Prabhat Patnaik at Stand With JNU, 9 March 2016. Photo by Subin Dennis.

Affirmative Action and the “Efficiency Argument”

Prabhat Patnaik

(In this article, Prabhat Patnaik demolishes the “efficiency argument” which claims that affirmative action, including reservations, lowers the “efficiency”, or quality of output, in any sphere where it is practised. The basis for an analytical objection to affirmative action on the basis of the “efficiency argument”, he says, can only be provided by a rejection of the presumption of even distribution of talent across social groups, in favour of an alternative presumption that either talks of the all-round superiority of some groups over others (which underlies racialism and concepts like the herrenvolk), or talks of different groups having different kinds of talent (which is supposed to justify the caste system). If both of these alternative presumptions are rejected as they should be, then the analytical objection to affirmative action on the efficiency argument not only disappears, but becomes its very opposite, namely an argument for affirmative action.)

Prabhat Patnaik at Stand With JNU, 17 February 2016. Photo by Subin Dennis.
Prabhat Patnaik speaking at the Freedom Square in JNU, 17 February 2016.

Both the defenders and the critics of affirmative action, of which reservations are one particular instance, are paradoxically in agreement on one point, namely that such action lowers quality of output in any sphere where it is practised. For example if there is affirmative action practised in the recruitment of doctors, then it lowers the quality of the service offered by the medical profession; if there is affirmative action in the appointment of university teachers, then it lowers the quality of teaching, and so on. But while the proponents of affirmative action argue for it on the grounds of “social justice” and “equity”, the opponents profess to be unwilling to pay the price of a lowering of quality for the sake of such “social justice”.

The statement that even the proponents of affirmative action believe that it would lower the quality of output may appear surprising at first, but is nonetheless true. Indeed the very phrase “social justice” which has been advanced to defend reservations since the days of the V.P. Singh government in India, bears out this claim. Underlying the demand for affirmative action is a desire for “justice”, which is supposed to override all other considerations. Implicit in this view is the admission that these other considerations are likely to be pulling in the opposite direction, i.e. there is an implicit recognition of the existence of a cost in terms of loss of quality. Or putting it differently, since silences are also eloquent, the exclusive emphasis on “justice” unaccompanied by any critique of the “efficiency” argument, is tantamount to an acceptance of the “efficiency” argument.

Since any such loss in the quality of output is equivalent to a reduction in the amount of output being produced, measured in terms of “rupees’ worth “, it follows from this perception that in a society with affirmative action, the vector of outputs in any period is smaller, for any given endowment of inherited material and human resources, than in the absence of such action. This loss in output represents therefore an “inefficient” outcome. Affirmative action in other words is commonly conceded to produce an inefficient outcome, even when it is defended in terms of “equity” and “social justice”. I call this argument, which is based on a comparison of the output vectors in two situations, the “efficiency” argument.

The purpose of the present paper is to examine this “efficiency argument” in some detail. It argues in particular that the efficiency argument, far from going against affirmative action, actually supports it. In other words, the argument for affirmative action is not merely that justice overrides efficiency, though that is a perfectly legitimate position to take, but that both justice and efficiency demand it.


Since intelligence, or talent, is evenly distributed in any society across different groups, i.e. the probability distribution of talent within each group is pretty much the same for all groups, it follows that an inordinately large representation of one particular group, or a few particular groups, in any profession is ipso facto efficiency-lowering, for then it must be the case that less talented members of the “privileged” groups would be getting accommodated in lieu of more talented members of the less privileged groups. Putting the matter differently, as long as the selection from within each group is on the basis of talent, then the closer the group-wise composition of the members of a particular profession is to the group-wise composition of the population as a whole, the greater would be the level of efficiency in the economy. Efficiency enhancement, in other words, requires not only that selection be made on the basis of talent, but that the composition of the recruits into any sphere should be as close as possible to the composition to the population as a whole. Since in practice selection made on the basis of talent appears nonetheless to throw up recruitment patterns whose composition is unrepresentative of the population as a whole, then, given the presumption that talent is evenly divided among groups in a population, the only reason for it must be the existence of some implicit or explicit barriers to entry that certain groups must be facing. As the objective of affirmative action is to enable those groups which face barriers to entry to overcome these barriers, it follows that affirmative action is efficiency-enhancing.

