I hate these kind of catch-all posts, but I’m pressed for time here in Cambridge, Massachusetts, where I’m doing wall-to-wall research for my upcoming book project. If everything goes according to plan, then within the span of about 36 hours, I will have interviewed the 2008 American League MVP, the man who really did invent the Internet, and some of the world’s leading authorities on hurricanes, artificial intelligence, and string theory. I’ve been reluctant to talk too much about the book project out of a sort of fear of jinxing it, but if you start to see me make a few more Tyler Cowen-esque random asides about, say, earthquake forecasting or the Netflix Challenge, you shouldn’t be entirely surprised.
— A few people have alerted me to another oddity in Strategic Vision’s polls: their numbers — when you take all the candidates plus the undecided vote — always add up to exactly 100 percent (except for exactly one case — question 9 here — where they implausibly add up to 102). Adding up to 100 — that sounds like a good thing, right?
Well, not really. For instance, if your raw numbers are something like…
Oxendine 44.3
Barnes 39.4
Undecided 16.3
…those should round to 44, 39, and 16, which add up to 99 percent rather than 100. In a two-candidate poll, in fact, the numbers should not add up to 100 percent about a quarter of the time because of this rounding error, and the likelihood increases with every additional option you introduce into the poll.
For most pollsters, including Rasmussen, Gallup, CBS/NYT, ABC/Post, Pew, PPP, SurveyUSA, FOX News, and Quinnipiac, it’s very easy to find plenty of examples like these.
Nevertheless, there are a couple of exceptions. NBC/WSJ, and Research 2000, like Strategic Vision, seem to calculate their undecided/don’t know number only after having rounded all the other numbers off, and then subtracting that total from 100. In other words, they correct any rounding error in the undecided column. I don’t think that’s the proper way to do it … but it’s certainly not a huge deal.
So this piece of evidence alone isn’t particularly incriminating. Still, if you want to get all Bayesian about it, although perhaps 2 out of the 11 most prolific pollsters do things this way, I’d imagine that the likelihood would be much higher than that if a pollster were fudging numbers — they’d have to be fairly clever to build in examples of rounding error and this is the sort of thing they might easily overlook.
— I owe people more coverage of New Jersey and Virginia. My general thoughts: these races have taken on something of a life of their own because of what are frankly a pretty flawed set of candidates. So I don’t know how reliable of a barometer they’ll really be as we look toward 2010. If Bob McDonnell loses the governor’s race because of some uncouth things he said in a thesis in law school, when he might have won if he’d written something about — I don’t know — the Holy Roman Empire instead, does that suddenly imply that the Democrats are in a much stronger position for 2010? I don’t think so.
— In a feature over at Esquire, I look at declining gambling revenues in Las Vegas, Atlantic City, and elsewhere around the country. The conclusion: states may no longer be able to count on expanded gaming as a cash cow. But Vegas itself, where gambling is much more of a “luxury good” than is generally believed, has a reasonable shot of bouncing back along with the rest of the economy.