List - Multidimensional Welfare Aggregation

Multidimensional Welfare Aggregation Author(s): Christian List Reviewed work(s): Source: Public Choice, Vol. 119, No.
1/2 (Apr., 2004), pp. 119-142 Published by: Springer Stable URL: http://www.jstor.org/stable/30025816 . Accessed: 30/11/2012 16:07
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Springer is collaborating with JSTOR to digitize, preserve and extend access to Public Choice.
http://www.jstor.org
This content downloaded by the authorized user from 192.168.82.217 on Fri, 30 Nov 2012 16:07:04 PM All use subject to JSTOR Terms and Conditions
Public Choice 119: 119-142, 2004. a 2004 KluwerAcademicPublishers. Printed in the Netherlands.
119
Multidimensional welfare aggregation *

CHRISTIANLIST
NuffieldCollege, OxfordOX1INEFU.K.;e-mail: christian.list@nuffield.oxford.ac.uk 2003 Accepted 13 February Abstract. Most accountsof welfare aggregationin the traditionof Arrow's(1951/1963) and frameworks the welfareof an individualin Sen's (1970/1979) social-choice-theoretic represent welfarefunction.I develop a multitermsof a single welfareorderingor a single scalar-valued dimensionalgeneralizationof Arrow'sand Sen's frameworks, individualwelfare representing in terms of multiple personal welfare functions, correspondingto multiple 'dimensions' of welfare. I show that, as in the one-dimensionalcase, the existence of attractiveaggregation assumptions,specificallyaboutthe measurability procedures dependson certaininformational of welfare and its comparabilitynot only across individualsbut also across dimensions. I state several impossibility and possibility results. Under Arrow-typeconditions, insufficient of a single individual,while insufficient across individualsleads to dictatorship comparability dominance of a single dimension. Given sufficient across dimensions leads to comparability both acrossindividualsand across dimensions,a rangeof possibilitiesemerges. comparability I discuss the substantiveimplicationsof the results.
1. Introduction
The concern of this paper is the problem of aggregatingthe welfare of the individualmembersof a group into the correspondingwelfare of the group as a whole. Individualwelfare is typically assessed in termsof some normator ively relevantevaluationstandard. Examplesof such evaluationstandards, 'currenciesof welfare', are money, indices of resources,utility, or Rawlsian primarygoods. By Arrow's theorem (1951/1963), there exists no procedurefor aggregating individualwelfare orderingsover a set of alternativesinto collective,
* Previousversions of this paperwere presentedat the seminarof the PhilosophyProgram of the ResearchSchool of Social Sciences, AustralianNationalUniversity,Canberra, in June 2000, at an NSF Workshopon Models of Individualand Public Choice, held at the University of Californiaat Irvine,in July 2000, and at the AnnualMeeting of the Public Choice Society in New Orleans, Louisiana, in March 1999. The authorwishes to express his gratitudeto the seminar and conference participantsat these occasions and to A.B. Atkinson, Geoffrey Brennan, John Dryzek, Robert Goodin, Iain McLean, Kevin Roberts, John Weymark,and an anonymousreviewer for very helpful comments and discussion. Financial supportof the (British)Economic and Social ResearchCouncil, the GermanNationalMeritFoundationand the GoodhartFundis gratefullyacknowledged.
120 or social, ones where the procedure satisfies a set of minimal conditions (transitivityof social orderings, universal domain, the weak Pareto principle, independence of irrelevant alternatives and non-dictatorship).Sen (1970/1979) has shown that Arrow's theorem depends crucially on the assumption that individual welfare is not interpersonallycomparable:given of welfarelevels or units, the impossibilityresult interpersonal comparability The or not interpersonally of whether question significantinformdisappears. ationis accessible dependson the chosen 'currencyof welfare'. The question of individualwelabout the significantcontent, personal and interpersonal, fare informationis called the question of measurabilityand interpersonal comparability. Sen's (1970/1979) frameworkis more general than Arrow's in that it allows alternativeassumptionson measurabilityand interpersonalcomparability. But there is one assumptionthat both accounts share; namely the assumptionthat individual welfare can be expressed in terms of a single welfare function.This assumption welfare orderingor a single scalar-valued 'dimension' or a single 'currency'of welexistence of a the requires single fare, in respect of which each individual'swelfare over a set of alternatives can be assessed. This paper provides a multidimensionalgeneralizationof Arrow's and The welfare of each individualwill be expressedin terms Sen's frameworks. of multiple personal welfare functions, one for each relevant 'dimension'. frameworkraises Like a one-dimensionalframework,the multidimensional Unlike a oneand interpersonal the questionof measurability comparability. dimensional framework,it raises an additional question. Is it possible to comparean individual'swelfarein one dimensionwith thatsame individual's welfare in anotherdimension?This, roughly,will be called the question of interdimensional comparability.1 I will first identify a multidimensionalArrow problem. In the absence of sufficientinterpersonal any aggregationproceduresatisfycomparability, individualdictatorial.And in the will make one conditions ing Arrow-type absence of sufficient interdimensional comparability,any such aggregation procedurewill make one dimensiondominant.An Arrow-typeimpossibility both across individualsand result can be avoided if sufficientcomparability and a rich set of possibilitiesemerges. and across dimensionsis admitted, Although the literaturecontains some formally closely related results, Such resis usually not developed as an interpretation. multidimensionality of multipleopinionsabout ults areRoberts's(1995) resultson the aggregation the welfare of a group of individualsinto a single social orderingand Khmelnitskaya's(1999) and Khmelnitskayaand Weymark's(2000) results on social welfareorderingsfor differentscales of individualutilitymeasurement
121 in distinct population subgroups.Amongst the first papers on multidimensional welfare are Plott, Little, and Parks (1975) and Sen (1980/81), the former providing an Arrow-style theorem for aggregationacross multiple dimensions within a single individual.Kelsey (1987) provides a relevantreof the literature on informational interpretation assumptionsin social choice. Less closely relatedprecursorsare resultsby Fishburn(1971) and Batraand Pattanaik(1972) on multi-stagemajoritydecisions, involving nested aggregation over nested subprofilesof a given profileof preferenceorderingsacross individuals.Finally,a companionpaperto this paperis concernedwith multidimensionalpreferenceaggregationwithout any forms of comparability, but with intradimensional single-peakedness(List, 2002). The paperis in five sections.In Section 2, I will brieflydiscuss the political theory backgroundof multidimensional aggregationproblems.In Section 3, will address the formalization of I measurability, interpersonal comparabilIn Section 4, I will state the results, ity and interdimensional comparability. and, in Section 5, I will draw some conclusions. The derivationof the main impossibilityresult will be statedin an appendix.
2. Multidimensionality and socialchoice

