Explaining Black-White Wage Convergence, 1940-1950: The Role of the Great Compression

Robert A. Margo

Department of Economics

Vanderbilt University

Between 1940 and 1980 the ratio of weekly wages of black males relative to white males increased by 29 %age points. Recent work by Donohue and Heckman indicates that black-white wage convergence did not occur at a continuous pace between 1940 and 1980, but rather was concentrated in two episodes: the 1940s, and 1964 to 1975. Post-1964 wage convergence has been attributed to racial convergence in the quantity and quality of schooling, and to the impact of federal anti-discrimination legislation enacted as a result of the Civil Rights Movement, while studies of wage convergence in the 1940s have emphasized black migration from the South, where wages were far below the national average. Wage convergence in the 1940s, however, may also have been aided by a set of forces unique to the decade that produced a marked erosion of wage differentials between skilled and unskilled labor -- the so-called "Great Compression". Because black workers were, on average, less skilled than white workers in 1940, wage compression would have benefitted blacks disproportionately. Although previous studies have noted this possibility, there has been no systematic attempt to quantitatively assess the effects of decade- specific shifts in the structure of wages on black-white wage convergence in the 1940s.

Using a decomposition technique invented by Chinhui Juhn, Kevin Murphy, and Brooks Pierce, I examine the sources of black-white wage convergence in the 1940s, based on samples drawn from the public use tapes of the 1940 and 1950 censuses. Wage compression contributed to racial convergence throughout the wage distribution, with a slightly greater impact observed in the upper tail among blacks. Computed at the sample means, between half and two-thirds of black- white wage convergence can be attributed to the general changes in wage structure induced by the Great Compression.

The Great Compression of the 1940s produced a substantial narrowing in wage differentials in the United States. I present a brief discussion of the quantitative dimensions of this economic phenomenon and the factors causing it. Readers desiring additional evidence and a fuller treatment should consult the paper by Goldin and Margo (QJE, February 1992). Goldin and Margo demonstrate that wage compression took place at both tails of the wage distribution during the 1940s. The gap between the median wage and the 90th percentile fell by -0.15 in logs, and the gap between the 10th percentile and the median decreased by -0.12 in logs. Consistent with these patterns, the average earnings of less-skilled and less-educated workers increased relative to the average earnings of skilled and educated workers, as indicated by the decline in the ratio of average weekly wages of college and high school graduates and in the earnings of white-collar workers relative to the non-farm average. The Great Compression, however, was not solely a narrowing of mean wage differentials between education or occupation groups; wage compression also occurred within groups.

Although their relative significance is a matter of debate, the factors behind the Great Compression are not difficult to identify. Some portion of the compression occurred early in the decade in response to wartime shifts in labor demand and of government regulation of the wartime economy. As the war progressed, however, firms could not simply bid up the relative wages of workers in short supply, because of wage and price controls. The National War Labor Board (NWLB), established in 1942, was responsible for approving all wage increases. Analysis of case studies performed by the Bureau of Labor Statistics before, during, and after World War Two reveals that, with a few exceptions, the NWLB did indeed compress the wage structure at its left tail. The data also demonstrate, however, that compression took place in the right tail of the wage distribution during the war and, most importantly, continued immediately after the war. Postwar compression was aided by a rising federal minimum wage, strong unions, and an unexpectedly large increase in the relative supply of educated workers fueled by the GI Bill of Rights, which subsidized college attendance by veterans. Despite these forces, a shift in labor demand towards better-educated workers had become evident in the early 1950s, and by 1960 the effects of Great Compression were partly reversed.

My empirical analysis derives from samples of black and white adult males from the public use micro-data samples (PUMS) of the 1940 and 1950 censuses. Both public use samples are large random samples of the population in their respective census years. To be included in the analysis, an individual had to be between the ages of 25 and 64, and a wage or salary earner who worked at least 40 weeks in the year prior to the census and who earned at least one-half of the federal minimum wage on a weekly basis. Following Goldin and Margo, the earnings of individuals whose reported wage and salary income exceeded the highest census earnings level (so-called "top-coded" individuals) were estimated by multiplying their reported earnings by 1.4.

The limitations of the above sampling criteria should be noted. In particular, the restriction to wage and salary workers eliminates the vast majority of persons in agriculture, relatively more of whom were black than white in 1940. Unfortunately, the restriction to wage and salary workers cannot be relaxed, because the 1940 census did not report sources of income other than wages and salaries.

I first compute the extent of racial wage convergence in the samples at various locations in the wage distribution. At the 10th percentile the racial gap in log wages actually widened over the decade. At all other percentiles, however, the gap fell, with the greatest convergence observed in the upper tail of the black wage distribution (50th-90th percentiles).

The earlier discussion that the Great Compression contributed to racial wage convergence between 1940 and 1950. To investigate this possibility, I first present race-specific calculations of interdecile ranges. The effects of the Great Compression are clearly evident among whites, particularly in the lower tail of the wage distribution; the gap in wages between the 10th and 50th percentiles fell by 0.236 in log terms between 1940 and 1950 while the gap in log wages between the 50th and 90th percentiles fell somewhat less, by 0.097. Wage compression also occurred among blacks, but only in the upper tail of the black wage distribution. The gap in black wages between the 10th percentile and the median was larger in 1950 and 1940, so much so that the spread between the 10th and 90th percentiles was only slightly smaller (0.029 in logs) in 1950 than in 1940.

