Is Ranked-Choice Voting Too Complicated?


Elections in which voters can rank their choices instead of voting for only one candidate per office are spreading, and have public support. But they raise legal difficulties in light of how people actually treat elections. Ranked-choice voting is much harder than it looks.

The biggest problem may be in how to count ambiguous votes. The voters usually are not personally too ambiguous about their views, but most of the voters are not present in the counting rooms; and many of the ballots by themselves may not clearly reveal voters’ intentions to counters. To deal with this requires one set of rules that will be carried out by several bodies of people with different needs that are almost in conflict and hard, possibly impossible, to resolve.

Triplets Count

Counts have to be managed by election officials; courts; and computer programmers.

Count by Election Authorities

The election authorities have their own people who will prepare the initial official counts. If the winners win by large margins, the courts will likely not get involved.

Because election authorities need many more people on election day and in the few days before and after than they need at almost any other time of year, they add people to help the authorities for those few days, and some for only one day.

The people who work year-round for the authorities are likely to spend much of their time supervising the temporary workers. So, the permanent staff is unlikely to do much hands-on counting themselves.

The people they add are people who are available. They may do the jobs that are not done any other day, so the added people are the ones doing the counting. So, while they may include people currently active in relevant professions, they’re superproportionately likely to be retirees, unemployed individuals, and people in largely-irrelevant occupations. They’re also likely to have diverse educations, ranging from a lot to a little, and degrees of commitment to the work manifesting in carefulness or shortage thereof (e.g., if rushing) and in neutrality or bias on a candidate.

High commitment can also have a deleterious effect. For example, people who were instructed to copy numbers kept separate by some factor (e.g., election district or poll machine) may still combine them because they believe that they’re saving other people from unnecessary work (adding up numbers). The reason they believe that is that they think they know the subsequent process, but they don’t.

People with unfortunate commitment levels can be fired. But then new people have to do the jobs. The jobs have to be done within just a few days already underway or on a certain day already begun, and it may be functionally too late to recruit replacements. That makes the temps almost irreplaceable. Managers’ preference for hiring qualified people means election agencies will hire temps familiar with politics and recommended by political people. So, some temps are likely to be politically connected with people who would be offended at a firing and apt to make themselves heard at higher echelons about it. The political connections are important to other work of the election agencies, like government funding, and that leads the agencies to keep and compliment a low-quality low-productivity temporary employee.

Essentially, the same instructions have to be given to everyone. Therefore, the instructions have to be adequate for the minimally-educated and minimally-committed temps. Since they’re probably paid by the hour or equivalent, the instructions should not require much time to learn. One hopes for the instructions to be brief.

Machines used for voting or counting sometimes break down or are unavailable. When that happens, manual processes will be needed for all steps. For these cases, instructions have to apply to manual processes.

For the officials, instructions on how to evaluate and count the ballots, especially manually, have to be complete and yet relatively simple.

Count by Courts

Courts include judges and juries. Possibly, these cases don’t use juries but have judges do what juries would do, to save time through expedited processing. But that depends on the jurisdiction, so planning may have to allow for a jury.

If a loser loses by only a slight margin, they may invest in litigating the count as the route to winning.

Judges are well-educated and work in a legal environment requiring laws to have some legal precision and clarity. They generally have access to a law library.

Juries, per empirical evidence, are disproportionately retirees and civil service workers who read tabloids rather than broadsheets. They’re told not to weigh the evidence when heard but to wait until the judge has instructed them on the law (although some jurors have said they don’t wait and that’s likely consistent with scientific findings in psychology on how rapidly people respond to important stimuli). Judges’ legal instructions, as received by the juries, are usually unwritten, but only verbal and stated only in advance of deliberation, thus available during deliberations only from memory. Juries usually do not have access to law libraries during instruction or deliberation. Jurors may even be limited in asking questions of judges, and barred entirely from asking parties or attorneys.

For judges, the law must be relatively precise and clear.

For juries, the law must be easy to remember without reference to a written guide.

For the judge and the jury in a case, the law must be the same.

Count by Computer Algorithm

Computers are growing in their roles. While counts can be manual and one backup procedure is likely to be manual, computers are likely to be the first counters.

Computers have to be programmed. The programming must articulate steps humans don’t normally itemize.

