Open sandys opened 3 years ago
francois-durand
commented
3 years ago Hi, Please find attached an example of jupyter notebook showing how to do this. You have my authorization to do what you want with it (publish, modify, etc.) with no restriction. Best, François
sandys
commented
3 years ago this is mindblowingly helpful. I cant begin to thank you !
one question here: ultimately im trying to make a case of FPTP vs approval here. So what is the visualization that best reflects this ?
Caveat - this is a simple model since indian elections have typically 50+ candidates. So I'm not sure if the poisson method will work. In which case, if i had to get visualizations from your normal model for 50 candidates...what should i do ?
francois-durand
commented
3 years ago Hi,
Sorry for the late reply, I was in a deadline for submitting a paper and then I was on vacations. Here is an example using a probabilistic model known as "Von Mises-Fisher", where high concentration means that all voters agree, whereas concentration close to 0 means that voters are uniformly distributed. You see that Approval yields a better social welfare than Plurality. Note however that this analysis does not take strategic voting into account. Best,
sandys
commented
3 years ago hi @francois-durand thanks for this. Just to set context at what we are working on - we are introducing this in our local govt body to actually induce a change in voting system, so this has been super super useful for us. If successful, we will have started a change in the world's largest democracy.
We are EXTENSIVELY mapping out different kinds of simulations - because obviously politicians cant read code.
Your last line is something im struggling to figure out.
You see that Approval yields a better social welfare than Plurality. Note however that this analysis does not take strategic voting into account.
how do we model different strategic voting situations here ? I mean is it as straightforward as changing the concentrations here ? i thought of doing it that way...but i wasnt sure if that is the right way.
francois-durand
commented
3 years ago Hi,
Short answer: strategic voting is a very complicated issue, and scientists of the discipline of social choice are still struggling to address this question correctly.
Less short answer now... Disclaimer: I talk only about single-winner voting rules here. The question of multi-winner voting rules (e.g. to elect a parliament, for example) is even more complicated.
Consider a voting situation with fixed candidates and fixed preferences of the voters. From a theoretical point of view, it can be seen as a "game" for the voters, in the sense of game theory. Unfortunately, this game generally has a lot a Nash equilibria, and for each candidate, at least one equilibrium where she is the winner. So who will actually win, i.e. which equilibrium will actually happen? On the contrary, if you choose a stricter notion of equilibrium, like the strong Nash equilibrium, it does not always exist in a voting game, so it is not a good predicting notion either.
Furthermore, it is not even clear that this theoretical approach is fully justified. Experiments with actual human voters tend to show that the model of "strategic voter" (called "rational agent" in economics) is not totally accurate: a lot of people vote simply "sincerely" (without strategic considerations), or with a mix of sincerity and strategy.
From my point of view, a big question is: given a voting rule, is there a big difference of "voting power" between a non-informed sincere voter and a well-informed strategic one? I.e. to what extend does the strategic questions involves a violation of the "one person, one vote" principle?
By the way, the question of strategic voting raises a lot of other questions, like:
Despite all the scientific difficulties about all this, I think I can safely say that social choice theorists generally agree that the following voting rules have arguments in their favor:
From the point of view of immunity to strategic voting, IRV is clearly superior. Cf. for example my PhD thesis (https://hal.archives-ouvertes.fr/tel-01242440). It is in French, but you can look for example at Figure 9.5 (page 225). Here, IRV is called "VTI" and the y-axis represents the proportion of real-life situations where it is sensitive to strategic voting. "CVTI" and "SE" are variants of IRV/VTI, but you can ignore them. Another argument in favor of IRV is that it is used since several decades in many (mostly English-speaking) countries, without any dramatic issue.
Approval voting has the great advantage of being extremely simple to explain, count, etc. It is very sensitive to strategic voting, in the sense that, from a strategic point of view, it is important to choose wisely where you put the limit between the candidates that you approve and the ones you do not approve. But at least, this dilemma is quite easy to understand for everyone and, in some sense, all voters are at equality in front of this dilemma. However, to reach a informed decision, you need to rely on the polls, so it is better if you can ensure that polls will not be dishonest. Approval voting begins to be used in political elections, e.g. in Fargo (North Dakota).
