Developing Intelligence

Update: Diebold effect explained.

Jon Stewart famously accused the Crossfire co-hosts as “hurting America” by imitating the style and appearance of political debate to disguise partisan hackery and vacuous strawman arguments.

In the case of the recent NH primary, the same criticism can be leveled at the mainstream media (e.g. this discussion of a red herring issue) – but also at some usually-thoughtful corners of the internet, including our very own Scienceblogs.

Mark at GoodMathBadMath, whom I eminently respect, has contributed trivial high-school anecdotes and parroted the essentially-refuted meme that the Diebold effect can be explained by differences in population density.

Another scienceblogger, whom I also respect, has omitted statistics on the assumption that “Demographic factors …. don’t always explain all of an election result” – but not addressed why a new peer-reviewed paper uses statistical methods to show there was not vote fraud in the 2004 NH primary, nor why the Diebold effect remains after controlling for whatever little influence demographic and geographic factors may have.

Slashdot, a site whose readers are technically adept, has fallen into the same trap, focusing on the apparent symmetry of Obama/Clinton votes across townships, which is effectively meaningless due to lack of demographic controls. Slashdot’s comments indicate willful ignorance of the statistics that have been done, or focus on patently irrelevant statistics like the R-square value for the Diebold variable.

So, to those on soapboxes: please stop hurr-rrr-rrrting America, do a little reading, and discuss the real issues:

- The aforementioned statistically-significant anomaly known as the Diebold Effect;
- The REMARKABLE “chain of custody” issues with the recount, such that a sole private contractor with minimal oversight handles ALL paper ballots;
- Encouraging those with statistical expertise to analyze the data;
- Or, most productive of all, the acquisition of additional demographic and geographic factors to be used as covariates, such as race, gender, and precinct-by-precinct 2004 NH primary voting patterns.

This frequently-replicated statistical anomaly warrants intelligent, reasoned, and informed discussion – which has been, and continues to be, disappointingly absent.


  1. #1 Brian Mingus
    January 17, 2008

    Even if there is voting fraud it is only of academic interest at this point. Not only is there no way to prove it, but the media blitz that this is all really about is already over. Candidates want to win the front loading states to initiate a snowball effect that will help them win their party’s nomination on super duper tuesday. Statistical significance here is not going to change that outcome…

    That aside, I read the study funded by California where they were easily able to crack the Diebold machines. The techniques were excessively simple and this has been shown more than once. You can’t use your experience as a company who builds ATM’s to build voting machines. There is WAY more at stake than a stack of 20 dollar bills.

  2. #2 vanderleun
    January 17, 2008

    Dear Brian,
    It may have escaped your notice that the primary elections do not mean we are done with elections this year. Since we will have, in November, an important and determinative election, it would behoove the country to understand what is or is not going on with Diebold election machines since they will doubtless be used again.

  3. #3 CHCH
    January 17, 2008

    Agreed with both of you. In many ways, understanding this data may be our best opportunity to understand the Diebold effect, because on non-optical scan machines, any apparent Diebold effect is confounded with issues related to human computer interaction. So I think it’s of more than academic interest.

    But I do think brian has a point – most people fail to understand the nature of the statistical tests, and the mainstream media will in all likelihood NOT cover these issues. Even arstechnica downplayed the stats in favor of the (godawful) descriptive approach (i.e., the supposed “symmetry” of votes, which is meaningless for reasons mentioned above).