Sen has arguedthatthe choice of a normativelyrelevantevaluationstandard, or 'currencyof welfare', may be as consequentialas the choice of an aggregation procedureitself. The locus classicus is his famous paper "Equalityof What?' (Sen, 1982). People may agree that equality of some form matters in the design of an aggregationprocedure,and yet disagree on the question total"Equalityof What?".Income egalitarians, marginal-utility egalitarians, and Rawlsianprimarygoods egalitarians all claim to utility egalitarians, may use 'egalitarian'aggregationprocedures.But, in light of theirdifferentviews on what 'currency'should be equalized,they derive fundamentally different conclusions as to what social arrangements shouldbe pursued. Let me illustratehow the choice of a 'currencyof welfare' affects the recommendedsocial arrangements. Income egalitariansseek to equalize income acrossindividuals.This may lead to an unequaldistribution of utility,as the capacityto convertincome into utility may differ across individuals.For instance, someone with a certainmedical condition may need more income to sustain a particularutility level than someone without that medical condition. Marginal-utility egalitariansseek to equalize marginalutility across individuals,which typically amountsto maximizing the sum-totalof utility across individuals.This will not in general lead to an equal distributionof income, nor to an equal distributionof utility. Those individuals who are more efficient in convertingincome into utility may receive more income
122 underthis proposal, and those who are less efficient may receive less. Further, a distributionthat equalizes marginal utility across individuals (and thereby maximizes total utility) is not typically one in which utility itself is equally distributed. Total-utilityegalitarians,by contrast,seek to equalize which maximizes utility itself. This will not in general lead to a distribution of income. Total-utilityegalitarians total utility, nor to an equal distribution endorseincome inequalitiesbetween individualsif these income inequalities of utility.Undertotal-utilityegalitarianism, lead to a more equal distribution someone with a medical condition as in the example above should receive more income than someone without that medical condition. However,totalutility egalitariansface the challenge of expensive tastes. If a person has an expensive taste and requirescaviar to achieve the same utility level that seems to othersachieve by consumingbread,then total-utilityegalitarianism entail thatthatpersonought to receive additionalresourcesin orderto afford caviar.Many find this conclusionunacceptableand thereforeseek to identify a 'currency'that is less 'subjectivist'thanutility (so that it is immuneto the thanincome (so that problemof expensive tastes), but more welfare-relevant it is sensitive to special needs, like medical ones). Whethera 'currency'with the desiredpropertiesexists is a matterof philosophicaldebate,but Rawlsian primarygoods, as discussedbelow, are sometimesheld to be a plausiblesuch currency. Sen uses the termcompetitive pluralityto referto this diversityin views on what the relevant'currencyof welfare' is: differentproposalsstandas rivals to each other.According to the classical assumptionof one-dimensionality, one 'currencyof welfare' is to be selected as the relevantone from amongst this competitiveplurality. Againstthis assumption,Sen (e.g., 1985, 1987), Walzer(1983) and others for many problemsof welfareevalhave arguedthatthe relevantinformation on social arrangements The effects of alternative uationis multidimensional. an individual'sincome, nutritionand shelter,health, educationalprospects, social status, and so on, may all be relevant.But it may be impossible to representthis informationin terms of a single one-dimensional'currencyof welfare'. The use of multipleevaluationstandards may thereforebe warranted. Sen introducesthe term constitutiveplurality to refer to this view, that is internallydiverse (e.g., 1987: 2-3). the relevantevaluationstandard Walzer both defend constitutivelyplural conceptions Sen and Although of welfare, their accounts are quite different.Sen's account is motivatedby the Aristotelianessentialist view that several humanfunctionings form an essential part of a 'good life' and are thus relevantdimensions for welfare evaluation(see also Nussbaum, 1992). Such functionings may range from being well nourishedand being free from avoidable disease to being able
123 to participatein social life and having self-respect. Walzer's plurality of and contexspheres of goods, by contrast,is motivatedby a communitarian tualist theory of the good. The relevantspheres are determinedby contextspecific social meanings in the relevant society. Examples of such spheres might be health,education,employment,political influence,and money. Sen acknowledgesthe need for aggregationand cross-dimensional indexing (e.g., 1991; 1997: section A.7.3), whereasWalzerdefendsthe mutualseparateness, and arguablyincomparability, of differentdimensions.In particular, Walzer that different of be and that differrequires spheres goods kept separate ent (dimension/goods-specific)principles of justice apply to different such spheres. The evaluation standardof Rawls's Theory of Justice (1971), an index of primarygoods, is also constitutivelyplural,includingrights, libertiesand income andwealth,andthe social bases of self-respect(Rawls, opportunities, 1971: 60-65). However,Rawls makes the tacit assumptionthat summarizing a pluralityof primarygoods into a single index (which is ordinallymeasurable and interpersonally comparable)is possible, but is silent on how to constructsuch an index. Formally,Rawls's social-choice-theoretic proposals, the in are a one-dimensional form. particularly difference principle, presented A numberof social choice theoristshave addressedthe problemof indexing primarygoods and derivedsome Arrow-typeimpossibilityresults, all based on the assumptionthat the comparability between differentprimarygoods is limited (for instance, Plott, 1978; Gibbard,1979; Blair, 1988; but see Sen, 1991). At a practicallevel, the HumanDevelopmentIndex(HDI), as employedby the United Nations DevelopmentProgramme,is also a constitutivelyplural 'currencyof welfare', albeit an aggregateone. The HumanDevelopmentIndex combinesthreecomponents,each of which is itself an aggregatemeasure across people in a country or region. The three components are longevity, knowledge, and standardof living. Longevity is measuredin terms of life expectancy at birth. Knowledge is measuredin terms of a combination of the adultliteracyrate and enrolmentin primary,secondary,and tertiaryeducation. Standardof living is measuredin terms of GDP per capita (UNDP, 2002). The presentapproachallows to formalize a constitutivelypluralconception of welfare at a micro-level:it allows the separaterepresentation of the effects of each alternative on each individualin each one of multiplerelevant dimensions.
124
and interdimensional 3. Measurability, interpersonal comparability comparability

