-
Features
Evaluating voting methods
- Thread starter Blahface
- Start date
ronburgundy
Contributor
In the immediate short-term, range voting should produce the lowest average regret. However, over the longer term, the nature of a voting system heavily impacts the type of candidates that are put forth as voting options, which in turn impacts how people rank them, and together these interact to determine the type of candidates that wind up holding office. It appears these simulations account for none of this. They presume a pre-set slate of candidates and then test how different voting systems on the same set of candidates impacts short term regret related to that particular election. It wrongly assumes that the slate of candidates and the type of system are independent factors, but they are not.
I'm not saying that range voting impacts candidate options in a negative way that winds up increasing regret in the long run. Just that the model isn't capable of predicting long term impact, and it really cannot be modified to do so, since there is no scientifically validated theories about how a different political interests that shape who the candidates are would react to different voting systems. These kinds of statistical simulation models are only as valid as the already existing theories based upon actually observed behavioral data.
I'm not saying that range voting impacts candidate options in a negative way that winds up increasing regret in the long run. Just that the model isn't capable of predicting long term impact, and it really cannot be modified to do so, since there is no scientifically validated theories about how a different political interests that shape who the candidates are would react to different voting systems. These kinds of statistical simulation models are only as valid as the already existing theories based upon actually observed behavioral data.
lpetrich
Contributor
Maybe you could try to fill in that gap.
lpetrich
Contributor
Should the Victor Share the Spoils? - The Atlantic
The Atlantic article then got into redistricting and gerrymandering, and noted that multimember districts can make gerrymandering much less of a problem. Then this problem:
RangeVoting.org - Proportional Representation
Then about how common proportional representation is, noting Proportional Representation Systems
The Atlantic article then got into redistricting and gerrymandering, and noted that multimember districts can make gerrymandering much less of a problem. Then this problem:
RangeVoting.org - Proportional Representation
I recall from somewhat that that was directed at bloc-vote elections. in such multiseat elections, one votes for as many candidates as there are seats. It is as bad as first-past-the-post. There was a missed opportunity here, to allow proportional representation for multimember House districts.
fromderinside
Mazzie Daius
- Joined
- Oct 6, 2008
- Messages
- 15,945
- Basic Beliefs
- optimist
OK. Here's where the shit hits the fan. Voters usually don't bother to pick among candidates based on, say, candidate's values and voting record. Most come in with one candidate in mind or with an idea put in their heads by those they trust on what candidate is right. Most also come with notions of those they don't like most. There is little rationality in their process. They will do whatever they need to do to game the system to favor their choice. If this results in a 'best' result let us pray.
The form of play I just outlined is much worse than, say, a person committing the gambler's fallacy increasing her likelihood to go broke or break the voting system.
ronburgundy
Contributor
I doubt there is sufficient scientific knowledge base to do more than speculate exactly how the candidate options would change, but we know enough to know that they would almost certainly change in both number and type. The kinds of candidates that can win in a range system is more variable than the kind that can win in a first past the post system. Range systems allow candidates that are not anyone's favorite to still win. Also, powerful interests would also seek to add foil candidates to the field that wind up helping the candidate they want to win. For example, conservatives could (and likely would) field fake liberal candidates that their polling numbers show cannot win but would also soak up some points from the most threatening liberal candidate.
ronburgundy
Contributor
So, separate from the issue I already raised (that voting system impacts the number and type of candidates available to vote for), I have some other questions about the assumptions of the simulations.
The system must make many assumptions about how preferences are distributed, determined, co-constraining, influence post-outcome feelings, influence voting, how voting influences post-outcome feelings somewhat independent of preferences, and if and how all of these are moderated by the context of the voting system constraints themselves. The outcome of the system is completely determined by such assumptions. There is nothing magical about computer modeling. The models spit out and can only spit out whatever they were programmed to spit out. The assumptions are inherent in the logical code and equations written by the programmer. They are not themselves science. They are math and computer programming.
To be a viable scientific theory not only must each of these assumptions be grounded in and coherent with existing psychological science, but then its predicted outcomes must be tested against empirical behavioral data. Looking at the pretty graphs spit out by such programs deludes some people (often the non-scientist mathematicians and economist who write them) into thinking they are looking at empirical data that tells us something about the theories. It doesn't. It tells us only about the programmer themselves and what assumptions they made to produce those outcomes.
Lets take a brief look on some of these assumptions.
