scieee Open visual document viewer

Group strategy-proof social choice functions

Barberà, Salvador; Berga, Dolors; Moreno, Bernardo

Abstract

We define different concepts of group strategy-proofness for social choice functions. We discuss the connections between the defined concepts under different assumptions on their domains of definition. We characterize the social choice functions that satisfy each one of them and whose ranges consist of two alternatives, in terms of two types of basic properties.

Full text

G oup s a egy-p oo social choice unc ions wi h bina y anges and a bi a y domains: cha - ac e iza ion esul s1 Sal ado Ba be ày Dolo s Be gaz and Be na do Mo enox Ap il 28 h, 2010 Abs ac : We de…ne di¤e en concep s o g oup s a egy-p oo ness o social choice unc ions. We discuss he connec ions be ween he de…ned concep s unde di¤e en assump ions on hei domains o de…ni ion. We cha ac e ize he social choice unc ions ha sa is y each one o hem and whose anges consis o wo al e na i es, in e ms o wo ypes o basic p ope ies. JEL Classi…ca ion Numbe : D71. Keywo ds: Social choice unc ions, Bina y anges, g oup s a egy-p oo ness, xy-mono onici y, xy-based ules. 1We would like o hank he commen s o Wal e Bosse and Luis Co chón. Sal ado Ba be à g a e ully acknowledges suppo om he Spanish Minis y o Science and Inno- a ion h ough g an "Consolida ed G oup-C" ECO2008-04756, and by he Gene ali a de Ca alunya, Depa amen d’Uni e si a s, Rece ca i Socie a de la In o mació h ough he Dis inció pe a la P omoció de la Rece ca Uni e si à ia, g an SGR2009-0419 and he Ba celona GSE Resea ch Ne wo k. Dolo s Be ga acknowledges he suppo o he Spanish Minis y o Science and Inno a ion h ough g an SEJ2007-60671 and o Gene ali a de Ca alunya, h ough g an SGR2009-0189. She also acknowledges he Resea ch Recogni- ion P og amme o he Ba celona GSE. Be na do Mo eno acknowledges …nancial suppo om he Spanish Minis y o Science and Inno a ion h ough g an ECO2008-03674. yMOVE, Uni e si a Au ònoma de Ba celona, and Ba celona GSE, Edi…ci B, 08193 Bella e a, Spain. E-mail: sal ado .ba b[email p o ec ed] zDepa amen d’Economia, Campus de Mon ili i, Uni e si a de Gi ona, 17071 Gi ona, Spain. E-mail: dolo s.b[email p o ec ed] xDepa amen o de Teo ía e His o ia Económica, Facul ad de Ciencias Económicas y Emp esa iales, Campus de El Ejido, 29071 Málaga, Spain. E-mail: b[email p o ec ed] 1 In oduc ion The Gibba d-Sa e hwai e Theo em es ablishes ha , when a social choice unc ion is de…ned on he uni e sal se o p e e ence p o…les o e kal e na- i es (k > 2), and i s ange con ains a leas h ee al e na i es, i can only be s a egy-p oo i i is dic a o ial. This esul is subjec o di¤e en quali…ca ions. One is ha , when he ule is de…ned on smalle se s o p o…les, he e may o may no exis o he ules ha a e s a egy-p oo , in addi ion o he dic a o ial ones. This is he case unde a a ie y o domains, ha include he ones o med by he Ca e- sian p oduc o single-peaked p e e ences, o o single-dipped p e e ences, o o sepa able p e e ences, among o he s. Ou s a emen s in his pape will be essen ially ue o unc ions de…ned on any domain, howe e small, asym- me ic o special i may be.1 A second quali…ca ion conce ns he ange o he social choice unc ion. In his pape we conside he subclass o unc ions ha ail o mee Gibba d and Sa e hwai e’s equi emen ha hei ange should con ain a leas h ee al e na i es. Speci…cally, we concen a e on ules ha a e no cons an and whose ange consis s o exac ly wo al e na i es, xand y. Because o his, i is known ha in ha case he e a e possibili ies o design non-dic a o ial s a egy-p oo ules. We wan o cha ac e ize hem all. This leads us o no ice ha he ange o a social choice unc ion may be bina y because he e a e only wo al e na i es in he ele an wo ld, bu i may also be bina y in he p esence o mo e han wo al e na i es, in which case his may be conside ed as one pa o he possible choices open o he mechanism designe . As we shall see, he cha ac e iza ion o bina y ules in his con ex equi es a numbe o p ecisions