Let N = {1, 2, ... , n} be a set of individuals,and X = {xl, x2,. . . a set of thatthereare k relevantdimensions,contained alternatives. Suppose,further, in K = {1, 2, ... k}. We will assume n > 1, IXI > 2, and, unless stated otherwise,k > 1.2 To each individuali e N, there correspondsa k-tuple Wi :=< Wij >jEK = < Wi, W2, ... , Wik > of personalwelfarefunctions,containingone Wij : X - R for each dimensionj E K. For each individuali E N and each dimensionj E K, the function Wij assigns to each x e X a real number the welfareof individuali in dimensionj under Wij(x). HereWij(x) represents x. alternative For example, let X be a set of employmentpolicy alternatives,and let 1, 2, 3 e K be the dimensionsof income, healthand social status.Then Wil, Wi2 and Wi3 representindividuali's welfare over the alternativesin X from the perspectivesof income, healthand social status. = A profile of k-tuplesof personal welfarefunctions is an n-tuple {WiliEN for ineach one of such k-tuple Wi k-tuples, containing IW1, W2, ..., Wn} N. As a notationalconvention,we use { }-bracketsto denote dividual i >-brackets to denote k-tuples across n-tuples across individuals and < dimensions. A multidimensional social welfarefunctional (MSWFL)is a function F which maps each profileof k-tuplesof personalwelfarefunctions(in a given on X, domain), IWijiEN,to a correspondingsocial orderingR = F({Wi}iEN) where R is reflexive,transitiveand connected.R induces a strongorderingP and an indifferencerelationI, definedas follows: for all xl, x2 e X, not x2Rx1; x1Px2if and only if xlRx2 and xlIX2 if andonlyif xlRx2 andx2Rx1. A one-dimensional social welfare functional (SWFL) is simply an MSWFL for the special case k = 1. How can we formalize assumptions on measurability, interpersonal comparability and interdimensional comparabilityof welfare? The formalizationto be developed is directly analogous to Sen's well known method of formalizing assumptions on measurabilityand interpersonalcomparabilityin a one-dimensionalsocial choice framework(e.g., Sen, 1970/1979; Sen, 1982: Ch. 11; see also List, 2001). We first explain the idea behind the formalizationand then proceed to stating the formalization properly. Let us consider statements of the following forms: Level Comparisons (LC). Alternativex1 from the perspectiveof individual
125 ii in dimensionji is at least as good as alternative x2 from the perspectiveof individuali2 in dimensionj2; formallyWi,j (x1) > Wi22 (X2). Unit Comparisons (UC). The ratioof [individualiI's gain/loss in dimension x1 to x2] to [individuali2's gain/loss jl as a resultof switchingfromalternative in dimensionj2as a result of switchingfrom alternative yi to Y2] is X, where k is a realnumber;formally(Wi j1(X2)-Wi j1 (Xl))/(Wi2~2 (y2)Wi2j2 ((Y)) = k. The key idea is that different assumptionson measurability,interpersonal and interdimensional of welfare imply different comparability comparability conditions under which (LC)- and (UC)-statements are meaningful. Specifically,we have the following: Ordinalmeasurability(0) implies that(LC)- statementsare meaningful when iI = i2 andj1 = j2. Cardinal measurability(C) implies that (LC)- statements and (UC)statementsare meaningfulwhen il = i2 andjl = j2. level that are (LC)-statements Interpersonal comparability (Lper)implies = meaningfulwhenjI j2 but iI i2. unit comparability are Interpersonal implies that (UC)-statements (Uper) = when but meaningful jl j2 il # i2. Interdimensional level comparability (Ldim) implies that (LC)statementsare meaningfulwhen = i2 butjl j2 il Interdimensional unitcomparability (Udim) implies that(UC)-statements are meaningfulwhen = i2 butjl # j2. i1
We use the labels (Nper) and (Ndim) to refer to, respectively,no interpersonal and no interdimensional comparability comparability. Once we have assigned a profileof k-tuplesof personalwelfare functions {WiliENto a set of individuals, we can of course make (LC)- and (UC)statements relative to that profile. However, whether such statements are meaningfuldepends on how unique the profile {Wi}iN is. Suppose, for instance, that each Wij in the profile {Wili}N is unique only up to a positive monotonic transformation, for different possibly a different transformation individualsand dimensions. Then interpersonal or interdimensional level or unit comparisons are not well-defined and thus not meaningful, as these underthe specifiedtransformations. comparisonsare not in generalinvariant Suppose, more generally,we specify the class of transformations1 up to
which each profile {Wi}ijN is unique. We can then ask whether each kind of
in P. A (LC)- or (UC)-statementis invariantunder all the transformations
126 kind of (LC)- or (UC)-statementis said to be meaningfulif and particular underall transformations in P. only if it is invariant We state assumptionson measurability, interpersonal comparabilityand interdimensional comparability by specifying the class of transformations ) is takento be unique.The smallerthis class up to which each profile {WiliEN the more informationis contained in a profile. Table 1 of transformations, lists severalalternative such classes of transformations.3 An example shows how to read Table 1. Consider OLdimNper, the asandthe columnlabelled to the row labelledLdimNper sumptioncorresponding Then interdimenis the of ordinal OLdim Nper measurability, assumption O. sional level comparability,but no interpersonalcomparability.According to OLdimNper, P is the class of all n-tuples of k-tuples of transformations, {< Oij >jEK}iEN such that each 4ij : R -- R is a positivemonotonic transformation, and, for each i e N, Oil = Oi2 = ... = =ik. in ci. Now suppose that each {WiJiEN is uniqueup to the transformations The We then requirethata MSWFLbe invariant underthese transformations. into can be transformed is this. Suppose {WiliEN idea behindthis requirement in D. Then {Wi}iEN and {[W}ieNare taken {W*}iiNby some transformation to contain exactly the same relevant information.Therefore our MSWFL to the same social ordering.Formally,we should map {Wi}iEN and {fW}iEN as follows. can state this requirement and {W*}iEN Invariance assumption with respect to (D. For any IWiIiEN C c such that, for each in the domain of F, if there exists {< Oij >jEKIiEN then F({WidiEN) = F({W*}ieN). i N and eachj K, W = Oij(Wij), Thus each assumptionin Table 1 defines a specific class of transformations (D, and we can consider the correspondinginvariance assumption with respect to D. We use the name of each assumptionin round brackets,for invariance to denote the corresponding assumption.4 example (OLdimNper), (D and Let me make one finalremark.Suppose I are two classes of transand formationssuch that C 4. For example, D correspondsto OLdimNper, T corresponds to CLdimUdimNper.Then anyMSWFLsatisfyingthe invariance assumptionwith assumptionwith respectto ) will also satisfy the invariance respectto 4. 4. Results Before we can presentimpossibilityand possibility results, we need to state multidimensional generalizationsof Arrow'sconditions. The generalization of universal domain, the weak Pareto principle and independence of
127
4' is the class of all n-tuplesof k-tuples of transformaTable1. The class of transformations tions, {< ij >jeK}ieN, with the following properties: O
C
Ordinalmeasurability Cardinalmeasurability Each ij : R - R is a Each oij : R - R is a positive affine transformation, and... positive monotonic transformation, and...
NdimNper
no furtherassumption no furtherassumptionrequired required
UdimNper
not applicable for each i E N, oil = bi2 = ... = oik not applicable not applicable
oil
LdimNper
for each i e N, Bai , ai2, ..., aik E R such that + ail = i2 + ai2 = ... = oik + aik not applicable
for each i E N, #il = oi2 = ... = ik
LdimUdimNper NdimUper
for eachj e K, Balj, a2j... anj e R such that 4lj + alj = 42j + a2j = ... = Qnj+ anj not applicable for eachj e K, =lj = 2j not applicable 3al ... alk, a21,..., a2k,anl ,...., ank E R such that411 + all = ... = 'lk + alk = 21 + a21= ... = 2k + a2k = 'nl + anl = ... = nk + ank
for each j e K, blj = =2j
NdimLper
for each j E K, 4lj =

q2j = ...
= nj
NdimLperUper LdimLper UdimUper
not applicable all fij are identical not applicable
= ... = nj
UdimLperUper
not applicable
... = Pnj
and 3al, a2.... ak E R such that, for each i E N, il + al = 'i2 + a2 = ...

