The Treasury

Global Navigation

Personal tools

Treasury
Publication

Adult literacy and economic growth - WP 04/24

4  Macroeconomic studies of aggregate returns to literacy skills

4.1  Cross-country growth studies

This chapter looks at whether cross-country studies of economic growth provide evidence that increasing literacy skills would be good for the economy. It is one of three chapters looking at the economic effects of increased literacy skills (these were introduced in Chapter 3).

Cross-country growth regressions relate GDP growth in various countries, or the level of GDP at a particular time, to various features of the countries such as their rate of savings, growth in population, size of government, and even the degree of religiosity in the country (Barro and McCleary 2003). In a very recent study, Coulombe, Tremblay and Marchand (2004) use literacy as an explanatory variable in their cross-country growth regression, and find that the average literacy score in a country is positively associated with economic growth. This result appears extremely promising, and is discussed below, but it is first necessary to rehearse a little of the theory of economic growth.

Human capital is a key part of contemporary theories of economic growth, of which there are two main types: neo-classical growth models and endogenous growth models. In an influential and much-cited paper, Mankiw, Romer and Weil (1992) give human capital equal billing with physical capital, labour and technology in an augmented neo-classical growth model.[21] In this model, the long-run level of GDP per capita in a country is determined by the proportion of GDP the country saves (and therefore invests in physical capital), the proportion it invests in human capital, and the rate at which its population grows. The higher the proportion of GDP saved, or invested in human capital, the richer the country. The higher the rate of population growth, the poorer the country. These factors affect the long-run steady state level of GDP per capita, and also the rate at which a country ‘converges’ to this steady state level over time. Only the fourth factor in the model, technology, affects long-run growth in GDP per capita – a country’s steady state level will grow each year as technology (knowledge, for example, about production methods) increases.

Neo-classical growth models treat human capital as an investment good in much the same way as a farmer might consider investing in tractors. And, just as there are diminishing returns to a farmer buying more and more tractors, these models hold that there are diminishing returns to human capital accumulation. Suppose a country devotes a fixed 10% of its GDP each year to accumulating human capital. More human capital means the country can produce more output, which means that more is spent on human capital, which leads to more output, and so on. However, at each iteration the country gets less and less return for its additional investment in human capital, until the process grinds to a halt some years later at the steady state level of human capital investment (and the steady state level of GDP). If the investment rate were to increase from 10% to 12% the process of human capital accumulation and increases in GDP would crank up again and continue until the country reaches a new steady state of human capital investment (and GDP).

Endogenous growth models, on the other hand, make much more of the role of human capital, and consider the accumulation of ideas and skills to be quite different from the accumulation of tractors. In particular, they assume constant or increasing returns to investments in human capital. Consider again the country which devotes 10% of its GDP each year to accumulating human capital. As the country gets richer it devotes more to human capital, which in turn continues to increase output, and so on, indefinitely. In endogenous growth models, human capital accumulation leads to a sustained increase in GDP, that is, to sustained economic growth in the long term. Increasing investment in human capital to 12% of GDP would give an even bigger boost to this continuing cycle of growth. Dowrick (2003) discusses features of endogenous growth models which generate constant or increasing returns to human capital investments.

Temple (1999) provides an excellent review and discussion of empirical work on economic growth. Cross-country growth regressions constitute a relatively new field of study in economics, having only been conducted over the past 15 years. Studies typically try to explain countries’ growth experiences from 1960 onwards, either using a large sample of countries or, less often, a sample of OECD countries. Some studies use a formal framework derived from one of the theoretical growth models as discussed above but others use a more or less ad hoc specification. Almost all cross-country studies include some measure of human capital in their regressions, such as school enrolment rates or the average years of schooling in the working-age population. Hanushek and Kimko (2000) and Barro (2001) depart from the norm by using the results of international student tests of achievement conducted from the 1960’s to the 1990’s.

