Wednesday, June 30, 2010

Pro-Approval Video from NYU Professor

Steven Brams, Professor of Politics at New York University, runs down the advantages of approval voting and refutes many of the arguments against it, in this video. Worth the five minutes.


25 comments:

  1. Here's a way we could make the presidential election an "open" primary that would determine the final three candidates, with the final winner determined by the electoral college.

    dlw

    ReplyDelete
  2. Yes, but: IRV still gets wrong answers with 3 candidates.

    Are you familiar with the Gibbard-Saiterwaithe theorem? It's possibly more-important than even Arrow's:

    http://rangevoting.org/GibbSat.html

    No ranked-order ballot method (including IRV) satisfies G-S when there are more than 2 candidates.

    Cardinal ballot methods (like approval and score) satisfy G-S for 3 candidates (but not for more), so if you can narrow it down to three, you ABSOLUTELY should use one of these methods.

    ReplyDelete
  3. DSH,

    A couple of thoughts.

    1. The dictator criterion is not a deal killer so long as there is some uncertainty and turn-around as to who plays "dictator"....

    2. Methinks that criterion 2: If every voter ranks X top, then X wins the election is not the problem when there's only three options and IRV is the option under consideration.

    3. Strategic voting is not that bad, it's inevitable at some point and not the same thing as lying. It's good to mitigate it, but it doesn't poison outcomes in terms of making the outcome reflect general viewpoints.

    I'm afraid that you're thinking a bit like an engineer, rather than an activist. You want the best system that meets the quality specifications, but there's room for discretion on some of these things.

    So I'm saying keep it simple. Tabulations of 1st choices and 2nd choices from 3 candidates are very easy to tabulate. Thus, the vote counting can be done easily in a decentralized manner that helps to prevent gaming of the system. Range-voting, by providing oodles of options to voters, doesn't summarize as easily and can increase the scope for wrong-doing by unscrupulous party-machines. They can both create votes and create high-votes. And I strongly doubt that most voters are going to put the time and thought into their range-votes so as to choose other options besides the top-ranking very often. In which case, it'll tend to be a wash for the most part...

    dlw

    ReplyDelete
  4. Good eye; I am an engineer. Here's my engineer-opinion on your points:

    1. Psychologist find that, with three options, ranking or rating is about as cognitively difficult; with more than three, rating pulls ahead. Therefore, I think your intuition on this is wrong.

    2. IRV, even with only three choices, still requires centralized coordination to adjudicate. Approval (and range), since they have no "elimination rounds", do not. In fact, approval requires no-more information to be communicated from precinct to HQ than a plurality election does: the number of votes each candidate received. IRV requires, at least, the number of votes each permutation of candidates received; with only 3 candidates, that's just 3!=6, but, as I'm sure you're aware, that number increases VERY quickly. So, I believe, your intuition is wrong. All of your other claims of unscrupulous behavior are equally applicable to EVERY voting system: approval, irv, plurality, etc.; so there applicability as disadvantages specific to approval is nil.

    Enough for point-by-point though,
    as you seem to be ignoring my central premise:

    It is a mathematical fact that, in 3 candidate elections, IRV runs up against a fundamental impossibility inherent in social-decision science; one which approval voting and score voting are not beholden to, and because of this I believe that, in a three candidate election, they will lead to better outcomes. Computer simulations back up this assertion.

    I don't care what your "activist intuition" says; we only get to address these sorts of issues once in a lifetime, and IRV would be nothing but a colossal waste of effort.

    Your intuition is wrong. I have presented evidence in support of my claim that your intuition is wrong, and your attempts to justify your intuition are grasps at straw. I am not insulting your intelligence--and please do not insult mine--but you need to know that, in this case, your gut is wrong.

    If you want to fight for IRV for the sake of fighting for IRV, please, go ahead.

    If you want to fight for the best possible electoral reform in a hope that this will bring about political change for the better, than please, listen to me, and fight for approval or score (or any form of proportional representation), not for IRV.

    ReplyDelete
  5. 1. Psychologist find that, with three options, ranking or rating is about as cognitively difficult; with more than three, rating pulls ahead. Therefore, I think your intuition on this is wrong.

    dlw: But we're arguing on the case with 3 choices, like with my "top 3 IRV", right?

    2. IRV, even with only three choices, still requires centralized coordination to adjudicate.

    dlw: no, there are 6 options, besides the "none of the above". the seven possible choices can be spelled on a ballot like we do today.