For example: You learned to walk when you were around one year old and mainly by observing and imitating while someone kept you from falling too many times. But a robot that looks human has to be taught to recognize the word “walk”, retrieve the definition from somewhere, check if the planned direction is forward, if so check if one foot is significantly ahead of the other, if so lift the rear foot by bending that leg at the knee but not more than a certain number of degrees, swing the elevated shin and foot ahead, visually scan the ground ahead for obstructions and holes, compare the near terrain to a set of specifications for what constitutes sufficient flatness, straighten the knee within a certain engineering tolerance in order to deposit the foot onto the ground, check the robot’s center of gravity relative to the feet, check whether the ground is slippery, etc. The election procedure for a computer needs to be just as detailed, and probably more so.

Multiple computers may each do a separate count, as assurance against a breakdown. The multiple computers may be programmed by different people not communicating with each other and therefore will almost certainly have different programming for the purpose. (NASA reputedly has done that for computers on board long-distance satellites.) The use of different programming by programmers who are not talking with each other is justified by the need to reduce the odds of all the programs containing the same error. Whether one computer is considered primary with others used only if the primary system reveals a critical failure or all counting computers are considered equal with arbitration led by a specialty computer designed to resolve disputes among counting computers, or some other system for handling differences is applied, each program must be more precise than humans usually are.

Accuracy is also needed. If computers are going to read handwriting, or even just detect it, that is a challenging area for programming unless a high rate of false positives is acceptable on the ground that the system will invite humans to confirm the presence or content of the handwriting and then resolve the false positives. But if the rate of false positives is so high that humans are needed for almost all ballots with apparent writing and therefore for almost all ballots, maybe computers shouldn’t bother trying to interpret handwriting and someone should hire more people instead.

General Parameters For All Methods of Counting

Some requirements apply to all three methods.

Machinery (including computers) to cast ballots is becoming more pervasive, but paper ballots are likely to continue in use when machinery is failing or where special needs, such as a particular kind of disability, are not adequately accommodated by machinery. We have to design for every technology likely to have some use, even the most primitive.

Perfection is not required, but credibility is. The larger the system the less likely it will be perfect, but, for the public to consider the system reliable, permissible error or tolerance has to be low. Especially when a victory is slim, accuracy matters greatly. To get high accuracy, it has to be designed into the system before voting begins. That includes being ready for early voting.

Whether paper trails are provided beyond what’s needed for the prototype test and whether receipts are issued to voters are irrelevant.

Law

Laws, including instructions (a form of law), have to be available. Because human memories are imperfect, laws including instructions have to be written and published. The major ones usually are, but all of them have to be (if colorably relevant). It won’t be enough for some Bob to testify that surely he must have said such-and-such and that he doesn’t know why no one remembers it.

A computer algorithm would determine which votes to count and voters’ intentions and those decisions would be enforced unless overturned. Thus, the algorithm would be law. While programmers would be taught the law, they probably are not lawyers, law is lengthy and complex, and the programmers would likely make a legal error in the algorithm. While, in general, a law can be drafted with some reasonable leeway relative to higher law, that authority is not unlimited. Programmers’ work therefore would have to be studied by a lawyer for lawfulness. This means that a lawyer must know how to read a program, but not many do. The program is likely to be lengthy and complex. This alone is a recipe for a mess.

While laws often have statements of purpose or intent or histories useful to courts, and programs should have similar documentation, programs often don’t, or have not enough of it. Programmers can be ordered to testify but many would be living out of state and may not have to testify and may have forgotten many of the considerations, especially for finer points, that were not documented.

Computers

What to Publish

For computer programs, specifications (based on law but written to guide programmers who usually are not lawyers or legal researchers) should be written and complete. It’s possible to create a program with the specs being all in someone’s head, but, for a big or critical program, specs being mental is usually a bad idea. Specs are a far cry from an elevator pitch or a brochure of benefits and features. Specifications get very detailed. They reveal the logic expected of the program, with particulars. The specs should be available, or at least large parts of them should be. If they’re not, anyone analyzing the software will have to guess them and, if they’re right, that will make the specs public anyway. We’ll need to know if they’re right, so one requirement is for the specs to be public in the first place.

Source code (what programmers create and use to verify exactly what’s supposed to happen inside a computer, and coming immediately before object code, which is what a computer executes) has to be available. Available or not, it exists. There’s no point to denying to information technology experts that source code ever existed.

