Click here to read an important correction in which the author partially reverses his original conclusion.
Last weekend the Miami Herald ran a fascinating story about what might have happened in a "flawless" Florida election. The article highlights a study done for the paper by Stephen Doig, a journalist-turned-academic who specializes in computer-assisted research. Based on Doig's precinct-by-precinct analysis of 185,000 uncounted Florida votes, Al Gore would have defeated George W. Bush in Florida by 23,468 votes if every voter had succeeded in voting the way he or she intended to vote. Doig's evidence is compelling and goes a long way toward justifying Gore's refusal, thus far, to concede the election. Gore not only won the popular vote nationally; more people left Florida voting booths thinking they had voted for him than did for Bush.
But Doig tries to project a notional result of the Florida election by speculating about the likely intent behind both "undervotes" (where tabulating machines picked up no choice for president) and the larger number of "overvotes" (which were rejected by the machine because a voter punched two or more holes). The assumption that both undervotes and overvotes would be likely to mirror the precinct-by-precinct breakdown for the candidates is certainly plausible. But it's fanciful to include overvotes, which are spoiled and can't be counted under election laws anywhere in the country. Thus the Herald article may have answered the question of which candidate should have won the presidency in an ideal world of perfect elections. But it fails to answer the more immediately relevant question of which candidate might win the election in the increasingly unlikely event of all legally valid votes being counted.
To try to answer this question, you have to do pretty much the opposite of what the Herald study does. You have to start by tossing overboard the overvotes, which represent approximately two-thirds of the 185,000 uncounted ballots. Then you must focus on the undervotes, which may or may not contain a legally valid vote for president.
The Bush legal team spent its weekend in Judge Sauls' courtroom arguing that undervotes typically represent a conscious decision by a voter to choose no candidate for president. As I argue here, this is a fallacy. In fact, there exists the basis for a pretty good guess about how many of the Florida undervotes represent an intentional no-vote and how many of them represent a vote simply not counted. In the 37 Florida counties that used the superior optical-scan voting system, only .3 percent of ballots, or 3 out of a 1,000, contained no vote for president. In the 15 counties that used the Votomatic punch-card voting system, the percentage was 1.5 percent, or 15 ballots out of 1,000, according to the New York Times. In the 13 counties that used the somewhat less unreliable Datavote punch-card system, the average was .9 percent, or nine votes out of 1,000. In the absence of any compelling argument about why punch-card voters should be less likely to cast ballots for president than optical-scan voters, it seems fair to assume that the difference in those numbers represent an attempt to vote for president. In other words, you can assume that for every 15 Votomatic voters whose ballots failed to yield a vote for president, only three of them, at most, intended to not vote. Blink twice if you're still with me.
It also seems plausible to assume that if it were possible to count all the undervoted ballots that are not intentional nonvotes, these votes would break down according to the percentage of machine-counted votes in each county. Evidence: In Broward County, the one large county that carried out an aggressive hand recount, trying to read as many dimpled votes as possible, this was the case. The result from 1,721 votes picked up in the hand recount nearly mirrored the results of the nearly 600,000 votes tabulated by machine. Gore got 66 percent of the machine-counted votes while Bush got 30 percent. In the hand recount, Gore got his same 66 percent while Bush actually did slightly better, winning 33 percent. In this case, at least, the undervotes closely tracked the results of counted votes.
Now for some arithmetic. For each punch-card county, subtract the baseline of .3 percent (the average number recorded under the superior optical-scan system) from the undervote. Then award the remainder of the undervotes proportionally according to the machine-counted vote for the two candidates. As an example, I'll use Sumter County, a Votomatic county that had 23,023 total votes cast and reported a 2.58 percent undervote rate in the presidential race. Multiply the 23,028 by 2.28 percent--the total undervote minus the share of the undervote attributable to intentional nonvoting for president. That yields at total of 525 Sumter County undervotes that could be legally counted, at least in theory. Break those down according to the vote shares won in Sumter County by the two candidates--41.8 percent for Gore and 52.7 percent for Bush. That means an additional 220 votes for Gore, an additional 277 votes for Bush, and a net gain of 57 for Bush.
Now let's perform the same math on the 18 of the 28 punch-card counties that reported separate undervote percentages (as distinct from a single tally for undervotes and overvotes). By my calculation, Gore would gain 2,520 votes in these 18 counties. The other 10 punch-card counties that I'm not including for lack of an undervote statistic would not be likely to change this number much. The 10 are all Datavote counties that are either very small (under 10,000 votes cast in total) or that had a very small percentage of votes that weren't counted by the tabulating machines in the first place.
After the second machine count, but before including the results of any hand recounts, Bush was ahead in Florida by 961 votes. Absentee ballots increased his margin by 630. So based on this analysis, if all the undervotes that could theoretically be counted were counted, Gore would win Florida by 929 votes. This is the best answer I can come up with to the question, "Who would win Florida in the case of an ideal statewide recount?"
The problem with that question is that there is no such thing as an ideal recount, at least with a voting system as flawed as Florida's. Those 2,500 additional Gore votes are potentially legal and valid votes, since a dimple or hanging chad can be counted according to Florida's voter-intent standard. But they exist fully only in the ether--there is no earthly means of capturing them all. Most of these votes are, in practical terms, not recoverable because they don't contain readable impressions on their chad.