A small addition to the IZA working paper with Barry R. Chiswick.
To deal with the endogeneity between Dutch language ability and earnings, we use an interaction term between age at migration (AAM) and a dummy for non-Dutch-speaking (NDS) origin as the identifying instrument. In mathematical terms, it is a numerical value of max(0, AAM-5)*NDS. As argued in Bleakley and Chin (2004), upon arrival in the Netherlands, immigrants originating from Dutch-speaking countries encounter everything that the immigrants from non-Dutch-speaking countries except a new language. Any difference in wages between young and old arrivers in non-Dutch-speaking countries that is different from the wage difference in Dutch-speaking countries can plausibly be attributed to Dutch language ability.
The figure below shows the relationship between age at migration and Dutch language ability. Children with early exposure to Dutch language attain higher levels of Dutch language ability. For early arrivers (before the age of 5), the Dutch language ability of immigrants from non-Dutch-speaking countries is comparable to that of immigrants from Dutch-speaking countries. However, there tends to be a significant difference in Dutch language ability for late arrivers. Our estimate for Dutch language proficiency remains positive and statistically significant.
Comments and suggestions are extremely welcome.
Bleakley H. and Chin A. (2004), Language Skills and Earnings: Evidence from Childhood Immigrants, The Review of Economics and Statistics, 86(2), pp.481-496.