    So punch (a, b), (a, c), (b, a), (b, c), (c, a), (c, b) or none. Then, the results can be tabulated very easily in a decentralized manner...

    DSH: IRV requires, at least, the number of votes each permutation of candidates received; with only 3 candidates, that's just 3!=6, but, as I'm sure you're aware, that number increases VERY quickly.

    dlw: Hence, the reason I'm arguing for "top 3 IRV", not IRV3, except for municipal elections where there won't be too many candidates...(and even there I'd prefer a limit based on ability to gather signatures and/or incumbency).

    DSHLSo, I believe, your intuition is wrong. All of your other claims of unscrupulous behavior are equally applicable to EVERY voting system: approval, irv, plurality, etc.; so there applicability as disadvantages specific to approval is nil.

    dlw: The more options, the more opps for unscrupulous behavior that's my intuition and you haven't disproved it one whit... Plurality only let's folks stuff ballots, they can't rank the "right ballots" or hide it by varying which of the nobody candidates is the first pick. It is preferable to other options in that regard, since it puts some natural limits on vote-stuffing.

    DSH: IRV runs up against a fundamental impossibility inherent in social-decision science

    dlw: In practice that fundamental impossibility is less of a hindrance than in theory. And so it pays to keep it simple. It also pays to support election reforms strategically, since we're still using a first-past-the-post system and the alternatives we're discussing are all better than first-past-the-post in most respects.

    dlw

    ReplyDelete
  6. IRV is a disappointment, in my opinion, largely due to how it's still a winner-take-all election and there tend to be one or two major candidates in those elections, along with a whole bunch of me-toos that most voters disregard. But when it's set up to be a 3-way election then that changes things. It makes it so that the incentive for the top two candidates to campaign negatively is removed and the third candidate has a fighting chance of winning. Especially when all three are relatively "centrists", so the election becomes more interesting and uncertain. Then, moneyed-intere$t$ need to hedge and inevitably get a lower return on their investments and minority voter-groups, willing to "vote strategically" can get more attention to their key issues that perhaps were neglected in the primaries.

    dlw

    ReplyDelete
  7. @dlw

    The dictator criterion is not a deal killer

    Any system where you pick a "dictator" voter to choose the winner will essentially have a social utility efficiency of zero. The same quality you'd get by having non-democratic random selection. I.e. terrible.

    Methinks that criterion 2: If every voter ranks X top, then X wins the election is not the problem when there's only three options and IRV is the option under consideration.

    IRV satisfies this criterion, if that's what you meant.

    Strategic voting is not that bad, it's inevitable at some point and not the same thing as lying. It's good to mitigate it, but it doesn't poison outcomes in terms of making the outcome reflect general viewpoints.

    This commentary is mathematically meaningless. In the vast majority of voting methods, strategic behavior results in less socially pleasing outcomes.

    Score Voting with 100% sincere voters gets about a 96% social utility efficiency, whereas with 100% tactical electorates it gets about a 78% SUE. Approval Voting goes from 85% to 78% (SV = AV with 100% tactical voters).

    IRV goes from 77% to 39%. So not only is it much more harmed by tactical voting than SV or AV, but it also starts off worse "out the gate"".

    These are concrete metrics rather than vague and/or subjective sentiments.

    I'm afraid that you're thinking a bit like an engineer, rather than an activist. You want the best system that meets the quality specifications, but there's room for discretion on some of these things.

    Voting science is an extremely mathematical subject, which so many people misunderstand largely because they lack expertise in skill in mathematics and general logical reasoning, which form the bedrock of engineering.

    So I'm saying keep it simple.

    Score Voting, and especially Approval Voting, are simpler than most ranked methods, especially IRV. This can be objectively demonstrated. First, write the shortest possible computer program to tabulate a set of SV or AV ballots, then do the same for IRV. The SV/AV program(s) will be shorter. Also, IRV typically causes more spoiled ballots, whereas SV/AV experimentally reduce ballot spoilage. Voters clearly find SV and AV simpler, because they screw them up less often.
    http://scorevoting.net/SPRates.html

    Tabulations of 1st choices and 2nd choices from 3 candidates are very easy to tabulate. Thus, the vote counting can be done easily in a decentralized manner that helps to prevent gaming of the system Range-voting, by providing oodles of options to voters, doesn't summarize as easily and can increase the scope for wrong-doing by unscrupulous party-machines.