= 'ik + ak LdimUdimUper
not applicable
for each i E N, 'il = Oi2= ... = ik and 3al, a2..., an ER such that, for each j e K,'lj + al = 2j + a2 =... = 'nj + an all 'ij are identical
LdimUdimLperUper not applicable
128 irrelevantalternativesis straightforward. Given any profile of k-tuples of we define R := F({WiiEN). personalwelfare functions, {WiliEN, Universal Domain (U). The domain of F is the set of all logically possible profilesof k-tuplesof personalwelfarefunctions. WeakPareto Principle (P). Let {WiliENbe any profile in the domain of F. For any x1, x2 e X, we have x1Px2 whenever, for all i E N and all j E K, Wij(xl) > Wij(x2). and Independence of Irrelevant Alternatives (I). Let {WiLiEN {W*}iEN be any profilesin the domainof F. Supposex1, x2 E X such that,for all i EN and all j e K, Wij(x1) - W (x1) and Wij(x2) = W*(x2). Then xlRx2 if and only if x1R*X2. is less straightforward The generalizationof non-dictatorship (see also List, 2002). In the multidimensionalframework, non-dictatorshipcorresponds to two conditions: non-dictatorshipand non-dominance.Non-dictatorship requires the non-existence of a fixed single individual whose k-tuple of personal welfare functions always determines the social ordering. Nondominancerequiresthe non-existenceof a fixed single dimension such that the personalwelfare functions (across individuals)in that dimensionalways determinethe social ordering. Non-Dictatorship (D). There does not exist an individual i e N (a in the domain of F and all dictator for F) such that,for all profiles IWiliEN where > f(< Wij(X2)>jEK) implies X1PX2, X1, X2 E X, f(< Wij(x1) >jeK) f: Rk --+ R is a strictlyincreasingfunction. Non-Dominance (Dom). There does not exist a dimension j E K (a in the domainof dominantdimensionfor F) such that,for all profiles {WiliEN F and all x1, X2 E X, f({Wij(X1)}iEN)> f({Wij(X2)xiEN) impliesx1Px2,where f: Rn -> R is a strictlyincreasingfunction. Non-double-dictatorship,finally, requires the non-existence of a fixed single individual and a fixed single dimension such that this individual's personal welfare function in that dimension always determinesthe social ordering. (DD). Theredoes not exist an individuali e N and Non-Double-Dictatorship a dimensionj E K such that, for all profiles {Wi}i~N in the domainof F and
129 all x1, x2 E X, Wij(x1) > Wij(x2) implies x1Px2. Under all these generalized non-dictatorshipconditions, personal welfare functions other than the dictatorialor dominant ones can act at most as tie-breakers,namely in those cases in which the dictatorial or dominant functionsreflectindifference. In the absence of both interpersonal comparabilityand interdimensional Arrow'stheorem(in the version of Sen, 1970/1979; see also comparability, List, 2002) implies the following result. Theorem1. There exists no MSWFL satisfying (ONdimNper) or (CNdinNper), and (U), (P), (I) and (DD). A multidimensional aggregation problem with n individuals and k dimensions, with (ONdimNper) or (CNdimNper), is equivalent to a onedimensional aggregation problem with nk individuals and without Theorem 1 thus follows directly from Arrow's interpersonal comparability. theorem. 4.1. Lexicographicdictatorshipsand lexicographichierarchiesof dimensions Is interpersonalcomparability,in analogy to the one-dimensional case, sufficientfor avoiding the multidimensional Arrowproblem?The answerto this questionis negative.If we assumeordinalor even cardinalmeasurability, with interpersonalcomparability of both levels and units, but without interdimensional comparability, any MSWFLwhich satisfies (U), (P) and (I) still violates (Dom). Theorem 2. There exists no MSWFL satisfying (ONdimLper) or (CNdimLperUper), and (U), (P), (I) and (Dom). The result can be seen as a reinterpretation of results by Roberts (1995, theorem 6) and Khmelnitskaya and Weymark (2000, theorem 2). See Appendix 1. Similarly, if we assume ordinal or even cardinal measurability,with interdimensional comparability of both levels and units, but without interpersonalcomparability,any MSWFL satisfying (U), (P) and (I) still violates (D). Theorem 3. There exists no MSWFL satisfying (OLdimNper) or (CLdimUdimNper),and (U), (P), (I) and (D).
130
See also appendix 1. Given comparabilityacross individualsbut not across

dimensions (i.e. (ONdimLper) or (CNdimUper)), there are MSWFLs satisfying
(U), (P), (I) and (D), but not (Dom). Examples are suitable lexicographic hierarchies of dimensions, as defined below. Given comparabilityacross dimensionsbutnot acrossindividuals(i.e. (OLdimNper) or (CUdimNper)), there are MSWFLs satisfying (U), (P), (I) and (Dom), but not (D). Examples are suitablelexicographicdictatorships,also definedbelow. A MSWFL F is a lexicographichierarchyof dimensionsif there exist k
... strictly increasing functions fl, f2,
fk : Rn --+ R (possibly different), one
a of K such that,for any {Wi}iEN for each dimensionin K, and a permutation and any x1, x2 E X,
x1Px2
if and only if
focrj)({Witj)(X1)isN) > fej)({Wiaj)(X2)}ieN)
fa(h)({Wia(h)(X1)}iEN)
and
= f(h)({Wia(h)(X2)ieN)
for somej e K for all h < j.
Under this definition,the dimensions are rankedin a fixed hierarchyof importance.For each dimension j, there exists a function fj (to be called an intradimensional aggregationfunction) for aggregatingthe dimension-jspecific welfare informationfor each alternativex across individualsinto a single aggregate figure for that dimension. The alternativesin X are then ranked,lexically, accordingto the aggregatefiguresfor the dimensionswhich are first, second, third, and so on in the hierarchy.As noted above, a lexicographic hierarchyof dimensions need not violate (D). Given sufficient one-dimensionalSWFL can any non-dictatorial interpersonal comparability,
be the basis for the intradimensional aggregation functions fl, f2 ,..., fk-
Prominent examples are:

if (at least) (CNdimUper) is satisfied:
a 'utilitarian'intradimensional function: aggregation

fj({Wij(X)}iEN):= XlWlj(x)
where 1, 2, . * > 0; - if (at least) (ONdimLper)iS satisfied : a 'maximin'intradimensional aggregation function:

fj({Wij(x)}ieN) := min(Wlj(x), W2j(x),...,
+ o2W2j(x) + ... + ,nWnj(x),
Wnj(x)).
In a lexicographic hierarchyof dimensions, different intradimensional aggregation functions can be chosen for different dimensions. Given interpersonalcomparabilitywithin each dimension, but no interdimensional a lexicographichierarchyof dimensions may in some cases comparability, be an attractiveMSWFL. In Theoryof Justice, Rawls contraststhe 'special conception'with the 'generalconception'of justice. The 'specialconception'
131 is intendedto apply to socio-economicallywell developedsocieties, whereas the 'general conception' is intended to apply to socio-economically less developed societies. The 'general conception' requiresthat "the difference principle [be] applied to all primarygoods including liberty and opportunity" (Rawls, 1971: 83; my italics). This presumablyinvolves aggregation,in accordancewith (DOM), across all primarygoods. Accordingto theorem2, if the differentdimensions are constitutedby differentprimarygoods and if these differentprimarygoods are mutuallyincommensurable (as capturedby
(ONdimLper) or (CNdimLperUper)),then Rawls's 'general conception' of justice
leads to an impossibility result. The 'special conception' of justice, on the other hand, assigns lexical priorityto some primarygoods (e.g., liberty and opportunity)over others. A lexicographichierarchyof dimensions captures precisely this idea. A MSWFL F is a lexicographicdictatorshipif there exist n strictly in... creasing functions fl, f2,
fn : Rk -- R (possibly different), one for each
individualin N, and a permutation a of N such that,for any {Wi}ieN and any

