Aug 22, 2001 - actions of individuals have been taken, and according to the promise of g ..... and so I would not abandon race-based affirmative action policies.
August 22, 2001 “Defending equality of opportunity” by John E. Roemer1
I will take up various criticisms that have been leveled against the theory of equal opportunity as I have expounded it in Roemer (1993, 1996, 1998, in press).
1. I begin with a criticism of Mathias Risse (2001), as it serves to point out a weakness in my conceptualization of ‘circumstances,’ or , if not a weakness, then at least a failure to point out certain possible pitfalls associated with the concept. Risse states my view as including, importantly, two postulates:
MR 1. There is a distinction between choices and circumstances, such that success with regard to the objective should depend only on one’s choices (effort), whereas negative effects of circumstances provide grounds for compensation; and
MR 2. The extent to which an individual should achieve the objective is assessed in competition with the performance of other individuals. That is, choices (effort) are evaluated in a way that essentially involves consideration of an individual’s choices in competition with other individuals’ choices.
1
Department of Political Science, Yale University. I am grateful to G.A. Cohen for comments on an earlier draft.
2 For concreteness’ sake, let us take the example of wage-earning capacity as the objective, and let us assume, for simplicity, that a person’s wages are precisely and monotonically related to the effort she expends in school, and her socio-economic status (SES). Thus, there is a ‘production function’ that, holding other inputs constant, relates an individual’s future wage to her effort in school and her SES, in the obvious way: the wage is a monotone increasing function of both effort and her family’s SES. I said, in this case, that the individual’s circumstance was the SES of her family, and that her effort was measured by the number of years of school she attended. I also said that circumstances are aspects of a person’s environment that are beyond her control and have an effect on her achievement of the objective, while effort is that constellation of behaviors that are taken explicitly to be within her control. The use of ‘control’ implies that a person has the power to change certain of her behaviors, whether because not all of her choices are deterministically explainable, or because – the compatibilist view—determinism is consistent with genuine choice. I believe that is the most sensible premise for the theory. But I sometimes said that a society might hold a person responsible for certain behaviors, even if they were beyond her control. In this case, effort can be construed as those behaviors for which the society holds the person responsible. Besides effort and circumstances, the third input which affects the value of the objective the individual achieves is the intervening policy – in this example, the resources the state devotes to education. Thus, I often wrote the complete production function as:
v i = v (e i , C t ( i ), x t ( i ) )
(1)
3
where ei is the value of i’s effort, t(i) is the type i belongs to, C t(i) is the circumstances of i’s type, , and xt(i) is the value of the policy intervention that i’s type enjoys. Now Risse points out that this is a very special case; more generally, the success of an individual depends on the efforts of others – we can say that the efforts of others are an externality that is germane to each individual. For example, if the objective is a wage level, then each person’s wage level may be a function of the entire distribution of skills in the labor force, which is clearly a function of the efforts of others.
Thus, taking Risse’s point on board, (1) must be amended. Let Ft be the distribution of effort of the tth type, and let F = ( F 1, F 2 ,..., F T ) be the vector of distributions of effort of the various types. Then we should write:
v i = v (e i , C t ( i ), x t ( i ); F , C − t ( i ), x − t ( i ) ),
(2)
where C-t(i) is the vector of circumstances of all types other than i’s type, and x-t(i)is the vector of policy values for all types other than i’s. Equation (2) allows the outcome for i to be influenced by the efforts of others, the circumstances of others, and the policy values enjoyed by others. To reiterate, the difference between (1) and (2) is the following. Equation (1) models a world in which each individual is an island, producing ‘output’ ( the objective) in his own backyard and consuming it. The output an individual produces is not affected by the activities of others. Equation (2), in contrast, models a world where there are
4 interactions between the productivity of individuals. Clearly the second world is the more general and interesting case. My terminology was ambiguous, because I sometimes said (sticking to the above example) that a person’s circumstances were his family’s SES, his race, his sex, and so on, while I other times said that his circumstances were those aspects of his environment, beyond his control, that affected his outcome. Under the latter construal, circumstances should contain the whole vector (C t ( i ), x t ( i ); F , C − t ( i ), x − t ( i ) ) -- all components of which are beyond his control and affect his outcome. I here assume there are large numbers of individuals in each type, so that the distribution of effort in a person’s type is beyond his control, even though we assume he can alter his own effort. I suggest that this ambiguity be resolved by continuing to refer to the characteristics of individual i that are beyond his control as his circumstances, but to refer to the whole vector (C t ( i ), x t ( i ), F , C − t ( i ), x − t ( i ) ) as i’s environment. Thus, i’s circumstances are only one component of i’s environment. All components of i’s environment are beyond her control – assuming that each type has a large number of individuals, and so no individual can change the distribution of effort in her type by her choices. Risse’s critique of my view is, as I understand it, essentially the following. Ft(i) is beyond the control of i. Therefore, F t(i)should be considered a component of i’s circumstances.
Therefore i should be compensated if the efforts of others in his type
negatively affect his outcome, by MR1. But this conflicts directly with MR2, in which we decide how much of a person’s outcome is deserved by ranking his effort against the efforts of others in his type. Thus, my theory is internally inconsistent.