Based on these findings, a prima facie argument that wage compression was a factor behind racial wage convergence could be made if the upper tail (50th-90th percentiles) of the black wage distribution overlapped with the lower tail (10th-50th percentiles) of the white wage distribution. Accordingly, I locate the race-specific percentiles, measured in terms of the number of standard deviations (î) from the median white wage (î refers to the year-specific standard deviation of white wages). In 1940 the median black wage fell 1.282 standard deviations below the median white wage -- almost exactly the distance in standard deviations between the 10th and 50th quintiles of the white wage distribution -- while the black wage at the 90th percentile fell -0.193 standard deviations behind the white median. In 1950 the upper tail of the black wage distribution continued to overlap the lower tail of the white wage distribution, except at the 90th percentile where the black wage exceeded the median white wage by a small amount (0.037 standard deviations). Although the evidence is consistent with the hypothesis that wage compression served to reduce the wage gap between white and black workers, it does not measure the quantitative impact per se of shifts in the structure of wages. To measure this impact, I employ the decomposition technique developed for this purpose by Juhn, Murphy and Pierce.

The first step in the Juhn-Murphy-Pierce procedure is to estimate a standard log wage regression for white workers:

yt = Xt§t + åtît[1]

where yt = log of weekly wages for person i in year t (the subscript i is suppressed), Xt = characteristics of person i in year t included in the regression, §t = "prices" (i.e. regression coefficients) of the characteristics X, ît = "standardized" residual and åt = standard deviation of the (unstandardized) residuals. Also, let a "d" in front of a variable - - for example, dyt -- indicate the difference between blacks and whites, computed at some location in the wage distribution or at the sample means. Suppose that equation [1] has been estimated for both 1940 and 1950. I compute the following predicted wage for each individual, black and white, in 1950, y1:

y1 = X50á40 + å40î50[2]

The difference between y1 and y40 is made up of two terms

y1 - y40 = (X50 - X40)á40 + å40(î50 - î40)[3]

The first term on the right hand side measures the effect of changes in "observable" quantities (the X's). The second term measures the effect of movement up (î50-î40>0) or down (î50-î40<0) the residual distribution, valued (in log wage terms) in the base year (å40). Therefore, the difference between blacks and whites

dy1 - dy40 = (dX50 - dX40)á40 + å40(dî50 - dî40)[4]

measures the impact of racial convergence (or divergence) in observable quantities and relative (compared with whites) movement by blacks in the residual wage distribution. Following Juhn, Murphy, and Pierce, I label the first term on the right hand side of [4], "X's" and the second term, "Gap".

Next, I compute a second predicted wage, y2

y2 = X50á50 + å40î50[5]

The difference between y2 and y1 (= X50(á50-á40) measures the impact of changes in observable prices (the regression coefficients) on wages. Hence dy2 - dy1 measures the impact of changes in observable prices on racial wage differentials. I label this term, "Prices", in the decomposition.

The difference between y50 and y2

y50 - y2 = î50(å50 - å40)[6]

measures the impact of compression (å50-å40<0) or widening (å50-å40>0) in the residual distribution. Thus, the final term in the decomposition, which is labeled "Residuals", is dy50 - dy2; it measures the effect on the wage gap of a change in the residual wage distribution among whites, holding fixed the relative quintile positions of black workers. By convention, the Prices and Residual terms are what is meant by shifts in wage structure.

The effects of changes in wage structure -- the prices and residual terms -- are evident throughout the wage distribution; indeed, had only the wage structure changes occurred, the black- white wage gap at the 10th percentile would have fallen, rather than increased. Although the relative impact varied somewhat through the distribution, compression in observed prices (the regression coefficients) was three or four times as important as compression in the residuals. Overall, at the sample means, wage compression accounted for 52 to 62 % of black-white wage convergence between 1940 and 1950, depending on the base year.

Because the numerical effect of wage structure changes was similar throughout the wage distribution, variation in wage convergence across percentiles was due primarily to observable quantities (X's) and by movement by blacks up the residual white wage distribution (the Gap term). At the 10th percentiles the impact of observable quantities served to widen the racial wage gap; at the median and upper quartile, blacks succeeded in narrowing the gap with whites in observable quantities, thereby causing substantial wage convergence. At the sample means, however, the impact of changes in observable quantities was more modest, accounting for 22 to 30 % of the narrowing in the racial wage gap.

