I want to provide a bit of analysis here for this year’s OSMF board election results.
The election was ultimately uninteresting in terms of STV voting dynamics since the final results were the same as what they would have been based on first round results only. None the less i want to look at the second choices here.
In STV voters give a priority list of candidates. It essentially says: I want to vote for <position 1> on my list but in case that vote would be pointless for some reason i give my vote to <position 2> and so on. One of the most interesting thing to look at in the ballots is the second choices grouped by first choices. Below you have for example the 189 people who voted for Tobias on position 1 (27.7 percent) grouped by who they voted for at position 2. The most popular choices were Joost with 82 votes (43.3 percent) and Guillaume with 64 votes (33.8 percent). Same for the other candidates. Here are the numbers:
Overall Joost was a popular second choice among voters across most of the spectrum which allowed him to ultimately maintain his lead on Miriam - even though she got a significant number of second choice votes in round 4 of the STV process when Geoffrey was eliminated.
Relatively speaking Miriam was also a popular second choice across voters of different candidates, in particular Geoffrey, Jo and Joost. Tobias was comparatively popular among voters for Joost, Guillaume and Nuno but not such a popular second choice overall.
If you sum the first and second choice votes (as a theoretical exercise) Joost is ahead of Tobias but they’d both still win.
If you look at it in terms of candidates who - relatively seen - form popular pairs (A-B), i.e. where candidate B is popular among voters of candidate A and candidate A is also popular among voters of candidate B the most popular pairs seem to be
Note this is relative numbers - in terms of absolute numbers the combination Tobias-Joost is obviously the most popular with a hundred votes overall having these two at position 1 and 2 in either order.
Comment from Peda on 16 December 2018 at 16:17
I also had a quick look at the numbers. Btw, do you happen to know what the first column and the last column in the ballots raw data are meant to mean?
If you’re looking at ballots that gave 7 options, Tobias and Nuno had the most mentions there. As the 6th option Tobias was mentioned 58 times and Nuno 68 times and as 7th option Tobias was mentioned 72 times and Nuno 73 times. The next most frequent one was Jo with 62 mentions with overall 331 ballots filling the list up with 7 candidates.
The most common duplicated ballots voted Tobias for position 1. E.g. 29 ballots with Tobias, Joost, Stereo; 11 with Tobias, Joost; 8 with Tobias, Stereo, Joost and further 8 with Tobias, Stereo. There also seem to be 19 empty ballots (I thought it had been less).
If you look at uniqueness of ballots (as Rory made some statements regarding the possibility to sell your vote): 398 ballots are unique! Of those 78 had Tobias as option 1, 77 Joost as option 1, 73 had Miriam option 1 and still 66 had Stereo as option 1. So no statistical conspicuity for sold voted :-)
Further there had been 14 ballots naming only a single candidate: 5 of which named Tobias, 3 named Joost and 3 named Geoffrey, 2 named Miriam and 1 named Stereo.
Comment from imagico on 16 December 2018 at 17:27
The raw data is relatively easy to interpret: All lines containing a vote start with 1 so you can just grep for ^1 to get the data lines. After the 1 follows the vote with the candidates represented as numbers. The first empty place on the ballot is marked with a 0.
It is clear that since Tobias got by far the most first priority votes most of the identical ballots were with him on position one. Of the 189 people who voted Tobias first as i mentioned 82 voted Joost second and of these as you say 11 had no third choice and 29 had Stereo on third and then empty. This seems a pretty natural distribution if you take into account the political similarities (i.e. that people who voted for Tobias first have a higher preference for some candidates than for others).
Regarding “So no statistical conspicuity for sold voted” - that is not something you can necessarily observe as an outsider. The technique Rory discussed would be along the lines of “You should vote for A first and for us to verify you actually did so please fill the rest of your ballot with the following random sequence” Since there are 6!=720 possibilities for this specific combination together with the desired first position candidate you would have a relatively high risk of no other voter incidentally voting the same combination. But only the person who actually assigned someone to vote this way would be able to detect if the instructed voter obeyed the instruction.
Similar things apply for the possibility of collective voting instructions. Such instructions would usually call for who to vote for on position one and maybe two but there is not that much gain in instructing people to a specific whole sequence. The distribution of pairs of first and second choice is - as i analyzed - pretty broad and while there are combinations more frequent than others (which is natural given similarity and dissimilarity in what candidates represent) there is no single one that stands out specifically.
If you’d think about how much effort it would have taken to change the election to a different outcome - the distance between Joost and Miriam in the end was about 45 votes. That is is about how much votes you would have needed to add or remove to change the result - assuming that this specific change (Miriam instead of Joost) is what you want to accomplish. All other potential goals would have been much more expensive to accomplish.