Free and private cardinal polling

Is there a place where there are free and private tools for making cardinal polls. Either STAR or Score would work. I think doodle only has approval

Whatever you find, add it to the list https://electowiki.org/wiki/Voting_links#Polling_sites

1 Like

This is kind of not exactly on topic, but I think the best system for internet voting would use score but allow voters to change their vote according to the current result, which gives a cleaner system that would get rid of the need for any sort of clunky run-off.

The current standings would be constantly updated so voters could see exactly how strategically optimal their current vote is. Obviously a lot of people might leave their vote right until the end so that it’s harder for people to do this, but that can be mitigated against by having a random end point.

Say you wanted the vote to last about a week. The first five days could be completely free to vote and change your vote. Then at the end of five days, no new votes can be added but changes are permitted. Then after six days the vote will randomly end with a half-life of, say, two hours.

And while I’m here, for a proportional vote, you’d probably use an allocation system. Voters would have, say, 10 points that they can distribute among the candidates. Then instead of there being a single end point, there would be eliminations (using the same random time system as above) where the candidate in last place would be eliminated and voters could then reallocate points given to these candidates.

Sujjest does this but for Approval Voting.

Sounds like Asset but with direct negotiation? Seems like you could just have voters rank the candidates and run CPO-STV, and if there’s a Condorcet cycle of outcomes, the voters could become involved again in figuring out who should win.

I disagree. I think the half-life should be measured in days because otherwise there will be many people who, say, are away from 09:00-17:00 and only have a specific window of time in the evenings to get on the poll.

That was just an example of how it could work, and someone setting the poll could determine the values for each variable. You might be right though that a longer half-life would work better in general.

Would that give similar results? I quite like the idea of there being a score for each candidate that everyone can see allowing them to adjust accordingly. I’m not really a big fan of this sort of allocation rather than plain score voting, but this gives a more transparent way to tell exactly what changing your vote would do to the result.

I think so? Asking voters to continually adjust point allocation (which is really just fractions of their whole vote) until they reach agreement on who should win is just a style of Asset negotiation, except the voters, not the candidates, are negotiating. Asset is Smith-efficient for negotiators’ preferences if they are given enough time, since eventually they’ll reach a winner set or set of winner sets which consistently get more votes (points) than any other possible winner set. So the only way CPO-STV and Asset could diverge (supposing rational and honest negotiators) is if, during the negotiations, some voters’ preferences changed (which could help to resolve Smith Set cycling if there is any.)

You could do “cumulative Asset” to decide the winners, but use CPO-STV as a guideline just to show voters which winner sets are most likely to be in the Smith Set. Any PR method should work for that purpose, but a Smith-efficient PR method will demonstrate what the negotiating equilibrium will look like if every negotiator is maximally strategic. To see the more honest utilitarian outcomes, you could ask voters to submit rated ballots and compute the Monroe, Thiele, and CPO-STV outcomes from those.

What is the Smith set when you’re talking about multi-winner elections?

With the single winner version (where I’d use score voting), I see it as a feature that it wouldn’t necessarily always elect the Condorcet winner, even if it would in theory with people optimising their votes. Take the following score ballots (out of 10):

49: A=10, B=0, C=1
49: A=0, B=10, C=1
2: A=0, B=0, C=1

This might be people’s initial votes before they change them according to the current result. C is the Condorcet winner, but I’d be very surprised indeed if C ended up winning the election, nor would I want them to.

I’d also like something similar to happen in the multi-winner case (I might try and concoct an analogous multi-winner example), which is why I’d rather not use a ranked-ballot method. I’m not sure how well the points allocation method would mitigate against it though.

For example, we could have an election like this with two to elect and the following honest scores out of 10:

10 voters: A=10, B=0 C=9, D=9
10 voters: A=0, B=10, C=9, D=9

An STV method would automatically elect A and B, whereas C and D is arguably more desirable. If voters have 10 points to allocate, it might look something like this to start with:

10 voters: A=4, B=0, C=3, D=3
10 voters: A=0, B=4, C=3, D=3

At this point, A & B would have an average of 2, and C & D would have an average of 3. Of course, voters might at this point start reducing their C/D scores and increasing their A/B scores, and they might not have given such an initial vote and maybe just bullet voted for their favourite. So we might still get AB rather than CD.

In fact with the elimination phase, voters probably would start to drop support of one of C/D and then with just ABC (or ABD) left, it probably would be AB that got elected.

So I don’t know. I still think there is a chance CD could be elected, and there might be a better way of doing it than allocating points, but STV would certainly elect AB.

Take this example (duplicated candidates are not affiliated):

8 D>A>B>C
(writing the ranks without the > sign below)
7 BCAD
5 CBAD
8 D2>A2>B2>C2
7 B2C2A2D2
5 C2B2A2D2

Initializing an algorithmic Asset negotiation, D and D2 have the most votes (1st choices). But within each Hare Quota, ABC and A2 B2 C2 are in a Condorcet cycle. Therefore, once a few negotiating moves (pairwise comparisons) have been done, D and D2 won’t be in any of the 2-winner winner sets the negotiators cycle through. For example, if the negotiators are currently supporting (B, B2), and D and D2 attempt to gather enough support to win, 12 ballots prefer B to 8 for D or D2, and the same for B2. So the Smith Set here is all of the outcomes in the Condorcet cycle, which is a proper subset of all possible 2-winner sets.

The new ranked preferences of the voters, if they allow C to win, would be something like:
49 A>B=C
49 B>A=C
2 C
In other words, C isn’t the Condorcet winner anymore when the negotiators’ preferences change, but one of A and B, both being “weak Condorcet winners”, will probably win, so Asset still picks from the negotiators’ Smith Set here.