5 But it is wrong to say that, because the distribution of effort in one’s type is beyond one’s control, therefore one’s rank in that distribution is beyond one’s control. By presumption, i’s rank in that distribution is within his control. It is both the case that every other individual’s effort level is beyond i’s control, and that his rank among those effort levels is within his control. The view, here, is that the range of effort levels for a type (or, what is called the support of the distribution of effort for the type) indicates the range of effort which is reasonably available to members of that type, and that a member of that type can voluntarily choose any level in that range. ‘Reasonably available’ is intended as shorthand for ‘what a person can be expected to strive for, given the beliefs induced by, and constraints associated with, the circumstances of the type.’ Let me try to use the slippery language of moral arbitrariness. My circumstances, the circumstances of others, the distribution of efforts of other types, and, if my type is large, the distribution of effort of my type, are all morally arbitrary for me, while my own effort level if not morally arbitrary for me. Because my effort level is not morally arbitrary for me the relationship between my effort level and the effort level of someone else is not morally arbitrary for me: I could alter that relationship.
This is consistent
with saying that the distribution of effort levels of others in my large type is morally arbitrary for me. It is relevantly morally arbitrary, for this theory, because that distribution reflects the circumstances of my type, and hence, I may be due compensation because that distribution has, for example, a low mean value, which in part explains why my own effort was lower than that of others in more advantaged types.
6 2.
Proceeding to a new point, some challenge the functional form (2), and say that
no two persons – even of the same type – face the same environment, or circumstances, because ‘local conditions’ are immensely variable. But that critique is an extremely general one, which applies to all social science, for it is a critique of the procedure of aggregation of individuals into categories of individual, based upon their having similar environments (in some sense). If we took that critique seriously, we would have to abandon all statistical inference in social science. If one believes that the differences between the circumstances of individuals are always more important than their similarities, then each individual should be classed as the unique member of his type, and equality of opportunity collapses into equality of outcomes. Another way of stating this objection is that luck is all important, and that luck comes in so much variety that it cannot be captured by denoting a discrete list of circumstances (like SES, race, and sex). I adopt the view that circumstances are the predictable aspects of a person’s situation, that are morally arbitrary, and that are, as well, protracted or chronic. Luck, although morally arbitrary (at least, the kind we are here contemplating) is both unpredictable and episodic. With this convention, we may reasonably capture a person’s circumstances with a discrete list. The substantive challenge remains, that luck (those morally arbitrary episodes that are not captured in the list) may be extremely important. I will return to this issue in section 7.
7 3.
Let me continue discussing the EOp algorithm. If two persons of the same type
differ in their efforts, then I say, according to the EOp principle, the one with higher effort deserves to enjoy a higher value of the objective. But how do we decide, between two individuals of different types, which one has expended the higher effort? My solution to this problem was as follows. Let the types to which these two individuals belong be denoted t and q. Then the distributions of effort in t and q are characteristics of their types. An individual should not be disadvantaged by characteristics of her type, according to the EOp view. I proposed that we factor an individual’s effort into two components: her centile (more generally, quantile2) on the effort distribution of her type, and the effort distribution of her type. Formally, we characterize individual i’s effort by the ordered pair ( π i , F t ( i ) ) , where Ft(i) is the distribution of effort in i’s type, and π i is i’s rank on that effort distribution (that is, i’s effort quantile). Now I said that Ft(i) is a characteristic of i’s type, while π i is i’s voluntary choice. It is therefore only appropriate to hold an individual responsible for her rank on the effort distribution of her type, and not the absolute value of her effort, which is in part the consequence of the factors accounting for the distribution of effort in her type. Hence, I declared that two persons in different types should be deemed to have expended the same effort (I said , degree of effort) if their ranks on their respective effort distributions were the same. I claimed that we can only tell how hard a person has tried ‘on her own hook’ by comparing her to others of her type, because a large part of a person’s actions may well be determined by her circumstances. By comparing a person only to others of her type,
8 we hold constant that circumstantial part, and hence we can measure relative effort. Thus, we can unequivocally say that
JR1. Two persons of the same type have tried equally hard if and only if they sit at the same rank of the effort distribution of their type.
I then said, in addition, that we declare
JR2 .Two persons of different types have tried equally hard if and only if they sit at the same rank of their respective effort distributions.
I noted that JR2 does not logically follow from JR1. I invented an ‘assumption of charity,’ which I have since abandoned, to move from JR1 to JR23. The assumption of charity has been challenged by a number of people, including Risse. From JR2 (which for now, absent the assumption of charity, lacks a justification, but will be argued for below), we can conclude that:
JR3. Equality of opportunity prevails when, for every rank π, all individuals at rank π on the effort distributions of their respective types enjoy the same value of the objective, and that value is increasing in π.
2
A quantile is a rank on a distribution, normalized to be in the interval[0,1]. It is the infinitesimal version of the discrete analogues of quintile, decile, centile, etc. 3 The assumption charity (see Roemer[1998, p.15]) begins with a premise that persons possess some ‘deep individuality’ beneath their circumstances, and that that individuality includes a propensity to expend
9
When the situation described in JR3 prevails, then any two individuals who have tried equally hard enjoy the same outcome, regardless of their circumstances, and those who expend more effort enjoy better outcomes. The reader will note that many distributions of the objective could satisfy JR3. So JR3 will eventually need some refinement, unless we wish to admit a multiplicity of objective distributions as all satisfying EOp. In particular, JR3 ignores the issue of whether a particular set of distributions of the objective, among the types, is ‘efficient’ or unsurpassably good.
