their mistakes of 2012, instead of tracking who would actually vote in 2014.

[Mr. Morden]

Herding should not be discounted, but it looked like there was at least one other factor influencing the polling after Labor Day. After 2012, many pollsters realized that their likely voter screens were too tight and they underestimated Dem turnout compared to 2008. In reaction to 2012 likely voter screens were relaxed for 2014, but that overestimated turnout, and with it the Dem vote. As it turned out turnout plummeted, and with it the Dem vote in many states. Pollsters couldn't catch that because they were trying to avoid their mistakes of 2012, instead of tracking who would actually vote in 2014.

That's one possible explanation for why the polls might be systematically off in favor of one party.  But it doesn't explain why the scatter among different polls for the same race would show less scatter than the margin of error predicts.  How else do you explain this graph, except by herding?:

[Attorney General, LGC Speaker, and Former PPT Dwarven Dragon]

As it is, if PPP was herding, they would never release outliers close to the election...something they did this year in CT (had a Malloy +8 poll in October)

[Mr. Morden]

As it is, if PPP was herding, they would never release outliers close to the election...something they did this year in CT (had a Malloy +8 poll in October)

Well, sure, you can pick out individual examples like that, but how do you explain this graph of PPP's Senate polls?

[Ljube]

So, PPP is a sham. They are basically lazy, don't do any public polling and simply invent the public polling numbers based on their private polls and the prevailing consensus.

[🐒Gods of Prosperity🔱🐲💸]

So, PPP is a sham. They are basically lazy, don't do any public polling and simply invent the public polling numbers based on their private polls and the prevailing consensus.

no, that's not what this means. They weight their samples and/or choose not to release what they consider to be outliers.  Read the article, it's worthwhile.

Morden, is that a graph of just PPP's polls, or all polls?  It's quite ambiguous.

[Landslide Lyndon]

What the heck is wrong with Silver? He seems to have an axe to grind with everyone. One day he trashes PPP, the next Sam Wang and who knows who will be next.

[Mr. Morden]

Morden, is that a graph of just PPP's polls, or all polls?  It's quite ambiguous.

The first graph I posted is all polls, with the orange line showing the mean deviation from the polling average.  The second graph is just PPP polls, but it compares against the same orange line for all polls, showing how PPP hugs the polling average more closely than it should.

[Lurker]

What the heck is wrong with Silver? He seems to have an axe to grind with everyone. One day he trashes PPP, the next Sam Wang and who knows who will be next.

To be fair, in Sam Wang's case it was not Silver who started that "flame war". Wang had been trashing Silver long before the vice versa occured.

Silver's dislike of PPP is another matter though.

[○∙◄☻¥tπ[╪AV┼cVê└]

If you weight a poll for party or race, and those weights are correct, then you can get more accurate than the 2.8%. The herding might not have been on the actual poll results, but how many of each race or party to include.

[Nichlemn]

What the heck is wrong with Silver? He seems to have an axe to grind with everyone. One day he trashes PPP, the next Sam Wang and who knows who will be next.

What is wrong with this exactly? In the quest for accurate prognostication, there are lots of people who will lead you astray. If he's right, he's making the world a better place by keeping the shysters honest. What, do you think there should be like, a gentlemen's agreement where no-one in the predictionsphere attacks each other? The people who noticed odd things about Research2000 polls, they should have let them continue with their fraudulent polling rather than risk being seen as having "an axe to grind"?

[Mr. Morden]

If you weight a poll for party or race, and those weights are correct, then you can get more accurate than the 2.8%. The herding might not have been on the actual poll results, but how many of each race or party to include.

Except, why would the weighting of all these different pollsters converge closer to the same answer as the election gets closer?  With demographic weighting, shouldn't those demographic weights be the same one month before the election and one day before the election?  There's also the problem that the identity of the other polls you're comparing against changes over time, and from state to state, because in some cases the most recent polls are Fox, Quinnipiac, and SUSA.  In others, they're Des Moines Register, Marist, and Yougov.  Or some other combination.  How do you keep getting close to the recent polling average if the polling you're comparing to consists of pollsters with a different mix of demographic assumptions each time?

[KCDem]

Hahaha coming from the guy who was accurate than PPP. I guess he has to take his frustration out on someone for ing up so bad. Are you ok, Natypoo?

[Panda Express]

bring back research 2000

[Landslide Lyndon]

What the heck is wrong with Silver? He seems to have an axe to grind with everyone. One day he trashes PPP, the next Sam Wang and who knows who will be next.

What is wrong with this exactly? In the quest for accurate prognostication, there are lots of people who will lead you astray. If he's right, he's making the world a better place by keeping the shysters honest. What, do you think there should be like, a gentlemen's agreement where no-one in the predictionsphere attacks each other? The people who noticed odd things about Research2000 polls, they should have let them continue with their fraudulent polling rather than risk being seen as having "an axe to grind"?

It's the attitude more than the fact that he attacks them. Let's not forget that Silver even trashed Paul Krugman when the latter "dared" to say that his new site was somewhat disappointing.

[Lief 🗽]

Nate Silver once again desperately drumming up controversy so that he can get some page views for his failure of a website.

[Vosem]

Nate Silver is one of the biggest FFs among today's prominent political prognosticators. Right up there with Sean Trende.

[King]

They stole Nate Silver's gig.