Abstract
This article explores under what conditions an agent can derive a transitive all-things-considered preference from a plurality of non-comparable objectives, values or judgements mirroring her plural identity. In contrast to existing contributions, the multiple values are reflected by partial (viz. incomplete) orderings. It is shown that a slight modification of the conditions employed by Arrow implies a spread of the dictate of one identity to its ‘incomplete parts’. The second result reveals that if one requires the decision making power to be spread a bit more equally across the various parts of a person’s identity, then even the derivation of an acyclic all-things-considered preference is rendered impossible. A third result shows that requiring some minimal consistency among a person’s plural identities, introduced via a domain restriction, allows the avoidance of the highlighted impossibilities.
Similar content being viewed by others
Notes
Arlegi and Teschl (2011) pursue a slightly different line of inquiry: they explore how at first sight inconsistent choice behaviour can be explained as the result of a person’s plural (possibly conflicting) motivations.
In contrast to the other contributions, Plott et al. (1975) adopt a fixed profile approach.
Note that others have questioned whether Arrow’s result is applicable to an intra-personal setting on other grounds. Hurley (1985) argues, for instance, that Arrow’s conditions cannot be reconciled with the supervenience of evaluative judgments on the non-evaluative characteristics of the alternatives.
Sen, however, as far as we know, never used this point as part of an argument that Arrow’s impossibility is less problematic in an intra-personal context. He is making this point in response to Arrovian-type impossibilities in the social context.
(Raz (1986), pp. 347–348) raises a similar case when he argues that for many people it is inconsistent with their identity as parents to rank (compare) their child(ren) with sums of money.
Note that the widespread definition of acyclicity, that is \(R\) on \(X\) is acyclic if, and only if, for all \( x_{1},x_{2},...,x_{j}\), \(j\) being a finite positive integer, \(x_{1}Px_{2}, x_{2}Px_{3},\ldots , x_{j-1}Px_{j}\) implies \( x_{1}Rx_{j}\), is equivalent to the one adopted here if the preference relation is complete.
Note, however, that as discussed in the introduction of this article, the plausibility of the preferences in the profile depends on their interpretation: it might be more plausible to allow for incompleteness if one adopts an interpretation of plural identities, reasons or values than under an interpretation of plural criteria or characteristics.
The sole way in which a change in a person’s identity over time enters the analysis is via a monotonicity condition, introduced in section 3. It requires some continuity of the choices a person derives from her identity, in case the composition of her plural identity changes. For an analysis of a person’s plural identities, or multiple selves over time, see Heilmann (2010).
There are other methodological reasons to impose Universal Domain: one is that we are interested in conditions which the Identity Aggregation Procedure of any person is supposed to satisfy. Thus the procedure is supposed to work for different persons, who are characterized by different identity profiles. Another (methodological) reason relates to the question the framework aims to address, namely under which conditions a person can indeed derive a transitive/acyclic all-things-considered preference. In order to gain a clear understanding of the structure of the aggregation problem at hand (and possible problems in a person’s preference formation) it is fruitful to start with the most admissible case, in order to then step-wisely restrict it in the light of emerging impossibility results.
Kelsey (1986) discusses the counter-intuitive flavour of a weaker form of imposition in the case in which the various identity parts, or objectives as he calls them, are defined very narrowly.
It can be shown that the condition of Strong Imposition coincides with Chichilnisky (1982) dictatorship property.
The term ‘lexicographic procedure’ refers to the following aggregation procedure (based on Houy and Tadenuma 2009, p. 1771), \(xPy\) if, and only if, (1) \(xP_{1}y\) or (2) \(xN_{1}y\) and \(xP_{2}y\), or \(\ldots ,\) or (m) \(xN_{1}y, \ldots , xN_{m-1}y\) and \(xP_{m}y\). The relation \(N\) on \(X\) is transitive if, and only if, for all \(x,y,z \in X, xNy\) and \(yNz\) implies \(xNz\).