4.
Now in practice, effort is not unidimensional, and so it is not correct to speak of
effort as something which can be linearly ordered, and in addition, the conglomerate of actions comprising effort is often very hard to observe completely. Therefore, although JR1 and even JR2 may be appealing, they are not very useful in applications, where we endeavor to compute what the EOp policy intervention is.
For example, saying that
effort can, as in the example above, be measured by the years of school attended, is clearly a gross simplification. Let us define precisely a policy. A policy consists of policy components, one for each type. A policy component is a function that associates certain actions that individuals may take, that are observable aspects of effort, with the allocation of a resource that the policy maker is endowed with. Thus, let these actions be denoted a; then a policy component is a function gt(a). A policy is thus a vector of policy
effort. The assumption then states that the distribution of this propensity to expend effort is the same in all types. I now believe that the premise upon which the assumption is based is incoherent.
10 components g= ( g1,..., gT ) , one for each type. For a policy g to be feasible, it must be the case, first, that it is legal for the policy to be implemented , that the actions a can be observed, and that the total amount of resource the policy maker must distribute, once the actions of individuals have been taken, and according to the promise of g, will not overspend the policy maker’s budget (the budget constraint). Although effort is not unidimensional, it is usual that the objective for which we seek to equalize opportunities is unidimensional (an income level or a wage rate or a life expectancy , to name some important applications), and so I attempted to implement JR1 and JR2 by adopting the following convention:
JR4. Within a type, if all individuals face the same policy component, then the outcome [value of the objective] observed is an increasing function of ‘effort.’
JR4 is a convention, because it says, even if effort is a vector of actions, we will say that ‘more’ effort has been exerted by the person in a type who achieves a higher value of the objective. In other words, JR4 is a method of creating a unidimensional index of effort by mapping vectors of specific efforts, in a reasonable way, onto outcomes: whatever effort is, in a given situation, its expenditure should augment the objective under consideration. Now JR4 is not sufficient to provide us with a cardinal, unidimensional measure of effort. It provides us with only an ordinal measure: we can say that if individuals i and j in a given type faced the same policy component, and i achieved a higher outcome, then i expended more effort that j. But, it turns out, that is all we need.
For observe:
11
JR5. If all persons in a type face the same policy component, then a person’s rank on the effort distribution of her type is equal to her rank on the outcome distribution of her type.
JR5 follows from JR4. Since, according to JR3, all we need to implement the EOp policy is knowledge of the ranks of persons on the effort distributions of their types, and not the cardinal values of ‘effort,’ it follows from JR5 that we need only observe a person’s rank on the outcome distribution of her type. And that is a readily available datum.
Thus, from JR5 it follows that:
JR6. Equality of opportunity prevails when the outcome distributions of all types are identical.
JR6 follows from JR5 and JR3. Now one might say: Why go through this contentious process of distinguishing degrees of effort from levels of effort, and not just define equality of opportunity by JR6? The answer is that the argument I have outlined provides microfoundations for JR6, in the sense that it relates the claim of JR6 to the more fundamental view that what should count is effort and what should not count are circumstances. making inter-type comparisons of effort will not lead to JR6.
In particular, other ways of
12 It is, indeed, a practice among some social scientists to treat JR6 as a definition of equality of opportunity.
Consider the concept of an inter-generational transition matrix.
The rows of this matrix are labeled by the incomes of parents, the columns are the incomes of their children, and the ijth element in the matrix is the fraction of children of parents with income i who earned income j. Sociologists and economists sometimes say that equality of opportunity holds in a society when the rows of the transition matrix are identical. This means precisely that the distributions of income of children, in each type (where type is defined by parent’s income) are identical: thus, JR6. The difference between equality of opportunity and equality of outcomes is that the latter prevails when the outcomes of all individuals are identical. But EOp, which is to say JR6, does not attempt to eliminate differences in outcomes within any type; thus, it is content to equalize distributions of outcomes, because it interprets the different outcomes along a distribution (within a type) as due to differential effort, and therefore not compensable. Equality of outcomes holds when all columns, except one, in the transition matrix consist of zeroes.
5.
Now, there are two obvious problems with JR6. The first is one of efficiency.
We might be able to achieve JR6 by starving everyone. Thus, we must be concerned not only with achieving EOp, but with achieving it at a level which is, in some sense, unsurpassable4. The second is that, in most situations, there is no feasible policy that will achieve the identity of all type-outcome distributions. My way of addressing these two problems was as follows.
4
We need not include (Pareto) efficiency as an aspect of equality of opportunity, but I am concerned with proposing an allocation of resources that both equalizes opportunities and is efficient.
13 If it were not the case that every individual in a type faced the same policy component, then we could not say that the environments of any two individuals, in a type, were the same. Thus, it is not necessary that all individuals in a type receive the same amount of resource from the policy maker, but rather that they face the same menu (function) which relates their chosen action to the resource they will receive. Across different types, we may, however, offer different menus. Of course, the actions taken by individuals (their efforts) will depend, inter alia, on the announced policy. In particular, it also follows that the rank of an individual on the effort distribution of his type may depend on the policy. Some people are bothered by this, thinking it introduces an undesirable degree of arbitrariness into the view; they believe that the equality-of-opportunity ethos is only compelling if the rank of a person on the effort distribution of her type is independent of the policy.
