While recounts have always existed in some form or another, they seem to have become much more common (or at least much more publicized) ever since the 2000 presidential election in which Al Gore pressed for a recount in Florida until the U.S. Supreme Court effectively put a stop to the effort. Al Gore's demand to "count every vote" seemed reasonable enough, but was it? Ever since then it seems that recounts have become the accepted second step of any election. Even the July 2006 elections in Mexico lead to the looser echoing Al Gore's strategy.

And in 2004, a recount in the Washington governor's race was initiated when just 261 votes separated the two candidates out of a total of 2.8 million cast. The initial count gave the race to one candidate by 261 votes , an automatic recount then ensued which confirmed the candidate's lead, only to have a second manual recount give the race to his opponent by a margin of an even slimmer 133 votes .

The margin of victory in the Washington race was 0.0045%--this is like two sharpshooters attempting to hit the same target from 500 feet away and one missing by a quarter of an inch and the other missing by about half an inch. Does this prove that the first shooter is really the better shooter? Of course not--it simply proves that they both demonstrated very similar performance and, in another run, it could have gone the other way. In fact, in Washington they "fired" three times and candidate "A" had the better shot the first two times, but on the third shot candidate "B" had a better showing and was declared the winner.

Without getting too deep into statistics, this is a question of the "margin of error" of the system. All systems have errors regardless of what those systems do. If you have two cars driving side by side down the road at the exact same speed, it's possible that the speedometer of one car might read exactly 70 miles per hour while the speedometer of the other car might read 72. This is why, in most states, the police will not pull you over for going up to 5 miles per hour over the speed limit. This is a recognition of the margin of error of the speedometer in any given vehicle and that while the policeman's radar clocked the driver at 74mph, it is entirely possible the driver's speedometer read 70. It is not reasonable to make a legal decision when the difference involved is within the recognized margin of error the system.

Unfortunately, we do not apply the same logic to elections. The closer our elections are--and thus the more likely it is that we're within the margin of error of the system--the more apt we are to insist on a recount. In other words, we insist on scrutinizing the results when the system is least able to give us the accuracy we need.

For example, if candidate "A" wins 60% of the vote and candidate "B" wins 40% of the vote, it's clear that candidate "A" won handily. A recount would obviously confirm that and everyone knows it, so the loser graciously concedes and we all move on.

However, when candidate "A" wins 50.0001% of the vote and candidate "B" wins 49.9999% of the vote, all the sudden candidate "B" is going to push for a recount. If he's honest, he thinks the vote count was so close that there could have easily been enough errors in the original count that it could swing his way in a recount. If he's not honest, he thinks that enough errors could be made in the recount itself to swing things his way. In either case, papers are filed with the court, a bitter dispute ensues, and roughly half the electorate is eventually unhappy with the result which is decided after weeks or months in a court, often based on the political makeup of that court.

The reality is, regardless of whether an election is close or not, when you're counting millions of ballots you are going to get a different result every single time. It's just not possible to count millions of ballots and not make a single mistake. As this paper at Cal-Tech indicates , any election that is won by a margin of less than 0.5% has a non-trivial chance of being reversed if the ballots are simply recounted.

Statistically speaking, if the election is won by less than the margin of error, the election was a tie. Each recount may very well reverse the previous count, and subsequent recounts may reverse the reversal. When dealing with 2.8 million votes, such as in the Washington election mentioned earlier, winning or losing by 260 votes is statistically irrelevant. Given the exact same set of ballots, either candidate could have conceivably won and it's entirely probable that each recount will result in a different winner. A recount is not inherently more accurate than the initial count, and a second recount is not inherently more accurate than the first recount. In fact, the more the ballots are handled in recounting, the more probable it is that some ballots will be damaged, lost, or fraudulently manipulated. If anything, recounts are prone to be less accurate than the original count yet, ironically, the recounts are what are considered to be the final word rather than the initial count.

Given these facts, it would make sense to establish laws that say that recounts should only be ordered when there is serious evidence of fraud that gives society just reason to be skeptical of the results. A close election, in and of itself, is not evidence of fraud and should not automatically trigger a recount which will not improve the statistical accuracy of the election but may very well undermine the perceived legitimacy of the eventual winner. Further, it may make sense to establish laws that define an election as a tie if the margin of victory is less than, say, 0.1%. In those cases, the election could be decided by a coin toss or a drawing of straws or some other random game of chance. This would be no less accurate than counting and recounting until one side or the other achieves the results it wants and would eliminate the long court battles and divisiveness that follows.

The purpose of voting is to divine the will of the people. As good as it may seem to say, "Count every vote," the reality is that if every single vote has to be counted to determine the winner, the election is so close that either of the candidates could be the winner. The will of the people is clearly divided, almost precisely, between the two candidates. Either candidate could win and either candidate could lose, and it's folly to claim that one candidate is somehow better than the other, and it's equally folly to suggest that a recount that gives the other candidate the win by an equally small margin is somehow a more accurate representation of the will of the people than the original count.

In 2000, Bush won the election under the original count and under some recount scenarios while Gore won the election under other recount criteria. This only proves the fact that every recount is going to provide a different result. In 2000, Bush won, but Gore could just as easily have won and that would have been just as valid (or just as invalid, depending on your point of view). It's obviously frustrating to lose by such a small margin, but it's statistically dishonest to suggest that a recount in those scenarios is somehow going to magically result in the will of the people being more accurately determined.

These days, in the United States, the recount mechanism is being abused by people that do not understand statistical realities. We have laws on the books that mandate a recount precisely in those cases when a recount is least useful and even if there is no indication of fraud. And we have candidates winning elections with a margin of victory that is smaller than the margin of error of the system and which should be considered a tie and decided by lot.

In a democracy, counting the votes is important, but so is unity and acceptance. Fighting for weeks and months over the exact number of votes each candidate received, when it's impossible to know with that level of precision, is useless if that process leads to suspicion, division, lack of unity, and extreme partisanship for the entire term of the elected official. In such close races, my respect goes to the candidate who voluntarily recognizes the statistical tie and allows society to move on without prolonging a statistically pointless process.

Rather than engaging in a recount every time an election is close, it would be far more healthy for our democracy if we would recognize a statistical tie when we see it and simply draw lots to determine the winner. Save the recounts for when there is compelling evidence of fraud and the recount is ordered simply to make sure that the fraud did not materially change the outcome of the election.