[EM] Scoring (was Re: OpenSTV 2.1.0 released)

[Juho Laatu]

On 23.9.2012, at 8.01, Michael Ossipoff wrote:

> On Sat, Sep 22, 2012 at 6:11 PM, Juho Laatu <juho4880 at yahoo.co.uk> wrote:
>> On 22.9.2012, at 22.06, Michael Ossipoff wrote:
>> 
>>> 2. Your statement above implies that Symmetrical ICT doesn't choose as
>>> well as [...what?] when
>>> people rank sincerely. That statement requires specification of what
>>> method(s) choose(s) better than SITC under sincere voting, and why
>>> that is so.
>> 
>> Give me a description of who would be the best winner with sincere votes in the election that we talk about.
> 
> You're the one who wants to use the notion of "the best winner with
> sincere votes". Odd that you need to ask me to describe your ideal
> sincere winner. If you want to object that ICT and SITC don't choose
> the ideal sincere winnner well enough, then you're the one who needs
> to say what you mean by "the best winner with sincere votes".

Note that my opinion is that different elections may have different criteria. I mentioned one possible criterion of the best winner as an example, but that need not be your target (or not in all elctions). You could also in principle declare the operative definition of ICT or SITC as the definition of an ideal winner (with sincere ballots), but I'd have some doubts on the genuity of that claim since I believe that those methods have some strategy defence flavour embedded in them, and it doesn't sound probable that the strategy defence algoritms and ideal winner definition would coincide. If you pick some other definition of ideal winner, then it is obvious that ICT and SITC sometimes deviate from that ideal.

> 
> But, with all sincere ballots, many like the idea of electing the CW:
> The candidate who pair-beats each of the others, when such a candidate
> exists. When "CW" is legitimately defined, when equal top and equal
> bottom ranking are interpreted consistent with the preferences, intent
> and wishes of people voting in that way, then SITC elects the CW.
> Beatpath, VoteFair, and all unimproved Condorcet methods fail to elect
> the legitimately-defined CW.
> 
> So, there is a popular "ideal sincere winner": the CW.

CW could indeed be part of the definiton. All Condorcet methods would be partially ideal (i.e. when there is a sincere CW).

> 
> You said:
> 
> Then one can tell what the best method with sincere votes is (or at
> least give some directions). I don't know what the philosophy of ICT
> and the other mentioned related methods is if we assume sincere votes.
> 
> [endquote]
> 
> A sincere-voting property of SITC is that it elects the
> legitimately-defined CW. It's the method that does that.
> 
> You see, what you're missing is that the same disregard for voters
> preference, wishes and intent tha makes unimproved Condorcet fail FBC,
> also makes it fail the legitmately-defined Condorcet Criterion, and
> fail to elect the legimately-defined CW.
> 
> So, meeting FBC doesn't require some sort of violation of the choice
> of ideal sincere winner. On the contrary, it comes with the election
> of the ideal sincere winner, because both gains come from respecting
> the voters' intent and preference.

Do you mean that the voter should help e.g. by falsifying her sincere preferences by voting some candidates tied at top? :-)

> 
> You said:
> 
> My understanding is that they have been designed to resist certain
> strategies, not only to pick the best winner with sincere votes.
> 
> [endquote]
> 
> As explained above, Symmetrical ICT avoids favorite-burial need
> precisely _because_ it respects voter wishes. And if the CW and the
> Condorcet Criterion are defined according to voter wishes, then
> Symmetrical ICT is the method that elects the CW when there is one,
> and meets the Condorcet Criterion.
> 
> You said:
> 
> Therefore there must be another method that elects the best winner
> (based on the definition that you gave) more often than they do.
> 
> [endquote]
> 
> So that's your best argument: That, because SITC meets FBC, there must
> be a method (unspecified by you) that does better under sincere
> voting.
> 
> The reason why you don't specify a method that does better than SITC
> under sincere voting is because you don't even know what a method
> should do under sincere voting. You ask me to describe the ideal
> sincere winner, because you don't have any idea what the ideal sincere
> winner should be.

