School policy: implications of recent research for human capital investments in South Asia and other developing countries

Abstract

Concentration on school attainment goals without close attention to school quality has hurt developing countries. Recent evidence shows that individual incomes, the distribution of income, and economic growth rates are all closely related to the cognitive skills of the population. While direct evidence from developing countries is thin, the evidence from developed countries points to the central importance of improving teacher quality in any reform strategies.

Keywords:

• economic growth
• school quality
• cognitive skills
• teacher quality

Acknowledgements

This paper has benefited from ongoing collaboration with Ludger Woessmann and from comments at the Conference on Development Research: Lessons for Indian Policy, Stanford Center for International Development, and at the South Asia Regional Conference on Education Quality. Helpful comments by Michelle Riboud and Geeta Kingdon on an earlier draft improved the paper. Support has been provided by the Packard Humanities Institute.

Notes

1. The Millennium Declaration has eight objectives, initially set by a UN resolution in 2000 and adopted by 189 world leaders during the world summit in 2005: eradicate extreme poverty and hunger; achieve universal primary education; promote gender equality and empower women; reduce child mortality; improve maternal health; combat HIV/AIDS, malaria, and other diseases; ensure environmental sustainability; and develop a global partnership for development.

2. Details of the underlying statistical analyses plus an extended set of references can be found in Hanushek and Woessmann (2008).

3. The Mincer earnings function relates the logarithm of earnings to years of schooling, potential labor market experience, and other factors specific to individual studies (Mincer 1974). The coefficient on years of schooling in this regression can, under specific circumstances, be interpreted as the rate of return to schooling. (See, however, Heckman, Lochner, and Todd [2006], who offer a critique and interpretation of these analyses.)

4. For extensive reviews of the literature, see for example Topel (1999), Temple (2001), Krueger and Lindahl (2001), and Sianesi and Van Reenen (2003). Early studies used adult literacy rates (for example, Azariadis and Drazen 1990; Romer 1990) or school enrollment ratios (for example, Barro 1991; Mankiw, Romer, and Weil 1992; Levine and Renelt 1992) as proxies for the human capital of an economy. An important innovation by Barro and Lee (1993, 2001) was the development of internationally comparable data on average years of schooling for a large sample of countries and years, based on a combination of census or survey data on educational attainment wherever possible and using literacy and enrollment data to fill gaps in the census data.

5. See the general conceptual model in Hanushek (1979) and the review in Hanushek (1986).

6. The clearest analyses are found in several references for the United States (analyzed in Hanushek 2002). See Bishop 1989, 1991; O’Neill 1990; Grogger and Eide 1993; Blackburn and Neumark 1993, 1995; Murnane, Willett, and Levy 1995; Neal and Johnson 1996; Mulligan 1999; Murnane et al. 2000, 2001; Altonji and Pierret 2001; Lazear 2003).

7. Scores are standardized to mean zero and standard deviation one for comparative purposes. A one‐half standard deviation change would move somebody from the middle of the distribution (the 50th percentile) to the 69th percentile; a one standard deviation change would move this person to the 84th percentile. Because tests tend to follow a normal distribution, the percentile movements are largest at the center of the distribution.

8. See, for example, Dugan (1976) and Manski and Wise (1983)) for early analyses. Murnane et al. (2000) separate the direct returns to measured skill from the indirect returns of more schooling and suggest that perhaps one‐third to one‐half of the full return to higher achievement comes from further schooling. Similarly, Rivkin (1995) finds that variations in test scores capture a considerable proportion of the systematic variation in high school completion and in college continuation, so that test score differences can fully explain black‐white differences in schooling. See further discussion and references in Hanushek (2006)

9. See Glewwe (1996), Jolliffe (1998), Boissiere, Knight, and Sabot (1985), Knight and Sabot (1990), Angrist and Lavy (1997), Alderman et al. (1996), Behrman, Ross, and Sabot (2008) and Moll (1998).

10. At the same time, the estimates of the return to years of schooling from models that incorporate families and ability to allow for other inputs to cognitive skills show noticeably lower Mincer returns to school attainment – consistent with the general model of student achievement.

11. As discussed below and in Hanushek and Woessmann (2008), there are some difficult issues in putting these results on a common scale.

