All right, 12 responses are sufficient for me to do the analysis, so here it is...
The purpose of this experiment was twofold: 1) to see whether the presentation of the vote increased the "rationality" of cumulative count ballots; 2) to see whether cumulative count stacked up well against the two alternative methods most often promoted.
I'll deal with the second first. It appears that cumulative count performs much better than its critics expect.
Instant runoff voting (also called "alternative vote" and "single transferable vote") uses the rankings in the first column. It works by eliminating choices with the least amount of first-place votes and then allocating second preferences among those candidates left. Then the process is repeated, until a winner is found. When this done to the above ballots, New Hampshire and Montana are the two left standing at the end, and Montana wins handily, 8-4.
Under cumulative count, New Hampshire and Montana are again the top two, and Montana wins, 281-257. The third place choice is North Dakota with 179; under IRV North Dakota and Alaska are tied for third. Alaska is a close fifth under cumulative count, getting beaten by Idaho 121-120. So the cumulative count result closely matches the IRV result.
Under approval vote, which simply adds up the absolute ratings in the third column, Montana and New Hampshire are again the top two candidates, and again Montana wins, 860-731. However, Idaho sneaks all the way into third place with 715, not far behind New Hampshire. North Dakota and Alaska are fourth and fifth, with 685 and 625 respectively.
Apart from the anomaly of Idaho, cumulative count and approval vote are very, very similar. The Pearson r correlation between the two methods for all 10 candidate states was 0.89, which is quite good: 1.0 is a perfect correlation, and 0.0 means no correlation.
My conclusion: the hullabuloo over voting method is much ado about nothing. All 3 methods yield essentially the same result. (By the way, pure plurality voting yields a tie between Montana and New Hampshire for first-place votes. Plurality vote is perhaps the worst voting method we could possibly choose.
)
When it comes to rationality of the cumulative count ballots, however, there are some real puzzles, the solutions to which may have important implications for how the Research Committee designs the ultimate state ballot, and for how we go about educating people about cumulative count.
The main potential problem with just about any voting system is strategic voting. Strategic voting is the phenomenon whereby people refrain from voting for candidates that they think won't win, and instead vote for the lesser of the evils that they think can win.
[In what follows, please recognize that I am not criticizing how anyone voted. Your ballot is inviolate, and neither I nor anyone else has the right to criticize the way you choose to cast it. However, paradoxes in the way certain people voted are instructive for us in learning how to describe cumulative count and how to design the "real" ballot.]
Now then...the most obvious case of strategic voting was Eddie's ballot. He gave all his CC votes to 2 states, and 50 more CC votes to New Hampshire than to Delaware, even though Delaware for him was almost as good a candidate as New Hampshire (rating 98 versus 100). Alaska, Idaho, Vermont, Montana, Wyoming, and Maine did not get any votes from him even though he considered them to have some value as candidates (ratings>0). Phyllis' ballot was also apparently strategic: even though North Dakota and Idaho were almost as good for her as Montana, she gave 65 more CC points to Montana than to either of them. Two very instructive cases! Still, Eddie's and Phyllis' strategic voting was a lot less than what people have been doing on the practice poll on the State Data page. A few days ago I examined the individual votes in the practice poll - fully 42% of them gave all 10 of their points to a single candidate! This, of course, defeats the whole point of cumulative count. I'm heartened to see that no one here gave all 100 votes to a single state. There are two potential reasons for this: having people rank states made them think about candidates that were good but not their first choice, and giving people 100 points rather than 10 made them feel as if they could spread them out more. If this is true, we should do the same thing on the final ballot. Matt Cheselka is designing a new practice poll for the website to determine this more scientifically, over a bigger sample. But the results from this experiment are encouraging.
There were some minor paradoxes in almost everyone's ballot. Mike Lilback gave 15 CC points to Maine and 10 to Vermont even though they are equally good candidates according to his ratings. Michelle gave 5 points to 3 states even though 1 fell below the others in terms of rating. Robert gives 10 points to a state he rates 75 and 10 points to a state he rates 40 - wow! There seems to be a strong tendency to give "round numbers" in the CC vote (and in the ratings for that matter), even when people are clearly not indifferent among two or more candidates they may "round them both off" to the nearest 5.
Something everyone except Tim and Kim did was to give 0 points to more than 1 state. Theoretically, this shouldn't happen: you should want to give at least 1 point to your #9 choice in order to help it out, even if just a very little bit, over your #10 choice. (Tim's and Kim's ballots have their own paradox, in that they gave votes to their worst choices!
)
So I'm interested in hearing from people about some of these paradoxes, especially why you didn't give any CC votes to states that were low on your list but still rated much higher than your worst choice. Again, I'm not criticising the fact that you did this - far from it. But getting to know people's logic will help us figure out whether we need to put more explicit instructions in the final ballot, or undertake more education about how best to cast a cumulative count ballot given any array of preferences.