Contributions in the literature (Gibbard 1969; Weymark 1984) have shown that—over a domain of complete and transitive orderings—requiring the social ordering to be complete and quasi-transitive leads to oligarchies. An oligarchy is a set of individuals whose (i) unanimous strict preference over any two alternatives \(x\) over \(y\), implies \(x\) to be socially preferred to \(y\); and (ii) the members of the oligarchy have veto-power, i.e. if some member of the oligarchy strictly prefers \(x\) over \(y\), this implies that it is not the case that \(y\) is socially preferred to \(x\). Furthermore, results in the judgment aggregation literature (Gärdenfors 2006; Dietrich and List 2008) indicate the emergence of ‘strong oligarchies’ if one allows for a domain of possibly incomplete judgment sets (Dietrich and List 2008, p. 27). In a social choice framework, strong oligarchies can be defined as a set of individuals who do not only dictate their unanimously held strict preferences on the social preference, but also dictate incompleteness between any two alternatives \(x\) and \(y\) which are not ranked by all members of the strong oligarchy. It can be shown that a strengthening of the Arrovian conditions to those employed in this article will imply the strong oligarchy to be a singleton which is equivalent to strong imposition as defined in this article.
Fishburn (1970) considers inter-personal aggregation problems over a domain of quasi-transitive preference profiles. The difference between theorem 2 and his example (Fishburn 1970, p.482) is that he illustrates the occurrence of cycles for \(m =3\) and \(\# X = 3\), whereas theorem 2 of this article holds for \(m \ge 2\) and \(\# X > 3\). I am very grateful to an anonymous referee for making me aware of this point made by Fishburn.
In this section we focus on acyclicity and allow a person’s all-things-considered preference to be incomplete. If acyclicity were satisfied, however, it can be shown that a person can always complete the resulting all-things considered preference such that the imposed conditions are satisfied.
Costello (2004) uses the notion of ‘identity dissonance’ in relation to a problem that differs from the one addressed in this paper, namely the possibility that ‘identity dissonance’ might explain the persistent failure of affirmative action programmes in the schoolroom.
To be more precise, Fishburn (1970) is concerned with inter-personal aggregation problems and considers a sub-domain of profiles of quasi-transitive orderings. He shows that for his sub-domain simple majority rule yields a quasi-transitive social ordering. If indifference (\(I\) on \(X\)) between alternatives is replaced by incompleteness (\(N\) on \(X\)) in his definition, then it can be shown that Value Overlap is contained in his sub-domain (Definition \(A\) in Fishburn 1970, p. 484), which in turn guarantees a transitive all-things-considered preference.
The condition Value Overlap in this article was inspired by Sen (1969) whose domain condition Value Restriction is sufficient for Majority Rule to generate quasi-transitive social preference orderings over a domain of profiles of weak orders. Pattanaik (1970) provides a generalization by establishing sufficient conditions for the class of binary, decisive, neutral, and non-negatively responsive social decision procedures to yield quasi-transitive social preferences over a domain of profiles of quasi-transitive individual preferences. It can be shown that the domain generated by Value Overlap is contained in Pattanaik’s domain condition VR* (Pattanaik 1970, p. 270).
Note that the formal results obtained in this paper can also be interpreted in an inter-personal context. Especially the interpretation of theorem 3 might be of particular interest in this respect, since it shows how possible restrictions upon the plurality of a society can lead to an escape from Arrovian impossibility results.
