Education quality and the empirics of economic growth: reconciling Mankiw-Romer-Weil estimates with microeconometric evidence

ABSTRACT

We extend Mankiw, Romer and Weil’s classic paper by introducing differences in education quality (proxied by students’ performance on the PISA test). This substantially reduces the role of human capital investment rates in explaining cross-country income differences. More importantly, the coefficient on this variable is now consistent with microeconometric evidence on returns to education.

KEYWORDS:

• Education quality
• human capital
• cross-country income differences

JEL CLASSIFICATION:

• O47
• E24
• I29

Acknowledgement

We thank an anonymous referee for helpful comments and suggestions. Financial support from CAPES and CNPq is gratefully acknowledged.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ International standardized tests compare students at the same education level – therefore with the same quantity of education. Differences in average scores can then be interpreted as differences in education quality. It is important to notice that scores are affected not only by school inputs. Family and community characteristics, for instance, can play a role in shaping students’ learning and performance in these tests. Therefore, as in Hanushek and Woessmann (2008), our measure intends to capture education quality in a broader sense, encompassing both school and non-school features.

² Breton (2015) also estimates MRW regressions controlling for education quality. He however uses growth rates as dependent variable instead of GDP per capita levels.

³ Following MRW, we assume that $g+\delta$ does not vary across countries. Specifically, $g+\delta =5\mathrm{\%}$.

⁴ $Y/L$ is real GDP (from PWT 7.1) divided by population aged 15–64 (from the World Bank). MRW original study used PWT version 4.0.

⁵ Specifically, the human capital investment rate ($s_h$) is the secondary gross enrolment rate (from UNESCO) multiplied by the share of population between ages 15–19 (from the World Development Indicators – Health and Population Statistics).

⁶ To compute such averages we use data whenever available within the time frame considered. Data on $Y/L$, investment rates and population growth rates are available on a yearly basis, while data on schooling are available every five years.

⁷ We take the simple average between Science and Maths scores and divide by 10. Our quality measure thus ranges from 0 to 100 (since PISA scores go from 0 to 1000). This strategy is similar to the one employed by Caselli (2005).

⁸ Implied $\beta$’s are larger but not far from values suggested by MRW, which claim it should be somewhere between $1/3$ and $0.5$.

⁹ Regression results are not shown for these cases. We use data on $Y/L$ from the last year of the time frame, and averages over the whole period of estimation for $s_k$, $s_h$ and $n$. When the last year of the period is 2010, we use PISA scores from 2009; when the period ends in 2000, we use PISA scores from 2000.

¹⁰ Data on years of schooling are from Barro and Lee (2013). In this exercise, we use average years of schooling from persons aged 15 or older. We always use data on years of schooling from the last year of the period considered (1985 in the MRW sample).