Except at the extreme left tail, the Gap terms are positive: blacks moved up the percentiles of the distribution of residual wages among whites. Regardless of the base year, the magnitude of such movement was greater above, than below, the median. At the sample means, the Gap term accounts for approximately 17 % of wage convergence, regardless of the base year. Studies of racial wage convergence since 1960 have emphasized the importance of racial differences in the school characteristics, such as the length of the school year, teacher-pupil ratios, or per pupil expenditures. It seems unlikely that changing racial differences in school characteristics could have played a similar quantitative role prior to 1950 because the pre- 1950 pattern of change was not one of monotonic convergence. In particular, racial differences in school characteristics actually widened for blacks born in the South between 1886 and 1910. Racial differences in school characteristics did decline somewhat among cohorts born in the South after 1910, but a substantial narrowing was not apparent (among labor market participants) until after 1960. Still, it is possible that changes in school characteristics could account for a portion of the Gap term even in the 1940s, if for no other reason that the numerical representation of the 1886-1910 southern cohorts in the black sample fell from 64 % in 1940 to 44 % in 1950.

A defensible estimate of the effects of school characteristics can be made by redefining black educational attainment downward relative to white educational attainment, and re-computing the decomposition. In particular, I shall assume that, for blacks born in the South before 1910 completing more than five years of schooling, the appropriate reduction is four years; for five years or less, the appropriate reduction is to one year of schooling. Thus, for example, a black man with eight years of schooling is assumed to have received an education functionally equivalent to that of a southern white with four years of schooling. For blacks born in the South after 1910, I assume that the corresponding reduction is three years. As a result of this adjustment, racial convergence in school quality will occur as older black cohorts are replaced by younger cohorts. No adjustment for school quality is made for northern-born blacks or for whites, regardless of region.

Measured at the sample means, the adjustment for school quality can explain about three- quarters of the Gap term; the explanatory power is somewhat less at the median or 75th percentile, but still substantial. Adjusting for school quality also increases the quantitative significance of the wage structure terms (the sum the prices and residual terms) but the magnitude of the increases is small. Based on these results, I conclude that adjusting for school quality can account for the majority of the Gap term, except perhaps at the median. Put another way, virtually all of the narrowing in the racial wage gap at the sample means can be explained by a narrowing of racial differences in observable quantities (including school quality), prices, and within-group residual inequality (the residual terms). The remainder of the gap term is presumably due to changes in (unobserved) quantities other than school quality and possibly to a decline in racial wage discrimination over the decade.

I have demonstrated that the Great Compression increased the relative wages of adult black males in the short run (i.e. the 1940s). The compression may also have helped in the long run. Recent research indicates that the educational achievement of black children during the first half of the 20th century was a positive function of their parents' economic status. By (temporarily) boosting the economic status of black parents, the Great Compression may have increased the schooling levels of black teenagers beyond what would have occurred otherwise in the 1940s. If so, these generations would have been better equipped to take advantage of government anti- discrimination efforts that enhanced the demand for educated black labor in the 1960s.

To investigate the relationship between schooling levels of black teens and parental earnings, I estimate equation [7], using data from the 1940 PUMS:

ln (S/S*) = Xë + î[7]

where S = schooling level (highest grade completed); S* = "expected" schooling level for a person of age t; X = determinants of schooling, including weekly wages of fathers; ë = regression coefficients, and î = random error. The expected schooling level is the modern reference standard: individuals enter the first grade at age six and complete each grade in a year's time. Negative values of ln (S/S*), therefore, indicate (in log terms) age-in-grade retardation (for those still in school) or dropouts.

The regression results confirm a positive and statistically significant association between child schooling and adult earnings: the elasticity between ln(S/S*) and the father's weekly wages is about 0.1. I use this coefficient to predict how much larger the average schooling level would have been in 1940, had the increased in adult earnings associated with the Great Compression. For this purpose, I use the gain in earnings unadjusted for school quality (0.114). The predicted increase in ln(S/S*) is 0.011 [= 0.114 x 0.098]. Compared with the mean schooling level among black teens in 1940 (-0.331), the impact of the Great Compression was very small. Compared with the increase in schooling levels over the decade, however, the impact would have been larger. Unfortunately, because of a peculiarity of the 1950 PUMS it is impossible to create a matched sample of black teens (whose fathers were full-time wage and salary workers and whose mothers were present) in both years. For all black teens ages 14 to 19, the value of ln(S/S*) rose by 0.11 between 1940 and 1950. Using this figure as a base, the Great Compression can explain 10 % [=0.011/10] of the growth in black schooling levels between 1940 and 1950.

Like the experience of the post-1975 period, the 1940s illustrates the sensitivity of black- white earnings differences to shifts in the overall structure of wages. In the 1940s, however, shifts in wage structure served to narrow racial differences in earnings. By narrowing wage differences between skilled and unskilled workers, and by compressing wages within occupations and other labor market groups, the Great Compression led to greater racial wage convergence between 1940 and 1950 among adults than would have occurred otherwise. By indirectly raising the schooling levels of black teens, the Great Compression also contributed to racial wage convergence in subsequent decades, although its effects here were more modest.

The results of this paper could be extended. In particular, census data cannot determine the exact timing (within the 1940s) of the changes documented in the paper, nor can census data reveal the precise institutional mechanisms at work (for example, the NWLB vs. the minimum wage vs. unions). Further analysis of wage studies performed by the Bureau of Labor Statistics, specifically of industries that employed large numbers of blacks, may shed light on both issues.