The question that immediately arises is: what exactly constitutes a “group”? Given the analytical nature of the above argument, this question too requires first of all an analytical answer. It is obvious that between two groups, one of which has ten million members and the other ten, the proposition that talent is evenly distributed across groups, is likely to lack general validity. The proposition acquires greater general validity as each group acquires a larger size. A group must have a certain minimum size before it can be said that the distribution of talent within it approximates the distribution of talent for the population as a whole, or the distribution of talent within any other group with a similarly above-minimum size. It can be larger than this minimum size but not smaller, since the very fact of the group being large enough to approximate the distribution of talent within any other group or the population as a whole implies that its becoming still larger will not deprive it of this property. The answer to the question what exactly constitutes a group may then be given in a manner that appears at first sight to constitute circular reasoning but is actually far from it, namely that a group in this argument consists of that conglomeration of individuals the distribution of talent within which is almost exactly the same as in the population as a whole.

The whole debate about affirmative action in the Indian context of course has occurred around social groups, or caste categories. But since these social categories around which the debate has taken place are very large in size anyway, and since the basis for demarcating these categories is such that the presumption about the even distribution of talent across categories remains valid, it follows that these categories qualify to be called groups in the analytical sense we have just defined, in which case we can unhesitatingly assert that affirmative action, such as is proposed in the Indian situation, is efficiency-enhancing.

Affirmative action, it was mentioned above, is meant to overcome barriers to entry. But in a situation where talent has unequal scope across individuals and groups for revealing itself, any selection exclusively on the basis of revealed talent itself constitutes a barrier to entry. In other words the imposition of formal equality in a situation that is fundamentally unequal, acts in an inequalizing manner, i.e. in a manner that attains the very opposite of equality. Affirmative action here must take the form of overcoming the effects of the fundamental inequality, for which it must go beyond formal equality in order to achieve substantive equality. Since the revelation, or expression, or nurturing, of talent takes time, affirmative action may entail some short-term “efficiency loss”, but it would necessarily achieve long-term efficiency gain. The short-term “efficiency loss” arises from the fact of violation of the principle of recruitment on the basis of existing “revealed talent”; the long term efficiency gain arises from the fact that the distribution of “revealed talent” changes over time through affirmative action. (The affirmative action required for the “nurturing” of talent, in the sense of bridging the gap between “intrinsic talent” which is evenly distributed, and “revealed talent” which is not, to start with, will in practice entail much more than reservations alone; it will need an additional remedial effort in the form of “bridge courses”, pre-requisite courses, special programmes and such like).

But then how can we compare the short-term loss with the long-term gain? If the agency making the choice was characterized by extreme short-sightedness, then the consideration of loss would triumph. The economist Arthur Cecil Pigou had called such short-sightedness “a lack of telescopic faculty”. Whether or not individuals are characterized by such a lack of telescopic faculty is itself debatable. Many writers including Joseph Schumpeter doubted this strongly. They denied that individuals had any innate positive rate of time preference in the sense of preferring a unit of additional consumption today to a unit of additional consumption tomorrow if the consumption stream in the absence of this additional unit is otherwise identical between today and tomorrow; they ipso facto therefore also denied that individuals always work with finite time horizons. If individuals do not have such a lack of telescopic faculty, then attributing such a lack to society as a whole of course makes no sense, even when society is seen as a mere agglomeration of atomized individuals. But even if individuals as atomized entities have such a lack, but society is seen not as an agglomeration of atomized individuals but as a supra-“atomized individual” entity, then again there is no reason why society should have such a lack, since society in this sense lives for ever. (Societies of course may actually be hit by calamities, but this fact never enters into their reckoning).

For societies then, the question of having a short-sight, or a lack of telescopic faculty, or even a positive rate of time preference, simply does not arise. Consequently, the long-term efficiency gain in their reckoning should always necessarily outweigh the short-term efficiency loss, since the efficiency loss is transitional but the efficiency gain, once it starts accruing, keeps accruing in every period ad infinitum. It follows then that, as far societies are concerned, their time-horizons must necessarily be always so long that affirmative action for them is efficiency-enhancing.