    IRV is not precinct summable. For instance, there are 5040 ways to order 7 candidates, not even allowing for truncation. You think you can get around that by artificially restricting the number of candidates a voter may rank. Aside from the obvious problem that this requires limiting voter expressiveness (SV/AV are precinct summable no matter how many candidates voters rate), it doesn't really even fix the problem. For instance, say you have 8 candidates and ballot truncation is allowed. That's 8+56+336 = 400 different "pseudo-candidates" you'd have to keep track of to use your hack.

    ReplyDelete
  8. @dlw

    Continuing...

    Here in my home of San Francisco, we use IRV with up to 3 rankings allowed. If you went to the San Francisco government website to seek results of their 2008 November 6th ranked-choice voting elections, you STILL got this message for many months:

    "Due to the requirement that all ballots must be centrally tallied in City Hall and not at the polling places, the Department of Elections has not set a date for releasing any preliminary results using the ranked-choice voting method."

    And I strongly doubt that most voters are going to put the time and thought into their range-votes so as to choose other options besides the top-ranking very often. In which case, it'll tend to be a wash for the most part...

    What is the evidence for your "serious doubt"?

    Warren Smith's Bayesian regret calculations hold true even with high ignorance factors (which simulate these sort of rough approximations).

    Here's a simple example to demonstrate why your intuition is wrong here. Say the honest scores I'd give, were I well-informed about the issues and the candidates, were Paul=10, Obama=8, Richardson=4, Clinton=0. Now, being lazy and/or not well educated, I fudge this up a little, into Obama=10, Paul=9, Richardson=6, Clinton=0. That may seem to you like a significant distortion of the "real values". But then consider that if I were to use Borda then EVEN WITH PERFECT KNOWLEDGE (i.e. knowing the scores that really best represent my well-informed view), I would still cast a ranked vote which the Borda system would effectively translate into equally spaced scores, like Paul=10, Obama=6.67, Richardson=3.33, Clinton=0. That is MORE of a distortion caused by a bad voting method than caused by the human error/laziness you are worried about.

    With IRV, the case is far worse, as it throws away TONS of ballot data.
    http://scorevoting.net/IgnoreExec.html

    It gets even worse when you take tactical exaggeration into account. Say the polls suggest Richardson and Clinton are the "frontrunners". With Score Voting, I would obviously want to give Richardson a 10, to be strategic. But then I could still give Paul and Obama as high a score as I want, with no fear that it would help Clinton to defeat Richardson.

    But in IRV and Borda and Condorcet (pretty much any ranked method that is commonly discussed), maximizing the distance between Richardson and Clinton REQUIRES that I insincerely rank Paul and Obama behind Richardson. That is a far more severe distortion of my preferences than what you're talking about.

    ReplyDelete
  9. @dlw

    IRV is totally susceptible to tactical behavior, even with just 3 candidates.
    http://www.electology.org/debate/IrvPlurality

    In practice that fundamental impossibility is less of a hindrance than in theory.

    Gibb-Sat and Arrow aren't theories, they're theorems. They are mathematically proven. And their consequences are REAL. E.g. when a voting method violates Independence of Irrelevant Alternatives, you know with mathematical certainty that the wrong winner has been chosen in at least one of the two alternative scenarios. The average voter is then literally worse off.

    And so it pays to keep it simple.

    Approval Voting is vastly simpler than any ranked method, particularly IRV. Score Voting is also much simpler than IRV.

    It also pays to support election reforms strategically, since we're still using a first-past-the-post system and the alternatives we're discussing are all better than first-past-the-post in most respects.

    That common line of reasoning ignores opportunity cost. For instance, 4 municipalities (Burlington, Pierce County, Aspen, Cary) reverted from IRV to Plurality recently. Those wells have been poisoned, because now those voters won't be interested in trying an alternative voting method again for a generation or more. So you have to focus voting reform energy on the right alternative, not just any alternative.

    And there is some evidence that Plurality with runoff elections is better than IRV.
    http://scorevoting.net/TTRvIRVstats.html

    ReplyDelete
  10. DSH:Any system where you pick a "dictator" voter to choose the winner will essentially have a social utility efficiency of zero. The same quality you'd get by having non-democratic random selection. I.e. terrible.

    dlw2: The point is that it ain't predetermined which voter is the "median" voter and it is normally not the same voter across elections. So long as there's adequate turnaround in the key issues, the fact that there is a voter whose vote effective decides an election doesn't matter that much in practice. Now, we don't have a healthy democracy right now and there have been some tails that have wagged the dog too much for too long. Election reform will deal with that in part, but it's not all of the answer and the answer won't per se get rid of there being a "median" voter.

