-2-
Productivity Convergence and Education:
Evidence from OECD Countries
Edward N. Wolff
September 2000 Draft
1. Introduction
There are three main views of the role of education in economic growth: the first stems from human capital theory; the second can be classified as a catch-up model; and the third important approach stresses the interactions between education and technological innovation.
Human capital theory views schooling as an investment in skills and hence as a way of augmenting worker productivity (see, for example, Schultz 1960 and 1961, and Becker, 1975).[1] This line of reasoning leads to growth accounting models in which productivity or output growth is derived as a function of the change in educational attainment.
The early studies on this subject showed very powerful effects of educational change on economic growth. Griliches (1970) estimated that the increased educational attainment of the U.S. labor force accounted for one-third of the Solow residual, the portion of the growth of output that could not be attributed to the growth in (unadjusted) labor hours or capital stock, between 1940 and 1967. Denison (1979) estimated that about one-fifth of the growth in U.S. national income per person employed between 1948 and 1973 could be attributed to increases in educational levels of the labor force.[2] Jorgenson and Fraumeni (1993) calculated that improvements in labor quality accounted for one fourth of U.S. economic growth between 1948 and 1986. Maddison (1987), in a growth accounting study of six OECD countries, covering the years 1913 to 1984 generally found that increases in educational attainment explained a significant proportion of productivity growth, though the contributions varied by country and sub-period.
Yet, some anomalies have appeared in this line of inquiry. Denison (1983) in his analysis of the productivity slowdown in the U.S. between 1973 and 1981, reported that the growth in national income per person employed (NIPPE) fell by 0.2 percentage points whereas increases in educational attainment contributed a positive 0.6 percentage points to the growth in NIPPE. In other words, whereas educational attainment was increasing, labor productivity growth was falling. Maddison (1982) reported similar results for other OECD countries for the 1970-1979 period.
The second strand views the role of education in the context of a productivity "catch-up" or "convergence" model. Previous explanations of the productivity convergence process almost all involve the so-called "advantages of backwardness", by which it is meant that much of the catch-up can be explained by the diffusion of technical knowledge from the leading economies to the more backward ones (see Gerschenkron 1952 and Kuznets 1973, for example). Competitive pressures in the international economy ensure rapid dissemination of superior productive techniques from one country to another. Through the constant transfer of knowledge, countries learn about the latest technology from each other, but virtually by definition the followers have more to learn from the leaders than the leaders have to learn from the laggards. One direct implication of this view is that countries which lag behind the leaders can be expected to increase their productivity performance toward the level of the leading nations and, ceteris paribus, should experience higher rates of productivity growth.
However, being backward does not itself guarantee that a nation will catch up. Other factors must be present, such as strong investment, an educated and well trained work force, research and development activity, developed trading relations with advanced countries, a receptive political structure, low population growth, and the like. Indeed, Abramovitz (1986 and 1994) has summarized this group of characteristics under the rubric of social capability.[3]
In this context, education is viewed as one index of the social capability of the labor force to borrow existing technology. One of the prime reasons for the relatively weak growth performance of the less developed countries is their failure to keep up with, absorb and utilize new technological and product information, and to benefit from the international dissemination of technology. One of the elements that can be expected to explain an economy's ability to absorb information and new technology is the education of its populace. Indeed, Nelson and Wright (1994) put particular emphasis on the educational attainment of the U.S. population, particularly at the university level, as a crucial factor in America's catch-up and eventual overtaking of the British economy in the latter part of the nineteenth century.
In this context, education may be viewed as a threshold effect in that a certain level of education input might be considered a necessary condition for the borrowing of advanced technology. Moreover, varying levels of schooling might be required to implement technologies of varying sophistication. On an econometric level, the correct specification would then relate the rate of productivity growth to the level of educational attainment.
As far as I am ware, I was the first to report an extremely strong effect of educational level on the growth in per capita income among a cross-section of countries covering all levels of development (see Chapter 9 of Baumol, Blackman and Wolff,1989, originally written and circulated in 1986). For our educational variable, we used enrollment rates for primary school, secondary school, and higher education Since that time, many other studies have reported similar results on educational enrollment (see, for example, Barro 1991; and Mankiw, Romer, and Weil 1992). In these two, as well as in most others, the secondary school enrollment rate has been used as the measure of educational input.
However, several cracks appear to have formed in this strand of research (see Wolff and Gittleman 1993, for details). First, the introduction of a number of "auxiliary" variables -- most notably, investment -- appears to mitigate the importance of education in the growth process. Second, whereas primary and secondary school enrollment rates both remain statistically significant as a factor in explaining economic growth, the university enrollment rate often appears statistically insignificant.
Third, the use of enrollment rates in productivity growth regressions has been aptly criticized because they are not indices of the educational attainment of the current labor force but of the future labor force. Moreover, high enrollment rates may be a consequence of high productivity growth -- that is, the causation may go the other way. As a result, several studies have used educational attainment at a particular point in time instead of educational enrollment rates in cross-country regressions in which growth in GDP per capita is the dependent variable. However, measures of the direct educational attainment of the labor force (or of the adult population) often produce weaker results than the use of enrollment rates (see Wolff and Gittleman 1993, for details).
A third strand emanates from the work of Arrow (1962) and Nelson and Phelps (1966). Arrow introduced the notion of learning-by-doing, which implies that experience in the application of a given technology or new technology in the production process leads to increased efficiencies over time. As a result, measured labor productivity in an industry should increase over time, at least until diminishing returns set in. One implication of this is that an educated labor force should "learn faster" than a less educated group and thus increase efficiency faster.
