Part Two: Retired NSA Analyst Proves GOP Is Stealing Elections

by Denis G. Campbell and
Michael Duniho

Retired NSA Analyst Michael Duniho’s work was the subject of Part One. Overnight he finished a detailed essay and further testing. His focus and 9-detailed charts are from Pima County, Arizona where he resides.  

This is a developing story. This website has been under consistent and repeated attack for the Part One in this series. We encourage readers to download the spreadsheet, acquire data from their own elections board and run these same analyses and report back. If we’re wrong, we’ll admit it. So far, it’s been spot on. Someone is doing their damnedest to make sure your vote is manipulated.

When you publish something like this story, the other side does its best to completely dismiss and discredit it. Charley James and I spent hours and I a sleepless night making sure we got this right.

This publication is known for a commitment to depth. We are told time and again that UK Progressive is where people come to get a complete story. So it is our pleasure to reproduce in full the essay Michael wrote detailing his work. Yes it is partly to silence the comment trolls living in denial and we have a responsibility to our readers to go beyond the original 1,600 words.

Our readers also hold our feet to the fire and when two sections were found to be confusing, we updated the article to make sure it was clearer. For this commitment on your part to make sure we get it right, our gratitude knows no bounds. -Ed.

Evidence of Vote Counting Fraud in Pima County?

by Michael Duniho

A paper published by Francois Choquette (Aerospace Engineer, Statistics, California) and James Johnson (Senior Quantitative Financial Analyst, California) outlined anomalies found throughout the United States in the Republican Presidential Primary that always favored Mitt Romney. The favoritism correlated strongly with precinct size, and did not correlate with any other logical choices (such as population density, income levels, race, etc). The paper asked readers to confirm their analysis and report on findings.

I analyzed the 2012 Presidential Preference Primary in Pima County and confirmed their analysis. I also analyzed two 2012 Board of Supervisors Republican primaries and found no apparent anomalies. I analyzed many 2010 races and found anomalies in all statewide races, always apparently favoring Republicans (except for propositions). I found no anomalies in the LD28 State Senate race.

The correlation of Republican (or in the case of the Republican Presidential Primary, with Mitt Romney) strongly suggests vote count fraud, which would occur in larger precincts because it is easier to hide it there.

Pima County has persistently refused my requests to sort early ballots before doing the state-required hand-count audit because they say it is too much trouble. These statistics suggest another reason: they may be hiding vote-counting fraud.

We need to re-examine early ballots for the 2010 election, if they are still available, to confirm the hypothesis of vote-count fraud or to certify that there was none. This would involve removing the ballots from the vault under a court order, sorting them by precinct, and then hand counting some of the larger precincts. It appears from the data that we might discover vote-count fraud in every state-wide race in that election.

We need to re-examine the Republican Presidential primary early ballots in the same way, sorting them by precinct and then hand counting some of the larger precincts.

If the contention of vote-counting fraud is correct, we could be victimized by vote-continuing fraud in the November election, affecting not just the Presidential race but also state-wide races and possibly county-wide races. Anomalies in the 2010 general election appeared to show about ten percentage points taken from the Democratic candidate and added to the Republican candidate’s total. This changes the outcome of any race that is closer than 20 percentage points.

Here are some charts showing apparent vote flipping and others confirming no fraud in a race. Each chart shows a different race. Lines that move up from left to right represent votes added through fraud. Lines that move down from left to right represent votes taken away from a victim through fraud.

The first chart shows a race that was clearly not flipped, the LD28 State Senate race. It is not a county-wide race, so there were fewer precincts to analyze. The wiggles on the left of the chart represent small precincts with so few votes as to render statistical analysis less useful. In the larger precincts, the lines are flat, showing no votes transferred from one candidate to another.

The second chart is another race that apparently had no evidence of fraud. This is the 2012 Pima County Board of Supervisors race between the incumbent Ray Carroll and challenger Sean Collins. This race had even fewer precincts (62) than the LD28 State Senate race (92 precincts), so the lines are not perfectly flat, but they do not show any significant slope, leading me to believe there is no evidence of fraud.

The next chart is another Board of Supervisors primary race (District 1, Republican) that shows no evidence of fraud. The small number of precincts (65) causes what would be flat lines in a larger data set to appear wiggly, but again there is no discernible slope for any candidates.

The next chart is the 2010 Giffords-Kelly race, showing an apparent shift of ten percent of the votes from Giffords to Kelly. Unfortunately for Kelly, Giffords appears to have run more than 20 percent better than Kelly so that even with the shift she still won the election by a small percentage. Making an assumption of fraud, this suggests that it might have been programmed into the computer before the counting begins and could not easily be changed during the election process.

[Note: I have been advised by the author of the report I was following that I need to account for relative party registration numbers in this chart, which might lessen the steepness of the slopes if Democrats tend to be found in larger numbers in smaller precincts and Republicans in larger numbers in larger precincts. I am working on getting precinct-level voter registration data for 2010 so that I can recalculate this chart to more accurately reflect the anomaly.]

Here is the same voting data adjusted for voter registration. On this chart it appears that large precincts provided Kelly with about 6% more votes than small ones did, and they provided Giffords with about 6% fewer votes than small ones did. The voter registration data, then, accounts for about 4% of the apparent 10% shift of votes.

The following chart represents the 2012 Republican Presidential Preference Primary. It shows a slope upward for Mitt Romney and downward for Rick Santorum. Although there is always the possibility that these slopes could be explained by something such as moderate Republicans being found in larger numbers in large precincts and conservative Republicans being found in larger numbers in smaller precincts, it has not yet been possible to correlate the vote count data with a convincing and numerically precise explanation. This argues for sorting early ballots by precinct so that a proper hand count audit can be done. Then whatever anomalies are found in statistical analyses, we can confirm with statistically comforting precision that no fraud occurred in any election contest.

