While I don’t necessarily think the Condorcet criterion is the be-all, end-all of elections (you aren’t comparing candidates in isolated pairs), if a system violates the majority criterion it’s undemocratic. If 40% of voters prefer Trump, and give him an average score of 5/5, and 60% of voters prefer Biden, and give him an average score of 3/5, then Trump wins. That’s even worse than the current system.
Condorcet winner should be thrown out of the window. But looking at the average of grades is a stupid idea too. We should look for a person that is described by everyone as "above average" and avoid candidates that are hated by some and loved by some. It's like choosing a film for a film party. You may love your arthouse with all your heart but if it triggers Epilepsia - put it in the no-no shelf. Even if it's just one person with Epilepsia there.
@@rucksplash12 I disagree that it’s undemocratic. In the example given, it actually seems as though MORE people are happy with the result. The majority’s first pick may not have been chosen, but their second pick was a very close preference for them, and for the entire group as a whole was most preferred. Wouldn’t it be more democratic for the chosen candidate to be one that is liked the most overall by all people, rather than one that may be the first pick of the largest group, but isn’t the most liked by everyone?
Because it would be bad policy. If 40% of voters prefer Trump, and give him an average score of 5/5, and 60% of voters prefer Biden, and give him an average score of 3/5, then Trump wins. That’s even worse than the current system.
Your biggest mistake is assuming the candidates would be the same. Biden is the candidate because he has the best chance of winning under the current system. Second, why are you assuming people with give the candidate they prefer to Trump 3/5? You’re also ignoring the fact that people who don’t like Trump are going to give him a low rating.
@@ganondorfchampin I'm assuming that people will vote sincerely. There is absolutely no advantage to replacing first past the post with score voting when in order to get your top candidate elected you have to give the max score to your first choice and the lowest score to any other choice. It has devolved to FPTP, but a more complicated one where someone honestly voting and indicate that they tolerate Biden but don't love him can end up with the person that they hate winning. Biden and Trump are just interchangeable names to show that the system is undemocratic. They could be any two people. It's possible there are different nominees with score than FPTP. But score is still undemocratic. We shouldn't replace the electoral college with a different system where someone with fewer supporters can beat someone with more supporters; we should change it to a system where that can no longer happen.
The idea is that there would be way more than two candidates, and again, you’ve been completely ignoring that fact that people who like Trump will be balanced by people who absolutely hate him.
@@ganondorfchampin there are more than two candidates now. It doesn't matter what you think might happen. It's what the system allows to happen. A system which would allow a group of 40% to defeat the will of 60% is undemocratic and we shouldn't adopt it.
STAR voting (Score Then Automatic Run-off) solves the condorcet candidate. it works the exact same way, but then the automatic run-off takes the top two candidates, then regroups everyone's preferences between the two. so, the candidate that is popular and unpopular would lose to the popular candidate with the broadest support. this encourages candidates who prefer coalition building.
The Nash equilibrium of Range Voting tends to approval voting. This is why the Condorset winner comes up. Approval Voting only fails to get the Condorset winner when the compromise has a lower Bayesian Regret
Underrated comment. At 5:10, the pro-Charmander would not rationally give 4 stars to Squirtle because they know they would lose their preferred candidate. Knowing this, the pro-Squirtle would do the opposite. Approval voting under perfect strategy and knowledge gets both the Condorset criterion, and all the Arrow paradox properties right, except maybe the independence of irrelevant alternatives.
@@jibraan123100 If they have perfect knowledge, they know that giving Squirtle a 4 will not elect Charmander, while giving Squirtle a 0 will. So they will be incentivized to give Squirtle a 0.
@@darbyl3872 People are strategic. If they get a choice between electing a candidate they like for 4 stars, or one they like for 5 stars, they will choose the 5 star candidate and vote accordingly - voting the 4 star candidate lower than they like him.
@@darbyl3872 Strategic voting gives them _more_ power for the candidate they want to vote to the top, compared to any candidate they like less. While star voting can let you give 5 stars to everyone up to and including the candidate in your personal preference list that you want to push, you are still encouraged to give everyone below that pushed candidate lower ratings than if you were voting honestly.
I link people to this video quite often when they tell me to vote third party. Basically my response is "Don't lose focus after the election. Things need to be changed, so pay attention outside of the media and keep working for a system that makes third parties viable. This is one."
I saw this after I saw the rank based voting system on represent. Us. Honestly I prefer this one slightly more. I love your style, great video! I honestly think majority of the group happy is better because people constantly complain about picking the lesser of two evil.
Condorcet is suppose to be how the multicandidate election winner compares to head to head elections. The head to head winners should also be selected by range voting. There is no reason to fixate on ordinal positions. In your example the majority's second choice won because the majority didn't care about the difference between the two candidates (not because the candidate was polarizing). If you care strongly between two candidates, score them differently. It's as simple as that.
its not that simple actually. if your true opinion is charmander 5 starts and squirtle 4 starts, and you are able to predict that one of these 2 candidates will win, you should vote charmander 5 stars and squirtle 1 star. It is often quite easy to predict the top candidates.
Voters shouldn’t have to vote strategically in order to get their honest preference elected. In score voting, If 40% of voters prefer Trump, and give him an average score of 5/5, and 60% of voters prefer Biden, and give him an average score of 3/5, then Trump wins. That’s even worse than the current system. If a voter’s honest preference doesn’t deliver the result they want, the system is flawed. If 60% of voters preferring a candidate doesn’t get that candidate elected, the system is flawed.
@@rucksplash12 I disagree. If 60% of people barely prefer candidate A and the other 40% would flee the country if candidate A won, but candidate B would be so awesome that 100% of everyone would be so happy that they'd personally make a statue to commemorate candidate B in their space and want to sing and dance as much as possible. Then candidate B sounds pretty good. Candidate A sounds so bad that you are basically are deciding whether to keep having a country at all. Here is an example that might be different enough to challenge your bias about how elections are "supposed" to work. What if your vote wasn't 100% secret. In fact, what if the government set aside US$100,000.00 to give to every voter and the range you assigned to the winner would proportionally decrease your tax refund. So if you gave the winner a maximum vote, you get $0.00 out of that $100,000.00 (but you are still happy since that's truly how much you loved your candidate) and if you gave the winner half the maximum vote you get only $50,000 (but you are still happy since that's how much you loved the winner). In this example, the range vote winner is the one that saves the government the most money right out of the gate. But the election resulted in every single person being super happy since they were all compensated for exactly how much they preferred the candidate that won. A country with everyone happy about the result. That is what range voting is designed for. Other reasons to have other voting systems is that we don't have to have just one winner. Maybe only one person can be commander in chief at a time, but there is no reason that multiple candidates couldn't share at least some powers. God didn't come down and say the point of an election is to pick a condorcet winner that merely has the property that if a majority of people don't disprefer one candidate by even the tiniest amount (i.e. one person might prefer the candidate by the tiniest of tiny tiny tiny amounts and everyone else in the majority might find them equally good) that that one has to be the sole winner of everything no matter how much the minority hates them. It seems like you have a bias towards systems that can result in most of the country outraged (since not 100% of everyone will be able to vote) while a near majority didn't really care much. And the fact that some people will be deeply busy in the hospital, or deeply sequestered in a jury or deep undercover or such and so won't be able to vote that least a couple votes won't change the total happiness of the country much if they get the wrong answer. Because range voting is about maximizing happiness of the country. That's why I used the money example. You aren't going to vote away more money for a candidate unless you really value that candidate more.
@@TimaeusBouma but elections aren't about getting money. An election system which allows the passion of 40% to override even a weak preference by 60% is unacceptable. I have a bias toward systems which are democratic. People have a right to be outraged. But a system which allows a minority to overturn a majority is going to leave a lot more people outraged. In the example I gave, 60% of the population gave the winner the lowest popular score. A system which allows such a winner is not a system which recognizes the will of the people. You can't base a system on how it would work in an ideal made up scenario where the majority have a weak preference for their chosen candidate but are OK with the alternative. For starters, even in that case, it's still not their desired outcome. But secondly, it's not representative of the system's overall performance. A system which would deliver an undemocratic result even in one particular scenario is unacceptable.
@@rucksplash12 Is there a reason you didn't state that the 60% gave the lowest possible score? To me, if you take a system designed to express preferences for more than two candidates and people use it badly that just seems like it is bad on them. Because, honestly. What it you quantified it is as hate minimization instead of happiness maximizing. Then if 60% of people voted 3/5 for one candidate and minimum for another canidate than you see the amount of hate from that 60% based on who wins (60% maximum hate of the other candidate and (2/5)*60% hate if their allegedly "preferred" candidate wins). If there is truly less hate generated by the candidate that generates 0% hate in 40% of the population and so the overall hate generated is only 60% then that sounds terrible ... sounds like maybe a third candidate should have run. But if that candidate generated the least hate. Sounds better than the condorcet candidate. The revolution was bot fought based on democracy. It was based on enough people hating rule by England enough that they were willing to to fight and die. It was based on the collective amount of dissastisfaftion. And your example shows more aggregate dissatisfaction for the candidate with 40% with zero hate and 60% with maximum hate. Since the only other person willing to run had 64% of the maximum possible hatred (your numbers dude) then the person with only 60% of the maximum hatred wins. Makes sense. Find a less hated candidate if you want a winner. The video was pretty clear that things degenerate into voting against people if you use ordinal voting without taking into account how *much* people hate a candiate. Basically you completely ignored how much people hated that candidate that hat 64% of the maximal hate. You could probably have run a ficus and gotten better results. Saying 3/5 for candidate B and 0/5 for candidate A means that that is exactly how different you consider them to be. The money just forced people to be honest. But why be honest if people like you are going to ignore their feelings anyway
Ranked choice voting was my favorite voting system until I saw this video. I wish more people knew about alternative voting systems beside plurality. I think ranked choice voting could be a stepping stone to range voting.
This is what happens when a trusted education channel only thinks the topic halfway through. As hinted towards in the example at 5:00, this voting system still enocurages strategic voting. And with strategic voting, you actually lose independence of irrelevant alternatives, because even a least favourite candidate can change how you vote when voting strategically. Ultimately, voting in this system becomes an advanced strategic task, disfavouring not only less educated people, but also allowing polls and media to manipulate strategic votes by painting the wrong picture of who is at risk of winning
@@darbyl3872 This video is incomplete, as it does not address the strategic voting that would result from a rating-based election. If you have really studied this topic for years, I hope you can fill out the blanks for anyone who asks (including me).
@@darbyl3872 It is true that strategy goes into every part of voting. But that is only what makes it so important to look at HOW strategic voting impacts each particular system. With "STAR", you are left with a system that does not have the advantages over ranked choice that the video outlines. This is a big part of the picture, and people should at least be made aware of those drawbacks. I've outlined examples for how I can know that "STAR" does not have independence of irrelevant candidates, but the mathematical proofs on this topic surely go further than my own casual thoughts. I have read comments in here saying that the "Nash equilibrium" (that means the combination of optimal strategies for all voters) for STAR voting is Approval voting. If that is true, the main difference between STAR voting and Approval voting is that people with lower education will be voting in a less optimal way, giving them less voting power than if the voting system was Approval voting, rather than STAR voting. And if that is the case: Why do you prefer STAR voting over Approval voting if you have spent years on this topic?
@@iwersonsch5131 Strategical voting is usually formulated and thought of in terms of ranked preferences, as if every voter has infinitely precise preferences (they can make perfect comparisons), they have perfect information and certainty about the results of the election, and would always want to maximally impose themselves regarding their preferences, no matter how minor they are. *NONE* of these assumptions are even remotely true in real life, otherwise society itself wouldn't exist. A better analysis of strategical voting is based on Decision Theory. It recognizes that people have fuzzy beliefs about the world and themselves, which are constantly updated as new information arrives (which can be anything from reading a policy proposal, reading a book about economics or the environment, reading an opinion poll, talking to friends and family, watching debates, etc.). Still, they always have imperfect information and fuzzy preferences, with degrees of indifference. They will also only try to impose themselves (what strategical voting is *actually about*) IF they care about the results strongly enough, in which case it is a legitimate form of expression, not an "undesirable" feature. An election is a choice under risk and uncertainty. Thus, we WANT voters to adjust their ballots as more information arrives so they can make their tradeoffs, as accurately as possible. This is not a bad thing. This is a voter being informative and thoughtful. We do not want voters to make uninformed selfish decisions, and all information they get about candidates, reading about policy and its effects on society and others, or following pre-election polls, are identically valuable for the voter to consider what is a priority and what is not, and for them to genuinely express their beliefs. If voters existed in a vacuum, casting ballots as if they were the only people in the world and had no concern for any external information, then the election would not be a *collective choice*, and ballots would, rigorously speaking, carry no information that could be aggregated across voters. it is the shared context which gives meaning and information content to the ballots, which is another reason we want voters to actively adjust to external information. The same concerns also exists under ranked methods, but in a ranked method there is no adjustment under risk other than reversing preferences. In this sense, rankings are *less* responsive and representative of voters' genuine beliefs and behaviors than cardinal voting, and when adjustments do occur they are rank reversals, which completely distort the actual information we want! Cardinal voting only seems to pose a problem in terms of "strategy" if a minority side really cares about making a distinction between two options, and the majority does not, and so the minority wins. But this is the correct outcome! First, the majority can still always enforce their will if they want to. Methods that are deemed "resistant to strategy" (like ranked methods) force the majority to always impose themselves over the minority, even if they don't feel strongly about doing so. The only difference in cardinal voting is that this imposition is made OPTIONAL. That's it. This is, at its core, the only thing people complain with respect to strategy. But it is not up to us to decide when the majority should impose themselves. If the majority, given information about the election and the likelihood of outcomes, deliberately and rationally chooses to not impose themselves, then that is a clear sign that the minority overriding their preferences is something acceptable to do, according to the majority itself. So the "failure" of the method is, by assumption, not a failure at all, and was deemed a legitimate outcome by those who had a say in it. With respect to score voting encouraging people to vote min/max scores only, that is only true under the 100% certainty, zero information, perfect preference assumptions. In realistic scenarios under uncertainty and imperfect information, it can be shown, mathematically, that the strategical equilibrium requires voters to cast more diverse ballots.
@@1ucasvb Humans don't have to have perfect information to gravitate towards a Nash equilibrium. If they see something doesn't work, they do something that works a bit better, and that way they gradually get closer to the optimal strategy
So would we be able to do away with both the Electoral College AND the Presidential Primary Elections?? That would be awesome. The primaries are a ridiculous mess.. every state so different and VERY restrictive. Closed primaries, open primaries, caucuses. Insane! We should just be able to vote for who we want!!
Correct: Score voting doesn't require primaries. Primaries exist because of our voting system "splitting the vote" between similar candidates. Score voting doesn't do that, since you can give similar candidates the same score.
This wouldn't do away with primaries... you'd still have two major parties, which are private entities, deciding which candidate to put forward for the general election. Getting them to agree to change our voting system on such a fundamental level will never happen, regardless of how much it would improve things. The realistic alternative we can work towards *right now* is the *_National Popular Vote Interstate Compact_* which has been covered on this channel before... *_EVERYONE_* should contact their state legislature and demand they pass the NPVIC so we can, once and for all, get rid of the insanity that is the Electoral College and go to a popular vote winner *_like every other country on the planet._*
"So would we be able to do away with both the Electoral College AND the Presidential Primary Elections?? That would be awesome." Argentina has both the primaries and elections at the SAME time under the SAME system. How it works. Round 1: Every possible candidate from every party run and people vote at any candidate. If an candidate of an party X get a vote, this vote goes to the party. If the party has 1.5% of all votes or more, the party is the winner, and the winner of the party is the guy with more votes. Round 2: People pick an candidate between the remaining ones at second round. If someone gets 45%+ of the votes he is the winner (dont ask me why 45%+ and not 50%+ and dont ask me what happens if two candidates have 45%+ of the votes). If someone gets 40%+ but less than 45% and has more than 10% than the second place, he is the winner, if none of the previous two happens, there is an third round. Round 3: At third round the top 2 of previous round to election and the most voted at this round become the winner. One problem with this method is that usa allow people to become candidates without being at a party and it would be a problem at this system.
I do not think that the electoral college should be done away with -- it seems to be doing its job of protecting rural communities from the whims of large urban centers. However, it is possible to adopt range voting within the current electoral system: the local elections produce average scores for the candidates, and the electors then use those average scores as their scores in the electoral college vote. The result would then be an electoral system that combines the merits of both systems.
@@k98killer that’s neither its job nor what it’s doing. Right now only 8 states get any real say in the electoral college; everyone else’s vote is taken for granted. Rural voters shouldn’t get more votes than urban voters. Everyone’s vote should have the same value. Trump shouldn’t be able to win an election with 3 million fewer votes just because more of his voters lived in certain areas. That’s fundamentally undemocratic.
This is a nice video, but range voting does NOT violate Arrow's Theorem. The reason is that you have to assume people will adjust their ballots to fill the dynamic range, so when an irrelevant candidate is removed scores can be adjusted and this can change the winner.
its weird that these alternative vote organizations have popped up in US deliberately ignoring scored voting and campaigning inferior systems instead. (democratically). What gives ? Its great that they are stirring up awareness, but why do some organizations insist on being shy of what is optimal ? Why is the best method getting the last glance ?
They don’t deliberately ignore them. They’ve done the analysis and concluded that score is not good policy. In score voting, if 40% of voters prefer Trump, and give him an average score of 5/5, and 60% of voters prefer Biden, and give him an average score of 3/5, then Trump wins. That’s even worse than the current system.
@@rucksplash12 Why is that a bad thing? If average happiness in the country is higher, I think it's a better outcome. I think it's easier to think about this without the political connotations of using real candidates (since it makes a system where your preferred candidate loses seem bad). That being said, I also don't believe we would have ended up with Trump vs Biden if we used range voting. I don't think a candidate that averages below 3/5 would be at all viable in the system, and we'd get stronger choices from both sides of the aisle.