    DLW1:Strategic voting is not that bad, it's inevitable at some point and not the same thing as lying. It's good to mitigate it, but it doesn't poison outcomes in terms of making the outcome reflect general viewpoints.

    DSH: This commentary is mathematically meaningless. In the vast majority of voting methods, strategic behavior results in less socially pleasing outcomes.

    DLW2: That's if everyone voted strategically all of the time. I'm simply saying that trying to exorcise the incentive to vote strategically is a chasing after of the wind.

    DSH: Score Voting with 100% sincere voters gets about a 96% social utility efficiency, whereas with 100% tactical electorates it gets about a 78% SUE. Approval Voting goes from 85% to 78% (SV = AV with 100% tactical voters).

    IRV goes from 77% to 39%. So not only is it much more harmed by tactical voting than SV or AV, but it also starts off worse "out the gate"".

    dlw2: Only if voters truly vote rationally wrt AV or SV. What I have learned from Rob Richie is that in practice, most voters don't pick up on the nuances of AV right away. They tend to vote for one of the candidates. Then a subset of the voters who do get AV tend to garner a disproportionate influence that elicits a negative impression of AV from the other voters.

    Likewise, I reckon most voters might vote all of their votes for one candidate, what they'll be encouraged to do so. After all, who wants their vote not to count 100%? In which case, the rule would tend to devolve back into a first-past-the-post rule.

    We simply do not have in our heart of hearts what are the true rankings of all of the candidates. This is a utopic. or heroic assumption, and drives why it seems ideal for us be able to rank all of the candidates. In real life, we all see in part as through a mirror darkly when it comes to assessing candidates and we're often overcome by misinformation. All of which makes it so that giving voters more options a mixed bag, for: Garbage In, Garbage Out.

    to be continued.

    ReplyDelete
  11. This comment has been removed by the author.

    ReplyDelete
  12. This comment has been removed by the author.

    ReplyDelete
  13. DSH:These are concrete metrics rather than vague and/or subjective sentiments.

    dlw:They are metrics, but that doesn't mean they are concrete. When dealing with a complicated subject, words and numbers all fail us at times and we cannot divorce the subjective from the objective.

    I am not driven by subjective sentiments. I see the use of winner-take-all elections as causing increasing returns to scale in election-winning. It leads to there being two dominant parties (aka like Duverger's law) and when only winner-take-all elections are used, it leads to there being only one dominant party. The latter situation is much worse than the former. Thus, we must find ways to use both types of elections. And, due to the inherent fuzziness of candidate options, as well as our tendencies to not do our homework before voting, the number and types of options we give voters turn out to be of second order importance for election reform.

    DSH: Voting science is an extremely mathematical subject, which so many people misunderstand largely because they lack expertise in skill in mathematics and general logical reasoning, which form the bedrock of engineering.

    dlw2: Look, I have a PhD in Economics and was a double major in math/econ as an undergraduate. I get the math and logical reasoning. But all of that can still be folly if the presumptions are misleading. Election reformers need to draw on both deduction and induction and take leaps of faith in experimenting with new ideas.

    dlw1:So I'm saying keep it simple.

    dsh: Score Voting, and especially Approval Voting, are simpler than most ranked methods, especially IRV. This can be objectively demonstrated. First, write the shortest possible computer program to tabulate a set of SV or AV ballots, then do the same for IRV. The SV/AV program(s) will be shorter. Also, IRV typically causes more spoiled ballots, whereas SV/AV experimentally reduce ballot spoilage. Voters clearly find SV and AV simpler, because they screw them up less often.
    http://scorevoting.net/SPRates.html

    dlw2: I'm not surprised AV wd outperform IRV with larger nos of candidates, as the number of options for a combination, instead of a permutation, are smaller and more reasonable.

    But I am mainly arguing here for "top 3 IRV" with three candidates in the general election. Let's say that there's pretty much a "tie" between IRV and AV when there's only 3 candidates. Then, IRV wd trump AV, simply because it's closer to what people are used to.
    This is also why when I use AV in the primary, I constrain people to picking three of the seven candidates. It is easy to explain. The primary needs to pick three candidates so pick which three you'd like to see in the general election the most. Or don't vote...
    --
    dlw

    ReplyDelete
  14. And I'm saying that among highly educated and well informed folks like you, true scoring of the candidates bears a decent semblance to reality, but for most folks that just is not true.

