Ranking vs. Rating — Responding to Seth Alan Woolley

My purpose in posting this is to add to an interaction about comparing voting systems. The interaction started on Facebook. Only people who have signed up for Facebook and then joined the specific discussion forum thereon can see it, but for the record, here is the link to the start of the debate.

Two aspects of argumentation are (1) trying to carry forward on ones own points and (2) responding to false statements and/or dishonest rhetoric from the other party. Up to this point, I have optimistically tried, much of the time, to concentrate on (1) and I have let some of (2) just sort of slide for the time being. I feel it is time to catch up on (2), in order to do my best against the possibility of some readers being snowed.

Seth Allen Woolley says:

No. Ranking is better than rating. Rating is vulnerable to strategic manipulation and the intermediate values are not well-defined and thus incomparable. It’s equal to approval voting in most features. The corporations funding its advocacy bore me.

What does “vulnerable to strategic manipulation” mean? Perhaps the concept carries no meaning, and the insertion of the words amounts to a word salad intended to distract from the portions of our dialog that do have meaning. Or, perhaps the concept of “vulnerable to strategic manipulation” has something to do with an accusation that the voting system under study accords some parties more power than it does others, to manipulate the result. I suggest that the Greens and other “third” parties and independents need, in order ever to gain in popularity and power, to concern themselves with the question of which voting systems accord one voter equal power to another voter. I could ask Wolley to explain how to test, mathematically, a proposed voting system for “vulnerable to strategic manipulation”, however, I have asked him for clarifications before of some of his expressions and generally he has not provided any. So, based on experience, there is little reason to think that asking him now would bear fruit.

and the intermediate values are not well-defined and thus incomparable.

This is manifestly false. The voting system itself uses the intermediate values, summed, in the very comparison that determines the final (partial) ranking that determines who is elected (unless there is a tie). Given that the system compares the values, they must be comparable. Recall that voting is an exercise in political power. The information it yields has (primarily) the semantics of command, not that of declaration or indication (more nuance on this later — actually there is some declaration or indication, but I argue that the perfection of the accuracy of that part is of secondary importance). The (primary) significance of a vote lies in its effect on the command given by the election, to put so-and-so in office. Whatever is compared for that effect is obviously comparable. If it were not comparable, it would not be possible to compare it, and comparing it is what happens. Therefore, your statement is wrong.

It’s equal to approval voting in most features.


The corporations funding its advocacy

Just because Bush says something does not make it false. You would serve the cause of honest debate better if you focused on the questions related to the comparison of voting systems and did not try to distract attention by referring to peripheral matters such as who is in favor of something. This particular instance of attempted diversion falls under the well-known category of argumentum ad hominem.

bore me.

Your emotional response is irrelevant to the matter at hand.

Waugh cites http://zesty.ca/voting/sim/ and Wolley responds:

Simulations have premises, and the rating system simulations in particular have serious errors in their premises. Voters act together and lie about their desires. When you take that into full account, IRV and STV and party-list turn into the best systems. I’ve studied these systems for decades and have done my own experiments. Rating systems in real elections are very bad, from personal experience experimenting in local Green elections.

Piece by piece:

Simulations have premises, and the rating system simulations in particular have serious errors in their premises.

The simulation I cited certainly uses a simplified model of how voters vote. However, in general outlines it models that area. It assumes that people are interested in how close in opinion the candidates are to themselves. In reality there may be more or fewer than two dimensions of that opinion, and some of the dimensions might be be discrete rather than anything like continuous. The model uses two continuous dimensions. However, what this simulation reveals is that even using inputs and assumptions that are smooth and continuous, some voting systems produce grotesque fractured results. As the author of the simulations points out, it seems very unlikely that a more-complicated real-life situation will produce less pathological results than the simplified simulation does. Rather, I add, real life is likely to be worse. This particular simulation provides insight into the natures of the mathematical functions at the centers of the voting systems under comparison.

Voters act together and lie about their desires. When you take that into full account, IRV and STV and party-list turn into the best systems.

In our discussion of single-winner systems, the only STV that is relevant is IRV, and party-list isn’t relevant. So let’s look at “lie about their desires” and IRV.

The concept of “lie about their desires” requires a nuanced analysis. If a voting system could serve two purposes perfectly, of on the one hand polling people for an accurate account of their desires, and on the other hand according the people equal political power, then the concept of “lie about their desires” would be simple to understand. In that hypothetical (but unrealistic) world, to “lie about ones desires” would mean to provide input to the system (your ballot) tending to make the system not reflect your real desires in its reporting of aggregate desires of people. However, it is a valid mathematical conclusion that no voting system can do both of those functions perfectly. As progressives, we need the system to elect the person based on equal power, and we need that at the expense of whatever we might have to give up in terms of the indicative information provided in the results.

