In defense of standard welfare measures

In defense of standard welfare measurement

Don Ross[1]

School of Economics, University of Cape Town

Center for Ethics and Values in the Sciences and Department of Philosophy, University of Alabama at Birmingham

dross@commerce.uct.ac.za

[Note: XXX in text refers to Ross, for facilitation of blind refereeing. XXX (2005), referred to in the text and suppressed in the references, is Ross, D. (2005). Economic Theory and Cognitive Science, Volume One: Microexplanation. Cambridge, MA: MIT Press.]

Abstract: This paper critically discusses Amartya Sen’s case for broadening the basket of well-being indicators in development policy beyond income and consumption expenditure. I first argue that, contrary to what Sen has suggested, the theoretical and practical motivations that he gives for this do not form a mutually complementary set. In the second, policy-focused, part of the paper I present problems Sen’s approach to measurement raises in the context of a case study from rural South Africa. I conclude by suggesting that development indicators should be regarded as narrower – in particular, that we should give privileged attention to women’s incomes – rather than broader.

Keywords: development policy; well-being proxies; welfare; Amartya Sen

Word count: 11,315
In defense of standard welfare measurement

Due in large part to arguments given over many years by Amartya Sen[2], increasing numbers of development theorists have embraced the idea that `traditional’ measures for welfare comparisons[3] (amongst individuals, and amongst variously distinguishable groups) are too narrow and should be replaced by indicators that reflect philosophically and anthropologically richer conceptions of human flourishing. The policy-relevant instrument that most directly incorporates Sen’s approach is the annual United Nations Human Development Report, which ranks countries’ populations on a scale that is intended to capture the relative `qualities of life’ available to (though not necessarily actually enjoyed by all or most[4]) representative members of these national populations. Recently, the World Bank has adopted the language of this broad conception of flourishing in commenting on development initiatives and policies, while economists throughout the developing world mine ever more complex panel data in search of broader social measurements of poverty and its causes, constituent elements, and circumstances of alleviation.[5]

This paper will criticize Sen’s campaign for the broadening of development measurement in two ways. First, I will try to bring some analytical clarity to the issues by distinguishing between two different bases for the campaign that Sen has presented. My aim here will be to show that these bases don’t have nearly as much to do with each other as Sen suggests and as many policy-implementers appear to suppose – that, indeed, they are so diverse in their underlying justifications and in their implications that they do not lend conciliating support to one another as a package. I will then describe a case study in current development planning, in a region of rural South Africa, in which objectives are sufficiently general that any or all of Sen’s justifications for broad indexing are potentially relevant. Nevertheless, I will argue, there are persuasive reasons, found in coordination problems that plague program design and implementation, for favoring restriction to a very narrow instrument of development measurement, namely, changes in women’s incomes.

Some readers may find the structure of this paper schizophrenic, in the sense of its seeming to roll two barely related problems arbitrarily into one text. The first part of the paper will be highly abstract, and the second part almost relentlessly focused on practicalities. In this respect, however, I just follow Sen’s lead. His semi-popular book[6], based on a seminar series he gave for World Bank officials, is plausibly the most influential development policy document of the past two decades. He there tries to combine all of his reasoned motivations for index broadening into a comprehensive package, and suggests a direct connection between the way in which the ends of development are implicitly conceptualized in economic theory, and inadequacies of large-scale anti-poverty interventions.

That the latter have often failed is not to be doubted. (Easterly[7] provides an elegant historical synopsis.) What I aim to question here is whether these difficulties have a significant philosophical component. My view, put in a nutshell, is that development theory is still on the upward-sloping part of a learning curve with respect both to modeling relevant incentive structures and implementing them in policy rollout, and that neither economists’ prevailing view of economic agency, nor deep philosophical controversies concerning the conceptualization of human flourishing, are important contributors to policy frustrations. I will argue that, at least in the one case I consider, the frustrations have a straightforward institutional explanation.

According to Development as Freedom, both the theoretical and the practical paths of argument converge on the claim that a person’s well being consists in her freedom as an agent, and a community’s well being supervenes on the extent to which its members are free to exercise agency. Sen is not clear as to whether the latter measure refers to spreading freedom’s reach across all members of a given community, or to maximizing average freedom in it, or to maximizing the freedom of a representative member. This issue would likely be the main focus of a traditional political philosopher setting out to criticize Sen’s position. It will not be the issue pursued here, however. While difficult problems of application arise for special zones of conflict amongst these normative benchmarks – especially concerning children and other people whose freedom as agents is inherently restricted by exogenous factors – I will leave these problems black-boxed. My justification for passing over standard philosophers’ `hard cases’ has two strands. First, it had better be possible to say something useful about goals and benchmarks to economists and other development agents in advance of settling deep conflicts along the frontiers where utilitarians, libertarians and Rawlsians seek transcendent or compromise principles in the face of one another’s’ counter-examples to proposed universal principles of evaluation. Second, and closely connected to this point, most people’s social and political ethics, where they are universalist at all, are simultaneously committed to the core elements of all the mainstream philosophical currents in political and social moral theory. A society that developed along absolutely consistent utilitarian or libertarian lines might be regarded by majorities of people as evolving unattractively; however, a society that consistently improved along core dimensions of both bodies of principles, however ersatz its procedures for sorting out instances of conflict between them, would likely strike most non-parochial observers as obviously making progress. I am thus in sympathy with Sen’s[8] refusal to choose amongst the standard over-arching moral frameworks, while acknowledging that his attempt to assimilate their core commitments is not philosophically rigorous.

Another aspect of Sen’s view I will not dispute is its emphasis on freedom as the most plausible single end of political activity – to whatever extent it is plausible to think that political activity could have any single over-arching end. The plausibility of freedom as such an end relies precisely on its flexibility as a conceptual container. Thus a conception of freedom that is mainly formal, or mainly negative, or demands too much willfulness or Nietzschean heroism, is not a plausible unifying focus of political norms. Sen’s favored conception has none of these limitations; the freedom of Sen’s well-off agent consists in her actual ability to play a substantial, and continuous, role in fashioning a life that socially `normal’ observers (including, by virtue of the meaning of `normal’, the agent herself in most cases) regard as worthwhile. Since Sen stresses that almost all people regard the embeddedness of their lives in social networks as being essential to living worthwhile lives, the freedom Sen exalts is not normatively atomistic. At the same time, he fully recognizes that most people do not think they can flourish without material comforts beyond biological necessities, and even that in rich communities, a person’s flourishing might require her being materially rich herself by historical standards.[9] Given this breadth of inclusiveness with respect to particular possible goods, who could object to Sen’s equation of freedom with flourishing?