Agreed. Consider also if you have
51 A5 B4
49 A0 B5
More than likely the majority will allow the result to stand.

The outcome CPO-STV comes up with could be optionally ratified by the voters.

Why not just have the voters vote with regular score ballots, do Monroe or Thiele, and then allow them to visualize adjusting their score for any given candidate(s) and observe what impact that has on the winner set? It’d be computationally intensive, but maybe not prohibitively so.

Consider that Condorcet forms of STV are compatible with equal-ranking, so it’s possible voters would equally rank their pick of A or B along with both C and D 1st. Also:

Just to clarify my stance, point allocation could be the decision-making system, but in addition, the voters could submit rated or ranked ballots to get guidance on which candidates are more likely to be in a final winner set, and thus should be focused on for point allocation and negotiation. It also helps the voters figure out if their favorites are viable or not at that point in time, since then they can say “please take a second look at my favorite, re-evaluate, and consider giving more support to them! they’re better than (whichever is the most similar candidate who currently seems to have a better chance of winning)”

Congratulations. You have managed to go so far off topic from my original post you have actually gotten on topic for not one but two of my other posts.

You method is essentially Cumulative Voting. I brought up Cumulative Voting here. I do not think it works because it does not solve the concept of vote splitting.

Putting aside the concept of making continual updates to votes, the bigger issue is if you want the voter to handle the strategy and negotiations or if you want to get their honest scores and compute the negotiation fairly. @Toby_Pereira, your PAMSAC is one such method and I think these are the way to go. Cumulative voting and variants of it seem to force the strategy on the voter meaning that they can play it wrong. If we just ask for honest utility and then compute the best winner set we are likely to be much better off in the end. Admittedly free riding is an issue to be mitigated but I do not think it is any better in the cumulative voting situation. In fact, I could see vote management becoming an issue.

So this bring us to the work of the Wolf Committee. The goal of which is to try to find the best method to select a winner set. The current top contenders would all select C and D not A and B. STV has a known polarization bias.

Wouldn’t SMV do A and B, since they have better quota scores?

That brings up the possibility of a consensus modification: calculate both the quota score and 2-quota score, and if some candidate’s 2-quota score is higher than the 2-quota score of the candidate with the highest quota score, and the first candidate has a quota score that is almost as good as the latter candidate, elect them instead and spend their quota (possibly with SSS reweighting).

Just so people are aware: When replying, you can click the reply icon above-left of the message box next to either the person’s name or the topic name. The option “Reply as linked topic” will appear.

1 Like

Yes you are right my mistake.

For an online cardinal negotiation tool, the added complexity and computation cost of Marylander’s anti-freeriding cardinal modification shouldn’t be too bad, right? That idea (if no candidate can get a full quota in a particular round of a sequential PR method, restore ballot weight to previously spent ballots proportionally if they had spent more to elect a candidate than the ballots in the current round would, and recompute scores for each candidate this way, picking the best outcome) can be applied to SMV and other cardinal PR methods.

I suppose my problem with this is that all you can see is the winner set, so the voter would be blind to how close other sets are to becoming the winner set. Where each potential set of candidates has its own score under the current vote, you couldn’t list them all with their current score because there would be too many, and even if you listed the top 10 or whatever, it would still not be clear what would need to happen for the result to be overturned.

In what I am proposing, each individual candidate would have a score so voters would understand the result better.

Perhaps giving voters too many strategy options could be counterproductive. But one change you could make is that if a candidate is eliminated then a voter who had allocated points to that candidate gets them automatically redistributed to the other candidates they have already allocated points to in the same ratio that they’ve already given them points (subject to rounding). They can still also change it further themselves, but it means that they haven’t got to act if one of their candidates is eliminated.

Could that be partly because the methods being considered by the Wolf Committee are sequential rather than because they lack polarisation bias? I would expect most sequential score methods to elect C and D because they have higher totals.

Also, for the single-winner version I mentioned above with score voting rather than cumulative voting, I think the ongoing vote-changing mechanic would still work well, partly because score voting itself is a good method and those voters not strategising optimally are just engaging in score voting. And all other methods (whether Condorcet, STAR etc.) have their own problems, and are really only needed because under score voting people don’t necessarily know the optimal strategy. But this method would give them that tool. It wouldn’t work for all types of poll, but for stuff that’s being decided online anyway by internet savvy people, I think it would be a very good method.

Maybe some (approximate) dynamic metrics such as “this candidate appeared in x% of Thiele’s upper half of winner sets when you scored them a 5/10, but y% if you score them a 7/10” or such would work?

There are sequential systems which have this polarization bias. Other than STV, it was pointed out above that Sequential Monroe has this bias and it is score.

It’s really “select the candidate with the best most-supporting Hare Quota” selection that’s at the root of this, since SSS or other methods with Hare selection also have this (though if the reweighting is changed from allocation and maybe parts of the example are tweaked, it seems like a Hare selection PR method will pick a polarizing candidate for the first seat but could pick a consensus candidate for the second seat).

Added after receiving the first heart:

I believe utilitarian advocates may be missing the possibility that voters themselves value consensus (when it is very good for them too) and oppose the election of “universally regarded as terrible” candidates, even if they’re theoretically a little better than the other side (even if only because supporting such candidates could help those candidates defeat their favorite). Their own honest preferences will likely reflect that, if they’re only allowed to express them more fully, and particularly when equal ranking or rating is allowed.

Also, it’s possible to take rated ballots, eliminate all candidates who don’t have at least 50% as much utility as the Score winner, then do Condorcet among the uneliminated candidates. Or, simply elect the Score winner if they have 30% more utility than the Condorcet winner (or the Score winner in the Smith Set).

1 Like