Nonetheless, we have important progressive reasons to be interested in the secondary information that rating systems provide in addition to the primary signal, which is the command signal to put so-and-so in office. This secondary information gives all readers a measure of how well the losing candidates did and how popular they are. This measure is useful in drawing peoples’ attention to candidates they had ignored as “fringe”, to consider it worth the time to see what those candidates stand for. Such drawing of attention is necessary so that Green positions and candidacies can rise in popularity, and not be stuck on the fringe forever.

I have referred you to arguments by others that demonstrate that given the mathematics of IRV, there are times when a party or faction’s best short-term interests are served by betraying their favorite candidate(s). If that is not “lying about their desires”, I don’t know what is.

The only lying that Range provides an incentive for is shifting the voter’s degree of support or opposition of compromise candidates, without inverting any orders in the (partial) ranking derivable from the ballot. There is no incentive to move down your favorites, nor to move up those you hate the most (love and hate in Range are symmetric). This is a sharp contrast with IRV, where the incentive is to reverse the order of your preferences involving your favorite(s). In exchange for putting up with this degree of partial blindness that Range imposes on analysts of the results in regard to voters’ real opinions about some of the candidates, we get that Range works best in the primary purpose of the system, that is, determining the winner. It would not make sense to confuse an election with an opinion poll and choose a system based on how it reflects opinion at the expense of ending up with a system that elects the wrong candidate. However, Range reveals support for losing candidates much better than vote-for-one Plurality does, to a degree that progressives should see as crucial. The New York poll shows that the voting system matters much in determining outcomes; I can cite that if you ask.

So, IRV comes out much the worse when we look at “lying”. It causes the kind of “lying” that results in electing the wrong candidate in addition to providing wrong information about the levels of popularity of the losing candidates. Range does better in both regards.

I’ve studied these systems for decades and have done my own experiments. Rating systems in real elections are very bad, from personal experience experimenting in local Green elections.

By the parts:

I’ve studied these systems for decades

The amount of time you have spent is not relevant and is a distraction from the argument. No matter how long you have studied, you could have erred. You might just have personality flaws that make you overlook important considerations, even for decades. Let’s please stick to the topic and the logic that bears on it, not something peripheral that looks like an attempt to either build up someone as an authority or tear down someone as an authority. Those are just distractions. Respect for those who take the time to read our conversation requires staying on topic.

I’ve studied these systems for decades and have done my own experiments. Rating systems in real elections are very bad, from personal experience experimenting in local Green elections.

Here you cite work you say you have done, but despite my questions, you keep this work secret. You have not revealed the nature of the experiments you claim to have performed. Accounts of experiments can contribute very validly to discussions, but to do so, they have to give the usual information included in reports of experiments and their outcomes. And of course the most useful experimental results are those that have been repeated. You haven’t exactly earned my trust here as an honest reporter of experiments and their results given the rhetorical devices you have resorted to in this discussion. But if you do describe experimental method, there may be an opportunity for analysis.

Here in Oregon we had range advocates team up with a long list of corporations to try to get range and range-like top two systems in place. The progressives battled them back and won.

Quoting again by pieces:

long list of corporations

ad hominem

get range and range-like top two systems

Here you put “range” and “top two” in the same breath, trying to conflate things that aren’t very related and confuse our readers. A top-two scheme involves more than one election or stage of election in which the voters receive the opportunity to cast ballots. Let us confine our discussion to what happens in one stage or election where that opportunity exists. When we began the discussion, that is what it was about, not proposals involving more than one stage of balloting. If we should find reason to veer during the conversation into such proposals, then what happens at each stage will be at the crux. To mention a two-stage election without saying what happens at each stage just throws a wall in front of what matters.

The progressives battled them back and won.

Maybe some of the people who sided against Measure 90 thought they were serving progressive values, but there is reason to think that if they thought that, they were mistaken. Had the measure passed, there would have been increased opportunity to convince people to support equality of voting power (which progressives should want), because the preamble of the measure stated that as a requirement. Therefore I was convinced that the progressive position to take on it was in favor. Since our debate to start with is about what is in the Greens’ interest, and Greens are presumably progressive, you somewhat beg the question when you insert an offhand assumption about which side of the Oregon question would have served progressive values.

A range vote closer to zero is equal to not voting at all.

This is demonstrably false. No matter what ballot you write for a Range election, I can show an example of the other votes such that adding yours will convert whichever candidate you name between winning and tying or between tying and losing. Please do not make false mathematical statements. I thanked you for correcting me on terminology regarding NP-complete and related concepts (of which I am a very poor student to date), and to behave reciprocally to that would mean you would take this opportunity for correction of your understanding regarding the effect of a Range ballot on the electoral outcome.