x1,X2 E X,
x1Px2 if and only if

and > fo(i)(< WT(i)j(x2) >jEK) fa(i)({Wcj(i)j(xi)}jGK)
fa(h)(< Wa(h)j(X1) >jeK)
=
fa(h)(<
We(h)j(X2)
>jeK)
for some i E N for all h < i.
In a lexicographicdictatorship, the individualsare arranged in a fixed hierarchy of decisiveness. For each individuali, there exists a function fi (to be called an intrapersonal this individual's aggregation function) for aggregating of welfare information for each alternative x into a k-tuple single aggregate figure for that individual.The alternativesin X are then ranked, lexically, accordingto the aggregatefigures for the individualswho are first, second, third, and so on in the hierarchy.A lexicographicdictatorshipneed not violate (Dom). Given sufficientinterdimensional comparability, intrapersonal aggregationfunctions that focus on more than one dimension are available,
e.g. if (at least) (CUdimNper) is satisfied:
a 'utilitarian'intrapersonalaggregation function:
fi(< Wij(x) >jeK) := olWil(x)
+ X2Wi2(X) +
...
+ +kWik(X),
where -
.1, if (at least) (OLdimNper) is satisfied: a 'maximin'intrapersonalaggregation function: fi(< Wij(x) >jeK) := min(Wil(x), Wi2(X),..., Wik(X)).
X2,...
,Xk
> 0;
In a lexicographicdictatorship,differentintrapersonal aggregationfunctions can be chosen for different individuals. However, unless there are
132 particularreasons why certain individuals should be given lexical priority over others, lexicographicdictatorshipsare hardly attractivefrom a liberal viewpoint. egalitarian 4.2. Stronger possibility results both across individualsand across dimensions is requiredif Comparability we regardboth (D) and (Dom) and the otherArrow-typeconditionsas indispensable. To state strongerpossibility results, we must distinguishbetween differentways of aggregatingacross individualsand dimensions. Aggregation may take place either in two steps or in one step. Two-stepaggregation can mean one of two things: (1) for each individual, the k-tuple of welfare informationfor each alternativeis first aggregatedacross dimensions into an aggregatefigure for that individual,and these aggregatefigures are then aggregated across individuals; (2) for each separate dimension, the welfareinformation for each alternative is firstaggregated dimension-specific across individualsinto an aggregatefigurefor that dimension,and these aggregatefiguresare then aggregatedacross dimensions.One-stepaggregation means that the entire profile of k-dimensionalpersonalwelfare functions is aggregateddirectlyinto an overall social ordering.5 utilitarianrules 4.3. Multidimensional (U), (P), (I), (D) Proposition4. Thereexist MSWFLssatisfying(CUdimUper), and (Dom). Given interdimensional comparability and interpersonal comparability of units, suitable multidimensionalutilitarianrules satisfy (U), (P), (I), (D) and (Dom). All multidimensionalutilitarianrules are expressible as two-step aggregationmechanisms. We will distinguish between those that aggregationand those thatprioritiseintradimensional prioritiseintrapersonal aggregation. A MSWFL F is a multidimensionalutilitarian rule that prioritises intrapersonal aggregationif thereexist (i) n strictly increasingintrapersonal aggregationfunctions fl, f2, ... Rk 4 R; (ii) all, a2, ... , an individuals) 0 (with= fn
1) (weights corresponding to the n
such that, for any {Wi}ieN and any xl, x2 e X.
133
X1Rx2
if and only if CieNaifi(<
Wij(x1)
>jeK)
1 CiNaifi (< Wij(x2)
>jeK).
Possible choices for fl, f2, ... ,fn are, for example, 'utilitarian'intrapersonal aggregation functionsor 'maximin'intrapersonal functions aggregation at least as defined above. (requiring (CLdimUdimUper)) A SWFL F is a multidimensional utilitarian rule that prioritises intradimensional aggregationif thereexist (i) k (ii) strictly increasing intradimensional aggregation functions
fk : Rn
fl, f2,
R;
-0
. , P2, ..., Sk 8l, dimensions)
(with lijeK
1) (weights correspondingto the k
such that,for any {Wi}iEN and xl, x2 e X,

jeK jfj({Wij(Xl)}iN) > LjeK jfj({Wij (x2)ieN). Possible choices for fl, f2, ... ,fk are, for example, 'utilitarian' intradiif and only if X1RX2
mensionalaggregationfunctionsor 'maximin'intradimensional aggregation

functions (requiring at least (CUdimLperUper))as defined above.
Multidimensionalutilitarianrules are versions of the classical utilitarian principle:make social choices so as to maximize the sum-totalof welfare.In the case of rules thatprioritiseintrapersonal aggregation,an aggregatefigure is first determinedfor each individual (by intrapersonalaggregation),and these aggregatefigures are then added up across individuals.In the case of rules that prioritiseintradimensional aggregation,an aggregatefigure is first determinedfor each dimension(by intradimensional aggregation),and these are then added across dimensions. aggregatefigures up The concept of two-step aggregationcombinedwith differenttypes of inor intradimensional functions(e.g., utilitarian ones as trapersonal aggregation well as maximinones) greatlyextendsthe varietyof possible utilitarian rules as comparedwith the one-dimensionalcase. Suppose, for example, we have interdimensional not only of units but also of levels. We might comparability then use maximinintrapersonal aggregationfunctions for each individualas the basis for utilitarian welfare summationacross individuals.For each individual and each alternative, we find the dimensionin which the individualis worst off and define the individual'saggregatefigureto be his or her welfare in thatdimension.The sum-totalof welfareis then determined by summation of these aggregatefiguresacrossindividuals.To considera differentexample, suppose we have interpersonalcomparabilitynot only of units but also of levels. We might then use maximin intradimensional aggregationfunctions
134
as the basis for utilitarianwelfare summationacross dimensions:for each dimensionandeach alternative, we identifythe individualwho is worstoff in thatdimensionand define the dimension'saggregatefigureto be the welfare of the worst-off individual in that dimension. The sum-total of welfare is then determined by summationof these aggregatefiguresacross dimensions. Othercombinationsare possible. The first of the two rules capturesthe view that the overall welfare of an individualshouldbe identifiedwith the lowest componentof this individual's welfare k-tuple, but that subsequentaggregationacross individuals should takethe form of utilitarian summation.The second rulecapturesthe view that - in a Rawlsianspirit- the overall 'social' welfarein each dimensionshould be identifiedwith the welfare of the worst-off individualin that dimension, but that- divergingfrom Rawls - subsequentaggregationacross dimensions summation. shouldtake the form of utilitarian 4.4. Multidimensional leximinrules
(U), (P), (I), (D) Proposition 5. There exist MSWFLs satisfying (OLdimLper),
and (Dom). Given interdimensional comparability and interpersonal comparability of levels, suitable multidimensionalpositional rules, in particularleximin rules, satisfy (U), (P), (I), (D) and (Dom).6 We will distinguish between three types of multidimensionalleximin rules; two-step rules that prioritise intrapersonalaggregation, two-step rules that prioritise intradimensional aggregation,and one-steprules. A MSWFL F is a multidimensional leximin rule that prioritises aggregationif thereexist intrapersonal
... (i) n strictly increasingintrapersonal aggregationfunctions fl, f2,
fn
"
Rk -+ R; and
(ii) for each x e X, a permutationi - [i] of N (dependingon x) such that f[l](< W[l]j(x) >jeK) I f[2](< W[2]j(X) >jcK) * ... < f[n] 7 (x) > jeK); (< Wtnlj
such that, for any
{Wi}iEN
and any x1, X2 E X,
if and only if X1PX2 f[i](< W[ilj(x1) >jeK) > f[i](< W[ij(x2) >jeK)
and f[h](< W[hbj(X1) >jeK)
-= f[h](< W[hl(X2) >jeK)
for some i E N
for all h < i.
135 Possible choices for fl, f2,