4.2  Literacy in cross-country regressions

Coulombe et al (2004) perform cross-country growth regressions using literacy scores, obtained from IALS, as their measure of human capital. The study includes those 14 OECD countries which participated in the 1994 and 1996 rounds of IALS. Coulombe et al use a specification which is closely based on that tested in Mankiw et al (1992), the main difference being that Coulombe et al look at growth in GDP over five-year periods between 1960 and 1995 (Mankiw et al look at growth over the whole period between 1960 and 1985). For each country, growth in GDP over these five-year periods is modelled as a function of GDP at the beginning of the period, the mean rate of savings over that period, an indicator of literacy over that period, and the mean fertility rate (which performs a similar role to population growth) over that period.

Leaving aside the issue of the validity of IALS comparisons (see section 2.5), the availability of historical literacy data presents an obvious problem for this analysis since IALS has only been conducted once. Coulombe et al therefore assume that the mean literacy score for the 51-59 age group in IALS in 1994 would have been the same for this cohort 34 years earlier in 1960, when the cohort was aged 17-25, and they take this 17-25 age group score to be the literacy indicator in the model for the period 1960-65. Similarly, they use the mean literacy score for the 46-54 age group in IALS as the literacy indicator for the period 1965-70, and so on. This use of synthetic cohorts does involve some brave assumptions, not least that people’s literacy skills persist over quite long periods of time, neither increasing nor decreasing as they get older. This may not be such a problem for the analysis if any loss or increase in literacy over time occurs in a similar fashion for each country, but we have no knowledge about whether or not this is the case.

Regardless of the difficulties involved in using IALS literacy scores it must be stressed that literacy is included in this model, not for its own sake, but as a proxy for the proportion of GDP spent on human capital accumulation. This latter variable appears in the augmented neo-classical model but is very difficult to measure directly. Mankiw et al use the proportion of the population enrolled at secondary school as their proxy measure, while other studies use average years of schooling of the adult population. These measures are clearly imperfect but are reasonable proxies if the relative position of countries on the proxy measure is more or less the same as their relative position with respect to the proportion of GDP spent on human capital. Coulombe et al are testing the explanatory power of the neo-classical growth model when using literacy skills as a proxy for the human capital investment rate rather than enrolment rates or years of schooling. This is in response to a number of studies which find that schooling-based proxies for human capital are statistically insignificant, or have a negative sign, when five-year periods of growth are analysed, or when the cross-country sample consists only of OECD countries, eg Islam (1995).

Coulombe et al find that literacy scores, constructed as described above, are positively and significantly associated with the rate of convergence to a country’s steady state level of GDP. This is the case regardless of whether prose, document or quantitative literacy is used in the regressions. Literacy is also a determinant of the steady state level of GDP. A country that achieves literacy scores one percent higher than the average is estimated, all else equal, to reach a steady state with around 1.3% higher GDP per capita.

These findings do not show, however, that literacy skills in and of themselves are good for the economy. As discussed above, the authors are not trying to isolate the effect of literacy skills on growth but rather to test the use of literacy as a proxy for human capital accumulation. They conclude that literacy performs well in this role:

The central result of the paper is that direct measures of human capital based on literacy scores outperform measures based on years of schooling in growth regressions of a sub-set of OECD countries. Furthermore, it appears that, overall, human capital indicators based on literacy scores have a positive and significant effect on the transitory growth path, and on the long run levels of GDP per capita and labour productivity. The key economic policy implication that comes out of this result is that, in contrast to previous findings… human capital accumulation matters for the long run wellbeing of developed nations. (p39).

4.3  Conclusion

The aggregate data used in cross-country growth regressions kind is necessarily crude and gives little helpful guidance on detailed policy questions. In particular, some cross-country regressions show that the rate of human capital accumulation is important for economic performance but offer little advice on what to invest in (basic literacy? PhDs? learning Spanish?). The study undertaken by Coulombe et al (2004) suggests that literacy is a reasonable proxy for human capital accumulation in cross-country growth regressions but it may be that other aspects of human capital, correlated with literacy scores, actually drive economic performance. Microeconomic studies provide a more detailed investigation of these issues and we turn to such studies in the next chapter.

Notes

  • [21]Previous to this, neo-classical growth models in the tradition of Solow (1956) and Swan (1956) did not explicitly refer to human capital, and had as factors only capital, labour and technology.
Page top