Second-Generation Outcomes of the Great Migration

We occasionally refer to the “South” and “non-South,” using the latter to refer to all states outside of the census-defined South. In some cases, we distinguish between the “North” and “West” as distinct nonsouthern locations. When we use the generic terms of “North” and “northern,” without additional distinction, we are referring to all areas outside the South.

We borrow this term from Portes and Rumbaut (1990).

The study by Restifo et al. (2013) is based on public use microdata samples for 1910, 1920, and 1930 for New York City only. They defined the second-generation as those born to one southern-born parent, and the third-generation as those born to two northern-born parents. Their data imposed a number of important limitations, including (1) a lack of early life characteristics for second-generation migrants and (2) an inability to clearly delineate when individuals migrated. Further, females and southern-born whites are not included in their study.

The Minnesota Population Center and Ancestry.com (2013) provided the complete-count 1940 census.

We did not extend our analysis to the 2001–2015 American Community Surveys (ACS) because we believed this would exaggerate selection into the matched sample that occurs by requiring survival between 1940 and 2000. Longevity is positively associated with higher education outcomes (Lleras-Muney 2005) and income (Chetty et al. 2016). Given the age of our sample, the relationship among education, income, and mortality may lead to our estimating upper bounds for the benefits experienced by second-generation migrants.

See Alexander et al. (2014) and Wagner and Layne (2014) for more information about PVS and how PIKs were assigned to the 1940 and 2000 census files.

Because the 2000 census long form includes roughly one in six U.S. households, the maximum number of children with PIKs from 1940 who could be linked to their 2000 census records is 4,526,938 (i.e., 0.167 × 27,107,415 = 4,526,938). In principle, we successfully linked roughly 70 % (i.e., 3,169,843 / 4,526,938 = .70) of the children from the 1940 census who had PIKs assigned. Because we do not account for migration or death in these estimates, a match rate of 70 % is a worst-case scenario, or lower bound, on how many individuals we should link from 1940 to 2000.

We follow a reweighting procedure similar to that proposed by DiNardo et al. (1996) to reweight the matched sample to exhibit the distribution of characteristics of the full population. We construct the weights using probit regressions of matched status on characteristics in 1940, including migrant group, age, parental education, parental occupational standing, and whether the respondent lives with one or both parents (results available by request). We also include polynomials in age, parent education, and parent occupation score up to the fourth degree. Due to computational constraints, we took a 10 % random sample of the 31 million total black and white children in the United States to reweight against. Our probit regressions estimate the propensity of being matched or unmatched (p). We then reweight the matched cases by (p / (1 – p) × (1 – q) / q), where q is the share of matched cases. Only three point estimates are statistically different from their original values as a result of the reweighting exercise. These include the coefficients on white northern stayers for the probability of graduating high school (without covariates), the coefficients on whites who migrated before birth and white northern stayers in the income regressions (without covariates), and the coefficients on migrated before birth in the probability of being in poverty regressions (without covariates) for whites.

Approximately 0.5 % of the total cases have nonmatching race responses in 1940 and 2000. Because this is not a sufficient number of cases to support a “both” racial category in the analysis, we exclude the cases.

We also examined whether migrants who moved to the West differed from those who moved to the North (Midwest and Northeast). However, we found few significant differences between these groups in our multivariate models, and so those comparisons are not examined here.

We estimated our models while both including and excluding southerners who migrated out of the South between 1940 and 2000 from the group of southern stayers. However, those results yielded very similar results. In fact, our decision to include in our primary analyses individuals who may have migrated out of the South between 1940 and 2000 slightly attenuates the advantages enjoyed by black migrants.

Income includes wages, salaries, commissions, bonuses, tips, self-employment income, public assistance, Social Security income, and Supplemental Security Income.

We control for the top five destination cities for both black and white migrant groups to control for network effects caused by previous migration streams as well as location-specific differences in income and employment opportunities. The top five cities for blacks are Detroit, Chicago, St. Louis, Philadelphia, and New York City. The top five cities for whites are Chicago, Los Angeles, Cincinnati, Detroit, and Cleveland.

We measure occupational status using occupation scores constructed using median income by occupation from the 1950 census (Sobek 1995).

We also examine the relationship between migration status and educational outcomes using OLS regression, with years of educational attainment as the dependent variable. We find that the returns to migration are somewhat narrower for black second-generation migrants using this specification. However, the results are substantively very similar to the models examining blacks’ and whites’ probabilities of graduating from high school, with second-generation black migrants exhibiting modestly higher years of educational attainment relative to southern stayers, and white second generation migrants exhibiting statistically nonsignificant differences in educational attainment relative to white southern stayers after we control for the 1940 characteristics. We present these results in Table S5 and Fig. S1 in Online Resource 1.