BBV: A Statistical Analysis of Possible Electronic Ballot Box Stuffing

[ramblin_dave]

A Statistical Analysis of Possible Electronic Ballot Box Stuffing, The Case of Baldwin County Alabama Governor’s Race in 2002

By James H. Gundlach, Auburn University

Paper Presented at the Annual Meetings of The Alabama Political Science Association, Troy, Alabama, April 11, 2003

In this paper I demonstrate how relatively simple statistical techniques can identify apparent systematic electronic manipulation of voting results. This paper consists of four parts. The first part is an overview of the election; the second part is an analysis of county level data that suggests that both sets of results from Baldwin County are anomalous. The third part of the paper is a set of analyses of results from voting districts that identifies and describes some clear patterns in the anomalous Baldwin County final results. The final part of the paper discusses the possibilities of electronic vote manipulation and suggests mechanisms for preventing it in the future.

Full article - pdf format

[Eloriel]

1. Oh, this looks yummy

I hope it's not too, well, uh, mathematical. (Never mind, I know it is.)

So where's our favorite resident election statistician?

Eloriel

[BevHarris]

2. More and more info is rolling in. Thank you for this!

And now duh.

[DEMActivist]

3. Ooomf! There it is!

The final point that raises suspicions is that there should be no way to produce two different results with the computerized vote tabulation. That is, the system should not allow access to computer code or procedures that can produce different results. Computers do not accidentally produce different totals. Someone is controlling the computer to produce the different results. Once any computer produces different election results, any results produced by the same equipment operated by the same people should be considered too suspect to certify without an independently supervised recount.

[grasswire]

4. can someone summarize...

...for those of us who can't open PDF files because of certain firewall quirks?

Thanks a lot.

[DEMActivist]

5. It's actually pretty simple

He takes the 1998 numbers and compares them to the 2002 numbers.

Then he takes the voting shift to Bush in 2000 and charts that in.

Then he charts the probabilities of the reported outcome. Final determination is:

The combined findings of a strong relationship between the 1998 and 2002 votes in Baldwin County as well as outside, and the different slope strongly suggests a systematic manipulation of the voting results. In addition, a comparison of the slopes provides a way to estimate the apparent nature of the manipulation of the results. By dividing the Baldwin County slope, .697, by the slope for the other voting districts, .854, and subtracting the results from 1.00 you get an estimate of the proportion of the Siegelman vote in each voting district that that apparently disappeared from the official Baldwin County results. This yields a result of .18, which is about half as much as predicted from the hypothesis based on the first reported results. This raised the question about how could a process of moving X number of votes from one candidate to the other results in the mysterious production of erroneous results that were 2X above the final reported results?

The answer is surprisingly simple. This is a common error pattern to appear in programming spreadsheet calculations. My hypothesis is that someone was moving a little more than 3,000 Baldwin County votes from Siegelman to Riley by calculating a fifth of Siegelman’s votes in each voting district, rounding it to a whole number, adding the resulting value to Riley’s votes in that district and then subtracting that number from Siegelman’s vote. However instead of subtracting the calculated number they added it to the vote for Siegelman. This is a common error created by using copy and paste to produce the invisible formulas for cells of spreadsheets. The result was a first report of county vote totals that had percentage distributions close to what was expected but a total vote that much higher than expected. Once they went back and fixed the procedure so that it performed as they desired, a reasonable total vote and Riley winning the election, the difference between the first and second reporting of Siegelman’s vote was twice the number of electronically shifted votes. If what I hypothesized happened, then the total votes for Baldwin County was 27,866 votes for Riley and 15,283 votes for Siegelman. This would have produced state totals of 669,039 votes for Riley and 671,652 votes for Siegelman. The only way we will know for sure is if the paper ballots for Baldwin County are recounted.

[LearnedHand]

6. Summary:

Gundlach (statistician, Auburn University) became suspicious of voting results in Baldwin County AL when election supervisors reported two separate, different sets of returns. He conducted a number of statistical analyses with freely available data and determined in two major ways that the results were most likely tampered with. He includes scatter graphs in his text showing deviations about one-third beyond the norm for certain candidates.

He guessed from his results that someone manipulated the returns, then discussed several ways in which this might have happened:

1. The cartridges carrying the raw data could have been physically tampered with between the precincts and the central tabulation office.

2. Someone could have installed a worm or virus to change data in the initial data stream (between the voter's selection and the writing of that vote onto the cartridge).

3. Someone could have accessed the compiled votes on the tabulating computer and changed the data (which would have required access to the machine and a keyboard/monitor for the tabulating computer).

