IV. Some Components of the Income Disparities

Age Structure

In fact, the nonwhite population has a very different age structure from that of whites. For example, in 1967 the median age for Negroes (who make up the overwhelming bulk of the nonwhite population) was 21.1 compared with 29.1 for whites. The median age for Negro men was 19.7 compared with 28 for whites. These relationships apparently have been altering drastically in recent years, and it seems that race differences in median ages used to be considerably smaller. At any rate, they are smaller now for the part of the population that is over 14 years old and receives income: in 1967 Negro men with income had a median age of 38.9 compared with 41.8 for white men. As a consequence of the differing age structures of income receivers, if Negro men had the same lifetime earnings -- that is, the same earnings as whites for each age group -- then median income would be some 98 percent of white median income; and income levels would be lower than those of whites for most of the income distribution, especially at the low end. Near the 7th percentile, for example, it would be less than 84 percent of whites. (This calculation was made by taking white incomes at each age group and assigning these incomes to a population with the Negro age structure.) Yet these do not appear to be welfare differences nor, therefore, differences in welfare resulting from discrimination. With such a transformation, there would still be inequalities but not necessarily inequities. In fact, it seems likely that such income inequalities due to differences in age structure may continue to increase for the next few years. This, at any rate, is suggested by the greater disparity between the age structures of the Negro and white populations as a whole, as distinct from the age disparity of the populations of working age with income.

However, if the adjustment is made the other way around -- that is, if instead of assigning white incomes to Negro age cohorts, one used the white age structure with Negro incomes -- then this hypothetically older Negro population would experience, as the result of that artificial advance to later stages in lifetime earning careers, little or no improvement in income for most of the distribution. For men, the improvement in the ratio of median incomes would be only l/4 of 1 percentage point and would average about 1 percentage point across the distribution. It is of interest that this adjustment shows a more sizable improvement at the lowest percentiles. For example, it makes a ratio difference of .07 near the 11th percentile, but less than .01 for all of the upper 2/3 of the distribution (see Fig. 20). For women, the ratio of median incomes would actually drop by about 8 percentage points with this adjustment.

We observed above that the difference in white and black median ages for income receivers is less than the difference for median ages of the two entire populations. This suggests that measuring changes over time for distributions adjusted for age differences for men would show a greater relative improvement than in the case of the unadjusted distributions and particularly at the low end. This is in fact the case, as shown in Fig. 21, which compares adjusted distributions for men for the years 1949 and 1967.[2] This result confirms and amplifies our earlier observations about improvements at the low end of the income distribution based on data not controlled for age.

Present and Past Discrimination, Genetics, and So On

Moreover, these various disabilities interact. The chances for benefiting from years in school are likely to be limited by disadvantages in the knowledge and motivations gained outside of school. The combined disadvantage therefore is not simply a sum, but, as Thurow suggests, more like a product of many separate disadvantages. Finally, many of the factors are hard to define or measure. "Education," especially "informal education," covers not only the transfer of information but the instilling of motives or values that improve the chances for doing productive work in our society; the quality of formal schooling is not measured adequately by the pay of teachers or the cost of school buildings or scores in reading or arithmetic; and the inadequacies of measurement both in respect to quality and content seem likely to increase with the increasing numbers of years of schooling measured. And on many of the component factors there are few or no data.

These examples suggest how hard it is to disentangle the effects of current discrimination in the marketplace from the various results of multiple past discriminations that may in turn have made it unlikely that a minority can compete currently on equal terms. While the fact of current as well as past discrimination is clear enough, the precise magnitudes involved are not clear; and, it appears to us, given our present state of knowledge, they are likely to resist exact determination. The proportion of the current income differences that is attributable to current discrimination in the market place is extremely difficult to determine, and in spite of several attempts, it does not seem to us to have been measured convincingly.

