# Where do the (many) voters in the 2014 community moderator election come from?

Togeter with the results, it would be interesting to get an analysis to see where generally the many active voters in this election have come from.

This could for example be done by looking (automatically and of course only by people who have access to these data) on which site in the SE network (including Meta Stack Exchange) each voter has the largest amount of rep (his home site) and displaying these data in an appropriate voters versus (say 10 most important) home sites histogram.

Can somebody provide such a statistical analysis and post a corresponding histogram as an answer, after the results are out?

• Probably it is possible to ask for this sort of statistical analysis, and it's probably going to be useful after the election has ended. I agree that this election has a larger than ever turn out, which can be a good sign or a very bad sign. – Asaf Karagila Dec 21 '14 at 11:23
• I was shocked by the amount of votes in the primaries. There's no way the meta-MSE community has that many active members (active in the sense that they at least follow meta) and yet an enormous amount of people felt qualified to vote. – Git Gud Dec 21 '14 at 11:47
• What exactly would the purpose of that be? Certain statistics about the voters would be interesting of course, but I don't see what the "home site" of voters would tell us? I also find the whole concept of "home site" to be rather dubious, there are many different activity patterns for SE users, not all of them have something you could call a home site. – user9733 Dec 21 '14 at 12:22
• Not answering the question ... one reason for large number of voters may be email notices sent to users. – GEdgar Dec 21 '14 at 13:13
• The data needed is completely public: via the constituent badge it is easy to know who voted and the activity on sites is obviously also public. – quid Dec 21 '14 at 14:17
• About half of the constituent badges currently given (2358) were given for the recent election. This means that in the past 18 months since the last election, infinitely many more people became aware of the process. I do agree, however, that most people are not qualified to make this decision on the grounds that they are most likely not sufficiently familiar with meta and the candidates. – Asaf Karagila Dec 21 '14 at 16:17
• So-called "low information voters" are a problem in any election and I don't know that there's anything that can be done about them. Though at some point it becomes more expedient to simply appoint the three candidates with highest reputation and forget about the election. – user7530 Dec 21 '14 at 17:16
• @user7530: I agree. But when you have over all 600 voters, and about 200 of them are reasonably informed, and when you have 1500 voters and only 300 of them are reasonably informed... that's when you start getting a bit worried about the process (although it doesn't mean that the results are going to be bad, for this we'll have to wait and see). – Asaf Karagila Dec 21 '14 at 17:46
• @Dilaton what's interesting about it? – user7530 Dec 21 '14 at 18:34
• @GitGud, what's the basis for your claim that "there's no way the meta-MSE community has that many active members"? The number of people who voted in the primary doesn't necessarily reach 1000; there are over 100 people who got the "Supporter" badge on meta.MSE just in the past 3 weeks. Are you sure you're not vastly underestimating the number of people who read meta but don't post much? – Peter Taylor Dec 21 '14 at 19:58
• As one of the so called "unqualified voters" I voted as follows: 1. Daniel F., 2. Jyrki L., 3. Jack A. Even if one is unqualified it is easy to see which candidates are unfit/unqualified: e.g Pedro T. because he admits himself that he has a lot of quarrels (so he doesn't even satisfy the general criteria mentioned, albeit he has a lot of "reputation", but who cares about this). Others are sort of random candidates, e.g. Dustin or Ahaan where it is not clear why anyone would spend votes on them. There is not a lot of things to know in order to understand who is fit and who is not fit – Mister Benjamin Dover Dec 22 '14 at 0:43
• @313: No, it's not that easy to judge a candidate. Just look one election back, and all the fuss that ensued. – Asaf Karagila Dec 22 '14 at 7:39
• @AsafKaragila Yes, I guess very few would like to repeat the 2012 experience. But last year's election [the last one I referred to] hasn't caused such an upheaval, as far as I know. – Daniel Fischer Dec 22 '14 at 13:17
• @Asaf: So infinitely many is on the order of $1300$ or so? :-) – Brian M. Scott Dec 23 '14 at 15:19
• @Brian: It's closer to $3$, actually. – Asaf Karagila Dec 23 '14 at 15:21

