Here's a test for determining whether an electoral system operates successfully: unless the initial count of ballots shows the two leading candidates separated by less than one-hundredth of one percent (< 0.01%) of total votes cast, is the system able to identify the correct winner and resolve any disputes in a way the losing candidate accepts as fair — and to do so before the date on which the winner is supposed to take office?
There are three noteworthy features of this test.
Tolerable margin of error. First, and perhaps most obvious, this test does not apply when the initial count shows a difference of less than one-hundredth of one percent between the top two candidates. When a race is that close, the test essentially lets the system off the hook. In this situation, the system is not expected to be able to identify an undisputed winner by the time the winner is supposed to take office. Therefore, it does not condemn the system as a failure if it unable to do so when the margin of victory is so low.
In a statewide race with 4 million votes (as Ohio's Senate and gubernatorial elections may be this year), this threshold is 400 votes. In a congressional race with 200,000 votes (as Ohio's competitive races this year will have), the threshold drops to 20 votes. The 537-vote difference between Bush and Gore out of Florida's almost 6 million votes cast (5,962,657 according to certified totals) would have been just below this threshold. Likewise, the 129 vote margin in the 2004 Washington gubernatorial election, out of 2.8 million ballots cast, is below this threshold. As other observers have said about these two recent controversial elections, we could not have expected our vote-counting systems to be indisputably accurate when the results are this close.
Other kinds of systems are held to stricter standards of accuracy. In industry, many major businesses, including Motorola and General Electric, have adopted a standard known as "Six Sigma," which requires tolerates error rates of only 3.4 per million, which is less than one-thousandth of one percent (0.001%). In an election with 4 million votes, Six Sigma would require an accurate count unless the margin were less than 14 votes, and in a race with only 200,000 votes cast, there would be no room for error at all. Six Sigma would not have exempted the 2000 presidential election in Florida from the requirement of an accurate count. Nor would it have exempted the 2004 gubernatorial election in Washington.
Six Sigma, however, seems too stringent a standard for electoral systems. Indeed, in contemplating this column, I wondered whether the aforementioned threshold of one-hundredth of one percent is too demanding and thought perhaps the easier test of one-tenth of one percent (0.1%) would be more appropriate. But that would mean that an electoral system would be off the hook — not expected to identify the undisputed winner before the time to take office — if fewer than 4000 votes separate the two leading candidates in a statewide race with 4 million ballots cast, or if there are fewer than 200 votes separating congressional candidates in an election with 200,000 votes cast.
I'm not at all sure that our electoral systems are currently capable of meeting even this easier standard. For example, it is easy to imagine the initial count in Ohio's Senate election this year showing more than 4000 votes separating incumbent Mike DeWine and challenger Sherrod Brown, and yet the outcome of the race could be in dispute until well after the time for the new Senator to take his seat. Likewise, it is all too conceivable that a competitive congressional election in Ohio could produce an initial margin of victory greater than 200 votes, with that result remaining disputed until long after the date on which the new Representative is to be sworn in to Congress.
Ohio law calls for a mandatory recount in a statewide race when the margin of victory in the initial count is less than one-quarter of one percent, and in a district race when the margin of victory is less than one-half a percent. Moreover, Ohio law now requires that any recounts of votes cast on a touchscreen (DRE) machine use the "voter verified paper audit trail" as the official ballot to be recounted. Based on an effort to recount a sample of precincts in Cuyahoga County after the May primary, it could take at least two months to recount these paper trails in either the Senate race or a congressional race. Thus, even if 10,000 votes separate the two Senate candidates, or 500 votes separate incumbent Deborah Pryce and challenger Mary Jo Kilroy (to take one competitive congressional race as an example), the mandatory recounts of these races might well not be complete by the time in January when the new Congress is supposed to begin business.
Still, I'm not inclined to opt for the easier standard for evaluating whether the electoral system is successful. Even if only 1000 votes separate DeWine and Brown in the initial count, my judgment is that a well-designed and well-implemented electoral system should be able to identify the correct winner and resolve all disputes to the satisfaction of the losing candidate by January, when the new Senate is scheduled to convene. Likewise, even if the initial count shows only 150 votes separating Pryce and Kilroy, the voters as well as the candidates should have confidence in the electoral system's ability to identify the true winner of the race and to settle fairly any disputes arising from the election before the time one of these candidate is to take her seat in Congress.