@@darbyl3872 there are lots of voters who don't fully understand electoral systems and who would give an "honest" rating rather than an effective one. For instance a Bernie fan might give Biden 3/5 and Trump 1/5 without realizing they're essentially just giving 60% of a vote to Biden and 20% of a vote to Trump. Ranks may not be the ideal system, but they don't force voters to bullet vote the highest score for one candidate or else lower their effectiveness.
Range voting is clearly the best. Even in the last part of the video where you present a possible flaw it actually isn't a flaw ... If 60% is like 'I prefer Red (5)but I don't really mind Blue(4)' and 40% is like 'I like Blue (4) and hate Red (1)' then the fair and just outcome is that Blue wins. If Blue was really that bad the 60% wouldn't have given him 4. I believe this system is the best or one of the best out there.
I love range (or as i learned about it scored voting) just for the fact that it doesn't require you to choose 1 candidate over another you can support the mainstream party as well as supporting the smaller party you like the most equally.
@@barnacles1352 No, ranked is significantly worse, and in some cases no better than FPTP voting. rangevoting.org/rangeVirv.html rangevoting.org/CompChart.html
Thanks for running through that explanation, I have been looking for a voting system with two qualities: a candidate that loses every head to head matchup must not win, and 'controversial charmanders' like the one in the example, even if they are the favorite of 51%, must not automatically win, squirtles have to have a shot.
Great video! I very much liked your explanation (4:37) addressing the majority criterion criticism with regard to range voting. Given that the choice between voting systems is essentially a tradeoff between different drawbacks, I would much prefer the one whose flaw biases in favor of a consensus candidate over a divisive one, rather than a system such as instant runoff, in which voting sincerely can potentially be self-defeating. Your explanation satisfied my concerns with that critique in a way that other discussions on the topic had so far failed to do. Thanks for the clarity!
Nice video! I prefer STAR Voting though It's a 2 round system with only one ballot required to be filled out. To fill out the ballot: rate your candidates 0-5 stars 1st round: the two candidates with the highest total stars are the finalists 2nd round: the finalist most preferred wins (most preferred means higher stars than other finalist on the majority of ballots) Each ballot counts as one vote only (if a ballot has the two finalists tied in stars given to them, that vote is counted as a vote of no preference)
There actually is a system, has a bit of both, although it a bit more involved. Its called something like: "single peaked preferences" and it uses the ordered list to compute the median ballet, the favourite on that ballet is automatically the condorcet winner.
Great video other than presenting average score as the logical option rather than total score which would eliminate the risk of low visibility fringe candidates winning.
Average is better for presentation anyway. But if blanks default to zero, there's no risk at all. A better approach, if you include blanks, is that when you compute an average score you have a minimum denominator of a simple majority of total number of voters. Say 1000 voters participated, so you have a simple majority of 501. If a candidate gets 900 blanks and 100 10's, they would get an average score of 100 x 10 / 501= 1.99, instead of 100 x 10/ 100 = 10. Any candidate which received more than 501 non-blank scores would get the average computed normally. This means candidates that are unknown to a majority get penalized here, and candidates known by a majority get the rating they deserve. It is reasonable because if a majority of people know a candidate and think it is on average a score of X, it is reasonable to assume those people giving blank scores would also average it as an X if they were to give a score.
what about proportinal range voting for parlmantary elections? the idea is range voting for parties and then simply award seats proportinal to the party's score. it sounds simple but strangely i don't find many ressources on it..
There is no need for that because in a proportional system everyone votes for his personal preference and the party then gets seats to represent this number of voters. There are no spoilers or anything like that. Moreover, a majority of voters should not be able to surpress a minority by rating a small party badly, because the point of a parliament is to display everyones opinions.
Great video! I have a question about the Condorcet winner example. Charmander would have won had each ballot been normalized so that the lowest ranking given to a candidate was the minimum, and that the highest ranking given to a candidate was the maximum. So, using 1 as the minimum score, instead of 60% giving Charmander 5/5 and Squirtle 4/5, and 40% giving Charmander 1/5 and Squirtle 4/5 It would be 60% giving Charmander 5/5 and Squirtle 1/5, and 40% giving Charmander 1/5 and Squirtle 5/5 Is there any reason someone may not utilize the full range of ratings allowed on their ballot unless they thought all candidates are completely equal? Is it still possible that the Condorcet winner could still lose after applying this transformation?
There is a theorem that, if the supporters of the would-be Condorcet winner are sufficiently strategic (including only giving min and max scores) they will win with Range/Score and Approval Voting. In a two-candidate election, normalizing ballots would have the effect of giving one candidate the minimum score and the other the maximum (unless they are tied, in which case the ballot cannot change the winner) which basically forces everyone to be strategic. With three or more candidate, normalization alone might not do this. For example, what if Bulbasaur also ran, and this was the outcome: 60% Charmander 5, Squirtle 4, Bulbasaur 1 40% Charmander 1, Squirtle 4, Bulbasaur 5 Averages: Charmander: 3.4 Squirtle: 4.0 Bulbasaur: 2.6 This doesn't make Bulbasaur the winner, in fact he's dead last, but now how would you normalize the ballots? I definitely think the traditional majority and Condorcet criterions are naive, and not inherently desirable (though they were good heuristics when we didn't have anything better). In these examples, paradoxically, a supposed minority candidate had *broader* appeal than the so-called *majority* candidate. If anything, this sounds like a "truer" majority to me. As for why anyone would WANT to do only use a portion of the range? For two-candidate races, it probably doesn't make much sense. But some voters may want to do it for the sake of honesty, or they might not be confident, or are willing to cooperate and deffer to others.
If you dislike all candidates equally, it won't effect the results if you score them all at minimum score or maximum score. You're scoring candidates relative to each other, not some arbitrary universal standard of quality (such as how Amazon reviews are scored)
No, it won't effect the ranking, but it will effect the final scores, and I might want exactly that; no one should have the right to change my vote if they disagree with me on that. Secondly, I don't have to dislike them all equally, I can dislike them _almost_ equally (depending on how fine a scale one uses here). E.g. if my votes are 0/10 and 1/10, then I bloody well meant to give 1/10. What gives anyone the right to change my 1/10 into a 10/10? If that's what I wanted, that's how I would have voted.
You would be within your right to do that, but I'd still like to note two things: 1. The score given to a candidate only makes sense within the context of that particular election. If in one election, candidate A wins with a score of 6.5/10, and in another election, candidate B wins with a score of 7.1/10, that doesn't mean that people generally liked candidate B more than candidate A overall. Similarly, a candidate that won with 7.1 in a hypothetical election might have lost with 7.5 in the same election had different/additional candidates been running. 2. If you like the 1/10 candidate better than the 0/10 candidate, why wouldn't you want the 1/10 candidate to have a higher chance of winning? 10/10 doesn't mean "Ideal candidate who I agree with on everything" nor does 0/10 mean "worst most evil candidate in the history of the world". 10/10 means "favorite of the options provided", 0/10 means "least favorite of the options provided"
I don't think the analysis would be any different. If blanks are kept abstentions, then they would function exactly the same. If you don't use abstentions but blanks are treated as the minimum score, they'd be the same in that case as well. If blanks are 0s, then they'd be a bit different.
I think this ignores the fact that people are likely to get pretty extreme on rankings. (if you look at patterns of ratings on a lot of real data, you get huge spikes of 1- and 5-star rankings, with much fewer 2-4.)
If you're talking about ratings outside of elections, part of that may be because people only bother giving any score at all if they already had strong feelings. Anyways, if people only use the minimum and maximums, then it's equivalent to Approval Voting, which is still great. It is true that this style of voting is always strategically optimum---in theory. In practice, it can require a lot of confidence that your information is extremely accurate. Otherwise, it can be safer to give at least a few partial scores.
Not really, and not really an issue. This isn't like rating products, it is fundamentally different. When people rate products, they don't buy and test every variant available before giving out ratings. They buy and rate just one. That's what inflates the rating scales. With voting, it is different. You have an exhaustive comparative scale, because everyone is giving a rating on all elements of the same set of options. This is what makes the ratings comparable and what incentivizes differences. People hate giving the same amount of support to candidates who don't deserve it. That's one of the main criticisms approval voting receives. People are also observed to use the comparative rating scale to differentiate candidates they will accept, and they collapse everyone who is unacceptable to the minimum rating. This naturally makes the system slightly less susceptible to naive exaggeration of ratings.
In a 2 party system, yeah. Here in Germany we have 6 significant parties in parliament. I'd dive one of them 5 stars, two of them 4 stars and the rest 1 star. It's definitely scaled to give my favorite 5 stars and my least favorite 1 to make the most out of my vote. But It also makes it very clear that I do not only approve of that party.
doesn't matter. if you actually think this is a good argument, why would you then prefer plurality voting or approval voting where you are forced to give every party either max or min score?
to fix the condorcet winner, look at the top two highest ranked candidates, and see who is most preferred over the other, and make them the winner (even if their ranking is lower)
A two step voting system first the well known system and the second page would have a range voting. Than you could decide weather you want to have a Condorcet winner or not. Did i get something wrong?
Range Voting is not a condorcet winner, because the winner of the pairwise battles at condorcet winner is selected according to first past the post (or some voting system that with 2 candidates become first past the post). And this is a HUGE problem, because if first past the post was considered better than range we would be using it. So we use an voting system to decide the winner when more than 2 candidates are running for election, but when we want to decide the winner between just 2 we use something else for some reason. Maybe we should create something called Self-Referential Condorcet Winner, where it would test that if someone that won all pairwise election using the voting system we are testing would be the winner of the election. Also with even 2 candidates, range voting is better, two examples. 3 childrens pick an animal to be their pet. The choices is cat and spider. Kid 1 and 2 rate spider with 10 and cat with 8, kid 3 rates cat with 8 and he is agoraphobic and rate spider with 1. Here spider get 7 and cat get an 8 and the cat win, and under first past the post, the spider win. Another example: 2 childrens pick an animal to be their pet, between cat and dog. Kid 1 say 9 to cat and 3 to dog, and kid 2, give 9 to both cat and dog. Cat get 9 and dog get a 6. Under first past the post, kid 2 would be forced to pick someone between cat and dog that he think is 100% equal, so there is 50% chance of a tie that shouldnt happen and will only happen because he can't pick both cat and dog.
Another good, informative video, Undefined Behavior. Something I haven't heard score voting proponents tackle is how do you prevent the system from becoming an approval vote? Anything less than a top of bottom score is effectively tallied as a fractional vote. In your five star system, why would any voter give anything other then 0 or 5 starts to each candidate?
Thanks for the question! This is a really good one. If voters all vote strategically, you are absolutely correct that range voting would devolve into approval voting. According to rangevoting.org (rangevoting.org/rangeVapp.html, unfortunately some of their links appear broken right now), experimentally we see that most people do not vote strategically. There's seems to be a psychological effect that leads people to want to be more truthful with range voting, since the perceived payoff of exaggerating appears small. Of course this isn't 100% proof that this is how things would pan out in the real world, but it's certainly interesting.
> It is a fallacy to asses voting methods based on subjective/arbitrary criteria like this. Okay > The voting systems analog for this is Bayesian Regret. But Bayseian Regret is just another arbitrary criteria. On a 0 to 10 scale, I don't agree that five people getting a 6 rated candidate balances out someone stuck with a 0. I'm not even sure a single-scale utility score for all candidates is possible within a single voter. > your naive intuitions about "fairness" turn out to be all wrong I'm sure it seems that way when measuring with your yardstick but you can't just dismiss mine and claim yours as truth. We need to consider the criteria itself. How about we move forward one of two ways? Either you try to convince me that a score voting system doesn't give honest voters fractional votes or I will try to convince you that the Bayesian Regret system is a flawed analysis of a voting system. Thoughts? > This is a non-issue. A great deal of empirical data tells us that a significant number of voters will be honest... Unless you can show me that dishonest voters will never swing an score vote it is very much an issue. You lean on scorevoting.net for your sources. I appreciate the site but I think many of the pages do not speak to my point or have assumptions I do not agree with. For example, my concern isn't that people won't vote honestly but that people who do will not get as much say in the election outcome. It also suggests that people who don't register a party affiliation also won't score vote strategically which I think is specious. Thanks for your thoughts. I am always interested in discussing these issues with interested parties.
>> I don't agree that five people getting a 6 rated candidate balances out someone stuck with a 0 > It is mathematically proven that you are wrong. Great. You can prove a utility function is true if you assume a utility function. > I'm sorry to break it to you, but you're headed down a common naive line of reasoning that has been rigorously explored and debunked. Yet all you can do is toss strawmen my way and insult me. We're done here.
The biggest issue of this system is, that difference it is not clear what 1 start means. 3 is I do not care or I will be ok with this candidate? 4 means I will be really satisfy or a little bit. Also when we should vote for more parties, it is hard to express differencess between them in this scale (but it is much easiler then in voting ranking system).
There are ways to penalize unknown candidates. He even went over one in this video. By starting all candidates with, say, 1000 scores of '0' to start with, if someone gets nothing but a single max score (say, 5) instead of having an average of (total score divided by non-abstentions) = 5/1 = 5 (the max), it'd be 5/1001 ~ 0.005 --- hopelessly low. There are other ways too, like counting abstentions as scores who's value depends on the raw average and the portion of non-abstentions (opinions). abstention_value = raw_average * opinion% Abstentions are always worth less than the raw average, so they can never help. But the fewer there are, the less the difference is, and the less they hurt. This way you don't need to pick an arbitrary amount of 0s to give to everyone, which they would presumably have to agree on (and the most powerful ones could try to give themselves an advantage).
I'm confused. You've outlined IIA in the way that is irrelevant to Arrow's theorem. IIA for Arrow's theorem says that the choice between A and B ought to depend solely on the pairwise preferences between A and B in the profile, not on where other alternatives C,D, etc. fall relative to them. And you seem to be assuming that Arrow's framework requires voters to 'force a ranking' (force a strict preference). But it doesn't, hence his 'weak preference relation' which incorporates both strict preference and indifference.
This system is a type of approval voting. While better than First-past-the-post (what most of the USA uses), it's susceptible to *strategic voting,* where you rank your favourite candidate with 5-stars, and then give everyone else 1-star. If everyone votes like that, Range voting will basically turn into First-past-the-post. Ultimately Ranked voting (instant run-off) is a better method
This is a "bullet voting" argument, and I'm pretty sure it's not quite right. Strategic Range does involve giving only maxes and mins, but not necessarily just one max. Imagine if your "true favorite" is otherwise unpopular even in a Range election, while you have a close 2nd who's a real front-runner. Giving only your favorite any points is once again throwing your vote away, so you should definitely also give some points to your close 2nd. But you can still also give points to your true favorite, so their overall amount of support doesn't appear artificially lower in the final results. IMO Range behaves well with strategy, while I'm not very fond of IRV. It's basically just FPTP but automatically strategic on the voter's behalf using a simple heuristic -- if your current top choice is last place, it goes through another round as if you strategically voted for someone who was more of a front-runner instead, repeat. So it's less that it's "resistant to strategy" and more that it's "nearly maximally strategic by default", but we don't think of it that way because the method does it itself instead of relying on the voter to do it. IMO the only reason it's not even worse is because some methods arguably _benefit_ from some strategy, including FPTP. (Also I'd probably say that Approval is a type of Range/Score voting, rather than the other way around.) (Yes, I posted almost the same thing on another comment just a little before this one, the OPs are pretty similar)
There's one problem with this analysis. It assumes everyone is voting honestly. Even though there's never an incentive to rank your favorite candidate lower, there can be, and typically are, incentives to rate others lower. The real question becomes how altruistic the average person is, which is up for debate. In the best case, range voting maximizes overall pleasure. In the worst case, where everyone acts 100% selfishly, the system is no different from plurality voting (you get only one vote). The one benefit of ranked choice is that it has less issues with incentivizing lying (though it's by no means impervious to the problem). I'm not saying that range voting is bad, I'm just saying it's not the end all be all, even if maximizing overall pleasure, which range theoretically excels at, is your priority.
This is a "bullet voting" argument, and I'm pretty sure it's not quite right. Strategic Range does involve giving only maxes and mins, but not necessarily just one max. Imagine if your "true favorite" is otherwise rather unpopular even in a Range election, while you a close 2nd who's a real frontrunner. Giving only your favorite any points is once again throwing your vote away, so you should definitely also give some points to your close 2nd. But you can still also give points to your true favorite, so their overall amount of support doesn't appear artificially lower in the final results. IMO Range behaves well with strategy. If by ranked choice you mean instant runoff, then I'm torn on how much it incentivizes honesty, or how much we should care. It's basically just FPTP but automatically strategic on the voter's behalf using a simple heuristic -- if your current top choice is last place, it goes through another round as if you strategically voted for someone who was more of a frontrunner instead, repeat. So it's less that it's "resistant to strategy" and more that it's "nearly maximally strategic by default", but we don't think of it that way because the method does it itself instead of relying on the voter to do it. IMO the only reason it's not even worse is because FPTP is one of the few systems which sort of benefits from strategy a bit.
@@MustSeto Imagine everyone is 100% selfish. A has a rating of 4/5 B has a rating of 3.9/5 C has a rating of 3.2/5 If everyone votes honestly, A wins. But there will be plenty of people who would've preferred B, but didn't mind A, so gave, for example, B 4/5 and A 3/5. If some of those voters rank A a bit lower and/or B a bit higher than their true preferences, they can make B win. They're happy, B is happy, but everyone who preferred A is now upset. This is misrepresenting voters' preferences and incentivizes lying, which breaks the whole system. That being said, how much this would effect real world elections is uncertain, and perhaps range voting is actually the best option. I'm simply pointing out a potential downside of the system that this video doesn't mention, which, in my opinion, is a bit disingenuous. I'm not saying range is objectively not a good system.
