A Question for Pinheads.....

I am not sure I understand what you mean here.

What I said was that the system would print out a verifiable list of votes by the voter that stays with the system showing who the vote was cast for. This printout stays with the system and a random sampling of the precincts should be taken to check on the accuracy of the system count.

Immie
Actually our touchscreen machines do this. I was talking about the receipt for voting that another poster suggested. We cannot give those, it is against the law because of the reason I gave...
 
Btu I am pretty sure the touchscreen ones are programmable. Now if the code were burned into a rom and soldered into the board with no external means of reprogramming. I know computers , that is why I don't trust a programmable device for balloting.
I don't really agree with the hand out slip either, but perhaps a final verification of your choices before you commit your votes. And a hard copy trail that is updated as you vote, not dumped later.
 
Thats truely amazing...
Dixie names a thread "A question for pinheads"...and actually gets them all to answer.....damn...you are good Dix...
 
As some of us stated. It is an honor to be called a pinhead by Dixie.
Sort of like the devil disagreeing with ya.
 
I disagree. If you take random samples of precincts and compare the electronic count to the printout of the machines you would know if there had been any source code problems which would lead to investigations and major troubles for the culprit.

For example say they sampled 1500 precincts in the next presidential election. The machines showed significant errors in 800 of those precincts. It would lead to an investigation and I am sure a review of the source code.

I have no problem with reviewing the source code, but I am not sure that, that alone is sufficient to guarantee integrity of the system.

Immie
Random errors have a Normal distribution. A skewed data report would likely resemble a Weibull distribution (a long tail on one side). It wouldn't take 800 of 1500 to reveal a bias or flaw. Easy to do a non-parametric analysis (Kolmogorov - Smirnoff test) of the cumulative distribution function of the data and compare to the cdf of Normal +/- K bands. In some of my research, I've had to back into the sample size needed to generate certain sensitivity / ability to discrminate between Normal and Weibulll. I have found that, if all the samples distribute Weibull, then it takes less than 25 samples to discriminate between the two distrinutions. However, we can not assume that every sample has been manipulated, and would thus distribute Weibull.

So, since the width of the Kolmogorov band "d(a,n)" is nearly the same for an infinite number of points as it is for 120 data points, we could probably get a decent indication using as few as 120 random samples nation wide.

The analysis of this could be done by any competent statistician, or you could just plug it into off-the-shelf software (e.g., MiniTab).
 
Random errors have a Normal distribution. A skewed data report would likely resemble a Weibull distribution (a long tail on one side). It wouldn't take 800 of 1500 to reveal a bias or flaw. Easy to do a non-parametric analysis (Kolmogorov - Smirnoff test) of the cumulative distribution function of the data and compare to the cdf of Normal +/- K bands. In some of my research, I've had to back into the sample size needed to generate certain sensitivity / ability to discrminate between Normal and Weibulll. I have found that, if all the samples distribute Weibull, then it takes less than 25 samples to discriminate between the two distrinutions. However, we can not assume that every sample has been manipulated, and would thus distribute Weibull.

So, since the width of the Kolmogorov band "d(a,n)" is nearly the same for an infinite number of points as it is for 120 data points, we could probably get a decent indication using as few as 120 random samples nation wide.

The analysis of this could be done by any competent statistician, or you could just plug it into off-the-shelf software (e.g., MiniTab).

Since it's highly likely that the raw data "as few as 120 random samples nation wide" is available, why not do something like that in your spare time, if you ever have any, and let us know what you come up with. I would certainly be interested to see what you get.
 
Troglodyte 27 wrote:
Random errors have a Normal distribution. A skewed data report would likely resemble a Weibull distribution (a long tail on one side). It wouldn't take 800 of 1500 to reveal a bias or flaw. Easy to do a non-parametric analysis (Kolmogorov - Smirnoff test) of the cumulative distribution function of the data and compare to the cdf of Normal +/- K bands. In some of my research, I've had to back into the sample size needed to generate certain sensitivity / ability to discrminate between Normal and Weibulll. I have found that, if all the samples distribute Weibull, then it takes less than 25 samples to discriminate between the two distrinutions. However, we can not assume that every sample has been manipulated, and would thus distribute Weibull.

So, since the width of the Kolmogorov band "d(a,n)" is nearly the same for an infinite number of points as it is for 120 data points, we could probably get a decent indication using as few as 120 random samples nation wide.

The analysis of this could be done by any competent statistician, or you could just plug it into off-the-shelf software (e.g., MiniTab).

Since it's highly likely that the raw data "as few as 120 random samples nation wide" is available, why not do something like that in your spare time, if you ever have any, and let us know what you come up with. I would certainly be interested to see what you get.
If there is a bias toward one particular candidate embedded in the system (widespread fraud), this method would likely detect it, but if there are isolated incidents, this method is not real good for that purpose. You would have to do pointwise comparisons (likely an NP-hard problem - translation: a royal pain in the ...).

I'll think about it doing the simple one though.

Ornot - reality check on the methodology, please

(P.S. Yes, I am a math geek)
 
Last edited:
Back
Top