Paradoxically, the logical structure of many models that have been used to measure current discrimination against nonwhites has been almost identical with that of models used to measure a supposed nonwhite innately inferior productivity. In both cases, the model builder attempts to show that only part of the difference in the reward or achievement can be explained as a result of some list of factors explicitly treated. The fact that the model leaves a sizable part of the variance unexplained may be taken in the one case as a measure of discrimination, and in the other as a measure of the genetic difference between races. This odd logical identity was illustrated recently in Arthur Jensen's well-known controversial article, "How Much Can We Boost IQ and Scholastic Achievement?"[4] Jensen mustered a miscellany of study results that he interprets as evidence for innately lower Negro average intelligence, scholastic performance, and, apparently (the article is not precise in this respect), occupational status and differential earnings of men with the same schooling in the same line of work. The cited miscellany included studies of quite a few scholars who themselves interpret their results as measuring the effects of discrimination: James Coleman, Otis Dudley Duncan, Rashi Fein, and Daniel Patrick Moynihan.

We are not arguing against the use of simple models. Our view is that simple models are fine, and to the extent that they work, the simpler the better. but if a model does not provide an adequate explanation of a complex process, there is no logical basis to use its inadequacies as if they explained and precisely measured something not actually included in the model.

The logical problem is quite analogous to that affecting some economic models designed in the past ten or fifteen years to show the contribution of a specified list of inputs to the measured growth in output. There was a tendency, especially early on, to identify the residual growth of output not explained by changes in input quantities as an increase in the productivity of the inputs and, specifically, as a measure of the contribution of "new knowledge" to increases in output. But, of course, the residuals included errors in estimates of the influence on growth of those factors taken into account, and the influence of all factors not taken into account; the simpler and more inadequate the model the larger the apparent influence of "new knowledge."

Careful attempts to explain differences in earnings between various groups, say between professional and non-professional workers, run into quite analogous problems. Friedman's and Kuznets' careful, early investigation into professional incomes before World War II[5] estimated that (1) professionals earned on the average between 85 percent and 180 percent more than non-professionals; but (2) the extra costs, direct and indirect, of the long training needed for professional work would have called for, at most, a 70 percent extra return to make professional and non-professional pursuits equally attractive financially. A very plausible explanation of at least part of this difference between extra returns and extra costs is that the professions are "non-competing groups." (Young men are not equally able to finance professional training, are not equally aware of opportunities for professional work, and social and economic stratification makes it much easier for some to enter professions and to succeed in them.) But it is also not implausible to assume, for example, that professional work attracts people with higher level abilities than non-professional work. Inference from this analysis had to be qualified accordingly.

It is plain that both public and private institutions have offered nonwhites lesser opportunities than whites to get productive training and education, and that nonwhites have had a much more restricted range of choice among occupations. Further, it's clear that the prejudices of customers, employers, and fellow workers reduce nonwhite productivity in many jobs. Color prejudice, then, has much the same effect on the earnings of nonwhites as would a difference in ability. But the economic and educational processes involved in this result are enormously complicated and the forces that affect learning and the ability to earn money in the marketplace are highly confounded. Our understanding is quite limited. The empirical models available employ, for the most part, variance or correlation analysis or regressions with rather crude linear or sometimes log linear relations, a rather short list of quite coarse variables, and with a good many relevant variables left out of account altogether.

When we say, for example, that we are "controlling for" (or "adjusting for" or "holding constant") "occupation" or "education," it is important to bear in mind that we are not literally talking of "...men with the same schooling and in the same line of work"[6] who exhibit differences in income or employment rates, or the like. Rather, so far as occupation is concerned, we are talking about people within some very broad category of disparate jobs, one of a small number of sets of jobs into which we have petitioned gainful work. Becker[7] uses 3 categories; Hiestand, 7; Gilman, 10 (as do we in the occupational adjustments presented below); Blau and Duncan, 17; and so on for larger numbers of still heterogeneous sets.[8] But even in a 17 occupation breakdown of males, a "single line of work" such as "salaried professionals" includes physicians and surgeons as well as primary and secondary school teachers. In 1968 the mean earnings of the former were $16,273, nearly double that of the $8,240 mean for the teachers. And of course nonwhites are much scarcer among physicians and surgeons than they are among teachers. The distinction between occupational differences and differences in skill and reward within the same occupation is to a considerable extent arbitrary, even when jobs are quite narrowly defined. But the broad categories of jobs for which relevant current data are available neglect a variety of differences that need to be controlled for a moderately convincing analysis of either the effects of current market discrimination or of differences in ability, to say nothing of distinguishing between the effects of discrimination and those of ability.