True in some sense. But if you allow me to define the sincere winner for you, and for your election, you could take my example definition and compare it to SITC. That could be valid for some needs. There is a difference between those two definitions.

> 
> But I gave you a suggestion: The legimately-defined CW.
> 
> Do you see the irony here? Someone who doesn't know what the ideal
> sincere winner is, wants say that surely there's some method that
> chooses it better than Symmetrical ICT. He just doesn't know what that
> method is, because he doesn't have a suggestion for what the ideal
> sincere winner would be.
> 
>>>> This
>>>> means that one should deviate from the method that picks the best winner
>>>> with sincere votes...
>>> 
>>> ...which is _what_ method?
>> 
>> There are different best methods for different needs.
> 
> Translation: You can't name one.

I already named one (see two lines below).

> 
> You said:
> 
>> (In my text above I asked you to provide a definition of the best candidate. A simple Condorcet oriented definition could be e.g. "the candidate that requires least additional support/votes to beat any of the other candidates in a pairwise comparison/battle should be elected". This target could be selected because it gives one rational argument why the winner could be able to rule well (= only little bit of additional support needed (if any) to gain majority support for his proposals while in office).)
> 
> [endquote]
> 
> That sounds like Dodgson. Sure, it can be justified as you say.

That was supposed to be equal to Minmax(margins).

> 
> You want to elect the best winner under sincere voting. Is that what
> you claim is the best winner under sincere voting (ideal sincere
> winner)?

That could be the definition for some needs.

> 
> You described a circular tie solution, for when there's no CW. And
> presumably the CW that you'd choose would be the traditional (as
> opposed to legitimate) one.

That definition covers also the CW (= no additional support needed).

> 
> So, for one thing, your method wouldn't choose the candidate who
> legitimately beats each one of the others under sincere voting.
> 
> And, if you say that SITC's justification doesn't apply in sincere
> voting, then you'd be mistaken.
> 
> Say you like X and Y equally, and you rank them both in 1st place.The
> fact that you like them equally and rank them sincerely in 1st place
> doesn't mean that you want one to beat the other and give the election
> to your last choice. SITC respects your wishes better than unimproved
> Condorcet, even if your equal top ranking is sincere.
> 
> For another thing, everyone agrees that the election of the CW under
> sincere voting is a lot more important than how you choose when there
> _is_ no CW. So the circular-tie-breaker isn't so important.

Maybe that is your definition. Maybe you mean that all candidates are equally good if there is a top loop. In that case the ideal method would be one that picks a random candidate if there is no CW. (Or do you want to limit the lottery to a smaller set of candidates?)

> 
> For yet another thing, natural circular ties are considered to be
> uncommon. Certainly they'd be rare, to the extent that the scenario
> approximates a 1-dimensional political spectrum, in which issues are
> strongly correlated (as they tend to actually be).
> 
>>>> ...only if one is certain that otherwise the method would
>>>> give even worse results because of strategic voting.
>>> 
>>> ..even worse than what??
>> 
>> A method that has been modified to cope with strategies does not elect the ideal sincere winner > always.
> 
> Does any method? You don't know, because you don't know what an ideal
> sincere winner would be.

Take my example definition above.

> 
> Aside from that, I answered that statement above.SITC elects the ideal
> winner better than unimproved Condorcet does, because it better
> respects voters' intent and preference.
> 
> When no candidate beats everyone, SITC elects from the set of unbeaten
> candidates. When no one is beaten, SITC doesn't necessarily choose
> from the Smith set--probably not even from a legitimately-defined
> Smith set. Choosing from the Smith set would lose one or more of
> SITC's desirable properties. But I remind you that the time when it
> matters who wins under sincere voting is when there is an unbeaten
> candidate. Or more than one unbeaten candidate. When there's no CW,
> it's anyone's guess who should win, even under sincere voting.
> 
> Dodgson's choice is one of many opinions about who should win under
> sincere voting when there's no CW.
> 
>>> You haven't said what are the bad results of
>>> SITC and ITC that we need to avoid.
>> 
>> See my first comments above. Their deviation from the ideal should become visible after one defines the ideal sincere winner.
> 
> Ok, let's define the ideal sincere winner as the legitimately-defined
> CW.

