By Clay Shentrup & Raphfrk, 13 April 2007
Haiku is an open source "clone" (currently in beta stage) of the now defunct BeOS (Be Operating System). An icon set contest was held some months back with approximately 2100 voters. They used Range Voting. But it wasn't just a poll – it actually determined which icon set would be used, so it was a "real election" hence with genuine incentive for strategic voting.
In order to try to keep the contest somewhat meaningful, the voters were to give a 0-4 score for 7 criteria, for each icon set – thus it can be viewed as a Range Voting election with a 0-28 total score (or as an RV election with 7 times more "voters" and 0-4 scores, except that each voter is really casting 7 votes).
The criteria:
The contest is here. The result dataset (reformatted) is here. There were 2138 voters and 16 "candidates" (A,B,...,P) and a lot of "abstentions." Problem is, we can't tell abstentions from 0's. (There are no 0's, so that's my assessment.)
It looks like it is just an internet poll. The "electorate" was just the people who visited the site, or were only developers or users connected with the project allowed to vote? The results were presumably available while the vote was ongoing.
Also, they list at the end both types of range result (abstain=0 and abstain = X). The overall results here can be dramatically different depending on how you treat abstentions. The biggest difference was for "Ikonki" which scored 75% of the winners score for abstain=X and 8.3% of the winner's score for abstain=0. He would have been eliminated by the quorum rule (if he didn't lose anyway).
The result would have been:
highest sum = 1.2, so eliminate all below 0.6
Actual Rank and average score shown
3) Stippi 3.4
1) Honey 3.3
4) zuMi 3.3
5) Mc Clintock 3.3
2) Tango 3.1
6) Dropline Neu 2.8
Would be eliminated by quorum rule
7) "575" 2.9So, averaging range with quorum would have elected a different winner.
The winner got an average of 24.29 (out of 28), with 512 total votes. Coming in close was a theme that got a 23.48 average, but with 790 votes.
The lowest rated candidate got an 8.19 from 74 voters. Coming in close was a theme that got an 8.73 from 190 voters.
Using the 50% total score quorum rule, only 5 of the 16 candidates were valid.
The valid candidate with the fewest votes received 409, with a 23.68 average.
The invalid candidate with the most votes received 393, with a 20.78 average.
The lowest rated valid candidate got a 21.98 average from 662 voters, and was also the first choice in the list.
The highest rated invalid candidate got a 21.06 average from 293 voters.
3 of the 5 valid candidates were the first 3 in the list.
29.25% of the ballots were approval-style. 24.01% were bullet votes i.e. plurality-style. So most of the strategizers thought it strategic to bullet vote? Apparently.
The biggest problem here was that non-votes are indiscernable from 0's. So if we look at the averages treating the non-votes as 0's instead of abstentions we get a different result:
The winner (who got second place previously) got an average of 8.75, and by far the most total points, with 18,552. Coming in second was a theme that got a 6.86.
The lowest rated candidate (which didn't change from above) got a 0.29. Coming in close was a theme that got a 0.76.
The valid candidate with the fewest points received 9,684 for a 4.57 average.
The invalid candidate with the most points received 8,675 for a 4.09 average.