and ca e ul ea men ha can be a oided in wo lds whe e only wo al e na i es a e p esen o begin wi h. A hi d quali…ca ion e e s o he no ion o s a egy-p oo ness o be used. When we concen a e on ules wi h bina y anges, he e exis a numbe o a ac i e s a egy-p oo ules, and i becomes hen much mo e in e es ing o explo e he ex en o which some o hem may also be immune o ma- nipula ion by g oups. We analyze his ques ion ca e ully, unde a numbe o di¤e en possible no ions o g oup s a egy-p oo ness, and also by keeping in mind ha we wan ou s a emen s o hold o unc ions de…ned on any ype 1By "essen ially ue" we mean ha hey a e ei he ue wi hou quali…ca ion, o ue unde e y mino assump ions, o be discussed case by case. 1 o domains. One de…ni ion o g oup s a egy-p oo ness equi es ha i should no be possible o a g oup o agen s o de ia e om decla ing hei ue p e e ences and ge a s ic gain o each one o hem. Social choice unc ions a oiding his s ong ype o manipula ion a e called Weakly G oup S a egy-P oo . A second de…ni ion s a s om conside ing ha a g oup can p o… ably de i- a e i some o i s membe s de i e a s ic gain om doing so, while o he s simply emain indi¤e en while helping hei pa ne s. Rules ha a oid his weake o m o manipula ion a e called S ongly G oup S a egy-P oo . In an in e media e e sion o he p ope y, ha we simply call G oup S a egy- P oo ness, we allow ha only some agen s may gain om he de ia ion, bu we equi e ha all agen s in ol ed in ge ing he change should ac i ely pa icipa e in he manipula ion by ac ually de ia ing om hei u h ul p e - e ence. We p o ide cha ac e iza ions o he classes o social choice unc ions ha sa is y each one o hese h ee p ope ies, and also we elabo a e on why we single ou hese pa icula de…ni ions. Ou main cha ac e iza ion esul s iden i y wo ypes o basic p ope ies ha hese ules mus sa is y. These p ope ies mus be quali…ed in each case. Since we allow indi iduals o be indi¤e en be ween xand y, in some cases we will equi e ha hey a e sa is…ed "essen ially", and in o he cases no . By "essen ially" we mean ha he p ope ies will hold condi ional o he ac ha he p e e ences o indi iduals ha a e indi¤e en be ween he wo al e na i es in he ange emain cons an . Ou … s condi ion is ha o essen ial xy-mono onici y: i xob ains a a p o…le, and hen some people change hei p e e ences so ha he suppo o xinc eases, while he suppo o ydoes no , hen xmus s ill ob ain a he new p o…le. A mo e demanding equi emen in a simila spi i is ha o xy-s ong mono onici y. In ha case i xob ains a a p o…le, and p e e ence changes induce la ge suppo o x; hen xs ill be chosen a he new p o…le e en i suppo o ymay ha e also inc eased.2A second ype o equi emen e e s o he ype o in o ma ion on which ou ules may be based. We say ha hey a e xy-based i wha hey choose a each p e e ence p o…le only depends on he ela i e posi ion o x wi h espec o y o each indi idual. I is essen ially xy-based i he p ope y holds when we only compa e p o…les whe e indi iduals indi¤e en be ween 2In his second de…ni ion we d op he quali…ca ion o he p ope y being essen ial because he s a emen is no longe condi ioned o he p e e ences o indi¤e en indi iduals emaining cons an . 2 xand ykeep hei p e e ences unchanged. No ice also ha he equi emen will no apply in he case whe e all indi iduals a e indi¤e en be ween bo h al e na i es in he ange. We es ablish h ee cha ac e iza ion esul s in e ms o he abo e condi- ions, one o each o ou h ee ypes o g oup s a egy-p oo ness equi e- men s. A social choice unc ion is weakly g oup s a egy-p oo i and only i i is essen ially xy-based and essen ially xy-mono onic. I is s ongly g oup s a egy-p oo i and only i i is xy-based and xy-s ong mono onic. Finally, we show ha , when n3and unde a mild condi ion on he ichness o he domain, ules ha mee ou in e media e no ion o g oup s a egy-p oo ness a e also s ongly g oup s a egy-p oo , and hus sa is y he same p ope ies. The sophis ica ed eade will ealize ha ou condi ions a e pa o a la ge se o di¤e en equi emen s ha ha e been used by di¤e