This would be another partial definition.

> Symmetrical ICT is the method that chooses the
> legitimately-defined CW. Unimproved Condorcet doesn't.
> 
>>>> Obviously you believe
>>>> that basic [unimproved] Condorcet methods would attract certain strategies to the extent
>>>> that those methods must be fixed. And I believe that in most societies it is
>>>> more likely that strategic voting will be marginal.
>> 
>>> Because I don't know what method you're referring to, of course
>>> there's no way to answer your expression of belief.
>> 
>> I referred to basic Condorcet methods. (Ranked Pairs, MInmax,...)
> 
> And remember that unimproved Condorcet also has the chicken dilemma.
> You said that you don't think that people would favorite-bury. I've
> answered that amply, but do you also believe that there will be no
> chicken dilemma?

Condorcet methods are not very prone to that.

> Because, if the chicken dilemma will happen, then it
> will happen in unimproved Condorcet, because unimproved Condorcet has
> the chicken dilemma. SITC doesn't have the chicken dilemma.
> 
> Your talk of sincere voting loses even what relevance it had before,
> when I remind you that unimproved Condorcet has the chicken dilemma.

What claim are you referring to and what is the problem with it?

> 
>>> ...aside from the fact that I make no claim to know what's true about
>>> more than 1/2 of all societies.
>> 
>> With "most" I wanted to say that I don't expect many societies to converge towards widespread >strategic voting.
> 
> That's nice. I don't propose Approval, Score and SITC for "many
> societies". I propose them for the U.S.  (Approval and (maybe) Score
> for official public elections; and SITC for informational polling--but
> Score for informational polling when there isn't agreement about
> choice of rank-count).
> 
> You said:
> 
> I start from that assumption and I want evidence before deviating from
> that assumption.
> 
> [endquote]
> 
> Unsupported assumptions tend to lead to error.
> 
> If you choose some (unspecified) method that doesn't offer good
> strategy-guarantees, and it turns out that your strategy-free public
> is a mistaken assumption, then you've made a mistake.
> 
> SITC doesn't depend on an all-voters-sincere assumption. It would work
> fine with or without strategically-inclined voters. It would work fine
> with or without everyone completely non-strategic.
> 
> And remember, I asked if you think that the chicken dilemma, too,
> won't happen.  ...In other words, I ask if you think that there
> wouldn't be defection in the familiar chicken dilemma example, in the
> U.S. in particular, or in "most societies".
> 
> You depend on an assumption. You think that maybe your assumption is
> right. Sometimes "maybe" isn't good enough. You don't claim that there
> wouldn't be favorite-burial in the U.S. You say you've had
> conversations with Americans. Does it occur to you that there could be
> some "sample bias" there?

There certainly will be. (some)

> How often does Joe Sixpack really get to
> Europe?
> 
> Do you also assume no chicken dilemma?

So far I don't expect that to be a major problem in Condorcet in the U.S.

> 
> SITC doesn't need an assumption of sincere voting. Nor does it fail to
> choose well when everyone votes sincerely.
> 
> All the fighting about which unimproved Condorcet version is best is
> quibbling about what is best when there is no CW. ...in other words,
> about the least important circumstance, the time when there is a
> natural (sincere) circular-tie.

Also artificial loops are part of that debate.