The answer isn’t quite `nobody’. First, a fully committed subjectivist about value might argue that we have no basis for pronouncing any conception of the good, no matter how broad, as the `end’ of social-political activity from a standpoint independent of the dynamics of that very activity.[10] Now, while Sen’s frequent criticisms of traditional neoclassical subjectivism in economics, and almost equally frequent nods at Aristotle, indicate that he does mean to exalt freedom from such a standpoint, it isn’t obvious that, in the context of his work on development theory, much of importance hangs on this. Hardly anyone, no matter how sternly opposed they are to moral objectivism, would reject a set of development policies just because they didn’t take the preferences of people with anthropologically eccentric preferences – say, unusually selfish people – into serious account. This can be justified on grounds that don’t involve meta-ethical arguments at all. We need rely merely on some everyday anthropological observation: given Sen’s appropriately inclusive notion of freedom, even the unusually selfish would generally consider themselves well off – at least with respect to everything amenable by social or economic policy – if they were confident that their social, political and economic institutions were promoting their freedom in Sen’s sense. Again, this point highlights the way in which Sen buys the plausibility of his normative standard at the cost of making it worryingly flexible.

We can, however, note for completeness a more anthropologically significant source of objections to flourishing as freedom. Sen’s over-arching good is essentially secular, since it emphasizes the conclusion about values that most people come to if they take seriously the idea that their one life is all there will be for them. But the majority of human beings do not take this idea seriously. This certainly is relevant to development policies, especially when they are being rolled out in, for example, Afghanistan. For purely practical reasons alluded to above, development policy can ignore the views of the infinitesimal proportion of secular people who feel oppressed by (for instance) measures that discourage them from murdering promiscuous sisters and daughters. However, it is patently obvious that no such proposition applies to non-secular people who think themselves obligated or entitled to sew girls’ clitorises shut, massacre infidels, or deprive gay people of civil rights. There are plenty of them. The point here is not that there aren’t significant numbers of secular bullies. Rather, it is that, with Stalinism now discredited, there are few secular bullies who care at all about consistency and public argumentation but don’t acknowledge that their victims have at least prima facie claims against them based on appeals to the good of freedom in Sen’s sense; and policy-makers can justifiably ignore those who refuse to engage in public argumentation. Many of the non-secular, by contrast, argue from premises that do not lead to Sen’s conclusion.

This is yet another philosophical problem – though by no means a pragmatically idle one – which I will bypass in this discussion. One can reach the conclusion that development policy, at least as pursued by public agencies, should be based on non-parochial normative ends by one of two routes. There are familiar philosophical arguments for impartiality, which probably turn more real political gears among well-educated public servants than is often acknowledged. And there are complementary pragmatic arguments from the need to focus on elements of shared agreement amongst special parochial values. Neither of these considerations is always effective – witness interference with family planning programs in the developing world by recent Republican administrations in America, for example – but I think it can safely be said that non-negotiable differences over fundamental values are far from being the major obstacles to implementation of successful development policies, at least outside a few special parts of the world. One just does not run across these sorts of differences much amongst development theorists, agency officials, or (importantly) community spokespeople (again, outside of special parts of the world where religious identities massively influence economic aspirations).

There is, however, a sting in the tail that comes with endorsing Sen’s claims about value in this way. As noted above, a crucial factor allowing anthropological observation to support Sen is the fact that `freedom’, as he understands it, is a container so flexible as to hold everything that is likely to matter deeply to a thoughtful secular person, or at least to a thoughtful development administrator. Development policy is made difficult precisely because there is a plethora of different specific things we can try to do for poor people, and choices amongst these must be made. `Flourishing is freedom’ is something we can let Sen say because he defines `freedom’ so broadly as to make it more or less synonymous with `(secular) flourishing’. However, I take it that development problems have not repeatedly stumped us because we have been missing the appropriate tautology. Since the beginning of development policy, its practitioners and theorists have wanted to help people flourish. So now let us agree to say that we want to make them freer (both on average, because core utilitarian insights are insights, and by ensuring some universal minima, because core libertarian and Rawlsian insights are also insights). How shall we do so? And how shall we measure different extents to which people are free, or are becoming more or less free?

Sen’s arguments imply broad changes in prevailing development policy, as opposed just to adjustments in the way we philosophically integrate that policy with more general conceptions of the good, if and only if some presently prevailing approaches to development measurement systematically fail to promote core components of freedom, and in doing so over-emphasize goods that are inessential to freedom. Sen attempts to convince us of precisely this. Obviously, this requires appeal to premises and considerations going beyond those for the abstract claim that flourishing can be conceptualized as being synonymous with broad freedom.

These premises and considerations, as they feature in Sen’s arguments, can be divided into two broad classes. First, he argues that (mainstream) microeconomic theory forces us to model people in a fundamentally inaccurate way, such that the value people attach to freedom is not represented, or is perhaps not representable, in the logical vocabulary of that theory. This can be directly relevant to criticism of development policy only to the extent that policy is relatively directly informed by microeconomic theory, and systematically fails to introduce in application whatever it is about people that theoretical microeconomic models don’t incorporate. Second, Sen argues that traditional development policy leads to distorted objectives because it focuses exclusively on one member of the vector of influences on poverty, namely, poverty of monetary incomes and consumption in tradable goods and services. Argument for this proposition leads to Sen’s main conclusion only if income and consumption poverty is not a good proxy for `freedom poverty’ (or `capability poverty’, in his refined terminology), and if measures to alleviate income poverty fail to be causally effective against the other components of capability poverty. (A stronger claim, sometimes implied by Sen, is that they often fail to be causally effective against income poverty itself, because they often fail to be effective against other components of capability poverty.)

I will provide reasons for being skeptical about both of these generic arguments. On the first argument, I will not claim to do justice to the range of relevant complexities, both because space precludes my doing so and because I have provided detailed technical analysis elsewhere[11]. My purpose in reviewing this argument here is just to show that superficial affinities between the claim that microeconomic theory models people as utility maximizers and the idea that development practice models people as income maximizers are indeed superficial, and merely superficial. This is the specific thrust of my earlier remark that the separate pieces of Sen’s argument package do not constitute the mutually reinforcing set of considerations that Sen suggests they do.

III

It is widely supposed that mainstream (`neoclassical’) economic theory has been based on the assumption that every individual person is a rational maximizer of their self-interest. The content of this supposition is nearly empty unless both `rational maximizer’ and `self-interest’ are clearly interpreted; and the history of economic theory, even restricting consideration to unequivocally mainstream theory, offers several interpretations of both notions. Since the 1930s, `rational’ has meant merely `consistent in ranking states of the world with respect to their desirability’. To apply microeconomic theory to a person, we need not suppose that that person approximates one rational agent throughout her whole biography; she need merely be consistent in her preferences throughout the domain (in time and space) in which we want to apply a given exercise in economic modeling to her situation. Nor need a rational economic agent be omniscient or blessed with perfect mathematical competence (though it is sometimes useful to assume this when designing policy mechanisms one wants to be proof against subversion by the very clever or the very well resourced). Fortunately, Sen does not rest his arguments (though numerous others who share aspects of his critique have done so) on criticisms of the straw thesis that standard theory requires people to be prodigies of logic and self-mastery.[12] Thus we may bypass these issues here.

