Yes, I can see how that would simplify vote tampering.
Or vote verification. If a party adds ballots to the national heap, it's going to be really hard to discern them from the valid counts. But if they add ballots to a local heap so there are more ballots than voters, it's easy to see something is screwy, and it's going to be really hard for the party to tamper with enough local counts to offset the national count as much as in the first scenario. Similarly, if some ballot boxes get "lost" (only those supporting your opponent, of course) while transporting to the national count, then nobody will notice, but the party will have to corrupt many local areas in order to get away with something on the same scale.
So it makes sense in an adversary environment. But it also makes sense in a benign one; it makes logistics a lot simpler.
As opposed to what, a bunch of corn farmers standing in a corner to choose (non-secretly) which $1B presidential candidate(s) gets a guernsey to run for an election with less than 50% turnout?
As opposed to a system where you only have to rank or rate the candidates you have an opinion about. IRV can be retrofitted to work this way, but Condorcet and Range works like this by default. You can keep the compulsory voting, but it'll make it simpler for people with different opinions to accurately give that opinion, and it'll deter donkey voting.
There's also the strong defensive criterion, which means it shouldn't hurt if a majority votes for two parties instead of just the one they prefer the most. Range fails it because if you vote for say, Green and Democrat, your Democrat vote will counter your Green (and if all Green voters do the same to hedge, then the Green can never win). Condorcet passes...
... but in the case of cycles, Condorcet is complex. Since Range dynamics matter, not all is lost, however; when the parties become about equal in size, some of the Greens may stop rating the Democrat highly. So both Condorcet and Range are better than plurality and (one'd expect) escape the third party spoiler problem.
The above shows that Range (and its reduction, Approval) is not flawless -- and Bayesian regret only makes sense if you accept that you can sum up utilities. Arrow didn't, so his theorem doesn't apply to Range; he constructs a mostly-circular tie and then shows that excluding a certain candidate makes a certain other win, where the situation is symmetric, so the system can't be perfect.
It would be interesting to see a theorem showing that no rated voting system is perfect either (or that there is some perfect system), but Arrow isn't it.
Like in any constantly changing system, the pressures matter. One of Warren's other papers say that in a two-party state, the two parties have to show opinions that look similar to that of the usual voter (with one party slightly on the left and another slightly on the right). But since the parties aren't made of usual voters, that means they have to lie, and often quite severely, to affect that picture.
It's also easier for third parties to appear when the voters know that their vote aren't wasted, and the results inform others that third parties can be a viable choice. This is what happened in New York in 1936 when they introduced the Single Transferable Vote; prior to it, the democrats pretty much controlled everything, but afterwards, many parties appeared. (It was eventually repealed - the main parties' Red Scare tactics with regards to elected Communists worked.)
The point here is that a shift does not only change the situation today, but it changes the preconditions for the situation tomorrow. Changing the method could lead to more parties, and more parties would mean it's harder to bribe them all, weakening the power of capital, for instance.
People should be able to follow Range voting; practically every review has an "out of five" or "out of ten" verdict.
It's not as good as Condorcet (IMHO, but Warren disagrees). Still, it's a lot easier to explain if there's any chance of a cycle in a Condorcet election. (If not, both are probably as easy to explain: "Rate each candidate, and the candidate who has the highest average wins", versus "Rank the candidates, and the candidate who wins against all the others in a round robin wins".)
Systems like the Condorcet Method, Meek's Algorithm and Range Voting have some theoretical advantages, but they fail in one crucial respect: they are hard to count. Range Voting creates possibly hundreds of rounds of counting. The Condorcet Method creates exponential numbers of counts.
That's completely wrong. Range Voting consists of adding up the numbers and then taking the average. As anyone knows, that's linear in the number of candidates and votes. Even if you do it by counting "pseudovotes" (this candidate got that many ones, twos, threes, etc up to nines), the granularity of the ballot is a constant, so it's still linear.