> Aside from that, as I said, natural
> circular ties are considered uncommon.
> 
> You said:
> 
> This is a very traditional process. Nothing new in it. I'll give an
> approximate description of the process in Finland. In the polling
> station there are many representatives from many parties, monitoring
> the process. The votes are counted (information collected) right after
> the election ends, again together on one table by multiple people from
> multiple parties. After that the votes are sealed and sent for
> storage. I don't recall any serious problems or complaints. With
> complex votes the process would take more time, and there could be a
> need to double check, but the principles would probably stay the same.
> 
> [endquote]
> 
> I guess I must have missed the part about digitization. Anyway, though
> the ballots should, of course, be saved, they should also be _tallied_
> right there in public, in front of the multiparty observers. That's
> what I suggest for Approval counts.
> 
> You can't do that for a rank method. You can do it for Approval.
> 
>> 
>>> Did the voter hirself make out a paper ballot? Or
>>> was it made electronically by a voting machine (presumably, but not
>>> necessarily based on the voter's voting)?
>> 
>> Use of paper ballots is the traditional and reliable method. Ballots that are printed by a machine >are still theoretically quite safe,
> 
> ...only if the machine is printing what the voter says :-)
> 
> You continued:
> but in practice not quite at the same level.
> 
> [endquote]
> 
> No, not quite :-)
> 
> 
>> 
>>> And, if the voter made out the paper ballot, then you've got millions
>>> of paper ballots, distributed around the country (our large country).
>> 
>> No need for transport around the country. They could as well be sealed and stored locally after >ballot data has been collected.
> 
> Juho, our local precincts are all around the country.
> 
> You continued:
> 
> And maybe better so (to store them under the eyes of all the locals
> who don't want their opinions to be falsified).
> 
> [endquote]
> 
> Not wanting their opinions falsified, and not getting their opinions
> falsified, aren't quite the same thing.
> 
> Security can't be perfect.
> 
> Best to do the actual tally of the candidates totals during the public
> count process. You still save and store the ballots, but you don't
> depend on them for count verifiability.
> 
> That tally during the public count procedure pretty much disqualifies
> all rank methods, and anything other than Approval or Plurality. Maybe
> Score.

Getting the ballot content is enough. After the ballot data has been safely recorded and shared, algorithmic computations can be done by computers, and can take place anywhere and by anyone.

> 
> 
>> 
>>> How do you propose to show them to anyone who wants to look at them?
>> 
>> I don't. For most EM experts the digitized information should be enough. Only if the end product of the digitization is really what the voter wrote on hir ballot.
> 
> You said:
> 
> A different procedure applies for the ballot digitization.
> 
> [endquote]
> 
> Yes, that's the problem, isn't it.

I think both digitization and result calculation can be made pretty safe. In the digitization and data sharing part I assume a well organized and only mildly corrupted society.

> 
>> 
>>> And, as I asked before, why do you want to be guessing and hoping,
>>> when it's possible to simply be rid of favorite-burial incentive?
>> 
>> There are numerous different threats, and we should minimize them, starting from the worst >problems.
> 
> What are these other threats that you refer to? Let me guess: You
> don't know, and you were going to ask me :-)

Depends on the method. The EM list is full of various threat discussions.

If you want to know which problems are the worst problems, my most frequent comment is to request practical examples instead of listing theoretical vulnerabilities. That is because practical examples make it easier to estimate how serious some threat is in practical elections in the given environment. That means that you have to estimate all the probabilities, including ability of voters to determine when and how to use the strategy, probability of success, risk of making the result worse etc. It is not easy to give an exact answer to which threats are the worst ones, but it is possible to come quite close. And if not on paper, the actual elections could point out where the leaks are (if any).

> 
> I've compared Symmetrical ICT to unimproved Condorcet, by a number of
> criteria, including Condorcet's Criterion. I've compared them in
> regards to electing the legitimately-defined CW.

I might repeat my most frequent comment here :-). Well, the point is that we should not defend against theoretical vulnerabilities but against ones that are likely to hurt us in real elections. And we should keep in mind that there is also an interest to elect the best candidate, not just to tweak the method so that a selected set of threats are not possible even in theory.

