The size of human capital externalities: cross-country evidence

Seminal contributions include Romer (1986), Lucas (1988), Stokey (1988), and Young (1991).

This insight has been formally proved by Baumol (1964). The Coase theorem (Coase 1960) also formulates the conditions for Pareto efficient outcomes in the presence of an externality without government intervention. Buchanan (1973) demonstrates that voluntary contractual arrangements may emerge under a variety of institutional conditions in some large-number sets. For a critique of the theory of public goods, see also Goldin (1977).

For a recent summary of the empirical literature, see Patrinos and Psacharopoulos (2011).

On the distinction between pecuniary and technological externalities, see the seminal paper by Scitovsky (1954).

Examples for recent within-country studies that estimate human capital (educational) externalities by comparing private and local social returns include Baum-Snow and Pavan (2012), Heuermann (2011), and Canton (2009) on the income of workers across cities and agglomerations; Martins and Jin (2010) on the income of workers across firms; and Kirby and Riley (2008) on the productivity of firms within industries. Most studies in this strand of literature find no evidence for human capital (educational) externalities. A different strand of the literature uses within-country data to infer the potential relevance of human capital externalities from a correlation between measures of education and variables such as political participation, health, and crime. Examples for papers estimating education’s effect on political participation include Dee (2004), who considers the local availability of colleges and changes in teen exposure to local child labor laws, and Milligan et al. (2004), who consider extra schooling induced through compulsory schooling laws. A recent survey of the literature in this field (Lochner 2011) finds that more schooling tends to foster a politically more engaged citizenry, better health, and less crime, which points to the presence of human capital externalities.

To keep the analysis as simple as possible, a quadratic experience term is omitted from Eq. (4).

The macro-Mincer specification is criticized by Ciccone and Peri (2006) for falsely identifying human capital externalities when there are none at all. In their reading, the macro-Mincer specification confounds positive externalities with neoclassical supply effects. They show that the wages of low-skilled workers should rise in response to an increased supply of high-skilled workers, due to a downward sloping demand curve for relatively skilled labor, even in the absence of any positive externality. To the best of our knowledge, this argument has been tested only on the basis of within-country data. As discussed in Sect. 2, this may introduce a bias against finding evidence for a human capital externality. This is not to deny that the macro-Mincer specification may generate upwardly biased estimates of the human capital externality as long as cross-country differences in the composition of skill-types are not accounted for. However, the systematic cross-country variation in the composition of skill types may be partly captured by the continental dummies discussed below (see Sect. 4.1).

The assumption of equal private returns to schooling and experience is problematic but necessary for the identification of the externality. Using country-specific returns instead of averages would be possible in principle, but apart from reducing the sample one would face the problem that country-specific estimates also show some degree of variation, so the problem of averaging partly would remain. The systematic variation in cross-country rates of return, namely lower average returns in richer countries, will be captured in part by the continental dummies discussed below (see Sect. 4.1). In addition, cross-country differences in schooling systems, which should affect returns, are accounted for by an index of schooling quality (see Sect. 4.1).

This mistake apparently results from not taking into account that in the absence of a human capital externality, the estimated social macro-Mincer return is already given by the production function context as the product of the private micro-Mincer return and the labor share; see Eq. (7).

The average schooling years in Barro and Lee (2010) are based on population-weighted schooling years for ten age groups, ranging from 15–19 years to 60–64 years. This is consistent with the weighting used for the construction of the median age.

Detailed results are available upon request.

The estimated coefficients may also be downwardly biased because of classical measurement error, which would reflect that quality-adjusted schooling and experience may be imperfect proxies for a true measure of human capital. Compared to other estimators, OLS estimation may actually lead to less biased estimates in the presence of multiple data problems; see Hauk and Wacziarg (2009) for simulation results in the context of convergence regressions.

The instrumental variables are taken from a variety of sources, which are listed in Table A1 in the Appendix.

The combination of a high partial R ² and a p-value of 0.98 in column (3) signals a high correlation of all included instruments, which reduces the power of the Sargan test.

For details on the measures of geography, see Table A1 in the Appendix.