    There are no true values to compare the de facto rankings against for most folks. It is a matter of learning by doing, how we develop our habits and we don't get to learn by voting very often.

    I'm all for evolving beyond IRV3 for San Francisco. But maybe it'd be better to push for using AV or SV in a primary first while keeping IRV for the final election? Or you could focus more on making there be fewer wards with multiple candidates. This can be done, while keeping IRV3. You'll just end up eliminating fewer candidates...

    DSH: Gibb-Sat and Arrow aren't theories, they're theorems. They are mathematically proven. And their consequences are REAL. E.g. when a voting method violates Independence of Irrelevant Alternatives, you know with mathematical certainty that the wrong winner has been chosen in at least one of the two alternative scenarios. The average voter is then literally worse off.

    dlw: It is mathematically proven that the incentive to strategically vote does exist for some voters. This says nothing about whether they actually do strategically vote or how perverse such would end up becoming. Likewise, if the independence of irrelevant alternatives is violated then some may try to game the system. Whether they do this or it works out very dysfunctionally in practice is a matter of experience. That a voting system has potential weaknesses does not mean that those weaknesses get manifested in full blooming glory.

    dlw

    ReplyDelete
  15. Perhaps, you could explain the algorithm used for AV in calculating the regret functions.

    I believe folks end up, in practice, heuristically guessing how many of the candidates they want to approve of out of the field when they are given the full field of opportunities....

    dlw

    ReplyDelete
  16. @dlw

    They are metrics, but that doesn't mean they are concrete. When dealing with a complicated subject, words and numbers all fail us at times and we cannot divorce the subjective from the objective.

    You used subjective unquantifiable terms like "good" and "poison" and "general viewpoints". I cited Bayesian regret figures. It is as if you said "that city is far away" and I said "that city is 253 kilometers away". The second statement is

    I see the use of winner-take-all elections as causing increasing returns to scale in election-winning. It leads to there being two dominant parties (aka like Duverger's law)

    Duverger's Law does not say that single-winner (presumably what you mean by "winner-take-all") elections lead to two-party domination. It says that Plurality (aka first-past-the-post) Voting leads to two-party domination, whereas top-two runoff (aka "TTR", or "Plurality with a runoff election if there is no majority winner") tends to escape duopoly. Here are Duverger's words, verbatim:
    http://scorevoting.net/DuvTrans.html

    IRV also maintains duopoly, but it seems probable that Score Voting (and by extension, Approval Voting) would not have this problem, largely because they pass the Favorite Betrayal Criterion.

    due to the inherent fuzziness of candidate options, as well as our tendencies to not do our homework before voting, the number and types of options we give voters turn out to be of second order importance for election reform.

    There is no evidence to support this. Bayesian regret calculations have shown that Score Voting and Approval Voting offer large improvements over e.g. Plurality and IRV, even with a wide range of ignorance factors. And the improvement caused by upgrading to SV/AV is larger than the improvement you get by making the electorate "do their homework".

    And in fact a big part of the reason voters don't do their homework, is that they are wasting their time if they learn about any candidate besides the two obvious frontrunners. That is a result of having a bad voting method. And another reason they tend to think in black-and-white unwavering terms, is that two-party domination lends itself to the human tendency to think in binary "us vs. them"/"good vs. evil" terms. Again, that's a result of the voting method. So you need Score Voting or Approval Voting so that you can break out of that.

    Proportional Representation is not a solution for the USA, because it doesn't apply to single-member positions like governor, senator (staggered terms), mayor, etc. And because PR is federally illegal in the US Congress, hence it probably will not be possible until we FIRST break out of duopoly.

    Look, I have a PhD in Economics and was a double major in math/econ as an undergraduate. I get the math and logical reasoning.

    Well, many of your statements demonstrate gross misunderstandings of the subject matter. I don't think that means you're not capable of understanding it, but you do need to read up on things like Duverger's Law and other core principles if you're going to be taking any kind of position on these things.

    But all of that can still be folly if the presumptions are misleading. Election reformers need to draw on both deduction and induction and take leaps of faith in experimenting with new ideas.

    ReplyDelete
  17. But I am mainly arguing here for "top 3 IRV" with three candidates in the general election. Let's say that there's pretty much a "tie" between IRV and AV when there's only 3 candidates.

    There isn't, because IRV isn't precinct summable, hence the message from our election office about why the results took a long time to appear.