Issues surrounding interpretation of the `self-interest’ part of the standard supposition are more significant for present purposes. Sen is himself among the builders of formal revealed preference theory, and so speaks with unusually powerful authority on questions relating to its foundations. He has long argued that preference profiles, as economists have constructed these since the 1940s, are unreliable guides to people’s well being because preferences range only over `internally’ generated and anchored metrics of evaluation, whereas a morally and socially healthy person’s choices are strongly influenced by prevailing social norms that she may incrementally influence but does not choose[13]. It is by recourse to this thesis that Sen connects issues in microeconomic theory to questions about welfare measurement in his more recent writings on development. Briefly summarizing arguments I have given elsewhere[14], I will indicate why I find the thesis unpersuasive, and the connection therefore spurious.

Note first that Sen’s thesis does not concern the formal structure of standard utility analysis, which he of course understands at least as well as anyone. Rather, it concerns the way in which we should interpret that formal structure in empirical models. Following usage in recent philosophy of science, let us refer to the theory of interpretation of microeconomic theory’s formal elements as its model theory. (See Ruttkamp[15] for the clearest available interpretation of the relationship between mathematical models and empirical models.) Sen reads a strong level of continuity into the history of microeconomic model theory, supposing that the various formal theories that have succeeded one another through the history of neoclassicism are all intended to be isomorphic to a common model of the economic agent, which is in turn to be understood as typically coextensive with at least a temporal slice of a psychological person. Thus, Sen[16] presents the following sketch of the microeconomic foundations of welfare theory from early neoclassicism through to the present. Initially, agents coextensive with (at least) temporal slices of psychological people were taken by Jevons, Walras and Fisher, following Bentham, to be maximizers of hedonic pleasure, and it was supposed that psychologists would eventually provide a scale, based on empirical research, for quantitatively comparing any two hedonic states, including hedonic states of different people. However, in the 1930s economists under the influence of positivism came to doubt that hedonic states were, or could be, scientifically observable. They thus operationalized utility by reference to actually observed episodes of choice amongst alternative baskets of goods, thereby making the impossibility of interpersonal comparisons of utility a self-fulfilling prophecy, since (1) few if any people confront identical choice sets, and (2) the new framework allowed merely ordinal rankings of utility judgments, which radically underdetermine comparative evaluations of people’s absolute levels of utility.

Sen frequently implies that if we can’t perform interpersonal welfare comparisons it follows that we can’t do welfare economics at all (or can do so only by sweeping interpersonal differences under the carpet[17]). It should thus be noted that this was not the view of the 1930s economists who attacked the validity of interpersonal utility comparisons. In the most influential of the many manifestos on the subject, Robbins[18] argued that because no available data could favor treating one person’s preference satisfaction as more important to them than someone else’s, all utility profiles should be weighted equally in making policy, on grounds that in conditions of ignorance symmetry assumptions are the only safeguard against the influence of special prejudice. As to how the preference rankings themselves were to be discovered, Robbins and his contemporaries supposed that we could ask people about them. While this method raises issues around mechanism design for avoiding strategic preference revelation, we have since Robbins’s time learned a good deal about techniques for dealing with such issues.[19] Note that weighing all utility functions equally in determining ultimate policy goals does not imply that any particular means to utility maximization, such as income, should be weighted equally across persons for policy purposes. As I will argue below, for example, there is good reason to think that in many impoverished communities, overall welfare is best promoted by paying disproportionate attention to women’s income and ignoring men’s.

This is a side issue for now, however, because in my view Sen’s capsule history of microeconomic utility theory involves more serious distortions. The 1930s methodological campaign against interpersonal comparisons had a specific part of the then prevailing microeconomic formalism as its main target: it aimed at showing that the representational device of indifference curves could be defended independently of Jevons’s empirical psychological claim that utility from all commodity streams diminishes on the margin for each individual. A standard 1930s argument was that failure of diminishing marginal substitutability implies the rationality of monomaniac preferences, and this was taken to be a valid reductio, based not on any psychological hypothesis but on what would happen to an agent with monomaniac preferences in a typical market. This argument was later shown to be invalid. However, it marked an important shift in the interpretation of what utility theory models were supposed to be about: not the internal psychology of people, but the descriptive dynamics of behavior amongst agents competing for scarce resources. This shift was completed with Samuelson’s Foundations[20], in which utility functions, as formal representational devices, explicitly cease to have anything at all to do with psychological profiles of individuals, and simply become inputs to the computation of demand in allocation machines involving distributed control. One might fault Samuelson and his immediate successors (especially general equilibrium theorists) for devoting insufficient attention to the question of which (if any) empirical systems their machinery might describe. What one cannot accuse them of is failure to promote radical revision of the traditional microeconomic model theory. Samuelson, indeed, urged retirement of the word `utility’ altogether, on grounds that the model theory in which that concept originally emerged had been rendered obsolete. Simple semantic inertia, of little substantive significance, defeated him in this ambition.

One would gather little or nothing of this from Sen’s critical discussions of utility in his writings on development. We are repeatedly told by him that congruent demand functions and identical consumption profiles are consistent with different `levels’ of utility for different people[21]. However the very concept of a `level’ of utility makes no sense in the revealed-preference interpretation of the formalism, since in that interpretation `utility’ does not denote a quantity in the first place. In Sen’s more dedicated criticisms of utility theory[22], the key arguments are based on the idea that a fully rational person bases her choice behavior on all sorts of socially governed contextual factors – e.g., is the apple in the basket being passed around the table, which I’d otherwise choose over an orange, the last apple available to all? – to which it is assumed that utility functions must be insensitive. This is the basis for Sen’s famous contention that a person whose choice behavior is accurately modeled by a function from choices to commodity rankings that satisfies the weak axioms of revealed preference would, in social context, be a `rational fool’.

What justifies the assumption that utility functions can’t register concern for social norms? In the formalism of post-Samuelson utility theory, there is no restriction at all over the kinds of objects that can appear in the scope of a utility function; `last-apple-in-the communal-basket’ is just as well qualified an object of preference as `apple-in-a-bunch-of-apples-and-oranges’. It is true that if we insist that the model theory for the formalism must be that of Jevons or Fisher, then use of the revealed-preference interpretation of utility will force us to choose between (1) making application of utility theory to individual psychologies empirically vacuous and (2) restricting the scopes of utility functions to objects of preference that preserve rankings invariant under mere shifts of context.[23] This is clearly a central premise in Sen’s argument. But the model theory of postwar microeconomics is not that of Jevons or Fisher (or of the transition figures of the 1930s, Robbins, Hicks and Kaldor). There are interesting questions to be asked about just what the new (implicit) model theory is; XXX[24] tries to provide answers to these, and I will summarize them shortly. The crucial point first is that Sen’s interpretation is poorly supported by the intellectual history.