4. Someone could have installed an "802.11 card, along with enabling software..."), to change results on the tabulating computer via a laptop computer in a nearby room. This was, in his estimation, the most likely scenario.

[DEMActivist]

7. A slight correction...

#4 wasn't the "most likely scenario" but his preferred method.

His exact words were "the one I would take if I were to do it,"

[LearnedHand]

8. Oh, right.

Sorry.

[shirlden]

9. kick

n/t :kick:

[DEMActivist]

10. Kick, important stuff

.

[BevHarris]

11. Statisticians are another arrow in our quiver

Because so many of us are weak in math, we get a little glazed looking at their work. Yet look at all the areas where they can be helpful in this:

Georgia anomalies
Baldwin County anomaly
San Luis Obispo - validation that those were the real votes

Somehow pollsters have managed to make statistics understandable to the masses. We face the same task in these other areas.

Bev

[leftchick]

12. kick

:kick:

[ramblin_dave]

13. Has anyone done an analysis of the Georgia 2002 election?

What this guy at Auburn did for the Alabama election should be applicable to the 2002 election in Georgia. Are the Georgia results available online? Has anyone tried such an analysis?

[never cry wolf]

16. It would be interesting to know

how Baldwin County did in the Senate race, were there 2 results there as well??

[ramblin_dave]

18. There were no changes in the Senate race

In fact the only total that changed was the total for Siegelman (D) and by enough to reverse the outcome. The totals for Riley (R) and the Libertarian did not change.

[DEMActivist]

20. I'm trying to locate the author

if anyone has suggestions on method to do so, they would be greatly appreciated.

I don't even know if he's a student, undergrad or professor at Auburn.

[ramblin_dave]

22. He's a sociolgy prof listed here...

http://www.auburn.edu/academic/liberal_arts/sociology/faculty.htm

[DEMActivist]

23. Thank you! I've sent him an email.

Asking for help.

[kimchi]

14. kick

[gristy]

15. I'm printing the paper now

I look forward to reading it. Am also now listening to Democracy Now. Bev should be coming on soon.

[RedEagle]

17. Kick

(I don't have time to read this now and want to make sure it sticks around!)

[ArkDem]

19. Can you find an analysis of paper ballot box stuffing?

I think there is a possibility..

[althecat]

21. Bev on Democracy Now... Link...

http://www.democracynow.org/streampage.pl

She comes in on Real player at 33.30... am listening now..

[ProfessorGAC]

24. Ummmmm! I Don't Know About This

If someone turned this paper in as work in applied statistics, they'd get a C, at best!

There are some incredibly broad assumptions that are wholly untested, without which the conclusions are not reachable.

In addition, there are some errors in statistical interpretation within this paper.

Also, he provides no confidence boundry for the various fits, and i even question some of the techniques used in the early part of the paper. These fall into the "what difference does this make?" category.

Also, the causation logic requires an enormous leap of faith that the writer is an expert in the field of digital data transfer. Yet, no credentials of such are provided.

Lastly, the inference that the problem is traced to a spreadsheet formula transfer error is original, but completely unsupported by any of the data presented. That's a guess that happens to fit the hypothesis.

I'm not saying that nothing dirty happened in the AL election. I'm just saying this is fairly "squishy" proof.
The Professor

[ramblin_dave]

25. The guy does have a PhD from Univ of Texas

So I wouldn't be so quick to badmouth his analysis.

The Alabama election results are online for anyone to examine and challenge what this paper claims:

http://www.sos.state.al.us/downloads/dl2.cfm?div1=Elections%20Division&types=Data

[ProfessorGAC]

26. Whatever You Say, Dave

However, i lecture on statistical mechanics around the world, and i build statistical models to explain the behavior of complex industrial systems.

So, i know of whence i speak. This is NOT a sound statistical analysis.

Just remember, Lynn Cheney and Condi Rice have Ph.D.'s. Since this is one of my fields of expertise, i will choose to dismiss his analysis, thank you very much.
The Professor

[Eloriel]

28. Well, then what can you turn out here?

Or do you have no interest in assisting this project?

Eloriel

[T Roosevelt]

27. Here's a suggestion

I'm sure that we have ample statistical experience here at DU (we have at least the Prof) that we could use this paper as a tentative first step in conducting some analyses of various votes across the country (the 18,181 results spring to mind, as does the GA governor race).

I know that I've expressed interest to Bev regarding development of some statistical techniques to help identify races that may have been tampered with - my initial thought is from an engineering quality control (I'm an engineer working on my PhD), but I recognize this is likely in and of itself insufficient to identify purposeful tampering.