6.1 Introduction
This chapter will survey empirical studies on the link between human capital and economic growth. In order to render this survey manageable and to provide a bridge to the empirical work in the remainder of this book, it will concentrate mainly on studies that use statistical data from a cross section of countries and employ econometric estimation techniques. Thus, it will not consider the many studies that focus solely on one or a handful of countries and/or studies that use calibration or simulation methods. The empirical literature on the human capital-growth nexus is differentiated by the specification of the estimating equation, the way human capital is defined, the time frame considered, and the countries included in the sample. The scheme for classifying empirical studies adopted in this chapter is mainly for convenience. In practice the various empirical approaches share several common features and there is overlap between them.
Both intuition and various theories of endogenous growth discussed in Chapters 3 and 4 point, a priori, toward a positive effect of human capital on economic growth. Empirical evidence on this issue has been mixed. To provide a flavor for the various approaches discussed in this chapter we note that early contributions proved quite successful in establishing a robust link between enrollment rates (the proportion of adults enrolled in secondary education) and growth of GDP per worker (for example, the influential work of Mankiw, Romer, and Weil discussed in Section 3.2.3). Subsequent studies have questioned this result by broadening the definition of human capital (to include education at different levels) as well as alternative measures of human capital. They find that human capital explains a much smaller proportion of the variation in income per capita than claimed earlier. The early studies tended to emphasize the use of enrollment rates (flows) for primary or secondary education. More recent studies have used stock measures, that is, mean years of schooling of a country’s adult population. Some researchers have differentiated measures of human capital, not only by level of education (primary/secondary/tertiary), but also by sex. Studies that treat human capital as a direct input to the production function have shown that human capital accumulation exerts an insignificant or sometimes even negative effect on growth (Benhabib and Spiegel 1994; Pritchett 2001). These studies, however, have been criticized for their use of a log–log specification. The time dimension over which the growth rate is calculated has also come under scrutiny. Studies range from those utilizing pure cross-section data to those with panel data of varying frequencies. As the frequency over which growth rates are calculated increases (5-year changes vs. 10 or 20 years) the statistical evidence on human capital accumulation and growth tends to weaken: when the temporal dimension of human capital variables is incorporated into growth regressions, outcomes of either statistical insignificance or negative sign have surfaced. In sum, in his survey of the growth literature, Temple (1999) contrasts the success of micro-level studies that have established a positive effect of schooling on wages with the failures of studies at the macro level to do so.
In the following sections we provide a general framework to assess empirically the human capital-growth link and pay particular attention to empirical tests derived from some of the theoretical propositions of the models discussed in earlier chapters. We discuss briefly the estimates obtained from the parametric framework that characterize the linear approach. These results will provide the benchmark against which we will measure the nonlinear specifications presented in later chapters. On the whole, the empirical studies reviewed in Sections 6.2–6.5 use linear estimation techniques. Following up on the theoretical models discussed in Chapter 5 and the nonlinear approach that we will discuss in greater detail in subsequent chapters, in Section 6.6 we review threshold effects from an empirical perspective. These effects act as the main engine for the presence of nonlinearities both from a theoretical and empirical standpoint, even though they describe fairly simple forms of nonlinear behavior. Most attempts to tackle nonlinear behavior empirically are based on methods that are closely related to threshold effects and their smoother generalizations. Section 6.6 serves by way of introduction to the investigation of nonlinearities in Chapters 8 and 9.
6.2 A General Empirical Framework for Estimating the Contribution of Human Capital to Economic Growth
Most of the empirical literature on growth and human capital takes as a starting point the dual question of, first, whether the quantity of education (the empirical measure most closely associated with the concept of human capital) has a positive impact on the rate of economic growth and, second, the magnitude of any such effect. One of the most popular ways of modeling empirically the macroeconomic contribution of human capital emphasizes the process of convergence to the steady state. In its general form (e.g., Barro 1991), it claims that the deviation of the rate of growth from its steady-state level depends on the distance between the initial and steady-state level of output per worker or
where (ln yi,τ +ν–ln yi,τ ) is the growth of output per worker of country i between period τ and τ + v, gA is the steady-state growth rate, yi,τ is initial and is steady-state level of output per worker, a1 is a measure of the rate of convergence, and ∈i, τ is a random error term that obeys standard assumptions. The steady-state level of output per worker, , depends on the level of human capital per worker (hi) and a vector of variables (Zi ) that capture differences in the economic, institutional, and political environments across countries. With this in mind, the estimable form of the cross-country growth-regression approach becomes: Our interest is in the estimate of a2.
Topel (1999) and Krueger and Lindahl (2001) point out that interpreting the estimate of a2 is not straightforward. It is consistent with various hypotheses about the role of human capital in the process of economic growth. As Barro (1991) points out, human capital is a direct determinant of the steady-state level of output per worker. Therefore, given the current (initial) level of output per worker, a higher level of human capital also raises the distance from steady state and convergence would thus imply higher growth, or a2 > 0. But a...