I see no reason to
require this. Let me give an analogy. Some people believe that a person deserves what he gets in a competitive equilibrium of a market economy with private ownership of assets. Others object that this cannot be so, because more than one equilibrium is possible. If several equilibria are possible, in which an individual receives very different incomes, then we would have to say that this individual deserves many incomes at the same time. Although I do not agree with the proposition that a person deserves what he gets at the competitive equilibrium, I do not think that that argument defeats the proposition. There is no inconsistency in saying, “At a particular market equilibrium, a person deserves what he earns.” Multiplicity of equilibrium just means that what a person deserves is not
14 determined entirely by the institution of competitive markets; it is also determined as well by history, that is, the path that leads to a particular competitive equilibrium. In like manner, I have no problem with the fact that there may be several policies that lead to different distributions, all of which equalize opportunities, although the fate of a particular individual may differ among them, because, in part, her degree of effort differs among them. To define the EOp policy that solves the two problems raised above, let us proceed in two steps. First, fix an outcome rank π, a number in the interval [0,1]. Find that policy that maximizes the minimum level of the objective achieved by individuals at rank π of the outcome distribution across types. Call this policy gπ . This is a ‘maximin’ policy, with respect to the achievements of all those who exert a particular degree of effort. In general, it will not be the case that all outcomes of those at rank π will be equal at gπ : this may be due to the usual kinds of incentive problem that force us to settle for maximin rather than equality. We interpret gπ as the policy that equalizes opportunities (for acquisition of the objective) at the highest possible level for those at rank π of society’s effort distribution – society’s π-effort tranche, so to speak. If the policies gπ were identical, for every π, then that single policy would be, unequivocally, the EOp policy. But that will almost never happen. Thus, we seek a second compromise. Here, there are a number of equally good moves. I will mention three. The first, which I have expounded in the writings referred to in the paper’s first sentence, is that
15 policy that maximizes the sum of minimum values, over π, of the objective achieved across types, for each π. (See the mathematical statement in equation (5) below.) The second, which I have used in some applications, is to take the average of all the policies gπ , over π. The third, expounded by Van der Gaer(), is to take that policy that maximizes
the minimum of type-averages of the objective, over types. I do not think there is a robust argument to favor any of these choices over the others. They are all efficient second-best solutions to the non-existence of an unambiguous equal-opportunity policy. I conjecture that, for most real-world problems, these three procedures will yield quite similar solutions, though I have no theorem to that effect.
6.
There are, I believe, (at least) three important problems with the theory as I have
thus far outlined it that are left to consider. These are:
A. Does not the occurrence of luck (other than the luck of what circumstances one has) destroy JR4? B. What is the argument for JR2? C. JR6 expresses no concern with the degree of inequality in outcomes that occurs within types at the EOp policy. Is the inequality within types that EOp allows too severe?
7.
Question A was raised, initially, in section 2 above.
Good(bad) luck, in the
sense of non-predictable, episodic, morally arbitrary influences , may put a person at a
16 higher(lower) rank on the outcome distribution of her type, than would be expected from the effort she expended. This is not a problem for ‘ideal theory,’ where by that term, I mean when we ignore the problems of implementing our policies with real data. But it is a substantial problem, given real data sets, which usually do not permit a great deal of disaggregation into a large multitude of types – large enough to account for the myriad forms that luck can take. We must usually be satisfied with a fairly coarse partition of the relevant population into types, based upon one or two or three variables of circumstance. The answer to A is, I think, that there will always be mistakes in the implementation of EOp policy: some who had bad luck will not be compensated as much as they should be, and some with good luck will receive too much compensation. We can eliminate some of these errors by constructing data sets in which we attempt to measure the most important kinds of luck. The presumption must be that luck is not the major determinant of the outcome – that, once we have accounted for measurable circumstances, most of the variance in outcomes is due to what society would be happy to call effort. This is a presumption. If it is false, then the whole project of implementing equal opportunity, as I have conceived of it, is called into question. Indeed, if one believes that idiosyncratic luck is the overwhelming determinant of outcomes, then one might well advocate equality of outcomes as the desirable kind of equality. Indeed, this would follow from insisting that, by virtue of the importance of luck, each individual is in a type by himself, for, as I wrote earlier, in that case, equality of opportunity reduces to equality of outcomes.
17 8.