... ,
fn are, for example, 'maximin' intraper-
sonal aggregationfunctions or 'utilitarian'intrapersonal aggregationfunctions (requiring at least (CLdimUdimLperUper))as defined above.
A MSWFL F is a multidimensional leximin rule that prioritises intradimensional aggregationif thereexist (i) k strictly increasing intradimensional aggregation functions [j] of K (dependingon x) such that
<...<
: Rn -+ R; and fl, f2,....fk (ii) for each x e X, a permutation jf[l]({Wi[1](X)}iEN)

<
f[2]({Wi[2](X)}ieN)
f[k]({Wi[k](X)}iEN);
such that, for any {WilieN and any x1, X2 E X,

fj3({Wij](xl)}ieN) > fmj({Wiuj(X2)lieN) for somej e K and f[h]({Wi[h](Xl)}ieN) = f[h]({Wi[h](X2)1ieN) for all h < j.
Possible choices for fl, f2, ... fk are, for example, 'maximin' intradimensionalaggregation functionsor 'utilitarian' intradimensional aggregation
functions (requiring at least (CLdimUdimLperUper)) as defined above.
A MSWFLF is a (one-step)multidimensional leximinrule if thereexists (i) for each x e X, a bijectionh - [h] = (i, j) from {1, 2, ..., nk} to Nx K (depending on x) such that W[l](x) < W[2](x) < ... <W W[nk](X) such that,for any {WiliENand any xl, x2 E X,
> W[h](X2) for some h e {1, 2, ..., = and Wyg](xl) W[g](x2) for all g < h.
W[h](X1)
nk}
Multidimensionalleximin rules are versions of Rawls's (lexicographic) differenceprinciple:makesocial choices so as to maximizethe lowest welfare levels; if there are ties, maximize, in a lexicographichierarchy,the second lowest, third lowest, ..., welfare levels. These welfare levels are overall personalwelfarelevels in the case of rulesthatprioritiseintrapersonal aggregation and overalldimensionalwelfarelevels in the case of rules thatprioritise intradimensional aggregation. Here the concepts of one-step and two-step aggregationcombined with different types of intrapersonalor intradimensionalaggregationfunctions (e.g., utilitarianones as well as maximin ones) also extend the variety of possible leximin rules as comparedwith the one-dimensionalcase. Suppose, for example, we have interdimensionalcomparabilityof levels and units.
136
We might then use utilitarianintrapersonal aggregationfunctions for each individualas the basis for leximin aggregationacross individuals.For each individualand each alternative, we take the (possibly weighted) sum of this individual'swelfare levels in all dimensions to be the aggregatefigure for that individual.The leximin rule is then applied to these aggregatefigures across individuals:make social choices so as to maximize the lowest such aggregatefiguresacross individuals;if there are ties, maximize, in a lexicothe second lowest, thirdlowest, ..., such aggregatefigures. graphichierarchy, To considera differentexample,supposewe have interpersonal comparability of levels and units. We might then use utilitarianintradimensional aggregation functionsfor each dimensionas the basis for leximin aggregationacross dimensions. For each dimension and each alternative,we take the (possibly weighted) sum of dimension-specificwelfare levels across individualsto be the aggregatefigure for that dimension. The leximin rule is now appliedto these aggregatefiguresacross dimensions:make social choices so as to maximize the lowest such aggregatefigures across dimensions;if there are ties, the second lowest, thirdlowest, ... maximize, in a lexicographichierarchy, such aggregatefigures. The first of the two rules captures the view that the welfare levels of an individual should be identified with the (possibly weighted) sum of the componentsof this individual'swelfare k-tuple,but that subsequentaggregation acrossindividualsshouldfocus, lexically, on the lowest, second lowest, third lowest, ..., such individualwelfare levels. This is one version of the Rawlsian difference principle, where the primarygoods indexing problem for each individualis solved by summationacross differentprimarygoods. The second rule capturesthe view thatfor each dimensionthe overallwelfare in that dimension should be identifiedwith the (possibly weighted) sum of dimension-specificwelfare across individuals,but that subsequentaggregation across dimensionsshould focus, lexically, on the lowest, second lowest, thirdlowest, ..., such dimensionalwelfare levels. A one-step multidimensionalleximin rule, finally, treats the multidimensional aggregationproblem as if a one-dimensionalleximin rule were appliedto a society of nk individuals,effectively consideringeach of the nk individual-dimension pairs as a distinct individualand focussing, lexically, on the lowest, second lowest, thirdlowest, ..., welfare levels amongstthese nk individual-dimension pairs.
5. Some implications What conclusions can be drawn from this? Table 2 summarizesthe basic resultsof this paper.
137 In short, if we assume comparabilityacross individuals,but not across dimensions, the only aggregation procedures satisfying Arrow-type conditions are (possibly lexicographic) hierarchiesof dimensions. And if we assumecomparability acrossdimensions,but not acrossindividuals,the only aggregationproceduressatisfying Arrow-typeconditions are (possibly lexicographic)dictatorshipsof individuals.If we assume comparabilityacross both individualsand dimensions, a range of aggregationproceduressatisfyand non-dominanceemerge, such as multidimensional ing non-dictatorship leximin rules. utilitarian rules and multidimensional Is it plausibleto assumeinterpersonal and interdimensional comparability of me turn welfare? Let first to comparability interpersonal comparability. Arrow himself excludes interpersonalcomparisonsof welfare from his framework,holding "thatinterpersonal comparisonof utilities has no meanin that is no there ing and, fact, meaningrelevantto welfare comparisonsin the measurability of individualutility"(Arrow,1951/1963: 9). In responseto this view, two points must be noted. First, it is not obvious that interpersonal comparisonsof utility are indeed as meaninglessas Arrow claims they are, but even if interpersonal comparisonsof utility were empirically meaningless,this would not entail thatthereis no othernon-empirical butnonethelessnormatively significantandnon-arbitrary way of makingsuch comparisons. Second, as noted above, the question of whether interpersonalcomparisons are meaningful in a given sense depends crucially on what welfare evaluationstandard we choose to compare.To give just two examples,interpersonal comparisonsof money are unproblematic(leaving practicalissues aside), and interpersonal comparisonsof the amountof health care or education a personhas access to may also be unproblematic (againleavingpractical issues aside). This means that, even if we concede Arrow'sview that "interpersonal comparisonof utilities has no meaning" (my italics), this entails not that interpersonal comparisonsare in principleimpossible, but only that such comparisonsare possible only for certainwelfare evaluationstandards other than utility.And severalwelfare evaluationstandards with the required have been Rawls's index of properties proposed,including primarygoods and Sen's functionings. acrossdimensionsis a more seriousproblem.An important Comparability intuition underlyingmany argumentsfor a constitutivelyplural conception of welfare is that welfare has several, possibly mutuallyincommensurable, aspects, and that those aspects cannot easily be combined into a single overall index. Accordingly,the results of the present paper raise a tension that needs to be taken seriously. In particular, they raise a tension between
138
Table2. SupposeF is a MSWFLsatisfying(I), (P), and (U). Do we have interdimensional comparability?
Yes
No
Do we have interpersonal comparability?
Do we have interpersonal comparability?
Yes
No
Yes
No
Whichtype of interpersonal andinterdimensional comparability do we have? comparability
of units
of levels
F can be a F can be a iis a multidimensionalmultidimensional(possibly utilitarian rule. leximinrule lexicographic) rule. dictatorship.
F is a' (possibly lexicographic) of hierarchy dimensions.
F is a (possibly lexicographic) double dictatorship.