@@Owen_loves_Butters By all means, point out anything shortcomings you see, or anything you think could be good to know Yes, as mentioned, theoretical maximum strategy does involve giving only maximum and minimum scores. And in some cases that _can_ mean only giving one max, such as possibly in the case you mention here. But this is not _always_ strategic, for every voter in every election, so it is not "no different from plurality voting" even in the worst case, and I'm not sure this qualifies as "breaking the whole system" Also, IIRC being "maximally strategic" like this is only guaranteed to be safe if you have enough information. But if there's enough uncertainty about precisely how other people will vote, then the most strategic, selfish thing to do can involve reducing the risk of greater evils by giving compromises partial scores. Though not necessarily the _exact_ scores you would give if you were 100% honest
@@Owen_loves_Butters The problem with that argument is that it assumes only the voters of one candidate vote strategically. If everyone votes strategically, the voters of A will in turn rank B a bit lower, thus restoring the original order.
Wouldn't non-condorcet winner creates strategic voting? In this case, the Charmander voters (who are in the majority) are encouraged to give Squirtle 1 star. So we back to square one?
Score Then Automatic Runoff (STAR) Voting. You give each candidate a score of 0 to 5. The two of the top scores go an instant runoff round. If you voted higher score than the other, then your vote is counted for that candidate. If your score is equal, then you have abstained. The one with the most wins. This far as I can tell helps with the last problem. It discourages strategic voting. It is most accurate in reflecting votes, -and it does not cause a favorite betrayal- edit: STAR does fail favorite betrayal and later-no-harm to minimize the problems with both and spoiler effect. Explained here: www.starvoting.us/farewell_to_pass_fail
STAR does reintroduce favorite betrayal. If your favorite would lose in the runoff when a compromise would win, then it's in your interest to put your compromise above your favorite. And STAR is basically similar to (forced, evenly distributed) strategic Score. It's mostly only better than Score when there's one-sided strategy (Score's worst-case). So STAR has a better worst-case and worse best-case. One-sided strategy like that seems pretty unlikely to me, and IMO even it's worst-case isn't that bad compared to other methods, so I think there wouldn't be much risk in at least trying plain score first. But it's nice to know we have a good backup in case it really does have problems in practice.
Here are some links to explanations of how and why IRV sucks. wp.me/p23U97-dD Here is a description of a voting system that I propose as having the perceived advantages of IRV and the real advantages of Range. wp.me/p23U97-eZ
@@darbyl3872 Made this comment 3 years ago and have looked much deeper into this and I firmly believe approval voting is the way to go. Cheapest to implement and better results than instant runoff
If you truly want 3rd parties, then what's needed is some form of multi-member districts via proportional representation. Any single-menber system always favours the two-largest parties.
To counter a unknown winner, maybe you have a first round where people give a vote (yes vote) to a max of 9 votes to candidates (can give yes to 0) if wanted. The best 9 most voted goes to the second round, and there a huge amount of time between first and second round to make less known candidates more known. At second round, everyone must give a score between 1 and 10 to each candidate (not giving score to some candidate make your votes not valid/count) The one with best average win. The number 9 come from the fact you want to have a not so huge amount of candidates so everyone can know and also be able to be able to be FORCED to vote at ALL them at second round. But the number is not so low that it would basically turn this into something that is mostly approval system. The number 9 was selected because its the amount of sides at political compass, Left/Authoritarian, Left/Center, Left/Libertarian, Center/Authoritarian, Center/Center, Center/Libertarian, Right/Authoritarian, Right/Center, Right/Libertarian. This method also deal with countries like usa that have an extreme amount of candidates, almost all voting system have problems with alot of candidates, the few ones that "don't" are First Past The Post, Anti-First Past The Post (vote the worse, less voted candidate win), two round system, anti-two round, coombs version of two rounds system. And those systems sucks.
this could also lead to the same result coz if i have several options and i am worried that some other option might get elected and there is an alternativ which i dont like but i prefer than the guy i hate, i would rate more stars to him than i would normaly do, to get him on top!
The largest issue of no condorcet winner is having lukewarm candidates who hardly anyone feels strongly about thus leading to disparity amongst voters.
"Range" used to be the more common name. From the upload time I think this video was made around the time the transition was happening (around the same time I was getting interested, now that I think about it).
How about turning votes into a turnament style dilema, and we vote for every instance of preferance? I can already see some gaps but my smooth brain can't process it properly :)
I like range voting but STAR is a slight improvement that makes it a lot better Let's say Charmander, and Squirtle are the 2 favorites, but I'm a 3rd party supporter (Bulbasaur). If I vote 0-3-5 respectively, then I'm essentially weakening my vote in a Charmander vs. Squirtle race (I'm only putting 3 points toward Squirtle), and if I don't think Bulbasaur has a good chance then I'd be incentivized to give 5 points to Squirtle and minimize Charmander's chances STAR is "Score then Automatic Runoff", the score part is just like range voting, but then once all the averages are calculated, there's a automatic runoff between the top 2 where every vote only counts once. So my 0-3-5 Bulbasaur vote would count as 1 vote for Squirtle during the runoff and I'd have the same voting power as the Charmander and Squirtle voters
I don't think range or approval voting passes IIA. If voters voted on each candidate independently, yeah, it would, but voters take into consideration all the other candidates. Let's say that there are 3 candidates, A, B, and C. You might like C, not like B, and really hate A. If it were just A and B, you would likely give full support to B. Since C is in the race, you might give less support support to B which could help A win. One of the things that bothers me about range is voters may give unknown candidates a middle score. You say that this experimentally doesn't happen. Do you mind sourcing that? Btw, I think Range is one of the better voting systems. Ideally though I prefer a top two approval voting system. You would have a single non-partisan primary that everyone runs in and allow the two most approved candidates to advance to the general election regardless of party.
Hey, thanks for watching! These are great questions. So IIA refers more to removing the options after the ballots have been cast. With the standard of method of voting in US elections, if you remove a third party, the votes that went to the third party need to go somewhere, and where they go could end up causing the winner to change. Similarly, votes that use ranked preferences will typically see situations where the result will change if a candidate is removed, and the votes are recounted. With range voting, each candidates average is independent of the others so removing one candidate won't change the overall ranking. However, you are generally correct in that there are still ways for someone to vote tactically under range voting, and that were we to redo the entire election with one candidate removed, many people might change their scores for various candidates. I think that range voting strikes a nice balance between simplicity and the type of tactical voting that it's susceptible to. Here's a link from rangevoting.org about the unknown lunatic wins scenario being unlikely (rangevoting.org/Averaging.html#unlikely). I think that the soft quorum as a safety valve is still important, just in case someone does manage to get a high average with very few actual voters.
I would also point out that a "score runoff" or "approval runoff" voting system would prevent a "lunatic wins" scenario as well as the scenario where the score voting picks a Condorcet loser. Washington and California could become powerful testing grounds for ARV or SRV if they modify their blanket primary balloting systems.
Is it even possible to pass IIA with deterministic, no-dictator elections? If you're considering that voters change their votes depending on the available candidates, I don't see how anything could pass it.
+Skyval Ream, yes! As mentioned above, range voting does satisfy IIA. However, if you add in the unanimity and restrict the ballot to a ranked ordering, then no it's impossible (which is the statement of Arrow's Theorem).
Sorry, my reply was to Mutex50's and their "version" of IIA (where voters get to re-cast their votes. I've seen others say the same thing). I know if you don't allow voters to "adjust" their scores then Range passes IIA (and I think this is the better definition). What I was getting at is that I don't know how useful Mutex's version of IIA is. As long as the voting method allows strategic voting, I don't think any method can pass that version of IIA. But that's more a problem with strategic voting, which no voting deterministic, no-dictator method can completely eliminate.
Unanimity explanation is confusing. You mentioned the converse rule of majority loser criterion or actually the weaker rule, unanimous loser criterion. That one is also sort of satisfied (additional simple condition) by Majority Judgement but not by plain Range Voting as described. This mode of failure is quite hard to achieve in practice. Crucial difference is the use of median rather than average. The fix to give 0 stars is wrong as it prevents explaining anti-preferences. The correct fix is to give a median score (impossible to know beforehand) or a fixed middle score. Presumably extending the scale to say -3 through +3. It does not actually enhance any of tactical votes and helps with honesty.
If one of the goals is to keep voters honest...what's to keep literally everyone from rating their top choice 5 stars and everyone else 1 star to specifically avoid giving a higher score to a more universally like candidate?
@@MustSeto but that's the problem with first past the post to begin with...people not being able to vote their actual preference out of concern of a less-liked candidate winning.
@@wolfpackjew In FPTP, supporting a front-runner means not supporting anyone else, so candidates "steal" votes from each other (vote splitting). In Approval you can continue to support your favorite. In general, approving of more candidates decreases the chances of any specific one winning, but increases the chances of at least one winning against those you like less. In the extreme this becomes the opposite of bullet voting: an anti-vote, where you approve of _everyone except_ the one you hate most, which makes just as much sense as bullet voting. Everything between can make sense too, e.g. voting for everyone from your favorite faction and every sufficiently similar or more electable compromise factions.
@@MustSeto That still doesn't prevent strategic voting. I'm not saying that's a deal breaker for any system, but it's disingenuous to try to sell it this way.
@@wolfpackjew Sure! I don't think anyone is trying to claim this system is completely immune to strategic voting or anything. In fact a different theorem, Gibbard's Theorem, says that's outright impossible for any system whatsoever. But approving just your favorite isn't strategic in general
Range voting is perfect in theory, but in practice it has too many problems: 1) to satisfy the independence of irrelevant alternatives the range must be absolute but this is not the case: If I vote like this (range 0-10): A-0, B-5, C-10 removing candidate A, my vote would have been like this: B-0, C-10 because with only those two candidates, I want to favor maximum C and minimum B. 2) tactical votes because an unknown party is better than a hated party, therefore a person who hates A will give a non-zero score, e.g. 5, to all unknown candidates and 10 to all favorite candidates. Giving points to unknown parties to disadvantage A is a huge problem. 3) the votes don't all have the same value, that is, there may be votes that affect the results more than others. 4) equal interests can be represented differently, depending on how a person defines 0 and 10 in the range.
1) IIA is not well-defined for rating-based systems. You could, very reasonably I think, define it in such a way that your example does not violate it, because they decreased the score for B, increasing its distance from C, and in many cases it makes complete sense for such a change to change the winner. This form of IIA still works with Arrow's Theorem. Of course, this isn't ideal behavior whatever you call it. Though a different theorem basically proves that no system can be strategy-proof (Gibbard's theorem), and IIRC it doesn't depend on IIA. 2) Seems a bit speculative, and personal. I would probably agree with you that an unknown could be worse than the hated party and would probably give them minimum points (or close to it). And I expect most people who agree with us to do roughly the same---in which case there is no problem. But others could disagree, and honestly believe an unknown candidate would likely be better than the most hated party. That's their perogative, and heck, maybe they could have a point. 3) I think you're talking about voter equality? Like If a voter doesn't use the max score for any candidate? If this ended up relatively common & unintentional, I could understand the argument of this being a problem (and there are modifications like STAR which alleviate it) but I don't think it's an issue in general. Someone could do it intentionally, sort of like a partial abstention. OTOH, there is a sense in which systems like this are more equal than FPTP and RCV/IRV. Any vote can be canceled about by one other person's vote: Just invert each candidate's rating. FPTP can't do this, which is basically why it has the spoiler effect. 4) I don't think this is much of a problem. For one, this applies to pretty much any system. Two people who vote for the same candidate in FPTP could feel differently about them, even if they are honest votes. Same for ranks. Ultimately, the "meaning" of any system is defined by its rules. With Score, you increase the chances of candidates you give higher scores to, which decreases the chances of candidates you give lower scores to. How a voter chooses to do this depends on how they feel about the candidates.
range voting is the same as Arrow Theorem minus IIA. 3 candidates = range of 3, 5 candidates = range of 5 etc. IIA is too strong. Plus Arrow's dictator is morally the same as the +1 vote in a 50% + 1 simple majority.
I guess we're going through some of the same videos. I don't understand what you mean about range being "the same as Arrow Theorem minus IIA". The +1 vote in a 50%+1 majority not is "morally" the same as Arrow's dictator. The +1 vote only gets to decide when another 50% agree with them. Arrow's dictator gets to decide even if 0% of the others agree with them.
But even in score voting it would still be best if I am not completely honest, I would sure give my favorite condidate a top qcore but I would want to give some candidates a score that is way less than the one I believe they deserve just so my favorite candidate can have a better chance
That's only safe if your honest favorite is already strong, otherwise it's throwing your vote away. If duopoly is ended, and there's a good pool of candidates, some which are fairly similar to each other, than it starts making sense to support a few of them.
@@MustSeto actually I thought about it more and it seems to me that you are correct, this scenario could only happen if there are only two candidates woch kinda defeats the purpose of why the method was proposed. However when I told my brother about this he sqid to me that this could facilitate fraud because those who count would easily change the papers where as with a one vote system if you try to modify anything the vote is immediately canceled so that could be a problem.
@@destroctiveblade843 If it was implemented exactly as shown in the video, you'd you be right, you could "add more stars" to candidates. But in the places that I've seen actually propose it, it doesn't work that way, You just fill in one bubble. If you want to give them a 3, then you fill in the "3", etc. No candidate should have multiple bubbles filled in. You could arguably permit having none filled it for a candidate, so that you don't have to spoil their entire ballot, but should probably at least "encourage" filling in a "0", to prevent fraud.
What if parties were not important? - What if 'Dictatorship' were handled in a different way to arrive at the end result in a way that is different from a single-vote?
Honestly I don't like Star voting. Voting candidates that you like other than your favourite high could mean that your vote harms your favourite candidate. This isn't possible under STV.
STAR voting resists it better than Range voting. And it is possible under STV, although it's a multi-winner method with different applications than Range/STAR anyways
How about Balanced Elimination Voting? wp.me/p23U97-eZ I designed it to address a perception (correct?) on the part of IRV advocates that in Range Voting, you can't provide any support to a compromise candidate over the worst (in your opinion) candidates without sacrificing some of your political power to support your favorite candidate should that candidate have a chance to win.
Condorcet Ranked Pairs is better and still simple enough. The issue with Range Voting is that you are incentivized to vote contrary to your actual preferences depending on how you think others will vote. Lets say you have a candidates A, B, and C. You think A=5, B=3, and C=1. But if you think other people like B and it's going to be a close race between A and B (and that C won't have a shot), then you will want to rank B=1 to maximize the chances of your favorite A. Or, if you think all 3 have a chance of winning, you might want to rate B=5 to minimize the chances of your least favorite C.
Not that much of a problem in practice, considering opinion polls and other dynamics. But it's important to highlight you can ALWAYS fully support your favorite in score voting. The only way to completely address that type of problem you mention is to never allow your vote to support someone other than your favorite unless your favorite can't win. But ALL reasonably simple voting systems CANNOT do that without breaking the "you can always support your favorite" feature. So which is more important: 1) Always be able to support your favorite honestly, or 2) Always being able to support other candidates beyond your favorite You cannot have both. I think most people really want 1, and with time, score voting will allow more and more good candidates and parties to thrive, which means supporting another candidate and having them win over your favorite isn't going to be the end of the world to you. In practice, in score voting, at worst scared voters will lie about indifference. People would play it safe and rate A and B max. In all ranked systems, including Ranked Pairs, there will be many situations in which it is advantageous to lie, and that lie means supporting A over B when you really mean B over A. The lie has you flipping your preferences. This is a much bigger lie than lying about indifference.
There is a sort of latter-no-harm type violation, where you can rank someone as a 3, and your preferred as a 5, and your 3 could be the deciding factor in your second choice's victory. It seems like it will behave exactly like FPTP in the following way -> Voters know that not giving a rating/rating zero hurts the average of their non-preferred candidates. To increase their preferred candidate's chances, they rank their preferred candidate 5 and don't rank/give a 0 to every other candidate. Everyone does this because of the prisoner's dilemma. Is there experimental evidence that people give their honest preferences without strategic voting on a range vote even when there is this mathematical reason not to?
That's only strategic when you preferred is already very strong. Otherwise it's wasting your vote, just like in FPTP. My understanding is that being maximally strategic under Score goes like this: Before you vote, give honest scores to every candidate, and determine each's probability of winning. Multiply each candidate's score by their probability of winning, and add the results to get the expected value of the election. When you vote, give a max score to everyone who's honest score was higher than the expected value, and a minimum score to everyone else. Even this only makes sense when you have very good information (can accurately assess probabilities). Otherwise you could give a max score to someone you shouldn't, or give no support to someone you should have. In these cases, it can be safer to give some more honest, partial scores. Also it's not quite the prisoner's dilemma, it's a related but different dilemma called the chicken dilemma.
I can see that makes sense, but when I say "Voters know...", I don't mean maximally strategic voters or polisci major voters or game theoretician voters. I just mean a certain number of typical voters, who don't think about voting much except for about a couple weeks/days before the election, but think about it enough to try "bullet voting". I think my critique is pretty limited, because it's doubtful that EVERYONE would do that, but it seems like it could be a concern if a significant number did. Thank you for the correction on the chicken dilemma, that's a good one to know!
Giving a zero to your second favorite candidate increases the chance of all other candidates compared to your second choice. This means that your first choice is more likely to win, but also that all other candidates which you prefer less than your second choice are more likely to win. It's a risk-reward-consideration which every voter has to make. Do you want to maximize the chances of your favorite candidate winning or minimize the chances of your least favorite candidate winning?
