Bob Fitrakis Talks out of Both Sides of his Ass

Self-appointed guardian of the vote and stolen election uber-troll Dave Hickman has been spamming the blogosphere again, at least when he's not cozying up to right wing bloggers. He put us on to a Bob Fitrakis post, reproduced here on ProgOH, that supposedly debunks a Dispatch article about how Stolen Election Guy (and in particular Stolen Election Documentarian Guy) is immune to evidence.

The Fitrakis article is mostly the usual collection of dark mutterings, half-truths and apple/orange swaps. But this in particular stood out:

For example, they [the collective personas of reporter Mark Niquette, apparently] report without embarrassment that the official
response for the highly accurate exit polls being wrong in Ohio - and
so outside the margin of error that it would only happen in one in
959,000 presidential elections - was that "…exit polls are based only
on responses from voters who agree to participate."

The Dispatch's own polls, that they brag about as being highly
accurate, are based on only those who agree to participate.

Hmmm. Where have I heard Fitrakis talk about the Dispatch poll before? Oh, yeah:

Despite Democratic victories in five of six statewide partisan offices, an analysis by the Free Press shows a statistically implausible shift of votes away from the Democratic Party statewide candidates on Election Day, contrasted with the results of the Columbus Dispatch's final poll. The Dispatch poll predicted Democratic gubernatorial candidate Ted Strickland winning with 67% of the vote. His actual percentage was 60%. The odds of this occurring are one in 604 million.

This is the essence of Fitrakis. Evidence is selectively important based on whether it supports a pre-determined conclusion. Today the Dispatch poll is to be mocked. In 2006 it was gold.

Fitrakis falling back on the Dispatch tracking poll in '06 was the point at which I went from thinking that Fitrakis tended to overclaim his evidence to thinking he's a publicity-happy hack. In arguing that this one tracking poll proved fraud, Fitrakis ignored the results of every other tracking poll, plus the exit polls that reflected the final result. Three problems here. First, his claim of a one in 604 million result based on one poll falls apart when you mix in the other polls. If the Dispatch poll was accurate, and Strickland in fact won by 67 percent, it means that all other polls are wrong, and that wrongness is similarly statistically impossible.

Second problem explains all that. Pollsters offer (to us mortals, anyway) two figures about how reliable the poll is. One is the margin of error. Fitrakis's "calculations" are based on how far out of the MOE a final result is. The other measure, the confidence interval, functions an estimate of the chance that the poll itself is accurate. Usually it's something in the ninties, so for example a 95% confidence interval means a one in twenty chance that the poll is a big pile of hooey. A longshot chance to be sure, but far more likely than one in 604 million. (h/t Redhorse who explained all this in the earlier incarnation of PBD.)

The third and biggest problem is that the "evidence" for fraud in 2006 is the dead opposite of the evidence for 2004. Prior to the '04 election all the tracking polls showed Kerry trailing, though within the MOE. The result matched the tracking polls, but not the exit polls. I'll accept someone arguing that one or the other type of poll is more reliable. But when he flogs the exit polls in '04 and one tracking poll in '06, survey says he has no credibility.

If Fitrakis and his minions stop by, their first response will be that I'm not a polling expert. Exactly. I'm not an expert and I can see the Mack truck-sized holes in his argument. If you want polling experts to explain the exit poll anomaly in '04, start here.