    IRV wd trump AV, simply because it's closer to what people are used to.

    That's both unsound and invalid. It's unsound because Approval Voting uses the same ballot as Plurality, and just removes one single rule (the rule that says you have to stop at one). So it's more similar than IRV is to our current system.

    It's invalid because, even if it were true, that's a very minor consideration compared to performance, simplicity/cost, and transparency. These are areas where Approval Voting dominates IRV, and Score Voting is substantially better.

    This is also why when I use AV in the primary, I constrain people to picking three of the seven candidates. It is easy to explain.

    Exactly. You're using a complex voting system, and trying to simplify it by limiting voter choice. With Approval Voting, you could let them vote for as many candidates as they wanted to, and still have a simpler election process than if they used this "IRV3" system (called "Ranked Choice Voting" here in San Francisco, and I think some other places where it's been adopted).

    The primary needs to pick three candidates so pick which three you'd like to see in the general election the most. Or don't vote...

    There is no reason that the number of candidates the voters get to rank should correspond to the number that the system needs to pick. That makes absolutely no sense.



    And I'm saying that among highly educated and well informed folks like you, true scoring of the candidates bears a decent semblance to reality, but for most folks that just is not true.

    I've done a Score Voting exit poll on a bunch of people in Beaumont, Texas, and they seemed to have absolutely no problem scoring the 5 candidates in the 2006 gubernatorial election.
    http://scorevoting.net/Beaumont.html

    And like I already said, Warren Smith's Bayesian regret calculations accounted for this problem by using ignorance factors, and Score Voting still came out on top. Logically, that makes perfect sense, since voter ignorance is also a handicap for ranked systems. Just as ignorance might turn your SV ballot from {X=10, Y=7, Z=0} to {Y=10, X=5, Z=0}, it can turn your ranked ballot from X>Y>Z to Y>X>Z.

    And another major factor that leads to distortion of sincere preferences is tactical voting. Score Voting is less affected by tactical behavior than IRV, largely because it passes the Favorite Betrayal Criterion.

    And another major factor of voting method performance is tabulation efficiency. IRV discards a great deal of ballot information, and has extremely poor tabulation efficiency.

    These factors have a much larger effect on voting method quality than does voter ignorance, which may skew ballots away from their otherwise accurate "semblance of reality".

    I explained this in my previous post, by the way.

    ReplyDelete
  18. I'm all for evolving beyond IRV3 for San Francisco. But maybe it'd be better to push for using AV or SV in a primary first while keeping IRV for the final election?

    That would be a really irrational thing to do, since SV and AV are so much vastly better than IRV.

    It is mathematically proven that the incentive to strategically vote does exist for some voters. This says nothing about whether they actually do strategically vote or how perverse such would end up becoming.

    The "perverseness" of election outcomes as a result of strategic behavior is measured via Bayesian regret.
    http://scorevoting.net/StratHonMix.html

    You are correct that mathematical theorems can't tell us whether voters will actually be tactical. But consider that IRV behaves worse with 100% sincere voters than Score Voting and Approval Voting do with 100% tactical voters. In that case, even if we make the worst case assumption, SV/AV are still better than IRV.

    And empirical data can tell us a lot about how often voters actually do use tactics. For instance, ANES exit polling from the 200 US presidential election showed that 90% of voters who claimed to support Nader also said they voted for someone else, most of them for Gore.

    With IRV, you can look at examples from countries like Australia:
    http://scorevoting.net/AusAboveTheLine07.html

    I've actually called up Australian political parties and talked with their activists.
    http://scorevoting.net/AusIRV.html

    Likewise, if the independence of irrelevant alternatives is violated then some may try to game the system.

    You again illustrate your lack of experience with this subject. The problem with violations of consistency-based criteria is NOT that they may be tactically exploited, but that they show with 100% certainty that the wrong candidate is being elected in some situations. For instance, say candidate X wins in Scenario 1, but candidate Y wins in Scenario 2, which is identical to Scenario 1 except that some voters have increased their support for X relative to Y. You then know with 100% certainty that the wrong candidate was elected in Scenario 1 and/or Scenario 2.