Sen often[25] raises tautology objections to interpretations of neoclassical model theory that emphasize the openness of formal RPT. It is indeed the case that if one is motivated to treat some unit as an agent in a model of the formalism, then it follows axiomatically that a consistent utility function will characterize that unit’s behavior. But formalisms are supposed to define tautologies; that’s what’s formal about them. Relative to the formal theory of an economic agent, it is an empirical question to what extent the behavior (not the subjective psychology) of people (or firms, or households, or trade unions or fish) approximates economic agency; we do not assume that people are paradigm instantiations of economic agency and then tailor the concept to that assumption. In science, the need for consistent formal regimentation requires that we not be prisoners of our folk ontology. This principle applies in economics just as elsewhere.

I have been arguing that neoclassical theory can’t force us to misrepresent the facts about real people if it tells us nothing definite about how to represent real people in its terms to begin with. But now the following objection can readily be raised in the context of policy: of course welfare policy is directly about people; it isn’t disinterested scientific inquiry. Doesn’t this concede Sen’s whole point – that people, in the sense in which we care about them when we worry about their practical welfare, aren’t the utility-maximizing entities we meet in theoretical models of whole markets? In one sense, yes; but it does so in a way that gives him too much to make his critique of utility theory relevant to issues in development theory. Worse, as I will go on to point out, it exposes a problematic sense in which Sen’s model of people is actually highly conservative.

Let us start with the first point. Sen’s case against what he calls `welfarism’ rests on the idea that economists are committed to the idea that `utility’ denotes all that matters to the welfare of a person, and that standard development theory then takes either income or consumption to be the most reliable proxy measures of utility. But, I have argued above, postwar economic theory (by contrast with prewar theory) abandons the first idea by dropping the requirement that bearers of utility functions must be coextensive with psychological people. Economists applying microeconomic theory to practical development issues seldom construct formal utility functions as parts of their models at all. This fact is not unrelated to Samuelson’s reasons for thinking that the term `utility’ should have been retired from use by economists. Essentially, utility functions are now mainly technical constructs lurking in the deep theoretical background that play a role in recursively fixing the operational meaning of other concepts, such as `demand’ and `opportunity cost’, that are directly interpreted with respect to empirical variables – but aggregate ones that needn’t decompose with respect to individual preference profiles in any particular way. While a specific concept of utility is thus relevant to what is meant by demand in economic models, this utility concept plays no role in normatively justifying the significance given to demand in economics. That is, we don’t regard measurement of aggregate demand schedules as important because these schedules compose individual demand functions that are in turn proxies for hypothesized individual utility functions. Rather, the normative justification for caring about aggregate demand is pragmatic: we obviously couldn’t model an economic process without an estimate of what will get consumed under various possible production and price regimes. Most working economists would regard it as both naïve and pointless to literally hypothesize a person’s utility function, in circumstances where the testable value of a prediction hinged on the accuracy of this hypothesis. Utility functions are often explicitly constructed in possibility proofs of general relations, but this is exactly the sort of use where assumed coextensivity of economic agents and natural persons does not arise.

The point I am making here admits of easy misunderstanding, so at the risk of pedantry let me stress it very explicitly. A political philosopher has every reason to wonder why a focus on aggregate demand must be normatively appropriate if no one typically has any firm idea how it relates to the well being of any particular individuals. I am not arguing that under-theorizing of this relationship is of no interest in general. Rather, I am questioning Sen’s contention that the normative emphasis on demand in policy economics is inherited from a supposed normative emphasis on formal utility in microeconomic theory. The overwhelming majority of development economists are normatively motivated to study poor countries because most of the people who live in them are self-evidently materially poor. All policy-relevant measures of economies require estimates of aggregate demand. And that is about as far as normative justification typically goes. Thus I contend that the formal concept of utility maximization has little or nothing to do with controversial issues of measurement relevant to development policy. Therefore Sen’s policy-based case against the use of narrow proxies for well being in development theory is not lent any additional plausibility by his criticisms of standard microeconomic foundations.

On the second point introduced above, it is slightly ironic that Sen’s case against standard utility theory is widely thought to gain plausibility through its emphasis on the importance of social context, and through its implied attack on `standard’ microeconomics as being unduly atomistic. That early neoclassical thinking was strongly individualistic, both ontologically and normatively, is not to be doubted. However, as both Davis[26] and XXX[27] argue at length, the model theory by which contemporary microeconomic theory is interpreted has become so ontologically agnostic that Davis feels the need to devote a volume to trying to rescue any systematic role for the human individual within the framework. As is demonstrated in XXX[28], the recent widespread turn to evolutionary dynamics for predicting and explaining preference structures and microequilibrium stability patterns has convinced many economic theorists that individual rationality is a residual folk myth that has no proper role to play in either theoretical generalizations or empirical applications.[29] Interestingly, when Davis sets out to construct an interpretation of economic theory that preserves a central role for the individual, it is to Sen that he turns for help. Why is this? The answer is that, as explained in the preceding paragraphs, Sen continues to interpret the model theory of microeconomics according to the early neoclassical conception, in which utility levels are well defined properties of individuals. This is why Sen is able to coherently contrast the utility an individual might derive from an apple with the wider sense of well being that same individual might derive from observing social norms and leaving the last apple in the basket for others. The traditional neoclassical individual is still very much in Sen’s picture, but her behavior is understood by him as produced by a vector composed from traditional, internal motivations and external, social pressures.

As argued at length in XXX[30], the contrast on which this vector composition is based finds little or no support from the contemporary behavioral sciences.[31] Nor does most current microeconomics either presuppose it or derive predictions from it. The formalism merely requires units that stabilize their control dynamics long enough for their behavior to be represented by demand functions. In the model theory, it is common to represent all pressures on the relevant units as external, though some idiosyncratic informational access will typically be explained by reference to internal properties and processes. In this sort of picture, economic agents are not individual atoms that are then dropped into social contexts, or bundles of internally instantiated preferences with social wrap-arounds. They are just (relatively) stable sites of informational integration that function as nodes in the social-evolutionary dynamics of larger systems governed by distributed control. In this framework, Sen’s distinction between self-centered preferences and socially governed behavioral norms and dispositions collapses. For the purposes of the present argument, it is not necessary to maintain that the new picture is necessarily the most ultimately satisfactory one. It suffices to point out, first, that it is a consistent model of the standard formalism, and that we can demonstrably do standard microeconomic analysis with it[32]; thus it cannot be true that the formal theory requires Sen’s interpretation. Second, Sen’s conservative strategy should not be regarded as the `natural’ or inevitable way to factor social influences into economic models. We can just begin by taking social influences as ontologically fundamental to economic agency – as, for instance, in the currently very popular models in which agents acquire their utility functions and their strategy sets by more or less unintelligent copying of those of others, with change occurring through interactive effects controlled or planned by no individuals – and thus face no problem of `factoring them in’ in the first place.