Range Voting is rather worse in terms of strategic voting. In particular, roughly speaking, strategic voters who use only the extreme grades have more powers than more nuanced voters. Pro: in field experiments, voters appreciate to have the possibility to give a nuanced point of view. However, this voting rule has never been used in large-scale political elections, so we cannot predict what would happen once the voters get used to it. Will they end up using only extreme grades? No one can tell.
Finally, Condorcet voting rules have diverse vulnerability to strategic voting, but their main disadvantage is that they generally are quite complicated to explain. For this reason, they seem relatively ill-suited for a large political election.
I'm sorry if, after my answer, you stay with more questions than those you had at the beginning :-). But as a scientist, I would not do my job if I told you: ok, science has solved the question, the solution is [X]. This is unfortunately not true for voting systems. Most of us scientists agree on the fact that some rules are bad, such as Plurality=First-Past-The-Post, but there is not a voting rule that is superior to the others from all points of view.
If you want to talk about all this (e.g. by videoconference), do not hesitate to tell me. Best,
François
sandys
commented
3 years ago Hi Thanks for an awesome reply.
So here's the specific context for India.
Basically, it is either approval Voting or FPTP. There are no other choices that are viable here.
Part of what I'm trying (and I'm bothering you so much!) is model out approval vs fptp in strategic voting scenarios. Like the scenario I posted at the beginning of this GitHub issue is a real (however a very simple scenario).
In your answer before this, you said your code didnt model for strategic voting..but I need to model EXHAUSTIVELY all strategic voting scenarios and write them down.
Is there a way to do this ? Your code is the only real good resource I have found here.
On Thu, 14 Oct, 2021, 20:21 francois-durand, @.***> wrote:
Hi,
Short answer: strategic voting is a very complicated issue, and scientists of the discipline of social choice are still struggling to address this question correctly.
Less short answer now... Disclaimer: I talk only about single-winner voting rules here. The question of multi-winner voting rules (e.g. to elect a parliament, for example) is even more complicated.
Consider a voting situation with fixed candidates and fixed preferences of the voters. From a theoretical point of view, it can be seen as a "game" for the voters, in the sense of game theory. Unfortunately, this game generally has a lot a Nash equilibria, and for each candidate, at least one equilibrium where she is the winner. So who will actually win, i.e. which equilibrium will actually happen? On the contrary, if you choose a stricter notion of equilibrium, like the strong Nash equilibrium, it does not always exist in a voting game, so it is not a good predicting notion either.
Furthermore, it is not even clear that this theoretical approach is fully justified. Experiments with actual human voters tend to show that the model of "strategic voter" (called "rational agent" in economics) is not totally accurate: a lot of people vote simply "sincerely" (without strategic considerations), or with a mix of sincerity and strategy.
From my point of view, a big question is: given a voting rule, is there a big difference of "voting power" between a non-informed sincere voter and a well-informed strategic one? I.e. to what extend does the strategic questions involves a violation of the "one person, one vote" principle?
By the way, the question of strategic voting raises a lot of other questions, like:
- What power does it give to polling institutes? For example, in first-past-the-post, if a poll says that the main contest will be between candidates A and B, then strategic voters will vote either for A or B, which makes the prediction self-fulfilling. Of course, this gives a lot of illegitimate power to a polling institute that would be dishonest.
- Will the voters regret their ballot a posteriori, seeing that a different ballot would have defended their preferences better?
- Will the winning candidate have legitimacy, i.e. will he/she appear as a legitimate aggregation of the preferences?
- If the voting situation can lead to a "reasonable" result, for example a Condorcet winner, will the voters be able to reach this equilibrium?
- Is it difficult to compute a strategic ballot? Does it lead to unbalanced powers between educated and less educated people ?
- Do voters of similar interest need to communicate in order to compute their strategic ballots?
- Given the properties of the voting game, how will the political offer will evolve? Do candidates will move to the political center? Will only two big parties survive, or many small political parties?