I believe that this Data Explorer query answers your question: "Home site" of election voters. The SQL is monstrous, but it basically just looks at voters in the election and identifies which site they have the highest reputation on. (I'm not sure how it behaves when it comes to picking a site when a user has the same, highest rep on two sites, but that's an edge case and should be rare. Also, the total number of voters reported by this query will not necessarily agree with the count at https://math.stackexchange.com/election because some voters may have deleted their accounts at some point in time after the election happened.)

You can examine the data in detail on the query's page; here are the top 10 sites:

Count SiteName
----- ----------------------------
991   StackExchange.Math
207   StackOverflow
35    StackExchange.Mathoverflow
33    StackExchange.Physics
14    StackExchange.Tex
11    StackExchange.Stats
9     StackExchange.Mathematica
8     StackExchange.English
8     StackExchange.Codegolf
...


The results are much as you would expect - a large majority from this site; a large minority from SO; and then a long tail starting with sites on topics of interest to mathematicians.

The underlying data in #out is a list of (SE-wide account ID, site name, reputation on that site), for all users that voted in this election (as identified by having a Constituent badge awarded since the election started). If you want to slice and dice it in some other way (for example, bucketing by rep on Math.SE or something), you probably only need to edit the query at the bottom (beneath -- Reap results) - though, obviously, you'll need to know SQL. If you want to do something more complex, let me know and I'll take a look.

For comparison, here are the same data for the 2013 election:

Count SiteName
----- --------------------------
311   StackExchange.Math
48    StackOverflow
7     StackExchange.Physics
5     StackExchange.Stats
5     StackExchange.Tex
5     StackExchange.Mathoverflow
4     StackExchange.Mathematica
3     StackExchange.Cs
3     StackExchange.English
...


Also, I noticed that some users wanted additional data about meta participation of voters and so forth. If any of you have specific ideas about what kind of stuff you'd like to know, I can probably pull something together (provided, of course, that the necessary information is exposed in the Data Explorer).

• Additionally, here is some raw data about meta participation among election-voters that may be of interest to some: data.stackexchange.com/math/query/261475/…. You would probably be best served by downloading a CSV and examining the data in your Excel-analogue of choice. – senshin Dec 29 '14 at 22:00
• Very interesting stats about meta. Too bad they can't actually be downloaded due to a known SEDE bug. But sorting by various columns and scrolling around gives some idea of the data. For example, the majority of election voters have voted on Meta at least once. – user147263 Dec 29 '14 at 22:06
• It's perhaps interesting to note that, out of the 10 candidates in the final election round this year, two (Daniel Fischer and me) currently have SO as our "home site" by this definition, and one (dustin) has TeX.SE. – Ilmari Karonen Dec 29 '14 at 22:09
• @Behaviour Hm. That is a strange bug. There is a workaround, though: go to data.stackexchange.com/math/query/261475 (i.e. the no-parameters version of the query); then type in the parameters into the fields and run the query. Once you have done this, you will be able to download the CSV. – senshin Dec 29 '14 at 22:09
• Thanks, that worked. One can also avoid typing by using your original link, then remove the parameters from the URL, click Run Query again, and download. – user147263 Dec 29 '14 at 22:16
• Thanks for analysis, this is exactly what I wanted to get at. – Dilaton Dec 30 '14 at 8:51
• I was a bit surprised to see that I have the most comments on meta by a margin, and most votes by a nontrivial margin as well; and second-most posts. – Asaf Karagila Dec 30 '14 at 16:34
• Could anyone explain why Meta activity is a useful parameter to look at? I agree that being uninformed when voting in an election can be a problem. However, activity here doesn't seem to be a good indicator. Just considering myself, for instance, I regularly visit Meta, but simply never feel the need to add anything at all, which accounts to very low Meta activity. (I wouldn't even be surprised if this comment is the first text I write on Meta.) And yet, I'd still consider myself to be reasonably well-informed. – HSN Jan 2 '15 at 12:12