en au ho s unde di¤e en names o he cha ac e iza ion o s a egy-p oo ules o e uni e sal domains. Names like Maskin mono onici y, s ong posi i e associa ion, and o he s ha e been used o deno e a ia ions o p ope ies ha one expec s o be sa is…ed by ules ha a e s a egy-p oo . And, indeed, many combina ions o p ope ies end up cha ac e izing he same ules when hese a e de…ned on ich enough domains. We eel ha ou choice o p ope ies is especially … , because hey allow us o cha ac e ize ules de…ned on all kinds o domains, possibly e y asymme ic and con aining ew p e e ences. The equi alence be ween ou s and o he p ope ies is no g an ed unde hese ci cums ances. Also no ice ha we do no insis on indi idual s a egy-p oo ness as a special case o cha ac e ize. This is because by a ecen esul o ou s, i is an es ab- lished ac ha indi idual and weak g oup s a egy-p oo ness a e equi alen when he ange o he social choice unc ion consis s o only wo (o h ee) elemen s (see Ba be à, Be ga, and Mo eno, 2010). A di¤e en ype o cha ac e iza ion esul s a e based on desc ip ions o how he ules would choose al e na i es a each p e e ence p o…le. The e exis wo ele an pape s ha ake his poin o iew. One is by La sson and S ensson (2006), who p o ide a cha ac e iza ion o s a egy-p oo ules: unde ou assump ion ha he ange is bina y, s a egy-p oo ness is equi - alen o weak g oup s a egy-p oo ness, as p o en in Ba be à, Be ga, and Mo eno (2010). Hence, hei cha ac e iza ion in e ms o he unc ional o m p o ides an al e na i e o he one we p esen he e. A second esul , his one due o Manjuna h (2009a), cha ac e izes he unc ional o m o s ong g oup s a egy-p oo ules when he e a e only wo al e na i es. We e-s a e he esul wi h some addi ional p ecisions and in o de o co e he case whe e 3 he ange is bina y bu p e e ences a e de…ned on a la ge se o al e na i es, and p o ide a no el p oo o i . The pape p oceeds as ollows. In Sec ion 2, we p o ide he amewo k, we p esen di¤e en e sions o g oup s a egy-p oo ness and discuss hei ela ionships unde di¤e en domain assump ions. In Sec ion 3 we p o ide he cha ac e iza ions in e ms o p ope ies. In Sec ion 4 we p o ide he announced addi ional cha ac e iza ion o s ongly g oup s a egy-p oo ules, he no el p oo , ha also allows us o comple e he p oo o one o he heo ems in he p eceding sec ion. Sec ion 5 concludes. 2 The se up and de…ni ions Le Abe a …ni e se o al e na i es A= x; y; z; w:::g:Le Nbe a …ni e se o agen s N= 1;2; :::; ng:Le Ube he se o all p eo de s on A(comple e, e‡exi e, and ansi i e bina y ela ions on A). Le Ri U be he se o admissible p e e ences o agen i2Nand le R  i2NRi. Fo any p e e ence ela ion Ri2 Ri, we deno e by Piand Ii he s ic and indi¤e ence pa o Ri, espec i ely. A p e e ence p o…le is deno ed by R= (R1; ::; Rn)2 R o also by R= (RC; RC)2 R when we wan o s ess he ole o a coali ion CN. Then RC2 RC i2CRiand RC2 RNnC deno e he p e e ences o agen s in Cand in NnC, espec i ely. Asocial choice unc ion (o ule)on a domain Ris a unc ion :R ! A. The ange o is deno ed by A . In his pape we concen a e on he amily o social choice unc ions wi h bina y ange, ha is, whose ange consis s o exac ly wo elemen s, ha we call xand y om now on. Le Rx ijRibe he subse o p e e ences such ha o any Rx i2 Rx i, xPx iy. Simila ly, de…ne Ry i. Le Rxy ijRibe he subse o p e e ences such ha o any Rxy i2 Rxy i,xIxy iy. We s a e ou esul s unde he ollowing minimal assump ion on he domain o admissible p e e ences: each indi idual has a leas one admissi- ble p e e ence whe e xis p e e ed o y, one whe e yis p e e ed o x; and one whe e he is indi¤e en be ween he wo. Tha is, o any i2Nand any 2 x; y; xyg,R i6=?.3 3Fo se e al o ou esul s, we could e en weaken his minimal condi ion on he domain and allow o some o he se s R i o be emp y. 4 The bes known nonmanipulabili y axiom is s a egy-p oo ness. I e- qui es he u h o be a dominan s a egy and i is a necessa y condi ion o implemen a ion in dominan s a egies (Gibba d, 1973 and Sa e hwai e, 1975). De…ni ion 1 An agen i2Ncan manipula e a social