As for Condorcet, counting a ballot takes quadratic (0.5*n^2) time with respect to the number of candidates. If A, B, and C are ranked on a ballot, then you just check if A is more highly ranked than B, A more highly ranked than C, and B more highly ranked than C. Finding out who the winner is is linear in the best case - that there's a candidate who's preferred to all the others one-on-one and that's the first candidate you checked, and quadratic in the worst case if there's still a candidate who's preferred to all the others. If there is a cycle, the methods vary, but in public elections, that would be exceedingly rare. Though for the case of completion, I'll note that most of the good Condorcet methods (like CSSD which Debian uses) are n^3 in the very worst case. In either case, determining the winner once the votes have been totaled up into the matrix takes logarithmic time in terms of the number of ballots (since all you have to do is compare numbers in the matrix or the averages list).
Another advantage with Range or Condorcet is that you can count the ballots where they're gathered and then only transmit a small amount of data (the pairwise counts for Condorcet, or the numerators and denominators for the average for Range), instead of having to count everything at the central place as in IRV.
That it is compulsory in Australia helps to moderate our politics by ensuring that the almost the whole population turns out to vote, not just ultra-motivated special interest groups (churchies, to pick a purely random example).
Too bad about the how-to-vote cards then, no? Though there's nothing about IRV that demands you have to rank absolutely all the candidates, the implementation you have is flawed.
Even worse, it's not just Pascal's dilemma with regards to God, it's Pascal's dilemma with regards to God, Allah, Yahweh, and many more.
It's not merely to either sacrifice the amulet at the altar or be hacked up by one of the four horsemen, but you have to pick the right altar!
(At this stage the odds seem pretty slim already, and maybe the god, if he exists, like frank unbelievers more than intentional heretics believing in some other god. The wager is getting harder by the minute.)
As an example of the above, I am working on a programming language, and one thing this programming language will feature is different subsets for different niches. One such subset will allow any program to be written, so long as it doesn't change the state of anything outside the program that was not passed into it as a modifiable data structure.
That sounds a lot like capabilities, which is a very good idea. It's just too bad none of the larger OSes support them by default, which means we're stuck with having to add compartmentalization explicitly.
I know that was a joke, but it just has to be said. The engineers messed Chernobyl up. Basically, they turned off all but one of the safety overrides to see if the remaining would work on its own. And all this in a reactor without a containment dome, moderated with something that burns really nicely in the atmosphere.
I'm not so sure. Or if it will, it won't be as easy as one might imagine.
The great strength of Linux and OSS projects is that the developers are the users. But the coders rarely use low-end hardware, which causes software to have errors or performance problems even though it works for the coders. There's an example of this, as it relates to file systems, here, and I think the "there's no memory leak/fragmentation!" stuff in Firefox can be partly blamed on this effect too -- if you have gigs of memory, you won't see it.
As the parent says, when you reach a peak on U-235, you switch over to breeding U-238 or Thorium, or use an Energy Amplifier (accelerator driven system). And once you reach the peak on those, well, given that estimates for U-238 reach from a thousand years and up, by that time you should have fusion.. or parallel universe extraction, or a ZPM, or who knows?
If you're trying to imply that the Canadian political system is somehow immune to such excesses, you're wrong. The reason companies spend a boatload of money on US elections is because US elections matter a great deal to their bottom line; on the other hand, who governs Canada simply doesn't matter much to corporations or anybody outside Canada.
No. Unless you count these as unimportant, the United Kingdom used about $1.34 per capita for campaign finance in 2004. Canada used $1.50 per capita. The United States, on the other hand, used a much larger number, $16 per capita! So even when adjusted for population, the United States is way off the rest.
The true problem is money influences much more than votes here. With the game stacked against the voters (and against third parties), how could it not?
Slashdot Mirror
SMB? Oh... I know! Just use Samba, that's open! :-)
Aww, but I prefer Omega 13!
Yes, I can see how that would simplify vote tampering.