In the Nelson-Phelps model, it is argued that a more educated workforce may make it easier for a firm to adopt and implement new technologies. Firms value workers with education because they are more able to evaluate and adapt innovations and to learn new functions and routines than less educated ones. Thus, by implication, countries with more educated labor forces should be more successful in implementing new technologies.
The Arrow and Nelson-Phelps line of reasoning suggests that there may be interaction effects between the educational level of the work force and measures of technological activity, such as the R&D intensity of a country. Several studies provide some corroboration of this effect. Welch (1970) analyzed the returns to education in U.S. farming in 1959 and concluded that a portion of the returns to schooling results from the greater ability of more educated workers to adapt to new production technologies. Bartel and Lichtenberg (1987), using industry-level data for 61 U.S. manufacturing industries over the 1960-1980 period, found that the relative demand for educated workers was greater in sectors with newer vintages of capital. They inferred from this that highly educated workers have a comparative advantage with regard to the implementation of new technologies.
A related finding is reported by Mincer and Higuchi (1988), using U.S. and Japanese employment data, that returns to education are higher in sectors undergoing more rapid technical change. Another is from Gill (1989), who calculated on the basis of U.S. Current Population Survey data for 1969-1984 that returns to education for highly schooled employees are greater in industries with higher rates of technological change. Wolff (1996), using industry level data for 43 industries covering the period 1970-1985, found that the growth of cognitive skill levels (as defined by the Dictionary of Occupational Titles) of employees was positively related to indices of industry technological change, including computer intensity, capital vintage, and R&D activity.
There are several methodological problems in the types of cross-country growth regressions cited in the literature above (see Levine and Renelt 1992). First, there may be problems of comparability with cross-country measures of many of the independent variables used in this type of analysis, particularly between countries at very different levels of development. Behrman and Rosenzweig (1994), for example, stress the difficulties in comparing educational measures across countries, particularly in regard to the quality of schooling. Second, a related problem is that the availability of educational attainment data is much more limited than that of enrollment data. This may bias the sample of countries and the regression results. Also, imputations of missing educational data can also be misleading (again, see, Behrman and Rosenzweig 1994). Third, there may be specification problems in the equations that relate education and other variables to productivity or output growth. Levine and Renelt (1992) report that econometric results for certain exogenous variables can be very sensitive to the form in which they are entered into the equation.
The objective of this paper is to subject the three alternative models of the relation of education to economic growth described above to empirical analysis. I will also try to account, at least in part, for any discrepancies in results and, perhaps, to shed some new light on the role of education in economic growth.
In this study, I will confine the analysis to OECD countries. This has two methodological advantages. First, it will provide a relatively consistent sample of countries to be used in testing a wide range of models (though, in some cases, missing observations will force the exclusion of one or more of these countries). Second, it will mitigate, to some extent, problems of comparability of educational data. However, it should be stressed even at this point that educational systems do differ even among OECD countries. For example, as Maddison (1991) notes, standardized tests of cognitive achievement are usually much lower in the U.S. than in other industrialized countries at the same grade level. Moreover, some countries, such as Germany, have an extensive system of apprentice training integrated with part-time education, which is not reflected in the figures for formal schooling.[4] Thus, comparisons of standard measures of formal schooling even in this select sample of countries must be interpreted cautiously. We shall return to this point again in the conclusion of the chapter.
The remainder of this paper is organized as follows: The following section (Section 2) provides descriptive statistics on productivity levels, educational enrollment and attainment, and research and development for OECD countries. Section 3 reports econometric results on the effects of educational enrollment and attainment levels on per capita income growth. Section 4 analyzes the role of the growth in educational capital on productivity and output growth. Section 5 investigates evidence on interactive effects between education and R&D. Implications are outlined in the concluding part.
2. Comparative Statistics among OECD Countries
We begin with measures of real GDP (1985 dollar equivalents) per worker (RGDPW). Computations for 1950 to 1990 are based on the variable RGDPW from Penn World Table (PWT) Mark 5.6 (see Summers and Heston, 1991, for a description of the database). Computations for 1995 are based on the variables GDPD (Gross Value Added in US Dollars) and ET (Total Employment) from the OECD ISDB (International Intersectoral) Database for 14 OECD countries. The PWT figures for 1990 are used as benchmarks, and the figures are updated to 1995 on the basis of the growth rate of labor productivity for 1990-1995 computed from the ISDB data.
The now familiar convergence story is evident. The coefficient of variation (the ratio of standard deviation to mean) among the 24 OECD countries listed in Table 1 declines by more than half between 1950 and 1990. The 14 ISDB countries are by and large the biggest OECD economies, excluding countries such as Austria, Greece, Iceland, Ireland, New Zealand, Portugal, and Turkey. As a result, the convergence results for the ISDB countries are even stronger than among the 24 OECD countries, with the coefficient of variation fallying by almost three fourths. However, after 1980, the rate of convergence in labor productivity slows markedly in both samples.
Catch-up is also evident, as indicated by the correlation coefficient between the 1950 RGDPW level and the rate of growth of RGDPW after 1950. The correlation coefficient is -0.93 among all OECD countries and -0.93 among the ISDB sample. The results indicate that the countries with the lowest productivity levels in 1950 experienced the fastest increase in labor productivity.
Table 2 shows measures of educational levels among OECD countries. Panel A shows gross enrollment rates, defined as the ratio of the number of persons enrolled in school to the population of the corresponding age group by educational level. There was almost 100 percent enrollment at the primary school level and almost no variation among countries in this group. In contrast, the average secondary school enrollment rate increased from 59 percent in 1950 to 94 percent in 1991 among all OECD countries. The standard deviation remains fairly constant over time, while the coefficient of variation falls from 0.26 in 1965 to 0.15 in 1991 --- a reflection of the rising mean.