The Distortion of Cardinal Preferences in Voting Ariel D. Procaccia and Jeffrey S. Rosenschein Lecture outline Introduction Distortion • Introduction to Voting • Distortion – Definition and intuition – Discouraging results • Misrepresentation – Definition and intuition – Results • Conclusions Misrepresentation Conclusions What is voting? Introduction Distortion Misrepresentation Conclusions • n voters and m candidates. • Each voter expresses ordinal preferences by ranking the candidates. • Winner of election determined according to a voting rule. – Plurality. – Borda. • Applications in multiagent systems (candidates are beliefs, schedules [Haynes et al. 97], movies [Ghosh et al. 99]). Got it, so what’s distortion? Introduction Distortion Misrepresentation Conclusions • Humans don’t evaluate candidates in terms of utility, but agents do! • With voting, agents’ cardinal preferences are embedded into space of ordinal preferences. • This leads to a distortion in the preferences. Distortion illustrated 11 Misrepresentation 1 11 c2 9 8 8 7 7 c3 2 5 utility 9 6 6 4 3 3 c1 2 1 0 1 2 5 4 2 c3 Conclusions 10 rank utility 10 Distortion c1 1 3 0 c2 3 rank Introduction Distortion defined (informally) Introduction Distortion Misrepresentation Conclusions • Candidate with max SW usually not the winner. – Depends on voting rule. • Informally, the distortion of a rule is the worst-case ratio between the maximal SW and SW of winner. Distortion Defined (formally) Introduction • • • Distortion Misrepresentation Each voter has preferences ui=<ui1,…,uim>; uij = utility of candidate j. Denote uj = i uij. Ordinal prefs denoted by Ri. j Ri k = voter i prefers candidate j to k. An ordinal pref. profile R is derived from a cardinal pref profile u iff: 1. i,j,k, uij > uik j Ri k 2. i,j,k, uij = uik j Ri k xor k Ri j • Conclusions (F,u) = maxjuj/uF(R). An unfortunate truth Distortion Introduction 1 1 5 c1 2 c2 1 3 c1 2 c2 1 c2 2 0 c2 c2 1 c1 c1 2 4 utility 3 utility 4 rank 4 0 c1 5 3 rank c1 5 utility Conclusions F = Plurality. argmaxjuj = 2, but 1 is elected. Ratio is 9/6. rank • Misrepresentation 2 1 c2 2 0 Distortion Defined (formally) Introduction • • • • • Distortion Misrepresentation Conclusions Each voter has preferences ui=<ui1,…,uim>; uij = utility of candidate j. Denote uj = i uij. Ordinal prefs denoted by Ri. j Ri k = voter i prefers candidate j to k. An ordinal pref. profile R is derived from a cardinal pref profile u iff: 1. i,j,k, uij > uik j Ri k 2. i,j,k, uij = uik j Ri k xor k Ri j. (F,u) = maxjuj/uF(R). nm(F)=maxu (F,u). – S.t. j uij = K. An unfortunate truth Distortion Introduction Misrepresentation Conclusions • Theorem: F, 32(F)>1. 1 1 5 2 c2 1 3 c1 2 c2 1 c2 2 0 utility c1 c2 c2 1 c1 c1 2 4 rank 3 utility 4 rank utility 4 0 c1 5 3 rank c1 5 2 1 c2 2 0 Scoring rules – a short aside Introduction Distortion Misrepresentation Conclusions • Scoring rule defined by vector = <1,…,m>. Voter awards l points to candidate l’th-ranked candidate. • Examples of scoring rules: – Plurality: = <1,0,…,0> – Borda: = <m-1,m-2,…,0> – Veto: = <1,1,…,1,0> Distortion of scoring rules – the plot thickens Introduction Distortion Misrepresentation Conclusions • F has unbounded distortion if there exists m such that for all d, nm(F)>d for infinitely many values of n. • Theorem: F = scoring protocol with 2 1/(m-1)l2l. Then F has unbounded distortion. • Corollary: Borda and Veto have unbounded distortion. An alternative model Introduction Distortion Misrepresentation Conclusions • So far, have analyzed profiles u s.t. i, juij=K. • Weighted voting: voter with weight K counts as K identical voters. • juij=Ki. Voter i has weight Ki. • Define nm(F) analogously to previous def. • Theorem: For all F, n1, m, n1m ≤ n1m, and there exists n2 s.t. n1m ≤ n2m. • Corollary: For all F, 32(F)>1. • Corollary: F has unbounded F has unbounded . Introducing misrepresentation Introduction Distortion Misrepresentation Conclusions • A voter’s misrepresentation w.r.t. l’th ranked candidate is ij = l-1. Denote j = i ij. • Misrep. can be interpreted as (restricted) cardinal prefs. – e.g. uij = m - ij - 1. • nm(F)=maxR (F(R)/minj j). Misrepresentation illustrated Introduction Distortion Misrepresentation Conclusions 9:00 9:00 9:00 9:00 10:00 10:00 10:00 10:00 11:00 11:00 11:00 11:00 12:00 12:00 12:00 12:00 13:00 13:00 13:00 13:00 14:00 14:00 14:00 14:00 15:00 15:00 15:00 15:00 16:00 16:00 16:00 16:00 17:00 17:00 17:00 17:00 18:00 18:00 18:00 18:00 19:00 19:00 19:00 19:00 Misrepresentation of scoring rules Introduction Distortion Misrepresentation Conclusions • Borda has misrepresentation 1. – Denote by lij candidate j’s ranking in Ri. – j’s Borda score is i(m-lij)=i(m-ij-1)=n(m-1)-iij=n(m-1)-j – j minimizes misrep. j maximizes score. – Borda has undesirable properties. • Scoring protocols with = 1 are fully characterized in the paper. • Theorem: F is a scoring rule. F has unbounded misrep. iff 1=2. – Corollary: Veto has unbounded misrep. Summary of misrepresentation results Introduction Distortion Misrepresentation Conclusions Voting Rule Misrepresentation Borda =1 Veto Unbounded Plurality = m-1 Plurality w. Runoff = m-1 Copeland m-1 Bucklin m Maximin 1.62 (m-1) STV 1.5 (m-1) Conclusions Introduction Distortion Misrepresentation Conclusions • Computational issues discussed in paper, but exact characterization remains open. • Distortion may be an obstacle for applying voting in multiagent systems. • If prefs are constrained, still an important consideration. – In scheduling example with m=3, in STV there might be 3 times as much conflicts as in Borda.