I'm not sure how to do the relevant analysis you asked for (given that it requires data from multiple sites), but I wrote a nice little query for voter reputation here. The data there's not real time (per Asaf's comment, it's presumably from ~7:00 AM UTC, Sunday), but at the moment, the reputation distribution looks as follows (which is a plot of the data $\log_2$):

Obviously, voting drops off relatively quickly as reputation increases, but that's not too surprising given that there are less people with high reputation than with low reputation (The median reputation of voters is $980.5$). I think the main takeaway here is that André Nicolas voted, being the only person with reputation greater than $2^{18}$. (Actually, I don't know what to make of this; I mainly just posted an answer so that I could post the histogram)

Per anorton's comment, I created a graph of users who could have voted (i.e. had 150 reputation and visited the site during the election). That looks like so:
Which drops off way more quickly than the other graph - the median reputation here is $519$). Where roughly $1/6$ of the users in the $150-256$ range voted, more than $1/3$ voted in the $4096-9182$ range, and every user in the $262144-524288$ range voted.

I will update this post when the election is over and all the data can be accessed via the data explorer.

• "Data updated 11 hours ago", meaning the data is relatively up to date. But we will have to wait a week before we can get the full information using the query. – Asaf Karagila Dec 21 '14 at 17:44
• Another interesting histogram would be number of users by reputation $\log_2$. – apnorton Dec 21 '14 at 18:32
• @anorton Done. (Though I filtered out users who have too little reputation to vote or who didn't visit during the election - I figure knowing who could have voted is more useful than just knowing everyone) – Milo Brandt Dec 21 '14 at 19:10
• Another interesting histogram would be the number of numbers with same integral value of $\log_2$. :-) – Asaf Karagila Dec 21 '14 at 19:14
• Rep has little import here. Instead one should plot some meta participation numbers, maybe number of meta Q&A, possibly also including number of comments. No doubt the results will show that most of the voters have very little meta participation, so are probably not well-informed on the major issues (true also for some of the candidates). Such problems are difficult if not impossible to solve well in systems with one-dimensional rep systems. Better systems might allow votes on other important factors, e.g. a user's mediation skills (resolving meta disputes), etc. – Gone Dec 21 '14 at 21:09
• @Bill I agree and that was initially my thought too, but that's a non-trivial statistic to get, since the data explorer only queries one site (and considers math.SE and meta.math.SE different). It would only require two queries and some processing on my computer to splice them together; I could try to do that once all the election data's in (so I don't have to do so twice). I'd be interested in counting votes on meta - any evidence of reading meta is worth seeing, since I'll bet many voters have no meta activity, no matter how generously you count (>1000 voters = more than on meta recently) – Milo Brandt Dec 21 '14 at 21:17
• @Meelo Any data analysis that may help serve to improve future elections is, imho, time very well spent. I'm grateful for your above work, and any future work you may do. – Gone Dec 21 '14 at 21:23
• I have looked roughly at the data and want to add that lower quartile is around 300 and only around 150 voters have less than 200 points. Fears that the election is unduly influence by voters not at all invested in the site seem exaggerated. – quid Dec 22 '14 at 0:39
• The database is updated since yesterday, in case you feel like drawing graphs in $\log_2$ again. – Asaf Karagila Dec 29 '14 at 17:57

I checked the "main site" of the most recent page of voters (at the point of writing) as well as the first few that have not many points on this site (I took the threshold to be 1000, since this is easy to parse while scrolling and also about at the median).

• 28 this site

• 13 SO

• 3 Physics

• 10 other (one from each of Math Educators, Statistics, Electrical Engineering, Programmers, Sustainable Living, SU, English, Christianity, Computational Sciences, Music).

Given the relative sizes of the sites this does not strike me as unusual.

Note: These are only the voters with few points; according to the data on the median in the other answer should make up about half of all voters.