I turn to problem B, which is the still open problem of finding a level-comparable
measure of effort, across types. Effort comes naturally as a multi-dimensional vector of quantities that are in principle measurable: years spent in school, hours per night spent on homework, hours spent planning one’s future, and so on. There are also aspects of effort that are much harder to quantify: units of will power expended to avoid engaging in dangerous practices, and so on. Perhaps some of these dimensions of effort are not even quantifiable in principle. Our theory requires that we be able to compare the effort levels of different persons, because it will declare that all those who expended the same effort level should receive the same outcome. There are two problems in constructing such a levelcomparable measure of effort: first, collapsing a multi-dimensional variable into a single variable, and second, purging that single variable of the effects of circumstances. The second task refers to the following problem. It may be much harder for a poor girl to finish university than a rich girl, other things equal, because of the extra sacrifices she faces in doing so. But ‘sacrifices’ is probably too narrow a designation. In addition to the fact that the poor girl must take the costly decision to defer her entry into the labor market in order to finish university, which is not a costly decision for the rich girl, there may be an expectation from family and friends that the rich girl finish university, while the poor girl will be supported by no such expectation, and indeed, may have to act against the expectations of her community. Thus, of two such girls, both of whom completed university, we would naturally presume, without more information, that the poor girl expended more effort: she tried harder on her own hook. How can we shape
18 this intuition into a precise, level-comparable measure of effort of individuals across types? Surely, there is no unique way of doing so, unless, perhaps, we could discover through neurophysiology some center in the brain which emits a chemical, or fires synapses, in proportion to the degree of psychologically salient effort expended. Short of such a discovery, we require a rule of thumb, such as JR2. ( Below I express skepticism that such a neurophysiological solution will ever be found.) I will reduce JR2 to two somewhat more primitive assumptions that together imply it. Let us call ‘sterilized effort’ the unidimensional measure of effort that is levelcomparable across types – the kind of effort measure we seek. (Sterilization refers both to boiling the circumstances out of effort, and collapsing the many dimensions of effort into one.) My two assumptions are:
JR2a. The lowest sterilized effort levels expended in any two types are the same and the highest sterilized effort levels expended in any two types are the same.
Denote these two levels by zero and one, respectively.
JR2b. The mapping from the quantile of the outcome distribution to the sterilized effort level expended by individuals, in a type, is linear.
We have already agreed that the quantile of the outcome level in a type is an ordinal measure of the sterilized effort level. We seek a function e t ( π) , which will be the level
19 of inter-type-level-comparable sterilized effort expended by the individuals at quantile π of the outcome distribution in type t . JR2a says that
for all t, e t (0) = 0 and e t (1) = 1.
(3)
JR2b says e t ( π) is a linear function of π; given (3), this determines e t ( π) to be
e t (π) = π.
Hence JR2 follows. What is the alternative to JR2b, if we endorse JR2a? If the mapping from the quantile of the outcome distribution to the sterilized effort level were quadratic, for instance, then we would have:
e t ( π) = a t π + (1 − a t ) π 2 , for some a t ∈[0, 2] . (4)
The functional form (4) is the unique quadratic form that has two properties that we require: it obeys (3), that is, JR2a, and it gives a measure of effort that is monotone increasing in π. Thus, we see, with a quadratic relationship, we have one degree of freedom in defining sterilized effort, the number at. In other words, we can change the way we compare sterilized effort levels, across types, by changing the vector
( a1, a 2 ,..., aT ) .
20
In like manner, if we allow a cubic relationship between π and sterilized effort, we will have two degrees of freedom for the mapping et in each type. In sum, the assumption JR2 can be derived from a much weaker assumption than JR2, but one of the same kind – that is, JR2a, which asserts only that the ‘laziest’ and ‘most industrious’ persons, across types, have exerted the same degrees of effort —and an assumption of mathematical simplicity of the mapping from outcome levels to sterilized effort, JR2b. My justification for JR2b is one of simplicity: if we have no clear way of solving a problem, we should take a simple convention in place of a solution: Procrustes’ razor, if you will.
9.
I turn to problem C. Even with JR2, there is, as I wrote, in general no
unequivocal meaning to ‘equalize at the highest possible level, for each degree of effort, the outcome’. We cannot equalize (or even maximin) many interdependent quantities simultaneously. I mentioned earlier three possible solutions to the problem. The one I have advocated is:
1
Max T
( x ,..., x )∈X 1
∫ Min v√ (π; x ,..., x t
0
t
1
T
) dπ ,
(5)
where X is the set of feasible policies, and v√t ( π, x1,..., x T ) is the level of the objective for individuals of type t at sterilized effort level π of their type when the policy is ( x1,..., x T ). In particular, in this formula, the minimum objective levels across types are simply
21 summed. Thus, (5) is , as I wrote, ‘Rawlsian’ with respect to differences due to circumstances, but ‘utilitarian’ with respect to differences due to effort (how to treat the different effort tranches in the population). Now some, such as Marc Fleurbaey (in press), would prefer that, in the social objective, we give more weight to the interests of those who expended less effort, substituting something like
1
Max 1 T
∫ [ Min v√ (π; x ,..., x t
( x ,..., x )∈X
0
t
1
T
)]ρ dπ ,
where ρ is some number between 0 and 1, for (5).
(6) Formula (6) will force the
mechanism to be more concerned with inter-effort-level inequality than (5) is. My feeling about (6) is that it mixes two principles: equality of opportunity and equality of outcomes. I certainly do not contend that we should be unconcerned with the inequality that remains once opportunities have been equalized; but that has no bearing on whether (5) is an optimal statement of what equality of opportunity requires. Perhaps equality of opportunity, as captured by (5), leaves too much intra-type inequality. More generally, is the equality-of-opportunity view too unforgiving? Should society refuse to pay for the surgery of the motor cyclist who, for the fifth time, has crashed and sustained head injuries while not wearing his helmet? (This is the epitome of a low effort individual, who might well not receive such surgery.) It probably is. No rule, I contend, will give the right answer in all cases. Even if equality of opportunity is justice, we may prefer a society which makes room for charity, and is thereby less just.
22 10.
