Handbook of Well-Being and Public Policy: Extended Preferences Analysis

Extended Preferences Oxford Handbook of Well-Being and Public Policy Matthew D. Adler, Duke University Princeton Conference, Febuary 2014

What Ill cover (core of chapter) Harsanyi s account of extended preferences Criticisms of Harsanyi My own account of extended preferences (Adler 2012, 2014, this chapter) A simple case A comparison with other measures of well- being: objective indices; utility as currently measured in SWF literature; happiness; equivalent incomes

Harsanyis account A history is a possible life: a hybrid bundle of attributes and tastes. h = (a, R) A deliberator (observer, spectator) k, in the course of ethical deliberation, formulates extended preferences : a ranking of histories and lotteries, consistent with vNM utility theory, represented by wk(h). Principle of acceptance. wk(a,R) wk(a*,R) iff a R a*, and similarly for lotteries. The deliberator formulates her extended preferences via empathetic projection, standing in the shoes. All deliberators have the same extended preferences. Thus there is a single w(.), unique up to a positive affine transformation, and determinate, deliberator-independent interpersonal (and intra-personal) comparisons of well-being levels and differences. w(.) is the input to a utilitarian SWF

Criticisms of Harsanyi Nonshareable/essential attributes. h = (a, R) may include attributes that the deliberator k, given her theory T of personal identity, can t imagine acquiring Why assume that the deliberator s measure of well-being differences must emerge from a vNM function tracking her preferences over history lotteries? The principle of acceptance ignores the robust philosophical tradition of objective goods : ways in which individuals might benefit even in the teeth of their (laundered) preferences No reason to believe that deliberators will have the same extended preferences wk(.) can be used as the input to a non-utilitarian social evaluation function (e.g., a prioritarian SWF, or an inequality or poverty metric). Harsanyi s arguments for utilitarianism remain controversial.

A Revised Account A is a set of attributes. For each (maximal) shareable subset As, there is an associated set of tastes Rs. A history is a combination h = (a, R), with bundle a from some Asand R the associated Rs. H is the set of histories. The deliberator has an extended-preference structure k, which includes a quasiordering of H and a difference quasiordering of H x H, and which she arrives at via explicit judgments of well-being levels/differences, or proxies. I judge h at least as good for well-being as h*. I judge the difference between h and h* to be at least as large as the difference between h** and h***. A rich literature on difference quasiorderings, although economists tend to be suspicious. Must conform to substative limitations, e.g., Reversal: (h, h*) k(h**, h***) iff (h***, h**) k(h*, h). If kis complete and also satisifes technical axioms, will be representable by wk(.) unique up to positive affine transformation. If zeroed out, unique up to ratio trans.

A Simple Case A is shareable. Everyone in the population could, conceivably, have any bundle in A. Bernoulli . The deliberator can use her vNM function vk(.), representing her ranking of history lotteries, to proxy her judgments of well-being differences. (h, h*) k(h**, h***) iff vk(h) vk(h*) vk(h**) vk(h***) Sympathy connection. The deliberator ranks one history above a second iff she would self-interestedly prefer the first (and similarly for lotteries). Three versions of this case: Taste Independent: vk(a, R) = uk(a) Sovereignty-Respecting: vk(a, R) = sk(R)uR(a) + tk(R) Nuanced: vk(a, R) = sk(R)uk,R(a) + tk(R) Set wk(.) = vk(.) or, if we need ratio-scale, wk(a, R) = vk(a,R) vk(hzero)

A Simple Case, continued Harsanyi s approach emerges as one specification of a much broader approach. Shareability (deliberator can think in terms of well-being judgments regarding histories and lotteries, or self-sympathetic preferences to have attributes); Bernoulli (deliberator proxies judgments of well-being differences with vNM function representing ranking of lotteries); Sovereignty Respecting = Principle of Acceptance (deliberator s ranking of R-containing histories and lotteries defers to R). In this case, her only task is to identify the scaling factors sk(R) and tk(R) via cross-taste judgments: (a, R) just as good as (a*, R*), etc. To simplify this further? Two bundles a+, a++ such that (a+, R) just as good as (a++, R*) for all R, R*.

Pooling extended preferences? The deliberator k ( social planner ) is someone engaged in ethical/moral evaluation: a government official, a philanthropist, a citizen. The extended-preference framework is simply a normative recommendation to such a person. Specifically, it is one part of a broader recommendation that the deliberator s evaluation should be welfarist and consequentialist: that she evaluate choices (policies) via a ranking of outcomes that combines her measure of well-being wk(.) with a moral evaluation Mk(an SWF or, perhaps, an inequality metric or poverty metric). Outcome x morally at least as good as outcome y iff: (wk(h1(x)), ,wk(hN(x))) Mk (wk(h1(y)), ,wk(hN(y))) or more generally for all wk(.) in Wk. Each deliberator makes her own well-being judgments; but (in a process of moral dialogue) might include others well-being measures in Wk

A comparison to other well-being measures An objective index: ok(hi) = ok(ai). Corresponds to taste-independent version of wk(.) Utility in SWF literature: uk(hi) = uk(ai). Best interpreted as sovereignty-respecting version of wk(.) and simplified assumption of homogeneous preferences R, i.e., uk(ai) = uR(ai) Self-rated happiness: qk(hi) = swbi(ai). Two possibilities swbi(ai) a measure of the quality of the mental states produced by ai swbi(ai) a measure of i s preference satisfaction, i.e., uRi(ai) Equivalent income: ek(hi) = ek(ai, Ri) ek((mi, bi), Ri) = mequivs.t. (mequiv, bref) Ii(mi, bi) Monotone path: ek(ai, Ri) is a monotonically increasing numbering of the indifference curves of Ri