Analogous comments, of course, apply to controlling for years of schooling as a surrogate for education; and for a good many other variables that figure in the familiar models. And these are only some of the limitations of the familiar models. (1) Some models regress income on variables such as "occupational status" as well as schooling, but determine "occupational status" as a linear function of income and schooling in some past year. (2) The forms of regression used, in general, lend themselves mainly or solely to the use of a summary statistic for the dependent variable (for example, mean or median incomes or the nonwhite to white ratio of means or medians), even though comparisons at other points in the distribution are quite relevant to the policy points at issue. It is not surprising, then, that in general a substantial part of the variance in nonwhite-white income differences (or unemployment rates or scholastic achievement or IQ) is not associated with the coarse variables of the models.[9]

Fortunately, policy choices need not wait for complete understanding. We don't have to know the exact dividing line between the effects of past and current discrimination to support programs to reduce both. It is plain that both are very substantial and that they reinforce each other.

As for possible genetic difference, our view is not unlike that of Eckland and of Duncan, and even more that of the geneticists James. F. Crow and Joshua Lederberg.[10] There are, of course, native differences among individuals in abilities of various sorts. However, so far as race differences are concerned, genetic theory does not imply that each of the various traits associated with skin color has persisted because it has had a survival value in past environments, nor that the effects are immutable. And genetic theory has even less to say about the relation of race differences to the complexities of contributing to the product of a modern industrial society. The mischief to be done by rejecting a true hypothesis that there are no substantial genetic factors disabling nonwhites from contributing on a par with whites is so large compared with the consequences of accepting that hypothesis if it is false, that the null hypothesis seems the appropriate one on moral and political grounds as well as scientific ones. The appropriateness of this stance is reinforced by a candid and realistic view of the many limitations in our studies and in our understanding of these matters.[11]

With these caveats in mind, we turn to a summary of preliminary work on white-nonwhite differences in income distributions associated with differences in the distribution of major occupations and years of schooling.

Occupational Distributions

While a very substantial occupational dissimilarity appears to be bad in itself, such as segregated housing and schools, the mere fact of dissimilarity doesn't offhand appear to explain why nonwhites have lower incomes than whites. It is conceivable that occupations might be "separate but equal" so far as income is concerned. The relation of dissimilarity, in fact, is symmetrical with respect to whites and nonwhites. Nonwhites may be scarce in "white occupations" because they regard them as nonwhite and inferior. Whatever the reasons for the segregation, the effects might be the same. A nonwhite in the jobs accessible to him has to compete with an extra supply of nonwhites who are capable of working in "white occupations" but are excluded from them. And he is free from the competition of the whites who chose to work elsewhere. A white, on the other hand, limited to some occupations by his own or community prejudices, would experience a parallel extra competition from members of his own race whose prejudices limit their choice, and would benefit by freedom from the competition of the excluded nonwhites.

The earnings disparity stems from the fact that nonwhites find it specifically hard to get into higher paying occupations and seem to be disproportionately limited to a range of occupations in which they can produce less and are paid less at the margin than whites.

There has been a considerable reduction in occupational dissimilarity in the last decade and a shift of nonwhites to higher paying occupations. The improvement by 1967 was larger than had been anticipated, for example, in the National Planning Association projection made early in the 1960s for the year 1972. Many shifts expected only by 1972 occurred by 1967 and the others were ahead of schedule.[13] Nonetheless in 1967 nonwhites were greatly under-represented in high paying occupations.

In Table 1, comparing white and nonwhite earnings for men in 1967 by major occupational categories, we note first that the relative status of nonwhites within occupations (as measured by the ratios of nonwhite to white median earnings) varies across different occupations. But second, the ratio of overall nonwhite to white median earnings is lower than any of the ratios for component occupation groups, with the exception of farmers and farm managers. The lower overall ratio is a result of the difference in the occupational distributions of whites and nonwhites. Nonwhites are proportionately over-represented in occupational categories such as operatives, laborers, and service workers, and under-represented in generally better paying categories such as professional and technical workers; managers, officials, and proprietors; craftsmen and foremen; and sales workers.