12. The rescaling uses performance of US students over time (as measured by the National Assessment of Educational Quality, or NAEP) to calibrate the US scores on different international tests. Then, by setting the variance of each test according to an OECD standardization group, each country and test can be equated. See the further description along with a listing of the separate tests in Hanushek and Woessmann (2008).

13. Details of the data and analysis are found in Hanushek and Woessmann (2008). The source of the income data is version 6.1 of the Penn World Tables (cf. Heston, Summers, and Aten 2002), and the data on years of schooling is an extended version of the Cohen and Soto (2001) data described in Jamison, Jamison, and Hanushek (2007).

14. These results are very close to the estimates by Hanushek and Kimko (2000), which reported estimates in terms of the country level standard deviation that is approximately one‐half as large as the individual level standard deviation; see Hanushek and Woessmann (2008).

15. Note that it is possible to have high rates of return for secondary and tertiary attainment without getting the gains through economic growth. With low quality, the growth effects can be small, even though the people with more school attainment get significantly higher incomes than those with low attainment. In the case of South Asia, see the returns in Riboud, Savchenko, and Tan (2007).

16. PISA is conducted by the OECD. It involves testing a representative group of 15 year olds in each participating country. The tests themselves are designed to measure practical skills rather than deeper conceptual skills. These scores from PISA also enter into the construction of aggregate country measures of cognitive skills used in Hanushek and Woessmann (2008).

17. See Krueger (1999) and Hanushek (1999) on the estimated impacts both from experimental manipulation and from econometric analyses.

18. The actual reform policy is presumed to operate linearly such that, for example, a 20‐year reform that ultimately yielded one‐half standard deviation higher achievement would see the performance of graduates increasing by 0.025 standard deviations each year over the period. It also assumes that the impact is proportional to the average achievement levels of prime age workers, based on workers in the first 35 years of their work life.

19. For a review of existing literature, albeit largely for developed countries, see Hanushek and Rivkin (2004). This paper describes various attempts to estimate the impact of teacher quality on student achievement.

20. The conventional measures of teacher experience and the level of teacher schooling are not closely related to student outcomes except for the first year or two of experience. Neither is teacher certification itself closely related. Hanushek (2003).

21. See, for example, Rivkin, Hanushek, and Kain (2005), Aaronson, Barrow, and Sander (2007), Boyd et al. (2006), and Gordon, Kane, and Staiger (2006) for added examples. Reviews of the US teacher quality research and of the policy implications can be found in Hanushek and Rivkin (2004, 2006).

22. The discussion presumes that quality improvements require changing the stock of teachers. It is possible that this could be done through professional development and training of existing teachers, but currently available evidence does not suggest that such an approach would be very effective.

23. A recent study on India suggests that performance pay could work to improve teachers’ performance in particular country settings; see Muralidharan and Sundararaman (2008)

24. There are of course many reasons for caution. The most important is that estimates of individual teacher value‐added contains substantial measurement error. For an analysis of how this can be integrated into policy, see Gordon, Kane, and Staiger (2006)

25. See Murnane (1975), Armor et al. (1976), and Jacob and Lefgren (2006).

26. These basic estimates are described in detail in Hanushek (2009).

27. These calculations assume that one standard deviation of teacher quality – moving from the center of the distribution to the 84th percentile – is 0.20 standard deviations of student achievement; using a calculation of 0.30 makes these conclusions even more grim.

28. These estimates apply the information on the distribution of teacher effectiveness for each year to a cumulative impact if teachers are improved in all grades. To obtain these steady‐state results, it would be necessary for a student to have a higher average teacher throughout school to Grade 12.

29. UNESCO data may not be the most up to date. According to the most recent household survey data, the primary net enrollment rate may have now reached 95% in India.

30. India also has almost the largest variance in performance, placing it behind just South Africa. Because it is so populace, India then produces a very large number of students at the top of the distribution (even though the proportions of Indian students there is small).

31. Similar concerns exist for mathematics; see Pratham (2005).

32. For other related works on teacher quality, see Hanushek and Rivkin (2007).

33. The innovative experimentation described in Muralidharan and Sundararaman (2008) amply demonstrates both the feasibility and utility of such experimentation. See also Kremer (2003).