This chart shows the data for the 2012 CD8 Special Election between Barber  and Kelly, with no adjustment for voter registration. It appears that larger precincts provided about 10% more votes to Kelly and small ones did, and that those same larger precincts provided about 10% fewer votes to Barber than smaller ones did.

This next chart shows the same vote count data adjusted for relative voter registration for the Democratic and Republican parties in each precinct:

We can see that voter registration variations between the two parties in larger precincts accounts for some of the variation seen in the first chart, but we are still left with an unexplained shift of about six percent of the vote from Barber to Kelly when we compare small precincts with large precincts. What causes this shift? We simply don’t know. Could there be another demographic cause that we have not yet analyzed? Yes, but what would it be? We are left with the question to be answered: Could this shift have been caused by fraud in the computer? The answer to that is that it could have been, and without a proper hand count audit, we cannot rule out that possibility.






Denis G Campbell View more

Denis G Campbell
Denis G. Campbell is founder and editor of UK Progressive magazine and co-host of The Three Muckrakers podcast. He is the author of 7 books and provides Americas, EU and Middle Eastern commentary to the BBC, itv, Al Jazeera English, CNN, CRI, MSNBC and others. He is CEO of Monknash Media and a principal with B2E Consulting in London. You can follow him on Twitter @UKProgressive and on Facebook.


  1. An inlaw of mine worked for several years at the Pima County Recorder’s Office as an Electronic Data. During that time he wrote the program for scanning and counting election ballots. I’m not well versed technically,so must ask are miscounts and fraud due to programming?


  2. Interesting discussion, however I propose that the factor causing the correlation is an effect caused by variations in voter turnout rates between different precincts. Since the analysis orders the precincts by the number of actual voters, (not the number of voters registered in the precinct), the trend line may be capturing voter motivation rather than fraud or any of the socioeconomic factors which were mentioned but discounted in the original analysis.

    Here’s a summary of the analysis which I performed:

    1. Choose an election for analysis. I chose to download the raw data for the Ohio 2006 governor election, Ted Strickland vs. Ken Blackwell. I completed the same analysis for the Ohio 2008 general presidential election, McCain vs. Obama, with similar results.

    2. Perform the original analysis described by the authors, the data shows the trend lines as described, indicating an alleged fraud.

    3. Calculate the voter turnout rate in each precinct (votes cast/registered voters).

    4. Calculate the ratio of votes in each precinct for Blackwell vs. Strickland (votes for Blackwell/votes for Strickland.)

    5. Generate a scatter plot for the data (turnout vs. candidate ratio.) Observe the wedge shape and the trend line which indicate a correlation between voter turnout and a vote for Ken Blackwell.

    As further confirmation of this analysis, note that While Blackwell lost the election overall by 60-37, Blackwell had more votes in 16 of the 25 precincts with the highest turnout ratios (excludes one precinct with only 2 voters.) It’s hard to believe that this anomaly was created solely by vote flipping, as fraud of this magnitude would be too obvious not to be detected.

    In the original analysis, the number of votes in a precinct was used to order the data. Voter turnout clearly affects the number of actual voters in a particular precinct. Two precincts may have a similar number of registered voters, but a very different number of votes cast. Here is one example from the Strickland/Blackwell data set of 2 precincts with approximately the same number of registered voters, but very different turnout rates:

    Darke County – CV-Z 1045 votes cast/1240 registered voters – ratio of Blackwell votes to Strickland votes: 1.644

    Franklin County – Columbus-12-B 237 votes cast/1246 registered voters – ratio of Blackwell votes to Strickland votes: 0.1487

    Another interesting analysis to perform is as follows:

    1. Download the raw data for the Ohio 2012 primary election.

    2. Order the precincts by number of votes cast, then calculate the cumulative totals for the numbers of Democrat and Republican ballots completed. Since this is a primary, voters had to request either a Dem or a Rep ballot. This number is not actually a vote, but rather a count of the number of ballots requested. It seems that this number would be very difficult to falsify since all of the other totals on ballot must sum to the total number of ballots completed for the particular party.

    3. Perform the cumulative analysis as described in the original paper, graphing the Democrat and Republican ballot counts. Note that these graphs show the same type of trends that were noted in the original analysis, with the number of Republican ballots sloping sharply upward as the number of votes in the precinct increases.

    I suggest that the trend is caused not by vote fraudulent vote flipping, but rather by a hidden correlation in the data, such as the voter turnout metric proposed above.

    Please try this analysis on other data sets and let me know if you agree that this refutes the authors’ “proof” of fraud.

  3. Data Loving Independent

    @Doris, I agree that just because the cumulative fraction of votes is not a flat line, that is not sufficient on its own to cast doubt on the integrity of the vote. Demographics and voter motivation in each precinct are mixed into the data.

    As I mentioned on the part one article on this topic, demographics in Pima county almost perfectly describe the trend lines observed in the CD8 Special Election votes. If one uses the author’s technique on party affiliation of registered voters, the exact same trend lines appear. This explains that demographics are very likely driving the trend lines, leaving very little up for suspicion unless there are particular precincts where the ratios of the registered voters party affiliations are way different than the ratios of the parties for which votes were cast. This method is sound with the assumption that turnout is relatively similar from precinct to precinct, which is not a perfect assumption, but it is plenty sufficient to explain the trend lines.

    And the authors have not shown any exapmles of precincts where registrations of parties are way out of whack with votes cast for parties…

    Once again, I absolutely agree with the authors that independent checks of more than just a few percent of the votes must be performed to check these privately owned computers with proprietary software. Evidence of fraud has been documented in many cases, specifically the 2000 and 2004 presidential elections.

Leave a Comment