Notes:
^a10.1 percent of men who had earnings in 1967 were nonwhite.
^bProportions for occupations are proportion of earners who belong to that occupation group.

The Effects of Improving the Job Distribution of Nonwhite Men

The Effects of Improving the Job Distribution of Nonwhite Women

Notes:
^a12.9 percent of men who had earnings in 1967 were nonwhite.
^bProportions for occupations are proportion of earners who belong to that occupation group.

The Distribution of Schooling

However, as in the case of income, so with years of schooling, the use of medians alone can be deceptive. The distributions of schooling remain very different; even for 25 to 29 year olds it appears that while the medians are very close all this means is that most nonwhites as well as whites finish high school.[14] However, in the upper half of the distribution of whites, much larger numbers go on to graduate school or at least graduate from college. In the case of men aged 25 to 29, 75.5 percent of whites and 58.1 of blacks had completed 12 or more years of schooling; the proportions that had completed 13 or more years are only 34.3 percent and 14.5 percent respectively. Among those who completed high school then, a much smaller proportion of blacks than whites received any additional schooling. The higher black dropout rate occurs at all schooling levels. The proportions that completed 16 or more years of schooling were 19.1 percent and 5.3 percent, and for 17 or more years they were 7.9 percent and 1.0 percent.

While the difference in years of schooling completed at the 50th percentile (the median) is only .4 years, the difference at the 80th percentile is 2.8 years (12.9 and 15.7 years for blacks and whites respectively). And even for the lower end, the differences are all larger than .4. For example, at the 20th percentile the difference is 1.5 years (9.6 and 11.1 years respectively).[15]

Increased schooling (a) is clearly useful for raising nonwhite dollar income in absolute terms and (b) is plausible for increasing the overall relative income of the nonwhite working population compared with whites. But (c) the income of nonwhites with a given level of schooling need not thereby rise compared with the income of whites with an equal number of years of schooling. (d) In fact it is commonplace in the literature to suggest that as nonwhites become better educated (or at least acquire more years of formal schooling), they increasingly find themselves with lower relative income than whites of equal education (or at any rate the same number of years of schooling). In this sense, nonwhite "relative poverty" might be expected to worsen as nonwhites catch up with whites in schooling and even as the nonwhite population as a whole gains on whites in income.

Marginal Returns to Schooling for Whites and Nonwhites

But first it should be noted that if education were more evenly distributed as between whites and nonwhites, such an income decline relative to whites with the same years of schooling, as schooling increases, would nonetheless go along with a very sizable increase not only in the absolute dollar income of nonwhites but also in the relative income of the nonwhite population as a whole.

Second, a decline in the nonwhite to white income ratios with increasing levels of formal schooling might in good part be accounted for by the relatively small investment made in on-the-job training for nonwhites. This would argue not against increased schooling for nonwhites, but in favor of following investments in increased formal schooling, as in the case of whites, with a larger investment in on-the-job training, enabling nonwhites to make better use of their schooling. Mincer estimates that for a nonwhite man with college level education in 1949, investment in formal schooling averaged $13,200, and investment in on-the-job training came to $7,870.[17] The corresponding per capita investments for all men in the United States were $15,900 and $24,300, about 1/5 higher in the case of schooling and more than three times as high for on-the-job training. Mincer's estimates very likely overstate nonwhite to white differences in on-the-job training.[18] Nonetheless the differences are surely substantial, and reducing these differences will complement a relative improvement in nonwhite formal schooling. The returns to the latter will be higher if these joint effects are not ignored.

