Black–White gap in self-employment. Does intra-race heterogeneity exist?

See Bates (1987), Borjas and Bronars (1989), Meyer (1990), Fairlie (1999), Fairlie and Meyer (1996, 2000), and Fairlie and Robb (2007).

Bates (1987), Bogan and Darity (2008), Meyer (1990), Fairlie (2005), Fairlie and Meyer (1996), and Fairlie and Robb (2008).

Note that all the groups we focus on are U.S.-born and not immigrants.

What we mean by “race does not matter per se” is that being Black is not a true predictor of lower self employment in the United States, as subgroups of U.S.-born Blacks share similar probability of self-employment as U.S.-born Whites. Instead, the dummy variable Black is a proxy for some other variable not measured by the econometrician but relevant for a portion of U.S.-born Blacks. We suggest that the Black dummy proxies for a past experience which a subset of Blacks were exposed to.

Differences in preferences, wealth, education, language proficiency, parental exposure to self-employment, size and efficiency of social networks, earnings in the wage sector relative to self-employment sector, discrimination in lending and consumer discrimination have all been shown to impact the relative probabilities of self-employment for minorities relative to Whites.

Evans and Jovanovic (1989), Bates (1997), Fairlie (1999), Holtz-Eakin and Rosen (2005), Dunn and Holtz-Eakin (2000), Blau and Graham (1990), and Gittleman and Wolf (2004).

For a more broad definition and impact of resource constraints, see Kim et al. (2006).

Although the CPS has ample information that could serve as potential controls in our estimation, it is important to mention that we cannot control for some factors that prior research has shown to affect self-employment but are not available in the CPS.

It is important to mention that the main inferences from our paper of heterogeneity in the self-employment probability among subgroups of Blacks still holds if we focus on the data period 2003–2010. However, the estimates for the subgroups change.

Our results are not dependent on using a probit model. In an earlier version of the paper we also employ logit and linear probability models. Given that our results do not change with either of these models, we were advised by a referee to take out both models from the paper.

For race dummy variables, the interpretations of marginal effects are slightly different. Estimates capture the change in the probability of self-employment for a particular racial group relative to the baseline group.

As mentioned earlier, using interest income as the wealth proxy provides a larger data sample and encompasses more dimensions of wealth. However, we acknowledge that both proxies are imperfect measures of wealth, yet represent the best choices for the data set being used.

Immigrants could have a different experience/exposure because they come to the U.S. with their home country experience (norms, culture, values).

Note that the group AAUBP includes those whose grandparents are foreign born.

It is important to mention that our main results do not change even if we maintain the four-group breakdown.

Notice that the results using the second wealth indicator seem to show that AAUBP and AAFBP share similar probabilities of self-employment. However, this result is not robust and disappears when further controls are included. Also, as mentioned previously, the sample of those with dividend income is small and suffers from selection bias because individuals select into owning stocks, while most adults own a bank account. Given the selectivity issue with this indicator, we are more apprehensive about results with this control and our preferred result uses the interest income indicator.

It is important to also mention that Foreign Black (FB) and Naturalized American Black (NAB) are the only other groups of Blacks with lower probabilities of self-employment than that of the base group at 2.8 and 2.7%, respectively. However, both these groups have a smaller gap than AAUBP. Though we also present the results of all immigrant groups, we do not discuss the results for all non U.S.-born groups because it is harder to properly control for wealth for this group, since we may not observe fully their savings in the United States. Immigrants, especially those who are not naturalized, are more likely to save both in the U.S. and their home country. In addition, given the constraints this group faces as immigrants that could affect self-employment and cannot be controlled for, it would not be useful to compare them with the U.S.-born. For both of these reasons, estimates for immigrants are more likely to be biased.

We do not worry about WAFB-mix also showing cohort differences because this result may not be robust. As a robustness check, we also test for the effect of cohort on self-employment probability by including an interaction between the cohort and the race dummies. Using this method, we find similar results as those summarized in Table 5. The coefficient on the interaction term between the AAUBP and cohort dummies is significant and positive but insignificant for all other groups including WAFBP-mix. The inference from this analysis suggests that belonging to the younger cohort matters more for the probability of self employment for AAUBP, in comparison to WAUBP.

Another possible related variable is institutional exposure or the experiences of a subgroup of people which can create differences.