choice unc ion on Ra R2 R i he e exis s R0 i2 Risuch ha Ri6=R0 iand (R0 i; Ri)Pi (R). A social choice unc ion is s a egy-p oo on Ri no agen i2Ncan manipula e on R. Ano he o m o manipula ion is by means o coali ions. The ollowing de…ni ions e e o cases whe e agen s may gain om join changes o de- cla ed p e e ences. They di¤e on wo accoun s: he equi ed gains om manipula ion and he ac ions expec ed om coali ion membe s. Rega ding gains om manipula ion we may equi e ha each membe om de ia ing coali ions ob ains a s ic gain o else ha only some o hem do wi h he es no losing. Rega ding de ia ions we may ask ha all membe s o a coali- ion mis ep esen hei p e e ences o ha jus some o hem do. The h ee de…ni ions below will e‡ec hese modelling choices.4 De…ni ion 2 A coali ion Ccan s ongly manipula e a social choice unc ion on Ra R2 R i he e exis s R0 C2 RCsuch ha o all agen i2C, Ri6=R0 iand (R0 C; RC)Pi (R). A social choice unc ion is weakly g oup s a egy-p oo on Ri no coali ion CNcan s ongly manipula e on R. De…ni ion 3 A coali ion Ccan manipula e a social choice unc ion on R a R2 R i he e exis s R0 C2 RCsuch ha o all agen i2C,Ri6=R0 i and (R0 C; RC)Ri (R), and o some j2C, (R0 C; RC)Pj (R). A social choice unc ion is g oup s a egy-p oo on Ri no coali ion CNcan manipula e on R. De…ni ion 4 A coali ion Ccan weakly manipula e a social choice unc ion on Ra R2 R i he e exis s R0 C2 RCsuch ha o some agen l2C, 4We shall omi wha could ha e been a ou h e sion o g oup s a egy-p oo ness, one ha would equi e all agen s o gain bu would allow o some o hem no o change hei p e e ences. Tha would u n ou o be equi alen o weak g oup s a egy-p oo ness (see De…ni ion 2). 5 Rl6=R0 l, o all agen i2C, (R0 C; RC)Ri (R), and o some j2C, (R0 C; RC)Pj (R). A social choice unc ion is s ongly g oup s a egy- p oo on Ri no coali ion CNcan weakly manipula e on R. Rema ks (1) S a egy-p oo ness and weak g oup s a egy-p oo ness a e equi alen o social choice unc ions wi h bina y ange (see P oposi ion 1 and Theo em 1 in Ba be à, Be ga, and Mo eno, 2010). (2) When indi¤e ences a e no allowed, all h ee de…ni ions o g oup s a egy- p oo ness collapse in a single one. (3) S ong g oup s a egy-p oo ness implies g oup s a egy-p oo ness and he la e implies weak g oup s a egy-p oo ness. The con e se implica ions do no hold in gene al, as shown by he ollowing examples. Example 1 A ule ha is g oup s a egy-p oo bu no s ongly. Le n2 and #A2,x; y 2A. Then, o any R2 UN, de…ne he social choice unc ion as ollows: (R) = xi xPiy o any i2N, yo he wise. We show ha is no s ongly g oup s a egy-p oo . Le Rbe such ha each agen s ic ly p e e s x o yand le R0be such ha n1agen s s ic ly p e e xo e y, and he o he agen is indi¤e en be ween xand y. Obse e ha (R) = xand (R0) = y: Then, coali ion Ncould weakly manipula e a R0 ia R. The eade may check ha he ule sa is…es he wo weake s a egic condi ions. Example 2 A ule ha is weakly g oup s a egy-p oo bu no g oup. Le n2,#A2and agen s’p e e ences such ha o any i2N,R i6=? o any 2 x; y; xyg. Le kbe a dic a o on x; yg, ha is, (R) = xwhen Rk2 Rx k[ Rxy kand (R) = yo he wise. No e ha is (weakly g oup) s a egy-p oo . Howe e , coali ion C= k; jg j6=kcould manipula e a (Rxy k; Ry j; R j;kg) ia (Ry k; R0 j; R j;kg) o any R0 j2 Rx i[ Rxy iand any R j;kg2 RNn j;kg. Thus, is no g oup s a egy- p oo ( hus no s ongly). Be o e cha ac e izing he ules ha sa is y ou di¤e en equi emen s, le us ema k ha g oup s a egy-p oo ness and s ong g oup s a egy-p oo ness become equi alen unde he mild complemen a y domain condi ion e- qui ed in he ollowing p oposi ion. 6 P oposi ion 1 Le #A3and Rbe such ha each indi idual has a leas wo admissible p e e ences in Riwhe e xis p e e ed o yand wo whe e y is p e e ed o x: Then, any g oup s a egy-p oo social choice unc ion on Rwi h a bina y ange is also s ongly g oup s a egy-p oo . P oo . Le be a g oup s a egy-p oo social choice unc ion. Suppose ha is no s ongly g oup s a egy-p oo . Tha is, he e exis R2 R, a coali ion CN, and R0 C2 RCsuch ha o some agen l2C,Rl6=R0 