I do not claim that problems A,B, and C exhaust the set of problems with the
EOp view. Let me reiterate another problem that I discussed in section 12 of Roemer(1998). What is the scope of EOp? As I wrote, I would not advocate equalizing opportunities with respect to the height of candidates, taken as a circumstance, for hiring by professional basketball teams. Thus, there is a proper scope for the application of the EOp principle. That scope, I said, cannot be determined without a complete theory of justice for the community. Equality of opportunity, as I have expounded it, is concerned with a sub-population of the community, and in the end we must consider the consequences of employing an EOp policy for that sub-population on the welfare of the community as a whole. In the case of professional basketball players, the community includes the fans, and it seems clear that the welfare of the many fans trumps the welfare of the relatively small number of short players whose bad luck we should not attempt to rectify by requiring teams to hire the hardest trying of them. I did not put forth the EOp view as an overall theory of justice for a community, but rather as an approach to policy that could be applied piecemeal in various spheres. I advocated, as a rule of thumb, that EOp be used when the objective was educating or training people for future competition in the labor market, but that, in that actual competition, we use meritocratic principles for assigning jobs: hire those most capable of performing the required functions (Roemer [1998, section 12]). Thus, I would employ EOp in admitting students to medical schools, but merit in the hiring of doctors. Recently, Richard Arneson (2000) has challenged this approach. Returning to the view he expounded in his celebrated piece on equality of opportunity (Arneson [1989]), he writes that the problem of scope can be avoided by not applying EOp in a piecemeal ,
23 ‘spherical’ fashion. Let us equalize the opportunities for all-round welfare in the whole population, he says. It will be in the vast majority’s interest to have highly competent surgeons (thus, in the interest of all but those few would-be incompetent but high-effort surgeons whom an EOp policy might admit to the surgery profession), and so equalizing opportunities for all-round welfare will indeed imply a meritocratic principle in hiring. Arneson (2000, p.344) writes:
The aim of social justice should be to improve the overall life prospects of individuals, not per se their prospects in this or that particular social practice or institutional setting. If we picture social life as divided into various institutional and social-practice spheres, then the more costly in terms of human benefit and harm it would be to equalize (maximin) within some particular sphere, the stronger the case for shifting the task of equalization to spheres where this can be done more efficiently.
This is surely correct, as ‘ideal theory.’ But, donning now the hat of a practical social scientist, I demur. For I believe that, if equality-of-opportunity is to become influential in social policy, it will surely be on a piecemeal basis, in one and then another sphere of social life. It is therefore incumbent upon a policy-oriented advocate to define the appropriate limits of EOp policy , that is, to provide some rules of thumb for its scope. Since publishing my book, interesting developments with respect to affirmative action in university admissions in the United States have occurred, about which it is useful to reflect, using the language of EOp. Policies which took race as a circumstance have come under heavy attack, and, in many universities, have been abandoned. In three states – Texas, Florida, and California—they have been replaced with affirmative action
24 policies that effectively take residential location as the key circumstance, which is probably a quite good proxy for social class. In those states, the public universities will now admit a substantial fraction of their students, not on the basis of scores on the SAT, but if they are in the top p% of their high school class, with regard to grades, where the number p varies across these states. In other words, students admitted in this way are compared only with others in their own high school. As high schools remain, for the most part, quite homogeneous with respect to social class in these states, this policy will tend to admit the high-effort students from each social class. To be sure, the objections to the race-based affirmative action admissions policies were in part due to an opposition to any equality –of- opportunity policy, in favor of a policy based strictly on merit. But they were also based on the perception that race was not a good measure of disadvantaged circumstance, and that is the issue I am addressing here. There was, in addition, a view that a society that wishes to teach its members to avoid distinctions based on race should not make such a distinction a basis for social policy.
My own view is that there is a disadvantage to being black in the United States
above and beyond its association with being poor, for which compensation is necessary, and so I would not abandon race-based affirmative action policies. (See Betts and Roemer [2000] for evidence.) Nevertheless, my point here is that the resolution of the problem was not to abandon equality- of- opportunity admissions policy, but rather to redefine the set of circumstances upon which it is based. Contrast this to the fate of affirmative action policies in hiring. These , too, have been opposed, and largely abandoned – but not replaced with another kind of ‘level
25 playing field’ policy. Rather, hiring is to be done on the meritocratic principle – cast the net broadly, but hire those who are best equipped for the job. American society seems to be saying (if I may ascribe it a common will), that it approves of leveling the playing field with regard to education, but not with regard to hiring. The social costs incurred by the first policy, but not the second, are deemed acceptable. To complete the picture on hiring, I should mention the Americans with Disabilities Act (ADA), passed in the early 1990s. This law is, indeed, based upon the EOp view. It takes disability as the circumstance, and requires employers to make investments to enable disabled employees to work productively in their establishments or firms.
It is perhaps unfortunate that the individual firm must pay the cost of such
enablement, rather than society as a whole, but it probably would be extremely costly to implement the law properly under the alternative arrangement.
Thus, American society
seems ready to extend the EOp principle to those with biological disabilities, but not those who are relatively disabled by virtue of social practices. Perhaps it is because the biological kind of disability seems, at least in most cases, to be both beyond the control of the individual and morally arbitrary. Many American citizens, however, still consider the dysfunction associated with poverty to be not beyond the control of the victim, or not beyond the control of the victim’s parents (and many would hold the parents, rather than society at large, responsible for the fate of the child).
11.
