Recently the long, strange story of former Alabama governor Don Siegelman has rightly received a lot of attention. There is good reason to believe that Siegelman's prosecution on corruption charges was politically motivated. But this diary isn't about that. It's about the claim that the 2002 gubernatorial election was stolen from Siegelman in Baldwin County, where a late change in the vote count took over 6,000 votes away from Siegelman and gave Republican challenger Bob Riley a narrow victory.
As Scott Horton noted in Harper's, an Auburn sociology professor (now retired) named James Gundlach wrote a conference paper arguing that the results in Baldwin County suggested that the results had been tampered with to give Riley the win. There is no way to prove that the results were accurate -- but I don't think Gundlach's analysis holds up. Based on the evidence available to me, I think Riley probably did beat Siegelman. Here's why.
At one point on election night 2002, Siegelman had 19,070 votes in Baldwin (compared to Riley's 31,052). In the final returns, Siegelman's total fell to 12,736 votes. Where did the votes go, and why?
In his paper, Gundlach does not assert that the first set of figures was correct: it adds up to more votes than ballots cast.* Instead, Gundlach hypothesizes that in each precinct in Baldwin County, someone intended to take 20% of Siegelman's votes and give them to Riley instead. But, he suggests, the hacker accidentally added the votes to Siegelman's totals as well as Riley's, instead of subtracting from Siegelman's totals. That is, someone tried to give over 3,000 of Siegelman's votes to Riley, but instead gave an extra 3,000 votes to both candidates.
*UPDATE: Someone apparently Jim Gundlach points out in comments that in fact there were three sets of totals; the first two sets both contained more votes for governor than ballots cast. I left that out as immaterial (especially because Gundlach's conference paper doesn't discuss it), but perhaps it offers further reason to doubt Gundlach's conference paper hypothesis of someone with a laptop in a nearby room fiddling around with a spreadsheet. Maybe some find it parsimonious to attribute two consecutive bizarre results to malevolent yet incompetent human intervention, but that would not be my first thought. Also, many people who express shock, shock that the numbers were changed ignore or downplay the fact that (by every report I can find, including Gundlach's) the numbers that gave Siegelman the win clearly were wrong. Of course, the final official numbers could be wrong as well.
Inconvenient facts
One problem with this story is that according to a contemporaneous report in the New York Times, the changes were confined to a single voting district:
The trouble started election night in Magnolia Springs, a town near the Gulf of Mexico, where unofficial poll numbers were released to the news media that erroneously put the incumbent ahead. The mix-up was straightened out a few hours later.... State election officials said there was no way he could have won the 7,000 votes he was at first reported to have gained from the Magnolia Springs precinct. Only 1,300 votes were cast there.
Jeffrey Gettleman, "Full Recount Asked in Race For Governor Of Alabama", The New York Times, 11/8/02
I'm not an investigative reporter, but I haven't found anything to challenge that account (except that the "7,000" seems overstated: probably it was 6,600+). Nothing in Gundlach's idea would account for over 6,000 extra Siegelman votes in Magnolia Springs -- and only in Magnolia Springs.
A second problem is that if Gundlach were literally correct, Siegelman could not (in the final returns) have received over 80% of the vote in any voting district. But at Bay Minette Intermediate School, Siegelman received 165 votes out of 192 ballots cast (86%).
The "outlier" that isn't
Well, what is Gundlach's evidence? First he claims that Baldwin County becomes an outlier in the final returns -- specifically, that when Siegelman's vote share in 2002 is plotted against his vote share in 1998, Baldwin County ends up farther from the best-fit line than any other county. This would suggest that the "correction" actually made the data wrong!
Gundlach is not entirely wrong here: Baldwin does move farther from the best-fit line in the final returns. However, it doesn't become an outlier, or even farthest from the best-fit line.
In the plot above, each 'point' represents one of Alabama's counties, except that the "I" shows Baldwin County in the initial returns, and the "F" shows Baldwin in the final returns. The dotted line shows where Siegelman's vote share was the same in 1998 as in 2002; you can see that he did worse in almost every county. The solid line is the Ordinary Least Squares (OLS) best-fit line, of course leaving out the initial returns from Baldwin County. Baldwin simply isn't farthest from this line: five counties lie farther below it, and four counties lie farther above it (which conceivably could mean that someone stole votes for Siegelman there). There are other ways of drawing a best-fit line, but there isn't any that would make Baldwin an "outlier."
The "anomaly" that isn't
Gundlach's other argument is based on results at the voting-district level. Basically, he says that he matched up (as closely as possible) voting districts from 1998 and 2002, in Baldwin County and in two other counties for the purpose of comparison. He says that Siegelman's vote shares are very highly correlated across years both in Baldwin County and in the other two counties. But, he adds, the slope is less in Baldwin County than in the other two counties: Siegelman's 2002 vote share doesn't climb as quickly. He takes this as evidence that some fraction of Siegelman's votes was shaved.
One problem with that, again, is that the stories about Magnolia Springs seem to contradict it. Another is that I could not replicate Gundlach's results, and I couldn't get anywhere near the correlation of 0.983 that he reports for Baldwin County (see the table on page 6). Vote shares tend to be highly correlated, but not that highly correlated. (Matching voting districts based on their names is crude enough that I wouldn't expect to get exactly the same results Gundlach did -- but that 0.983 is damn strange.)
But the biggest problem is that the argument doesn't make much sense to begin with. Gundlach seems to assume that the vote share slopes should be the same in every county -- or at least that if the correlations are high, the slopes should be similar. But he presents no evidence for any part of that. I tried a similar county-level analysis for several statewide races in Minnesota in 2004 and 2006, and the slopes varied considerably even between counties with similar high correlations. So, what Gundlach points to as an anomaly doesn't appear to be one, even assuming that he got the numbers right.
Conclusion
Assessing allegations of "anomalies" in election returns can be a frustrating task. Sometimes the anomalies are real and dramatic, as with the undervotes in Sarasota County, Florida in 2006, or the infamous "butterfly ballot" in 2000. More often, it seems that the critics haven't taken the time to establish the empirical baseline against which we could judge whether a result is unusual. When people are predisposed to think that an election was stolen -- and given what has happened to Don Siegelman since 2002, why not? -- the substantive debate can enter a death spiral. There is no way to know for sure whether Bob Riley really received more votes in 2002. I can only say that Gundlach's arguments that Siegelman was robbed aren't at all convincing, and so far I don't see better ones.