The returns to on-the-job training: are they the same for blacks and whites?

I. Introduction

The trend in earnings differentials among demographic groups in general, and the status of black workers relative to white workers specifically, is a timely and ongoing topic of research. Nearly two decades of evidence suggests that while wage gains associated with increases in human capital acquisition by blacks were steady and sustained from post-World War II through the rapid progress of the 1960s, and even throughout most of the 1970s, the progress of those years slowed or ceased altogether in the 1980s.(1)

Various explanations have been advanced in an attempt to explain the white/black wage gap and how it has changed over time. Some have focused on the important roles of education, literacy, and migration and have shown that the convergence in the earnings of blacks and whites can be largely attributed to the convergence in these human capital variables between the two groups.(2)

Other studies have suggested that the important factor in explaining changes in the white/black wage gap has been anti-discrimination legislation such as Title VII of the Civil Rights Act of 1964 [17; 2; 5]. Recently, some have suggested that other factors, such as family background [10], educational achievement and quality [18; 20], and occupational shifts [3] are playing an increasingly important role in explaining white/black earnings differentials.

In this paper, we contribute to the work previously done by focussing on the role that on-the-job training plays in explaining wage differentials among demographic groups and by investigating how shifts in the acquisition of training by different groups over time have impacted those differentials.(3) Specifically, we study the on-the-job training characteristics and wages of four race/sex demographic groups in 1976 and 1985: white males, black males, white females, and black females. The importance of on-the-job training on wage growth in general is demonstrated by Brown [6], and forms the basis for our hypothesis that such training models may be valuable in explaining wage differentials across demographic groups. Using a Brown-type model,(4) we confirm his fundamental finding that much of the tenure-related wage growth results from increased training. We are able to demonstrate this finding for all four race/sex groups in both 1976 and 1985.

However, we also find significant training differences between blacks and whites. For example, blacks tend to have acquired less current training than whites. We also find differences in the impact that prior training has on earnings for the two groups. Previous training acquired by whites tends to increase earnings in their current jobs. However, we find that previous training acquired by blacks has no significant impact on their current earnings. Thus, blacks are unable to bring their previous training, or the higher earnings associated with that training, with them when they change positions. Previous training is non-portable in this manner for blacks even when the previous training was acquired at the black worker's current firm. Both lower black levels of current training and black non-portability of training help explain lower black wages. Further, we find little evidence of progress in these two areas in the decade from 1976 to 1985.

In addition, we determine how much of the observed white/black wage differential can be attributed to training differentials and how the impact of these training differentials is changing over time. Most significantly, and especially for black men, we find that differences in human capital characteristics explain more of the white/black wage gap than do differences in the returns to those characteristics.

II. Data

This study employs cross-sectional data from the Michigan Panel Study of Income Dynamics (PSID). Separate samples are drawn in both 1976 (Wave 9) and 1985 (Wave 18) for four specific groups: white males, black males, white females, and black females. Waves 9 and 18 of the PSID are used because on-the-job training data is available in these two years for these four race/sex groups. We choose to estimate wage equations separately for 1976 and 1985 rather than building longitudinal data sets for each race/sex group because one of the central purposes of this study is to investigate the impact of training on white/black wage differentials and how that impact has changed over time. This is best done by estimating separate year-specific wage equations. We recognize that others have demonstrated the advantage of using longitudinal data to control for heterogeneity bias [1; 6; 14]. We note, however, that our findings are similar to those obtained by other researchers who have controlled for heterogeneity bias using PSID data [6].

Table I contains definitions for all variables included in the empirical work. These definitions are identical for both sample years and across the four race/sex groups. As indicated in Table I, in the interests of space we report on only the relevant variables in the empirical section. The samples used to estimate the wage equations which follow are limited to individuals who are 18 to 65 years of age and who are employed for wages or salaried in the nonagricultural and nonmilitary sectors.

Definition of Training and Experience Variables

Since we focus on training to explain existing white/black wage differentials and how they have changed over time, the derivation of the training variables is of primary importance. We generally follow previous researchers in our definitions of both the training and the experience variables in the PSID [6; 9]. For example, the PSID asks respondents the following question: "On a job like yours, how long would it take the average new person to become fully trained and qualified?" The worker's answer to this question equals the training required for their current position (TT). Total current training equals required training if the worker's experience in his current position (EXPJ) equals or exceeds his required training. Otherwise, total current training is assumed to equal the worker's experience in his current position.