> 
> You said:
> 
> Sufficient protection against favourite burial is needed, but 100%
> protection (against all possible strategic ballot sets) may well be
> too much. One need to defend only against practical threats. Same with
> all other threats.
> 
> [endquote]
> 
> With you, of course, deciding how much protection is practical. Such
> hoping isn't needed when a method outright doesn't have the problem.
> 
> You want to imply that, though SITC doesn't have a favorite-burial
> problem, or a chicken dilemma, or a Later-No-Help failure, and elects
> the legitimately-defined CW (unlike unimproved Condorcet), there is
> some other need that it doesn't meet, or some other threat that it
> doesn't protect against, or some other necessary guarantee that it
> doesn't offer.. But you can't tell us what it is. Your best argument
> is that there _might_ be one :-)

Oh no, I only made a theoretical claim that methods that aim at eliminating some strategies and definitions of ideal winners are usually not the same, and now you challenge me to make a complete analysis of your favourite method :-).

Maybe I'll make a very light analysys and ask you, do you expect voters to flatten some of their sincere preferences to ties to achieve all the advertised benefits of the method?

> 
>> 
>>>> I know at least one person that is inclined to bury and probably would bury
>>> 
>>> Then you know that the problem is genuine.
>>> 
>>>> :-).
>> 
>> Probably there are some random irrational strategic votes in every large election :-).
> 
> ...And it's only a question of how many "some" is. You don't know. You
> speculate that it won't be many, "in most societies" or "in many
> societies". I'm more concerned about this particular society, in this
> country. You admit that you don't know about that, and there's no
> reason for anyone to say that you should know about that. You clarify
> that you're only speaking in general, about "most societies" or "many
> societies". Fine.
> 
> But as for _random_ strategic votes, they aren't really random.
> They're predictable and systematic. They're heavily recommended by the
> mass media. You seem to be missing that.

I didn't exclude that.

> 
> And the favorite-burials are indeed irrational, in the sense that the
> premise for choosing them is irrational. But, given that premise, the
> favorite-burial is quite rational, optimal strategy.
> 
> 
>> 
>>>> But don't take medicine
>>>> that harms you more than the disease.
>>> 
>>> Ok, so then tell us how ICT and Symmetrical ICT would do harm.
>>> 
>>> And tell us your proposal for a method that wouldn't.
>> 
>> Again, see the beginning of this mail.
> 
> Translation: You can't name a way that SITC would do harm, and can't
> name a method that would do better.

Why should I be interested in doing that? My claim here was that strategy resistance usually means deviation from electing the best sincere winner (with free choice of the best winner criterion). This is a generic claim, not SITC specific nor U.S. specific. You can get around it by claiming that SITC actually is your definition of ideal winner with sincere votes (and SITC style interpretation of the ballot preferences).

Juho


> 
>>> "Next time, don't let that happen. Next time, rank the Democrat in 1st
>>> place, because otherwise the Progressive will win again."
> 
> You replied:
> 
>> My guess is that Republican leaders (in Burlington) would not make such a recommendation unless it would be very clear that the Republican candidate can not win.
> 
> [...]
> 
> Fine. For national elections, the mass media continually, daily,
> hammer home the message that the Dems and Repubs are "the two
> choices".  ...the message that no one can win other than the Dems and
> Repubs. With great success, in fact, the media often imply that there
> _are_ no other parties, and that there are no non-disastrous
> positions, no positions that should even be considered or heard, other
> than the Dem and Repub policies.
> 
> So yes, the media do indeed "make such a recommendation." They make it
> every day, on every channel and station, and everyone accepts it.
> 
> It would be nice to educate everyone differently. I don't think the
> mass media want to.
> 
> But there are voting systems that, even to someone who believes the
> media (which is most everyone), there would be no reason to do
> favorite-burial.
> 
> Approval, Score, and Symmetrical ICT are a few of those methods.
> 
> You speak of "most societies". Maybe "most societies" aren't
> (voluntarily) a captive audience of our mass media.
> 
> Mike Ossipoff