None of the points I have made in this section tell for or against Sen’s contention that traditional welfare proxies are inadequately narrow. That, indeed, is the point. The formalism of microeconomics is a highly flexible mathematical and logical instrument – just as it should be – which imposes few restrictions on what sorts of empirical states of affairs can be represented in its terms. Utility functions are formal objects, and their prevailing interpretation does not imply early neoclassical atomism. Thus, a reader of Sen who supposed that his long-fought critique of utility theory provides theoretical underpinning for his empirical and normative criticism of standard welfare measurement would be mistaken; the two controversies have little of substance to do with one another.

The reader may now be a bit impatient with me for having asked her to read so much for the sake of a wholly negative conclusion. Let me note, therefore, just how far some economists are prepared to go in the direction that Sen encourages. I will shortly be discussing policy issues around poverty alleviation in South Africa as a case study. Geeta Kingdon and John Knight[33] have recently addressed this topic, and have argued that we should aim to quantitatively measure subjective `happiness,’ which they maintain comes significantly apart from income as a measure of South Africans’ well-being, and direct our policy at alleviating that. This idea commits them to criticism of Sen, who is of course opposed to this sort of subjectivism about development values. However, from the perspective of the present paper this argument is a family quarrel among those who identify the target of economic policy with the atomistic early neoclassical person. The Kingdon and Knight approach, I allege, is deeply retrograde from the perspective of contemporary social and behavioral science. That science indeed has new and interesting things to say about the dynamics of subjective perceptions of satisfaction[34], including in national development policy contexts[35]. However, it complements other parts of behavioral science[36] in showing us how subjective personhood is a dynamic and feedback-mediated product of social and macroeconomic conditions, not the fixed ontological bedrock of Jevons’s day. I thus join Sen in thinking that if the goal of reducing poverty in poor countries is allowed to morph into the goal of maximizing subjective happiness in these countries, the basic political-economic objective of development will have been profoundly hijacked. I do not doubt that very poor people can get much pleasure out of life, and I am glad that surveys find people in, e.g., Nigeria, to be unusually happy. However, it is surely outrageous to conclude on this basis that we should stop working to make Nigerians less poor.

Sen would of course vigorously agree with me on this last point. But my contention is that, by encouraging return to an essentially Victorian interpretation of the objects of microeconomic theory, and then linking that interpretation directly to development policy questions, he invites authors like Kingdon and Knight to carry us all the way over the edge into the most conservative conclusion imaginable: things in Africa are actually alright.[37] That Kingdon and Knight do not think of themselves as conservatives any more than Sen does is beside the point.

Only an exceptionally thoughtless person might seriously imagine that a typical individual’s well being was entirely a function of the quantity and quality of her consumption of tradable goods. Aristotle argued long ago that a person could not be truly happy if he was ugly. Although that is now a complaint for which money buys exactly the right fix, there are no doubt more subtle personal imperfections that continue to vex the rich. Some wealthy people die young, as Sen points out, and they probably don’t think, in their last moments, that their generous bequests are satisfactory compensation. An anthropologist’s descriptive account of the sorts of circumstances in which people view themselves as well off must, to be adequate, explore complex and subtly textured psychological and social terrain. Sen reminds us that Adam Smith provided a splendid example of this in his Theory of Moral Sentiments, written before the word `anthropologist’ came into existence.

As Srinivasan[38] points out, development economists have always recognized these things to be truisms, just as clearly and consistently as have people in general. It is also a truism that the more complex something is, the harder and more expensive it is to accurately measure. Thus when we want to design policies that increase the supply of something thought to be desirable, whether that something is fulfilling human lives or scientific knowledge, we have to find reliable proxies in order to evaluate our progress. Sen has never denied this. As he says[39] `Some capabilities are harder to measure than others, and attempts at putting them on a `metric’ may sometimes hide more than they reveal. Quite often income levels – with corrections for price differences and variations of individual or group circumstances – can be a very useful way of getting started in practical appraisal. The need for pragmatism is quite strong in using the motivation underlying the capability perspective for the use of available data for practical evaluation and policy analysis.’

I agreed in the first section of this essay that if we find it helpful to label the end of all praiseworthy political action with one familiar word, `freedom’ is a relatively unobjectionable choice (though, as Binmore[40] forcefully reminds us, this choice mustn’t be allowed to obscure the fact that if we aim directly for freedom and forget that stability and predictability are necessary means to it, political action will likely get us less freedom rather than more). Let us further agree that explicit attention to the meaning of freedom can generate specific and substantial policy recommendations that we might otherwise miss. The work of Philippe van Parijs[41] is the outstanding example here. However, the thesis that increased freedom is an over-arching political and social goal is a highly general and abstract one. It is thus unsurprising that policy recommendations of the sort to which it leads, such as van Parijs’s, are also highly general, concerning, in this instance, the re-design of the whole political economy of the state. As Sen fully recognizes, if an emphasis on freedom – that is, in his terms, on the expansion of human capabilities within state or sub-state units – is to have practical implications for development policy, this must in large part consist in its leading us to identify improved proxy indicators for development that we otherwise might not consider.

Sen’s leading arguments on this subject, however, do not identify proxies that are claimed to be superior to traditional income-related or consumption-related ones; rather, they suggest that we need a wider array of proxies, amongst which we should flexibly choose as circumstances vary, because there are aspects of well being – for example, longevity – that are not reliably correlated with variations in income adjusted for purchasing power. As Srinivasan[42] forcefully points out, this is not mainly a conceptual issue – since most people would agree that longevity, at least up to a point, is an aspect of human well being regardless of whether they agree that emphasizing the concept of freedom illuminates this intuition – but an empirical one. Now, if we thought we could directly and independently measure freedom, then the conceptual claim would imply a method of empirical investigation: we would seek to find out which sets of proxies tend to yield the best ratios of set size to explained quantitative variance in freedom. But, of course, there is no direct, independent quantitative measure of freedom. So the situation seems to be as follows. Sen and others draw our attention to various things many people consider important aspects of well being, and then point out that they are unreliably correlated with income. We are thereby supposed to be persuaded to measure all of these things. Then, in wondering how to relatively weight them for purposes of policy design and evaluation, we are advised to reflect philosophically on their contribution to capability enhancement – that is, to freedom – and to be flexible, pragmatic and sensitive to special circumstances.