In addition to total current training (TOTALT), we also control for the job experience a worker has acquired since age 18. Job experience since age 18 is divided into three components: (1) job experience acquired in the worker's present position with his current firm (EXPJ), (2) job experience acquired in the worker's current firm prior to his present position (PEXPE), and (3) job experience acquired in previous firms (PEXP18) since age 18.

According to traditional human capital theory, training is either general in nature (i.e., it increases the productivity of a worker at any firm), or firm-specific (i.e., it enhances productivity only at the current firm). The human capital assumption that training enhances productivity, combined with a traditional assumption of diminishing marginal returns on any investment, implies a quadratic relationship between current training received by a worker and that worker's wage. The square of total current training and the three experience variables are included in the wage equations to allow for this possibility.

We recognize that current training alone will not account for the full impact that training has on wage growth. Workers with previous training should have higher productivity and earnings than workers without such training. Table I defines additional training variables which are used to control for the impact that prior training has on the returns to current training. For example, PSID respondents are asked: "Do you also have to get some work experience or special training to get a job like yours?" We assume that workers who answer yes to this question have received prior general training (PRVEXP = 1) while workers who answer no have not (PRVEXP = 0).

We also assume that workers who have worked with their current firm prior to their present position (PEXPE [is greater than] 0) have accumulated some firm specific on-the-job training. The dummy variable, PRIOR, measures whether workers have received such training; PRIOR equals one if workers report such previous experience and zero otherwise. We also use total required training (TT) as a measure of previous training; it controls for the possibility that workers reporting higher required training levels have also included previous training in their response.

The human capital model clearly indicates that controls for prior training are needed to correctly estimate the full impact of training on wages. In fact, the nature of the impact of prior training on wage growth depends on whether prior training is a substitute for or a complement to current training. If prior training is a substitute for current training, then the marginal returns to current training for workers who have prior training will likely be lower than marginal returns for those who do not. This result follows from the quadratic relationship between total training and earnings which suggests that a worker further out on the training curve (one with both current and prior training) will have higher total earnings but lower incremental returns.

If, on the other hand, prior training and current training are complementary, we would expect that the attainment of prior training would enhance the productivity of a worker currently receiving training. We would therefore anticipate higher marginal returns to current training among those with prior training as compared to those without such training. This case would be represented by a rotation upward of the earnings/training profile rather than by a movement along the profile.(5)

Table II. Percentage Growth in Selected Human Capital Variables: 1976 TO 1985

Variables             White Males   Black Males   White Females   Black
Females

NOMINAL HOURLY WAGE      79.14%        81.12%         85.34%          88.33%
REAL HOURLY WAGE(*)      -3.93%        -1.74%         -2.31%          -0.02%
EXPJ                     64.70%        51.34%         46.70%          52.19%
PEXP18                    5.14%       -10.85%         12.56%          -4.42%
PEXPE                   -50.49%       -66.33%        -36.84%         -28.82%
EXP18(**)                 3.19%        -8.30%         13.24%           6.15%
TOTALT                    1.70%        20.90%         17.48%          12.24%
TT                       -1.56%        26.86%         11.76%          13.66%
PRVEXP                   14.59%        44.97%         22.44%          34.30%
PRIOR                    -9.09%         4.25%         36.31%          21.50%
EDUC                      3.17%        16.60%          2.70%           9.24%
WEEKW                     0.69%         2.12%          6.46%           4.29%
SUPR                      0.64%        18.94%         13.10%          17.30%
PRTIME                   40.00%        62.50%        -25.26%          14.29%

* Deflated by the regional CPI-U: 1982 to 1984 = 100.

** EXP18 = EXPJ + PEXPE + PEXP18. Thus, EXP18 equals the worker's total job
experience since age 18.

Thus, previous training generally impacts wages through its impact on the returns to current training. As a result, we interact the prior training variables with total current training (TOTALT and TOTALT2) as did Brown [6] rather than including them separately in the wage equation. These interactions are easily understood. The estimated coefficients for total current training (TOTALT and TOTALT2) only fully measure the impact of training on wages for workers with no previous training. For workers with previous training, the interactions must also be considered. If the coefficient sign on the interactions with TOTALT is negative, this indicates that previous training decreases the marginal returns to current training. As a result, a negative sign indicates that prior and current training are substitutes. On the other hand, a positive coefficient for the interaction with TOTALT indicates that previous training increases the marginal returns to current training. This result suggests that previous and current training are complementary. In both cases, we would expect the opposite sign on the interactions with TOTALT2 as is found on the interactions with TOTALT.

