VOL. 80 NO. 2 THE"NEW" GROWTH THEORY the last 25 years and, in any case, a rough economic motive for capital to flow, there estimate is better than none at all. Her would be no motive for labor flows either method is to combine information on each Yet we see immigration at maximal allow country's mix of workers by level of educa- able rates and beyond from poor countries tion, age and sector with U. S. estimates of to wealthy ones. We do not want to resolve the way these factors affect worker produc- the puzzle of capital flows with a theory that tivity, as measured by relative earnings predicts, contrary to the evidence provided Krueger's main results are given in her by millions of Mexicans, that Mexican work Table Il(p. 653), that gives estimates of the ers can earn equal wages in the United States per capita income that each of the 28 coun- and in Mexico tries examined could attain, expressed as a fraction of U.S. income, if each country had Il. External Benefits of Human Capital the same physical capital per worker endow ment as did the United States. The estimates Obviously, we could resolve the puzzle of ange from around 38(India, Indonesia, the ina uac ows at any Ghana) to unity(Canada) and 84 (Israel). by assuming that marginal products of capi These numbers have the dimension of the tal are equalized, and using equation(2)and relative human capital stocks raised to the the estimated income differential to estimate ower of labor's share, so taking the latter at the relative levels of the intercept parameter 6(as I did in my introductory example), the A (often called the level of technology) in the two countries being compared. This is ments ranged from about. 2 to unity. That is, almost what I will do in this secto ach american or Canadian worker was esti- will do so in a way that has more content, by mated to be the productive equivalent of assuming that an economy's technology level about five Indians or Ghanians (Compensa- is just the average level of its workers'hu ion per employed civilian in the United man capital raised to a power. That is, I States in 1987 was about $24,000, so this assume(as in my 1988 paper), that the pro- estimate implies that a typical worker from duction function takes the form ndia or Ghana could earn about $4800 in the United States.) Krueger's estimated human capital diffe (3) tials, reinterpret y in equations (1)aden. where y is income per effective worker, x is capital per effective worker, and h is human s income per effective worker. Then the capital per worker. I interpret the term hYas ratio of y in the United States to y in India an external effect gust as in Paul romer, becomes 3 rather than 15, and the predicted 1986). It multiplies the productivity of rate of return ratio becomes(3)5=5 rather worker at any skill level h, exactly as does than 58. This is a substantial revision, but the intercept A in (3 even so, it leaves the original paradox very The marginal productivity of capital for much alive: a factor of 5 difference in rates mula implied by ( 3)is of return is still large enough to lead one to xpect capital flows much larger than any- (4) r=BA/y(8-1/hv/8 thing we observe If it had turned out that replacing labor I propose to estimate the parameter y using with effective labor had entirely eliminated Edward Denisons(1962)comparison of U. s stimated differences in the marginal prod- productivity in 1909 and 1958, and then to uct of capital, this would have answered the apply this estimate to (4) using Krueger's question with which I began this paper, but cross-country estimates of relative human only by replacing it with an even harder capital stocks in 1959 to obtain question. Under constant returns, equal cap diction on relative rates of return on capital i returns implies equa al wage rates for The estimation of y is as reported in my equally skilled labor, so that if there were no earlier paper(1988, p. 23). Using Denison,s