ABSTRACT
The study assesses the size and nature of the gap in incidence of employer-paid job training between Hispanic and non-Hispanic white workers. Using data on employer-paid training from the Survey of Income and Program Participation, Oaxaca/Blinder decompositions are estimated and analysed. The study finds that non-Hispanic white workers are more likely to engage in training than their Hispanic white counterparts, and that about half the difference in training incidence can be explained by average differences in educational attainment between the two groups. Besides educational attainment, being foreign born is another strong predictor of low training probabilities. The inability to explain a substantial part of the training gap suggests an economy-wide problem with human capital discrimination that leads to less job training among Hispanic whites compared to their non-Hispanic white counterparts. Policy efforts to increase formal education and reduce discriminatory behaviour both are consistent with the empirical findings.
Disclosure statement
No potential conflict of interest was reported by the author.
Notes
1 The multi-variate analysis later in the study focuses on the incidence of employer-paid training as the dependent variable. Results not presented in the paper reveal quantitatively and qualitatively similar findings when the most comprehensive measure of training is used as the dependent variable.
2 Frequency weights were used to arrive at estimates that more closely approximate characteristics for the underlying population.
3 The estimation does not control for industry and occupation variables. To add such controls would risk adding bias to the estimate on ‘Hispanic’ to the extent that some of the lack of integration or discriminatory animus manifests as occupational or industrial crowding. Nevertheless, specifications that control for industry and occupation yield qualitatively similar results as those reported in .
4 The Oaxaca/Blinder decompositions were estimated using the ‘Oaxaca’ command in Stata. For details please see Jann (Citation2008).
5 The results may be slightly different from those in Table 1 because frequency weights were not used when estimating the multi-variate models.