V. Summary and Conclusions

1. From the 1966, 1967, and 1968 data, the most powerful conclusion to be drawn is that nonwhite family and personal incomes are much inferior to white incomes along the entire distributions of each. It is not simply a matter of the middle of the distribution. Still less is the trouble confined to the low ends. The differences are pervasive and they are displayed most sharply in the existence of upper limits that tend to bound nonwhite personal income.

2. "Inequality" among nonwhites, measured in standard ways, is not much different from "inequality" among whites. In the case of families, nonwhite income is slightly more unequal. Nonwhite income to persons is slightly less unequal.

3. Nonwhite income is cyclically more unstable than white. But theory suggests that tight labor markets, like slack ones, have persistent effects; and in fact lasting gains have been made, especially during the very tight labor markets of World War II, Korea, and Vietnam.

4. The ratio of nonwhite to white income to persons did not return to prewar troughs after World War II and Korea, but it did decline with the loosening of the labor market. After Vietnam, it may do so once more, unless a well-aimed and executed policy prevents it.

5. Nonwhite family income has grown relatively faster than white income since before World War II and since the end of it. At the median, nonwhite family income has increased about a third more rapidly since 1947; and nonwhite income to persons twice a fast as white since 1948 -- the starting year for continuous (unpublished) annual data on income to persons.

6. Nonetheless, even at these rates, convergence would take place only in a distant future -- even for median income to persons at some time near the end of the century.

7. Relative gains over time have been largest at the low end of the distribution. In fact, compared with white, there has been little or no change in nonwhite income at the top.

8. This conclusion is emphasized if one takes into account the sizable differences in the current age structure of nonwhite and white populations and in their relative changes over time. Some of the income difference in the lower percentiles is accounted for by the larger proportion of young people among nonwhites. And, since this age difference has increased in the postwar years, the improvements at the bottom end are larger when income is controlled for age. On the other hand, the fact that such age adjustments do not significantly affect the middle and upper end of the distribution confirms the hypothesis that nonwhites get much lower money returns to increasing age and experience.

9. Adjusting nonwhite earnings for differences in distribution among major occupations improves the nonwhite to white income ratios for both men and women. For men it removes about a third of the disparity on the average throughout the distribution and almost half the disparity in the lowest fifth. For women, where the disparity is smaller, it is erased altogether in the lower half of the distribution. (In fact, the ratio is greater than unity there.) It does less in the high end of the distribution for women as well as for men. Nonetheless, the ratio at the high end for women is close to unity except in the last decile, the top of which is almost unaffected by the occupational adjustment.

10. Adjustments for differences in years of schooling show much the same character: for men, they eliminate about a third of the disparity on the average, less than this at the upper and more at the lower half. In the case of women they put the ratio well above unity for the lower half of the distribution and slightly below it for the top half. both the occupational and the schooling adjustments exhibit the particular difficulty nonwhites have in receiving high income.

11. Equality in years of schooling, or jobs in the same broad occupational category, only very crudely approximate identities in education or line of work. A substantial part of race income disparities, variously measured, may be associated with a few such coarse variables by simple linear or log-linear rules. Not surprisingly, this generally is not so for a still larger fraction of the disparity, however defined. Models that try to use the disparity unexplained by such relationships to separate precisely the effects of present from past discrimination have in general the same logical structure and the same defects as those that take the unexplained disparity as genetic and possibly immutable. In one case the unexplained residuals are attributed to present race discrimination in the marketplace and in the other to race inferiority. But these residuals plainly have a great deal to do with inadequacies in the standardization of the variables, with factors left out, and with differences between the nonlinearities in reality and the simple relations assumed in the models.

12. Some related though distinct comments apply to generalizations about nonwhite relative returns to extra years of schooling. In past decennial census years, several inquiries have found that nonwhite marginal returns appear generally, though not steadily, to decrease relative to white returns with increased years of schooling. However, it appears that the inadequate standardizations involved in equating years of schooling or broad job categories become even less adequate as schooling and skill levels increase. This increased differentiation of higher education and higher level skills (as well as prejudice against putting nonwhites in authoritative, supervisory or decision-making positions) may account for the behavior of the earlier census data, and also might help to explain why in these past studies relative returns to schooling, after a decline, turn and increase at the graduate level. For at the graduate level nonwhites in the past more frequently obtained their training in white schools. Graduate schools also tend to train people for professional rather than supervisory careers.

13. These relations are changing, and for recent years we find quite weak any decreasing trend in gross relative marginal returns for nonwhite men. But even if the returns to schooling were smaller for nonwhites, this would not imply that more schooling is unimportant for either the relative or the absolute income of the nonwhite population as a whole. The returns to schooling for nonwhites, whether smaller or larger than for whites, are nonetheless positive; and nonwhites are less schooled than whites. Equalizing the schooling distribution (as stated in point 10 above) closes the average income gap by about one-third. In the case of women it appears that nonwhite marginal returns are higher than white.

14. Some final comments on implications for policy: These comments are based on a larger study of which this monograph is a part. A good many aims of policy that are loosely connected with the problem of race differences in income have been confounded with it: reducing inequality in society as a whole, poverty, unemployment, ghetto riots, among others. These are all worthy or plausible goals but they are not the same goals.
    1. Eliminating the differences in income between whites and nonwhites need not affect inequality in society as a whole. In fact, if nonwhite men were given the same level and distribution of income as whites, this would in recent years have increased inequality (measured by the relative standard deviation or the variance of the natural logarithms) for the aggregate of whites and nonwhites.

    2. Although nonwhites are disproportionately poor, two-thirds of nonwhites are not poor, and more poor are white.

    3. If nonwhite unemployment rates were the same as those for whites, this would help some specific categories of nonwhites (teenagers and women especially) but would have very little direct effect on nonwhite income in the aggregate. In 1966 it would have increased it by less than 2 percent. On the other hand, of the 60% increase needed to close the gap, nearly half would be achieved if nonwhites were distributed among the major job categories in the same proportion as whites. And, even if there was no change in occupational distribution, more than half of the gap would be closed if nonwhites received the same rate of pay in each occupational category.

    4. Such a convergence of the income distributions of nonwhites with whites can be justified on grounds of equity. However, it has quite uncertain relations to the problem of reducing the incidence of riots and of increasing public order in the short run. The unsophisticated views that men riot because they are poor or getting poorer are plainly inadequate, but the more sophisticated and paradoxical theories that men riot because things are getting better are not much more persuasive as generalizations. Past and present theories of "relative deprivation" specify in advance too little about the reference group taken as standard to be proved wrong; or to provide much information about the effects on public tranquility of future decreases in the disparity in rewards to nonwhites and whites. Grounds of equity are quite adequate and offer the most persuasive justification for reducing race differences in income.