The formal argument for affirmative action advanced so far is curiously reminiscent of the “infant industries” argument for protection from free trade. The basic point about the “infant industry” argument is that the industry can become fully competitive, with equal or even lower costs of production compared to what the current producers in the industry, hitherto located in other countries, have. But to become so it needs time, to outgrow its infancy, as it were. Not to. protect it entails a long run loss of efficiency (if its costs would become lower than what currently prevails); but protecting it entails for a time, when it has not yet outgrown its infancy and hence is uncompetitive, with costs higher than those currently prevailing, a loss of efficiency. The infant industry argument states that such an industry must be protected; the short-run efficiency loss must be suffered for the sake of the long-run efficiency gain, which is exactly the argument for affirmative action, and indeed for reservations.

The infant industries argument is almost universally accepted. But the argument for reservations is not, even though the latter argument is even more powerful than the infant industry argument on at least two counts. First, the infant industry argument is typically an argument for protecting the infant industry of a nation against producers located elsewhere. Even if the eventual cost of production of the national producers will be the same, or of a similar order of magnitude, as the cost of production of producers located elsewhere, and no lower, there will still be an argument for protecting national producers. In other words, an actual lowering of costs relative to what currently prevails internationally is not required for the argument to be valid, and hence an actual lowering of cost is not necessarily postulated. But reservations entail an actual improvement in efficiency once the initial phase is over, because of the presumption of even distribution of talent. Once this presumption is granted, the case for reservations is even stronger than the case for infant industry protection on efficiency grounds. In other words, reservations necessarily improve efficiency in the long-run in a way that infant industry protection does not.

Secondly, which industries deserve infant industry protection is not always clear in advance. Nobody can say with certainty beforehand that the industry that gets infant industry protection will necessarily be able to achieve a level of cost over time that is comparable to what prevails internationally. But on the presumption of even distribution of talent across groups, the target groups for whom reservation should be enforced are quite clearly visible beforehand. True, there are always alternative methods of classification because of which one may arrive at alternative distinctions between the “privileged” and the “excluded”, but this practical problem of classification is qualitatively different from the analytical problem arising from the fact that there is no certainty about the target beneficiary not achieving the requisite level of efficiency to justify protection/reservation. But while the “infant industry” argument is universally accepted despite this shortcoming, the reservation argument is not so accepted, despite its being free of this shortcoming.

There is another important difference between the reservation argument and the infant industry argument. An infant industry, assuming it can achieve the requisite level of efficiency, will do so if allowed to come up. In other words, the argument itself postulates that the traverse from high immediate cost to low eventual cost in the case of the protected industry is an automatic process, a result, if you like, of “learning by doing”. Protection against international competition for a while is all that is needed for the infant industry to stand up and compete freely when protection is withdrawn. But in the case of affirmative action, as already suggested, reservation will have to be accompanied by a host of other remedial measures to ensure that the innate talent of the beneficiaries manages to get revealed. In the case of reservation in higher education, these measures will include “bridge courses”, “remedial courses”, more flexible programmes (such as provisions for repeating courses etc.). “Learning” here cannot simply be through “doing”. It will have to be additionally provided.

If we think of affirmative action in higher education as a package consisting of reservation together with such remedial measures, then it is difficult to see what analytical objections can be raised against it. There will to be sure be a host of practical problems, of classification, of the possible need for going backwards and having to introduce affirmative action even at a pre-higher education stage, of actually arranging the complementary remedial courses, and so on; but there cannot be any analytical objections on the basis of the “efficiency argument”, of the sort that have been raised in India. The only basis for such an analytical objection can be provided by a rejection of the presumption of even distribution of talent across social groups, in favour of an alternative presumption that either talks of the all-round superiority of some groups over others (which underlies racialism and concepts like the herrenvolk), or talks of different groups having different kinds of talent (which is supposed to justify the caste system). If both of these alternative presumptions are rejected as they should be, then the analytical objection to affirmative action on the efficiency argument not only disappears, but becomes its very opposite, namely an argument for affirmative action.


