In order to draw inferences about unobservable motives from observable choices, the external observer defines a system of conditional probabilities, \(p\left( \cdot \right) \), which defines the set of possible outcomes that can be generated by a given combinations of events (motives). These constitute the observer’s beliefs about the way in which motives generate observed choices. For tractability, we will impose some regularity assumptions (or rules for inference) on the set of conditional probabilities.
3.1 Linking motives to social outcomes
In the presence of dual preferences, the decision-making process will need to consist of at least three separate stages. In the first two stages, the options are assessed separately according to the two motives, \(R_{i}^{U}\) and \(R_{i}^{M}\). The specific order of the idiosyncratic and reasoned assessments does not influence the results of this paper. What matters is that two separate and independent assessments are carried out before the final choice is made, which is stage three.
Given that the external observer can only see the final choices made but not the underlying motives that have led to the observed choice, the observer needs to impose a set of plausible assumptions, which we refer to as properties, about the decision-making process linking underlying preferences to choice. The properties will be stated as a system of conditional probabilities characterising the decision-making process from the point of view of the external observer.
Anonymity (ANO) For all \(i,j\in N\) such that \(i\ne j\) and for \(x,y\in X\), \(p( \chi _{i}^{U}\mid \omega _{i}^{x}) =p( \chi _{j}^{U}\mid \omega _{j}^{x}) \) and \(p( \psi _{i}^{U}\mid \omega _{i}^{y}) =p( \psi _{j}^{U}\mid \omega _{j}^{y})\).
The first property (ANO) captures the external observer’s assumption that the probability that a given act is motivated by idiosyncratic motives is equal across all individuals.
Independence (IND) For any \({\mathcal {I}}\in Q\) such that \(\vert {\mathcal {I}}\vert >1\) and for \(x,y\in X\), \(p( \chi _{{\mathcal {I}}}^{U}\mid \omega _{{\mathcal {I}} }^{x}) =\prod _{i\in {\mathcal {I}}}p(\chi _{i}^{U}\mid \omega _{i}^{x})\) and \(p( \psi _{{\mathcal {I}}}^{U}\mid \omega _{\mathcal { I}}^{y}) =\prod _{i\in {\mathcal {I}}}p(\psi _{i}^{U}\mid \omega _{i}^{y})\).
According the second property (IND), the external observer assumes that idiosyncratic motives of individuals are jointly independent within any subgroup of size greater than or equal to 2. In other words, the external observer assumes that individual i’s preference for, say, x over y is not influenced by whether other individuals within the same group prefer x over y. This does not necessarily imply that the chooser’s idiosyncratic motives must be independent of others’ idiosyncratic motives, but simply that the external observer does not have any information on whether and how such relations exist. Inferring dependence in individual motives by the external observer will require further assumptions that we do not consider in this paper.
Completeness (COM) For any \({\mathcal {I}}\in Q\) and for \(x,y\in X\), \(p( \chi _{{\mathcal {I}} }^{U}\cup \chi _{{\mathcal {I}}}^{M}\mid \omega _{{\mathcal {I}}}^{\cdot }) +p(\psi _{{\mathcal {I}}}^{U}\cup \psi _{{\mathcal {I}}}^{M}\)
\(\mid \omega _{ {\mathcal {I}}}^{\cdot })=1\).
The third property (COM) states that a given observed act, \(\omega _{ {\mathcal {I}}}^{\cdot }\), must have been motivated either by \(\chi _{\mathcal {I }}^{U}\cup \chi _{{\mathcal {I}}}^{M}\) or by \(\psi _{{\mathcal {I}}}^{U}\cup \psi _{{\mathcal {I}}}^{M}\). Note that \(\omega _{{\mathcal {I}}}^{\cdot }\) may represent either \(\omega _{i}^{x}\) or \(\omega _{i}^{y}\) or \(\omega _{i}^{\varnothing }\).
Coherence (COH) For any \({\mathcal {I}}\in Q\) and for \(x,y\in X\), \(p( \psi _{{\mathcal {I}} }^{\cdot }\mid \omega _{{\mathcal {I}}}^{x}) =0\) and \(p( \chi _{ {\mathcal {I}}}^{\cdot }\mid \omega _{{\mathcal {I}}}^{y}) =0\).