4:14 obviously not voting on a candidate should equal 2.5/5. Why in god's name would a no vote be counted as 5/5? The real problem with range voting appears to me to be that if I dislike the Tories (I'm a Brit) then I might exaggerate my hatred for them, making it seem like I hate them more than Hitler; just to force them out the game. However, that would be my personal hangover from the horrible First Past The Post (plurality) system with strategic voting, because with range voting, maybe there is a tiny party I actually hate more than the Tories, who I never thought would have a chance of getting in, but now they actually could if I do weird strategic shit
The no votes weren't counted as 5/5s, they just didn't effect the average. The average is TotalScore/TotalVotes, and there was only one vote since abstentions aren't counted. Most proposals I know generally just use TotalScore, which would be like treating abstentions as 0
I've always liked range voting better than ranked choice voting, but both would benefit from the option to represent direct opposition to the election of a candidate, making them balanced range voting and balanced ranked choice voting respectively. To better understand this, consider the last example given in this video but imagine that nobody really wanted either of those two but were instead more afraid that one of those two would win than the other, but scared of them both. Now, add in a candidate who is wanted by 10% of the people and unknown to the other 90% of the people. If the 90% of the people who don't know anything about this new candidate think it would be better for an unknown candidate to win than either of the two that they're afraid of, but believe that the candidate is so unknown as to not have any reasonable chance of winning, and do not honestly SUPPORT the unknown candidate, then most of those people will probably give that candidate a vote as close to zero as the system allows, yet still vote FOR a candidate they are afraid of, in the hopes of preventing the election of the candidate they are terrified of. Since the 10% who adore this third candidate aren't enough to outnumber these fear-based votes, the "second worst" candidate is likely to win. Add in the ability to cast a negative vote, and such fear or dissent or disapproval can be represented and counted much more accurately, resulting in a better outcome.
Let's talk about MAL (my anime list) rating system. 3.1% of the people who voted for Gintama rated 1 star on purpose. That's because current top anime was FMA at that time (now it's Kaguya-sama love is war s3) and they didn't want Gintama to be the winner. It's not like they hate Gintama, but they love FMA more than Gintama and they thought FMA deserves the first spot. And I think if all people rated honestly, Gintama would have been the winner.
Isn't the biggest problem with range voting that voters can still get a benefit from voting dishonestly, by giving Squirtle 1 star in the last example?
I'm not found of the star method because obviously most voters will just give 5 stars to their candidates and 0 star to the rest. Ranked voting seems the best to me.
Eh, IIRC that's theoretically the optimal strategy if you have good enough information, but with more uncertainty in the details of how other people will vote I think it starts to make sense to give some partial scores. Anyways I'm not sure this would harm the method's overall performance much, Approval Voting is supposed to be pretty good as well. Plus, there's a method actually _named_ "STAR" (Score Then Automatic Runoff) which modifies plain Score slightly to try to encourage differentiating compromises a bit. First there's a plain Score step to find the top _two_ (instead of just electing first place), then whoever between those two is scored higher on the most ballots wins, even if they were 2nd place. This is using the same ballots as before; there's no separate round where you show up again or anything. So whether you say A=5, B=4 (difference of 1 point) or A=4, B=0 (difference of 4 points) that's one point to A in the 2nd round. But if you say A=5, B=5 then neither gets anything. So if A and B are competitive you're better off docking B at least one point, unless you're truly very indifferent between the two. It doesn't hurt B much against other candidates in the first part, but if it comes down to it, it helps A a lot in the end.
Sadly this video only tells half of the story. Independence of irrelevant alternatives is only true if voters are forced to vote honestly (which can't be done). Otherwise, additional candidates will affect how voters give ratings to the already existing candidates because of strategic voting. Let's illustrate an example. Say that there are two candidates, Charmander and Squirtle, and you like one for 5 stars and the other for 4. However, as seen at 5:00, voting 4 stars for Squirtle will give Squirtle a better chance of beating Charmander than voting 1 star for Squirtle. In general, there will never be a strategic reason to vote less than 5 stars for your favourite and more than 1 star for your least favourite candidate. Now let's add a third candidate, Bulbasaur. Say you like Bulbasaur for 2 stars, i.e. less than Squirtle. As mentioned before, giving Bulbasaur 1 star is only strategic, however, Squirtle can now get anywhere from 1 to 5 stars, and I'm sure there are scenarios where you want to give Squirtle more than 1 star. Thus, adding Bulbasaur has impacted how you vote for Squirtle, potentially changing the result between Squirtle and Charmander. Favourite betrayal is also only avoided to an extent. While it is true that you will always be encouraged to give your most favourite candidate 5 stars, your least favourite candidate 1 star, and give all candidates an honest ranking, the precise scores for the candidates in-between are up to strategy. You might betray your second favourite by giving him 2 stars just so your favourite has a higher chance of winning, even though there are other candidates you like far less than your second favourite. It's easy to imagine a situation where polls and media would have a lot of manipulative power not just by influencing how much people like each candidate, but also by lying about how popular and dangerous certain candidates are, and letting voters come to the wrong conclusion as to how strategically vote.
tl:dr The problem is not that people will vote honestly and that that will betray the condorcet winner, but that they will NOT vote honestly because they get better results from voting strategically.
I prefer this system en.wikipedia.org/wiki/Majoritarian_representation because it's not a matter if you put 1 star or 2 ; the problem is how many you have above 2.5 stars. But yes, we need to change our voting system
In the last example, people who like Charmander would never give Squirtle 4 stars, because they know it has a high approval among other voters and then their favorite candidate (Charmander) wouldn't win. IMO in practice ranged voting is actually more messy and worse than what we have right now.
If voters are strategic then in a two-candidate election Score is the same as Plurality/what we have right now. This is already sort of an improvement over most methods because they're the same as Plurality even if voters aren't strategic. If there are more candidates then it can be strategic to give multiple candidates points. Technically with perfectly strategic Score (with perfect information) decays into (Strategic) Approval, only giving the minimum and maximum score, though not necessarily just one maximum. In practice with less information it can be strategic to hedge your bets by giving some partial scores.
@@MustSeto you are correct, I even forgot I made this comment! hahahah I kept reading the other day about this topic and indeed, Score voting would degenerate into Approval voting in that perfect-information strategic scenario. Now I like Approval more than Plurality. But I also like Approval more than Score voting, because it's more simple and easier to understand. To make it even more simple, my fav option would be Approval voting with the choice of choosing 3 candidates max. This would make everything even easier to understand by citizens, and would still avoid favorite betrayal🥰
@@trincotrunco8932 Cool! Yeah voting methods is a real rabbit hole. It just keeps going... I largely agree about Approval being simpler than Score, and I also think that having larger and larger score ranges gives diminishing returns pretty quickly, even in theory. That said it seems like a lot of people _really_ hate being limited to just two options, and having to chose between putting compromises with either favorites or their hated, so I've wondered if 3 (e.g. for/neutral/against) would be more popular while still being simple. Maybe increase that to 4 if having an exact middle causes issues... I think there's been research on ratings scales like this (usually for surveys and things) that suggests people are unreasonably drawn towards middle scores, and you get better data on average if there's no true middle. Not sure how we'll that'd apply to this, or if voter frustration would outweigh the benefits, though. What do you mean about choosing 3 candidates max? You mean only 3 candidates can run, or each voter can only approve 3? I don't think I'd do either. I could see having a "unified primary" election to narrow the field a bit before having a "final" election being useful to let voters focus research on a more manageable number of candidates, though I don't know if I'd limit that to 3. Maybe 5, or maybe the top 20%, or something. A big draw of alternative methods and avoiding the spoiler effect is moving away from duopoly and towards more varied and nuanced races with multiple candidates. And I don't see the point of limiting a voter to just 3 approvals, especially if we do start getting a good spread of candidates.
Surely there is no favorite betrayal, but there is still strategic voting (second favorite betrayal, if you like). Let's say you want candidate C to win, are okay with candidate B and do not like candidate A. Then of course you will give C 5 stars and A 0 stars, but how much will you give B? If you're not too afraid of A winning, then it would be strategic to also give B 0 stars, since this increases the chance of your favorite candidate winning. But if you fear that C barely stands a chance and A is very likely to win, you'll also give B 5 stars to prevent A from winning. In the end, the systems probably converges to approval voting and with a huge incentive for strategic voting. Surely this system might work in a beauty contest or for Netflix, where the voters really just want to find the better candidates/films, but in something as divisive as politics, incentive for strategic voting should be kept to a minimum.
Maximally strategic Score does become maximally strategic Approval. But maximally strategic Approval elects Condorcet winners whenever they exist. It's not just about how much people are "incentivised" to be strategic, it's about what the effects are. For example, with Score/Approval, strategy is usually about Compression, while with Ranked systems, it's about explicitly misordering some candidates. So even a maximally strategic Score/Approval ballot arguably still contains more accurate, honest information than a minimally strategic ranked ballot---maybe even a fully honest one. In fact considering true Condorcet systems are not immune to strategy even when there is a Condorcet winner, Score/Approval might elect Condorcet winners more often than true Condorcet systems---except that when there is more honesty, Score and Approval can elect someone even better. Especially since being maximally strategic under Score/Approval can be risky, while being at least semi-honest still gives you most of your power without as much risk. I know I wouldn't have a problem giving a compromise some points if I did actually like them reasonably well.
I contend that strategic voting is not harmful and there is no reason to discourage or avoid it, provided that the voting system gives everyone the same amount of power over the outcome. I think the strategically best amount of exaggeration to give your compromise candidate depends on the popularity you estimate that your favorite candidate will have with the other voters.
I don't see how you can claim that strategic voting is harmless. In any scenario, it encourages voters to be dishonest, whereby the election results do not accurately reflect population wants. Not everyone is a voting theorist, so a lot of folks will vote honestly, whereby their vote has much less power compared to those who vote strategically (i.e. in this scenario only the appropriate mix of 5's and 0's). Campaigns will aim at explaining voters how they should vote (might e.g. encourage them to ditch all candidates but their favorite), possibly misinforming and confusing them, making them again angry at the 'needless' complexity of the system, and in any case taking a lot of time of the public debate, which should be spent on discussing actual policies, not voting tactics. There's more arguments to be made. I can see why one can claim that strategic voting might be the lesser evil in choosing a voting system, but you just cannot reasonably deny that it is something which is in principal very undesirable. For me, in something as divisive as politics, it is a larger put-off than e.g. failing 'Independence of Irrelevant Alternatives'.
The concepts of honesty and dishonesty do not apply to voting in an election to political office. The election is not an opinion poll; it is a channel for the exercise of political power by the individuals who vote, to produce an aggregate decision. To call it honest or dishonest is like asking whether the position of the throttle cable on a car is honest. It can't be honest or dishonest because it is not communicating an alleged fact, as for example the fuel level meter is, or the odometer. Those can in fact be dishonest; the odometer can be rigged. But the throttle cable is communicating a _command_. An election is the same kind of communication. In fact it is a game. The voters command the tally to take them into account in a certain way, within the freedom of movement that the system allows them. A command can't be dishonest. The question that matters is whether your command gets more sway than my command.
The problem is that imagine your candidate is approved by 60% and will kill the 40%, and the second candidate is slightly less approved by the first 60%, but are loved by all, the problem is that the first 60% has an incentive to rank the second voter horribly for a preferable outcome.
You’re assuming that the entirety of the 60% would vote 0 for the other candidate which is not true as far as real world studies go. 75% of people using range voting in studies vote using numbers that aren’t just the min or max. In fact the majority of range voters end up voting honestly in comparison to FPTP voting. With that being the case, the candidate liked by most will win. rangevoting.org/HonStrat.html
@@115zombies935 We don't really know how strategic these people really are. From a game-theoretic perspective, perfectly selfish vote from everyone results in 40% of the population being targeted.
@@swordwaker7749 We do know that though. It’s been shown many many times that range voters tend heavily towards more honest votes rather than purely strategic ones.
@@115zombies935 Yes, but it depends on the circumstance. A population might start out honest and get more strategic over time when they don't get what they want. This depends on a lot of factors. We don't really know what would happen in a given real-world situation.
If I want Charmander to win and give him 5 votes, and give Squirtle 3 votes ... what prevents me from forcing myself to give Squirtle 1 vote in case those 2 additional votes of mine would have made Squirtle win over Charmander? Wouldn't everybody be forced to lie and vote 5 to a single candidate and 1 to everybody else?
while this is likely to happen when there are two candidates as in the example, i don't think it would happen if you had, say, 10 candidates the fear that your five star loses is the same fear that your one star wins, so you rank someone else at 3. if you want.
We game systems. Range voting gets gamed by parties saying, "Vote 5 for these guys" and "Vote 1 for these guys". If either party fails to play the game, they lose. The best way to combat this is with approval voting. It produces the most acceptable alternative. It doesn't focus on finding the strong favorite of a small group. People often wonder why Netflix went away from 5 star voting, and why Google, UA-cam, Reddit, Twitter and Facebook never implemented it. It SEEMS like a five star system would give you richer data, but it only does in specialized circumstances. In high stakes contests, the results are always gamed to maximize the outcome. This invalidates the utility of the granular system. We should also have a minimum threshold for winning, such as 80%, and should require a quorum of 80% of the voters. (roughly). Other than that, approval voting for candidates is superior to plurality, ranked, or scored voting. Academics like to worry about pareto optimality and Condorcet winners. In the real world, approval voting produces the most acceptable alternative. It favors compromise candidates and solutions. Nuf said :) Okay... a little more. Point rating systems are great for when data is granular, and when a comment can be attached to the rating. In practice, they produce very close to the same score as an approval vote when aggregated and scaled. 80% up = 4.0/5.0 (4 stars out of 5).
Because Score/Range passes favorite betrayal, unlike other methods, voters don't have as much incentive to obey this. For one, there is never a reason not to give max points to your honest favorite, or indeed, if you give X points to candidate A, there's is not reason not to give at _least_ X points to every candidate you like better than A. So even if the front-runners are saying not to give some candidates any points, they don't have a leg to stand on. Especially since giving some points to a compromise candidate can reduce risk. Also, I don't see how Approval would help with this. "Approve these guys, don't approve these guys". If anything, it just makes your criticism mandatory. UA-cam used to have 5-star voting, and I think none of your examples are Approval, they're more like Score form -1 to +1. And IIRC Netflix and UA-cam's reasons for switching to this system was because they were mostly only getting extreme scores anyways, but that makes sense given their context. "Voters" aren't required to score videos or posts or whatever, so they only go out of their way to do it when they strong feelings about it. But for an election, I don't think this is applicable. Have you seen Warren Smith's Bayesian Regret simulations, or Jameson Quinn's Voter Satisfaction Efficiency simulations? Both put Range/Score above Approval in terms of the quality of the winner.
Yeah. UA-cam was 5 star. I guess Facebook, Google and Twitter are the approval voters, strictly speaking. +1/-1 is a type of range voting that is close to approval voting, but you're right that they're not equal. Broad score systems get gamed by giving max stars (points) to all the candidates you'd be okay with, and 0 to everyone else. In effect, if you play it intelligently, it turns into approval voting for the "players", who have a greater effect than the honest voters that give a real 0-5 point score. As long as we don't use Ranked or Plurality voting, I'm okay with the others. All the scored voting methods are in my favorite class of voting, and all product roughly the same data, except when they are gamed. When they are gamed, approval voting doesn't take power from the voters and transfer it to the "players". Thanks for the reply.
It's true that if you have perfect information about an election, strategic Score is identical to strategic Approval. But with more uncertainty, giving slightly more honest scores is more strategic because it reduces risk. This is an advantage over Approval. And Approval can still be gamed, by being more "selective" about who you approve, e.g. don't approve compromise candidates if someone you prefer has a better chance---which could prevent good compromises from winning. Smith's and Quinn's simulations suggest that when both Approval and Score are gamed, Score still does better on average, but Score is better yet when voters are more honest. I do agree that either should be acceptable, and Plurality Voting or most Ranked methods (especially IRV) should definitely be avoided. Condorcet methods seem like they might be okay though, if you ignore their complexity. And thanks for the compliment! It's always nice to have a good conversation about such an important, under-appreciated topic.
I would say maximizing happiness and the other benefits of range make it favorable to having a Condorcet winner.
While I don’t necessarily think the Condorcet criterion is the be-all, end-all of elections (you aren’t comparing candidates in isolated pairs), if a system violates the majority criterion it’s undemocratic.
If 40% of voters prefer Trump, and give him an average score of 5/5, and 60% of voters prefer Biden, and give him an average score of 3/5, then Trump wins.
That’s even worse than the current system.
@@rucksplash12 If 40% of your country is absolutely for the worst possible person, then you're doomed no matter the voting system
@@PhreackShadow not if they don't win power...
Condorcet winner should be thrown out of the window.
But looking at the average of grades is a stupid idea too.
We should look for a person that is described by everyone as "above average" and avoid candidates that are hated by some and loved by some.
It's like choosing a film for a film party. You may love your arthouse with all your heart but if it triggers Epilepsia - put it in the no-no shelf. Even if it's just one person with Epilepsia there.
@@rucksplash12 I disagree that it’s undemocratic. In the example given, it actually seems as though MORE people are happy with the result. The majority’s first pick may not have been chosen, but their second pick was a very close preference for them, and for the entire group as a whole was most preferred. Wouldn’t it be more democratic for the chosen candidate to be one that is liked the most overall by all people, rather than one that may be the first pick of the largest group, but isn’t the most liked by everyone?
...this system is so insanely simple and intuitive it's a surprise no one ever talks about it. Maybe powers that be like the way things currently are?
Because it would be bad policy.
If 40% of voters prefer Trump, and give him an average score of 5/5, and 60% of voters prefer Biden, and give him an average score of 3/5, then Trump wins.
That’s even worse than the current system.
Your biggest mistake is assuming the candidates would be the same. Biden is the candidate because he has the best chance of winning under the current system. Second, why are you assuming people with give the candidate they prefer to Trump 3/5? You’re also ignoring the fact that people who don’t like Trump are going to give him a low rating.
@@ganondorfchampin I'm assuming that people will vote sincerely.
There is absolutely no advantage to replacing first past the post with score voting when in order to get your top candidate elected you have to give the max score to your first choice and the lowest score to any other choice. It has devolved to FPTP, but a more complicated one where someone honestly voting and indicate that they tolerate Biden but don't love him can end up with the person that they hate winning.
Biden and Trump are just interchangeable names to show that the system is undemocratic. They could be any two people.
It's possible there are different nominees with score than FPTP. But score is still undemocratic. We shouldn't replace the electoral college with a different system where someone with fewer supporters can beat someone with more supporters; we should change it to a system where that can no longer happen.
The idea is that there would be way more than two candidates, and again, you’ve been completely ignoring that fact that people who like Trump will be balanced by people who absolutely hate him.