How are preferences for any given candidate are distributed in the population. Are they "normally" distributed with most people having no preference either way, and few having either a strong like or dislike? Or are that bi-modal, with most people having mod-to-strong like or dislike and few being neutral? Or is the distribution "flat" with all levels of preferences equally represented? Also, is the distribution of preferences the same for all candidates? It is highly plausible that the shapes of the preferences distributions among candidates themselves vary along several dimensions, from uni to bimodality, and centered around a neutral point to highly skewed in either direction, and somewhat flat to extremely peaked with most people sharing a similar perference for that candidate. Which of these sets of assumptions is closer to how preferences are actually distributed for candidates and how might the actual voting system impact the distribution of preferences?
Are the preferences of a particular voter for each candidate set independent from each other rather than co-constraining and correlated?
The simulations linked by Ipetrich suggest independent.
In simulations so far, all preferences have been random and independent".
Most relevant psychology would suggest that preferences for things are not independent because they are not determined by the inherent properties of that thing but by the particular context in which the evaluation is made and relative to other options. In fact, the very concept of a "preference" is inherently relativistic. Also, preferences likely co-constrain each other differently for different voting systems. In a single-vote system, voters have no motive to set relative preferences among all candidates, because all that matters for their voting is which single candidate is their favorite? In a range system, they must put every candidate into a relative rank. Thus, the system itself causally impacts how preferences are set and how the constrain each other.
In addition, what exactly is the algorythm the calculates outcome satisfaction based upon preferences?
Again, one of the links says:
"A voter's happiness with respect to the outcome of an election is equal to the voter's preference for the candidate that won the election."
That implies that only the value of the preference for the candidate who wins is the sole determinant of how voters feel about the outcome.
Like the other assumptions of the simulations, this one seems at best devoid of any basis in psychological science and more likely strongly refuted by much psychological science.
There are many alternatives to that assumption. For example, how much higher is my preference is for the winner relative to the other options? IF I love the winner but despise all others, then I am likely to be much happier with the outcome than if I love the winner but also like-to-love several others?
Again, not only are our preferences determined relativistically, but outcome satisfaction is likely determined by an interaction of the absolute level and relative level of preference for the outcome.
In addition, how does actually voting for a candidate impact outcome satisfaction? IOW, my preference will only partly determine my voting because the voting system constrains who, how and for how many I can vote for. Actions themselves are shown to impact emotion and preferences, not just the other way around. Thus, who I actually vote for is likely to impact outcome satisfaction, somewhat independent from and perhaps complexly interacting with the impact of my preferences on outcome satisfaction.
Again, type of voting system is itself something that likely has a causal impact upon preferences, how they are determined and co-constraining, how they impact outcome satisfaction, and voting actions themselves and thus their impact on satisfaction. Thus, it is likely that a separate set of algorithyms must be created for each type of system, each with psychologically plausible sets of assumptions for that context.
The programs in question do not even count as "simulations" if these various assumptions are not set based upon existing psychological data, which they don't appear to be.
They are predictions about what would happen in the fictional and implausible universe that the programmers have not very creatively imagined.
The system must make many assumptions about how preferences are distributed, determined, co-constraining, influence post-outcome feelings, influence voting, how voting influences post-outcome feelings somewhat independent of preferences, and if and how all of these are moderated by the context of the voting system constraints themselves. The outcome of the system is completely determined by such assumptions. There is nothing magical about computer modeling. The models spit out and can only spit out whatever they were programmed to spit out. The assumptions are inherent in the logical code and equations written by the programmer. They are not themselves science. They are math and computer programming.
To be a viable scientific theory not only must each of these assumptions be grounded in and coherent with existing psychological science, but then its predicted outcomes must be tested against empirical behavioral data. Looking at the pretty graphs spit out by such programs deludes some people (often the non-scientist mathematicians and economist who write them) into thinking they are looking at empirical data that tells us something about the theories. It doesn't. It tells us only about the programmer themselves and what assumptions they made to produce those outcomes.
Lets take a brief look on some of these assumptions.
How are preferences for any given candidate are distributed in the population. Are they "normally" distributed with most people having no preference either way, and few having either a strong like or dislike? Or are that bi-modal, with most people having mod-to-strong like or dislike and few being neutral? Or is the distribution "flat" with all levels of preferences equally represented? Also, is the distribution of preferences the same for all candidates? It is highly plausible that the shapes of the preferences distributions among candidates themselves vary along several dimensions, from uni to bimodality, and centered around a neutral point to highly skewed in either direction, and somewhat flat to extremely peaked with most people sharing a similar perference for that candidate. Which of these sets of assumptions is closer to how preferences are actually distributed for candidates and how might the actual voting system impact the distribution of preferences?