I favor a variant of Range that does not allow an explicit statement of abstention or “no opinion” on a candidacy as distinct from an action or inaction that results in a score being used in the tally for that voter-candidate pair. Lack of support amounts to full opposition. However, when I say this, I am not underhandedly switching definition of a Range election to make a statement of yours that was true in the first place look false. Your statement didn’t mention a voter taking the option that is present in some variants of Range that some people describe, to explicitly vote “no opinion” on a candidate, nor to have the default treatment regarding a write-in that others write in but whom you do not write in, to treat that as “no opinion” i. e. average without counting it. What you said is that voting zero amounts to abstention (I’m not sure whether you assume zero at the bottom of the range or in the middle of the range, but you are wrong in either event). I’m just throwing in this information about what version I favor, so that if you choose to take a new look at Range, you look at my favorite version rather than wasting time re-analyzing variants I don’t like.

Otherwise it behaves like first past the post systems.

You have not demonstrated this with any reasoning or data or examples or anything. I can lay out cases where the clear benefit to a voting faction is not to bullet-vote.

Further, a vote for a second choice person must be equal to your first choice in order to actually carry weight to your second choice in as much power as your vote has.

This is true, however, it is part of the price to be paid for according the voters equal power over who gets elected.

Range systems are mainly used in insignificant feedback systems without serious political consequences.

Even innovations that turn out to have merit take time to become properly understood for the value they can bring. The current pattern of usage and non-usage could be an accident of the point in history at which we have this discussion. What we’re discussing is the merits or demerits of advocating change in regard to that pattern. Saying where we are is therefore not very relevant to the discussion of in which direction we should be moving if we want to get elected.

According to at least one source, bees use Range to determine where to move as a colony. This is very “politically” significant for which alleles push out which others in the race for survival.

Comparison/order systems are the gold standard in social choice theory and are resistant to strategy since they must compare against others and create a strict ordering.

Comparison/order systems are the gold standard in social choice theory

“Gold standard” is a pretty loaded expression. What is a reader to take from it? Why should we consider it the gold standard? Are there reasons for such a high valuation? I think you are trying to insinuate here that there is some community of scientific authorities on this question just as there is for say quantum mechanics, that it is some largely settled science and the reader should think there is this huge weight of opinion based on expert knowledge that they should defer to. But it’s not clear to me that this field has that nature. Anyway, if there are reasons for such valuation of ranking systems, you should trot a few out instead of using suggestive wording.

and are resistant to strategy

You have not favored us with a definition of “resistant to strategy”. When you insert words and refuse to define them, you are inserting noise to tire out readers before they can find the energy to sift out the actual arguments.

since they must compare against others and create a strict ordering.

Non sequitur.

Without that strict ordering, information is lost.

What information? A better statement is, without that strict ordering, noise is not made up out of whole cloth. How is a voter to rank two candidates that the voter rates as equal in value?

The goal of reforms is to maximize [information] available to compare candidates against each other.

A secondary goal of revolution (anything that smacks of “reform” sounds like an effort to keep the current power structure in place) is to increase the expressiveness of the vote. But maximizing information available to compare candidates against each other is not the goal of changing the voting regime. The primary goals should be to implement voting equality, and take away, to the extent feasible, the incentive for lesser-evilism and such features that make people care about electability to the total detriment of the system revealing in the end something about people’s judgements about the merits of the ideas and candidacies. If you see maximizing information available to compare candidates as an intermediate step that is necessary to achieve the political goals, you should argue to show that that is true.

Expressiveness is either related to what you said about information or exactly the same thing. It is a secondary goal. In regard to expressiveness, ranking falls down compared to rating because ranking doesn’t gather any information about the degree to which the voter favors a candidate other than the ranking relative to the other candidates (and as I pointed about above, IRV provides an incentive to lie even about the ranking). So, in regard to quantity and quality of information collected, there are some regards in which it makes sense to see Range as outdoing IRV.

Partial rankings can be derived from ratings, but ratings cannot be derived from partial nor strict rankings. That should tell you something about which form carries more information, and which operations throw away information.

Without that strict ordering, information is lost. The goal of reforms is to maximize information available to compare candidates against each other. That is the actual metric one must use to determine the power of a vote. [spellings corrected]

Questions about the power of the vote can be answered by examining cases where a vote shifts the outcome. You have not brought reasoning to argue about under what circumstances a measure of the quantity of information bears either necessarily or sufficiently on the effect on the outcome. Obviously throwing away all the information would result in failing to convey the political command signal from the voter to the outcome. But reasoning about power should refer to the outcome, directly or indirectly.

If everybody voted first and second choice as max values, as they must, there will be no way to determine the real preference.

It is not always the case that they must. But I agree that in some cases, they must. Even when they really do have a preference. But the equality their power is more important than a complete extraction of their valuations, when these goals conflict.

Responding to all your statements like this is going to take me more than one session. I can’t devote my whole life to it. So I offer the present installment and hope to add later.

This entry was posted in Uncategorized. Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.