@@ganondorfchampin there are more than two candidates now.
It doesn't matter what you think might happen. It's what the system allows to happen. A system which would allow a group of 40% to defeat the will of 60% is undemocratic and we shouldn't adopt it.
STAR voting (Score Then Automatic Run-off) solves the condorcet candidate. it works the exact same way, but then the automatic run-off takes the top two candidates, then regroups everyone's preferences between the two. so, the candidate that is popular and unpopular would lose to the popular candidate with the broadest support.
this encourages candidates who prefer coalition building.
This is great. You should make one of these for STAR Voting!
The Nash equilibrium of Range Voting tends to approval voting. This is why the Condorset winner comes up. Approval Voting only fails to get the Condorset winner when the compromise has a lower Bayesian Regret
Underrated comment. At 5:10, the pro-Charmander would not rationally give 4 stars to Squirtle because they know they would lose their preferred candidate. Knowing this, the pro-Squirtle would do the opposite.
Approval voting under perfect strategy and knowledge gets both the Condorset criterion, and all the Arrow paradox properties right, except maybe the independence of irrelevant alternatives.
@@jibraan123100 If they have perfect knowledge, they know that giving Squirtle a 4 will not elect Charmander, while giving Squirtle a 0 will. So they will be incentivized to give Squirtle a 0.
Thank you for this comment. I was struggling to put my finger on exactly what was wrong with this.
@@darbyl3872 People are strategic. If they get a choice between electing a candidate they like for 4 stars, or one they like for 5 stars, they will choose the 5 star candidate and vote accordingly - voting the 4 star candidate lower than they like him.
@@darbyl3872 Strategic voting gives them _more_ power for the candidate they want to vote to the top, compared to any candidate they like less. While star voting can let you give 5 stars to everyone up to and including the candidate in your personal preference list that you want to push, you are still encouraged to give everyone below that pushed candidate lower ratings than if you were voting honestly.
I don't think the Condorcet criterion matters. A higher score matters more than the Condorcet winner in terms of Bayesian Regret.
I link people to this video quite often when they tell me to vote third party. Basically my response is "Don't lose focus after the election. Things need to be changed, so pay attention outside of the media and keep working for a system that makes third parties viable. This is one."
The world would be a better place if we could go back in time and replace mindless plurality voting methods with score (range) voting.
We have the present or the future to do that too.
score voting is nonsense tho
@@noneofyourbusiness6269how so? elaborate
I saw this after I saw the rank based voting system on represent. Us. Honestly I prefer this one slightly more. I love your style, great video! I honestly think majority of the group happy is better because people constantly complain about picking the lesser of two evil.
This. We need this. Everywhere.
Condorcet is suppose to be how the multicandidate election winner compares to head to head elections. The head to head winners should also be selected by range voting. There is no reason to fixate on ordinal positions. In your example the majority's second choice won because the majority didn't care about the difference between the two candidates (not because the candidate was polarizing). If you care strongly between two candidates, score them differently. It's as simple as that.
its not that simple actually. if your true opinion is charmander 5 starts and squirtle 4 starts, and you are able to predict that one of these 2 candidates will win, you should vote charmander 5 stars and squirtle 1 star. It is often quite easy to predict the top candidates.
Voters shouldn’t have to vote strategically in order to get their honest preference elected.
In score voting, If 40% of voters prefer Trump, and give him an average score of 5/5, and 60% of voters prefer Biden, and give him an average score of 3/5, then Trump wins.
That’s even worse than the current system.
If a voter’s honest preference doesn’t deliver the result they want, the system is flawed. If 60% of voters preferring a candidate doesn’t get that candidate elected, the system is flawed.
@@rucksplash12 I disagree. If 60% of people barely prefer candidate A and the other 40% would flee the country if candidate A won, but candidate B would be so awesome that 100% of everyone would be so happy that they'd personally make a statue to commemorate candidate B in their space and want to sing and dance as much as possible. Then candidate B sounds pretty good. Candidate A sounds so bad that you are basically are deciding whether to keep having a country at all.
Here is an example that might be different enough to challenge your bias about how elections are "supposed" to work. What if your vote wasn't 100% secret. In fact, what if the government set aside US$100,000.00 to give to every voter and the range you assigned to the winner would proportionally decrease your tax refund. So if you gave the winner a maximum vote, you get $0.00 out of that $100,000.00 (but you are still happy since that's truly how much you loved your candidate) and if you gave the winner half the maximum vote you get only $50,000 (but you are still happy since that's how much you loved the winner).
In this example, the range vote winner is the one that saves the government the most money right out of the gate. But the election resulted in every single person being super happy since they were all compensated for exactly how much they preferred the candidate that won. A country with everyone happy about the result. That is what range voting is designed for.
Other reasons to have other voting systems is that we don't have to have just one winner. Maybe only one person can be commander in chief at a time, but there is no reason that multiple candidates couldn't share at least some powers. God didn't come down and say the point of an election is to pick a condorcet winner that merely has the property that if a majority of people don't disprefer one candidate by even the tiniest amount (i.e. one person might prefer the candidate by the tiniest of tiny tiny tiny amounts and everyone else in the majority might find them equally good) that that one has to be the sole winner of everything no matter how much the minority hates them.
It seems like you have a bias towards systems that can result in most of the country outraged (since not 100% of everyone will be able to vote) while a near majority didn't really care much.
And the fact that some people will be deeply busy in the hospital, or deeply sequestered in a jury or deep undercover or such and so won't be able to vote that least a couple votes won't change the total happiness of the country much if they get the wrong answer.
Because range voting is about maximizing happiness of the country.
That's why I used the money example. You aren't going to vote away more money for a candidate unless you really value that candidate more.
@@TimaeusBouma but elections aren't about getting money.
An election system which allows the passion of 40% to override even a weak preference by 60% is unacceptable.
I have a bias toward systems which are democratic. People have a right to be outraged.
But a system which allows a minority to overturn a majority is going to leave a lot more people outraged.
In the example I gave, 60% of the population gave the winner the lowest popular score. A system which allows such a winner is not a system which recognizes the will of the people.
You can't base a system on how it would work in an ideal made up scenario where the majority have a weak preference for their chosen candidate but are OK with the alternative. For starters, even in that case, it's still not their desired outcome. But secondly, it's not representative of the system's overall performance.
A system which would deliver an undemocratic result even in one particular scenario is unacceptable.
@@rucksplash12 Is there a reason you didn't state that the 60% gave the lowest possible score?
To me, if you take a system designed to express preferences for more than two candidates and people use it badly that just seems like it is bad on them.
Because, honestly. What it you quantified it is as hate minimization instead of happiness maximizing.
Then if 60% of people voted 3/5 for one candidate and minimum for another canidate than you see the amount of hate from that 60% based on who wins (60% maximum hate of the other candidate and (2/5)*60% hate if their allegedly "preferred" candidate wins).
If there is truly less hate generated by the candidate that generates 0% hate in 40% of the population and so the overall hate generated is only 60% then that sounds terrible ... sounds like maybe a third candidate should have run.
But if that candidate generated the least hate. Sounds better than the condorcet candidate.
The revolution was bot fought based on democracy. It was based on enough people hating rule by England enough that they were willing to to fight and die. It was based on the collective amount of dissastisfaftion.
And your example shows more aggregate dissatisfaction for the candidate with 40% with zero hate and 60% with maximum hate. Since the only other person willing to run had 64% of the maximum possible hatred (your numbers dude) then the person with only 60% of the maximum hatred wins.
Makes sense. Find a less hated candidate if you want a winner. The video was pretty clear that things degenerate into voting against people if you use ordinal voting without taking into account how *much* people hate a candiate.
Basically you completely ignored how much people hated that candidate that hat 64% of the maximal hate. You could probably have run a ficus and gotten better results. Saying 3/5 for candidate B and 0/5 for candidate A means that that is exactly how different you consider them to be.
The money just forced people to be honest. But why be honest if people like you are going to ignore their feelings anyway
Ranked choice voting was my favorite voting system until I saw this video. I wish more people knew about alternative voting systems beside plurality. I think ranked choice voting could be a stepping stone to range voting.
This is what happens when a trusted education channel only thinks the topic halfway through.
As hinted towards in the example at 5:00, this voting system still enocurages strategic voting. And with strategic voting, you actually lose independence of irrelevant alternatives, because even a least favourite candidate can change how you vote when voting strategically.
Ultimately, voting in this system becomes an advanced strategic task, disfavouring not only less educated people, but also allowing polls and media to manipulate strategic votes by painting the wrong picture of who is at risk of winning
@@darbyl3872 This video is incomplete, as it does not address the strategic voting that would result from a rating-based election. If you have really studied this topic for years, I hope you can fill out the blanks for anyone who asks (including me).
@@darbyl3872 It is true that strategy goes into every part of voting.
But that is only what makes it so important to look at HOW strategic voting impacts each particular system.
With "STAR", you are left with a system that does not have the advantages over ranked choice that the video outlines. This is a big part of the picture, and people should at least be made aware of those drawbacks.
I've outlined examples for how I can know that "STAR" does not have independence of irrelevant candidates, but the mathematical proofs on this topic surely go further than my own casual thoughts.
I have read comments in here saying that the "Nash equilibrium" (that means the combination of optimal strategies for all voters) for STAR voting is Approval voting. If that is true, the main difference between STAR voting and Approval voting is that people with lower education will be voting in a less optimal way, giving them less voting power than if the voting system was Approval voting, rather than STAR voting.
And if that is the case: Why do you prefer STAR voting over Approval voting if you have spent years on this topic?
@@iwersonsch5131 Strategical voting is usually formulated and thought of in terms of ranked preferences, as if every voter has infinitely precise preferences (they can make perfect comparisons), they have perfect information and certainty about the results of the election, and would always want to maximally impose themselves regarding their preferences, no matter how minor they are. *NONE* of these assumptions are even remotely true in real life, otherwise society itself wouldn't exist.
A better analysis of strategical voting is based on Decision Theory. It recognizes that people have fuzzy beliefs about the world and themselves, which are constantly updated as new information arrives (which can be anything from reading a policy proposal, reading a book about economics or the environment, reading an opinion poll, talking to friends and family, watching debates, etc.). Still, they always have imperfect information and fuzzy preferences, with degrees of indifference. They will also only try to impose themselves (what strategical voting is *actually about*) IF they care about the results strongly enough, in which case it is a legitimate form of expression, not an "undesirable" feature.
An election is a choice under risk and uncertainty. Thus, we WANT voters to adjust their ballots as more information arrives so they can make their tradeoffs, as accurately as possible.
This is not a bad thing. This is a voter being informative and thoughtful. We do not want voters to make uninformed selfish decisions, and all information they get about candidates, reading about policy and its effects on society and others, or following pre-election polls, are identically valuable for the voter to consider what is a priority and what is not, and for them to genuinely express their beliefs.
If voters existed in a vacuum, casting ballots as if they were the only people in the world and had no concern for any external information, then the election would not be a *collective choice*, and ballots would, rigorously speaking, carry no information that could be aggregated across voters. it is the shared context which gives meaning and information content to the ballots, which is another reason we want voters to actively adjust to external information.
The same concerns also exists under ranked methods, but in a ranked method there is no adjustment under risk other than reversing preferences. In this sense, rankings are *less* responsive and representative of voters' genuine beliefs and behaviors than cardinal voting, and when adjustments do occur they are rank reversals, which completely distort the actual information we want!
Cardinal voting only seems to pose a problem in terms of "strategy" if a minority side really cares about making a distinction between two options, and the majority does not, and so the minority wins. But this is the correct outcome! First, the majority can still always enforce their will if they want to. Methods that are deemed "resistant to strategy" (like ranked methods) force the majority to always impose themselves over the minority, even if they don't feel strongly about doing so. The only difference in cardinal voting is that this imposition is made OPTIONAL. That's it. This is, at its core, the only thing people complain with respect to strategy.
But it is not up to us to decide when the majority should impose themselves. If the majority, given information about the election and the likelihood of outcomes, deliberately and rationally chooses to not impose themselves, then that is a clear sign that the minority overriding their preferences is something acceptable to do, according to the majority itself. So the "failure" of the method is, by assumption, not a failure at all, and was deemed a legitimate outcome by those who had a say in it.
With respect to score voting encouraging people to vote min/max scores only, that is only true under the 100% certainty, zero information, perfect preference assumptions. In realistic scenarios under uncertainty and imperfect information, it can be shown, mathematically, that the strategical equilibrium requires voters to cast more diverse ballots.
@@1ucasvb Humans don't have to have perfect information to gravitate towards a Nash equilibrium. If they see something doesn't work, they do something that works a bit better, and that way they gradually get closer to the optimal strategy
So would we be able to do away with both the Electoral College AND the Presidential Primary Elections?? That would be awesome. The primaries are a ridiculous mess.. every state so different and VERY restrictive. Closed primaries, open primaries, caucuses. Insane! We should just be able to vote for who we want!!
Correct: Score voting doesn't require primaries. Primaries exist because of our voting system "splitting the vote" between similar candidates. Score voting doesn't do that, since you can give similar candidates the same score.
This wouldn't do away with primaries... you'd still have two major parties, which are private entities, deciding which candidate to put forward for the general election. Getting them to agree to change our voting system on such a fundamental level will never happen, regardless of how much it would improve things. The realistic alternative we can work towards *right now* is the *_National Popular Vote Interstate Compact_* which has been covered on this channel before... *_EVERYONE_* should contact their state legislature and demand they pass the NPVIC so we can, once and for all, get rid of the insanity that is the Electoral College and go to a popular vote winner *_like every other country on the planet._*
"So would we be able to do away with both the Electoral College AND the Presidential Primary Elections?? That would be awesome."
Argentina has both the primaries and elections at the SAME time under the SAME system. How it works.
Round 1:
Every possible candidate from every party run and people vote at any candidate.
If an candidate of an party X get a vote, this vote goes to the party.
If the party has 1.5% of all votes or more, the party is the winner, and the winner of the party is the guy with more votes.
Round 2:
People pick an candidate between the remaining ones at second round.
If someone gets 45%+ of the votes he is the winner (dont ask me why 45%+ and not 50%+ and dont ask me what happens if two candidates have 45%+ of the votes).
If someone gets 40%+ but less than 45% and has more than 10% than the second place, he is the winner, if none of the previous two happens, there is an third round.
Round 3:
At third round the top 2 of previous round to election and the most voted at this round become the winner.
One problem with this method is that usa allow people to become candidates without being at a party and it would be a problem at this system.
I do not think that the electoral college should be done away with -- it seems to be doing its job of protecting rural communities from the whims of large urban centers. However, it is possible to adopt range voting within the current electoral system: the local elections produce average scores for the candidates, and the electors then use those average scores as their scores in the electoral college vote. The result would then be an electoral system that combines the merits of both systems.
@@k98killer that’s neither its job nor what it’s doing. Right now only 8 states get any real say in the electoral college; everyone else’s vote is taken for granted.
Rural voters shouldn’t get more votes than urban voters. Everyone’s vote should have the same value. Trump shouldn’t be able to win an election with 3 million fewer votes just because more of his voters lived in certain areas. That’s fundamentally undemocratic.
This is a nice video, but range voting does NOT violate Arrow's Theorem. The reason is that you have to assume people will adjust their ballots to fill the dynamic range, so when an irrelevant candidate is removed scores can be adjusted and this can change the winner.
For those of you who are watching now, there has been an innovation which fixes the Condorcet issue with Range Voting. Look up S.T.A.R. voting
its weird that these alternative vote organizations have popped up in US deliberately ignoring scored voting and campaigning inferior systems instead. (democratically). What gives ? Its great that they are stirring up awareness, but why do some organizations insist on being shy of what is optimal ? Why is the best method getting the last glance ?
Seriously! It's almost like FairVote doesn't want to unrig the system.
there's nothing optimal about cardinal utility
They don’t deliberately ignore them. They’ve done the analysis and concluded that score is not good policy.
In score voting, if 40% of voters prefer Trump, and give him an average score of 5/5, and 60% of voters prefer Biden, and give him an average score of 3/5, then Trump wins.
That’s even worse than the current system.
@@rucksplash12 Why is that a bad thing? If average happiness in the country is higher, I think it's a better outcome. I think it's easier to think about this without the political connotations of using real candidates (since it makes a system where your preferred candidate loses seem bad).
That being said, I also don't believe we would have ended up with Trump vs Biden if we used range voting. I don't think a candidate that averages below 3/5 would be at all viable in the system, and we'd get stronger choices from both sides of the aisle.
@@darbyl3872 there are lots of voters who don't fully understand electoral systems and who would give an "honest" rating rather than an effective one. For instance a Bernie fan might give Biden 3/5 and Trump 1/5 without realizing they're essentially just giving 60% of a vote to Biden and 20% of a vote to Trump.
Ranks may not be the ideal system, but they don't force voters to bullet vote the highest score for one candidate or else lower their effectiveness.
Range voting is clearly the best. Even in the last part of the video where you present a possible flaw it actually isn't a flaw ... If 60% is like 'I prefer Red (5)but I don't really mind Blue(4)' and 40% is like 'I like Blue (4) and hate Red (1)' then the fair and just outcome is that Blue wins. If Blue was really that bad the 60% wouldn't have given him 4. I believe this system is the best or one of the best out there.
I love range (or as i learned about it scored voting) just for the fact that it doesn't require you to choose 1 candidate over another you can support the mainstream party as well as supporting the smaller party you like the most equally.
Ranked is better
@@barnacles1352 No, ranked is significantly worse, and in some cases no better than FPTP voting.
rangevoting.org/rangeVirv.html
rangevoting.org/CompChart.html
Thanks for running through that explanation, I have been looking for a voting system with two qualities: a candidate that loses every head to head matchup must not win, and 'controversial charmanders' like the one in the example, even if they are the favorite of 51%, must not automatically win, squirtles have to have a shot.
This video should have more views!!
STAR Voting is the best system because it fixes the problems with Ranked Choice Voting & is simple.