References
Arlegi R, Teschl M (2011) Theory of choice under internal conflict. Working Papers of the Department of Economics DT 1208, Public University of Navarre
Arrow K (1963) Social choice and individual values. Yale University Press, New Haven
Baigent N (1995) Behind the veil of preference. Jpn Econ Rev 46:88–101
Bratman ME (2007) Structures of agency. Oxford University Press, Oxford
Costello CY (2004) Changing clothes: gender inequality and professional socialization. NWSA J 16: 138–155
Chichilnisky G (1982) The topological equivalence of the Pareto condition and the existence of a dictator. J Math Econ 9:223–233
Davis J (2003) The theory of the individual in economics. Identity and value. Routledge, New York
Dietrich F, List C (2008) Judgment aggregation without full rationality. Soc Choice Welf 31:15–39
Ellis S (2006) Multiple objectives: a neglected problem in the theory of human action. Synthese 153: 313–338
Fishburn PC (1970) Intransitive individual indifference and transitive majorities. Econometrica 38: 481–489
Gärdenfors P (2006) An Arrow-like theorem for voting with logical consequences. Econ Philos 22:181–190
Gibbard A (1969) Intransitive social indifference and the Arrow dilemma. Working Paper, University of Chicago
Heilmann C (2010) Rationality and time: a multiple-self model of personal identity over time for decision and game theory. Dissertation, London School of Economics
Houy N, Tadenuma K (2009) Lexicographic compositions of multiple criteria for decision making. J Econ Theory 144:1770–1782
Hurley SL (1985) Supervenience and the possibility of coherence. Mind 94:501–525
Jeffrey RC (1974) Preference among preferences. J Philos 71:377–391
Kelsey D (1986) Utility and the individual: an analysis of internal conflicts. Soc Choice Welf 3:77–87
May KO (1954) Intransitivity, utility, and the aggregation of preference patterns. Econometrica 22:1–13
Pattanaik PK (1970) On social choice with quasi-transitive individual preferences. J Econ Theory 2: 215–267
Plott CR, Little JT, Parks RP (1975) Individual choice when objects have “ordinal” properties. Rev Econ Stud 42:403–413
Raz J (1986) The morality of freedom. Oxford University Press, New York
Sen AK (1969) Quasi-transitivity, rational choice and collective decisions. Rev Econ Stud 36:381–393
Sen AK (1970) Collective choice and social welfare. Holden Day, Cambridge
Sen AK (1974) Choice, orderings and morality. In: Körner S (ed) Practical reason. Blackwell, Oxford, pp 54–67
Sen AK (1977) Rational fools. Philos Public Aff 6:317–344
Sen AK (1985) Commodities and capabilities. Elsevier, Amsterdam
Sen AK (1997) Maximization and the act of choice. Econometrica 65:745–779
Sen AK (2002) Rationality and freedom. Harvard University Press, Cambridge
Sen AK (2004) Incompleteness and reasoned choice. Synthese 140:43–59
Sen AK (2006) Identity and violence. Norton, New York
Steedman I, Krause U (1986) Goethe’s faust, Arrow’s possibility theorem and the individual decision taker. In: Elster J (ed) The multiple self. Cambridge University Press, New York, pp 197–232 Studies in Rationality and Social Change
Styron W (1979) Sophie’s choice. Random House, New York
Weymark J (1984) Arrow’s theorem with social quasi orderings. Public Choice 42:235–246
Acknowledgments
This work is part of research conducted at the faculty of philosophy of Groningen university. It formed part of the research programme ‘Modelling Freedom: Formal Analysis and Normative Philosophy’ which was financed by the Netherlands Organisation for Scientific Research (NWO). I am very grateful to Martin van Hees for his detailed comments on earlier versions of this paper. Furthermore, I am indebted to Nick Baigent, Matthew Braham, Boudewijn de Bruin, Franz Dietrich, Wulf Gaertner, Conrad Heilmann, Frank Hindriks, Christian List, Hendrik Rommeswinkel, Olivier Roy, Jan-Willem van der Rijt, Miriam Teschl, Yongsheng Xu and an anonymous referee for their comments and suggestions.
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
Definition 1
Identity Aggregation Procedure An Identity Aggregation Procedure (IAP) is a function \(f: \fancyscript{D} \rightarrow \fancyscript{R}_{a},\; \fancyscript{D}\subseteq \fancyscript{R}_{s}^{M}\).
Condition,
Unrestricted Domain \(f(.)\) satisfies Unrestricted Domain (UD) iff \(\fancyscript{D}=\fancyscript{R}_{s}^{M}\).
Definition 2
Minimal Decisiveness For all \(i \in M\), for all \(x\in X,\; i\) is minimally decisive for \(\{x,y\}\) (or for \(x\) against \(y\)) iff for all \(d \in \fancyscript{D}\): [\(xP_iy\) and \(xN_jy\) for all \(j \not = i\)], implies \(xPy\), and [\(yP_ix\) and \(xN_jy\) for all \(j \not = i\)], implies \(yPx\).