Susan Hurley (in press) has written that “Roemer’s account does not show how
the aim to neutralize luck could provide a basis for egalitarianism.” Her view is that,
26 absent luck, many possible distributions of the objective could have occurred, and one cannot claim that ‘neutralizing’ luck means to render outcomes sensitive only to degrees of effort.
Moreover, she writes that it is not an argument for EOp that it neutralizes the
effects of luck. Let me state clearly that the moral premise of the EOp view is that rewards should be sensitive only to the autonomous efforts of individuals. This is, I think, a special case of rewards according to deserts. People deserve, in the EOp view, to accomplish objectives in proportion to how hard they try. Thus, strictly speaking, the EOp view is not one whose fundamental primitive is equality: deservingness is fundamental , together with the normative thesis that justified inequality tracks deservingness.
Inequalities
that are not due to unequal efforts are defined as being due to luck: that is, luck is socalled because it is a cause of reward that is illegitimate from the EOp view. The statement that ‘EOp intends to neutralize the effects of luck on outcomes’ is therefore equivalent to the statement ‘EOp intends to render outcomes sensitive only to effort.’ So, for example, suppose a child, A, does well in life because his parents were rich, not because he exerted great effort, while another child, B, from a poor family, does well by virtue of exerting great effort. Some might argue that it may be no less a matter of luck that B was the kind of person who works hard than that A had rich parents, but that approach, whatever its merits, is not the sense in which responsibility-concerned egalitarians use the word luck. Good luck, for us, means the source of non-effort caused advantage5. To be sure, it is not an argument for EOp that it neutralizes luck, it is rather
5
Curiously, Hurley (2001) herself recognizes this when she writes: “However, there has also been a tendency simply to use ‘luck’ rather than ‘brute luck’ to refer to factors for which people are not responsible: Cohen himself slips into this usage in opposing choice to luck … I follow this latter usage.” (my italics)
27 definitive of the EOp view that it does so. The argument for EOp must be that is right to render outcomes sensitive only to effort6.
I have made no effort to provide an argument for this ethical premise. Indeed, I am unsure whether I endorse it, which means that I am unsure whether equality-ofopportunity, as I have been expounding it, is an ethically correct principle. My attempt has been to expound, in a precise way, a view that I believe has massive popularity in a vulgar (imprecise) form. I am also convinced that implementing the EOp algorithm would bring contemporary societies closer to justice than what currently stands in them, and for that reason I advocate the EOp view – not because it is necessarily the perfect form of justice, but because it will move us closer to it. In my 1998 book, I said that I put forth the EOp view not as a theory of justice, but as an algorithm that any society could use to implement a form of equality-of-opportunity consonant with its own views of responsibility (thus, how to make the cut between effort and circumstances). My advocacy of EOp was and is based upon the view that, most societies, if they should apply the EOp algorithm to an objective, with natural conceptions of what constitute circumstances, would eliminate what is in fact massive unfairness.
12.
My pragmatic view, of advertising the EOp algorithm as a tool for any society,
has invited some to ask whether there is a correct view of what constitute effort and circumstances, as opposed to a particular society’s view, and thus if there is a correct
6
The point in these two paragraphs was clarified for me by Cohen(2001).
28 application of the EOp algorithm, as opposed to a number of culturally relative applications. This, it seems to me, is a non-issue. EOp could only claim to be a theory of justice if the conception of effort it employed were the correct conception. Of course, each society that applies EOp will believe that its conception of effort is the correct one. The issue is, perhaps, why I advocate the application of the EOp algorithm under various different conceptions of responsibility, and I have answered that in section 11 above.
13.
I conclude with several examples of application of the EOp algorithm to
empirical problems.
In Llavador and Roemer (2001), we ask : how should international
aid be distributed in order to equalize the opportunities for growth of a set of 56 developing countries? Here, the objective is the growth rate of GNP of a country, the circumstances are aspects of a country that are only possible to change in the long-run, and that influence growth, and effort is taken to be what the World Bank defines as ‘good economic management,’ policies that can be changed in the short run. We compute three allocations of international aid: the present, observed allocation, the EOp allocation, and the utilitarian allocation, where the last allocates aid in order to maximize the growth rate of total GNP of the set of countries. It turns out that the EOp allocation is the most egalitarian of the three. Indeed, the utilitarian allocation is extremely inegalitarian: it would award aid to only three countries in the set! (If the world aid budget were larger, it would award aid to more countries.) The principal difference between the EOp allocation and the observed allocation is that the former would award less to African countries and
29 more to Asian and Latin American countries – because the ‘effort’ of African countries is so poor. This does not mean that we would allow the African countries to sink, but rather that they should be allocated trainers and advisers, to improve their economic management, rather than monetary aid. In other words, we admit that ‘effort’ as measured by the World Bank’s index might not be the most fundamental conception of effort, and that perhaps African countries are expending futile effort, in the sense that their effort does not result in good management policies. Indeed, in a number of African countries, increasing aid appears to decrease the growth rate, because it induces more rent-seeking and less productive activity. In Roemer et al (in press), we study eleven advanced countries (ten in Europe plus the US). We take as the objective the income received by individuals in the country; we use as the instrument to equalize opportunities for income, the fiscal system of taxes and transfers; we take as a person’s circumstances the level of education of his parents. We partition each country’s residents into three types, conditional upon the level of education of their parents. Thus, in each country, we compute three distributions of pre-tax income, one for each type. The aim is to use fiscal policy to render the post-fisc distributions of income of these three types as close as possible to being equal. We limit ourselves to a small set of policies, the set of affine tax policies that are revenue neutral. Thus, each tax policy in the feasible set involves taxing income at a constant marginal rate, and returning a lump-sum transfer to every household. We compute the optimal EOp tax policy, taking into account the fact that labor supply is elastic with respect to tax rates. We compare the optimal policy to the observed policy,
30 and thus we rank the eleven countries with respect to how closely they achieve the EOp optimum. It turns out that the northern European countries tax more than EOp, so defined, requires, while southern Europe, UK, and the US tax far less.
