Robert Shimer: George Borjas has written an ambitious paper, presenting some new empirical findings and developing a framework for quantifying the efficiency gains from immigration. The paper begins by documenting that immigrants tend to move to states where workers with their skills earn high wages. It then argues that the resulting increase in labor supply in high-wage states accelerates the convergence of wages across states. And finally it claims that the efficiency gain from immigration roughly equals conventional estimates of the immigration surplus. The bulk of my discussion will ask whether we should accept each of these findings. The short answer is that immigrants do move to high-wage states but, surprisingly, that native migrants do not—a piece of evidence that is inconsistent with Borjas's framework. The support for the other empirical result, that migration compresses interstate wage differentials, thereby generating an efficiency gain from immigration, is weak.
To analyze whether immigrants move to states where workers with their skills earn high wages, Borjas proceeds in two stages. First, he uses decennial census data on native-born workers to estimate the average wage that workers with a given skill level, as proxied by their educational attainment, earn in each state conditional on their age and sex. The paper implicitly maintains the hypothesis that these interstate wage differentials do not simply proxy for the unobserved characteristics of workers in different states. This implies that a worker who moves to a high-wage state can expect to earn a higher wage than he or she would have earned in a low-wage state.
In the second stage, Borjas examines whether immigrants' location decisions respond to wage differentials. If they do, we should see a large [End Page 120] number of immigrants in a state-education group some years after observing a high level of wages in that group. Borjas runs the appropriate regression in first differences (his tables 3 and 4), but he provided me with some additional results in terms of levels, which I find easier to interpret. Some of those results are shown in table 1 below. A state-education group with 10 percent higher wages than the average draws a 29 percent larger relative supply of immigrants than of natives. Comparing the second and third columns, we see that the effect is slightly stronger on new immigrants than on earlier ones. Since these regressions include state fixed effects, they do not just say that immigrants move to California, Florida, and New York, which happen to be high-wage states. Rather, states that pay relatively high wages to college graduates tend to draw a lot of college graduates relative to high school graduates.
The finding that both old and new immigrants tend to reside in high-wage states might appear inconsistent with the findings in Borjas's first-difference regressions, where he shows that an increase in wages attracts new immigrants but not old ones. The very slow convergence of wages across states reconciles these results. A highly educated immigrant drawn to New York in 1950 would continue to earn high wages in New York throughout the sample period. Thus these regressions support Borjas ' s model of immigrants' location decisions.
Borjas notes that what is true for immigrants should also be true for native migrants. Any worker who moves should move to a high-wage state, regardless of whether he or she came from another state or another country. His table 5 shows that this is not the case. The most straightforward
Click for larger view
View full resolution
[End Page 121]
regression in the first column of the table finds that native movers tend to move to low-wage states. This effect is large in magnitude and very significant, with a t statistic in excess of -6. Constructing the data in other ways, Borjas cannot reject the null hypothesis that native movers, native stayers, and earlier immigrants live in the same places—that is, that native movers do not seek out high-wage states. This is a puzzle.