Nuts and Bolts and Logic

Artificial intelligence or any other programming cannot be a mysterious black box. Knowing the inputs and the outputs is not enough. Exactly how the inputs produced the outputs must be transparent to programmers, election agency officials, jurors, and jurists, and election agency officials, jurors, and jurists probably never took a computer science class in their lives. For most of them, they’ll use a computer but having to know the difference between a bit and a byte verges on the inhumane.

Whether a computer program can be examined by the whole world, even Democrats and Republicans, will be argued. Security will be cited as a reason for withholding program details from public scrutiny. However, malicious actors who peek and poke into the system are evidently unrestrained by the wish for security, but the secrecy will make it harder for well-intentioned ethical members of the public, including journalists, computer scientists, and friendly hackers, to disclose major problems. Security by obscurity is a flawed approach. While access credentials (mainly passwords) need to be secret, that’s to prevent unauthorized reading of data (e.g., votes cast) or changes to the installed program or the data without permission. The programming itself needs to be vetted line by line before installation, including by people with no vested interest in agreeing with election officials. This can be permitted in a way that’s consistent with election security.

The technical explanation is that this is done by making a copy of the source code public. The source code is the last stage of program development that programmers can still read with eyeballs before conversion into what the machine will execute, which is object code. Conversion of source code into object code is reliably and securely done just before installing the program. Thus, even if someone randomly mangles a copy of the source code and no one knows exactly how it was mutilated, the person installing the program can use a remaining clean copy of the source code to do a safe installation.

Source code tends to be either closed or open. Proprietary products partly protected as trade secrets tend to have closed source code. Closed code can be forced open through subpoenas and other judicial process, but that’s after beginning a challenge in court, generally after an election, thus generally too late for pre-election preparation. (Shared source code from Microsoft is a variant of closed source code.) Open source code under an appropriate license of its intellectual property can be examined at any time by anyone. Thus, to check system integrity sufficiently, all of the source code must be open. Closed source code must be either highly trusted or absent.

Trusted closed source may be used for a well-known operating system, such as Microsoft Windows. Although reliant on closed source, Windows is highly trusted for this purpose despite its defects because it would be accepted almost everywhere as neutral with respect to elections. An application (an app or computer program for end users) for voting would not have that trust and therefore if it includes closed source the app would have to be absent, an open-source app being needed. If no good open-source app is available, it would have to be created, tested, and evaluated in advance.

Usability

Not only does the programming have to be logically and technologically correct, it has to be user-friendly, so that users are less likely to misunderstand and make errors. Most users are not nerds. If the software is not relatively easy to use, they will do something easy anyway, even if it’s wrong.

A well-designed user experience (UX) will not compromise the underlying logic, the integrity of the source code, or the quality of the results for the election. It will improve the a user experiences the electoral system, and that’s valuable for persuading voters to come back for future elections.

Defects

Many computer programs have not only features, they have undocumented and unsupported features. Perhaps a feature was created, then determined to be undesirable, and yet not entirely removed, although it was omitted from the documentation and from the user interface. This can yield confusion about whether the feature is present, with attorneys and programmers disagreeing. The program being highly complex does not make this easier to resolve.

Bugs, errors in a computer program, are almost certainly present. It is good practice for bugs to be publicly reported, triaged, and repaired as necessary. Some reports of bugs are themselves erroneous. Bugs implicating security should be reported secretly, although good practice also provides for security-implicating bugs to be publicly reported if, after a reasonable time, the secret report was not properly acted upon. If an election authority management decision is made not to permit any public reporting of bugs, or any reporting of bugs, the bugs may turn up being lawfully reported elsewhere, beyond reach of the prohibition, and the authority or programmer may or may not be aware of the reports in time. If the prohibition is effective, the result may be more bugs remaining in the program, never remedied but maybe secretly exploited by attackers.

Who Votes and When

Who may vote should be maximized. Who should be permitted to vote has been debated for centuries and still is, but that debate should be at the point of eligibility to vote and not at the level of how tabulation systems work. Therefore, the system for tabulating should scale up to any population size. The world population is around seven and a half billion people and growing. However, a system that is acceptable for a few million voters can probably be scaled up to cope with even the world population for not very much more development, when the logic is the same.