A priori we hypothesize that while blacks and whites might have different levels of current and prior training, and, therefore, have differing returns to each, the underlying relationship between prior and current training should be the same for both groups. We anticipate that if prior and current training are substitutes they will be so for both blacks and whites and, if they are complements, likewise.

Data Trends

We are primarily interested in learning how differences in on-the-job training levels affect both the existence of the white/black wage gap and the changing nature of the gap over time. Table II presents summary statistics in the form of percentage growth rates in the human capital variables from 1976 to 1985 as one way of examining changes in the acquisition of training and experience for blacks and whites.(6)

Notice first that, while each race/sex group had negative growth in their real hourly wage, the magnitude of the real loss was least for blacks and women. Thus, we observe a slight narrowing of the real white/black wage gap occurring over the decade. Table II presents evidence which suggests that blacks and white females generally experienced higher rates of growth in many of the human capital variables over this decade than did white men. A narrowing of the gap between whites and blacks in their acquisition of human capital may well explain the narrowing of the real white/black wage gap.

For example, blacks and women made significant progress relative to white males both in the acquisition of current training (TOTALT) over the decade and in the acquisition of jobs which required higher levels of training (TT). Table II also demonstrates that blacks progressed relative to whites in educational status (EDUC) and in becoming promoted to supervisors (SUPR). Furthermore, blacks and females made similar gains relative to males in the acquisition of prior training (PRVEXP and PRIOR).

The same trend, however, is not evident when looking at the experience variables. Overall, either blacks experienced decreases in total job experience (EXP18) as was the case with black males or blacks experienced lower increases than whites as was the case with black females. For all race/sex groups current job experience (EXPJ) rose substantially. However, whites, and especially white males, experienced larger increases in current job experience. Furthermore, while all race/sex groups had less previous experience with their current employer (PEXPE) over the decade, this decline is more evident for blacks and men than it is for whites and women. This trend, could partially be explained by the fact that blacks are receiving more education and are, therefore, substituting education for experience. Additionally, blacks and men may have changed jobs more frequently in 1985 than in 1976. Dependent upon the explanation, this trend may have helped or hurt black progress toward wage equality.

Thus, the data trends demonstrate that blacks made substantial gains in the acquisition of human capital during the decade from 1976 to 1985. However, during the same time period the white/black wage gap narrowed only slightly in both real and nominal terms. In the remaining sections we present regression results from logwage equations, and decompose the logwage regression results, in an attempt to explain these general data trends.

III. The Estimation of Wage Equations

We estimate log wage equations for each of the four race/sex groups using the variables reported in Table I. These equations are estimated using a standard technique which corrects for sample selectivity originally proposed by Heckman [11].(7) The results of the year specific race/sex wage regressions are reported in Tables III and IV for 1976 and 1985 respectively. We generally find support for the traditional human capital model with nearly universal concave relationships TABULAR DATA OMITTED between our various measures of experience and wages and training and wages.(8) Further, it is generally the case that those with higher levels of human capital characteristics receive lower marginal returns than those with lower levels.

However, contrary to our expectations, we find substantial differences between blacks and whites in the estimated impact of previous training on the marginal returns to current training. As noted above, the impact of previous training is measured by interacting current training (TOTALT and TOTALT2) with three measures of previous training (PRVEXP, PRIOR, and TT). Tables III and IV contain the results from three F tests on the joint significance of both the current and previous training variables (the six training interactions). Current training is found to be a significant factor in the determination of wages for both blacks and whites. Test 1 jointly tests the significance of the two current training variables (TOTALT and TOTALT2) and is significant for every race/sex group except for black males in 1976.