There is a second, methodological, objection to the “efficiency argument”. This objection is based on the same premise as the one that implicitly underlies the view that “social justice overrides efficiency considerations”. Let us look at this objection and in the process make this implicit premise more explicit. When we say that state A of society represents a more efficient point of production than state B, in the sense of entailing, for given endowments of material and human resources, a larger vector of outputs, that fact does not ipso facto make state A preferable as a social state compared to state B. What it means is that the output vector of state A, since it is larger (in the sense of >), can in principle be distributed among all members of a society in a manner such that at least some persons are better off, with none being worse off, than in state B. In other words, in state A everybody’s utility can in principle be vector-wise higher than in state B. (In technical economic parlance this is often expressed by saying that the utility possibility curve associated with state A is higher than the utility possibility curve associated with state B).

But to say that everybody’s utility can in principle be greater in state A than in state B is not the same as saying that it is actually greater than in state B. In other words, the claim that the output bundle available in state A can in principle be distributed in a manner such that everybody can be made better off compared to state B, is not the same as saying that the output bundle is actually so distributed. If A has two goods, say x and y, in amounts 100 and 80 respectively while B has the same goods in amounts 50 and 50, then clearly A represents a more efficient point of production compared to B, if the initial resource endowments are the same in the two cases. Clearly with the larger output bundle of A everybody can be made better off compared to his or her situation in B. But “can be made better off is not the same as saying “is better off. If there are only two persons in the society 1 and 2, and while the output bundle is shared equally between the two in state B but accrues entirely to l in state A, then clearly 2 is worse off in A than in B. So, even though A is a more efficient point of production, so that everybody can be better off in A, not everybody actually is better off in A.

In all such cases where not everybody is better off in one situation compared to another, the comparison between the two situations requires some kind of a social judgment. (Such a judgment incidentally is also required even when everybody is better off in one situation compared to another, but most would accept the judgment that the more efficient point in this case is socially preferable). It is quite possible therefore that while A is more efficient compared to B, social judgment may be such that B is considered socially preferable to A. A more efficient state in other words is not necessarily the socially preferred state. In a society which places a premium upon “distributive justice”, B will be socially preferred to A even though A happens to be the more efficient point of production. This is the sense in which the proponents of affirmative action argue that, even conceding that affirmative action may entail “inefficiency”, “distributive justice” overrides “efficiency considerations”, and hence affirmative action that will shift society from A to B is justified.

This extremely elementary principle of Welfare Economics would suggest that those who argue against the shift from A to B, because such a shift entails an efficiency loss, must either believe that distribution does not matter, which is nonsensical, or at best represents a value judgment which is socially unacceptable; or that, contrary to first impression, everybody is actually better off m state A than in state B. (“Everybody is better off” here should again be interpreted only in a vector-wise sense, i.e. it should be interpreted as “some are better off with nobody worse off”). Their argument for instance may be as follows: if there is reservation, say in the recruitment of doctors, then the quality of the medical treatment offered by the doctors will suffer, in which case everybody, including those who stand to gain from such reservation, will suffer in their capacity as consumers of such treatment; and what is more, the gain they make as beneficiaries of reservation in recruitment will be more than offset by the loss they make as consumers of such treatment. But if this was the case, then the latter group too (at least some of them) should be opposed to reservation that is ostensibly meant to benefit them. If they are not (and there has never yet been any instance in India of the purported beneficiaries of affirmative action being opposed to such action on the grounds that they would be net losers from it), then ipso facto they must be net gainers from it, in which case the claim that everybody is better off at state A rather than state B must be false.

It follows that the standard argument so often heard in India that reservation is socially undesirable because it will lower quality is invalid, whichever way one looks at it. What the argument refers to is a vector-wise loss in output, which will be socially undesirable (even assuming that we accept the basic Pareto- criterion that a vector-wise increase in the utilities of individuals entails an improvement in social welfare and a vector-wise reduction in utilities a loss in social welfare), only if it entails a vector-wise loss in utilities, i.e. some being worse off with nobody being better off. But since reservations do make some people better off, this vector-wise loss in output (if there is such a loss that is) does not per se entail a loss in social welfare. On the contrary if society values distributive justice, then it entails a gain in social welfare. (There may of course be ways other than reservations for bringing about the same improvement in distributive justice starting from state A, but those are not germane to the argument here).

