By Carl Bialik
When the votes were counted last week in the Democratic primary in Syracuse, N.Y., election officials found a tie: 6,001 votes for Sen. Hillary Clinton, and 6,001 votes for Sen. Barack Obama. Syracuse University mathematics Professor Hyune-Ju Kim said the result was less than one in a million, according to the Syracuse Post-Standard, which quoted the professor as saying, “It’s almost impossible.” Her comments were reprinted widely, as the Associated Press picked up the story.

When Syracuse election officials finish opening absentee ballots, as this California poll worker is doing, that unlikely tie may be broken. (Associated Press Photo)
But as more than one Post-Standard reader pointed out, a dead heat didn’t seem so unlikely. With 12,002 votes cast for Ms. Clinton and Mr. Obama, there were only 12,003 possible margins between them — one for every possible number of votes for each of them. (Another 344 Syracuse votes went to other Democrats.) If each margin was equally likely, each one would have a one in 12,003 shot of coming through. The result indicates that voter sentiment in the city is closely split between the candidates, so a small margin is more likely than a big difference between them. Prof. Kim told me that the chance of a tie — assuming that voter sentiment really is split — is one in 137. (Similarly, if you flip a quarter 12,002 times, you have no right to expect that it will come up heads exactly 6,001 times, because of random variation. But among all possible outcomes, that is the most likely one.)
Prof. Kim’s calculation for the Post-Standard was based on the assumption that Syracuse voters were likely to vote in equal proportions to the state as a whole, which went for Ms. Clinton, its junior senator, 57%-40%. By that assumption, the tie itself was nothing special; what was unlikely is that Mr. Obama came close to Ms. Clinton in one city in a heavily pro-Clinton state. Prof. Kim said she had little time to make the calculation, so she made the questionable assumption that Syracuse should have reflected the state-wide results for simplicity.
But even if the cliché “all politics is local” overstates the case, it’s certainly true that there are variations within regions and by types of localities. The tie looks even less likely if you zoom in, placing it in the context of Syracuse’s Onondaga County, where Ms. Clinton prevailed, 62%-36%; but more likely if you zoom out to the nation as a whole, where Ms. Clinton edged Mr. Obama by just tens of thousands of votes out of nearly 15 million.
Look at it not geographically, but demographically, and the Syracuse vote makes more sense. Mr. Obama has generally done better in cities, notably in Missouri, which he won thanks to a big margin in St. Louis County. Meanwhile, Mr. Obama won the black vote nationwide, 82%-17%; more than one-quarter of Syracuse residents in 2000 were black, compared with less than 16% in New York state as a whole.
It is always problematic to assign probabilities after events unfold; better to decide in advance what outcome you’d find interesting, and calculate that event’s likelihood. Before last Tuesday, political junkies might have said they’d find any tie interesting, not necessarily one specifically in Syracuse. By that broader measure, a tie isn’t that surprising — since Ms. Clinton and Mr. Obama were essentially tied nationally in votes, and the chance of a tie in Syracuse was one in 137.
As it happens, we’ve had ties before in presidential elections. In November 2004, both President Bush and Sen. John Kerry drew 2,200 votes in Mono County, Calif. Four years earlier, President Bush tied Al Gore at 2,025 in Cedar County, Iowa. In both cases, while the initial tie received wide publicity, the tie was broken upon a recount; Mr. Kerry won Mono County by seven votes, while Mr. Gore won Cedar County by two votes.
The Syracuse tie may also evaporate Friday, once all absentee votes are counted. Ed Ryan, commissioner of the board of elections in Onondaga County, told me: “It’s hard to believe that it will be a dead heat, but you never know.”
Hillary Clinton won the re-canvass by the way: 6,449 to 6,363.
The underlying problem with calculating the probability of a tie in Syracuse is that you have to say what that probability is conditional on. It seems that the consensus is that the right probability is 1/137 (or perhaps about half that, thanks to Verges’ point that you must have an even number of votes). But this can only be interpreted as the probability of a tie conditional on there being exactly 12,002 votes cast. Well, why is that the condition we are interested in? Why don’t we find it necessary to calculate the probability of 12,002 votes being cast? Not because we are not interested in it, but because there would be no good way to perform such a calculation. But if we were to do as Carl suggests, and calculate the probability of a tie before the votes are counted, then surely we wouldn’t have chosen to find the probability of a tie conditional on 12,002 votes. What could we have done?
It seems that the generally excepted “universe” that should be used as the condition for most public interest probability calculations is some sort of general knowledge level about the situation prior to the event, not too specific, nor too broad. For example, we accept it as reasonable to calculate the probability of a particular vote given publicly available polls at the time of the vote. Would it be too specific for us to instead accept as the condition the ACTUAL beliefs of all of the people in Syracuse, as well as their schedule for voting day (which would be relevant because it would affect their likelihood of finding time to vote), even though this was not knowable at the time of the vote? The fact that it was not knowable should not necessarily make us regard it as an irrelevant calculation. We are interested in the coincidence of an exact tie in the voting, but not as interested in the coincidence of a roughly half-and-half split in actual beliefs among Syracuse residents, right? That standard probably be too specific, and would therefore give a probability that is regarded as too high. On the other hand, completely ignoring the polls and all demographic info about Syracuse would be too broad, and if a probability could be calculated, it would be regarded as too low. What’s the right standard, or set of conditions, based on which the probability should be calculated?
This expert professor said it was nearly impossible! Obama probably cheated! Why doesn’t the national media report this!!!!!!!!!!!!!!It’s OJ Simpson all over again!
:P
I hope the absentee ballots keep it a tie so the Post-Standard can say the odds of that was one in a BILLION.
The odds of the 6,001-6,001 tie are no different than the odds of a 6,000-6,002 Obama win or a 6,002-6,000 Clinton win. It is simply a result that has a high probability of generating a headline.