There are a number of technical problems with the convergence regressions reported in Table 4 that deserve some discussion. First, many of the cells in the analysis contain relatively few observations. The dependent variable in each cell is constructed from wages reported in two different Censuses. Because the 1980 and 1990 immigrant extracts form a 5 percent random sample of the population (and because the immigrant population has grown rapidly over time), the sample size used in the construction of wage levels for the various cells is reasonable for most national origin groups. In particular, 19 observations were used to calculate the 1980 wage for the average cohort, and 24 observations were used to calculate the 1990 wage. The smaller size of the 1970 immigrant extract, however, implies that only 11 observations were used to calculate the 1970 wage for the average cohort.

As noted earlier, all the regressions are weighted by the factor {n~x +n~) )”‘ I reestimated the regression models using only the cells that are likely to have the least sampling error. In particular, I restricted the analysis to the 50 percent of the cells that have the largest value of the weights. As shown in Table 4, there is even stronger evidence of a positive correlation between entry wage levels and the unadjusted wage growth in this restricted sample. And, as before, the positive convergence coefficient turns negative when the regressions control for either educational attainment or country of origin.

A second potential problem is that the log “entry wage” measures different things for different cohorts. As noted above, the wage observed at the beginning of the decade (time t) is roughly the entry wage for those immigrants who arrived between 1965-69 or 1975-79. This wage, however, is not the entry wage for immigrants who arrived in either the first half of the 1960s or the first half of the 1970s. The bottom panel of Table 4 shows that the results do not change when the regressions are restricted to the immigrant cohorts that migrated in the last half of a particular decade. The correlation between the log entry wage and the rate of wage growth is positive when no controls are included in the regression, and turns negative when human capital controls are added.