Great video! I very much liked your explanation (4:37) addressing the majority criterion criticism with regard to range voting. Given that the choice between voting systems is essentially a tradeoff between different drawbacks, I would much prefer the one whose flaw biases in favor of a consensus candidate over a divisive one, rather than a system such as instant runoff, in which voting sincerely can potentially be self-defeating. Your explanation satisfied my concerns with that critique in a way that other discussions on the topic had so far failed to do. Thanks for the clarity!
I like how bulba was used to represent grass roots
Range Voting? This is literally just any rating system in existence - IMDB, MyAnimeList. This should be implemented EVERYWHERE NOW.
Historically it was used in the republic of venice and sparta. Today it is used in Latvia and the selection process for the UN Secretary General.
Nice video! I prefer STAR Voting though
It's a 2 round system with only one ballot required to be filled out.
To fill out the ballot: rate your candidates 0-5 stars
1st round: the two candidates with the highest total stars are the finalists
2nd round: the finalist most preferred wins (most preferred means higher stars than other finalist on the majority of ballots)
Each ballot counts as one vote only (if a ballot has the two finalists tied in stars given to them, that vote is counted as a vote of no preference)
There actually is a system, has a bit of both, although it a bit more involved. Its called something like: "single peaked preferences" and it uses the ordered list to compute the median ballet, the favourite on that ballet is automatically the condorcet winner.
Personally I think giving the Ok canditate more power is the best side effect a voting system can have, as it discourages extremism
I've set up a web site that uses range voting, similar to what StrawPoll does. You can find it here: rangevoting.info
Great video other than presenting average score as the logical option rather than total score which would eliminate the risk of low visibility fringe candidates winning.
I probably lean towards total score as well, but he did mention one method to try to avoid that.
Average is better for presentation anyway. But if blanks default to zero, there's no risk at all.
A better approach, if you include blanks, is that when you compute an average score you have a minimum denominator of a simple majority of total number of voters. Say 1000 voters participated, so you have a simple majority of 501. If a candidate gets 900 blanks and 100 10's, they would get an average score of 100 x 10 / 501= 1.99, instead of 100 x 10/ 100 = 10. Any candidate which received more than 501 non-blank scores would get the average computed normally.
This means candidates that are unknown to a majority get penalized here, and candidates known by a majority get the rating they deserve. It is reasonable because if a majority of people know a candidate and think it is on average a score of X, it is reasonable to assume those people giving blank scores would also average it as an X if they were to give a score.
Most underrated video on the internet.
what about proportinal range voting for parlmantary elections?
the idea is range voting for parties and then simply award seats proportinal to the party's score.
it sounds simple but strangely i don't find many ressources on it..
There is no need for that because in a proportional system everyone votes for his personal preference and the party then gets seats to represent this number of voters. There are no spoilers or anything like that.
Moreover, a majority of voters should not be able to surpress a minority by rating a small party badly, because the point of a parliament is to display everyones opinions.
Fantastic video. You deserve more views and subscribers!
6:00 You can implement STAR Voting, a simple score variant, to get both.
Great video!
I have a question about the Condorcet winner example. Charmander would have won had each ballot been normalized so that the lowest ranking given to a candidate was the minimum, and that the highest ranking given to a candidate was the maximum. So, using 1 as the minimum score, instead of
60% giving Charmander 5/5 and Squirtle 4/5, and
40% giving Charmander 1/5 and Squirtle 4/5
It would be
60% giving Charmander 5/5 and Squirtle 1/5, and
40% giving Charmander 1/5 and Squirtle 5/5
Is there any reason someone may not utilize the full range of ratings allowed on their ballot unless they thought all candidates are completely equal? Is it still possible that the Condorcet winner could still lose after applying this transformation?
There is a theorem that, if the supporters of the would-be Condorcet winner are sufficiently strategic (including only giving min and max scores) they will win with Range/Score and Approval Voting. In a two-candidate election, normalizing ballots would have the effect of giving one candidate the minimum score and the other the maximum (unless they are tied, in which case the ballot cannot change the winner) which basically forces everyone to be strategic. With three or more candidate, normalization alone might not do this. For example, what if Bulbasaur also ran, and this was the outcome:
60% Charmander 5, Squirtle 4, Bulbasaur 1
40% Charmander 1, Squirtle 4, Bulbasaur 5
Averages:
Charmander: 3.4
Squirtle: 4.0
Bulbasaur: 2.6
This doesn't make Bulbasaur the winner, in fact he's dead last, but now how would you normalize the ballots?
I definitely think the traditional majority and Condorcet criterions are naive, and not inherently desirable (though they were good heuristics when we didn't have anything better). In these examples, paradoxically, a supposed minority candidate had *broader* appeal than the so-called *majority* candidate. If anything, this sounds like a "truer" majority to me.
As for why anyone would WANT to do only use a portion of the range? For two-candidate races, it probably doesn't make much sense. But some voters may want to do it for the sake of honesty, or they might not be confident, or are willing to cooperate and deffer to others.
If I honestly dislike all candidates, you shouldn't be able to change my vote. If all options are bad, I want the final results to show that.
If you dislike all candidates equally, it won't effect the results if you score them all at minimum score or maximum score. You're scoring candidates relative to each other, not some arbitrary universal standard of quality (such as how Amazon reviews are scored)
No, it won't effect the ranking, but it will effect the final scores, and I might want exactly that; no one should have the right to change my vote if they disagree with me on that. Secondly, I don't have to dislike them all equally, I can dislike them _almost_ equally (depending on how fine a scale one uses here). E.g. if my votes are 0/10 and 1/10, then I bloody well meant to give 1/10. What gives anyone the right to change my 1/10 into a 10/10? If that's what I wanted, that's how I would have voted.
You would be within your right to do that, but I'd still like to note two things:
1. The score given to a candidate only makes sense within the context of that particular election. If in one election, candidate A wins with a score of 6.5/10, and in another election, candidate B wins with a score of 7.1/10, that doesn't mean that people generally liked candidate B more than candidate A overall. Similarly, a candidate that won with 7.1 in a hypothetical election might have lost with 7.5 in the same election had different/additional candidates been running.
2. If you like the 1/10 candidate better than the 0/10 candidate, why wouldn't you want the 1/10 candidate to have a higher chance of winning? 10/10 doesn't mean "Ideal candidate who I agree with on everything" nor does 0/10 mean "worst most evil candidate in the history of the world". 10/10 means "favorite of the options provided", 0/10 means "least favorite of the options provided"
I suggest a voting where the range include negative preference. Can you make a math analysis of this system?
I don't think the analysis would be any different. If blanks are kept abstentions, then they would function exactly the same.
If you don't use abstentions but blanks are treated as the minimum score, they'd be the same in that case as well.
If blanks are 0s, then they'd be a bit different.
I think this ignores the fact that people are likely to get pretty extreme on rankings. (if you look at patterns of ratings on a lot of real data, you get huge spikes of 1- and 5-star rankings, with much fewer 2-4.)
If you're talking about ratings outside of elections, part of that may be because people only bother giving any score at all if they already had strong feelings.
Anyways, if people only use the minimum and maximums, then it's equivalent to Approval Voting, which is still great.
It is true that this style of voting is always strategically optimum---in theory. In practice, it can require a lot of confidence that your information is extremely accurate. Otherwise, it can be safer to give at least a few partial scores.
Not really, and not really an issue. This isn't like rating products, it is fundamentally different. When people rate products, they don't buy and test every variant available before giving out ratings. They buy and rate just one. That's what inflates the rating scales.
With voting, it is different. You have an exhaustive comparative scale, because everyone is giving a rating on all elements of the same set of options. This is what makes the ratings comparable and what incentivizes differences. People hate giving the same amount of support to candidates who don't deserve it. That's one of the main criticisms approval voting receives. People are also observed to use the comparative rating scale to differentiate candidates they will accept, and they collapse everyone who is unacceptable to the minimum rating. This naturally makes the system slightly less susceptible to naive exaggeration of ratings.
In a 2 party system, yeah. Here in Germany we have 6 significant parties in parliament. I'd dive one of them 5 stars, two of them 4 stars and the rest 1 star. It's definitely scaled to give my favorite 5 stars and my least favorite 1 to make the most out of my vote. But It also makes it very clear that I do not only approve of that party.
doesn't matter. if you actually think this is a good argument, why would you then prefer plurality voting or approval voting where you are forced to give every party either max or min score?
@@opensocietyenjoyer I'd prefer to use a ranked-preference voting system
to fix the condorcet winner, look at the top two highest ranked candidates, and see who is most preferred over the other, and make them the winner (even if their ranking is lower)
A two step voting system first the well known system and the second page would have a range voting. Than you could decide weather you want to have a Condorcet winner or not.
Did i get something wrong?
Range Voting is not a condorcet winner, because the winner of the pairwise battles at condorcet winner is selected according to first past the post (or some voting system that with 2 candidates become first past the post). And this is a HUGE problem, because if first past the post was considered better than range we would be using it. So we use an voting system to decide the winner when more than 2 candidates are running for election, but when we want to decide the winner between just 2 we use something else for some reason.
Maybe we should create something called Self-Referential Condorcet Winner, where it would test that if someone that won all pairwise election using the voting system we are testing would be the winner of the election.
Also with even 2 candidates, range voting is better, two examples.
3 childrens pick an animal to be their pet. The choices is cat and spider. Kid 1 and 2 rate spider with 10 and cat with 8, kid 3 rates cat with 8 and he is agoraphobic and rate spider with 1. Here spider get 7 and cat get an 8 and the cat win, and under first past the post, the spider win.
Another example: 2 childrens pick an animal to be their pet, between cat and dog. Kid 1 say 9 to cat and 3 to dog, and kid 2, give 9 to both cat and dog. Cat get 9 and dog get a 6. Under first past the post, kid 2 would be forced to pick someone between cat and dog that he think is 100% equal, so there is 50% chance of a tie that shouldnt happen and will only happen because he can't pick both cat and dog.
Another good, informative video, Undefined Behavior.
Something I haven't heard score voting proponents tackle is how do you prevent the system from becoming an approval vote? Anything less than a top of bottom score is effectively tallied as a fractional vote. In your five star system, why would any voter give anything other then 0 or 5 starts to each candidate?
Thanks for the question! This is a really good one.
If voters all vote strategically, you are absolutely correct that range voting would devolve into approval voting. According to rangevoting.org (rangevoting.org/rangeVapp.html, unfortunately some of their links appear broken right now), experimentally we see that most people do not vote strategically. There's seems to be a psychological effect that leads people to want to be more truthful with range voting, since the perceived payoff of exaggerating appears small. Of course this isn't 100% proof that this is how things would pan out in the real world, but it's certainly interesting.
Would you agree that it is a flaw with the voting system if the opinions of honest voters are weighted less than the opinions of the dishonest ones?
Clay, up for continuing this discussion elsewhere? I am bkelly1984 on yahoo and reddit.
> It is a fallacy to asses voting methods based on subjective/arbitrary criteria like this.
Okay
> The voting systems analog for this is Bayesian Regret.
But Bayseian Regret is just another arbitrary criteria. On a 0 to 10 scale, I don't agree that five people getting a 6 rated candidate balances out someone stuck with a 0. I'm not even sure a single-scale utility score for all candidates is possible within a single voter.
> your naive intuitions about "fairness" turn out to be all wrong
I'm sure it seems that way when measuring with your yardstick but you can't just dismiss mine and claim yours as truth. We need to consider the criteria itself.
How about we move forward one of two ways? Either you try to convince me that a score voting system doesn't give honest voters fractional votes or I will try to convince you that the Bayesian Regret system is a flawed analysis of a voting system. Thoughts?
> This is a non-issue. A great deal of empirical data tells us that a significant number of voters will be honest...
Unless you can show me that dishonest voters will never swing an score vote it is very much an issue.
You lean on scorevoting.net for your sources. I appreciate the site but I think many of the pages do not speak to my point or have assumptions I do not agree with. For example, my concern isn't that people won't vote honestly but that people who do will not get as much say in the election outcome. It also suggests that people who don't register a party affiliation also won't score vote strategically which I think is specious.
Thanks for your thoughts. I am always interested in discussing these issues with interested parties.
>> I don't agree that five people getting a 6 rated candidate balances out someone stuck with a 0
> It is mathematically proven that you are wrong.
Great. You can prove a utility function is true if you assume a utility function.
> I'm sorry to break it to you, but you're headed down a common naive line of reasoning that has been rigorously explored and debunked.
Yet all you can do is toss strawmen my way and insult me. We're done here.
The biggest issue of this system is, that difference it is not clear what 1 start means. 3 is I do not care or I will be ok with this candidate? 4 means I will be really satisfy or a little bit.
Also when we should vote for more parties, it is hard to express differencess between them in this scale (but it is much easiler then in voting ranking system).
I have a question: Why is the average score ranked, and not the total score?
You could use total score, but then you couldn't have abstentions/no opinions.
Which is again a philosophical question: Should someone that nobody's heard of win if they have a high average score?
There are ways to penalize unknown candidates. He even went over one in this video. By starting all candidates with, say, 1000 scores of '0' to start with, if someone gets nothing but a single max score (say, 5) instead of having an average of (total score divided by non-abstentions) = 5/1 = 5 (the max), it'd be 5/1001 ~ 0.005 --- hopelessly low.
There are other ways too, like counting abstentions as scores who's value depends on the raw average and the portion of non-abstentions (opinions).
abstention_value = raw_average * opinion%
Abstentions are always worth less than the raw average, so they can never help. But the fewer there are, the less the difference is, and the less they hurt.
This way you don't need to pick an arbitrary amount of 0s to give to everyone, which they would presumably have to agree on (and the most powerful ones could try to give themselves an advantage).
He said everyone could start at zero, so that's unlikely if the most people give a zero.
I'm confused. You've outlined IIA in the way that is irrelevant to Arrow's theorem. IIA for Arrow's theorem says that the choice between A and B ought to depend solely on the pairwise preferences between A and B in the profile, not on where other alternatives C,D, etc. fall relative to them.
And you seem to be assuming that Arrow's framework requires voters to 'force a ranking' (force a strict preference). But it doesn't, hence his 'weak preference relation' which incorporates both strict preference and indifference.
well explained
This system is a type of approval voting. While better than First-past-the-post (what most of the USA uses), it's susceptible to *strategic voting,* where you rank your favourite candidate with 5-stars, and then give everyone else 1-star. If everyone votes like that, Range voting will basically turn into First-past-the-post.
Ultimately Ranked voting (instant run-off) is a better method
This is a "bullet voting" argument, and I'm pretty sure it's not quite right. Strategic Range does involve giving only maxes and mins, but not necessarily just one max. Imagine if your "true favorite" is otherwise unpopular even in a Range election, while you have a close 2nd who's a real front-runner. Giving only your favorite any points is once again throwing your vote away, so you should definitely also give some points to your close 2nd. But you can still also give points to your true favorite, so their overall amount of support doesn't appear artificially lower in the final results.
IMO Range behaves well with strategy, while I'm not very fond of IRV. It's basically just FPTP but automatically strategic on the voter's behalf using a simple heuristic -- if your current top choice is last place, it goes through another round as if you strategically voted for someone who was more of a front-runner instead, repeat. So it's less that it's "resistant to strategy" and more that it's "nearly maximally strategic by default", but we don't think of it that way because the method does it itself instead of relying on the voter to do it. IMO the only reason it's not even worse is because some methods arguably _benefit_ from some strategy, including FPTP.
(Also I'd probably say that Approval is a type of Range/Score voting, rather than the other way around.)
(Yes, I posted almost the same thing on another comment just a little before this one, the OPs are pretty similar)
The Dems would win forever because they would definitely give 0 star to all candidates but theirs
How does this compare with ranked pairs (i.e. the Tideman method)?
There's one problem with this analysis. It assumes everyone is voting honestly. Even though there's never an incentive to rank your favorite candidate lower, there can be, and typically are, incentives to rate others lower. The real question becomes how altruistic the average person is, which is up for debate. In the best case, range voting maximizes overall pleasure. In the worst case, where everyone acts 100% selfishly, the system is no different from plurality voting (you get only one vote). The one benefit of ranked choice is that it has less issues with incentivizing lying (though it's by no means impervious to the problem).
I'm not saying that range voting is bad, I'm just saying it's not the end all be all, even if maximizing overall pleasure, which range theoretically excels at, is your priority.
This is a "bullet voting" argument, and I'm pretty sure it's not quite right. Strategic Range does involve giving only maxes and mins, but not necessarily just one max. Imagine if your "true favorite" is otherwise rather unpopular even in a Range election, while you a close 2nd who's a real frontrunner. Giving only your favorite any points is once again throwing your vote away, so you should definitely also give some points to your close 2nd. But you can still also give points to your true favorite, so their overall amount of support doesn't appear artificially lower in the final results.
IMO Range behaves well with strategy. If by ranked choice you mean instant runoff, then I'm torn on how much it incentivizes honesty, or how much we should care. It's basically just FPTP but automatically strategic on the voter's behalf using a simple heuristic -- if your current top choice is last place, it goes through another round as if you strategically voted for someone who was more of a frontrunner instead, repeat. So it's less that it's "resistant to strategy" and more that it's "nearly maximally strategic by default", but we don't think of it that way because the method does it itself instead of relying on the voter to do it. IMO the only reason it's not even worse is because FPTP is one of the few systems which sort of benefits from strategy a bit.
@@MustSeto Imagine everyone is 100% selfish.
A has a rating of 4/5
B has a rating of 3.9/5
C has a rating of 3.2/5
If everyone votes honestly, A wins. But there will be plenty of people who would've preferred B, but didn't mind A, so gave, for example, B 4/5 and A 3/5. If some of those voters rank A a bit lower and/or B a bit higher than their true preferences, they can make B win. They're happy, B is happy, but everyone who preferred A is now upset. This is misrepresenting voters' preferences and incentivizes lying, which breaks the whole system.
