Monday, October 15, 2012

Nate Silver's Mythical "State-National Poll Stalemate"

by Deaniac83/The People's View

This weekend, I called out Nate Silver of 538 fame for his actual abysmal record of predicting contested, close state-wide races in federal elections. It seems though that Nate might be aware of the pitfalls of his model that unquestioningly counts all polls, no matter how out-of-whack and skewed they are, on the assumption that the biases in the polls eventually average out. So, he has come up with a new CYA strategy: claim that there is a "state-national poll stalemate."
 For the last week or so, we have been hoping to decode a confusing polling landscape.President Obama still appeared to hold a narrow Electoral College lead on the basis of state-by-state surveys, while national polls were suggestive of a tie or perhaps the slightest edge for Mitt Romney.

If the current polls hold, predicting the election outcome will boil down to making a series of educated guesses about the relationship between state and national polls, and between the Electoral College and the popular vote.
Here, I'll decode it for you, Nate: stop including every cooked up poll in your "analysis" and have a look at the internals and demographics. Figure out if a given poll warrants inclusion given the  The "likely voter" models, especially of national polls and state polls conducted by national pollsters, aredrastically out of whack with their own registered voter numbers, as well as actual turnout demographics of the 2008 elections. If you take out this large right-skewing effect, or stop including every astroturf poll in your model (not to mention account for the clobbering Mitt Romney is receiving from the early vote), you will see that there is no "stalemate" here.

Secondly, national pollsters, are underestimating the likelihood of the minority vote, especially the Hispanic vote. And it isn't like Nate Silver is completely unaware of this happening. The Tuscon Citizen, citing a poll from Latino Decisions showing President Obama picking up a staggering 80% share of the Hispanic vote, documented Nate Silver's own work in determining the extent of this discrepancy:
 Matt Barreto of Latino Decisions suggested to Nate Silver of the FiveThirtyEight blog at the NY Times how the polls all missed the impact of the Latino vote in Nevada in 2010: All the major polling firms conduct their polls in English only, while Latino Decisions conducts their polls in both English and Spanish, with the respondent selecting the language in which they prefer the poll to be conducted. The major polling firms missed the Latino voters who prefer to speak Spanish. About 40 percent of Latino voters in California meet this description, with likely similar numbers in Nevada and Arizona. Mr. Silver compiled results from the eight states with the largest share of Latinos in their population: these are Arizona, California, Colorado, Florida, Nevada, New Mexico, New York and Texas. He found that in 10 of the 15 races, the polling average underestimated the Democrat’s margin by at least 2.5 points.
Read that again. In 10 of the 15 races in the states with the largest Latino populations (of which at least four are battleground states, and likely five, if you count Arizona), pollsters underestimated the Democratic vote by at least 2.5 points. In 2010 - you know, the year of disaster for Democrats. If pollsters underestimated the Democratic margin by 2.5 points in the year of the red wave, just how much are they off by now, with the vaunted Obama turnout machine working in overdrive and the Latino population having increased at least a bit? Just how many Hispanic and other minority voters are these "likely voter" models missing? Could accounting for those possibly solve your "stalemate," Nate?

Yet another challenge for national pollsters has been the women vote. As soon as the national polls switched to a "likely voter" model, Romney suddenly started evening out in the female vote. Of course, that switch happened right at the same time as the first debate, giving the pundits and pollsters an out to claim that women were just mesmerized by Prince Charming's "performance" and are now moving to him in droves. You know, because everything they ever believed in - the right to health care (including reproductive health care), equal pay for equal work, a social responsibility to care for children, education, and leveling thee playing field for the middle class - just flew out the window when Romney made them weak at the knees.

Give me a break. The national media's and pollsters' ability to insult the intelligence of American women is breathtaking. In 2008, they told us that women were going to vote for McCain and Palin because, apparently, it's the body parts and not the policies that are important to American women voters. And now they are telling us they are voting for Romney because he went on stage and lied. Yes, last time I checked, women love a man that lies to them.

No, the sudden "evaporation of the gender gap" comes likely not from women actually moving so much as it comes from pollsters moving their samples as to which women they count as "likely" to vote... conservative, married, white women.

Generally speaking, state and local polls and pollsters have much less of a horse in the media's horserace and much more of an ability to know their state and prepare the right sample. They are far less interested in flashing headlines and far more interested in building credibility. They do so the right way - by actually choosing good, representative samples for their respective states. Instead of, you know, manufacturing a "state-national stalemate" to try to cover their behinds should their "models," along with their credibility, blow up in their faces.

So before I close, let me respond to one last point from Nate:
 Anyone in my business who is not a bit terrified by this set of facts is either lying to himself — or he doesn’t know what he’s doing.
Given that your "set of facts" is basically bad national data cranked by your formula conflicting with better state data, no, I'n not particularly terrified by them. And maybe you should focus on the actual facts of polling rather than just the top-lines, and worry a little less about your "business." Just a suggestion.
Post a Comment