l, o all agen s i2C; (R0 C; RC)Ri (R), and o some j2C, (R0 C; RC)Pj (R). I o any agen l2C,Rl6=R0 l, hen we ge a con adic ion o g oup s a egy- p oo ness. Thus, he e exis l2Csuch ha Rl=R0 l:De…ne CP= j2C:Rj=R0 jand (R0 C; RC)Pj (R)gand CI= k2C:Rk=R0 kand (R0 C; RC)Ik (R)g. By he complemen a y domain condi ion, o any j2CP, he e exis s R00 j2 RjnRjsuch ha (R0 C; RC)P00 j (R). I (R00 CP; R0 CnCP; RC) = (R) he e exis a coali ion CP, a p o…le (R00 CP; R0 CnCP; RC)2 R, and R0 CP=RCPsuch ha o any agen j2CPR00 j6=Rjand (R0 C; RC)Pj (R00 CP; R0 CnCP; RC) = (R)which is a con adic ion o g oup s a egy-p oo ness. Thus, (R00 CP; R0 CnCP; RC) = (R0 C; RC): I CI=? hen obse e ha he e exis R2 R, a coali ion CN, and R00 C(R00 CP; R0 CnCP)2 RCsuch ha o any agen i2C,Ri6=R00 iand (R00 CP; R0 CnCP; RC)Ri (R), and o some j2C, (R00 CP; R0 CnCP; RC)Pj (R). Then we ge a con adic ion o g oup s a egy-p oo ness. Thus, CI6=?. By he complemen a y domain condi ion, o any k2 CI, he e exis s R00 k2 RknRksuch ha (R00 CP; R0 CnCP; RC)P00 k (R). I (R00 CP[CI; R0 Cn(CP[CI); RC) = (R)coali ion CIcould manipula e ia RCI a (R00 CP[CI; R0 Cn(CP[CI); RC), which con adic s g oup s a egy-p oo ness. Thus, (R00 CP[CI; R0 Cn(CP[CI); RC) = (R0 C; RC). Then obse e ha he e exis R2 R, a coali ion CN, and R000 C(R00 CP[CI; R0 Cn(CP[CI))2 RCsuch ha o any agen i2C,Ri6=R000 iand (R00 CP[CI; R0 Cn(CP[CI); RC)Ri (R), and o some j2C, (R00 CP[CI; R0 Cn(CP[CI); RC)Pj (R), and hen we ge a con adic ion o g oup s a egy-p oo ness. Rema k 1 We ha e assumed in P oposi ion 1 ha #A3:This is because he complemen a y domain condi ion ha we assume in ou s a emen can only be sa is…ed in his case. When #A= 2, his condi ion canno be sa is…ed 7 and in ac he equi alence does no hold ( he ule in Example 1 when #A= 2 p o ides a coun e example). 3 Cha ac e iza ion esul s: p ope ies In his sec ion we p o ide ou … s se o cha ac e iza ion esul s. We p o e ha ou di¤e en e sions o he condi ion ha a ule should be xy-based and mono onic a e necessa y and su¢ cien o gua an ee ha hey sa is y ou di¤e en e sions o g oup s a egy-p oo ness.5 Fo each p e e ence p o…le R2 R;de…ne he se X(R) = i2N:xPiyg; Y(R) = j2N:yPjxg, and I(R) = k2N:yIkxg. We now de…ne he condi ions ha will cha ac e ize weak and s ong g oup s a egy-p oo ness. De…ni ion 5 A bina y social choice unc ion is essen ially xy-mono onic6 i and only i o any R; R02 R such ha Rh=R0 h o all h2I(R) I(R0); he ollowing holds: [X(R0)X(R); Y (R)Y(R0)(a leas one s ic inclusion), and (R) = x] ) (R0) = x; and [Y(R0)Y(R); X(R)X(R0)(a leas one s ic inclusion), and (R) = y] ) (R0) = y: De…ni ion 6 A bina y social choice unc ion is xy-s ongly mono onic i and only i o any R; R02 R he ollowing holds: [i ei he X(R0)X(R); Y (R)Y(R0)(a leas one s ic inclusion), o X(R0)X(R),?6=Y(R)$Y(R0)] and (R) = x) (R0) = x; [i ei he Y(R0)Y(R); X(R)X(R0)(a leas one s ic inclusion), o Y(R0)Y(R),?6=X(R)$X(R0)] and (R) = y) (R0) = y. 5Examples showing he ela ionship be ween he p ope ies de…ned in his sec ion a e a ailable upon eques . 6Lemma 7 in Manjuna h (2009b) shows ha when he se o admissible p e e ences is he se o all single-dipped p e e ences and a speci…c bina y ange es ic ion, some e sion o essen ially xy-mono onici y is a consequence o s a egy-p oo ness. 8 be ween xand y. Bu hen he second agen ob ains his bes ou come. This ends he p oo o S ep 1. S ep 2 Any s ongly g oup s a egy-p oo social choice unc ion wi h bina y ange is xy-based and xy-s ong mono onic. P oo o S ep 2: By Theo em 1, we know ha is essen ially xy-based and essen ially xy- mono onic. We now p o e by con adic ion ha is xy-based. Suppose no , hen he e exis R; R02 R such ha X(R)[Y(R)6=?,X(R) = X(R0); Y (R) = Y(R0), (R)6= (R0). Suppose … s ha X(R) = ?and hus Y(R)6=?