On Mon, 25 Mar 2002, Dave Ketchum wrote: > Thank you - this topic needs review occasionally. > > On Mon, 25 Mar 2002 03:13:05 +0000 MIKE OSSIPOFF wrote: > > > > > > > Maybe much improvement could be gotten by methods that use more > > than one balloting, especially more than 2 ballotings. Maybe > > idealness can be approached more if methods additionally use > > more than 1 kind of input.
Rob LeGrand recently experimented with simulations of repeated Approval balloting. He did both cumulative and non-cumulative versions. He found that when there is a CW, the non-cumulative version homes in on it quicker than the cumulative version, but they both eventually reach it. When there is no CW, then the cumulative version tends to give better results. Of course, a simulation of repeated balloting is not the same as repeated balloting itself, but presumably a voter doing repeated balloting would base her decisions on the results of the previous ballotings, her utilities for the various outcomes, and some kind of strategy. [Rob used the (admittedly non-optimal) strategy of approving down to the front runner, inclusive only if front runner is preferred to second placer.] So if voters were to submit their utility ratings and to select among several reasonable strategies, then the rest could be automated so as to avoid repeated trips to the polls, etc. This is the main idea behind Lorrie Cranor's Declared Strategy Voting (DSV). However, lamentably, in her dissertation she specialized to simulating repeated Plurality voting with only one available strategy. She did two versions of this repeated Plurality balloting simulation. One version could be described as random order cumulative, and the other as non-cumulative. The strategy was supposed to be near optimal for statistical information (but not detailed information) about the results of the previous steps (or step in the non-cumulative version). In my opinion, basing DSV on Approval would be a vast improvement over the Plurality based versions that Cranor explored. Cranor has put DSV on the back burner because of job and family, and it seems to me, because of unenthusiastic reception by participants in the study. Most of the students that participated in the study didn't understand that if they wanted to bullet vote for their favorite they could do so (simply by giving zero CR to all candidates except favorite). They didn't like the "machine" deciding when their vote would go for somebody other than their favorite, even though that was the purpose of the method, to help them figure out whom to vote for if not their favorite, and then to cast that vote for them, vicariously. They didn't like the stochastic aspect of the random version, either. Other than that they didn't have any consistent complaints. These complaints would be non-existent if DSV were based on Approval. No randomness is proposed or needed. No favorite betrayal is required. The "machine" automatically approves your favorite in every round; it just determines the approval cutoff according to your favorite strategy. If your CR ballot rates only at the extremes, then the DSV machine will cast the same Approval ballot for you every time, no matter which strategy you might have indicated as your favorite strategy. I think the best way to introduce the public to any form of repeated balloting or simulations thereof, would be through Proxy Repeated Balloting: You vote your favorite as your proxy. You participate in the repeated balloting vicariously for a couple of elections. Then you watch the live repeated ballotings turn into simulated versions based on candidate CR schedules and strategy choices. Next you start to get more options on your own ballot, allowing you to submit information to influence more directly the outcome. Ultimately, you submit your CR ballot and check off your preferred strategy for the DSV simulation, or continue to vote above the line if you don't want to go to all of that trouble. Forest