Third, it should be observed that none of the studies cited show that the relative marginal returns to nonwhites decrease steadily with higher education. For example, Hanoch, whose inquiry on earnings and schooling is the most elaborate and carefully qualified, shows a much higher relative increment for nonwhites than for whites at the graduate level (17 years of schooling and over). Miller's data show some increased marginal return at 17 or more years of schooling for the 18 to 64 year olds (the nonwhite to white earnings ratios were .576 for 5 years or more of college as distinct from .521 for 4 years) and this is even clearer for the 35 to 44 year olds (where the ratios are .612 by comparison with .527).[19] Hare found college graduates less dissimilar than college students and college students less than high school graduates. The largest marginal change in Siegel's occupational distributions was a sharp decrease in dissimilarity between those with 3 years of college and those with 4 years or more. Blau and Duncan suggest that though there is a marginal decline in relative occupational status for nonwhites with increasing education, their educational investment does begin to pay off with graduate studies.

Fourth, the gross figures for the population 25 years or over confound the difference between returns to education for given population cohorts of whites and nonwhites and differences due to an increased weight for nonwhites by comparison with whites of the younger cohorts. The more educated nonwhites are, on the average, at an early phase of their lifetime earnings cycle. They are farther from their peak income than the whites. The figures presented here do not correct for this.

Fifth, with notable exceptions, such as the studies of Hanoch and Mincer, most of the inquiries cited have looked at the gross relative returns to schooling without any explicit consideration of costs. But both direct costs such as tuition, and indirect opportunity costs in the form of earnings forgone or postponed during the years of schooling, appear to be lower for nonwhites. The relevant cost data are scarce, and they are extremely hard to estimate. Yet they have a quite critical influence on computations of private net internal rates of return. Hanoch emphasizes that his estimates of the marginal internal rates are most sensitive to errors in the rather arbitrarily estimated initial segments of his age profiles of earnings.[20] Such estimates for nonwhites are particularly doubtful.[21]

Sixth, the earnings forgone, left out of the gross figures on marginal returns to schooling, can be expected to affect the college and graduate level especially. It is only in the last years of high school, in college, or graduate school that either whites or nonwhites have earnings to forgo, and one might expect the difference between nonwhite and white forgone earnings to increase with an increase in schooling. Therefore not only do gross returns to schooling in general understate net returns of nonwhites compared with whites, but the bias very probably increases with increasing years of schooling.

Seventh, even reliably estimated net private money returns would not be decisive either for an individual or for public choice. For the individual the financial investment model of education only catches some of its major values; education is also a consumer good that may be directly enjoyed; and it offers positions in society that may be valued variously by different individuals and different ethnic groups at a given time. Moreover, a more nearly even distribution of education between whites and nonwhites has social as well as private returns.

None of the foregoing comments, moreover, take into account all the relevant variations in quality and content of schooling, and other inadequacies of our controls.

Finally, it appears that the gross returns to schooling are changing rapidly and it is not safe to generalize from a few years. Using a single year, say a decennial census date or a recent Current Population Survey year, is particularly hazardous. For example, comparing 1966 and 1967, using three categories for years of schooling, there were improvements for males at all levels, but particularly at the college level. The nonwhite-to-white ratios of median income in 1966 were .705, .701, and .653 for the elementary, high school, and college levels respectively, and in 1967 the corresponding figures were .734, .712, and .751. The improvement at the college level was statistically significant considering sampling errors, but the others were not. However, this is an important matter because it has been suggested that returns were small at the college level for nonwhite men compared with white men. For both years the ratio for all males 25 and over (.574 in 1966, .604 in 1967) was lower than for each of the three component classes because of the larger proportion of nonwhites in the lower educational groups and the higher proportion of whites in the higher educational classes. About 47 percent of nonwhite males in 1967 had only elementary school training whereas only about 28 percent of white males were at this level. The college level includes about 25 percent of whites and only 12 percent of nonwhites.