(a simila a gumen applies i Y(R) = ?). By Lemma 1, (R) = (R0) = ywhich is a con adic ion. Thus, X(R)6=?and Y(R)6=?. By essen ially xy-based, (R0 X(R)[Y(R); RI(R)) = (R). I n= 2,R0= (R0 X(R)[Y(R); RI(R))and we ge he desi ed con adic ion since (R0)mus be di¤e en om (R). I n3, since (R)6= (R0) he e mus exis an agen i2Nsuch ha xIiy ha is xy-pi o al o (R0 X(R)[Y(R); RI(R)). By Lemma 2, is no s ongly g oup s a egy-p oo which is a con adic ion. We now p o e by con adic ion ha is xy-s ong mono onic. Suppose no , ha is he e exis R; R02 R such ha ei he (1) X(R0)X(R), Y(R)Y(R0)(a leas one inclusion s ic ), (R) = xbu (R0) = y; o else (2) X(R0)X(R),?6=Y(R)$Y(R0); (R) = xand (R0) = y. A simila a gumen holds o he o he possibili y whe e he oles o xand y a e exchanged. Fi s obse e ha by Lemma 1, X(R)6=?and Y(R)6=? (o he wise, i Y(R) = ?, hen Y(R0) = ?and X(R0)%X(R). By Lemma 1, (R0) = xwhich is he desi ed con adic ion. I X(R) = ?and Y(R)6=?, hen by Lemma 1 (R) = ywhich is he desi ed con adic ion). I case (1) holds, by essen ial xy-mono onici y and essen ial xy-basedness, (R0 X(R)[Y(R); RI(R)) = x(de…ne R00 = (R0 X(R)[Y(R); RI(R)), i ei he X(R00)% X(R)and Y(R00)$Y(R)o X(R00) = X(R)and Y(R00)$Y(R)we apply essen ial xy-mono onici y. I X(R00) = X(R)and Y(R00) = Y(R)we apply essen ial xy-basedness). I n= 2,R0= (R0 X(R)[Y(R); RI(R))and we ge he desi ed con adic ion since (R0)mus be di¤e en om (R). I n3, since (R)6= (R0) he e mus exis an agen i2Nsuch ha xIiy ha is xy-pi o al o (R0 X(R)[Y(R); RI(R)). By Lemma 2, is no s ongly 15 g oup s a egy-p oo which is a con adic ion. I case (2) holds, by essen ial xy-mono onici y and essen ial xy-basedness, (R0 X(R0); RNnX(R0)) = x(i X(R0)%X(R), ha is, X(R0)includes some agen s in I(R), we apply essen ial xy-mono onici y. I X(R0) = X(R)we apply essen ial xy-basedness). Then, (R0 X(R0); R0 Y(R); RNn X(R0)[Y(R)g) = x by essen ial xy-basedness. I n= 2,R0= (R0 X(R0); R0 Y(R); RNn X(R0)[Y(R)g)and we ge he desi ed con- adic ion since (R0)mus be di¤e en om (R). I n3, since (R)6= (R0) he e mus exis an agen i2N ha is xy- pi o al agen o (R0 X(R0); R0 Y(R); RNn X(R0)[Y(R)g)such ha xIiy. By Lemma 2, is no s ongly g oup s a egy-p oo which is a con adic ion. This ends he p oo o S ep 2. S ep 3 Any xy-based and xy-s ong mono onic social choice unc ion wi h bina y ange can be desc ibed as a e o ule when n3:When n= 2, is ei he a e o ule o a se ial dic a o . To show S ep 3, we use he ollowing claims. Obse e … s ha since is xy-based hen o any R i,R i2 R i, (R i; Ri) = (R i; Ri) o any Ri2 RNn igwhe e 2 x; y; xyg. In wha ollows, when we use R iwe e e o any R i2 R iwi hou loss o gene ali y. This is because all he s a emen s we make in his p oo om now on hold wha e e he ep esen a i e o he se R iis. Claim 1 Le n2. I is xy-based and xy-s ong mono onic hen is xy-Pa e ian. P oo o Claim 1 Le Rx2 i2NRx i, ha is, X(Rx) = N. Suppose o ge a con adic ion ha (Rx) = y. No e ha by xy-based and xy-s ong mono onici y, o any o he p o…le R2 R, (R) = ywhich con adic s ha has a bina y ange. Thus, (Rx) = x. Suppose ha he e is Rsuch ha xRiy o any i2Nand xPjy o some j2N; X(R)6=Nand (R) = y. Then X(R)6=?and Y(R) = ?. No e ha X(R)$X(Rx)and Y(R) = Y(Rx), o any Rx2 i2NRx i. By xy- s ong mono onici y, (Rx) = ywhich con adic s wha we ha e jus p o ed. This ends he p oo o Claim 1. No e ha he coun e pa esul s o Claims 2 and 3 below exchanging he oles o xand ydo also hold. Claim 2 Le n2. I o some i2Nand some Ry i2 Ry i; (Ry i; Rx i) = y, 16 hen (Ry i; R0 i) = y o any R0 i2 RNn ig: P oo o Claim 2 Le R0 i2 RNn ig. Obse e ha (Ry i; R0 i) = yei he by xy-based i X(Ry i; R0 i) = Nn ig=X(Ry i; Rx i), o else by xy-s ong mono onici y i X(Ry i; R0 i)$Nn ig=X(Ry i; Rx i). This ends he p oo o Claim 2. Claim 3 Le n3. I o some i2Nand some Ry i2 Ry i; (Ry i; Rx i) = y, hen o any j2Nwe ha e ha (Ry j; Rx j) = y. P oo o Claim 3 By con adic ion, suppose ha (Ry i; Rx i) = yand (Ry j; Rx j) = x. I (Rxy i; Ry j; Rx  i;jg) = y hen (Ry j; Rx j) = yby xy-s ong mono onici y since ?6=X(Rxy i; Ry j; Rx  i;jg)$X(Ry j; Rx j)and Y(Rxy i; Ry j; Rx  i;jg) = Y(Ry j; Rx j). Thus, (Rxy i; Ry j; Rx  i;jg) = x. By xy-s ong mono onici y, (Ry i; Ry j; Rx  i;jg) = x, since X(Rxy i; Ry j; Rx  i;jg) = X(Ry i; Ry j; Rx  i;jg)and ?6=Y(Rxy i; Ry j; Rx  i;jg)$Y(Ry i; Ry j; Rx  i;jg). By Claim 2, since (Ry i; Rx i) = y hen (Ry i; Ry j; Rx  i;jg) = ywhich con adic s wha we ob ained abo e. This ends he p oo o Claim 3. Claim 4 Le n3. I o some i2Nand some Ry i2 Ry i, (Ry i; Rx i) = x hen (Ry C; Rx C) = x o any C,? $ C$N. P oo o Claim 4 Suppose, o ge a con adic ion, ha o some C,? $ C$N; (Ry C; Rx C) = y. Clea ly, C6= ig. No e also ha Ccan no be a single on (o he wise, i C= jg,j6=i, we would ge a con adic ion by Claim 3). Thus, #C > 1. No e also ha (Ry k; Rx k) = x o any k2N (o he wise, (Ry k; Rx k) = y, by Claim 4, (Ry i; Rx i) = ywhich is no he case). The e o e, wi hou loss o gene ali y, we can suppose ha i2C. We dis inguish wo subcases: Subcase 1 Le (Ry i; Rxy Cn ig; Rx C) = x. No e ha X(Ry i; Rxy Cn ig; Rx C) = X(Ry C; Rx C)and ?6=Y(Ry i; Rxy Cn ig; Rx C)$Y(Ry C; Rx C). The e o e, by xy- s ong mono onici y (Ry C; Rx C) = x, which is a con adic ion. Subcase 2 Le (Ry i; Rxy Cn ig; Rx C) = y. No e ha Y(Ry i; Rxy Cn ig; Rx C) = Y(Ry i; Rx i)and ?6=X(Ry i; Rxy Cn ig; Rx C)$X(Ry i; Rx i). The e o e, by xy- s ong mono onici y, (Ry i; Rx i) = y, which is a con adic ion. This ends he p oo o Claim 4. P oo o S ep 3: Fi s , by Claim 1, (R) = x o any Rsuch ha xRiy o any i2Nand xPjy o some j2Nand (R) = y o any Rsuch ha yRix o any i2N and yPjx o some j2N. Second, (R)can be any ou come o any Rwhe e 17 all agen s a e indi¤e en . Thi d, he a gumen di¤e s depending on nbeing wo o highe . I n= 2, suppose … s ha is such ha o some p o…le (Ry 1; Rx 2), whe e Ry 12 Ry 1and Rx 22 Rx 2, (Ry 1; Rx 2) = yand o some p o…le (Ry 2; Rx 1), whe e Ry 22 Ry 2and Rx 12 Rx 1, (Ry 2; Rx 1) = y. By Claim 2, o any Ry 12 Ry 1and Rx 22 Rx 2, (Ry 1; R2) = yand (Ry 2; R1) = y o any R22 R2and R12 R. Thus, can be ew i en as a e o ule o x. Second, suppose ha o some p o…le (Ry 2; Rx 1), whe e Ry 22 Ry 2and Rx 12 Rx 1, (Ry 2; Rx 1) = yand o any p o…le (Ry 1; Rx 2), whe e Ry 12 Ry 1and Rx 22 Rx 2, (Ry 1; Rx 2) = x. By Claim 2, o any Ry 22 Ry 2; (Ry 2; R1) = yand o any R12 R1. No e ha his ule can be ew i en as a se ial dic a o wi h o de 21. Thi d, suppose ha o some p o…le (Ry 1; Rx 2), whe e Ry 12 Ry 1and Rx 22 Rx 2, (Ry 1; Rx 2) = yand o any p o…le (Ry 2; Rx 1), whe e Ry 22 Ry 2and Rx 12 Rx 1, (Ry 2; Rx 1) = x. By Claim 2, o any Ry 12 Ry 1; (Ry 1; R2) = yand o any R22 R2. No e ha his ule can be ew i en as a se ial dic a o wi h o de 12. Finally, suppose ha o any p o…le (Ry 2; Rx 1), whe e Ry 22 Ry 2and Rx 12 Rx 1, (Ry 2; Rx 1) = xand o any p o…le (Ry 1; Rx 2), whe e Ry 12 Ry 1and Rx 22 Rx 2, (Ry 1; Rx 2) = x. By he coun e pa o Claim 2, o any Ry 22 Ry 2; (Ry 2; R1) = x o any R12 R1;and o any Ry 12 Ry 1; (Ry 1; R2) = x o any R22 R2. Thus, can be ew i en as a e o ule o y. I n3, suppose … s ha is such ha o some p o…le (Ry i; Rx i), whe e Ry i2 Ry iand Rx j2 Rx j o any j2Nn ig, (Ry i; Rx i) = y. Then, by Claims 2 and 3, (Ry k; Rk) = y o any k; any Rk2 Ry k;and any Rj2 Rj; j2Nn kg. Tha is, he ou come will be y o any p o…le whe e he e is one agen ha s ic ly suppo s yo e x. Thus, is a e o ule o x. Le now suppose ha is such ha o all p o…les (Ry i; Rx i), whe e Ry i2 Ry i and Rx j2 Rx j o any j2Nn ig, (Ry i; Rx i) = x. Then, by Claim 4 and he coun e pa s o Claims 2 and 3, he ou come will be x o any p o…le whe e he e is one agen ha s ic ly suppo s xo e y. Thus, is a e o ule o y. This ends p oo o S ep 3, and hence he p oo o Theo ems 2, 3, and 4. 18 5 Final Rema ks In his pape we ha e p o ided di¤e en de…ni ions o s a egy-p oo ness in on o possible manipula ions by g oups, and se e al cha ac e iza ions o ules sa is ying hese p ope ies when hei ange is es ic ed o co e wo al e na i es. We eel ha , when a ainable, non-manipulabili y by g oups (in i s di¤e - en o ms) is an a ac i e p ope y, since in many con ex s di¤e en agen s can be expec ed o explo e he possibili y o bene… ing om join ac ions, in addi ion o indi idual ones. Ea ly au ho s on he issue o s a egy-p oo ness did indeed e e o he in e es o a oiding such join s a egic beha io (Pa - anaik, 1978, Dasgup a, Hammond, and Maskin, 1979, Peleg, 1984 and 