Ultimately, however, whether to adopt an easier or stricter standard for evaluating the ability of the electoral system is a policy judgment. Despite the evident need since 2000 for public discussion of this policy issue, it has received regrettably little public attention. I offer my test, which incorporates the stricter (one-hundredth of one percent) threshold in the hope that doing so helps spark a policy debate on this issue.
Avoidance of extensive unfixable errors. Second, implicit in the way I stated my test at the outset is the requirement that, except for elections so close as to be below my stringent threshold, the electoral system must avoid uncorrectable errors that affect the accuracy of the result. A clear example of an uncorrectable error is the inability of voters who go to the polls to cast even a provisional or emergency ballot in the event that the voting machines break down (the kind of problem that occurred in the Maryland primary this year). If 1000 votes separate the two candidates, yet 20,000 voters who go to the polls are unable to cast ballots, the legitimacy of the result is irretrievably in doubt.
Suppose, for example, that the failure to supply emergency ballots occurs in urban precincts, and the 1000-vote margin favors the candidate with significantly weaker support in urban areas. It is quite likely that the opposing candidate could have made up this 1000-vote deficit if the 20,000 voters had been able to cast ballots as they intended. Yet there is nothing that can be done to fix this error, short of holding a new election, which by definition is a concession that the electoral system has failed in this instance.
Other examples of uncorrectable errors include the loss or destruction of paper ballots after they have been cast but before they are counted (as occurred in Cuyahoga County during the primary this year), or the loss or destruction of computerized vote totals when no back-up record exists. Even when lost ballots or vote totals are later found, the chain of custody has been broken in a way that cannot be undone. Therefore, if there are credible allegations that the ballots or vote totals were tampered with during the time they were lost, it may be impossible to refute these allegations to the satisfaction of the losing candidate. In this situation, this breach in the chain of custody would cause a failure of the electoral system's ability to identify an undisputed winner.
Long lines at polling places, which cause voters to leave before they are able to cast ballots, are another form of uncorrectable error. So, too, are defects in the design of the ballot itself — like Florida's infamous "butterfly ballot" in 2000 — which cause voters to register a vote for a candidate other than the one they intended. Indeed, in my judgment, the most serious problems that occurred in the 2000 presidential election in Florida were not the difficulties in recounting punch-card ballots with dimpled or hanging chads, but the flaws in the system that prevented voters from casting ballots for their preferred candidates in the first place. The butterfly ballot was one such problem; even worse was the lack of provisional ballots to enable those voters who had been erroneously purged from registration lists to record their preferences.
Thus, my test for evaluating electoral systems places a premium on their ability to avoid these sorts of uncorrectable errors. They don't need to eliminate these unfixable errors completely. Absolute perfection every time cannot reasonably be expected, and the notion of a threshold margin of error (as discussed above) recognizes that an electoral system cannot be condemned as a failure if a few isolated voters are improperly denied provisional ballots or encounter long lines that force them to leave before voting. But when these sorts of errors are not isolated, but instead become extensive enough that they affect hundreds or thousands of voters, then the electoral system should be labeled a failure in the event that these errors prevent the identification of a undisputed winner in a close race.
Timely correction of fixable errors. Third, my test for the success of an electoral system does not condemn correctable errors as long as they are corrected in a way deemed fair by the losing candidate by the time the winner is to take office. In this respect, my test is out of sync with contemporary American psychology regarding the operation of electoral systems. Any breakdown in voting machinery, or any mistake regarding the purging of voter registration lists, is viewed as a calamity by voting rights advocates and other watchdog groups, even if the problem is easily solved in such a way that all eligible voters are able to cast ballots that get properly counted before inauguration day. Likewise, the public wants to know definitively who has won immediately after the polls close, and thus views the system as having failed even if it accurately determines the winner two weeks later, when all the eligible provisional ballots are properly counted.
