- Four groups: L=Left, R=Right, N=North, S=South - Four candidates (or smaller groups): LN, RN, LS, RS - LN belongs to L and N, and similarly RN, LS and RS and belong to corresponding L, R, N and S groups - Votes: LN:30, RN:30, LS:30, RS:10
Case 1: One candidate elected - L gets more votes than R - N gets more votes than S - It makes sense to elect candidate LN (or candidate from group LN) Case 2: Two candidates elected - Electing RN and LS seems to be the best outcome - Electing LN and RN seems worse since N would get all the seats - Electing LN and LS seems worse since L would get all the seats - Electing LN and RS seems worse since RS got so few votes - Therefore it may be better not to elect LN although LN was maybe the best choice in the case where only one candidate was elected Pure serial (monotonic/serial) allocation of seats doesn't seem to work well in the described setting. Does this mean that in MultiGroup style of elections where candidates may represent multiple groups that all should be proportionally represented the Alabama Paradox (http://en.wikipedia.org/wiki/ Alabama_paradox) should be considered a feature whose presence is not a problem but maybe even a requirement? Juho ___________________________________________________________ Inbox full of spam? Get leading spam protection and 1GB storage with All New Yahoo! Mail. http://uk.docs.yahoo.com/nowyoucan.html ---- election-methods mailing list - see http://electorama.com/em for list info