Finer breakdowns of income by years of schooling completed in 1967 show that the nonwhite to white ratio of median incomes for men is not a monotonic function of years of schooling, although more often than not the ratio does decline with increasing years of schooling. There is improvement in the ratio with the 12th year of schooling, and again for 13 to 15 years of schooling, but the marginal change in the ratio is negative for all of the other schooling categories. See Table 3. The marginal changes between adjacent categories are not statistically significant for any, but fitting a straight line to these data results in a negative coefficient for years of schooling. However, this fit is also not statistically significant, so that it is not possible to make any strong statement about the relative marginal returns to schooling at the median for white and nonwhite men.[22]

Inferences about marginal returns to schooling, however, are usually based on mean incomes rather than medians. For the data points available, the ratio of nonwhite to white mean incomes declined steadily with increased schooling. Again, see Table 3. Fitting these data to a straight line results in a negative coefficient for years of schooling, and a much better fit than in the case of medians.[23]

Straight line fits on income by years of schooling are not warranted, since a year of schooling is not a uniform unit. The 8th, 12th, and 16th years are graduation years and are thus critical points. The marginal income returns for these particular years should be expected to be higher than for other years. A straight line fit simply averages everything out and thus cannot give an accurate picture of relative white and nonwhite marginal returns to additional years of schooling. Using the nonwhite to white ratios in the regression fit at least handles the problem of the nonuniformity of years of schooling, and the fit on ratios of mean incomes indicates that nonwhite and white marginal returns to schooling follow different patterns -- that is, that nonwhites have a lower marginal return to additional years of schooling. The differences in the fits for medians and for means suggest the importance of looking at income distributions.

The Effects of Relative Increases in Years of Schooling for Nnonwhite Men

The changes from 1967 to 1968 in the nonwhite to white ratios of median incomes for men by schooling categories resemble those for the Negro to white ratios, except at the college level. The nonwhite to white ratio for all men with one or more years of college hardly changed at all from 1967 to 1968, going from .751 to .753, but it did confirm the large change from 1966 to 1967. Moreover, the corresponding Negro to white ratio improved from .691 in 1967 to .731 in 1968. The increase in the total of all nonwhite men 25 and over with four or more years of college was about half that for black men, 18 percent compared with 34 percent. The number of nonwhite men with one or more years of college increased by 74,000 in that one year, and about 65,000, or 88 percent, of those were Negroes. This is a higher percentage of Negroes among nonwhites at the college level than in previous years.

Effects of Relative Increase in Years of Schooling for Nonwhite Women

Using an adjustment on the nonwhite income distribution for women similar to that used for men, we find that for the lower half of the distribution the ratios improve by an average of 22 percentage points, which puts the average ratio above unity in the lower half of the distribution. However, in the upper half of the distribution the adjusted ratios are again higher, by about 16 percentage points, than the unadjusted ratios, but they average about 93 percent. See Fig. 25. We find, therefore, that an adjustment for years of schooling brings about a much greater improvement in the income ratios for women than for men, although in both cases the improvement is considerably greater for the lower income levels than for the higher ones.

Concluding Comments on Schooling and Jobs and the Income Ratios at Higher Quantiles

Nonwhite to white income ratios tend to be lower for the higher categories of years of schooling and for the higher paid occupations, though we have seen that these tendencies are not uniform and moreover appear to be changing in recent years. Where nonwhite to white income does decline with increased schooling or increased occupational status, there are at least two ways of looking at the phenomenon.

First, barriers to entry and promotion may be higher at these higher reaches of society. Prejudice, as many have noted, may be particularly strong against admitting nonwhites to positions involving supervision or authoritative decision or high prestige. Whatever its quantitative extent, it is plausible that something like this phenomenon has been at work.

A second way of looking at it is connected, although not identical, with our earlier comments on unexplained residuals and the inadequacies in our standardization of the factors we use in explaining income disparities. Here we would observe that the shaky equivalence of "same years of schooling" and "same education," and the shaky equation of "same census occupation" with "same line of work," get even shakier with increased years of schooling and increasingly high paid categories of occupation.

Take education, and consider first educational "quality." There has been much controversy since the Coleman report on Equality of Educational Opportunity about the effects of school quality on scholastic achievement. We don't propose to enter the lists in that battle. However, it is worth observing that a recent analysis by Alex Mood of the data gathered for the report shows that the proportion of the variance in achievement between schools associated with indexes of school quality and peer quality increases with increasing grades of school through the 12th grade.[25] Mood's concern is to point out that most of the variance removed can be associated with either the peer or school characteristics and that essentially all of the school effect can be associated with the index of teacher quality. For our own purposes, the fact that the squared correlation coefficients generally increase with increased schooling -- that is, that an increased proportion of the variance (the total variance in achievement being almost constant over years of schooling) is associated with either peer quality or teacher quality or other school effects -- is of particular interest. For it suggests that the higher the number of years of schooling the less reliable are "years of schooling" as a measure of "education" or as a predictor of scholastic achievement.

