سرمایه گذاری سرمایه انسانی و رشد اقتصادی: بررسی شواهد بین کشوری

The paper investigates three models on the role of education in economic growth: human capital theory, a threshold effect, and interaction effects between education and technological activity. Data for 24 OECD countries on GDP, employment, and investment from the Penn World Tables over the period 1950 to 1990 was used. Five sources are used for educational data. The descriptive statistics suggest that the convergence in labor productivity levels among these nations appears to correspond to their convergence in schooling levels. However, econometric results showing a positive and significant effect of formal education on productivity growth among OECD countries are spotty at best. With only one or two exceptions, educational levels, the growth in educational attainment, and interaction effects between schooling and R&D were not found to be significant determinants of country labor productivity growth.

There are three paradigms which appear to dominate current discussions of the role of education in economic growth: the first has stemmed from human capital theory; the second could be classified as catch-up models; and the third important approach has stressed the interactions between education and technological innovation and change.

The descriptive statistics presented in this chapter suggest a positive association between years of formal education and labor productivity levels among OECD countries over the post World War II years. First, the increase in educational attainment seems to correspond to the growth in labor productivity over this period. Second, the convergence in labor productivity levels among these nations appears to correspond to their convergence in schooling levels. However, econometric results showing a positive and significant effect of formal education on productivity growth among OECD countries are spotty at best. Three models were considered in this paper. The first is the threshold model of education, in which a certain level of educational attainment of the work force is viewed as a necessary condition for the adoption of advanced technology, With regard to this model, none of the measures of mean years of schooling proved to be significant, and, indeed, in many instances, their coefficient is negative. Educational enrollment rates are generally found to be insignificant as determinants of productivity growth with the sole exception of the primary school enrollment rate in 1965 at the 10% level. Likewise, educational attainment rates are generally insignificant, with the sole exception of the primary school attainment rate. However, a variable measuring the number of scientists and engineers employed per capita is found to be significant across a wide range of specifications. There is also little support found for the second model, the human capital approach. The growth in levels of formal schooling appears to have no bearing on the growth in labor productivity. This is true despite the surface evidence that mean years of schooling in OECD countries continued to grow over the post World War II period, as did productivity levels, and schooling levels converged among these countries in conjunction with productivity levels. One exception is the change in university enrollment rates between 1965 and 1991, whose coefficient is significant at the 10% level. Another exception is that when a vintage effect for the capital stock is included in the regression equation, the growth in years of schooling becomes significant at the 10% level. These results also appear to be inconsistent with growth accounting models, which have attributed a substantial portion of the growth in productivity to increases in schooling levels. There are three potential reasons for the difference in results. The first is that growth accounting exercises typically exclude a catch-up effect. Evidence from the Penn World Table Mark 5.6 data (though not from the Maddison data) does suggest that with the absence of a catch-up effect, the growth in mean schooling does contribute significantly to the growth in productivity. The second stems from methodological differences in the two techniques. Growth accounting simply assigns to schooling (or measures of labor quality) a (positive) role in productivity growth based on the share of labor in total income. In contrast, regression analysis lets the data tell us whether a variable such as education is a significant factor in productivity growth. The third is that growth accounting is typically performed with data for a single country, whereas the regression results reported above are based on a cross-country sample. Insofar as there are problems in comparability of educational data across countries, individual country studies which use a consistent definition of education over time within a particular country may come up with more reliable results than cross-country comparison data (though see below for more discussion of the comparability problem). The third model is based on the argument that a more educated labor force might facilitate the adoption of new technology. However, the regression evidence provides no corroboration that there is any kind of interaction effect between the degree of technological activity, as measured by R&D intensity, and the educational level of the work force. On the other hand, R&D intensity by itself does play a powerful role in explaining differences in productivity growth among OECD countries. What may explain the generally poor results for measures of formal education in accounting for productivity growth in advanced countries? There are five possible reasons. The first is the poor quality of the education data. Though I have used five different sources for these figures, they all generally derive from UNESCO sources, which, in turn, are based on compilations of country responses to questionnaires. I have already noted some anomalies in the raw data for some of these educational indices — particularly, the figures on educational attainment rates and mean years of schooling of the labor force. However, it should be noted that when corrections were applied to these data — primarily, through averaging techniques — the econometric results remained largely unchanged. Moreover, as long as there is no systematic correlation between errors in measurement of schooling and productivity growth rates, the measurement errors should be captured in the stochastic error term of the regression and not bias the coefficient estimators. Another point is that even if there is a systematic bias in measures of education by country, as long as these biases remain relatively constant over time, they should not affect the results of the production function equations. In particular, assume that schooling is mismeasured and that the true educational attainment of country h at time t, Eth, is given by: View the MathML source where Sth is measured schooling of country h at time t and λth is the correction factor. Then, in a Cobb-Douglas production function, with quality-adjusted labor input, View the MathML source where Y is output, L is (unadjusted) labor input, K is capital input, and α, β, γ, and c are constant terms. Substituting for Eth, we obtain: View the MathML source It is now clear that as long as λh is constant over time (though it may differ across countries), the measurement error will be captured in the constant term: View the MathML source Such systematic measurement errors in education should not bias the results of regressions, such as Eq. (3), where labor productivity growth is regressed on the rate of growth of educational attainment. However, if λth varies over time, then biased coefficient estimates will result. A second and more likely reason may be problems of comparability in formal education measures across countries. As discussed in OECD (1998), there are substantial differences in schooling systems among OECD countries, and years of schooling may be a particularly unreliable indicator of achievement and knowledge gain (as measured by test scores). Moreover, the failure of measures of formal schooling to include time spent in apprenticeship programs will also create systematic biases across countries, particularly for a country like Germany. How do such problems affect the regression results? Here, one might suspect systematic errors in measurement across countries, so that regressions of labor productivity growth on educational levels would produce biased coefficient estimates. However, one might also suspect that this type of measurement error would be relatively fixed over time within country, so that the coefficient estimators of production function regressions, such as Eq. (3), would not be unduly biased. A third possibility is specification errors. Levine and Renelt (1992) have already shown the sensitivity of cross-country regression results to changes in specification. However, I have used a large number of alternative specifications in this work. With only a few exceptions (for example, adding a vintage effect variable to the production function equation), the coefficient estimators of the educational variables have been largely unaffected by adding or deleting variables. A fourth possibility is that the causal relation between productivity and schooling may be the reverse of what I have assumed — namely, that schooling levels respond to per capita income levels instead of productivity growth to educational levels. In many ways, the acquisition of schooling responds to social forces. This may be particularly true in an ‘elitist’-type schooling system such as France has developed. Insofar as schooling is a luxury good, rising per capita income would lead to greater availability of schooling opportunity and hence rising schooling levels, particularly at the university level. This hypothesis might account for the positive and significant coefficient on the change in university enrollment rates in the cross-country productivity growth regression (Table 10). It might also reconcile the apparent inconsistency between rising educational levels in OECD countries and the weak relation observed between productivity gains and the rise in schooling levels. The fifth possibility, and the one I think is the most likely candidate, is that formal education per se may not have much relevance to productivity growth among advanced industrial countries. Rather, it may be the case that only some forms of schooling (and training) are related to growth. Two types have already been highlighted in the results. The first of these is the attainment rate of primary education in the labor force. Primary schooling is the level at which the basic literacy and numerical skills required for almost all types of work are acquired. This result is consistent with an earlier study of ours (Wolff and Gittleman, 1993) which shows that primary schooling is the most powerful educational variable in explaining growth in per capita income among countries at all levels of development. The second is the number of engineers and scientists per capita, which I have interpreted as a measure of the level of R&D activity in a country. It is rather surprising that secondary education as a whole appears so weak as an explanatory factor of productivity growth. However, there is also strong suggestion that alternative institutional arrangements like worker-based or employer-based training, apprenticeship programs, and technical education may bear a stronger relation to productivity growth than average years of secondary schooling (see OECD, 1998). The lack of cross-country statistics on these programs prevents a systematic econometric analysis of their influence. It is also surprising that with the explosive growth of university education in OECD countries in recent years, its relation to economic growth, with the exception of scientific and engineering training, is so tenuous. I have previously suggested two possible explanations (see Wolff and Gittleman, 1993). The first is that higher education may perform more of a screening function (see Arrow, 1973 and Spence, 1973) than a training function. According to this explanation, the skills acquired in university education are not relevant to the work place, but serve employers mainly as a signal of potential productive ability. As enrollment rates rise, screening or educational credentials may gain in importance, and a higher proportion of university graduates may become over-educated relative to the actual skills required in the workplace. A second possibility is that university education may encourage more rent-seeking activities than directly productive ones. For example, a study by Murphy et al. (1991) reported that per capita GDP growth is negatively related to the number of lawyers per capita of a country. Here, too, with rising enrollment rates, the rent-seeking bias in higher education may have been increasing in recent years. A third and related explanation might be the increasing absorption of university graduates by ‘cost disease’ sectors characterized by low productivity growth, such as health services, teaching, law, and business services (see Baumol et al., 1989). These are essentially labor activities, and, as such, are not subject to the types of automation and mechanization that occur in manufacturing and other goods-producing industries. These sectors are also plagued by difficulties in measuring output and, therefore, the low productivity growth figures that are found for them may also reflect output measurement problems, particularly in regard to quality change. If this is so, there might, indeed, be a positive relation between productivity growth and university education, but inadequate output measures prevent us from observing it.