Commons talk:Sound Logo Vote/statistics

Closest bet

Interestingly, none of the individual votes got the final ranking exactly right. As far as I can tell, the nearest one was User:PBradley-WMF, whose ordering was VQ97 > AC54 > UN03 > GX13 > PK62 > FM76 > DS71 > BY23 > JW08 > OZ85. That is, the only difference was a swap between FM76 and DS71. Congrats on the winning bet! I guess this deserves a barnstar? 😁 --Waldyrious (talk) 14:49, 27 March 2023 (UTC)

User:Wanspirit tied! Nope, that actually tied with User:FOdeleye (WMF) here. 154.20.213.253 03:56, 28 March 2023 (UTC)

A barnstar for having the most distinctively average taste in sound logos. I'll take it! PBradley-WMF (talk) 14:04, 28 March 2023 (UTC)

Yayyyy! I want a barnstar. FOdeleye (WMF) (talk) 15:04, 28 March 2023 (UTC)

And you get a banrstar, and you get a barnstar, and you get a barnstar. MPourzaki (WMF) (talk) 16:59, 28 March 2023 (UTC)

Ooh, interesting. Waldyrious, just curious, how did you calculate this?

This brings up an interesting question: what's the optimal way of comparing an individual vote on this ballot to the final Schulze ranking? Given that it's just a list of pairwise comparisons, for example if I had voted consistent with final result except moved OZ85 to the front, it wouldn't be just one 'mistake', it would be 9 mistakes in the result of pairwise comparisons between OZ85 and the other 9 finalists. PBradley has just one of these mistakes and is certainly the winner regardless; because FM76 and DS71 are next to each other, the swap affects no other comparisons. It probably wouldn't be too difficult to write something in Python to rank votes by their 'mistakes'.

I wonder if the distribution of number of mistakes can be used as some sort of standard deviation-type metric for the extent to which the general voting populace agrees with the results. The margin of victory matrix provides insight into the comparisons between individual candidates, but there's no larger error metric that we can use. I'd also assume some inverse relationship between the mean number of mistakes and the net number of voters for each candidate n compared to candidate n-1, where n is its Schulze ranking. As in...if the margin of victory between each candidate n and n-1 is relatively large, I'd expect the mean number of 'mistakes' to be relatively small. The fun thing about voting methods such as Schulze's is that you get both an optimal result, and many ways to see the extent to which people agree with it. Best, RAdimer-WMF (talk) 16:49, 28 March 2023 (UTC)

Re-reading this...the distribution of number of mistakes wouldn't be a "standard deviation-type metric"; the standard deviation of the distribution of number of mistakes would simply be that metric. Though considering the likely pattern, something like skewness would be a better thing to look at. RAdimer-WMF (talk) 16:59, 28 March 2023 (UTC)

Here it is:

Top 16 by Kendall tau distance

Username	Distance
PBradley-WMF	1
Q2020	2
Paulina Studniczka (WMPL)	2
JAlberto (WMF)	2
D-Kuru	2
J. Patrick Fischer	2
Stekkmen	2
FOdeleye (WMF)	3
ILikelargeFries	3
Raymond august	3
Der rausch	3
Animataru	3
Al-Muqanna	3
Gennesect	3
Wanspirit	3
Szoltys	3
Tydalamb	3

(Top 16 because 17 through 32 is all a distance of 4)

Best, RAdimer-WMF (talk) 20:36, 29 March 2023 (UTC)