All of Mood's data refer to general education in the elementary and secondary schools. Higher education poses larger conceptual problems for measuring quality and achievement. On these grounds alone one would suspect the standardization implied in "same years of schooling" to be more doubtful for college and graduate schools.

In fact, while the content of the curriculum in the lower schools may for many practical purposes be taken as one among many aspects of "school quality" to be covered by a single index, this makes less sense at the college level or above. Education increases in differentiation as one goes from grade school to graduate school. Reading, writing, and arithmetic vary between regions, but they are being taught in grade schools in all regions, and nationwide scholastic achievement tests are feasible in principle. It is hard to think of a sensible, common measure or standardized achievement test for the progress of art historians and dentists. It strains usage to identify Fine Arts or English Lit or Classics as taught at Harvard as of a lower "quality" than Engineering or Business Administration or Medicine at the same institution. (It may strain Harvard usage especially.) Nonetheless, engineers, doctors, and business majors may in general make more money. For explaining disparities in income, "years of schooling," taken without qualification, is likely to lose explanatory force more or less steadily with the increasing level and differentiation of education.

Something like the forgoing comments about schooling at higher levels might apply to an increased shakiness in equating "same census occupational category" and "same line of work" for the higher paying categories. Within major occupational groups nonwhites are in detailed occupations that require less training and skill and "for this reason their earnings as well as earnings of the white incumbents of those (detailed) occupations will be low."[26] Moreover, the low paid work of the laborer or the household service worker seem much less differentiated than the highly varied skills of the professionals and managers. There is some evidence that there is a greater income variability among professional (and especially independent professional) than among nonprofessional workers,[27] but our own preliminary analysis of recent income data (along with a new measure of internal inequality)[28] suggest that this may not be so. Whether or not the income of professionals has a wider spread, their precise lines of work do.

The inadequacies in standardization of the education and occupation variables appear then to make for understating the gains that might result from a more even distribution of nonwhites and whites among jobs and schooling categories. The increased differentiation of schooling and work at higher levels, however, has some implications for policy: Attempts to reduce the income disparity along the entire income distribution might aim more precisely at the kinds of formal and informal education and training best calculated to prepare nonwhites for a substantially higher paying range of occupations. And, so far as jobs are concerned, private as well as governmental efforts need to focus not simply on entry level jobs capable of barely lifting nonwhites out of poverty, but on increased representation in those detailed occupations that might give them an even chance with whites for middle and higher incomes. It may be rather hard to apply fair employment practice laws to promotions for merit within a given line of work. But the more narrowly defined an occupation, the less troubling this may be and the more precisely the nonwhites may aim and prepare to earn higher incomes. This does not eliminate the problem of prejudice against nonwhites in supervisory jobs but it may circumscribe it somewhat.

Finally, we have stressed that the tendency for nonwhite to white income ratios to decline with higher categories of years of schooling has not in the past been uniform. And it appears now to be in the process of change. Hanoch, for example, believes that the reversal at the graduate level of the relative decline in his estimated nonwhite marginal returns to years of schooling may be quite meaningful in spite of all the uncertainties of the estimates. And we suspect he is right, given the analogous improvements at higher levels found in a number of independent studies.[29]

We might conjecture several explanations for such an improvement at the graduate level in past decennial census years. First, it is possible that nonwhite graduate schooling, much more than schooling at the college level or below, took place at predominantly white schools. Up to recently, no all-Negro college awarded Ph.D.s', most did not confer any graduate degrees or had been doing so only for a short while, or had been offering graduate as well as undergraduate training in a range of professions with relatively low prospects for lifetime earnings. (This could be so partly because they were professions servicing the low income Negro community or because they were professions that in general paid less.) Assuming this were so: