General Discussion
In reply to the discussion: DU Exclusive: Gravis Marketing exposed as a fraud Part I [View all]fiorello
(182 posts)Sam Wang has a great poll-aggregation site: election.princeton.edu . He also includes your friend at Gravis, and your info via dailykos is already there.
I added my own twist: a math-based analysis of whether Gravis poll numbers look real or made up. (They don't look real. He has his 'favorite numbers' for filling in the after-the-decimal-point digit that makes a poll number sound oh-so-real. I mean, if a pollster goes through the trouble to report values as exact as 43.9%, they sound 'sciencey' and must be real, right?)
Here are the nasty details:
I pulled up a poll from the Gravis web site, that gave a number of 3-significant-figure percent values. I took only the last digit of each poll number (- the number after the decimal point, which should be randomly distributed from zero to nine -) and it looked strange. I counted the occurrences of odd and even digits (32 odd, 10 even) and applied the "Monobit frequency test", and found that this distribution has a 1-in-one-thousand chance of occurring by accident likilihood of occurring by accident. That's way too low - suggesting the numbers were, uh, invented.
It is true that I 'polluted' this test because I first noticed the non-random pattern and then picked a test that was geared to the non-randomness that I found. Also, Monobit requires more than 100 entries; I just had 42. (And that a true random collection of numbers does sometimes look nonrandom.)
But in my defense: this was a quick first effort, from a polling sample picked at random, and a more sophisticated analysis would probably identify an even higher non-randomness.
Gravis poll used for evaluation:
http://www.realclearpolitics.com/docs/2012/Gravis_FL_1001.pdf
Last significant figure of each entry:
307173 94359 07797 0559
37 0394517 4913950 532372.
Frequency of occurrence:
0. =5
1. =3.
2. =2.
3. =8.
4. =3.
5. = 6
6. = 0.
7. = 8.
8. = 0.
9. = 7.
---