    Perhaps, you could explain the algorithm used for AV in calculating the regret functions.

    http://scorevoting.net/BayRegDum.html

    ReplyDelete
  19. dlw1:They are metrics, but that doesn't mean they are concrete. When dealing with a complicated subject, words and numbers all fail us at times and we cannot divorce the subjective from the objective.

    dsh:You used subjective unquantifiable terms like "good" and "poison" and "general viewpoints". I cited Bayesian regret figures. It is as if you said "that city is far away" and I said "that city is 253 kilometers away". The second statement is

    dlw2: Good and poison are quantifiable, they tend to be less forensic and more organic or multifaceted in their observable traits. When I say "good", or "poisoned", I'm referring to the percentage of the US population that votes or I'm referring to turnaround on the sorts of issues that are decisive in elections. These things can be corroborated, without reference to statistics based on conjured-up full sets of rankings among candidates by all voters.

    dlw1: I see the use of winner-take-all elections as causing increasing returns to scale in election-winning. It leads to there being two dominant parties (aka like Duverger's law)

    dsh: Duverger's Law does not say that single-winner (presumably what you mean by "winner-take-all") elections lead to two-party domination. It says that Plurality (aka first-past-the-post) Voting leads to two-party domination, whereas top-two runoff (aka "TTR", or "Plurality with a runoff election if there is no majority winner") tends to escape duopoly. Here are Duverger's words, verbatim:
    http://scorevoting.net/DuvTrans.html

    dlw2: I know, I'm dissenting from Duverger, hence the qualification above. I do not believe that models of voter rationality are realistic and put less stock in the effectiveness of alternatives to first-past-the-post to break the duopoly. I'm also taking the extra step of arguing that the system tilts to monopoly. As such, what matters most is to replace a contested monopoly with a contested duopoly.

    DSH:IRV also maintains duopoly, but it seems probable that Score Voting (and by extension, Approval Voting) would not have this problem, largely because they pass the Favorite Betrayal Criterion.

    DLW2: And my view is that the number of options won't matter, simply because the epistemic frailty problem alluded to before.

    DLW:due to the inherent fuzziness of candidate options, as well as our tendencies to not do our homework before voting, the number and types of options we give voters turn out to be of second order importance for election reform.

    DSH: There is no evidence to support this.

    DLW2: Bull Shit! Consider the results from googling "http://scholar.google.com/scholar?q=%22voter+irrationality%22&hl=en&btnG=Search&as_sdt=100000001">voter irrationality" in Google Scholar. Then, consider the results from googling "http://scholar.google.com/scholar?hl=en&as_sdt=100000001&q=elections+%22information+asymmetries%22" (elections and "information asymetries"). You simply cannot aggregate multiple policy issues into one metric uniformly for an entire population. And there are additional matters of personal character that always are relevant for voting decisions but are not easy to quantify.

    more later.

    I have an idea for common ground between us....

    ReplyDelete
  20. Good and poison are quantifiable, they tend to be less forensic and more organic or multifaceted in their observable traits.

    Yes "good" is quantifiable. The economic name for it is utility. I cited Bayesian regret metrics which actually do quantify it. Whereas you said:

    Strategic voting is not that bad

    What does "that bad" mean? I cited actual numerical expressions of the difference in performance between various voting methods, as a function of strategic voting. You invoked subjective-to-the-point-of-meaningless platitudes.

    You continued,
    It's good to mitigate it, but it doesn't poison outcomes in terms of making the outcome reflect general viewpoints.

    More subjective and totally unquantified platitudes. How good is it to reduce how much tactical voting? How high does the social utility efficiency have to be to count as reflecting "general" viewpoints?

    When I say "good", or "poisoned", I'm referring to the percentage of the US population that votes or I'm referring to turnaround on the sorts of issues that are decisive in elections.

    You didn't refer to any percentages or metrics of any kind whatsoever in that passage.

    These things can be corroborated, without reference to statistics based on conjured-up full sets of rankings among candidates by all voters.

    You cannot "corroborate" social utility efficiency, because you can't read human minds.
    http://scorevoting.net/WhyNoHumans.html

    dlw1: I see the use of winner-take-all elections as causing increasing returns to scale in election-winning. It leads to there being two dominant parties (aka like Duverger's law)

    dsh: Duverger's Law does not say that single-winner (presumably what you mean by "winner-take-all") elections lead to two-party domination. ..top-two runoff tends to escape duopoly...

    dlw2: I know, I'm dissenting from Duverger, hence the qualification above.


    What?! You said that winner-take-all elections lead to there being two dominant parties, and invoked Duverger's Law. But there are winner-take-all voting methods that do not maintain two-party domination, and that's actually part of Duverger's Law. You should have said "Plurality Voting" instead of "winner-take-all". So don't tell me you know. Clearly you did not know.