However, the notion that previous training has no impact on black marginal returns to current training is supported by the F tests in these two tables. Test 3, which jointly tests the significance of only the training interactions, is never significant for blacks in either 1976 or 1985. Further, Test 2, which jointly tests the significance of both current and previous training, is only significant for blacks in 1985. The F tests clearly demonstrate, though, that previous training has a significant impact on white returns to current training. For example, Test 2 is significant in both 1976 and 1985 for both white males and females; Test 3 is significant (at least at the 10 percent level) for all white groups except for white females in 1976.(9)

The importance of these findings, especially as they relate to both black acquisition of training and to black wage inequality, should not be understated. According to the human capital model, a worker who has previously acquired human capital will be more productive and, as a TABULAR DATA OMITTED result, will command higher wages.(10) However, our findings suggest that human capital is only portable in this manner for whites. Knowing that their training will only enhance their wages in their current position with their current firm, blacks have less of an incentive to invest in on-the-job training. Further, the non-portability of training for blacks significantly reduces their mobility and, hence, would allow their current firms to limit black wages more than for the more mobile whites [19]. In addition, future black earnings may be adversely influenced by this lack of training portability.

The explanation for why prior training does not increase the earnings of blacks as it does for whites is not obvious. While explanations relating to the differential types of training received by blacks and whites, the segmented labor market hypothesis, and other potential reasons exist, discrimination is an additional explanation which should be considered.

IV. Decompositions of the White/Black Wage Differential

One of our primary objectives was to determine the impact of training differentials between blacks and whites on the white/black wage gap and to see how this impact has changed over time. We use the procedure proposed by Blinder [4] to decompose the white/black wage gap into the portion of the gap resulting from differences in the characteristics of blacks and whites and the portion resulting from different market valuations of those characteristics. The Blinder decompositions for men and women are presented in Table V. Table V indicates that whites received 39% and 27% higher wages than blacks in 1976 for men and women respectively. However, by 1985 the white/black wage gap had only fallen to 37% for men and 25% for women. Thus, we again find that blacks experienced little economic progress between 1976 and 1985.

In addition, Table V indicates that in 1976 67 percent of the male white/black wage gap is explained by differences in mean characteristics; 33 percent is explained by differences in the returns to those characteristics. Although larger, the same trends are evident for changes in female white/black wage gaps between 1976 and 1985. Although the white/black wage gap did not decrease much from 1976 to 1985, a substantially lower proportion of the gap was explained by TABULAR DATA OMITTED differences in the mean characteristics of blacks and whites in 1976 than in 1985. This lends some support to the vintage notion that the human capital characteristic gap is narrowing between blacks and whites. Even though blacks made little economic gains from 1976 to 1985, economic gains may be experienced in the future as firms respond to the lessening characteristic gap. This prediction should be considered tentative, however, since an alternative explanation would be that the response to increased black acquisition of human capital has been to lower black returns for those characteristics.

The outlook for future decreases in white/black wage gaps is still uncertain. To illuminate this issue we present two additional decompositions which are also based upon the regression estimates presented in Tables III and IV but are not the traditional Blinder decompositions discussed above. These additional decompositions are informative in that they help determine the reasons for the lack of black progress between 1976 and 1985 by answering two questions. First, what would happen to the white/black wage gap in each year if blacks received white means, returns or both? Table VI contains these decompositions for various subsets of variables.

Generally speaking, blacks, both men and women, would have made wage gains relative to whites had they possessed the mean training characteristics of white workers in both 1976 and 1985. An even larger impact would have been made if blacks had possessed the white levels of education in those two years and been employed in the occupations and industries of the average white workers. It is interesting to note that all of the categories of variables which would enhance black earnings the most are human capital related: training, education, occupation, and industry.

Overall, and especially for black men, possessing the white mean characteristics would have benefitted blacks more than receiving the white returns to their mean characteristics. The fact that variable means tend to impact on black earnings more than returns to those mean characteristics tends to support the vintage notion, and gives hope for future reductions in the white/black wage gap.

Second, we also ask what would happen to the 1976 white/black wage gap if both blacks and whites had their own 1985 means or returns for various subsets of variables but retained their 1976 means and returns for all other variables. These decompositions are contained in Table VII. One way of looking at the values in the "Means" columns is to interpret a negative value as indicating that blacks made relative gains on whites in that particular characteristic. For instance, the negative value for both males and females in the EDUCATION variable indicates that if in 1976 blacks and whites were both assigned their 1985 education mean the white/black wage gap would have narrowed. This suggests that black progress in education over the time period was more pronounced TABULAR DATA OMITTED TABULAR DATA OMITTED than black progress in general. This was also the case for both males and females in the occupation and industry variables and was very pronounced for males in the experience variables.

