The raw data reveal substantial dispersion in the rate of wage convergence experienced by immigrants originating in different countries, arriving at different times, and at different ages. The remainder of the empirical analysis attempts to understand the source of these differences.
The dependent variable in this section is f), the rate of wage growth experienced by a particular immigrant cohort (from country i, arriving in year j, and at age к) over the intercensal ten-year period. The “cross-country” analysis is initially restricted to a sample that contains 85 countries (listed in the Appendix), four age groups, and 4 year-of-migration cohorts. The analysis “stacks” the data. The 85 countries used in the study are chosen because immigrants bom in these countries can be matched across two successive Censuses, and these countries have sufficiently large numbers of observations in the 1970,1980, and 1990 Censuses to allow reliable estimation of the first-stage regressions in equation (12). The issue of cell size will be discussed in more detail below. About 92 percent of the immigrants who entered the United States between 1960 and 1979 originate in one of these 85 countries.
I now use these data to examine some of the questions raised by the theory. First, what is the correlation between the rate of wage growth and the log entry wage? Consider the convergence regression model:
where 5jk is a fixed effect indicating if the immigrant cohort arrived in calendar year j at age t, and r\ is a stochastic error.
A number of technical details about the regression model in (15) are worth noting. First, the dependent variable may contain a great deal of sampling error. To account for the heteroscedasticity induced by this sampling error, I weigh all regressions by the factor (n~l + n~})“’, where nt is the sample size of the cell in Census year t. Note that the same country appears a number of times in the sample and the stochastic error r| might contain a country- specific component. The tables, therefore, report White-corrected standard errors that adjust for this sampling frame.