Whether voter registration is before election day or in the minute before a ballot is cast, or is altogether dispensed with, is irrelevant. By the time a ballot is offered for casting, whether the putative voter is allowed to cast should already have been decided. In the end, only two decisions should be allowed: affirmative, and therefore a ballot is offered, or negative, and therefore it is not offered. It is conceivable to offer a ballot tentatively, pending a later decision on whether to accept it for counting, but that decision presumably could precede the counting of anyone’s vote. This simplifies the work of tallying.

A society limits voting to being by adults, but that’s because of the responsibility attached to the consequences of votes, not generally because of minimal education standards to be expected of voters. Some voters may have very low levels of education. (One nation once dealt with low literacy among voters by adding candidates’ facial photographs to ballots.) Counters have to understand the intentions of voters even where low education makes filling out a ballot difficult regardless of candidates.

High turnout, meaning voting by more of the eligible voters, increases public confidence in the offices for which turnout is high. So, democracies generally want a system that does not depress turnout, either immediately or in subsequent elections. They don’t want voters to skip future elections. And the tally system must be of a scale that can take a higher turnout.

Voter Drop-Off

Drop-off, whether high or low, should be irrelevant to the counting process. Drop-off is when voters vote for fewer offices than the law allows them to do, and is common among voters who respond to appeals in certain races (e.g., Presidential) but don’t care about other races and are not party-line voters. The system should cope with high drop-off producing nonvotes on ballots cast and should not balk at a low drop-off rate.

Outcome Consistency

The various methods of counting votes, by officialdom manually, by courts, and by computer, have to produce almost the same results within very tight tolerances. Otherwise, all three methods will be publicly doubted and the overall system of counting votes (in this case, ranked-choice voting (RCV)) and the people responsible for the methods will be discredited and replaced in the interest of preserving the trustworthiness of elections. The people will be replaced first, a couple of times; if that doesn’t fix the failures, the system itself will be replaced. That’s already happened. It can happen again.

Educating Voters and the Effects

We can teach voters the procedure for voting. However, many will reject being instructed in how to do something they’ve done every year since they voted for a bathroom monitor in first grade or so. The instructional speech will be ignored by them, or considered rude. They’ll just go to the poll and vote like they always do. Even instructions posted at the poll site will likely be ignored by most voters, who will go straight to the ballot and start marking it (or pulling levers or whatever they see they’re supposed to do). If they think it’s too complicated, they’ll either walk away or, if they do something like voting that time, they’ll leave remembering that it wasted their time, they won’t come back the next time, and total voter participation will drop. So, the system where voters use it must operate properly even for voters who don’t read the instructions.

Some properly instructed voters will diverge from the instructions anyway. The society wants every vote to count. Therefore, the counting system has to try to accommodate voters’ divergences. It has to interpret voters’ intentions even when unclear but at least leaning in one direction. For voters whose intentions are completely unknown, those votes will have to go uncounted, but the larger objective will be to minimize how many are uncounted. So, interpretation will be the order of the day. But the interpretations have to be consistent across all three methods of counting and regardless of which human beings fill a given role.

Building a Prototype on Easy-Election Assumptions

Before anyone creates a full-scale system responsible for the real thing, it’s safer to create a prototype for testing. One aircraft manufacturer, before building a plane to fly a hundred or more passengers a thousand miles quickly through invisible air miles above the ground, built a wooden mockup to make sure that plumbing pipes didn’t intersect with cables pulling wing flaps. Before Boeing built the 747, someone named Wright flew a plane just a few hundred yards with one pilot and no passengers.

We need to craft a prototype suitable for at least one simple yet realistic election.

For our purposes, assume the following: Voters can rank any two candidates in a race among three. A voter is to assign the number “1” to any candidate (unless submitting a blank ballot) and may assign the number “2” to another. Each number must be in the designated space next to a candidate’s name. Any other number would be uncounted but would not invalidate a ballot. No write-in votes would be allowed. No other marks would be allowed. The presence of a write-in vote or another mark would invalidate a ballot. Ballots would be printed on white paper in black ink. Voting would be on the ballots with blue pens. No more than one ballot would be issued to a voter, even if the voter has made an error and can destroy the first ballot before it can be counted.