Condition,
Minimal Identity Decisiveness \(f(.)\) satisfies Minimal Identity Decisiveness (MID) iff for all \(x,y \in X\) there exists \(i \in M\) such that \(i\) is minimally decisive for \(x\) against \(y\).
Condition,
Monotonicity \(f(.)\) satisfies Monotonicity (M) iff for all \( d, d' \in \fancyscript{D},\) for all \(x,y \in X\), if \([\) for all \(i \in M (xP_{i}y\) implies \(xP'_{i}y)\) and there does not exist \(j \in M\) such that \((xN_{j}y\) and \(yP'_{j}x)]\) then \([xPy\) implies \(xP'y]\).
Condition,
Strong Imposition \(f(.)\) is strongly imposed iff there exists \(i \in M\) such that for all \(x,y \in X, xP_{i}y\) implies \(xPy\) and \(xN_{i}y\) implies \(xNy\).
Condition,
All-things-considered Transitivity \(f(.)\) satisfies All-things-considered Transitivity (TP) iff its range is restricted to \(\fancyscript{R}_{s}\).
Lemma 1
Suppose \(f(.)\) satisfies conditions TP, UD, MID, and M. For \(m \ge 2\) and \(\# X \ge 3\), for all \( x,y \in X\), if for all \(i \in M, xN_{i}y\), then \(xNy\).
Proof
Suppose to the contrary, namely that TP, UD, MID, and M are satisfied, but there exists \( x,y \in X\) and \(d^{1} \in \fancyscript{D}\) such that for all \(i \in M, xN_{i}y\) and \(xP^{1}y\). Let \(z \in X-\{x,y\}\).
By MID there is some element of \(M\) that is minimally decisive for \(\{y,z\}\). Say \(j\) is so, and consider a profile \(d^2\) in which
-
(a)
for all \(i \not = j\): \(xN^2_iy\), \(yN^2_iz\) and \(xN^2_iz\)
-
(b)
\(xN^2_jy\), \(yP^2_jz\) and \(xN^2_jz\).
By M we obtain \(xP^2y\), by MID \(yP^2z\), and by TP \(xP^2z\).
By MID some \(k \in M\) is minimally decisive for \(\{x,z\}\). Suppose \(j \not = k\) and consider a profile \(d'^2\) in which the following holds:
-
(a)
for all \(i \not = j, k\): \(xN'^2_iy\), \(yN'^2_iz\) and \(xN'^2_iz\)
-
(b)
\(xN'^2_jy\), \(yP'^2_jz\) and \(xN'^2_jz\)
-
(c)
\(xN'^2_ky\), \(yN'^2_kz\) and \(zP'^2_k x\).
By M we obtain \(xP'^2y\), by MID \(yP'^2z\), and \(zP'^2x\), violating TP. Hence \(j = k\) and \(j\) is also minimally decisive for \(\{x,z\}\).
Consider a profile \(d^3\) in which
-
(a)
for all \(i \not = j\): \(xN^3_iy, yN^3_iz, xN^3_iz\)
-
(b)
\(xN^3_jy, yN^3_jz, zP^3_jx\).
By M we obtain \(xP^3y\), by MID \(zP^3x\), and by TP \(zP^3y\).
By MID some \(l \in M\) is minimally decisive for \(\{x,y\}\). Take a profile \(d^4\):
-
(a)
for all \(i \not = l\): \(xN^4_iy, yN^4_iz, xN^4_iz\)
-
(b)
\(yP^4_lx, yN^4_lz, xN^4_lz\).
By MID we obtain \(yP^4x\), and by M, \(zP^3y\) and \(xP^2z\), we derive \(zP^4y\), and \(xP^4z\), violating TP. \(\square \)
Theorem 1
For \(m \ge 2\) and \(\# X \ge 3\), every \(f(.)\) that satisfies TP, UD, MID, and M is strongly imposed.
Proof
(1) In the first step we prove that for any triple \(\{x,y,z\}\) there is some \(i \in M\) that is minimally decisive for all distinct \(a, b \in \{x,y,z\}\).
