… if Nate Silver’s analysis is correct.
If a person is asked to make up a bunch of numbers … random numbers … s/he will tend to make up non-random numbers instead. So, for instance, if I ask you to state random numbers that have two digits, and I plot the second digits on a hisogram, and then I ask my computer to make up two digit random numbers and plot the second digits on a histogram, my computer’s histogram will show roughly the sane number of 0′s as 1′s as 2′s …. as 9′s (especially with a large sample size) but you, being a silly you-man, will make up numbers with more of one than another in a non-random fashion. If you are a typical Westerner you will come up with more sevens, most likely.
Nate Silver studied the polling data prodcued over a long period of time by the conservative publicity firm and polling company “Strategic Vision” and showed that their polling data is not random in the trailing digit like it should be.
Maybe polling data is just not random? If you studied the trailing digits of, say, prices of gasoline to the tenth of a cent in the United States, you’d find that all gasoline costs something-something-point-nine (very non random). If you studied the trailing digits of retail prices in most areas, again, you’d find more 9′s than expected. But, when Nate Silver studies Five Thirty Eight’s polling data, he gets a pretty even distribution across the trailing digits, which is what you’d expect in a large sample.
Nate’s article is here.
It will be interesting to see how Strategic Vision reacts to this . They’ll probably have a poll that shows that most people think they are being honest.