Or vote verification. If a party adds ballots to the national heap, it's going to be really hard to discern them from the valid counts. But if they add ballots to a local heap so there are more ballots than voters, it's easy to see something is screwy, and it's going to be really hard for the party to tamper with enough local counts to offset the national count as much as in the first scenario. Similarly, if some ballot boxes get "lost" (only those supporting your opponent, of course) while transporting to the national count, then nobody will notice, but the party will have to corrupt many local areas in order to get away with something on the same scale.
So it makes sense in an adversary environment. But it also makes sense in a benign one; it makes logistics a lot simpler.
As opposed to what, a bunch of corn farmers standing in a corner to choose (non-secretly) which $1B presidential candidate(s) gets a guernsey to run for an election with less than 50% turnout?
As opposed to a system where you only have to rank or rate the candidates you have an opinion about. IRV can be retrofitted to work this way, but Condorcet and Range works like this by default. You can keep the compulsory voting, but it'll make it simpler for people with different opinions to accurately give that opinion, and it'll deter donkey voting.
Perhaps he's a Putin supporter?
They give power to those with infrastructure, i.e. the parties. But if you want to have a party-based system, why not go straight to party list PR?
There's also the strong defensive criterion, which means it shouldn't hurt if a majority votes for two parties instead of just the one they prefer the most. Range fails it because if you vote for say, Green and Democrat, your Democrat vote will counter your Green (and if all Green voters do the same to hedge, then the Green can never win). Condorcet passes...
... but in the case of cycles, Condorcet is complex. Since Range dynamics matter, not all is lost, however; when the parties become about equal in size, some of the Greens may stop rating the Democrat highly. So both Condorcet and Range are better than plurality and (one'd expect) escape the third party spoiler problem.
The above shows that Range (and its reduction, Approval) is not flawless -- and Bayesian regret only makes sense if you accept that you can sum up utilities. Arrow didn't, so his theorem doesn't apply to Range; he constructs a mostly-circular tie and then shows that excluding a certain candidate makes a certain other win, where the situation is symmetric, so the system can't be perfect.
It would be interesting to see a theorem showing that no rated voting system is perfect either (or that there is some perfect system), but Arrow isn't it.
Like in any constantly changing system, the pressures matter. One of Warren's other papers say that in a two-party state, the two parties have to show opinions that look similar to that of the usual voter (with one party slightly on the left and another slightly on the right). But since the parties aren't made of usual voters, that means they have to lie, and often quite severely, to affect that picture.
It's also easier for third parties to appear when the voters know that their vote aren't wasted, and the results inform others that third parties can be a viable choice. This is what happened in New York in 1936 when they introduced the Single Transferable Vote; prior to it, the democrats pretty much controlled everything, but afterwards, many parties appeared. (It was eventually repealed - the main parties' Red Scare tactics with regards to elected Communists worked.)
The point here is that a shift does not only change the situation today, but it changes the preconditions for the situation tomorrow. Changing the method could lead to more parties, and more parties would mean it's harder to bribe them all, weakening the power of capital, for instance.
People should be able to follow Range voting; practically every review has an "out of five" or "out of ten" verdict. It's not as good as Condorcet (IMHO, but Warren disagrees). Still, it's a lot easier to explain if there's any chance of a cycle in a Condorcet election. (If not, both are probably as easy to explain: "Rate each candidate, and the candidate who has the highest average wins", versus "Rank the candidates, and the candidate who wins against all the others in a round robin wins".)
That's completely wrong. Range Voting consists of adding up the numbers and then taking the average. As anyone knows, that's linear in the number of candidates and votes. Even if you do it by counting "pseudovotes" (this candidate got that many ones, twos, threes, etc up to nines), the granularity of the ballot is a constant, so it's still linear.
As for Condorcet, counting a ballot takes quadratic (0.5*n^2) time with respect to the number of candidates. If A, B, and C are ranked on a ballot, then you just check if A is more highly ranked than B, A more highly ranked than C, and B more highly ranked than C.
