Burritos and political polls – who do you trust?

If you believe data reveals truths about everything, and you love a good burrito, you have to head over the Five Thirty Eight blog website and check out their quest to find America’s Best Burrito.

The site says it’s analyzed data on more than 67,000 restaurants and has identified the top 64 burritos in America for a March Madness-style playoff that I confess I lacked the patience to look carefully at.

But I just love the video above in which a correspondent for the survey performs an autopsy on a Taco Bell burrito supreme to explain the scoring system.

The Five Thirty Eight website was founded by data guru Nate Silver, who wrote a widely-read blog for the New York Times during the 2012 election cycle that analyzed polling data on a long list of political races. He was known for publishing a figure weeks leading up the election showing the likelihood of a Barack Obama victory.

Silver left the Times after the election, and has the new site.

Can an internet poll predict an election?

I went to the blog looking not for burritos, but for wisdom on whether to trust a couple of recent polls showing Pennsylvania Gov. Tom Corbett is gaining on his Democratic rival in the governor’s race. Neither were traditional polls that rely on telephone interviews with randomly-selected voters.

One poll, sponsored by the CBS and the New York Times tallied internet questionnaires filled out by a large group of participating voters. The other used robocalls to contact voters and ask them to punch in answers to questions.

Traditional pollsters look askance at those methods, but even they acknowledge that it’s getting harder to reach voters the traditional way. More and more people just don’t have land lines.

A traditonaly pollster suggested I check out this story about the polling debate on the Five Thirty Eight site, which is known for looking a lots of polls and finding trends. This piece reaches an interesting conclusion: In races where there are both traditional and non-traditional polls being taken, the non-traditional polls aren’t so far off. But in races where there are only non-traditional polls, they’re much farther off (compared to election outcomes).

The implication is expressed in the title of the article: “Are bad pollsters copying good pollsters?” The writer, Harry Enten says the data “raise(s) the possibility some pollsters may be peeking at their neighbors’ papers.”

The article is strictly a data analysis, and doesn’t ask any of the non-traditional pollsters if they’re fudging their numbers. But I guess they don’t interview Taco Bell about their burritos either.