However, blacks made little or no progress from 1976 to 1985 with respect to training. In fact, the impact of training for females was to increase the white/black wage gap while for males it had little impact. We note that if blacks would have made any gains in training (or not lost ground in the case of females) between 1976 and 1985, we would have seen a potentially substantial decrease in the white/black wage gap. Thus, Table VII indicates that future black improvement in training acquisition and their returns to training is a source for potential wage gains.

V. Conclusions

We find evidence that both black males and females would be better off if they possessed the amount of training that white males and females have. This is particularly true in 1985 and its impact is stronger for males than for females. The white/black differential in returns to training actually work to the advantage of black males in both 1976 and 1985 and has little impact on females in either year.

This appears to suggest that differences in human capital acquisition is more of a concern than the market responses to those characteristics. This does not mean, however, that blacks are much better off in 1985 or, indeed, better off at all. In fact, the wage gap between blacks and whites narrowed very little from 1976 to 1985. It is potentially encouraging that the proportion of the gap owing to differences in the mean characteristics between blacks and whites declined from 1976 to 1985 for both men and women. This may suggest a narrowing of the "characteristic gap" between blacks and whites which would support the vintage hypothesis. However, the same results might be found if the characteristic gap narrowed slightly and there were at the same time an increase in the amount of market discrimination.

Our results definitely establish the importance of training and experience on the overall success of both blacks and whites. For example, the lack of portability of training for blacks is a key component in explaining the lack of movement towards wage equality for blacks. In addition, blacks made little or no progress in the acquisition of training over the decade nor did black training become more portable from 1976 to 1985. Our decompositions demonstrate that the lack of black progress toward wage equality is at least partially due to lack of progress in training acquisition. Further, we find that changes in training means and returns that occurred for both blacks and whites from 1976 to 1985 actually cause increases in the white/black wage gap for women. Thus, not only are blacks making little progress in training relative to whites, some blacks are, in fact, experiencing relative declines in training and in wages.

We do note that even greater progress for blacks will rely on changes in other key characteristics. For example, increases in black education as well as increased returns to black education will both significantly increase black wages and, hence, decrease the white/black wage gap. The empirical work also suggests that black advancement in wages would occur as blacks move to jobs in occupations and industries formerly dominated by whites. Thus, while efforts to increase the acquisition of training among blacks and to increase its portability for blacks, in an effort to close the wage gap between blacks and whites, will undoubtedly be helpful, they cannot be expected to be a cure all.

1. An excellent survey is provided by Jaynes [12].

2. See, for example, Smith [22; 21], Smith and Welch [23], and Freeman [10]. While these papers do not exclude the possibility of a decline in discrimination over time, they rely more heavily on a "vintage"-type explanation, where the closing of the wage gap is explained by older, less educated black cohorts leaving the labor force and being replaced by younger cohorts whose human capital characteristics more closely resemble their white counterparts. For an "anti-vintage" viewpoint, see Darity and Myers [8].

3. Lazear [13] was one of the first to show how important a role on-the-job training can play in influencing the black/white wage gap. Corcoran and Duncan [7] and Duncan and Hoffman [9] are two additional examples of previous work that dealt with the role of on-the-job in explaining the wage gap. Neither of these studies examines changes over time and both vary significantly from our model specification.

4. Brown uses the longitudinal nature of the PSID with a fixed effects model to test for the returns to training. We assume that there are returns to training and use two cross-sections from the PSID to investigate how those returns have changed over time and across demographic groups.

5. For a more detailed discussion of this point see Olsen and Sexton [16].

6. Additional summary statistics, including the mean values for all variables in each race/sex group, are available from the authors upon request.

7. Although there is some debate about the robustness of such procedures [15] we use this technique owing to its computational ease and note the general similarity between the results presented below and OLS results obtained without using the Heckman procedure.

8. In fact, a careful analysis of the regression coefficients, available from the authors upon request, shows that all workers, including blacks, exhibit substantial wage gains as a result of their current training. For example, consider only workers who have previously received training (i.e., workers for whom PRVEXP = 1 and PRIOR = 1), whose current jobs require 4 years of on-the-job training (TT = 4), and who have completed 2 years of training in their current job. In 1976, such workers experienced cumulative wage gains due to their current training of 15.7, 11.5, 30.7, and 23.3 percent for white males, black males, white females, and black females respectively. In 1985, cumulative wage gains due to current training for such workers were 17.0, 18.1, 34.1, and 23.8 percent for white males, black males, white females, and black females respectively.

