The legacy of immigration: labour market performance and education in the second generation

Abstract

Previous research finds that the children of immigrants, or the second generation, earn at least as much as other native born but that there are persistent ethnic differences in the intergenerational transmission of education and wages. We explain why these results are not incompatible and extend the empirical evidence in several directions using the Canadian Survey of Labour and Income Dynamics. First, we estimate a model of wages, earnings and hours worked using modern econometric techniques to corroborate earlier US results of complete integration by the second generation in the labour market. We find that ethnic differences in labour market performance are significant, but that these difference do not alter conclusions about the relative performance of the second generation. Second, we find a source of superior labour market performance for the second generation in higher educational attainment, which constitutes an important legacy of immigration that should not be ignored. Third, we find that the definition of the second generation matters. Men and women with two immigrant parents achieve about one additional year of education, while those with one immigrant parent achieve about one-half that educational advantage. We conclude that the education effect of an immigrant mother or father are comparable.

Acknowledgements

We would like to thank Statistics Canada for providing access to the data and the Prairie Centre for Excellence in Research on Immigration and Integration for financial support. The authors take sole responsibility for errors, omissions and interpretation of the data.

Notes

¹ Beginning with Chiswick's (1978) initial estimates, many studies using cross-sectional, primarily census, microdata suggest that immigrants eventually do as well as, or better than, their native born counterparts (e.g. recently Yuengert, 1994 and Funkhauser and Trejo, 1995 for the United States, Baker and Benjamin, 1997 and Grant, 1999 for Canada, McDonald and Worswick, 1999 for Australia and Friedberg, 2000 for Israel).

² In a series of studies, Borjas (1985, 1993b, 1996) argues that the success of immigrants is overstated due to the declining productivity of successive cohorts of migrants, and this analysis is supported by Canadian research (e.g. Baker and Benjamin, 1994; Bloom et al ., 1995; Frenette and Morissette, 2003). Recent studies using panel data are also skeptical of the immigrant integration results based on cross-sectional data (e.g. Hum and Simpson, 2000, 2004a).

³ Other articles in this journal examine related issues of unemployment among immigrants to France (Gross, 2002), England (Price, 2001) and Australia (Nahid and Shamsuddin, 2001), but this related dimension of immigrant integration is not considered here, except insofar as immigrant earnings reflect both periods of employment and unemployment.

⁴ Some research includes both foreign and native born children of immigrants and refers to these children as ‘second generation immigrants’ (e.g. Kossoudji, 1989; Gang and Zimmerman, 2004). Our empirical analysis in Section IV excludes immigrants, for reasons we set out subseqently, and we remind readers that we are concentrating on native born children with immigrant parents by referring to this group as the ‘second generation.’ This definition of the second generation is consistent with the research on the labour market performance of the second generation in the United States by Chiswick (1977) and Carliner (1980) that we discuss subsequently. One advantage of restricting our analysis to the native born is that we do not need to define foreign-born children of immigrants in terms of age at arrival or otherwise control for age at arrival in our empirical analysis.

⁵ An additional layer of distinctions, which we ignore here, is between the children of immigrants who are born abroad but arrive at an early age and those who are native born. We consider only native born children of immigrants in our empirical analysis and avoid questions associated with what constitutes arrival at an early age.

⁶ Cohort bias arises in a cross section because cohort and years since migration are exactly correlated and their effects cannot be separated. The absence of variables to capture the cohort of entry introduces omitted variable bias to the estimates of included variables unless they are uncorrelated with the omitted cohort variables.

⁷ That is, children carry forward about 40% of the education and wage disparities from their parents’ generation by country-of-origin class.

⁸ If only age is used as a control variable, this will show up as an age premium for the second generation, i.e. a higher average return to age than their immigrant parents.

⁹ We use the internal or master file, which contains crucial information on immigrant parentage that is not available in the external or public file. An overview of SLID can be found at: http://www.ssc.uwo.ca/sociology/longitudinal/giles.pdf.