What Voters Will Really Do

Perhaps every voter will follow instructions to the letter. However, human beings generally are not utterly obedient.

What will actually happen:

Some voters will write with a blue pen that skips. They’ll think you can still read it.

Some voters will write with black ink, because it’s in their pockets and they always use it. Black ink from a pen may be hard to distinguish from black ink used for printing forms.

Some voters will write in pencil, because it’s in their pockets. Pencil lead may smudge or possibly the writing will be changed by a finger press.

Some voters will erase something, but ambivalently. This could confuse everyone.

Some voters will write “1” twice. They may prefer two candidates out of the three but not care which of the two is the winner. This is likelier when candidates run as a slate against their opponents, even if the slate is not identified on the ballot but only in campaign advertising seen before election day.

Some voters will write a “2” without a “1”. Whether to elevate the “2” to a “1”, invalidate the ballot, or neither has to be decided before the ballot is seen.

Some voters will apply check marks, because that’s what they usually do and because they may not want to rank anyone.

Some voters will apply an “x” next to a name. Some non-election forms, such as some tax forms, require an “x” where someone would otherwise use a check mark. Some people habitually use an “x” for other non-election preferences. They’ll do it here, too, without realizing, or when forgetting, that it’s not allowed.

Some voters will apply a check mark next to one name and an “x” next to another. They may do this because they use the marks interchangeably, as having the same meaning, or because they want one candidate and hate the other, but are neutral about anyone they left unmarked. This means the “x” can have either of two opposite meanings, but the one voter probably doesn’t realize this because the one voter is probably not thinking about how other voters and different officials think about an “x”. Meanwhile, the public wants us to count every vote.

Some voters, instead of writing “1” next to a name, will write “The Best!” (or something to like effect) in the same place. A decision will be needed on whether that’s equivalent or a violation. It can’t be both.

Some voters, presumably fluently bilingual, instead of writing “1” next to a name, will write “una”, “une”, “een”, “один”, or “एक” (and we’ll have to know that all of those mean ‘one’) or something in their native language that means “The Best!” (or something to like effect) in the same place. Probably, officials, juries, et al. can’t be counted on, or allowed, to translate between natural languages. The reason for not allowing it is that the same rule, in this case about whether to translate, has to be applied by everyone responsible for counting and not everyone will know the same non-English natural language, or even recognize it, or which one it is.

Some voters will write “1 The Best!” (or equivalent). The intent is clear but it includes unexpected writing. A rule is needed.

Some voters will write “1” next to one name and “2 The Best!” (or equivalent) next to another. The intent loses its clarity because the ballot includes unexpected writing conflicting with the numbers. A decision is needed.

Some voters will rank candidates with “A” and “B”, or lower-case equivalents, because they went right to the blank space and the officials drafting the ballot form didn’t repeat a “number only” instruction right next to every space; and, if they had, some voters would be annoyed or overwhelmed by so many redundant instructions that they’ll complain and/or not come back for the next election, and we don’t want them staying home in any year.

Some voters will circle or underline the names they like, with or without numbering.

Some voters will circle or underline one name and very emphatically circle or underline another name. Assume they don’t number. Maybe the emphasis is equivalent to a “1” and the less-emphatic mark is equivalent to a “2”. They vote; someone else will have to decide how they voted.

If anyone was campaigning near a poll site by handing out candidate stickers, stickers will appear on ballots. Some machines may respond to any extra thickness by rejecting the ballot, or by getting jammed, potentially tearing the ballot and making it unreadable, and also delaying subsequent voters, who may get discouraged and either leave before voting or not return the next year.

Some voters will place whatever mark they use near a name but not in the designated space, or as one mark overlapping designated spaces next to two names. This is especially likely with long ballots, since they are printed with either large paper or small fonts, and small fonts are cheaper to print.

Some voters will write something, cross it out, and maybe write something else. The crossed-out text is a mark that would invalidate a ballot, which, to voters, makes no sense.

Some voters will write something, perhaps kind of cross it out, and maybe write something else. The voter will have left, so no one can ask the voter whether that’s a cross-out. It will have to be judged in the voter’s absence.

Some voters will write editorial comments. They won’t understand why that’s a problem, although other voters likely will.