Finding out who the winner is is linear in the best case - that there's a candidate who's preferred to all the others one-on-one and that's the first candidate you checked, and quadratic in the worst case if there's still a candidate who's preferred to all the others. If there is a cycle, the methods vary, but in public elections, that would be exceedingly rare. Though for the case of completion, I'll note that most of the good Condorcet methods (like CSSD which Debian uses) are n^3 in the very worst case. In either case, determining the winner once the votes have been totaled up into the matrix takes logarithmic time in terms of the number of ballots (since all you have to do is compare numbers in the matrix or the averages list).
Another advantage with Range or Condorcet is that you can count the ballots where they're gathered and then only transmit a small amount of data (the pairwise counts for Condorcet, or the numerators and denominators for the average for Range), instead of having to count everything at the central place as in IRV.
That it is compulsory in Australia helps to moderate our politics by ensuring that the almost the whole population turns out to vote, not just ultra-motivated special interest groups (churchies, to pick a purely random example).
Too bad about the how-to-vote cards then, no? Though there's nothing about IRV that demands you have to rank absolutely all the candidates, the implementation you have is flawed.
Would that magic pixie dust be called Trusted Computing, by any chance?
The abbreviation's actually short for "Raze the F...er", since that's about the only thing you can do with the format.
Or a solar focus telescope, then you can use the sun as your lens. There's just that little detail about getting out to 550 AU!
That's Pascal's dilemma.
Even worse, it's not just Pascal's dilemma with regards to God, it's Pascal's dilemma with regards to God, Allah, Yahweh, and many more.
It's not merely to either sacrifice the amulet at the altar or be hacked up by one of the four horsemen, but you have to pick the right altar!
(At this stage the odds seem pretty slim already, and maybe the god, if he exists, like frank unbelievers more than intentional heretics believing in some other god. The wager is getting harder by the minute.)
As an example of the above, I am working on a programming language, and one thing this programming language will feature is different subsets for different niches. One such subset will allow any program to be written, so long as it doesn't change the state of anything outside the program that was not passed into it as a modifiable data structure.
That sounds a lot like capabilities, which is a very good idea. It's just too bad none of the larger OSes support them by default, which means we're stuck with having to add compartmentalization explicitly.
I know that was a joke, but it just has to be said. The engineers messed Chernobyl up. Basically, they turned off all but one of the safety overrides to see if the remaining would work on its own. And all this in a reactor without a containment dome, moderated with something that burns really nicely in the atmosphere.
I'm not so sure. Or if it will, it won't be as easy as one might imagine.
The great strength of Linux and OSS projects is that the developers are the users. But the coders rarely use low-end hardware, which causes software to have errors or performance problems even though it works for the coders. There's an example of this, as it relates to file systems, here, and I think the "there's no memory leak/fragmentation!" stuff in Firefox can be partly blamed on this effect too -- if you have gigs of memory, you won't see it.
As the parent says, when you reach a peak on U-235, you switch over to breeding U-238 or Thorium, or use an Energy Amplifier (accelerator driven system). And once you reach the peak on those, well, given that estimates for U-238 reach from a thousand years and up, by that time you should have fusion.. or parallel universe extraction, or a ZPM, or who knows?
If you're trying to imply that the Canadian political system is somehow immune to such excesses, you're wrong. The reason companies spend a boatload of money on US elections is because US elections matter a great deal to their bottom line; on the other hand, who governs Canada simply doesn't matter much to corporations or anybody outside Canada. No. Unless you count these as unimportant, the United Kingdom used about $1.34 per capita for campaign finance in 2004. Canada used $1.50 per capita. The United States, on the other hand, used a much larger number, $16 per capita! So even when adjusted for population, the United States is way off the rest. The true problem is money influences much more than votes here. With the game stacked against the voters (and against third parties), how could it not?