¹⁰ We further restricted our sample to adults 30 years of age and over, but the results showed little change and are available from the authors upon request.

¹¹ Among the native born there will also be ‘cohort’ effects associated with changing labour market conditions over time for different age groups, but these will be comparable across native born groups given education. There is nothing comparable to the across cohort bias associated with immigrant integration.

¹² We provide sample SDs to indicate the approximate significance of our results. Mean differences in wages represent more than seven SDs, while mean differences in earnings represent more than three SDs.

¹³ Figure 1 provides the predicted hourly wage from a locally weighted regression of hourly wages on age, based on Cleveland's (1979) tricube weighting function. This nonparametric method is also used in Figs 2–5.

¹⁴ That is, a set of variables that are not job-specific. In particular, z contains other household income and family circumstances (marital status, the presence of preschool children and the number of children in the household), which are distinctive characteristics of models of labour supply models but not Mincerian models of wages and earnings. Because a reduced-form hours worked equation will depend on factors which affect the wage rate, z also contains years of schooling and years of experience and its square. On the other hand, x will contain characteristics of employment union status, firm size and industry) that are not observed for those not working.

¹⁵ See, for example, Davidson and MacKinnon (2004, pp. 486–8) for the specification of the likelihood function for the sample selection problem.

¹⁶ An earlier version of this article (Hum and Simpson, 2004b) used a tobit model, which is inconsistent when the errors are heteroskedastic and considerably less flexible than the Heckman one-step estimator used here (Lee and Maddala, 1985, for example). The estimator used here, for example, allows us to expand the set of control variables to include job characteristics to explain hours. We thank an anonymous referee for this suggestion to improve the current version of the article.

¹⁷ The Wald test for independent selection (employment) and structural (wage, hours or earnings) equations also rejects the hypothesis of no sample selection bias at the 1% level of significance.

¹⁸ Hum and Simpson (2004a) estimate virtually identical returns to an additional year of schooling in Canada of 3.5% for men and 6% for women using a fixed effects model and the first SLID panel (1993–98). They also note that these estimates are consistent with other fixed effects estimates of the returns to education in the United States.

¹⁹ Using mean hours and wages for the other native born samples of men and women in Table 1, the coefficient for imputed log wage estimates a gross wage elasticity for men of 0.21 and 0.40 for women and the coefficient on other income yields an income elasticity of 0.04 for men and 0.02 for women. Hum and Simpson (1991, pp. 18–39) find comparable labour supply elasticity estimates from earlier studies using cross-sectional nonexperimental and panel experimental data in Canada and the United States.

²⁰ SLID does not ask about the specific ethnic backgrounds of parents but instead poses the following general question: ‘Canadians come from many ethnic, cultural and racial backgrounds. For example, English, French, North American Indian, Chinese, Black, Filipino or Lebanese. What is your background?’

²¹ Statistics Canada restricts the release of results involving ethnic categories to samples of 30 or more respondents. Using this criteria, Table 1 indicates that only the British and European ethnic categories would provide sufficient observations for analysis.

²² That is, we found a significant wage disadvantage for black men in an earlier version of this article (Hum and Simpson, 2004b), which used the same data set. In that version we used a tobit model to explain hours worked which could not allow for employment characteristics, since these characteristics are not observed for those who did not work in 1999. Thus, the finding that the wage disadvantage for black men is eliminated when we control for employment characteristics is not an artefact of the small sample size for black men in our data.

²³ We thank an anonymous referee for suggesting this additional line of evidence on the importance of education to the assessment of the integration achieved by the second generation.

²⁴ Estimates in Card et al . (2000) which use mean education by ethnic group from 1970 Census fathers and their 1994–1996 CPS offspring find higher correlations (82% for men and 87% for women), but these aggregated results are not strictly comparable and would be expected to be higher since they net out individual variation in educational attainment present in our results. We also cannot provide comparable estimates of the intergenerational assimilation rate represented by λ ^k in Equation 4 because we do not have a continuous measure of years of schooling for parents.