One of the joys of writing about politics for an online magazine is the sophisticated, nearly instantaneous feedback one gets from readers. When I arrived at the office today, I had e-mail messages waiting for me from a couple of professional statisticians as well as some shrewd amateurs. In the nicest possible way, they attempted to pick holes in the piece I posted last night about who actually won Florida.
So thanks to Andrew Lazarus, Jay Olsen, Lars Phillips, John Veboncoeur, and Stephen Feinson for helping me recognize two significant errors in my calculations.
The first was a dumb mistake. The 961-vote lead I said that George W. Bush held after the statewide machine recount did not include results from Palm Beach County, which was under court order not to certify its results when Katherine Harris released her second tally Nov. 10. If you include the results of Palm Beach County's machine recount as reported by the AP, Bush's lead before the hand recounting began shrinks from 961 to 327. The 634-vote difference means a more decisive answer to the question, "Who would have won Florida if we could count all the undervotes?" In that idealized circumstance, my model suggests that Al Gore would have won Florida by 1,563 votes, not 929.
My second mistake is slightly more complicated. I calculated that Broward County recovered 35 percent of its undervotes in its hand recount (1,721 of 4,939). But this number neglected to factor in my assumption that 3 out of every 1,000 votes represent intentional nonvotes, which are intrinsically unrecoverable. When you throw that in, Broward was actually able to recover 54 percent of its undervotes that weren't nonvotes (1,721 of 3,175). Extrapolating statewide from that percentage, Gore would have picked up 1,366 undervotes instead of 882--if every county had recounted its undervotes by hand using a Broward County standard that says yes to dimpled chad.
Then, I'm sorry to say, I compounded these two errors, deducting the 882 from an erroneous Bush "machine" lead of 1,591 (961 in the machine recount + his gain of 630 votes in the absentee count). I should have deducted 1,366 from a Bush machine lead of 957 (327 in the machine recount including Palm Beach + 630 absentees).
And that changes the final result. If all of Florida's counties had done a hand recount using the chad-counting rules employed by Broward County, Al Gore would not lose Florida by 709 votes, as I predicted yesterday. Gore would win Florida by 409 votes!
Of course, this remains a forecast, subject to a variety of caveats and what-ifs. One important footnote is that a precinct-by-precinct analysis of the kind done by the Miami Herald would be more precise and might yield a different result. Unfortunately, the data necessary to do this doesn't exist. But I think this county-level analysis represents a plausible, educated guess about the potential result of a hand recount that now will almost certainly never happen. And I think it also explains something that yesterday's erroneous prediction didn't, namely why the Gore side was so insistent on employing Broward County's dimpled-chad rules for the recount. Using that standard statewide, there's good reason to think Gore would win the election. Using any other standard, there's reason to think he would lose.