Andrew Gumbel’s Big Florida Lie

It’s day two of the blogstorm occasioned by Paul Krugman’s claim, following Andrew Gumbel, that:

Two different news media consortia reviewed Florida’s ballots; both found that a full manual recount would have given the [2000] election to Mr. Gore.

Mickey Kaus picks it up today, paying careful attention to Krugman’s wording:

The discomfiting truth is that, if you also recounted overvotes, the NORC media recount, under several “certainty” standards, showed Gore the winner.

Using the most inclusive standards, Bush actually gained more votes than Gore — about 300 net — from the examination of the undervote ballots. But Gore picked up 885 more votes than Bush from the examination of overvote ballots, 662 of those from optical scan ballots.

What’s more, there’s strong, near-smoking evidence that if the recount had been allowed to proceed overvotes would have been counted (despite the Gore camp’s revealingly idiotic, self-defeating focus on the “undervotes”).

Gumbel himself makes this claim on Amy Alkon’s blog (in comments:)

Re the media consortia recounts, yes, they suggested the outcome was inconclusive IF RESTRICTED TO THE FOUR SOUTHERN FLORIDA COUNTIES THE GORE CAMPAIGN WANTED RECOUNTED. Had the whole state been recounted, Gore would have won by any standard. [emphasis added]

But the claim is false. The Washington Post article that Kaus links says that ambiguity remains even when the overvotes were counted, the outcome depending on what standard is used to review the ballots:

But this is one case where disagreements among the reviewers affected the outcome. Gore won under this scenario when two of the reviewers agree on the markings. Under a standard in which all three were required to agree, Bush won by 219 votes.

Gumbel claims that Rick Hasen backs up his claim in this comment:

It is true that the NORC study found that had all the state’s undervotes and overvotes been counted, Al Gore would have come out ahead of George Bush.

…but Hasen makes this remark rather off-hand in the context of a posting arguing that the election was a statistical tie.

And this is in fact what the recounts showed. It’s impossible to perfectly create the same scenarios in an audited recount that prevail in a real election. The auditors used panels of three neutral parties to review ballots, and found that voter intent was hard to discern, hence the discrepancy between the 2-out-of-3 standard of review and the 3-out-of-3 standard. While one can certainly argue that voter intent in the real world is discerned by 2-out-of-3 canvassers, it’s not the same system as canvassers are partisans, so in effect 2 out of every 3 canvassers represent the majority party in that county.

But that being as it is, the fact remains that we can’t say that a full manual recount of all the ballots rejected by Florida voting machines in 2000 would have conclusively given the election to Gore. Including all uncounted votes – undervotes, overvotes, and absentees – the outcome depends on the standard for discerning voter intent, and that’s what both consortia found.

Krugman lied, Gumbel lied, and Kaus’ claim differs from theirs. Kaus says: under several “certainty” standards; Krugman’s boy Gumbel says under any standard. This is not the same claim at all, of course. Kaus puts it correctly, but Gumbel doesn’t.

(And BTW, how comfortable are we with elections in which the outcome is determined by a handful of people who can’t mark a ballot so that a voting machine can read it properly? If it were up to me, there would be no manual recounts because people are more prone to error and bias than machines. But that’s another point for another blog.)

Previous posting.

More at the James B.

UPDATE: Krugman builds a wall and hides behind it.