By MID for each pair of alternatives \(a, b \in \{x,y,z\}\), there is some part of the person’s identity that is minimally decisive for it. Assume there is no identity part that is minimally decisive for at least two pairs in \(\{x,y,z\}\). There must then be three different identity parts \(i\), \(j\) and \(k\), such that \(i\) is minimally decisive for \(x,y\), \(j\) for \(y, z\) and \(k\) for \(x,z\). Now consider a profile \(d\) satisfying:
-
(a)
\(xP_{i}y\) and \(aN_{i}b\) for all \(\{a,b\} \ne \{x,y\}\)
-
(b)
\(yP_{j}z\) and \(bN_{j}c\) for all \(\{a,b\} \ne \{y,z\}\)
-
(c)
\(zP_{k}x\) and \(aN_{k}c\) for all \(\{a,b\} \ne \{x,z\}\)
-
(d)
\(aN_lb\) for all \(a, b \in X\) and all \(l \not \in \{i,j,k\}\).
By UD, any such \(d\) is in the domain. By MID we get \(xPy, yPz\) and \(zPx\), which violates TP. Hence, there is some \(i\) that is minimally decisive for at least two pairs in \(\{x,y,z\}\).
We now show that \(i\) is minimally decisive for all pairs in \(\{x,y,z\}\). Without loss of generality, assume \(i\) is minimally decisive for the pairs \(\{x,y \}\) and \(\{y,z\}\). Consider any profile \(d\), such that:
-
(a)
\(xP_{i}y\), \(yP_{i}z\) and \(xP_{i}z\)
-
(b)
\(xN_{k}y\), \(yN_{k}z\) and \(zN_{k}x\) for all \(k\not = i\).
By MID we get \(xPy\) and \(yPz\) and thus by TP, \(xPz\).
In a similar way, we can construct a profile \(d''\) in which \(zP''_i x\) and \(xN''_jz\) for all \(j \not =i\) and in which \(zP''x\). Hence, \(i\) is minimally decisive for \(\{x,z\}\).
(2) We next show that for any triple \(\{x,y,z\}\) if some \(i \in N\) is minimally decisive for all distinct \(a, b \in \{x,y,z\}\), then for all profiles \(d\), and all \(a,b \in \{x,y,z\}\), \(aP_ib\) implies \(aPb\) and \(aN_ib\) implies \(aNb\), that is, \(i\) is decisive for \(a\) against \(b\).
Let \(i\) be as defined and assume \(xP_iy\). We want to show that \(xPy\). Take an arbitrary profile \(d'\) satisfying:
-
(a)
\(xP'_{i}y\), \(zP'_{i}y\), \(xP'_{i}z\)
-
(b)
\(yP'_{j}x\), \(yN'_jz\) \(xN'_{j}z\) for all \(j \not = i\).
By MID we get \(xP'z\) and \(zP'y\) and, by TP, \(xP'y\).
Since \(xP'y\), by M also \(xPy\), which had to be shown.
Next let \(d\) be an arbitrary profile in which \(xN_{i}y\). We now have to show that \(xNy\). Assume to the contrary that it is not the case that \(xNy\), say we have \(xPy\). Let \(D\) be the set of all profiles \(d'\) satisfying:
-
(a)
\(xN'_{i}y\), \(zN'_{i}y\), \(zP'_{i}x\)
-
(b)
\(zN'_{j}x\) and \(zN'_jy\), for all \(j \not = i\).
There must be some profile \(d' \in D\) such that the restriction of \(d'\) to \(\{x,y\}\) is exactly the same as it is in \(d\). Since \(xPy\), by M also \(xP'y\). By MID we have \(zP'x\), and thus by TP \(zP'y\), contradicting Lemma 1.
(3) We show that if there is some \(i\) such that \(i\) is decisive for any pair over the triple \(\{x,y,z\}\), then \(i\) is decisive for any pair of alternatives in \(X\).