This might not look like a procedure that could ever be rendered into an algorithm – but isn’t that precisely the intended point? Doesn’t it mainly amount to saying `When considering something as complex and multi-faceted as real human flourishing (which, again, has little to do with the purpose served in constructing the formal concept of utility), don’t aim to be robots?' Surely that must be good advice? Consider a homely analogy. Suppose people wanted to do better at monitoring the health of their marriages. Then suppose that some empirical psychologists identified three or four directly measurable proxies that most people judged to be independently and approximately equally well correlated with their intuitive conception of healthy marriage, but which correlated poorly with one another in identifiable ranges of their values for each pairwise comparison. Wouldn’t it obviously be a foolish idea for a person to just pick the statistically best predictor from among these proxies and then focus monomaniacally and robotically on it in governing their domestic behavior? Doesn’t every thoughtful person recognize that this policy, in each typical case, would be more likely to lead to the divorce court than to the golden anniversary column of the newspaper? Marital relationships are complex, dynamic and context-sensitive processes. Wise people who care a lot about theirs should try to be better philosophical psychologists, not search for the best measurement algorithms.

I want now to discuss an important reason for regarding this analogy as misleading. Before doing so, let me stress that I think the analogy does capture the most initially persuasive interpretation of Sen’s program for development policy planning and evaluation. Sen[43] makes clear that he does not think that he has identified a statistically superior metric for evaluating levels of economic development. The heterogeneity of capabilities, he argues, makes the `technocratic’ construction of any such metric an exercise that must always fall short of what deliberative democratic conversation, aided by philosophical reflection and (of course) rigorous empirical surveying, can accomplish. This is apt to sound overwhelmingly sensible to most people for, I suggest, exactly the same reasons that non-robotic marriage management really is sensible. The analogy doesn’t break down for conceptual reasons; it breaks down because of differences in the kinds of policy tools that are available to us.

In development policy, to a much greater extent than in personal life – and to a greater extent even than in many other areas of public policy, such as design of tax regimes – lies a profound gulf between the planning of schemes and their implementation. Yet another gulf, almost as large, opens between implementation and measurement of consequences. We know of no generally effective way of closing these gulfs. I will argue, by reference to a specific case, that their existence and their interaction with one another leads to a powerful argument in favor of the use of narrow and rigid proxies. In their light, I will maintain, failing to measure some conceptually persuasive aspects of well being is less problematic than are the bad consequences that stem from leaving ourselves free to sensitively adjust our benchmarks.

Before turning to the case study, I will tighten the topic focus slightly relative to Sen’s. One of his oft-cited pieces of evidence for the claim that obvious aspects of well being, in this case longevity, fail to correlate reliably with income, is that African-Americans have substantially lower average expected lifespans than are found in some developing-world populations in which income per head (adjusted for purchasing power) is much smaller than theirs. This fact certainly tells us something about an important social problem in the United States. It would be semantic quibbling to deny that it also tells us something about `development economics’ in the broadest interpretation of that enterprise. It is not obvious, however, that it tells us anything about development economics in the narrower, more clearly institutionalized, sense, that is, about how we should try to bring more countries up to the point where they are ready for OECD membership and would no longer be regarded as appropriate sites for foreign aid or special and differential treatment in international trade law. Of course, there are grey areas in the loose distinction I am drawing between development economics in the broad and narrow senses. If the western half of Alabama were a separate country, it would qualify for SDT in the WTO and would be nowhere near OECD status. The point remains that policy measures appropriate for dealing with inequality and complex racial and ethnic segregation in relatively wealthy countries are importantly different from the sorts of initiatives that governments and agencies try to implement in and for middle-income and least developed countries. Furthermore, if we don’t draw an important line by reference to countries, then `development economics’ risks semantically incorporating all of policy economics, since one can always keep subdividing a population until one or more subdivisions includes only poor people. For these reasons, the scope of my argument here is intended only to apply to the policy domain of mass poverty alleviation in non-rich countries.

The argument is based largely on my experience in managing infrastructure development projects in very poor rural districts of South Africa. Though South Africa sports first-world infrastructure in the cores of its large cities, the majority of its citizens have welfare far below OECD levels (on any plausible means of measurement), and substantial numbers of its rural citizens live in conditions similar to those prevailing in sub-Saharan Africa generally. Over the past few years I have been part of a team of economists and civil engineers that has been developing coordinated infrastructure projects for bringing roads, running water (for human and livestock consumption and sanitation, and for irrigation), and physical facilities for basic business establishment, to communities in which 60%-90% of the population is formally unemployed. Each area in which we have been working (uMkhanyakude District in the Province of KwaZulu-Natal and O.R.Tambo District in the Eastern Cape Province) has a population of between one and two millions. The primary sources of cash income for absolute majorities of adults in these areas are small state pensions provided to elderly citizens. These are supplemented in many cases by sales of handicrafts to small numbers of nature tourists who travel in the areas. Supply of these handicrafts relative to demand is such that craft producers are pure price takers, and investment of time in craft production is profitable at all only because opportunity costs of time (measured against other available sources of income) for producers are negligible. The principal source of food for most people is subsistence agriculture, which also generates small surpluses for roadside sale.

Based on a variety of political, economic and humanitarian motivations, there is an effective national consensus that alleviation of this poverty requires provision of substantial new basic infrastructure. Such infrastructure assets directly improve the welfare of those who presently lack access to them, and are prerequisites to the development of local employment-generating economic activity. It would certainly be correct to say that, on a very general level, infrastructure provision in the South African hinterland is intended to enhance the capabilities of the rural poor.

Public resources for provision are limited, of course. More to the point, capital stocks are insufficient to provide, in the short to medium term, all infrastructure assets that could, if they were already in place, be maintained at a cost that government would regard as less than the value of the social benefits. Thus we face a capital rationing problem, and decisions must be made concerning project priorities. South Africa’s current democratic ethos strongly inclines government, at least in principle, to try to elicit the views of the immediate local beneficiaries on these priorities, and to be guided by them as directly as possible. The broad approach to implementing such subsidiarity has been as follows. All personal residences in South Africa fall within the jurisdiction of one or another municipal council. Each municipal council falling outside of designated urban areas is entitled to grants for infrastructure provision according to a formula that correlates relative expenditure levels in regions with relative numbers of current direct welfare transfers. (The motivation for this is a mix of complementary humanitarian and fiscal considerations.) However, to receive all or part of its grant allocation, a municipality must produce and submit an `Infrastructure Development Plan’ (IDP) that prioritizes possible projects, estimates their initial capital costs and subsequent public costs for maintenance, and justifies their rankings. Implementation of projects, according to these municipally identified priorities, is then to be carried out by provincial governments, using resources coming primarily from taxes and charges leveled nationally and turned over to the provinces by the central government.[44]