the following threeclaims: (i) There exist several normativelyrelevantdimensions of welfare, which (for instance, as a result of their different social meanings) are not mutuallycommensurable. (ii) There are decisions which affect different dimensions, but which come in 'packages', indivisible into separatedimension-specificsubdecisions.
139 (iii) In making decisions of the kinds describedby (ii), there should be no dominantdimensionthathas lexical priorityin determining the outcome. It seems that a violation of at least one of (i), (ii), or (iii) is inevitable. Primafacie, Walzer'sapproachto constitutivepluralityseems more vulnerable to this tension than Sen's or Rawls's. Walzer's approachrequiresthat differentdimensions (spheresof goods) be kept separate.As a result of this the approachcannot accommodatecross-dimensionalindexing requirement, as easily as Sen's or Rawls's approaches.The most consistent Walzerian escape from the tension, presumablythe one pursuedin Spheres of Justice, would be to tryto avoidclaim (ii). Indeed,when Walzer'sarguesthatdifferent distributive principles should be applied to differentdimensions (spheresof on the basis of the different social meanings of these dimensions goods), (goods), Walzer's proposal might be interpretedas a version of the above discussed strategyof subdividingdecisions into several dimension-specific sub-decisions and thus 'defining away' the multidimensionalaggregation problem. But, as we have indicated above, while lexicographic dictatorshipsare hardlyattractivefrom a liberal-egalitarian viewpoint, there are situationsin which lexicographichierarchiesof dimensions (i.e. violations of claim (iii)) are defensible. (Thus there is an asymmetrybetween the normativeappeal of non-dictatorship and the normativeappeal of non-dominance.)Rawls's 'special conception' of justice explicitly assigns lexical priority to some primarygoods, such as liberty and opportunity,over others, thus in effect of dimensions.If thereis a tensionbetween defininga lexicographichierarchy a holding constitutivelyplural conception of welfare and accepting interdimensionalcomparability, then Rawls's defence of a lexicographichierarchy of differentprimarygoods in the 'special conception' of justice is clearly the most compelling solutionto thattension. social-choice-theoretically
Notes
1. Interpersonal and interdimensionalcomparabilitywill be treated in formally similar ways. According to some philosophical conventions, what I call comparabilityacross dimensions is called commensurability across dimensions. For simplicity, I will refer to in the case of both comparability persons and dimensions. 2. The questionof how the relevantdimensionsareto be identified(i.e., the questionof how the set K is to be interpreted) is an important butit will here be taken philosophicalmatter, to lie outside the scope of social choice theory.Nussbaum'sargument for a specific list of basic functionings (Nussbaum, 1992) and Walzer'saccount of how social meanings determinethe relevantdistributive to the spheres(Walzer,1983) aretwo differentapproaches identificationand demarcation of relevantdimensions,corresponding to the two different
140
accountsof constitutivepluralitybriefly introducedabove (essentialistand contextualist, respectively). A transformation 4 : R -+ R is positive monotonic if, for all s, t in R, s < t implies 4(s) < 4 (t).4 : R - R is positive affine if there exist a, b in R, with b > 0, such that, for all t in R, 4 (t) = a + bt. For a more detaileddiscussion of the logical relationbetween meaningfulstatementsand in the one-dimensionalcontext, see Section 5 of classes of admissible transformations Bossert and Weymark(1996). Note that all two-step aggregationmechanisms can also he interpretedas one-step aggregationmechanisms, but not all one-step aggregationmechanisms are expressible as two-step aggregationmechanisms. For simplicity, I will here focus only on leximin rules. The results can, however, be the leximin rules discussed here all satisfy generalizedto other positional rules. Further, of individuals underpermutation some form of an anonymityrequirement, i.e., invariance of dimensions.For the discussionof non-anonymous underpermutation and/orinvariance positional rules in the one-dimensionalcase, see Bossert and Weymark(1996). To define a leximax rule, we simply need to replace all the "<"-symbols with ">"[i]. Analogous remarksapply to the symbols in the definition of the permutationi definitionsbelow. Roberts proves this result for a domain more restrictedthan a universal domain (U), domain (OAU) (hence U is stated here in namely for an ordinalagreementunrestricted squarebrackets).He explains, however,that "OAUcan be changed to NPU or U to give the same resultas statedin Theorem6" (p. 164). Robertsuses the strongParetoprinciple,but points out that the weak Paretocondition is sufficientfor the resultsof his paper(p. 144). Under the specified identificationof Roberts's frameworkwith our framework,FC is equivalentto (CNdimLperUper).
3.
4.
5.
6.
7.
8.
9. 10.
References
Arrow,K. (1951/2nd ed. 1963). Social choice and individualvalues. New York:Wiley. d'Aspremont,C. (1985). Axioms for social welfare orderings.In L. Hurwicz,D. Schmeidler and H. Sonnenschein(Eds.), Social goods and social organization:Essays in memoryof Elisha Pazner, 19-76. Cambridge: CambridgeUniversityPress. P.K. R. and Pattanaik, Batra, (1972). Transitivemulti-stage majoritydecisions with quasitransitiveindividualpreferences.Econometrica40: 1121-1135. Blair, D.H. (1988). The primary-goodsindexationproblem in Rawls's 'Theory of Justice'. Theoryand Decision 24: 239-252. P.J. in S. Barbera, J.A. (1996). Utility in social choice. Forthcoming Bossert,W. andWeymark, Hammondand C. Seidel (Eds.), Handbookof utilitytheory,Volume2. Boston: Kluwer. P.C. (1971). The theoryof representative Fishburn, majoritydecision. Econometrica39: 273284. Gibbard, A. (1979). Disparate goods and Rawls's difference principle: A social choice theoretictreatment. Theoryand Decision 11: 267-288. The role of informationin social welfare judgements. OxfordEconomic D. Kelsey, (1987). 39: 301-317. Papers
141
Khmelnitskaya,A.B. (1999). Social welfare orderingsfor different subgrouputility scales. Discussion paper #198, Centerfor Rationalityand InteractiveDecision Theory, Hebrew Universityof Jerusalem. Khmelnitskaya,A.B. and Weymark,J.A. (2000). Social choice with independentsubgroup utility scales. Social Choice and Welfare17: 739-748. a 'zero-line' of welfareas an escape-routefromArrow's List, C. (2001). A note on introducing Theorem.Pacific Economic Review 6 (special section in honour of AmartyaSen): 223238. ArrowProblem. List, C. (2002). Intradimensional single-peakednessandthe multidimensional Theoryand Decision 52: 287-301. Nussbaum, M. (1992). Human functionings and social justice: In defense of Aristotelian essentialism.Political Theory20: 202-246. Plott, C.R. (1978). Rawls's theory of justice: An impossibility result. In H.W. Gottingerand W. Leinfellner(Eds.), Decision theoryand social ethics, 201-214. Dordrecht: D. Reidel. Plott, C.R., Little, J.T. and Parks,R.P. (1975). Individualchoice when objects have 'ordinal' values. Reviewof EconomicStudies42: 403-413. Rawls, J. (1971). A theoryof justice. Cambridge,MA: Harvard UniversityPress. Roberts,K. (1995). Valuedopinions or opinionatedvalues:The double aggregationproblem. In K. Basu, P.K. Pattanaikand K. Suzumura(Eds.), Choice, welfare and development: A Festschriftin Honour of AmartyaSen, 141-165. Oxford:OxfordUniversityPress. Sen, A.K. (1970/79). Collective choice and social welfare. San Francisco:Holden-Day. Sen, A.K. (1980/81). Pluralutility.Proceedings of the AristotelianSociety 81: 193-215. Oxford:Blackwell. Sen, A.K. (1982). Choice, welfare,and measurement. North-Holland. Sen, A.K. (1985). Commoditiesand capabilities. Amsterdam: Sen, A.K. (1987). Thestandardof living. Cambridge: CambridgeUniversityPress. Sen, A.K. (1991). On indexing primarygoods and capabilities.Mimeo. Harvard University. Sen, A.K. (1997). On economic inequality.Enlargeded. Oxford:OxfordUniversityPress. United Nations DevelopmentProgramme(UNDP) (2002). Humandevelopmentreport2002. New Yorkand Oxford:OxfordUniversityPress. Walzer,M. (1983). Spheresofjustice. New York:Basic Books.
142
Appendix 1: Proof of Theorems 4 and 5

