State-Salient Decision Rules

Somdeb Lahiri

This Version: October 8, 2019.

Abstract

We suggest that the celebrated Impossibility Theorem of Arrowian social choice theory is
after all not such a negative result since it can be interpreted as an axiomatic characterization
of state-salient decision rules of a decision maker with state-dependent rankings of
alternatives. We provide examples to show that there are meaningful choice functions which
are not state-salient decision rules.

Keywords: choice function, Arrow's Impossibility Theorem, state-salient decision rules

JEL Classification: D03, D81

1. Introduction

One implication of Arrow’s Impossibility Theorem for which an eclectic and reasonably up-
to-date bibliography of the vast literature associated with it is available at
https://plato.stanford.edu/entries/arrows-theorem/, may be that the underlying framework is
not appropriate for the study of group decision making. Perhaps group decision making
usually does not take place that way. I don’t know why I have never come across such a
statement. Ignorance on my part about comments about the Impossibility Theorem and the
fact that I may not have been able to appreciate the underlying philosophy (not the proof) of
the said theorem are equally likely reasons for me to have never seen a statement about the
inappropriateness of the Arrowian framework for the study of group decision making. In any
case, I don't think there is any harm in suggesting an alternative interpretation that works.
This paper is just that- an interpretation of the Arrowian framework and the impossibility
theorem that (also?) makes good sense. We provide a realistic example that satisfies the
conditions of the theorem and to emphasize that the theorem is meaningful and has content,
we provide two examples that do not satisfy the conditions. Let me emphasize that this paper
is not intended to be a contribution to the vast literature on social choice theory, and its
purpose would be defeated if one tried to view it that way. It is a paper concerned with
decision making under risk and uncertainty.

2. The Framework

The motivation for the framework of this note can be found in Lahiri (2019).
Let X be a non-empty finite set of alternatives containing at least three elements, from which
a decision maker is required to choose exactly one alternative. There is a non-empty finite set
of future states of nature, which unravel only after the decision maker has made his/her
choice. For a positive integer n 3, let N = {1,2,...,n} denote the set of states of nature.
A strict (preference) ranking on X is a reflexive, complete/connected/total, transitive and
ant-symmetric binary relation on X. A strict ranking is also known as a linear order.
Generally a preference relation is denoted by R with P and I denoting its asymmetric and
symmetric parts respectively. If for x,y∈X, it is the case that (x,y)∈R, then we shall denote it
by xRy and say that x is at least as good as y. Similarly xPy is interpreted as x is strictly
preferred to y, and xIy is interpreted as there is indifference between x and y.

Let ℒ denote the set of all strict rankings on X.

A (state-dependent) preference profile denoted RN is a function from N to ℒ. RN is

represented as the array <Ri|i∈N>, where Ri is the strict ranking of the decision maker of the
alternatives in X in state of nature i. The set of all preference profiles is denoted ℒ .

A choice function (CF) is a function f from ℒ to X.

A CF f is said to satisfy the Weak Dominance Criterion (WDC) if for all x,y∈X and
RN∈ℒ : [xPiy for all i∈N] implies [y≠f(RN)].
This criterion is available and discussed in Lahiri (2019).
The following criterion is one based on a more general property in Denicolo (1993).
A CF f is said to satisfy Independence of Irrelevant Alternatives in the sense of Denicolo
(D-IIA) if for all RN, R ∈ℒ and x,y∈X with x ≠ y: [Ri|{x,y} = R |{x,y} for all i∈N, x=
f(RN)] implies [y≠f(R )].
The following definition can be found in Lahiri (2019).
A CF f is said to be state-salient between a given pair of distinct alternatives at state of
nature i, if regardless of the state-dependent preference profile, the alternative in the given
pair that is ranked inferior of the two in state of nature i, is not chosen. Thus, if the two
alternatives are x and y and if it is the case that (a) if x is ranked above y in state i, then y is
not chosen regardless of the rankings in other states of nature, and (b) if y is ranked above x
in state i, then x is not chosen regardless of the rankings in other states of nature, then the
individual is decisive between x and y in state of nature i.
A CF f is said to be state-salient at a state of nature i, if it is state salient between every pair
of distinct alternatives at state of nature i.
A CF f is said to be a state-salient decision rule (S-SDR), if there exists a state of nature i,
such that f is state salient at state of nature i.
An important and realistic example of an S-SDR is the maximum likelihood decision rule.
A CF f is said to be a maximum likelihood state decision rule (MLSDR) if there exists a
probability distribution p on N (i.e. p∈ℝ with p = 1) and a state of nature i such that pi
pj for all j∈N\{i} (i.e. i∈argmax ∈ {p }) such that for all RN∈ℒ and x,y∈X: xPiy implies
y≠f(RN). In this case i is said to be a most likely state of nature (MLS).
The version of MLSDR when state-dependent rankings are represented by state-dependent
evaluation or utility functions is available on page 74 in chapter 4 of Mc Kenna (1980) under
the name “maximum likelihood criterion”. The same is also defined and discussed under a
different name – “modal outcome criterion”- on page 429 (section 9.4.2) of Monahan (2000).
However as correctly observed by several authors, using this criterion in a situation where
several states of nature exist, with probability of occurrence nearly or exactly equal to the
probability of occurrence of a most likely state of nature, may lead to serious mistakes.
3. The main result based on one a theorem due to V. Denicolo (1993)

Theorem 1: A CF satisfies WDC and D-IIA if and only if it is any S-SDR.

