[EM] [CES #8854] impossibility theorem for monetizing voting systems

[Abd ul-Rahman Lomax]

At 05:52 PM 6/19/2013, Warren D Smith wrote:
>[Note, the reason I am not sending this to E.Glen Weyl, is because of
>the following message from him dated 18 june 2013:
>"Warren,
>I am unwilling to continue my communication with you. I hope you will
>stop contacting me."]

Well, that's his choice. I'recently wrote to a published scientist, 
pointing out a major error in what had been printed from him, a 
blatant, face-palm, error (where this scientist was disagreeing with 
a published review of a field), where he actually showed the 
*opposite* of what he thought he had shown, and he responded with 
something like, "You will do anything to avoid the truth."

Takes one to know one.

The error was so bad that the phalanx of scientists who then 
responded to him clearly didn't even understand what he'd done, it 
was so stupid. It's fun, so I'll explain a little more.

(What follows has nothing to do with election methods.)

Confirmed research has shown, that in a certain type of experiment, 
result A is correlated with result B. The value of the ratio, A/B is 
of great theoretical interest. A can be measured reasonably 
accurately, B is difficult to measure. A is *unreliable.* I.e., it's 
apparently difficult enough to control experimental conditions that 
hidden variables have historically caused A to vary substantially. So 
in otherwise identical experiments, on the face, result A may vary 
from, within measurement error, zero to some substantial value. (The 
A "signal" in these experiments is often about ten times noise, 
sometimes far higher than that.)

So, in a published presentation of the data in a book, from a 
perticular experimental series, an author had presented a plot of A/B 
vs A, comparing this with a horizontal line that represented a 
certain value from a possible theory of what is happening in the 
experiments. The plot shows a certain scatter, which declines as A 
increases, as would be expected from the error in measuring B; error 
percentage declines as B increases.

The author I informed had criticized this work by calculating a 
correlation coefficient between the X-axis and Y-axis values of the 
data points in this chart. Why he used the chart, digitizing the 
values, instead of the actual data (presented on the next page of the 
book!), is beyond me. It shows the same thing, though: A/B has a very 
low correlation with A. I confirmed his math.

That is, A/B does not vary with A. He was claiming and apparently 
believing that the low correlation coefficient meant that A and B 
were not correlated. In fact, he showed that they were, because A/B 
is apparently a constant.

So I publicly pointed this out, and informed him as a courtesy.

He's an old man now, and, like many old men, has no time for fools. 
Literally and directly.