That being said, how much this would effect real world elections is uncertain, and perhaps range voting is actually the best option. I'm simply pointing out a potential downside of the system that this video doesn't mention, which, in my opinion, is a bit disingenuous. I'm not saying range is objectively not a good system.
@@Owen_loves_Butters By all means, point out anything shortcomings you see, or anything you think could be good to know
Yes, as mentioned, theoretical maximum strategy does involve giving only maximum and minimum scores. And in some cases that _can_ mean only giving one max, such as possibly in the case you mention here. But this is not _always_ strategic, for every voter in every election, so it is not "no different from plurality voting" even in the worst case, and I'm not sure this qualifies as "breaking the whole system"
Also, IIRC being "maximally strategic" like this is only guaranteed to be safe if you have enough information. But if there's enough uncertainty about precisely how other people will vote, then the most strategic, selfish thing to do can involve reducing the risk of greater evils by giving compromises partial scores. Though not necessarily the _exact_ scores you would give if you were 100% honest
@@Owen_loves_Butters The problem with that argument is that it assumes only the voters of one candidate vote strategically. If everyone votes strategically, the voters of A will in turn rank B a bit lower, thus restoring the original order.
@@aliveoutside5608 But you see how this is a never ending cycle?
Wouldn't non-condorcet winner creates strategic voting?
In this case, the Charmander voters (who are in the majority) are encouraged to give Squirtle 1 star. So we back to square one?
Score Then Automatic Runoff (STAR) Voting. You give each candidate a score of 0 to 5. The two of the top scores go an instant runoff round. If you voted higher score than the other, then your vote is counted for that candidate. If your score is equal, then you have abstained. The one with the most wins. This far as I can tell helps with the last problem. It discourages strategic voting. It is most accurate in reflecting votes, -and it does not cause a favorite betrayal- edit: STAR does fail favorite betrayal and later-no-harm to minimize the problems with both and spoiler effect. Explained here:
www.starvoting.us/farewell_to_pass_fail
STAR does reintroduce favorite betrayal. If your favorite would lose in the runoff when a compromise would win, then it's in your interest to put your compromise above your favorite. And STAR is basically similar to (forced, evenly distributed) strategic Score. It's mostly only better than Score when there's one-sided strategy (Score's worst-case). So STAR has a better worst-case and worse best-case. One-sided strategy like that seems pretty unlikely to me, and IMO even it's worst-case isn't that bad compared to other methods, so I think there wouldn't be much risk in at least trying plain score first. But it's nice to know we have a good backup in case it really does have problems in practice.
@@MustSeto
Thank you. I corrected my statement.
Ms. Liu, I am very impressed with your video. Would you be interested in collaborating on one?
I get it, your favorite is the fire dragon...why do so many people downplay Bulba? ):
I'd like to see one done on instant runoff voting as well! Very informative
IRV sucks
Here are some links to explanations of how and why IRV sucks.
wp.me/p23U97-dD
Here is a description of a voting system that I propose as having the perceived advantages of IRV and the real advantages of Range. wp.me/p23U97-eZ
@@darbyl3872 Made this comment 3 years ago and have looked much deeper into this and I firmly believe approval voting is the way to go. Cheapest to implement and better results than instant runoff
If you truly want 3rd parties, then what's needed is some form of multi-member districts via proportional representation. Any single-menber system always favours the two-largest parties.
I'd love this if the legislature themselves also used a better method internally
why not make a range vote and then count the total amount of stars for each candidate instead of looking at the average ranking?
Whats the difference between this and ranked voted?
To counter a unknown winner, maybe you have a first round where people give a vote (yes vote) to a max of 9 votes to candidates (can give yes to 0) if wanted.
The best 9 most voted goes to the second round, and there a huge amount of time between first and second round to make less known candidates more known.
At second round, everyone must give a score between 1 and 10 to each candidate (not giving score to some candidate make your votes not valid/count)
The one with best average win.
The number 9 come from the fact you want to have a not so huge amount of candidates so everyone can know and also be able to be able to be FORCED to vote at ALL them at second round. But the number is not so low that it would basically turn this into something that is mostly approval system. The number 9 was selected because its the amount of sides at political compass, Left/Authoritarian, Left/Center, Left/Libertarian, Center/Authoritarian, Center/Center, Center/Libertarian, Right/Authoritarian, Right/Center, Right/Libertarian.
This method also deal with countries like usa that have an extreme amount of candidates, almost all voting system have problems with alot of candidates, the few ones that "don't" are First Past The Post, Anti-First Past The Post (vote the worse, less voted candidate win), two round system, anti-two round, coombs version of two rounds system. And those systems sucks.
this could also lead to the same result coz if i have several options and i am worried that some other option might get elected and there is an alternativ which i dont like but i prefer than the guy i hate, i would rate more stars to him than i would normaly do, to get him on top!
The largest issue of no condorcet winner is having lukewarm candidates who hardly anyone feels strongly about thus leading to disparity amongst voters.
STAR Voting tho
I think the intrinsic value of a Condorcet winner is heavily undermined by its consistent delivery of 'two evils' to choose from.
the condorcet criterion is imo not something you need to fulfill. also, why did you call it range voting instead of score voting?
"Range" used to be the more common name. From the upload time I think this video was made around the time the transition was happening (around the same time I was getting interested, now that I think about it).
How about turning votes into a turnament style dilema, and we vote for every instance of preferance? I can already see some gaps but my smooth brain can't process it properly :)
This would sort of be a Condorcet method
I like range voting but STAR is a slight improvement that makes it a lot better
Let's say Charmander, and Squirtle are the 2 favorites, but I'm a 3rd party supporter (Bulbasaur). If I vote 0-3-5 respectively, then I'm essentially weakening my vote in a Charmander vs. Squirtle race (I'm only putting 3 points toward Squirtle), and if I don't think Bulbasaur has a good chance then I'd be incentivized to give 5 points to Squirtle and minimize Charmander's chances
STAR is "Score then Automatic Runoff", the score part is just like range voting, but then once all the averages are calculated, there's a automatic runoff between the top 2 where every vote only counts once. So my 0-3-5 Bulbasaur vote would count as 1 vote for Squirtle during the runoff and I'd have the same voting power as the Charmander and Squirtle voters
I like the idea of starters as candidate
I don't think range or approval voting passes IIA. If voters voted on each candidate independently, yeah, it would, but voters take into consideration all the other candidates.
Let's say that there are 3 candidates, A, B, and C. You might like C, not like B, and really hate A. If it were just A and B, you would likely give full support to B. Since C is in the race, you might give less support support to B which could help A win.
One of the things that bothers me about range is voters may give unknown candidates a middle score. You say that this experimentally doesn't happen. Do you mind sourcing that?
Btw, I think Range is one of the better voting systems. Ideally though I prefer a top two approval voting system. You would have a single non-partisan primary that everyone runs in and allow the two most approved candidates to advance to the general election regardless of party.
Hey, thanks for watching! These are great questions.
So IIA refers more to removing the options after the ballots have been cast. With the standard of method of voting in US elections, if you remove a third party, the votes that went to the third party need to go somewhere, and where they go could end up causing the winner to change. Similarly, votes that use ranked preferences will typically see situations where the result will change if a candidate is removed, and the votes are recounted. With range voting, each candidates average is independent of the others so removing one candidate won't change the overall ranking.
However, you are generally correct in that there are still ways for someone to vote tactically under range voting, and that were we to redo the entire election with one candidate removed, many people might change their scores for various candidates. I think that range voting strikes a nice balance between simplicity and the type of tactical voting that it's susceptible to.
Here's a link from rangevoting.org about the unknown lunatic wins scenario being unlikely (rangevoting.org/Averaging.html#unlikely). I think that the soft quorum as a safety valve is still important, just in case someone does manage to get a high average with very few actual voters.
I would also point out that a "score runoff" or "approval runoff" voting system would prevent a "lunatic wins" scenario as well as the scenario where the score voting picks a Condorcet loser.
Washington and California could become powerful testing grounds for ARV or SRV if they modify their blanket primary balloting systems.
Is it even possible to pass IIA with deterministic, no-dictator elections? If you're considering that voters change their votes depending on the available candidates, I don't see how anything could pass it.
+Skyval Ream, yes! As mentioned above, range voting does satisfy IIA.
However, if you add in the unanimity and restrict the ballot to a ranked ordering, then no it's impossible (which is the statement of Arrow's Theorem).
Sorry, my reply was to Mutex50's and their "version" of IIA (where voters get to re-cast their votes. I've seen others say the same thing). I know if you don't allow voters to "adjust" their scores then Range passes IIA (and I think this is the better definition). What I was getting at is that I don't know how useful Mutex's version of IIA is. As long as the voting method allows strategic voting, I don't think any method can pass that version of IIA. But that's more a problem with strategic voting, which no voting deterministic, no-dictator method can completely eliminate.
What about
Score voting
But multiple winners and each winners power in parliament =
winners average score /
(sum of average scores of all winners)
Isn't it condorset winners that *are* the problem?
Only when they’re divisive
cc on: "condor say" winner! 🙂
A hybrid of preference voting and two columns voting could fix it.
Unanimity explanation is confusing. You mentioned the converse rule of majority loser criterion or actually the weaker rule, unanimous loser criterion. That one is also sort of satisfied (additional simple condition) by Majority Judgement but not by plain Range Voting as described. This mode of failure is quite hard to achieve in practice. Crucial difference is the use of median rather than average. The fix to give 0 stars is wrong as it prevents explaining anti-preferences. The correct fix is to give a median score (impossible to know beforehand) or a fixed middle score. Presumably extending the scale to say -3 through +3. It does not actually enhance any of tactical votes and helps with honesty.
If one of the goals is to keep voters honest...what's to keep literally everyone from rating their top choice 5 stars and everyone else 1 star to specifically avoid giving a higher score to a more universally like candidate?
Unless their favorite is already very strong, this would be throwing their vote away. It's worth it to support a more electable compromise.
@@MustSeto but that's the problem with first past the post to begin with...people not being able to vote their actual preference out of concern of a less-liked candidate winning.
@@wolfpackjew In FPTP, supporting a front-runner means not supporting anyone else, so candidates "steal" votes from each other (vote splitting). In Approval you can continue to support your favorite.
In general, approving of more candidates decreases the chances of any specific one winning, but increases the chances of at least one winning against those you like less. In the extreme this becomes the opposite of bullet voting: an anti-vote, where you approve of _everyone except_ the one you hate most, which makes just as much sense as bullet voting. Everything between can make sense too, e.g. voting for everyone from your favorite faction and every sufficiently similar or more electable compromise factions.
@@MustSeto That still doesn't prevent strategic voting. I'm not saying that's a deal breaker for any system, but it's disingenuous to try to sell it this way.
@@wolfpackjew Sure! I don't think anyone is trying to claim this system is completely immune to strategic voting or anything. In fact a different theorem, Gibbard's Theorem, says that's outright impossible for any system whatsoever. But approving just your favorite isn't strategic in general
Range voting is perfect in theory, but in practice it has too many problems:
1) to satisfy the independence of irrelevant alternatives the range must be absolute but this is not the case:
If I vote like this (range 0-10): A-0, B-5, C-10
removing candidate A, my vote would have been like this: B-0, C-10 because with only those two candidates, I want to favor maximum C and minimum B.
2) tactical votes because an unknown party is better than a hated party, therefore a person who hates A will give a non-zero score, e.g. 5, to all unknown candidates and 10 to all favorite candidates.
Giving points to unknown parties to disadvantage A is a huge problem.
3) the votes don't all have the same value, that is, there may be votes that affect the results more than others.
4) equal interests can be represented differently, depending on how a person defines 0 and 10 in the range.
1) IIA is not well-defined for rating-based systems. You could, very reasonably I think, define it in such a way that your example does not violate it, because they decreased the score for B, increasing its distance from C, and in many cases it makes complete sense for such a change to change the winner. This form of IIA still works with Arrow's Theorem. Of course, this isn't ideal behavior whatever you call it. Though a different theorem basically proves that no system can be strategy-proof (Gibbard's theorem), and IIRC it doesn't depend on IIA.
2) Seems a bit speculative, and personal. I would probably agree with you that an unknown could be worse than the hated party and would probably give them minimum points (or close to it). And I expect most people who agree with us to do roughly the same---in which case there is no problem. But others could disagree, and honestly believe an unknown candidate would likely be better than the most hated party. That's their perogative, and heck, maybe they could have a point.
3) I think you're talking about voter equality? Like If a voter doesn't use the max score for any candidate? If this ended up relatively common & unintentional, I could understand the argument of this being a problem (and there are modifications like STAR which alleviate it) but I don't think it's an issue in general. Someone could do it intentionally, sort of like a partial abstention. OTOH, there is a sense in which systems like this are more equal than FPTP and RCV/IRV. Any vote can be canceled about by one other person's vote: Just invert each candidate's rating. FPTP can't do this, which is basically why it has the spoiler effect.
4) I don't think this is much of a problem. For one, this applies to pretty much any system. Two people who vote for the same candidate in FPTP could feel differently about them, even if they are honest votes. Same for ranks. Ultimately, the "meaning" of any system is defined by its rules. With Score, you increase the chances of candidates you give higher scores to, which decreases the chances of candidates you give lower scores to. How a voter chooses to do this depends on how they feel about the candidates.
range voting is the same as Arrow Theorem minus IIA. 3 candidates = range of 3, 5 candidates = range of 5 etc. IIA is too strong. Plus Arrow's dictator is morally the same as the +1 vote in a 50% + 1 simple majority.
I guess we're going through some of the same videos.
I don't understand what you mean about range being "the same as Arrow Theorem minus IIA".
The +1 vote in a 50%+1 majority not is "morally" the same as Arrow's dictator. The +1 vote only gets to decide when another 50% agree with them. Arrow's dictator gets to decide even if 0% of the others agree with them.
But even in score voting it would still be best if I am not completely honest, I would sure give my favorite condidate a top qcore but I would want to give some candidates a score that is way less than the one I believe they deserve just so my favorite candidate can have a better chance
That's only safe if your honest favorite is already strong, otherwise it's throwing your vote away. If duopoly is ended, and there's a good pool of candidates, some which are fairly similar to each other, than it starts making sense to support a few of them.
@@MustSeto actually I thought about it more and it seems to me that you are correct, this scenario could only happen if there are only two candidates woch kinda defeats the purpose of why the method was proposed. However when I told my brother about this he sqid to me that this could facilitate fraud because those who count would easily change the papers where as with a one vote system if you try to modify anything the vote is immediately canceled so that could be a problem.
@@destroctiveblade843 If it was implemented exactly as shown in the video, you'd you be right, you could "add more stars" to candidates. But in the places that I've seen actually propose it, it doesn't work that way, You just fill in one bubble. If you want to give them a 3, then you fill in the "3", etc. No candidate should have multiple bubbles filled in. You could arguably permit having none filled it for a candidate, so that you don't have to spoil their entire ballot, but should probably at least "encourage" filling in a "0", to prevent fraud.
@@MustSeto makes alot of sense
What if parties were not important? - What if 'Dictatorship' were handled in a different way to arrive at the end result in a way that is different from a single-vote?
You give majority rule as if it is better than utilitarian rule.... Utilitarian is clearly better.
Honestly I don't like Star voting. Voting candidates that you like other than your favourite high could mean that your vote harms your favourite candidate. This isn't possible under STV.
STAR voting resists it better than Range voting. And it is possible under STV, although it's a multi-winner method with different applications than Range/STAR anyways
How about Balanced Elimination Voting? wp.me/p23U97-eZ I designed it to address a perception (correct?) on the part of IRV advocates that in Range Voting, you can't provide any support to a compromise candidate over the worst (in your opinion) candidates without sacrificing some of your political power to support your favorite candidate should that candidate have a chance to win.
Condorcet Ranked Pairs is better and still simple enough.
The issue with Range Voting is that you are incentivized to vote contrary to your actual preferences depending on how you think others will vote. Lets say you have a candidates A, B, and C. You think A=5, B=3, and C=1. But if you think other people like B and it's going to be a close race between A and B (and that C won't have a shot), then you will want to rank B=1 to maximize the chances of your favorite A. Or, if you think all 3 have a chance of winning, you might want to rate B=5 to minimize the chances of your least favorite C.
Not that much of a problem in practice, considering opinion polls and other dynamics. But it's important to highlight you can ALWAYS fully support your favorite in score voting. The only way to completely address that type of problem you mention is to never allow your vote to support someone other than your favorite unless your favorite can't win. But ALL reasonably simple voting systems CANNOT do that without breaking the "you can always support your favorite" feature.
So which is more important: 1) Always be able to support your favorite honestly, or 2) Always being able to support other candidates beyond your favorite
You cannot have both. I think most people really want 1, and with time, score voting will allow more and more good candidates and parties to thrive, which means supporting another candidate and having them win over your favorite isn't going to be the end of the world to you.
In practice, in score voting, at worst scared voters will lie about indifference. People would play it safe and rate A and B max.
In all ranked systems, including Ranked Pairs, there will be many situations in which it is advantageous to lie, and that lie means supporting A over B when you really mean B over A. The lie has you flipping your preferences. This is a much bigger lie than lying about indifference.
There is a sort of latter-no-harm type violation, where you can rank someone as a 3, and your preferred as a 5, and your 3 could be the deciding factor in your second choice's victory. It seems like it will behave exactly like FPTP in the following way -> Voters know that not giving a rating/rating zero hurts the average of their non-preferred candidates. To increase their preferred candidate's chances, they rank their preferred candidate 5 and don't rank/give a 0 to every other candidate. Everyone does this because of the prisoner's dilemma. Is there experimental evidence that people give their honest preferences without strategic voting on a range vote even when there is this mathematical reason not to?
That's only strategic when you preferred is already very strong. Otherwise it's wasting your vote, just like in FPTP.
My understanding is that being maximally strategic under Score goes like this:
Before you vote, give honest scores to every candidate, and determine each's probability of winning.
Multiply each candidate's score by their probability of winning, and add the results to get the expected value of the election.
When you vote, give a max score to everyone who's honest score was higher than the expected value, and a minimum score to everyone else.