Are the preferences of a particular voter for each candidate set independent from each other rather than co-constraining and correlated?
The simulations linked by Ipetrich suggest independent.
In simulations so far, all preferences have been random and independent".
Most relevant psychology would suggest that preferences for things are not independent because they are not determined by the inherent properties of that thing but by the particular context in which the evaluation is made and relative to other options. In fact, the very concept of a "preference" is inherently relativistic. Also, preferences likely co-constrain each other differently for different voting systems. In a single-vote system, voters have no motive to set relative preferences among all candidates, because all that matters for their voting is which single candidate is their favorite? In a range system, they must put every candidate into a relative rank. Thus, the system itself causally impacts how preferences are set and how the constrain each other.
In addition, what exactly is the algorythm the calculates outcome satisfaction based upon preferences?
Again, one of the links says:
"A voter's happiness with respect to the outcome of an election is equal to the voter's preference for the candidate that won the election."
That implies that only the value of the preference for the candidate who wins is the sole determinant of how voters feel about the outcome.
Like the other assumptions of the simulations, this one seems at best devoid of any basis in psychological science and more likely strongly refuted by much psychological science.
There are many alternatives to that assumption. For example, how much higher is my preference is for the winner relative to the other options? IF I love the winner but despise all others, then I am likely to be much happier with the outcome than if I love the winner but also like-to-love several others?
Again, not only are our preferences determined relativistically, but outcome satisfaction is likely determined by an interaction of the absolute level and relative level of preference for the outcome.
In addition, how does actually voting for a candidate impact outcome satisfaction? IOW, my preference will only partly determine my voting because the voting system constrains who, how and for how many I can vote for. Actions themselves are shown to impact emotion and preferences, not just the other way around. Thus, who I actually vote for is likely to impact outcome satisfaction, somewhat independent from and perhaps complexly interacting with the impact of my preferences on outcome satisfaction.
Again, type of voting system is itself something that likely has a causal impact upon preferences, how they are determined and co-constraining, how they impact outcome satisfaction, and voting actions themselves and thus their impact on satisfaction. Thus, it is likely that a separate set of algorithyms must be created for each type of system, each with psychologically plausible sets of assumptions for that context.
The programs in question do not even count as "simulations" if these various assumptions are not set based upon existing psychological data, which they don't appear to be.
They are predictions about what would happen in the fictional and implausible universe that the programmers have not very creatively imagined.
lpetrich
Contributor
ronburgundy, all you seem to be saying is that big-money lobbies and interests will find different ways of gaming the system. That's out of the scope of these simulations. But please feel free to try to do so.
ronburgundy
Contributor
That is definitely not all I am saying. That is about 10% of what I am saying. Almost all the factors I have mentioned apply whether or not big-money games the system.
The fact is that these models all make tons of assumptions, including assumptions about every issue I raised plus others. If these assumptions have no valid basis (and they don't) in what relevant psychological and sociological empirical evidence we already have, then the entire models are just a meaningless fiction about what would be true in the purely fictional universe represented by the programmers code.
The output of the models are NOT empirical data. They are made up data, invented by the invented code that has no grounding in empirical reality.
Until the models are compared to the actual empirically observed data of actual humans voting under different systems, with actually measured human satisfaction and regret, then these models are nothing but a really boring form of fiction writing.
Duke Leto
Deleted
Well it seems to me that we're in a dilemma.
First, we don't have actual empirical data on what the untested alternative voting methods do in the real world.
Second, it would be grossly irresponsible to implement a voting method we have good reason to believe could have negative effects on a major government.
It looks to me like the only way out is to set up a clinical trial in self-government where several non-governmental organizations choose to administer themselves via the varying voting systems and then watch what happens in those contexts.
First, we don't have actual empirical data on what the untested alternative voting methods do in the real world.
Second, it would be grossly irresponsible to implement a voting method we have good reason to believe could have negative effects on a major government.
It looks to me like the only way out is to set up a clinical trial in self-government where several non-governmental organizations choose to administer themselves via the varying voting systems and then watch what happens in those contexts.