Please don your nerd hats folks...
In this post, labouchet suggested looking at Benford's law as a possible way of detecting vote fraud:
http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=104&topic_id=2639056#2644393Today, I ran 2 simulations with 6 million votes. I simulated 10,000 precincts with a uniform distribution between 100 and 1,100 voters per precinct. The number of voters were chosen at random and the actual vote was chosen at random, so different precincts would have different % of vote for each candidate.
In the first simulation, I made it a fair election, with the odds that 55% of the vote would go to candidate A, and 45% would go to candidate B. Then I looked at the frequency of digits showing up in the second column of the precinct totals for both candidates:
digit probability
0 0.1043
1 0.0997
2 0.0997
3 0.0977
4 0.0998
5 0.1022
6 0.1015
7 0.0998
8 0.1016
9 0.0938
In the second simulation, I made it a stolen election. The premise was that the voters would split evenly 50/50 for both candidates, but software would take 1 in 10 votes and put it in the Candidate A column, resulting in the same 55%/45% outcome as the fair election. I also re-seeded the random number generator so the number of voters and voter intents were the same as the first simulation. Here are the results for that simulation:
digit probability
0 0.1021
1 0.1017
2 0.1002
3 0.0993
4 0.1018
5 0.0997
6 0.0998
7 0.1055
8 0.0961
9 0.0939
Executive Summary:Benford's law, while useful in detecting human changes made in precinct totals, is not going to help find fraud commited electronically by stealing every Nth vote.
This does not mean it may not be useful in finding changes made at the central tabulators.
All you folks working one precinct at a time, one county at a time are our best hope. I thank each and every one of you.