I demonstrate that the literature on the racial wage gap has systematically overstated the gains made by African American men by ignoring their withdrawal from the labor force. Three sources of selection-bias are identified: imposing sample selection criteria based on labor supply, trimming wages on the basis of real-dollar cutoffs, and making inferences based on Current Population Survey (CPS) data whose truncated sampling design excludes the growing incarcerated population. To recover the counterfactual distribution of skill-prices for non-workers, I implement a quasi-bounds estimator that does not require the use of arbitrary exclusion restrictions for identification and find that: (1) Corrected estimates of the racial wage gap indicate a substantial role for the efficacy of the Civil Rights Act and related initiatives in affecting convergence in segregated states; ignoring selection causes estimates of convergence in the South as well as the within-cohort component of this change to be understated. (2) In contrast to the sharp convergence observed in standard wage series from 1970-90, selectivity corrected estimates indicate complete stagnation over this period with a divergence of 3.5 to 6 percentage points between 1980 and 1990. Almost half of this divergence is missed through the exclusion of the incarcerated population. The selective withdrawal hypothesis can explain 85 percent of the observed convergence between 1970 and 1990 and 40 percent of the 1960-90 convergence. (3) The disproportionate presence of highly skilled blacks in the armed forces (who are also excluded from CPS analysis) causes estimates of the racial gap to be overstated by 1 to 2 percentage points. (4) The relative increase in non-participation is a supplyside effect driven more by a massive increase in reservation wages for blacks at the bottom of the skill distribution, than by falling offer wages. (5) The significant gains made by black men during the 1960s and 1970s occured almost exclusively in the bottom offer wage decile, where significant numbers of black men were pushed out of the lowest white wage decile into higher quintiles. These gains constitute the primary location of black economic progress in the latter half of the 20th century. Amitabh Chandra Department of Economics 6106 Rockefeller Hall Dartmouth College Hanover, NH 03755 and NBER email@example.com In a highly influential paper, Richard Butler and James Heckman (1977) argued that expansions in the generosity of transfer programs over the decade of the 1960s had induced lower-skilled men to withdraw from the labor force. Because African-American men were more likely to be lower-skilled, observed relative wages would increase. Therefore, a preoccupation with the wages of workers would cause social scientists to overstate the success of Title VII Legislation, or spuriously conclude that discrimination against blacks had declined. This hypothesis was used to demonstrate that Richard Freeman’s landmark paper in (1973) was not consistent with the Civil Rights Act (CRA) raising the relative demand for black labor: whereas Freeman found a significant effect of Equal Employment Opportunity Commission (EEOC) expenditures on relative wages (at a time when EEOC budgets were small), Butler and Heckman argued that the selective withdrawal hypothesis could also rationalize the data.1 At the time of their writing, Butler and Heckman could not have anticipated the phenomenal increase in the returns to skill that would occur in the 1980s; a factor which would cause withdrawal to the extent that reservation wages were relatively fixed over this period. Nor could they have predicted the massive growth in the US prison population as a result of the “war on drugs” and the related Sentencing Reform Act of 1984 which introduced mandatory and longer sentencing guidelines for drug-related convictions. Together, these factors could generate convergence in observed wages since they disproportionately affect low-skilled blacks. Given that much of what is known on the convergence in the racial wage gap is based on a selected sample of workers, and even within that sample a group that typically meets additional criteria, it is important to understand the magnitude of possible biases that result from such sample-selection restrictions. Establishing the empirical magnitude of this hypothesized effect is the primary goal of this paper.2 Additionally, I seek to understand the degree to which ignoring nonemployment has contaminated the measurement of factors such as schooling levels, school-quality, and discrimination in affecting the convergence. Finally, this paper decomposes the extent to which supply shifts vs. demand side forces have 1 It is beyond the scope of this paper to review the enormous literature on the passage of the 1964 CRA and the related Voting Rights Acts of 1962 and 1965. For an introduction to this subject see the National Research Council commisioned volume A Common Destiny: Blacks and American Society [Jayes and Williams (1989), Chapter 6], and the rigorous reviews by Brown (1982) and Donohue and Heckman (1991). Briefly, Title VII of the CRA, which forbade discrimination in employment passed in 1964 and went into effect on July 2, 1965. Simultaneously, President Johnson’s Executive Order 11246 in 1965 formed the Office of Federal Contract Complicance (OFCC), which oversaw anti-discrimination efforts in government contracts. 2 In the spirit of this thesis, Katz and Krueger (1999) study the possibility that the 2.6 percent fall in the unemployment rate between 1985 and 1998 was a compositional effect that was driven by growing incarceration rates. Under alternative estimates of what the counterfactual labor force participation rate would be, they estimate that the true fall in the unemployment rate would have been between 2.1-2.5 percent.