The project I direct was established (from private industrial funding) mainly as a result of general recognition that the process described above, which authorities attempted to implement during 1999 – 2002, has completely failed to achieve its objectives in poor rural areas[45]. In most cases, municipalities lacked the planning and budgeting capacities necessary to develop their own IDPs. They were therefore given grants with which to hire consultants to assist them. However, consultants lacked quality-assurance incentives, because municipal officials were insufficiently experienced to be able to distinguish adequate from inadequate work. As a result, most IDPs as submitted by the national deadline in late 2002 did not prioritize project possibilities, or justify them, or even estimate their costs. Both provincial and municipal officials are under multifarious pressures, of the standard political kind, to see funds spread evenly amongst constituents. Furthermore, provincial officials often regard their mandates as substantially fulfilled if they succeed in exhausting their budgets, since difficulties in rolling out allocated funds for public goods have been a recurrent subject of media criticism during the first decade of South African democracy. These two influences have combined to produce a scenario in which, for example, road budgets are wholly consumed through addition of a few kilometers of new surface to each existing rural road, and in which road, water and other infrastructure projects are not integrated so as to capture complementarities or exploit multipliers. This has typically implied opportunity costs on the invested value of initial capital in the range of 80-90 % (my estimates, as made in recent reports to the Province of KwaZulu-Natal).

For these reasons, the IDP process has thus far had almost no impact on project selection or implementation. We are therefore now engaged in writing new IDPs for the municipalities in our pilot study areas, in which local preferences are gathered as inputs to the identification of possible projects, but in which economic principles of capital-value-maximization are the ultimate basis of prioritization. The rationale for this is that we are otherwise failing to buy levels of marginal community welfare for each unit of invested public funds, measured in terms of public assets available for use by communities, favorable enough to be fiscally justified.

The experience just described exemplifies the extent to which development planning typically involves, as its most difficult aspect, the simultaneous solution of various coordination problems. Incentives of, in this case, municipal bureaucrats, traditional leaders (tribal chiefs), provincial implementation officials, provincial planners, provincial politicians, national politicians, paid private consultants and everyday citizens, are at sufficient variance to produce largely unintended outcomes when they interact. This is despite the fact that almost all would agree that their efforts are for the sake of enhancing community welfare, and despite the fact that extensive and democratic consultation with local citizens has marked every stage of the process. With the exception of a few unusually cynical or self-serving bureaucrats and politicians, and private consultants interested overwhelmingly in minimizing their own labor expenditure per fees earned, almost every party involved can truthfully claim to have promoted a project agenda that would enhance some vector of capabilities among community residents. That fact is a major part of the problem, or so I will argue.

Sen maintains that development planning should be conceived in terms that are not narrowly technocratic, but are instead democratic, deliberative, and transparent to average citizens. One need not appeal to normative arguments to establish this conclusion. Though policy blueprints can be designed as technical exercises in World Bank offices or national ministries, implementation of these blueprints, even in relatively authoritarian states, inevitably flows through the activities of wide arrays of agents. If a policy’s only justification is a complex technical one, it is likely to fail because too many responsibly agents cannot understand it well enough to adapt its implementation to contingencies. So this part of Sen’s policy critique is correct. But do the multiple sites of agency in development practice imply the other, more distinctive, side of his recommended approach? I have told the reader what my answer is going to be here.

The current governing party in South Africa enjoys hegemonic administrative power and control over the social spending agenda, at least by comparison with political authorities in OECD countries. Nevertheless, accountability for rollout of aspects of programs is highly distributed, simply because much relevant information flows horizontally, and in both vertical directions. In this circumstance, the heterogeneity of success indicators produces the consequence that most responsible officials can sincerely find some set of capabilities that their activities have enhanced. However, these enhancements frequently work at cross-purposes to one another. Thus, for example, a tribal council might acquire increased control over the use of irrigation water from a local dam, but by the very same process reduce the capability of the municipal council to coordinate the transport and hydro networks. (This happened in my experience, in a Zulu village in 2001.) Thus capability to choose higher-value crops reduces capability to extract this higher value from sales, because the resulting coordination failure on transport shifts control of the distribution network to agents outside the region. Frustrated by these continuously recurring snarls, we look for some level of responsibility at which a fix should be implemented. This, however, requires that we locate not just failure itself, but somebody’s failure. Heterogeneity of intended outcomes, however, blocks these attempts far more often than not – indeed, almost always. In the case of the IDP process described above, my team of academics, working with direct agents of the governing party, and for cost rather than profit, have recently gained increased planning influence because the private consultants could be shown to have operated from goals that diverged from the broad consensus. This does not mean that we will now succeed where they failed; I am much more hopeful than optimistic. Note that this is so despite our being in the unusual circumstance that no one regards the IDP process as having succeeded, because the process has failed to produce any new infrastructure, and that was clearly what it was supposed to be mainly about. Nevertheless, it took four years to arrive at this negative consensus. In more typical cases, where programs enhance some capabilities but complementary capabilities are retarded by the very same programs, it is usually impossible to coordinate responsible officials on broad policy shifts, because majorities of them can claim at any given time to be broadly fulfilling their mandates.

In this sort of environment – which is always the environment except where policy scales are very small – proposed policy changes gain wide adherence only if and where agents cannot cherry-pick among success criteria. Yet cherry-pick they can, and do. In uMkhanyakude and OR Tambo, longevity rates cannot possibly be increased against historical norms by any immediate policy, due to the exogenous impact of the AIDS epidemic. At the same time, general health among those without AIDS has improved, because the campaign against HIV has brought additional diagnostic and clinical resources into the areas. On another measure favored by the Human Development Report, the IDP process certainly increased most people’s capabilities to influence identification of possible development projects, and to this extent politically empowered them. Yet there is a clear sense in which we can, and should, say that poverty has not been significantly addressed. Income per head among the people in each of the two areas has not increased. I do not expect that it will increase – or, at least, that public projects directed at welfare enhancement will succeed in raising it, as opposed to something exogenous to policy such as an increase in the world price of sugar – until and unless there is broad agreement that limited, local and partial improvements in other capabilities do not constitute entitlements to complacency by responsible agents. I thus view with unease the fact that Development As Freedom seems at the moment to be the most widely read book about development policy in South Africa. In its terms, one regularly encounters justification of a great deal of busy promotion of this or that capability of this or that group of poor people. Yet if real poverty is being alleviated at all just now, it is almost solely thanks to – of all things – supply-side trickle-down from the currently thriving urban-based national economy.[46] Relying on that process does not constitute successful anti-poverty policy; it constitutes failure to find or implement such a policy.