4. Choice functions which are not S-SDR’s

As for all other meaningful solution concepts, there are many realistic choice functions which
do not fall in the category of an S-SDR. The following leads to such a choice function.
Given any strict ranking R on X, let first (R) denote the alternative that is ranked first (at the
top) by R, i.e. xPy for all y∈X\{x}.
Let p be any probability distribution on N and suppose that i is a state of nature such that pi
pj for all j∈N\{i}, i.e. i is a most likely state of nature for p.

Let f be the CF such that for all RN∈ℒ , f (RN) = x if and only if either
∈{ ∈ | ! " #}
p > $ (i.e. the probability that the state of nature will be such that x is the
best choice is greater than $, in which case x will of necessity be the unique alternative to
satisfy this property) or {y∈X| ∈{ ∈ | !" %} p > } = φ and first (Ri) = x.
$

Clearly, f satisfies WDC. However it does not satisfy D-IIA as the following example
reveals.
$ '
Example 1: Let X = {x,y,z} and N = {1,2,3}. Let p1 = , p2 = p2 = . Clearly 1 is the most
& (
likely state of nature for p. Suppose xR1yR1z, zR2xR2y, yR3zR3x. Since none of the
alternatives is ranked first with probability greater than half, f chooses x and z is not chosen.
Now suppose we have a preference profile where in the first and second states of nature, the
rankings are identical to the ones in the earlier situation. However in the third state of nature
'
y and z interchange positions. In this new profile the probability that z will be ranked first is &
which is greater than half. Hence f chooses z now, although in both the profiles the relative
positions of x and z are the same.

Thus, f violates D-IIA and by Theorem 1, it cannot be a state-salient decision rule.

A CF f is said to satisfy the maximum likelihood alternative property (MLA) if there
exists a probability distribution p on N, such that for all RN∈ℒ , f(RN) = x implies
x∈argmax%∈) ∈{ ∈ | ! " %}
{p }.

A CF that satisfies MLAP is the following.

Given a probability distribution p on N, the CF f* is such that for all RN∈ℒ , f* (RN) = x if
and only if:

(i) x∈argmax%∈) ∈{ ∈ | !" %}{p }

(ii) min{j∈N|first (Rj) = x} min{j∈N|first (Rj) = z} for all
z∈argmax%∈) ∈{ ∈ | ! " %}
p.

If argmax%∈) ∈{ ∈ | !" %}{p } is a singleton then (ii) becomes redundant.

Clearly, f* satisfies WDC. However it does not satisfy D-IIA as the following example
reveals.
$
Example 2: Consider the scenario in example 1, i.e. X = {x,y,z}, N = {1,2,3}, p1 = , p2 = p2
&
'
= . Suppose xR1yR1z, zR2xR2y, yR3zR3x. Then, f* chooses x and z is not chosen.
(

Now as in example 1, suppose we have a preference profile where in the first and second
states of nature, the rankings are identical to the ones in the earlier situation. However in the
third state of nature y and z interchange positions. In this new profile the probability that z
'
will be ranked first is , which exceeds the probability that x will be ranked first and also the
&
probability that y will be ranked first. Hence f* chooses z although in both the profiles the
relative positions of x and z are the same.

Thus, f* violates D-IIA and by Theorem 1, it cannot be a state-salient decision rule.

Acknowledgment: I would like to thank Professor Prasanta Pattanaik for motivation and
inputs (both direct and indirect) in writing this paper. His role as my mentor for decision
theory in the Arrow-Sen tradition is undeniable. I would like to thank Carlos Bouza and
Roberto Veneziani for much encouragement and encouraging comments while and after
writing this paper. However responsibility for errors and views expressed, in this paper rests
entirely on the author of this paper and no one else.

References

1. Denicolo, Vincenzo (1993): Fixed Agenda Social Choice Theory: Correspondence and
Impossibility Theorems for Social Choice Correspondences and Social Decision Functions.
Journal of Economic Theory, vol. 59(2), pages 324-332, April.
2. Lahiri, Somdeb (2019): Re-interpreting the Arrow-Sen framework of social choice.
https://www.academia.edu/40224623/Re-interpreting_the_Arrow-
Sen_Framework_of_Social_Choice
3. Monahan, George (2000): Management Decision Making: Spreadsheet Modeling,
Analysis, and Application. Cambridge University Press.
4. Mc Kenna, Christopher (1980): Quantitative Methods for Public Decision Making (First
Edition). New York: Mc Graw Hill.