Both theoremscan be deducedfromtheorem6 in Roberts(1995). A formallyclosely related result is Khmelnitskaya(1999, theorem 2), also presentedwith a modified and Weymark(2000, theorem2), as mentionedabove. This proof in Khmelnitskaya resultuses a continuityconditionin additionto conditionsequivalentto the ones used here. The resultcan also he restatedfor the presentmultidimensional framework. "Theorem6. Let there be n opinions about the well-being of r individuals.(In n = r is permitted.)If f satisfies [U]8, I, P9, and FC10,then thereexists an particular, individuali and a strictly monotonicfunction W with r argumentssuch that, for all x, y X, W(u(x, i)) > W(u(y, i)) = xPy; i.e., individuali's opinions are dictatorial." We will explicitly deduce only theorem 2 from Roberts's theorem (hereafter simply referredto as R); theorem3 can be deducedanalogously. Assume, for a contradiction,that F is a MSWFL satisfying (CNdimLperUper), will also satisfy (U), (P), (I) and (Dom) (any MSWFL satisfying (ONdimLper) k (the numberof R of with We n in number (the opinions) (CNdimLperUper)). identify R n of individuals), r of in with number number (the dimensions), (the individuals) in i alternative about the welfare of under of x) R with Wij(x) u(x, i, j) (the opinion j f in R with F. It is and i in dimension under alternative welfare of individual x), (the j P R's and f satisfies conditions seen that FC, whence, by R, there exists U, I, easily a strictlyincreasingfunctionW with r (= n in our framework)argumentssuch that, for all x, y E X, W(u(x, -, i)) > W(u(y, ., i)) =i xPy (in R), i.e., for all x, y E X,
= xPy, f({Wij(x)}iEN)> f({Wij(y))iEN)

condition wherethe functionW in R is identifiedwith the functionf. This contradicts (Dom) in our framework. To deduce theorem3, simply identify n in R with n, r in R with k, u(x, j, i) in R with Wij(x), and f in R with F. O

List - Multidimensional Welfare Aggregation

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

List - Multidimensional Welfare Aggregation

Uploaded by

Copyright:

Available Formats

Multidimensional Welfare Aggregation Author(s): Christian List Reviewed work(s): Source: Public Choice, Vol. 119, No.

Multidimensional welfare aggregation *

2. Multidimensionality and socialchoice

and interdimensional 3. Measurability, interpersonal comparability comparability

in P. A (LC)- or (UC)-statementis invariantunder all the transformations

no furtherassumption no furtherassumptionrequired required

for each j E K, 4lj =

NdimLperUper LdimLper UdimUper

not applicable all fij are identical not applicable

and 3al, a2.... ak E R such that, for each i E N, il + al = 'i2 + a2 = ...

LdimUdimLperUper not applicable

See also appendix 1. Given comparabilityacross individualsbut not across

fk : Rn --+ R (possibly different), one

for somej e K for all h < j.

Prominent examples are:

a 'utilitarian'intradimensional function: aggregation

where 1, 2, . * > 0; - if (at least) (ONdimLper)iS satisfied : a 'maximin'intradimensional aggregation function:

+ o2W2j(x) + ... + ,nWnj(x),

fn : Rk -- R (possibly different), one for each

individualin N, and a permutation a of N such that,for any {Wi}ieN and any

x1Px2 if and only if

for some i E N for all h < i.

1) (weights corresponding to the n

such that, for any {Wi}ieN and any xl, x2 e X.

if and only if CieNaifi(<

1 CiNaifi (< Wij(x2)

. , P2, ..., Sk 8l, dimensions)

1) (weights correspondingto the k

such that,for any {Wi}iEN and xl, x2 e X,

mensionalaggregationfunctionsor 'maximin'intradimensional aggregation

and any x1, X2 E X,

135 Possible choices for fl, f2,

fn are, for example, 'maximin' intraper-

sonal aggregationfunctions or 'utilitarian'intrapersonal aggregationfunctions (requiring at least (CLdimUdimLperUper))as defined above.

: Rn -+ R; and fl, f2,....fk (ii) for each x e X, a permutation jf[l]({Wi[1](X)}iEN)

such that, for any {WilieN and any x1, X2 E X,

x1Px2 if and only if

Do we have interpersonal comparability?

Do we have interpersonal comparability?

Whichtype of interpersonal andinterdimensional comparability do we have? comparability

F is a' (possibly lexicographic) of hierarchy dimensions.

F is a (possibly lexicographic) double dictatorship.

Appendix 1: Proof of Theorems 4 and 5

= xPy, f({Wij(x)}iEN)> f({Wij(y))iEN)

You might also like