New Color Lines: Racial/Ethnic Inequality in Earnings among College-Educated Men

Abstract

Using the 2003 National Survey of College Graduates, this study examined four perspectives on new color lines in America—white–nonwhite, black–nonblack, tri-racial, and blurred—among college-educated white, black, Hispanic, and Asian men. Findings show that the color lines have not been consistently drawn but vary by nativity and migration status. Among the native born, the color line for earnings cuts mainly across white and nonwhite when field of study and Carnegie classification are controlled for in addition to other covariates. On the other hand, among members of the 1.5 generation, who obtained both their high school and highest degrees in the United States, the lines are most salient between black and nonblack. Among first-generation immigrants, who completed all their education in a foreign country, and 1.25-generation immigrants, who obtained their high school diploma in a foreign country but earned their highest degree in the United States, there is a gradation of the color line with whites at the top and blacks at the bottom. Despite these mixed results, blacks fall consistently at the bottom of the racial hierarchy and whites at the top, regardless of nativity and migration status. Implications of the findings are discussed.

ACKNOWLEDGMENTS

For helpful comments on this article, I am grateful to the editors, five anonymous reviewers of TSQ, Pyong Gap Min, and Samuel Lucas. I also thank the Institute for Policy & Social Research at the University of Kansas for their excellent research support. An earlier version of this article was presented at the 2013 annual meeting of the American Sociological Association.

NOTES

Notes

¹ All figures on racial composition in 1960 are from the author's calculations, for which he used the Integrated Public Use Microdata Series (Ruggles et al. 2010).

² Over this period, the proportion of whites with college degrees or higher increased from 0.113 to 0.303; for blacks, the increase was from 0.044 to 0.198, and for Hispanics it was from 0.045 to 0.139. Among men age 25 to 64 (the sample criterion for our study), 32.5 percent of native-born whites, 16.1 percent of native-born blacks, 17.8 percent of native-born Hispanics, and 51.6 percent of native-born Asian Americans were college educated as of 2003 (the author's own computation using the American Community Survey). Among immigrant men, 46.4 percent of whites, 30.6 percent of blacks, 11.5 percent of Hispanics, and 54.9 percent of Asians were college educated.

³ The 1930 Census listed “Mexican” as a separate race for the first and only time. In the 2010 Census Race and Hispanic Origin Alternative Questionnaire Experiment, it is recommended that persons of Hispanic, Latino, or Spanish origin be listed as belonging to separate races in Census questionnaires (U.S. Census Bureau 2012).

⁴ See Table A1 in Appendix A for specific sample sizes for the various groups and the numbers of cases dropped because of various sample restrictions.

⁵ To avoid the complexities surrounding nativity and where respondents were educated, we dropped a small number who are native born but earned their high school or highest degrees from foreign educational institutions. Including them in the analyses did not change the results.

⁶ The 0.565 correlation coefficient is observed only when parents' levels of education are treated as continuous variables. When dummy variables are used, none of the correlation coefficients of all dummy covariates controlled for in all models is higher than 0.40, and most of them are less than 0.10.

⁷ Among all the other control variables, only age and age-squared yield VIF values greater than 5. In spite of that, both age and age squared are statistically significant. When the high VIFs are caused by the inclusion of powers of other variables (e.g., age squared), the problem of multicollinearity can be safely ignored (Allison 2012). Also note that the multicollinearity issue arises because when variables are multicorrelated, the variance of their estimated coefficients is inflated, which tends to make the coefficients statistically nonsignificant. If the estimated coefficients are statistically significant, the effects of multicollinearity are mitigated. Large sample sizes decrease standard errors, thus making the coefficients more likely to be statistically significant. As sample sizes increases, multicollinearity becomes much less problematic. Indeed, “Worrying about high degrees of correlation among the independent variables in the sample is really no different from worrying about a small sample size” (Wooldridge 2003:98). Given the large sample size of the NSCG, the results of our study are not likely to be burdened by a multicollinearity problem, despite the fact that we controlled for many dummy variables. Furthermore, when the variables with high VIFs are dummy variables that represent a categorical variable with three or more categories which is the case for most of control variables in our models, multicollinearity can be safely ignored (Allison 2012).

⁸ Check the National Science Foundation's NSCG survey Web site (http://www.nsf.gov/statistics/srvygrads/) for a complete list of Carnegie classifications, fields of study, and occupational categories.

⁹ Except for Table 5, we do not present the results of the sensitivity tests. All results not shown here can be obtained from the author upon request.

¹⁰ The differences between native-born citizens and immigrants (naturalized citizens or permanent residents) are estimated by the coefficients of generational status. Therefore, legal status in our study is relevant only when there is a difference between naturalized citizens and permanent residents. We also note that the effects of legal status are statistically significant when immigration status is not controlled.

¹¹ The importance of African immigrants in shaping American color lines is growing, as evidenced by the fact that the proportion of immigrants in the African-American population grew from 0.7 percent in 1960 to 13.8 percent in 2006 (Sakamoto et al. 2010).