Let \(i\) be as defined. We have to show that \(i\) is decisive for any \(v,w \in X\). If \(\{v,w\} \subseteq \{x,y,z\}\), we already have the desired result. Say \(v \not \in \{x,y,z\}\). Consider the triple \(\{v,x,y\}\). By step 1 some \(j\) is minimally decisive for each pair in \(\{v,x,y\}\), and thus also for \(\{x,y\}\). By step 2, at most one individual can be decisive for \(x\) and \(y\), and therefore \(i = j\) and \(i\) is decisive for \(\{v,y\}\). If \(w \in \{v,x,y\}\), the result follows immediately. Let \(w \not \in \{v,x,y\}\). By step 1 some \(j\) is minimally decisive for each pair in \(\{v,w,y\}\). Again, by step 2, there is at most one individual decisive for \(v\) and \(y\) and we have just shown this to be \(i\). Hence, \(i\) is decisive for each pair in \(\{v,w,y\}\), and thus in particular for \(\{v, w\}\). \(\square \)
Condition,
Identity Decisiveness \(f(.)\) satisfies Identity Decisiveness (ID) iff for all \(d \in \fancyscript{D}\), all \(i \in M\), and all \(x,y \in X\): if \([xP_{i}y\) and \(xN_{j}y\) for all \(j \not =i\)], then \(xPy\).
Theorem 2
For \(m \ge 2\) and \(\# X > 3\), there is no \(f(.)\) that satisfies UD and ID.
Proof
Consider the set of profiles \(D\) satisfying for some distinct elements \(x, y, z, w\) of \(X\):
-
(a)
\(xP_{i}y\), \(yN_{i}z\), \(zP_{i}w\), \(wN_{i}x\)
-
(b)
\(xN_{j}y\), \(yP_{j}z\), \(zN_{j}w\), \(wP_{j}x\)
-
(c)
for all \(k \not = i, j \) and all \( x,y \in X\): \(xN_ky\).
By UD, the class of profiles \(D\) is within our domain. By ID, one obtains \(xPy\), \(yPz\), \(zPw\) and \(wPx\), violating acyclicity. \(\square \)
Condition,
Imposition \(f(.)\) is imposed iff there exists \(i \in M\) such that for all \(x,y \in X, xP_{i}y\) implies \(xPy\).
Condition,
Value Consistency \(f(.)\) satisfies Value Consistency (VC) iff \(\fancyscript{D}=\{d \in \fancyscript{R}_{s}^{M}\mid \) for all \(x,y,z \in X\), for all \(i,j \in M, xP_i y\) and \(yP_j z\) implies that there exists \(k \in M\) such that \(xP_{k}z\}\).
Theorem 3
If VC is satisfied there is an IAP that satisfies ID, M and TP and is not imposed.
Proof
An Identity Aggregation Procedure which satisfies the conditions is the Modified Intersection Rule \(\hat{f}\). It is defined as follows (where \(\hat{R}, \hat{P}\) and \(\hat{N}\) denote the relations \(R, P\) and \(N\) assigned by \(\hat{f}\)):
For all \(x, y \in X\), \(x \hat{R}y\) if, and only if, (a) \(x =y\), or (b) for all \(i \in M\), not \(yP_i x\) and for some \(j \in M\), \(xP_jy\).
The proof that the modified intersection rule satisfies ID, M and is not imposed is straightforward. Therefore we only prove TP.
Transitivity: We prove by contradiction. Suppose there exists some \(x,y,z \in X\) such that \(x\hat{P}y, y\hat{P}z\) and not \((x\hat{P}z)\). By definition of the modified intersection rule, we have
-
1.
For all \( i \in M,\) not \((yP_{i}x)\), not \((zP_{i}y)\);
-
2.
there exists \(i,j\in M\), such that \(xP_{i}y\) and \(yP_{j}z\).
By VC, \(xP_{i}y\) and \(yP_{j}z\) implies that there exists \(k \in M\) such that \(xP_{k}z\). Since, by assumption, not \((x\hat{P}z)\) there exists some \(l \in M\) such that \(zP_{l}x\). By VC, \(zP_{l}x\) and \(xP_{i}y\) imply that there exists \(m \in M\) such that \(zP_{m}y\), contradicting 1. \(\square \)
Rights and permissions
About this article
Cite this article
Binder, C. Plural identities and preference formation. Soc Choice Welf 42, 959–976 (2014). https://doi.org/10.1007/s00355-013-0761-z
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00355-013-0761-z