The problem of anti-poverty target coordination can be analyzed into two components, which reinforce one another. First, there is the absence of a clearly prioritized benchmark to begin with, which I have been emphasizing to this point. Equally important is the fact that few elements among the overly broad set of benchmarks are conceptually simple. `Autonomy enhancement’ is no doubt a good thing in most sincere officials’ hazy representation of it. As Sen would stress, nothing precludes our working to operationalize it in a sensible way so it could be quantitatively measured. However, even in this best-case circumstance, it inevitably would usually not be quantitatively measured in most situations where we were wondering about policy success along micro-dimensions. Furthermore, the fact that some technocrats agree on a stipulative measure for a variable cannot prevent other people from interpreting the concept in idiosyncratic ways; it is hardly in keeping with Sen’s preferences that technocrats overwhelm all other participants in policy discussions. For these reasons, `autonomy enhancement’ has a tendency in South African discussions to mean whatever best allows an advocate of a policy measure, or an official whose position incentivizes them to defend the measure in question, to argue for its effectiveness. Again, the point here is not philosophical; `autonomy enhancement’ might just be the best name for a very important aspect of well-being. The point is that to avoid systematic coordination failure in anti-poverty policy, we require highly transparent concepts, the meaning of which is not readily contested, to serve as benchmarks. Income is one such concept. There is of course some room for quibbling over what a person’s income is – but only some, and people tend to leave it to economists.

This point leads to further considerations in favor of narrowing, rather than broadening, the set of indices by which we measure welfare, at least in South Africa. Several years of conservative fiscal policy by the national government, which have reduced its debt and non-core obligations, along with greatly improved efficiency in gathering tax revenues, have enabled the number of direct welfare grants to be meaningfully increased.[47] Unfortunately, we see little visible evidence in uMkhanyakude and O.R. Tambo that these increased monetary transfers are translating into increased investment or savings, or into investment in human capital through increased consumption by children. I know of no evidence beyond the anecdotal that diagnoses the micro-dynamics of this situation within households.[48] However, the definitive work of Bhorat et al[49] establish that labor markets are the overwhelmingly important mechanisms governing poverty incidence; and formal unemployment rates in rural South Africa have been very stubborn. (There is unfortunately no panel data available on changes in volumes of remittances from urban-earned incomes to rural households, so we do not know much about the impact of national business-cycle movements on the rural economy.) But data from rural South Africa support more systematic evidence accumulated in other developing countries[50] suggesting that increased welfare of current female populations is a significantly more reliable indicator of increased expected welfare levels in the whole populations of which the women are members than are current increases in the welfare of the whole populations. The most obviously plausible explanation for this is that women whose incomes increase are much more likely to defer consumption and to invest higher proportions of resources under their control in social infrastructure improvement generally, and especially in the human capital of their children. Sen has, to his great credit, promoted and emphasized this point vigorously and effectively.[51] However, the same sorts of coordination problems arise with respect to policies aimed at improving women’s welfare as confront development policy generally. Most development initiatives in uMkhanyakude and OR Tambo that succeed in accomplishing anything improve the capabilities of some women along some dimension or other. Yet evidence for systematic improvement is scarce to nonexistent. A major complication for efforts to promote policy improvements through attention to this fact is heterogeneity of accepted benchmarks for progress, and the difficulty these raise for sourcing accountability amongst responsible agents. I believe that we would stand substantially better odds in this area if it could be agreed that success must be measured in terms of women’s income, in the pragmatically mixed utilitarian and Rawlsian evaluative framework endorsed in the first section of this essay. By this I mean that we should seek affirmation from political authorities for the proposition that if women’s incomes are not rising, both on average and with special respect to the poorest women, then current policy should not be regarded as adequate.

This conclusion is not based on my claiming to have new empirical evidence that income is the single best information proxy for welfare. Indeed, there is highly persuasive evidence furnished by Ravallion[52] that consumption expenditure measures are predictively superior because they are more stable (due to the fact that agents are better able to smooth out CE than income). My claim instead is that what we should want for the sake of policy coordination is a measure of something that (i) could not likely increase unless overall welfare were increasing, and (ii) is likely to be broadly causally efficacious against other components of poverty. Positive changes in women’s consumption expenditures would likely satisfy desideratum (i) even better than increases in women’s income; if women defer consumption expenditures, then seeing them rise should make us more confident that general welfare improvements have occurred. However, this ignores desideratum (ii).

I have been arguing that we should seek proxies that are difficult to multifariously represent because solving the severe coordination problems confronting policy implementation behooves us to pursue one agreed benchmark that is likely to be similarly conceptualized by most agents. Income and consumption expenditure plausibly meet this criterion equally well. Now, if our main basis for preferring a proxy is that it disciplines policy coordination, rather than just because it maximizes predictive leverage, then there is a good case for concentrating on women’s income. Agents evaluating policies have varying degrees of patience. The longer the lag between a policy intervention and its intended effect, the more agents will tend to drift away from coordination as their different patience thresholds are reached. Desideratum (i) above is stressed by concern with predictive accuracy and desideratum (ii) is stressed by concern with policy efficacy. My claim that coordination breakdown is the overarching general basis of anti-poverty policy failure leads me to the provisional conclusion that desideratum (ii) should be given increased weight unless the consequence appears to deeply compromise desideratum (i).

Does it? The best data we now have suggest that identification of poor South Africans by reference to per capita income tracks identification by reference to per capita consumption expenditures very closely: depending on which data set is analyzed, per capita income measurement identifies between 77% and 90% of the poor who are identified by reference to per capita consumption expenditure.[53] (Unfortunately, because these data are based on household surveys, we cannot use them to compare the predictive strength of women’s per capita income and consumption expenditure. The data do, however, permit one to distinguish between female- and male-headed households.)

I thus advance a necessarily speculative empirical hypothesis here. This is that if agents responsible for development policy were mandated to focus rigidly and in common on improvements in poor women’s income, to the exclusion of other goals whose partial achievement is a continuous basis for both miscoordination and excuses by officials, then we would produce faster and more durable improvements in the lives of people in poor countries than we do now, for any given level of welfare investment.

In summary: I have agreed with Sen that people are maximizers neither of utility nor of income. However, I have argued that mainstream economic theory has not been committed to the first idea for at least sixty years now, so it is unlikely that it is of significant portent for development policy. The second point suggests that development policy is indeed something we should think of ourselves as attending to for the sake of enhanced flourishing generally or, if you like, enhanced freedom. I have then argued, however, that one of the major impediments to policy success is systematic failure of coordination, both of measures themselves and of policy officials’ incentives. The standard device for solving a coordination problem is to establish a focal point. This implies that measurement proxies for welfare improvement had best be narrowed rather than broadened. Income improvement is likely not the statistically best proxy for increased freedom. However, income improvement is the best candidate for a proxy that could serve the function of being a relatively univocally interpreted focal point. We can then turn to empirical evidence about the value to whole societies of increased flourishing for women to justify the suggestion that, once we have usefully narrowed our focus to income, we might as well narrow it still further – since, with respect to the logic of focal points, the narrower the better, at least until we get so narrow as to confront visibility problems – and measure development progress strictly by reference to women’s wealth. It is, I think, highly likely indeed that as their women get richer, poor countries become reliably freer.[54]