The 538 Magic-BS-Spewing-Machine Takes Google Surveys Seriously

There’s been a lot of talk about Nate Silver’s Five Thirty Eight blog at The New York Times this week. I’ll spare you the arguments (pro and con) to Silver’s application of baseball’s sabermetrics to election polling to produce election projections. What I do want you to know though is that most of the articles don’t dive deep into which polls are included in his magic formulas.

So what do I see in the most recent column? A discussion of a Google Consumer Survey.

Perhaps the most intriguing result from this group is the poll from Google Consumer Surveys. (Yes, Google has begun to conduct surveys online.) That poll had Mr. Obama ahead by four percentage points, an improvement from a roughly 1-point deficit for Mr. Obama in their prior survey last week.

The Google survey could be an indication that the effects of the hurricane will play somewhat to Mr. Obama’s political advantage. But it will probably be Thursday or Friday, once power and some of the national tracking surveys that have been discontinued have come back online, before we can say so with much confidence.

Having never heard of this survey I decided to take a look. It’s very short – essentially two questions. The first question was how likely are respondents to vote. Here’s a quick tally of the response to that question.

100% likely 48.8% (+2.2 / -2.2)
Extremely likely 24.0% (+1.9 / -1.8)
Somewhat likely 7.8% (+1.3 / -1.1)
Not very likely 19.3% (+1.8 / -1.7)

Basically that’s a 72.8% to 80.7% voter turn-out number, which isn’t anywhere close to reality. In 2008 the turnout race for the Presidential race was 61.6%, in 2004 it was 60.1%. That question should be enough to discard the results of the survey itself, but not at Five Thirty Eight.

The second question is “Who would you be more likely to vote for.” Here’s the result table, taken from those who were either 100% likely to vote or extremely likely to vote (72.8%):

Barack Obama/Joe Biden, the Democrats 37.7% (+2.7 / -2.6)
Mitt Romney/Paul Ryan, the Republicans 34.0% (+2.7 / -2.6)
Leaning toward Obama/Biden 9.4% (+1.9 / -1.6)
Leaning toward Romney/Ryan 9.1% (+1.9 / -1.6)
Third party ticket 9.7% (+1.9 / -1.6)

Given the selection skew, and the fact this is an online survey it’s not surprising that the third-party ticket number is nearly 10%, but that number itself calls all of the other responses into question. There’s not going to be 10% of the vote going to third-party candidates in this election, just as there hasn’t been since Ross Perot ran in 1992 against Bill Clinton and George H. W. Bush.

So this whole poll is just online polling garbage, not that there’s anything necessarily wrong with it. What’s wrong is that it’s part of the Five Thirty Eight formula (with an unknown weight since the model is not published) used to produce that magic data that shows Obama with a 77% chance of winning the Electoral College vote.

As the old saying goes, “garbage in, garbage out.”

DeVotchKa: Rock Band Stage Show Features Romney Beheading
What Past Election Results Tell Us to Expect This Time