All Models
Serial Approval Vote Election (SAVE)
A new voting system for finding consensus
Index of all models on this site:
Descriptions of processes can be useful, but sometimes you just have to play with the ideas to truly understand them. That is the idea behind Explorable Explanations (not affiliated with this site) and the models developed in the process of developing and explaining serial approval vote election (SAVE).
Note: These models were developed using a browser frame about 6 inches wide and 12 inches tall. They have not (yet) been optimized for either phones or large screens.
The Models and Their History
Brief notes about why I built these models and what I've learned from them.
IRV and Voronoi: Illustrating Instant Runoff Ballot Counting ,,fold,,
The Voronoi IRV explorable uses a Voronoi diagram to illustrate the information considered during the counting phase of Instant Run-off Voting (IRV, also sometimes called Rank choice voting or RCV). Recent efforts have managed to get IRV used in some elections, and while it is under some circumstances better than plurality (or first-past-the-post) voting, it falls far short of an ideal voting system. (At least in my opinion.)
The model can be used to show several key ideas in voting theory, and it contributed to the development of SAVE in two different ways. It got me thinking about voting systems that iterate over successive rounds, and helped me realize that while any set of initial motions will always have a single best motion or a top cycle of motions, there is absolutely no guarantee that any of the motions on a given ballot are actually liked by more than a small minority of the electorate.
Tournaments, Profiles and Spaces, Oh My! ,,fold,,
These models were built to show the different levels of information content determined by the choice of model used to explore voting issues. The traditional choice has been preference profiles, and they are about the limit of what can be done easily without computers. There is a strict information hierarchy here, with tournaments having the least information and spatial models having the most. Any given spatial model will map to exactly one preference profile, and any given preference profile will map to exactly one tournament. Moreover, many different spatial models map to the same preference profile, and many different preference profiles map to the same tournament. There are constructive algorithms to create a preference profile that maps to a given arbitrary tournament, and constructive algorithms to create a spatial model that maps to a given arbitrary preference profile.
Spatial models allowed and supported the development of SAVE, and this set of three explorables for tournaments, preference profiles, and spatial models was created to encourage voting theorists to use spatial models instead of either preference profiles or tournaments.
Tournaments encode pairwise wins, ties, and losses, but provide no data whatsoever about vote counts, or preference strengths, or anything at all about individual voters.
Preference Profiles encode linear preference orderings of labeled items, and the number of voters for each ordering. This data is used to run definitive pairwise simple majority elections and derive a unique tournament, along with vote counts and margins, for each win, tie, or loss, but provide no information about the relative strength of voter preferences, and only aggregate counts of preference orders.
Spatial Models encode individual voter ideals and the voter-specific metric functions used to calculate the subjective distance between a voter's ideal and any arbitrary motion, including motions that have not yet been proposed. The distances between a voter ideal and a set of motions are used to generate preference orders in which closer motions are preferred over those that are farther from the voter's ideal.
The Three Serial Approval Vote Election models ,,fold,,
The first two of these three SAVE models show how SAVE works with a simple Euclidean metric, where the distance formula is: \(d = \sqrt{\Delta{x^2} + \Delta{y^2}}\). The third simulation allows for more complicated scenarios, and includes controls that let the user experiment with different rules for choosing the next focus, and to modify the simulated choices of the voters.
After playing with the SAVE Basics and Focus models, I suggest looking at the non-SAVE explorables on Voter Metrics, and Aggregation. That way you will have a better idea of how the voters calculate their preferences and how those individual voter preferences aggregate to determine better collective choices.
SAVE Basics Using a Standard Euclidean Metric Function ,,fold,,
The first, SAVE Basics explorable for Serial Approval Vote Election shows SAVE in a simplified scenario, and provides the user the opportunity to see how voters respond to the initial motions, and how new motions can be added to get a better final result when the best possible motion is not initially provided. This is a somewhat radical departure from other voting systems because it uses multiple rounds to converge on the final result, and takes full advantage of the multiple round to add new motions during the process instead of eliminating existing motions.
SAVE Focus Details on Focus Selection ,,fold,,
The second, SAVE Focus explorable lets you take a closer look at focus selection, and get a sense of why I chose the rules that I did. At this time, I think I made a good choice for the focus selection rules, but maybe you can find a better set of rules that converges more quickly to the best available motion.
SAVE Voters - Exploring Voter Behavior Possibilities ,,fold,,
This SAVE Voters explorable lets you play with the parameters governing the voters' choices. In particular, it lets you modify where voters draw their lines for their initial approval of motions, the rate at which their tolerance increases over time, and under what conditions a voter votes for the current focus. It also covers how voters propose motions, and how the path changes when voters propose motions in different orders.
Voter behavior is the part of SAVE that I am most uncertain about, because each voter is really making up their own rules as they go along. Although I can prove mathematically that an honest ballot is also the most strategic ballot, that does not mean voters will always vote in their own best interests. Yet even if a voter makes a mistake on their ballot, and that mistake is enough to change a result, quite often the voter will be able to correct that mistake and still get a good collective result.
Voter Metrics and Aggregation - Individual and Collective "Good" ,,fold,,
These two explorables let you play with how voters determine which motions are subjectively better for them.
The Voter Metrics explorable lets you play with how metric choices and weights can change individual voter preference orderings. In this model, a single voter and three motions are arbitrarily positioned, and an indifference boundary is drawn through the middle distance motion. Motions on this boundary line (also called an indifference curve) are equally distant from the voter, albeit in different directions. A motion inside the curve is considered better by the voter over the other two motions, and similarly a motion outside the curve is considered worse.
The Aggregation explorable shows how the individual preferences of several voters can combine to form a collective preference. This model is used to explore collective preferences independent of any voting system. The aggregation is based on the exact voter preferences and issue weights under the chosen metric. This data is generally not going to be available in any real election, so it cannot be used as an election method. But the data is useful when comparing voting methods under controlled conditions to see which voting methods are likely to be more functionally democratic.
The Aggregation explorable introduces the concept of a "best" motion, which is the motion (or motions) that minimize the total subjective distance between it and the collection of voters.