Some voters will write something that effectively calls for judgment about their intentions. If a voter writes next to a name without a number, “is he Catholic?” or “is she for tax reform?”, that could be to either favor or oppose the candidate, and which way to interpret it could depend on the neighborhood.

Some voters will sign their ballots. It will seem like an obvious thing to do with an official document. People won’t expect that doing so invalidates their ballots.

Some voters won’t be fluent enough in English, or written English. They may misunderstand the instructions, or not understand them at all. They’ll vote and other people will try to understand their intentions.

I’m doubtless missing a few more things voters are wont to do. The more voters there are, the more different things they’ll do. They’d usually just be doing what makes sense to them. We’ll need to decide on a consistent action for almost anything they do.

New York City History

New York City had two experiences with this kind of voting. I don’t know of any others in the city.

School Boards

The more recent experience was with Community School Board elections, running for over three decades starting in the late s. They had high compliance with voting instructions, reportedly 95%, but had low turnout. I think the voters tended to be committed to voting, most likely because of party affiliations, so they were more likely to accept being instructed in how to vote and then to cast compliant ballots. I learned that when votes were being counted it mattered which candidates were present or represented in person at the count, because that determined which ballots got counted first, and therefore which first-choice votes were left over from one victory and could be counted for second choices for the next victory, and so on through all the ballots.

Most of the school boards were not outstandingly respected, but were viewed as simply another piece of the large education bureaucracy. Neighborhood newspapers might cover the local races once or twice and citywide newspapers might cover a few highlights across the city, but generally there was not much publicity. Low news coverage indicates that the public was less interested in the boards’ role in children’s education. The public has always wanted good schools and is glad for teachers, but the boards themselves were not highly regarded.

I learned of a common practice suggesting that many candidates had less support than required to get on a ballot. They were supposed to submit petitions signed by a threshold number of signatories. Instead, a candidate would submit one petition with a few signatures and other petitions stapled to the first. The other petitions were blank, and the candidate would thus have submitted too few signatures to qualify to be on the ballot. Then, the Board of Elections handling this would not look at any pages after the first. The undersupported candidacy could be challenged by an opponent, but the law said that the challenge had to be submitted within three days after the submitting of the petition, which was not necessarily three days after the submission deadline. So, the candidate submitting the petition would do so at least three days before the deadline and would not tell anyone publicly about having submitted it. The three days would lapse, the candidate would then announce their candidacy, it would be too late to challenge it, and they’d be on the ballot; and often they’d get elected.

When the elections were abolished during one Mayoral administration, there was not much grief. Another Mayor was elected, he’s from a different political party than the prior Mayor’s, there’s no major support to bring the school board elections back, and they haven’t come back. There’s a little-noticed substitute now.

s

The other type of election was in ca. the

Modernity

Perhaps other ranked-voting systems now in effect, such as one in Maine, have lessons for us. But those only recently enacted need to be seasoned and results measured over some years before conclusions are reliable.

Wishing is Not Enough

What’s desirable is not always easy. Ranking gives voters more power. Candidates are more likely to admit to merit in their opponents; I don’t know if that means a greater willingness to compromise while in office in order to reach more agreements (and an argument is that often fewer agreements would leave more power elsewhere); but perhaps it would contribute to an atmosphere in which more agreements are politically palatable. I like ranking.

But I’ve written to some advocates about the problems. None have replied on point. My guess is that they haven’t given them much thought, probably leaving them up to local authorities. Local authorities have a role but should not be left alone, with no one else looking at the products they produce. If the public doesn’t look ahead of time at the plans, courts and the public will look at the rubble afterwards.

The people who do the most work in the government on electoral systems tend to be people who understand the subject best and are the most trusted, which, unfortunately, tends to be politicians and people trusted by politicians. They tend to lean strongly in favor of the candidates of the locally major political parties. They tend to know neutral things to say when required but also tend to be biased in favor of the candidates they consider as better, and this reflects in their judgment on marginal cases. There is evidence that they tend not to like systems that are too precise and secure. The saboteurs among them would want us to run an election with weak planning. If we do that, electoral corruption will rise and ranking will smell like a rat.

With any new system, candidates must know the new system in order to campaign in a way that relies on voters’ wielding their new power to rank their choices. Planning and testing must be finished before candidates arrive at their first deadline. Weaknesses in the total system must be sought, discovered, and repaired before they can cause harm.