Even this only makes sense when you have very good information (can accurately assess probabilities). Otherwise you could give a max score to someone you shouldn't, or give no support to someone you should have. In these cases, it can be safer to give some more honest, partial scores.
Also it's not quite the prisoner's dilemma, it's a related but different dilemma called the chicken dilemma.
I can see that makes sense, but when I say "Voters know...", I don't mean maximally strategic voters or polisci major voters or game theoretician voters. I just mean a certain number of typical voters, who don't think about voting much except for about a couple weeks/days before the election, but think about it enough to try "bullet voting". I think my critique is pretty limited, because it's doubtful that EVERYONE would do that, but it seems like it could be a concern if a significant number did.
Thank you for the correction on the chicken dilemma, that's a good one to know!
Giving a zero to your second favorite candidate increases the chance of all other candidates compared to your second choice. This means that your first choice is more likely to win, but also that all other candidates which you prefer less than your second choice are more likely to win.
It's a risk-reward-consideration which every voter has to make. Do you want to maximize the chances of your favorite candidate winning or minimize the chances of your least favorite candidate winning?
4:14 obviously not voting on a candidate should equal 2.5/5. Why in god's name would a no vote be counted as 5/5?
The real problem with range voting appears to me to be that if I dislike the Tories (I'm a Brit) then I might exaggerate my hatred for them, making it seem like I hate them more than Hitler; just to force them out the game. However, that would be my personal hangover from the horrible First Past The Post (plurality) system with strategic voting, because with range voting, maybe there is a tiny party I actually hate more than the Tories, who I never thought would have a chance of getting in, but now they actually could if I do weird strategic shit
The no votes weren't counted as 5/5s, they just didn't effect the average. The average is TotalScore/TotalVotes, and there was only one vote since abstentions aren't counted. Most proposals I know generally just use TotalScore, which would be like treating abstentions as 0
I've always liked range voting better than ranked choice voting, but both would benefit from the option to represent direct opposition to the election of a candidate, making them balanced range voting and balanced ranked choice voting respectively. To better understand this, consider the last example given in this video but imagine that nobody really wanted either of those two but were instead more afraid that one of those two would win than the other, but scared of them both. Now, add in a candidate who is wanted by 10% of the people and unknown to the other 90% of the people. If the 90% of the people who don't know anything about this new candidate think it would be better for an unknown candidate to win than either of the two that they're afraid of, but believe that the candidate is so unknown as to not have any reasonable chance of winning, and do not honestly SUPPORT the unknown candidate, then most of those people will probably give that candidate a vote as close to zero as the system allows, yet still vote FOR a candidate they are afraid of, in the hopes of preventing the election of the candidate they are terrified of. Since the 10% who adore this third candidate aren't enough to outnumber these fear-based votes, the "second worst" candidate is likely to win. Add in the ability to cast a negative vote, and such fear or dissent or disapproval can be represented and counted much more accurately, resulting in a better outcome.
Let's talk about MAL (my anime list) rating system.
3.1% of the people who voted for Gintama rated 1 star on purpose.
That's because current top anime was FMA at that time (now it's Kaguya-sama love is war s3) and they didn't want Gintama to be the winner.
It's not like they hate Gintama, but they love FMA more than Gintama and they thought FMA deserves the first spot.
And I think if all people rated honestly, Gintama would have been the winner.
Isn't the biggest problem with range voting that voters can still get a benefit from voting dishonestly, by giving Squirtle 1 star in the last example?
Also, can you compare this with STV?
I'm not found of the star method because obviously most voters will just give 5 stars to their candidates and 0 star to the rest. Ranked voting seems the best to me.
Eh, IIRC that's theoretically the optimal strategy if you have good enough information, but with more uncertainty in the details of how other people will vote I think it starts to make sense to give some partial scores. Anyways I'm not sure this would harm the method's overall performance much, Approval Voting is supposed to be pretty good as well.
Plus, there's a method actually _named_ "STAR" (Score Then Automatic Runoff) which modifies plain Score slightly to try to encourage differentiating compromises a bit. First there's a plain Score step to find the top _two_ (instead of just electing first place), then whoever between those two is scored higher on the most ballots wins, even if they were 2nd place. This is using the same ballots as before; there's no separate round where you show up again or anything.
So whether you say A=5, B=4 (difference of 1 point) or A=4, B=0 (difference of 4 points) that's one point to A in the 2nd round. But if you say A=5, B=5 then neither gets anything. So if A and B are competitive you're better off docking B at least one point, unless you're truly very indifferent between the two. It doesn't hurt B much against other candidates in the first part, but if it comes down to it, it helps A a lot in the end.
Sadly this video only tells half of the story.
Independence of irrelevant alternatives is only true if voters are forced to vote honestly (which can't be done). Otherwise, additional candidates will affect how voters give ratings to the already existing candidates because of strategic voting.
Let's illustrate an example. Say that there are two candidates, Charmander and Squirtle, and you like one for 5 stars and the other for 4. However, as seen at 5:00, voting 4 stars for Squirtle will give Squirtle a better chance of beating Charmander than voting 1 star for Squirtle. In general, there will never be a strategic reason to vote less than 5 stars for your favourite and more than 1 star for your least favourite candidate.
Now let's add a third candidate, Bulbasaur. Say you like Bulbasaur for 2 stars, i.e. less than Squirtle. As mentioned before, giving Bulbasaur 1 star is only strategic, however, Squirtle can now get anywhere from 1 to 5 stars, and I'm sure there are scenarios where you want to give Squirtle more than 1 star. Thus, adding Bulbasaur has impacted how you vote for Squirtle, potentially changing the result between Squirtle and Charmander.
Favourite betrayal is also only avoided to an extent. While it is true that you will always be encouraged to give your most favourite candidate 5 stars, your least favourite candidate 1 star, and give all candidates an honest ranking, the precise scores for the candidates in-between are up to strategy. You might betray your second favourite by giving him 2 stars just so your favourite has a higher chance of winning, even though there are other candidates you like far less than your second favourite.
It's easy to imagine a situation where polls and media would have a lot of manipulative power not just by influencing how much people like each candidate, but also by lying about how popular and dangerous certain candidates are, and letting voters come to the wrong conclusion as to how strategically vote.
tl:dr The problem is not that people will vote honestly and that that will betray the condorcet winner, but that they will NOT vote honestly because they get better results from voting strategically.
5:00 This example is not best, because 60 per cent of people never gave a blue 4 star. Their best interest is to give blue minimal numbers of stars.
5:10 - yep, Squirtle for the win. We don't want the population to be divided
5:50 - exactly!
Problem is that, 60 per cent of people never gave a blue 4 star. Their best interest is to give blue minimal numbers of stars.
People don't spend nearly as much thought on making an accurate low star rating
I prefer this system en.wikipedia.org/wiki/Majoritarian_representation
because it's not a matter if you put 1 star or 2 ; the problem is how many you have above 2.5 stars.
But yes, we need to change our voting system
C:
Star voting fixes the flaw in this. Support star voting.
This video has never been more relevant
In the last example, people who like Charmander would never give Squirtle 4 stars, because they know it has a high approval among other voters and then their favorite candidate (Charmander) wouldn't win. IMO in practice ranged voting is actually more messy and worse than what we have right now.
If voters are strategic then in a two-candidate election Score is the same as Plurality/what we have right now. This is already sort of an improvement over most methods because they're the same as Plurality even if voters aren't strategic.
If there are more candidates then it can be strategic to give multiple candidates points. Technically with perfectly strategic Score (with perfect information) decays into (Strategic) Approval, only giving the minimum and maximum score, though not necessarily just one maximum. In practice with less information it can be strategic to hedge your bets by giving some partial scores.
@@MustSeto you are correct, I even forgot I made this comment! hahahah
I kept reading the other day about this topic and indeed, Score voting would degenerate into Approval voting in that perfect-information strategic scenario.
Now I like Approval more than Plurality.
But I also like Approval more than Score voting, because it's more simple and easier to understand.
To make it even more simple, my fav option would be Approval voting with the choice of choosing 3 candidates max. This would make everything even easier to understand by citizens, and would still avoid favorite betrayal🥰
@@trincotrunco8932 Cool! Yeah voting methods is a real rabbit hole. It just keeps going...
I largely agree about Approval being simpler than Score, and I also think that having larger and larger score ranges gives diminishing returns pretty quickly, even in theory. That said it seems like a lot of people _really_ hate being limited to just two options, and having to chose between putting compromises with either favorites or their hated, so I've wondered if 3 (e.g. for/neutral/against) would be more popular while still being simple. Maybe increase that to 4 if having an exact middle causes issues... I think there's been research on ratings scales like this (usually for surveys and things) that suggests people are unreasonably drawn towards middle scores, and you get better data on average if there's no true middle. Not sure how we'll that'd apply to this, or if voter frustration would outweigh the benefits, though.
What do you mean about choosing 3 candidates max? You mean only 3 candidates can run, or each voter can only approve 3? I don't think I'd do either.
I could see having a "unified primary" election to narrow the field a bit before having a "final" election being useful to let voters focus research on a more manageable number of candidates, though I don't know if I'd limit that to 3. Maybe 5, or maybe the top 20%, or something. A big draw of alternative methods and avoiding the spoiler effect is moving away from duopoly and towards more varied and nuanced races with multiple candidates.
And I don't see the point of limiting a voter to just 3 approvals, especially if we do start getting a good spread of candidates.
Surely there is no favorite betrayal, but there is still strategic voting (second favorite betrayal, if you like). Let's say you want candidate C to win, are okay with candidate B and do not like candidate A. Then of course you will give C 5 stars and A 0 stars, but how much will you give B? If you're not too afraid of A winning, then it would be strategic to also give B 0 stars, since this increases the chance of your favorite candidate winning. But if you fear that C barely stands a chance and A is very likely to win, you'll also give B 5 stars to prevent A from winning.
In the end, the systems probably converges to approval voting and with a huge incentive for strategic voting. Surely this system might work in a beauty contest or for Netflix, where the voters really just want to find the better candidates/films, but in something as divisive as politics, incentive for strategic voting should be kept to a minimum.
Maximally strategic Score does become maximally strategic Approval. But maximally strategic Approval elects Condorcet winners whenever they exist. It's not just about how much people are "incentivised" to be strategic, it's about what the effects are. For example, with Score/Approval, strategy is usually about Compression, while with Ranked systems, it's about explicitly misordering some candidates. So even a maximally strategic Score/Approval ballot arguably still contains more accurate, honest information than a minimally strategic ranked ballot---maybe even a fully honest one. In fact considering true Condorcet systems are not immune to strategy even when there is a Condorcet winner, Score/Approval might elect Condorcet winners more often than true Condorcet systems---except that when there is more honesty, Score and Approval can elect someone even better. Especially since being maximally strategic under Score/Approval can be risky, while being at least semi-honest still gives you most of your power without as much risk. I know I wouldn't have a problem giving a compromise some points if I did actually like them reasonably well.
I contend that strategic voting is not harmful and there is no reason to discourage or avoid it, provided that the voting system gives everyone the same amount of power over the outcome.
I think the strategically best amount of exaggeration to give your compromise candidate depends on the popularity you estimate that your favorite candidate will have with the other voters.
I don't see how you can claim that strategic voting is harmless. In any scenario, it encourages voters to be dishonest, whereby the election results do not accurately reflect population wants. Not everyone is a voting theorist, so a lot of folks will vote honestly, whereby their vote has much less power compared to those who vote strategically (i.e. in this scenario only the appropriate mix of 5's and 0's). Campaigns will aim at explaining voters how they should vote (might e.g. encourage them to ditch all candidates but their favorite), possibly misinforming and confusing them, making them again angry at the 'needless' complexity of the system, and in any case taking a lot of time of the public debate, which should be spent on discussing actual policies, not voting tactics.
There's more arguments to be made. I can see why one can claim that strategic voting might be the lesser evil in choosing a voting system, but you just cannot reasonably deny that it is something which is in principal very undesirable. For me, in something as divisive as politics, it is a larger put-off than e.g. failing 'Independence of Irrelevant Alternatives'.
No system can be strategy-free (unless you allow randomness or something).
The concepts of honesty and dishonesty do not apply to voting in an election to political office. The election is not an opinion poll; it is a channel for the exercise of political power by the individuals who vote, to produce an aggregate decision. To call it honest or dishonest is like asking whether the position of the throttle cable on a car is honest. It can't be honest or dishonest because it is not communicating an alleged fact, as for example the fuel level meter is, or the odometer. Those can in fact be dishonest; the odometer can be rigged. But the throttle cable is communicating a _command_. An election is the same kind of communication. In fact it is a game. The voters command the tally to take them into account in a certain way, within the freedom of movement that the system allows them. A command can't be dishonest.
The question that matters is whether your command gets more sway than my command.
The problem is that imagine your candidate is approved by 60% and will kill the 40%, and the second candidate is slightly less approved by the first 60%, but are loved by all, the problem is that the first 60% has an incentive to rank the second voter horribly for a preferable outcome.
You’re assuming that the entirety of the 60% would vote 0 for the other candidate which is not true as far as real world studies go. 75% of people using range voting in studies vote using numbers that aren’t just the min or max. In fact the majority of range voters end up voting honestly in comparison to FPTP voting. With that being the case, the candidate liked by most will win.
rangevoting.org/HonStrat.html
@@115zombies935 We don't really know how strategic these people really are. From a game-theoretic perspective, perfectly selfish vote from everyone results in 40% of the population being targeted.
@@swordwaker7749 We do know that though. It’s been shown many many times that range voters tend heavily towards more honest votes rather than purely strategic ones.
@@115zombies935 Yes, but it depends on the circumstance. A population might start out honest and get more strategic over time when they don't get what they want. This depends on a lot of factors. We don't really know what would happen in a given real-world situation.
If I want Charmander to win and give him 5 votes, and give Squirtle 3 votes ... what prevents me from forcing myself to give Squirtle 1 vote in case those 2 additional votes of mine would have made Squirtle win over Charmander? Wouldn't everybody be forced to lie and vote 5 to a single candidate and 1 to everybody else?
while this is likely to happen when there are two candidates as in the example, i don't think it would happen if you had, say, 10 candidates
the fear that your five star loses is the same fear that your one star wins, so you rank someone else at 3. if you want.
We game systems.
Range voting gets gamed by parties saying, "Vote 5 for these guys" and "Vote 1 for these guys". If either party fails to play the game, they lose.
The best way to combat this is with approval voting. It produces the most acceptable alternative. It doesn't focus on finding the strong favorite of a small group. People often wonder why Netflix went away from 5 star voting, and why Google, UA-cam, Reddit, Twitter and Facebook never implemented it.
It SEEMS like a five star system would give you richer data, but it only does in specialized circumstances. In high stakes contests, the results are always gamed to maximize the outcome. This invalidates the utility of the granular system.
We should also have a minimum threshold for winning, such as 80%, and should require a quorum of 80% of the voters. (roughly). Other than that, approval voting for candidates is superior to plurality, ranked, or scored voting. Academics like to worry about pareto optimality and Condorcet winners. In the real world, approval voting produces the most acceptable alternative. It favors compromise candidates and solutions.
Nuf said :)
Okay... a little more. Point rating systems are great for when data is granular, and when a comment can be attached to the rating. In practice, they produce very close to the same score as an approval vote when aggregated and scaled. 80% up = 4.0/5.0 (4 stars out of 5).
Because Score/Range passes favorite betrayal, unlike other methods, voters don't have as much incentive to obey this. For one, there is never a reason not to give max points to your honest favorite, or indeed, if you give X points to candidate A, there's is not reason not to give at _least_ X points to every candidate you like better than A. So even if the front-runners are saying not to give some candidates any points, they don't have a leg to stand on. Especially since giving some points to a compromise candidate can reduce risk.
Also, I don't see how Approval would help with this. "Approve these guys, don't approve these guys". If anything, it just makes your criticism mandatory.
UA-cam used to have 5-star voting, and I think none of your examples are Approval, they're more like Score form -1 to +1.
And IIRC Netflix and UA-cam's reasons for switching to this system was because they were mostly only getting extreme scores anyways, but that makes sense given their context. "Voters" aren't required to score videos or posts or whatever, so they only go out of their way to do it when they strong feelings about it. But for an election, I don't think this is applicable.
Have you seen Warren Smith's Bayesian Regret simulations, or Jameson Quinn's Voter Satisfaction Efficiency simulations? Both put Range/Score above Approval in terms of the quality of the winner.
Yeah. UA-cam was 5 star. I guess Facebook, Google and Twitter are the approval voters, strictly speaking.
+1/-1 is a type of range voting that is close to approval voting, but you're right that they're not equal.
Broad score systems get gamed by giving max stars (points) to all the candidates you'd be okay with, and 0 to everyone else. In effect, if you play it intelligently, it turns into approval voting for the "players", who have a greater effect than the honest voters that give a real 0-5 point score.
As long as we don't use Ranked or Plurality voting, I'm okay with the others. All the scored voting methods are in my favorite class of voting, and all product roughly the same data, except when they are gamed. When they are gamed, approval voting doesn't take power from the voters and transfer it to the "players".
Thanks for the reply.
Excellent comment, btw.
It's true that if you have perfect information about an election, strategic Score is identical to strategic Approval. But with more uncertainty, giving slightly more honest scores is more strategic because it reduces risk. This is an advantage over Approval.
And Approval can still be gamed, by being more "selective" about who you approve, e.g. don't approve compromise candidates if someone you prefer has a better chance---which could prevent good compromises from winning. Smith's and Quinn's simulations suggest that when both Approval and Score are gamed, Score still does better on average, but Score is better yet when voters are more honest.
I do agree that either should be acceptable, and Plurality Voting or most Ranked methods (especially IRV) should definitely be avoided. Condorcet methods seem like they might be okay though, if you ignore their complexity.
And thanks for the compliment! It's always nice to have a good conversation about such an important, under-appreciated topic.
Jameson Quinn's simulations: electology.github.io/vse-sim/VSE/
it took pokemon to teach me this shit lmao