2002). T ue, in many domains, and o unc ions wi h non-bina y anges, i may be excessi e o ask o hese p ope ies. Bu no always! Fo in e es ing cases when hey may be ul…lled because o domain es ic ions, see Moulin (1999), Pápai (2000), Ba be à and Jackson (1995). In ac , ou pape con empla es ano he case whe e join manipula ions can be a oided, his ime because he anges o ou unc ions a e es ic ed. We ha e allowed o agen s o ha e p e e ences o e o he al e na i es ha a e no in he ange, and been ca e ul in ollowing up he implica ions o ha ex ension in he domains o he ules. This is in con as wi h he wo k o au ho s who assume ha only wo al e na i es a e a ailable when he ange consis s o wo o hem. We insis in he di¤e ence, because we wan o emphasize ha he choice o es ic he ange is indeed a possible ool o he mechanism designe , e en when mo e han wo choices a e in p inciple socially a ailable. We ha e also looked o cha ac e iza ions ha a e essen ially independen o he cha ac e is ics o he domains o de…ni ion o he ules. This is because he se s o ules sa is ying ou di¤e en e sions o non-manipulabili y by g oups could in p inciple be a ying as he domains o de…ni ion change om one applica ion o ano he . By selec ing p ope ies ha a e necessa y and su¢ cien o ou condi ions o be sa is…ed, we go o he essen ials o he ques ion. And, when needed, ou quali…ca ions on he minimal equi emen s on domains o ou esul s o hold a e made explici a each poin . We ha e also insis ed in examining he ole o indi iduals who a e in- di¤e en be ween he al e na i es in he ange (bu no iden ical in o he espec s). The p esence o indi¤e ences is always a sou ce o p oblems in social choice, and i also complica es and en iches ou analysis he e. 19 We lea e i o he in e es ed eade o examine how ou analysis would be simpli…ed (and some imes educed o p e iously exis ing esul s) when only wo al e na i es a e p esen a all, and/o when indi¤e ences among al e na i es a e uled ou . Le us also men ion ha we ha e concen a ed on he no ions o weak and s ong g oup s a egy-p oo ness, The in e media e no ion o g oup s a egy- p oo ness has been p o en o be equi alen o he s ong e sion unde mild domain assump ions, bu no o he pa icula case o wo al e na i es only. Cha ac e iza ions o ules sa is ying he in e media e p ope y in his pa ic- ula case a e le as an open p oblem. Re e ences [1] S. Ba be à, D. Be ga, and B. Mo eno, Indi idual e sus g oup s a egy- p oo ness: when do hey coincide?, J. Econ. Theo y (2010), o hcoming. [2] S. Ba be à and M. Jackson, S a egy-p oo Exchange, Econome ica 63 (1995), 51-87. [3] P. Dasgup a, P. Hammond, and E. Maskin, The Implemen a ion o So- cial Choice Rules: Some Gene al Resul s on Incen i e Compa ibili y, Re . Econ. S ud. 46 (1979), 185-216. [4] A. Gibba d, Manipula ion o Vo ing Schemes: A Gene al Resul , Econo- me ica 41 (1973), 587-601. [5] V. Manjuna h, G oup S a egy-p oo ness And Social Choice Be ween Two Al e na i es, Mimeo (2009a). [6] V. Manjuna h, E¢ cien and S a egy-p oo Social Choice When P e e - ences A e Single-dipped, Mimeo (2009b). [7] B. La sson and L.-G. S ensson, S a egy-p oo o ing on he ull p e - e ence domain, Ma h. Soc. Sci. 52 (2006), 272-287. [8] H. Moulin, Inc emen al cos -sha ing: Cha ac e iza ion by coali ion s a egy- p oo ness, Soc. Choice Wel a e 16 (1999), 279-320. [9] S. Pápai, S a egyp oo Assignmen by Hie a chical Exchange, Econo- me ica 68 (2000), 1403-1433. [10] P. K. Pa anaik, S a egy and G oup Choice, No h-Holland Publishing Company (1978). [11] B. Peleg, Game Theo e ic Analysis o Vo ing in Commi ees, Econo- 20 me ic Socie y monog aphs in pu e heo y-7, Camb idge Uni e si y P ess (1984). [12] B. Peleg, Game heo e ic analysis o o ing in commi ees, Handbook o Social Choice and Wel a e, olume 1, edi ed by K.J. A ow, A.K. Sen and K. Suzumu a, No h-Holand (2002). [13] M. Sa e hwai e, S a egy-P oo ness and A ow’s Condi ions: Exis- ence and Co espondence Theo ems o Vo ing P ocedu es and Social Wel- a e Func ions, J. Econ. Theo y 10 (1975), 187-217. 21