Democracy Building in the Worldbuilding World
It was recently announced that Worldbuilding.SE was to graduate. That is leaving the beta mode. One of the elements of the graduation is the election of moderators. So far, there were 4 pro-tempore moderators who were chosen by the StackExchange staff, to moderate the site during the beta period. Now the time came for the community to elect their moderators themselves. We will use that opportunity to details the process of an election, and provide some details.
An election on the StackExchange Network, follows three phases:
- Nominations: during 7 days, all users with more than 300 reputation points may nominate themselves. And provide some short description. Meanwhile the rest of the community at large collects a list of questions for all candidates to allow for some comparison.
- Primary election: during 4 days, all users who have more than 150 reputation points, may upvote, downvote or not vote for each candidate. There are no restriction on the number of votes, and the votes can be changed at any time.
- Main election: the top 10 from the primary election are selected for the main round of election. That one lasts for 4 days as well but all users with more than 150 reputation points are provided with a single vote, this can be shared among three candidates using Meek Single-Transferable-Vote (STV) principle.
Since we want to focus on the mechanism of the election, the following graphs are kept without the names of the candidates. To know them, the reader is advised to visit the corresponding site.
The first part of the election started on last February 1st. Candidates had until February 8th at 20:00 (UTC) to nominate themselves. The following shows the number of candidates as the nomination phase progressed.
The primary phase started with 19 candidates. We can observe that 16 of them nominated themselves in the first two days.
Even if you are not familiar with StackExchange/StackOverflow, from the first part of this post, you probably realised that reputation points are quite important within the network. Here is a distribution of the reputation points of the candidates, compared with the same distribution for all users.
A rather wide distribution of candidates reputation can be observed. But most interestingly, we can see that 5 from the 7 top users (in term of reputation) nominated themselves. Furthermore, 3 out of 4 pro-tempore moderators nominated themselves to follow on the moderator duties. It isn’t surprising to note that a large portion of those are regular chat users, and as such “know” each others reasonably well.
Alongside with the reputation, there are badges to reward the activity and achievements of the users. In particular for the election, 8 moderation-related badges, 6 editing badges and 6 participation badges are combined with the reputation (up to 20,000) to make a Candidate Score. Which is a number out of 40. The badges distribution of the candidates is the following
This leads to the candidate scores distributions
Badges and reputation often reflect the activity on the site. It is not necessarily the case, but highly reputed users tend to have many badges. Thus the distribution of candidate scores tend to resemble the reputation distribution. We see that three candidates are rather new to the site, and thus have lower candidate scores, and lower reputation points. If you are a StackExchange user, you can also find what would be your score.
Anecdotal statistics, the usual location of the candidates is the following
The 19 candidates faced the primary election. In that one, resulting votes are public, and users with some reputation can see the details of up-/down-votes. 16 candidates answered the questions collected during the nominations. Users of the site don’t have access to precise statistics of participation. About 1,500 users were allowed to participate in the primary. 600 users viewed the election page (this is also confirmed by received badges). And a minimum of 178 users casted a vote, based on the votes received by the most voted candidate and themselves.
To our knowledge, it is not possible to get the evolution of the votes, or participation, afterward. However, some users noted the scores of some/all candidates a few times throughout the primary. This allowed us to reconstitute that evolution. The votes (down-votes subtracted) are represented here
The most active users on the site are familiar with some/most of the candidates as the worldbuilding community is relatively small. However for less active users, one would expect the candidate score to be a strong factor for the votes. Following is a distribution of the primary score with candidate scores.
The blue line presents a linear fit of all candidates, with the exception of the pro-tempore, as their candidate scores do not necessarily compare to the other users. Those are indicated in red. Furthermore, the candidate score cannot be negative. So candidates with 0 votes may not fit. The green curves indicate a 1 sigma separation from the blue curve. Most users follow the linear fit to some extend, indicating a rather strong correlation between the candidate score and the votes.
The 10 most voted candidates went to the final round of the election. That round started on Friday the 12th of February (20:00 UTC). Voting is done using hidden ballots and no results are shown until the end of the election period, however we can see how many people have voted at any moment in time. Each voter is given three ordered votes. Those will be refered to as the first to the third choices.
The officials of the election gave the following values for the participation:
1,464 voters were eligible, 777 visited the site during the election, 640 visited the election page and 309 voted.
Details of the participation aren’t given, but two badges are important there, the Caucus which indicates the visit of the election page (those 640 visitors) and Constituent for the 309 who voted. From them, we can evaluate the visit and voting rates (includes nominations, primary and final election)
It is noteworthy to see the repartition of reputation of those who visited the page (in blue all users and in red users who visited the page, in green the ratio)
and of those who voted (orange) compared to the visit (still red)
In this section, each of the candidates was given a number between 1 and 10 in the raw data of the results. Following the previously discussed point, those numbers will be used in place of the candidates names.
Out of 309 votes, it is interesting to note that 1 user (0.3 %) made a second and a third choice, but no first choice. 8 users (2.6 %) on the contrary only used their first choices. Finally, 6 users (1.9%) were satisfied with two choices and did not select a third one.
The votes for the first round (first choices only) are given in the following
Many wonder what are the contributing factors for those first votes. To see some possible factors, we will see the relation with the result of the primary election (in red winners)
the relation with the candidate scores
and the relation with the reputation
It is also interesting to see how the first choices relate with the second and third choices. The matrices are as follow
and normalised to the first vote (in percent)
It can be observed, that the users who received more votes in the primary received many more votes in the main election as well. Some particular statistics of this effect are
- 293 users (94.9 %) had at least one of them among their three votes, 183 (59.2 %) had at least two of them, and 29 (9.4 %) had the three of them in any order,
- 214 users (69.3 %) had one of them as a first choice, 153 (49.5 %) placed one of them as a second choice, and 138 (44.7 %) had them as their third choice,
- 94 users (30.4 %) had them as first and second choice.
And how would it look if it wasn’t using a STV system, but a 3 votes system: the three votes count the same. In order to be comparable, the sum of the votes has been normalised ((1+2+3)/3). The dots are the results of the first choice (1) for comparison.
It can be observed that if such a system was used and the 4 most voted elected, the final result would remain unchanged. However, we see that the 3 most voted had an advantage by counting only one vote, as they would lose some of their advantage to all other candidates. In particular, number 3 and 6.
The STV system follows a series of rounds until all the candidates are either qualified (marked with w) or eliminated. In the following, some of the 22 rounds are illustrated. In red the threshold for the round is given.
It is of course hard to explain what were the determining factors for those four. Apart from reputation points, candidate scores and election presentation, it is interesting to note that
- two of the new team were pro-tem moderators (but that in itself isn’t enough as one other pro-tem moderator did not make it),
- two of the team are moderators on other StackExchange sites (but they weren’t the only ones in the final round).
The very first moderator team has been elected on Worldbuilding. And we wish them all the best for the challenges ahead.