One of my main contentions is that there needs to be a change in public psychology regarding the success or failure of electoral systems. It is important that the public become able to distinguish between correctable and uncorrectable errors — and to regard the timely correction of correctable errors as an indication that the electoral system's safety mechanisms are functioning properly, just as when a hospital's back-up generator keeps the electricity working in the event of a municipal power outage. It would be better not to have to rely on provisional ballots because voting machines fail or registration lists are faulty, just as it would be preferable not to have the power outage in the first place. But if the provisional ballots work to make sure that every eligible voter casts a ballot that gets counted, then no harm to voting rights has occurred, in the same way that the back-up generator prevents the municipal power outage from harming the hospital patients.
Similarly, the public should not be troubled if it takes several weeks to count and recount all the ballots in a close election. The processing of provisional ballots inevitably takes considerable time, as each ballot needs to be checked against available registration records, including (when necessary) original registration cards. The expectation of instantaneous results on Election Night is an unfortunate product of network television and exit polling. In the future, however, the public should recognize that Certification Day, not Election Night, is the time when the winner is officially announced, after all ballots have been counted and (when necessary) recounted.
But even if the public comes to understand that the accurate outcome of a close election cannot be expected for several weeks after the polls close, it is also essential that the electoral system be able to complete the counting process accurately before the date on which the winner is supposed to take office. If it really takes more than two months to conduct an accurate count, then some thought needs to be given to extending the time between election day and inauguration day. I am confident, however, that a well-designed electoral system can complete a fair and accurate count, worthy of acceptance by the losing candidate, in less than two months. If nothing else, using optical scan ballots (even if they are marked for voters by touchscreen devices) would expedite any recounts, avoiding the time-consuming process of manually counting paper trails.
Two months should also be long enough for a fair and acceptable resolution of any disputes arising from the voting process. Two months is enough time to evaluate the eligibility of provisional ballots in a way that is fair to the provisional voter and to the candidates. It should also be enough time to fix any correctable errors that occur in the counting process, like examining a back-up memory tape in the event that a machine malfunction erases the primary electronic record of votes cast on that machine.
Most states have a post-certification procedure for contesting the results of an election. Nonetheless, it is necessary that this procedure be completed by the time the winning candidate is supposed to take office. It does no one any good if a post-certification procedure is theoretically available after inauguration day, yet as a practical matter no court would order the removal of an already inaugurated officeholder. It would be better for the electoral system to require certification one month after the polls close and then require the completion of post-certification contests within the next month.
To be sure, whatever procedures the electoral system employs for resolving disputes that arise over election results, those procedures must be viewed as fair by the candidates who do not prevail after having invoked those procedures. The fairness of an electoral system must be measured, as many others have observed, by losing candidates and their supporters, not those who win. Consequently, I have built into my test for determining whether an electoral system is successful the requirement that it resolve any post-election disputes in a way that is regarded as fair and acceptable by defeated candidates and the eligible voters who cast ballots for them.
Losing candidates and their supporters, however, must be reasonable in their assessment of the system's fairness and acceptability. They cannot demand a dispute resolution procedure so elaborate that it extends beyond the date for taking office. Moreover, the time for assessing the fairness of post-election dispute resolution procedures is before the voting occurs, when the candidates and political parties do not know who will win. If beforehand the candidates and parties agree that the existing post-election dispute resolution procedures are fair, and if after the election these procedures operate as specified, then the losing candidate is not in a position to claim that the procedures were inadequate.
It is necessary, therefore, for the electoral system to specify in advance the procedures that it will employ in the event of a post-election dispute. These procedures should make clear that all post-election disputes will be resolved before the date for taking office and will be structured in a way to give the disputants a fair chance to present evidence that challenges the accuracy of the system's initial count. If in a close election these procedures are invoked, and if these procedures identify and correct fixable errors in the initial count, thereby producing a new accurate count before inauguration day, the public should regard the operation of its electoral system in this instance as a success, not a failure.
Building an Electoral System to Meet this Test. In future commentary, I hope to consider the chances that our current electoral systems are capable of meeting the test I have described. It will be interesting also to see how well our systems fare this election year, when measured by this test. Already, however, it is possible to identify improvements that could give our electoral systems increased chances for success in 2008. Thus, whether or not our electoral systems are fortunate to escape failure this year, legislatures should continue to make changes that would reduce the risk of electoral failure.