    ReplyDelete
  21. I do not believe that models of voter rationality are realistic and put less stock in the effectiveness of alternatives to first-past-the-post to break the duopoly.

    I never said anything about models of voter "rationality". Assuming you are referring to Smith's Bayesian regret calculations, I painstakingly explained that they incorporate ignorance factors, leading many voters to be irrational.

    If you claim the BR calculations are "unrealistic", then please show your evidence. The calculations used 720 different parameterizations ("knob settings") of five fundamental parameters. There were elections with 2 candidates, and 3 candidates, and 4 candidates, and so on. There were elections with 100% sincere voters, and ones with 100% tactical voters, and all mixtures in between. There were different types of utility generators, and varying levels of ignorance. Despite all of those differences, Score Voting came out on top in every parameterization. And the relative performance of the different systems was remarkably consistent. That consistency attests that the results were highly realistic, for if they were skewed from reality, the probability that they would all just happen to be skewed in the same direction would be exceedingly small.

    Moreover, if you don't put stock in the Bayesian regret calculations, then you effectively have no way to judge the relative performance of different voting methods. At that point, you can only judge them by their simplicity (e.g. Kolmogorov complexity of tabulation).

    I'm also taking the extra step of arguing that the system tilts to monopoly. As such, what matters most is to replace a contested monopoly with a contested duopoly.

    Our Congress is roughly 50/50 Democrats and Republicans. Australia's House of Representatives is also a duopoly (Labor and Coalition). I don't know what difference is made here with regard to monopoly. Actually, Ireland has used IRV since 1938 to elect their President, and in all but one bizarre fluke, the winner has been from the Fianna Fail party, so the biggest monopoly I know of is from an IRV election.


    DSH:IRV also maintains duopoly, but it seems probable that Score Voting (and by extension, Approval Voting) would not have this problem, largely because they pass the Favorite Betrayal Criterion.

    DLW2: And my view is that the number of options won't matter, simply because the epistemic frailty problem alluded to before.


    Of course it matters. It makes all the difference in the world.

    The reason Plurality Voting results in duopoly, is that voters don't want to throw their one vote away, so they usually support their favorite candidate between just the two candidates they think are electable. Being electable is mostly a matter of winning the nomination of one of the two major parties.

    But if people can vote for as many candidates as they want to, then a candidate who is preferred to both of the major party candidates can still win.

    This is one of the most obvious consequences of using a voting method that passes the Favorite Betrayal Criterion.

    ReplyDelete

  22. DLW2: Bull Shit! Consider the results from googling "http://scholar.google.com/scholar?q=%22voter+irrationality%22&hl=en&btnG=Search&as_sdt=100000001">voter irrationality" in Google Scholar. Then, consider the results from googling "http://scholar.google.com/scholar?hl=en&as_sdt=100000001&q=elections+%22information+asymmetries%22" (elections and "information asymetries").


    I am in no way disagreeing with you that there is rampant voter ignorance and irrational decision-making. The point is that this ignorance does not strongly affect the relative performance of the various voting methods, since it hurts all of them. It would help if your responses address my actual argument and/or evinced that you understood it.

    You simply cannot aggregate multiple policy issues into one metric uniformly for an entire population.

    Of course you can. You just sum up the voters' utilities for each candidate for each issue. Every time you pick a favorite candidate you are summing up your individual utilities for that candidate on each issue. Maybe you are in agreement with him on economics, foreign occupation, and drug policy, but you disagree with his stance on gay marriage. Choosing your favorite entails summing up the utility for each of these things, which is of course affected by the relative importance of the issue, and the distance by which you diverge from that candidate's way of thinking.

    And there are additional matters of personal character that always are relevant for voting decisions but are not easy to quantify.

    If something is hard to quantify, that just means there's a greater expected deviation between your assessment of your utility as a result of that factor, and your actual utility as a result of that factor. Bayesian regret calculations account for that, with ignorance factors.

    ReplyDelete
  23. 1st off, I have to apologize to broken ladder for not being more careful and distinguishing him from Dale.

    2nd off, I can't reply right now till probably Sunday, because I'm "busy".

    3rd off, I did write up the simple yet subversive algebra of AmPR at TPM.

    I hope you can help draw some attention to it, as has recently been done at Poli-Tea.

    yours,
    dlw

    ReplyDelete
  24. I delineated my thoughts on this exchange at length here.

    dlw

    ReplyDelete