This is due not
only to the level of taxation in these countries, but also, of course, to the degree of prefisc equality: in the Nordic countries, there is very little pre-fisc inequality among the three types, for various reasons (population homogeneity, and the solidaristic wage policy). Here is an instance where applying EOp would in fact reduce the degree of equality that exists in a society. There are several caveats to the conclusion that the northern European countries have gone farther than EOp requires. The first is that we have chosen a very truncated set of circumstances: we use only the educational level of the parents of the individual. The second is that we have chosen a set of policies that is very small: unidimensional, to be precise. If we considered a larger set of policies, then the result of over-taxation in northern Europe would evaporate.
Thus, there are many choices in
framing an EOp problem that will influence the conclusion: the set of circumstances, the set of feasible policies, and the precise definition of the objective. In Betts and Roemer (2000), we take US males as the population, wage-earning capacity at age thirty as the objective, circumstance as the race of the individual and/or the educational level of his parents, and educational expenditures as the instrument. We ask: if we could target different educational expenditures to students of different types, what distribution of educational expenditures would equalize opportunities for wageearning capacity? In this case, we deduce that EOp would be highly compensatory. In a
31 typology where we take four types, defined by whether the individual is black or white and whether his parents had ‘high’ or ‘low’ education, we find that the EOp expenditure would be about nine times as much for members of the ‘low black’ type than of members of the ‘high white’ type. Indeed, the High Black, Low Black, and Low White types would each receive more than their per capita share of educational resources. This is a far cry from what is conventionally thought to be the educational finance allocation that would equalize opportunities, namely, the allocation where the same amount is spent on each student (equal resources). In the US, the actual allocation is regressive in the sense that more is spent per capita on advantaged types than disadvantaged ones, due to the local financing of education through property taxation. We also compute the EOp allocation with respect to four types defined only with regard to the educational level of parents, and ignoring race. Here, the variation in expenditures across types would be much less than what was reported above. Indeed, we show that, under this EOp policy, the fraction of blacks in the lowest income quintile would hardly change from what we currently observe – which seems to say that, if we wish to change the relative economic position of US blacks (in which approximately 38% of blacks find themselves in the lowest income quintile), then we will have to target race, specifically, as a circumstance. There are, of course, implications for the American debate on affirmative action. There is, clearly, a process of reflective equilibrium transpiring here. The applications of the EOp algorithm lead to policy recommendations that are not crazy, which tends to re-enforce the view that the algorithm is sensible. Saying the applications are not crazy does not mean that they are politically realistic. But, if the
32 reasoning that has led to the algorithm is sound, and if it is true that the algorithm makes precise a popular view, then there is hope that the algorithm’s recommendations will become politically feasible, as people who advocate a level-the-playing-field view learn that their principle implies these recommendations.
33 References Arneson, Richard. 1989. “Equality and equality of opportunity for welfare,” Philosophical Studies 56, 77-93 --- 2000. “Economic analysis meets distributive justice,” Social theory and practice 26, No. 2, Summer, 327-345 Betts, Julian and J.E. Roemer, 2000. “Equalizing opportunities through educational finance reform,” Working Paper, available at http://pantheon.yale.edu/~jer39/ Cohen, G.A. 2001. “Reply to Hurley and Arneson,” (limited distribution) Fleurbaey, Marc. In press. “Egalitarian opportunities,” Law and Philosophy Hurley, Susan 2001. “Luck and equality,” Proceedings of the Aristotelian Society, Supp. vol 75 Hurley, Susan. in press. “Roemer on responsibility and equality,” Law and Philosophy Llavador, Humberto G. and J.E. Roemer, 2001. “An equal-opportunity approach to international aid,” Journal of Development Economics Risse, Mathias, 2001. “What equality of opportunity could not be,” Dept of Philosophy, Yale University Roemer, John E. 1993. “A pragmatic approach to responsibility for the egalitarian planner,” Philosophy & Public Affairs 10, 146-166 --- ,1996. Theories of distributive justice, Cambridge, Mass.: Harvard University Press ---, 1998. Equality of Opportunity, Cambridge, Mass.: Harvard University Press
34 ---, in press. “Equality of opportunity: A progress report,” Social Choice and Welfare , available at http://pantheon.yale.edu/~jer39/ --- and R. Aaberge, U. Colombino, J. Fritzell, S. Jenkins, A. Lefranc, I. Marx, M. Page, E. Pommer, J. Ruiz-Castillo, M.J. San Segundo, T. Tranaes, A. Trannoy, G. Wagner, and I. Zubiri (in press) “To what extent do fiscal systems equalize opportunities for income?” Journal of Public Economics, available at http://pantheon.yale.edu/~jer39/ Van der Gaer,
35