One of the most remarkable features of the debate on reservations in India has been that such reservations have been presented as being against the social good. But if this had actually been the case, then there should have been at least some members belonging to the social group that is the targeted beneficiary, who should have been against reservations. Those among them who would individually lose from such reservations even if it is for their group might have been expected to oppose it; but there is no such opposition. And if social good is defined in a supra-individual sense, then this argument, if it is compelling, should have persuaded at least some of them to rise above their individual interests and stand for this transcendental social good; but there has been no such instance. “Social good” in short has had no takers among the excluded groups, which suggests that this notion of “social good” has been a self-serving notion advanced by the privileged groups.


Let me pull together the threads of the argument advanced till now. The commonest argument used against affirmative action, which in the Indian context has essentially meant reservations, is that it lowers the quality of output, i.e. causes a vector- wise loss in output in terms of “rupees’ worth”. A vector-wise reduction in output in any given period, for given endowments of material and human resources, constitutes a loss of efficiency. Hence this argument against affirmative action can be called the “efficiency argument”.

There are two basic problems with the “efficiency argument”. The first is that a move from one state to another when there is a vector-wise loss in output, i.e. an efficiency loss, is not necessarily socially undesirable. It is only in the case where the loss in output makes some people worse off without making anyone better off that we can call it socially undesirable (though even this entails a value-judgment). But in a situation where the loss in output is accompanied by some becoming better off and some worse, the comparison of social states requires a specific distributive judgment. If the state entailing a loss of efficiency is associated with a distribution of output that is considered socially “better”, then this so-called “less efficient” state is socially preferable to the more efficient one. Efficiency in short is not synonymous with being socially preferable when we compare any two states. The “efficiency” argument in the way it is usually presented therefore is a false argument.

The second problem with the efficiency argument is the following. So far we have talked about just one period in which there is a vector-wise output loss. But since talent is evenly distributed across different social groups (each having a minimum size), reservation should have the very opposite effect of increasing rather than decreasing the vector-wise size of output. True, this does not happen immediately, so that the vector- wise output loss argument may hold in the short-run. But as soon as the talent of the recruits belonging to the hitherto excluded categories reveals itself, there should be an output gain relative to the situation prior to affirmative action. This is similar to the infant industry argument for trade protection: a short-run efficiency loss is more than made up by the long-run efficiency gain. In fact one can be more certain about this happening than of any particular infant industry making good. But oddly, while there is universal acceptance of the infant industry argument, there is little acceptance of the argument for reservations as a means of achieving efficiency gains (though for these gains to materialize, reservations have to be backed by a set of remedial measures).

Finally, we must note an asymmetry. While distributive considerations imply that the short-term efficiency loss is not per se a reduction in social welfare, but may constitute an improvement in social welfare relative to the initial situation, the long-term efficiency gain should certainly cause an improvement in social welfare relative both to the short-term situation and to the initial situation. Relative to the short-term situation it would entail ceteris paribus a larger output vector and hence an improvement in social welfare. Relative to the initial situation it will entail both a larger output vector and distributive justice, and hence an improvement in social welfare.

This last point can be established as follows. Suppose the government uses “lump sum” taxes to ensure that between the initial situation and the long-term situation, nobody is worse off. Then the fact of efficiency gain because of the sheer change in the composition of recruitment will make the long-term situation preferable. Now, if income distribution is allowed to change through an adjustment in the “lump sum” taxes, to reflect the change in the composition in recruitment, then this would make the long-term situation additionally preferable on grounds of distributive justice (as long as the change is justifiable on grounds of distributive justice). Affirmative action therefore can be defended on grounds of both efficiency and distributive justice.


(Valedictory Address to the Conference on “Affirmative Action in Higher Education in India, the United States and South Africa”, organised by the Programme for the Study of Discrimination and Exclusion, School of Social Sciences, Jawaharlal Nehru University, New Delhi and the Centre for Comparative Constitutionalism, University of Chicago on 19-21 March 2008 in New Delhi. This text of the address is reproduced here with the permission of the author.)