First, Back to Ohio is one of the few balanced pieces on the Web regarding the Kennedy article in Rolling Stone, so thank you.
However, there are certain facts that have been discussed really exclusively among those of us who practice the gematria called statistical analysis, and these you may not be aware of. The person who developed the explanation for the exit poll discrepancy for Warren Mitofsky is a graduate student (or, hopefully, former graduate student, since she submitted her thesis a few months ago), Elizabeth Liddell.
Elizabeth and I have had amicable discussions regarding what her work does and does not prove. The whole nine yards is deliveredat a forum established by Bruce O’Dell, but here’s a synopsis.
First, I stipulate that the reluctant responder hypothesis could explain the observations in Ohio. However, I have pointed out to Elizabeth that at that point, her reasoning does tend to get a little circular thereafter.
Also, the quality of the data is probably not good enough to draw hard conclusions. Unfortunately, Warren Mitofsky has withheld the raw data from public view, so that’s not certain. At any rate, Elizabeth does concede that the lower rates of completion in Democratic precincts are not what one would expect with the shy Republican hypothesis, but argues this might be within statistical error. My view is that it’s fine to speculate, but it would be much better to produce an estimate of exactly how unlikely that is.
Here are some specific points I have raised to Elizabeth:
a. If the hypothesis is correct, it should be tested in, say, neighboring Illinois and Indiana. Illinois went strongly for Kerry, Indiana strongly for Bush.
b. An eyeball analysis of the national data very strongly suggests to me that the hypothesis cannot be generally proven. While the precinct-level vs. state-level comparison is a bit apples and oranges, if the hypothesis is generally true, it should help to explain some of the observations.
c. There is some academic work on the shy responder hypothesis. Elizabeth cites the UK experience. There’s also some US research to support the idea, which I cite in the forum debate. However, the research is far from unambiguous and it may be dead wrong. There are also reasons to think that 2004 was different. The DNC report clearly describes an atmosphere of intimidation of African Americans. The report of the Texas Strike Force you mention– but also challengers at the polls, mass disenrollments, the very public savaging of an employee of the NAACP, and so on created an atmosphere of hostility. I think it’s more likely African Americans would fear talking to a stranger. One must ask why Warren Mitofsky– who has talked and talked about this issue– hasn’t contracted research on it, or requested that agencies like Census or CDC, who have to deal with the problem, make public their own research.
Next, Mitofsky’s failure to make the data public has fueled a lot of the rancor. I must say that I’m not favorably impressed at how he has hidden behind his respondents as a shield, and think, as I have said before, that the way he has approached this inevitably raises suspicions of professional misconduct.
Some of the rancor arises because Mitofsky has such disproportionate resources at his disposal. He’s an employer in the area of polling, creating potential conflicts of interest. As I understand it, Elizabeth was paid for her work, while people like Steve Freeman and Josh Mitteldorf are working pro bono; the latter feel they are being shouted down. As far as I am concerned, Elizabeth’s algebra speaks for itself: it’s correct. But the optics are terrible.
On Mark Blumenthal, I’m sorry, but–without in any way impugning his integrity– he is not what I would call a reliable source. He doesn’t hold an advanced degree, and his experience seems to be more in field polling. While I in no way hold the lack of a degree against Mark–since I write under a pen name, I would be a hypocrite if I did–his lack of formal training shows in his reasoning and his style of debate. I have, for example, panned Mark for claiming that O’Dell “eviscerated” the analysis of Mitteldorf et al. when the reality is that O’Dell, like most of the participants in the debate, was aware that between white and black, there is a shade called gray and said so.
Finally, I did some work on the Moyer/Connally story, which I didn’t publish because I felt that it was interesting but inconclusive. While what Kennedy describes is certainly not proof of skullduggery, Bob Fitrakis and Richard Hayes Phillips recently did some precinct-level ballot examination that is very suggestive that there was tampering. For example, they found that in one rural Republican precinct, voters who voted straight Republican voted not to illegalize gay marriage. Either there was one h–l of a mix-up in the Christian Coalition printing office, or– as Fitrakis and Hayes suggest– those ballots were generated in another precinct where, due to rotation of ballot position, the Republican and Democratic votes would have been switched. It’s this kind of close-to-the-earth full-on green eyeshades analysis that has been missing.
When all the facts are brought forward, I think the balance swings toward Kennedy’s view that Ohio was stolen. Through analysis of crossover and spoilage patterns, I have argued that there was widespread ballot tampering in Florida. But of course, unless there is a serious investigation, we’ll continue to have these debates which end up convincing no one. In my opinion, there will never be real voting reform until an investigation uncovers misdeeds so serious that the Congress can no longer play ostrich.
I have to shift gears and take care of other duties, but we can discuss the recent Fitrakis work or other topics if you are interested.