This fourth property (COH) states that if a group
\({\mathcal {I}}\) (which may consist of a single individual) is observed to have chosen
x over
y, then the external observer infers that
\({\mathcal {I}}\) could not have strongly preferred
y over
x, neither under idiosyncratic motives nor under moral motives. Similar statement applies when
y is observed to be chosen over
x. This is a Pareto axiom, stating that if preferences go in one direction, choice cannot go in the other direction.
4
Uncertainty (UNC) For any \({\mathcal {I}}\), \({\mathcal {J}}\in Q\backslash \{N\}\) and for \(x,y\in X\), \(p( \chi _{{\mathcal {I}}}^{U}\cap \chi _{{\mathcal {I}}}^{M}\mid \omega _{ {\mathcal {I}}}^{x}) =p( \chi _{{\mathcal {J}}}^{U}\cap \chi _{\mathcal { J}}^{M}\mid \omega _{{\mathcal {J}}}^{x}) \) and \(p( \psi _{\mathcal {I }}^{U}\cap \psi _{{\mathcal {I}}}^{M}\mid \omega _{{\mathcal {I}}}^{y}) =p(\psi _{{\mathcal {J}}}^{U}\cap \psi _{{\mathcal {J}}}^{M}\)
\(\mid \omega _{ {\mathcal {J}}}^{y})\).
The final property (UNC) states that when the external observer observes that a group has chosen, say,
x over
y, and it is possible for all idiosyncratic motives within that group to concord, the external observer acknowledges that there is a possibility of the decision being reached by a combination of both idiosyncratic and moral motives. However, in the absence of any information about the underlying decision making mechanisms, he assumes that the overlap between idiosyncratic and moral motives is the same across groups.
5 This property is only used to obtain the result in Corollary
1.
The five properties impose natural restrictions on beliefs based on observed choices, which will help the external observer to draw inference on unobservable motives from observed choice outcomes. In order to assign normative value to these choices, we must introduce an additional axiom that links the positive inference to a normative rule. If, like Sen (
1977), we consider that value stems from reason rather than from other preferences, it is logical to make the normative value of a given choice directly proportional on the probability that it is caused by
\(R_{i}^{M}\) and inversely proportional to the probability that it is caused by
\( R_{i}^{U} \). We do this by using the following normative axiom, called Probability-Based Rankings, which states that the external observer (1) normatively prefers the social outcome that is less likely to have been caused by idiosyncratic motives (i.e. a convergence of self-interests), and (2) is normatively indifferent between two social states that have equal probability of having been caused by idiosyncratic motives. In other words, from the perspective of the external observer, a greater concordance of choices made under idiosyncratic motives would not carry more normative strength, as such concordance would be merely coincidental, and could be reversed if the external interests leading to the choice were to change.
Probability-Based Rankings (PBR) For
\(x,y\in X\) and for any
\({\mathcal {I}},{\mathcal {J}}\in Q\) and
\(\omega _{ {\mathcal {I}}}^{x},\omega _{{\mathcal {J}}}^{x}\in Z\),
$$\begin{aligned} \omega _{{\mathcal {I}}}^{x}\succ & {} \omega _{{\mathcal {J}}}^{x}\Leftrightarrow p( \chi _{{\mathcal {I}}}^{U}\mid \omega _{{\mathcal {I}}}^{x}) <p( \chi _{{\mathcal {J}}}^{U}\mid \omega _{{\mathcal {J}}}^{x}) \end{aligned}$$
(1a)
$$\begin{aligned} \omega _{{\mathcal {I}}}^{x}\sim & {} \omega _{{\mathcal {J}}}^{x}\Leftrightarrow p( \chi _{{\mathcal {I}}}^{U}\mid \omega _{{\mathcal {I}}}^{x}) =p( \chi _{{\mathcal {J}}}^{U}\mid \omega _{{\mathcal {J}}}^{x}). \end{aligned}$$
(1b)