9. The impact of previous training does vary for whites from 1976 to 1985. For further discussion on this point see Olsen and Sexton [16].

10. This is true for previous human capital that is general in nature and previous human capital that is specific to a worker's current firm. However, previous human capital that is specific to a prior firm would not be expected to enhance either productivity or wages at a worker's current firm and, hence, would not be portable for any workers.

References

1. Altonji, Joseph G., and Robert A. Shakotko, "Do Wages Rise with Job Seniority?" Review of Economic Studies, July 1987, 437-59.

2. Beller, Andrea H., "The Economics of Enforcement of an Antidiscrimination Law: Title VII of the Civil Rights Act of 1964." Journal of Law and Economics, 1978, 359-380.

3. Blau, Francine D. and Andrea H. Beller, "Black-White Earnings Over the 1970s and 1980s: Gender Differences in Trends." Review of Economics and Statistics, May 1992, 276-86.

4. Blinder, Alan, "Wage Discrimination: Reduced Form and Structural Estimates." Journal of Human Resources, Fall 1973, 436-55.

5. Brown, Charles, "Black-White Earnings Ratios Since the Civil Rights Act of 1964: The Importance of Labor Market Dropouts." The Quarterly Journal of Economics, February 1984, 31-44.

6. Brown, James N., "Why Do Wages Increase with Tenure? On-the-Job Training and Life-Cycle Wage Growth Observed Within Firms." American Economic Review, December 1989, 971-91.

7. Corcoran, Mary and Greg J. Duncan, "Work History, Labor Force Attachment, and Earnings Differences Between the Races and Sexes." Journal of Human Resources, Winter 1979, 3-20.

8. Darity, William A., Jr. and Samuel L. Myers, Jr., "Changes in Black-White Income Inequality, 1968-78: A Decade of Progress?" The Review of Black Political Economy, Summer 1980, 354-79.

9. Duncan, Greg J. and Saul Hoffman, "On-the-Job Training and Earnings Differences by Race and Sex." The Review of Economics and Statistics, November 1979, 594-603.

10. Freeman, Richard B. "Black Economic Progress after 1964: Who has Gained and Why?," in Studies in Labor Markets, edited by Sherwin Rosen. Chicago: University of Chicago Press, 1981, pp. 247-94.

11. Heckman, James J., "Sample Selection Bias as a Specification Error." Econometrica, January 1979, 153-61.

12. Jaynes, Gerald D., "The Labor Market Status of Black Americans: 1939-1985." Journal of Economic Perspectives, Fall 1990, 9-24.

13. Lazear, Edward, "The Narrowing of Black-White Wage Differentials is Illusory." American Economic Review, September 1979, 553-64.

14. Lynch, Lisa M., "Private-Sector Training and the Earnings of Young Workers." American Economic Review, March 1992, 299-312.

15. Manski, Charles F., "Anatomy of the Selection Problem." Journal of Human Resources, Summer 1989, 343-60.

16. Olsen, Reed Neil and Edwin A. Sexton. "Gender Differences in the Returns to and the Acquisition of On-The-Job Training." Working paper, Southwest Missouri State University, 1993.

17. O'Neill, Dave M., "The Effect of Discrimination on Earnings: Evidence from Military Test Score Results." The Journal of Human Resources, Fall 1970, 475-86.

18. O'Neill, June, "The Role of Human Capital in Earnings Differences Between Black and White Men." Journal of Economic Perspectives, Fall 1990, 25-45.

19. Ransom, Michael R., "Seniority and Monopsony in the Academic Labor Market." American Economic Review, March 1993, 221-33.

20. Sexton, Edwin A. and Janet F. Nickel, "The Effects of School Location on the Earnings of Black and White Youths." Economics of Education Review, March 1992, 11-18.

21. Smith, James P., "Race and Human Capital." The American Economic Review, September 1984, 685-98.

22. -----, "The Improving Economic Status of Black Americans." American Economic Review, May 1978, 171-78.

23. ----- and Finis R. Welch. Closing the Gap: Forty Years of Economic Progress for Blacks. Santa Monica, California: The Rand Corporation, 1986.