- Etc.
Despite all the scientific difficulties about all this, I think I can safely say that social choice theorists generally agree that the following voting rules have arguments in their favor:
- Approval voting,
- Range voting with a short scale (like grades from 0 to 3, for example),
- Instant-Runoff Voting (IRV), which bears many other names depending on the country (I think it is used in India to elect the President, but I don't know how you call the voting rule),
- Condorcet voting rules.
From the point of view of immunity to strategic voting, IRV is clearly superior. Cf. for example my PhD thesis ( https://hal.archives-ouvertes.fr/tel-01242440). It is in French, but you can look for example at Figure 9.5 (page 225). Here, IRV is called "VTI" and the y-axis represents the proportion of real-life situations where it is sensitive to strategic voting. "CVTI" and "SE" are variants of IRV/VTI, but you can ignore them. Another argument in favor of IRV is that it is used since several decades in many (mostly English-speaking) countries, without any dramatic issue.
Approval voting has the great advantage of being extremely simple to explain, count, etc. It is very sensitive to strategic voting, in the sense that, from a strategic point of view, it is important to choose wisely where you put the limit between the candidates that you approve and the ones you do not approve. But at least, this dilemma is quite easy to understand for everyone and, in some sense, all voters are at equality in front of this dilemma. However, to reach a informed decision, you need to rely on the polls, so it is better if you can ensure that polls will not be dishonest. Approval voting begins to be used in political elections, e.g. in Fargo (North Dakota).
Range Voting is rather worse in terms of strategic voting. In particular, roughly speaking, strategic voters who use only the extreme grades have more powers than more nuanced voters. Pro: in field experiments, voters appreciate to have the possibility to give a nuanced point of view. However, this voting rule has never been used in large-scale political elections, so we cannot predict what would happen once the voters get used to it. Will they end up using only extreme grades? No one can tell.
Finally, Condorcet voting rules have diverse vulnerability to strategic voting, but their main disadvantage is that they generally are quite complicated to explain. For this reason, they seem relatively ill-suited for a large political election.
I'm sorry if, after my answer, you stay with more questions than those you had at the beginning :-). But as a scientist, I would not do my job if I told you: ok, science has solved the question, the solution is [X]. This is unfortunately not true for voting systems. Most of us scientists agree on the fact that some rules are bad, such as Plurality=First-Past-The-Post, but there is not a voting rule that is superior to the others from all points of view.
If you want to talk about all this (e.g. by videoconference), do not hesitate to tell me. Best,
François
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/francois-durand/svvamp/issues/4#issuecomment-943434200, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAASYU3ZDZ4HZE3DBWUVTZDUG3U5JANCNFSM5BYDFA4A .
francois-durand
commented
3 years ago Hi,
If you have essentially the choice between Plurality and Approval voting, I think that any social choice theorist will recommend to use Approval voting. In attachment, please find an example of a simple yet not-so-stupid model of strategic behavior, based on very limited information: voters do not have access to polls, but they can "guess" the Leader and the Runner-Up, i.e. the two candidates with most chances of winning (a priori). This can be by their experience of previous elections for example. Then Approval tends to elect "better" candidates than Plurality in terms of social welfare.
Best,
François
hi I'm posting from here - https://www.reddit.com/r/EndFPTP/comments/ozybgn/need_help_creating_a_simulation_of_approval/
I have had a couple of discussions at my local govt level. They want to see a simulation something on these lines:
100 total voters :
30 Group-1
20 Group-2
10 Group-3
20 Group-4
15 Group-5
5 Group-6
(these represents a typical ethnic/political distribution of voters in parts of India)
take 4 candidates A B C D
A is loved by Group 1,2,3 but hated by others
B is loved by Group 5 but hated by Group 1,2,3
C is loved by Group 1,2,3 but not disliked by others
D is not loved by any ...but not disliked by any either
then show how result will be under current FPTP and proposed Approval Voting.
Is this possible under svvamp ? I was not sure if it is possible to simulate entire demographics with a love/like/hate/neutral kind of preferences (which is very typical of approval) and compare the output to FPTP