Is This Pollster Forging Their Results?

Nate Silver has my respect based on his past performance and results. He knows how to crunch numbers more accurately than any other polling firm. If he thinks there's a problem, there's probably a real problem...

So the top 3 results are the 8 and the keys on either side. And the 5 key groups with the keys on each side of it. With the largest key on most 10 pads (0 key) getting a nice spike in. I can see from the chart how the randomness might be in question. The only flier in the center keypad bias is the high 3 key, but its between the 2 and the enter key on most computers, so that might not be so random as well. I'm no math wiz, so if I can see the problem, it must be bad...

(would have been nice to have a better labeled graph, had to read the article 3 times to figure out why it was there.)

This is a classic mistake about random numbers. Random DOES NOT mean an even distribution. In fact, I would suspect the numbers if there was an even distribution in a stated random list. Meaningless patterns can and do exist in random numbers, and should be expected, rather than surprising.

There is nothing here, please move along.

Data from sources with a fair bit of random noise - such as polls - tend to have a distribution of numerals which is predictable within a certain error of margin.

With a very high number of poll makers, even if all were guaranteed to be playing to the rules, you'd almost certainly get a few who'd show anomalies such as this, but given the comparative rarity of anomalies versus cheaters the latter tends to be more likely for any given case.

> "Random DOES NOT mean an even distribution."

Yes, it actually does mean an even distribution, given enough random data points, and given no reason for bias. 5 years of data is a pretty good data set.

Testing for even distribution is a common practive for determining the strength of a hashing algorithm, for instance. If the output of a process does not tend to evenly distribute numbers over larger and larger data sets, it indicates a bias in the process.

You're not looking for an even distribution. Polls like this should follow Benford's Law, an unusual property of many sets of data that indicates that most results will end in 1. The trailing digits logarithmically decrease down the line.

For polls like this, a perfectly random distribution of trailing digits would be unusual and likely indicative of fraud as well. I believe a uniform distribution is what Enron showed in their books a few years ago.

This almost sinusoidal pattern definitely isn't what you'd expect to see, however.

Hashes do not use randomness but rather have one-way characteristics. If hash results had anything to do with randomness they would be useless for the verification of the original plain-text data. Perhaps you are thinking about key generation.

"This is a classic mistake about random numbers. Random DOES NOT mean an even distribution."

No one is proposing that an exactly even distribution is expected. The point is that the distribution we are shown is much too far away from even to be the work of chance. The article makes this very clear.

I suggest you work on both your reading comprehension skills and your knowledge of statistics.

OK, so where is this data set available so other statisticians can test it? Without seeing the actual data it's difficult to assess what is going on or whether Silver's conclusions are repeatable. Also not listed is the method Sliver applied to test for randomness. I presume by "even distribution" y'all mean "Gaussian" or bell curve? I see a lot of comments here but nothing based on the posters having actually examined the data or tested Silver's assumptions.

Jim Galloway describes Strategic Vision as "Republican-oriented".

I believe that it is a generally accepted rule that being "Republican-oriented" absolves one from any responsibility towards reality.

I.e., if poll numbers suggest a reality that is not "Republican-oriented", then obviously reality has been subverted by ACORN and the data set that describes reality should be discarded.

This is probably why the results from Strategic Vision are 8-up.

Reprimanding a non-member? In and of itself, that makes me highly suspicious of the motives of the people involved in the AAPOR's "ethics panel".

You need to quote a bit more carefully. Your first pull-quote from Nate's post suggests that he has directly accused SV of cooking the books. But in context, what he suggests is that there is the data look falsified, not that they are. The evidence is there and pretty damning, and I think he (and I and you and a lot of others) believe SV isn't playing straight, but he does not actually directly accuse the firm of falsifying.