NN40, May 2008
Education for Sustainable Development? Or The Sustainability of Education Investment? A Special Issue
The Impact of Compulsory Education: a Layman?s Explanation of ?Instrumental Variable? Techniques and Findings from Taipei,China
By Chris Spohr, Asian Development Bank Resident Mission in the PRC
Email: cspohr@adb.org [1]I. Assessing the Value of Education
In the development literature, there is a large body of research to indicate that education, perhaps particularly basic education, matters. In the economics literature, much of this research has focused on estimating the value of, or ?economic returns? to, education in terms of things like adult wages, and generally finds that returns to primary and secondary schooling are larger in developing vis-à-vis industrialized economies.
At the same time, while the basic finding that ?education is good? confirms what we?d all expect, several issues emerge as we look more closely into interpreting the numbers. The first set of these remains within the realm of economics?more specifically the field of econometrics?and deals with how much we can trust the numbers or the ?leap of faith? between correlation and causality. Namely, much of the earlier literature runs statistical regressions of (for example) adult wages versus years of schooling and a range of other variables for which data is available, and interprets the coefficient on schooling as the increase in wages caused by each additional year of schooling. However, such estimates may be confounded by problems such as biases from omitted variables. For example, a child?s intellect (which by itself has implications for career prospects) may influence whether he/she stays in school, what economists term the problem of endogeneity, which would confound any simple attribution that school causes higher wages. Similarly, complex factors like parents? social status clearly affect children?s development prospects above and beyond their parallel affect on children?s schooling; so if the former goes unmeasured, a simple regression may attribute the former effect to the schooling variable. Econometrics has provided an increasingly sophisticated array of tools to deal with such potential biases, including the technique of instrumental variables (IV) noted below.
For the development community more broadly, a second set of issues is whether wage returns ?tell the whole story?, particularly in terms of implications for national policy or project interventions. For example, research (and perhaps more universally practitioners? intuition) suggest that while the benefits of tertiary education accrue largely to the individual (e.g., higher wages), basic education can also yield society-wide and cross-generational benefits. These appear particularly strong in the case of girl?s schooling and female literacy: e.g., evidence from an array of countries almost universally finds that an educated mother invests more in the ?human capital? of her children. The presence of such spillovers or ?externalities? (i.e., which explain why the economist?s ?rational decision-maker? may under-invest in public goods from a society?s standpoint) provides a strong justification for state investment. Of course, in untangling the causal role of formal education vis-à-vis other factors, one faces the same issues noted above.
II. The Case of Taipei,China?s 1968 Reform
While falling short of the hard scientist?s controlled laboratory experiment (or social scientists? analogue of controlled pilot testing of multiple interventions in the field), so-called ?policy experiments? sometimes exist that allow similar comparisons across experimental and control groups of people. Reforms in Taipei,China in 1968 that extended compulsory education from 6 to 9 years (i.e., adding lower secondary schooling), combined with the availability of extensive household data, provide such an opportunity to untangle a more credibly causal effect between education and various workforce and other outcomes.
To purposively oversimplify the perhaps rather daunting econometrics, start in the year 1950 and imagine that all children in a society born in that and subsequent years were assigned and received exactly six years of state-provided primary education. A graphical profile would show a flat line at six years of education across cohorts born in 1950, 1951, and so on. If, in this hypothetical case, the norm were switched to 9 years of schooling for cohorts born in 1955 onwards, the profile would trace out a step (an upward shift). Assuming this were the only major external shock that differentially affected those born before and after 1955, this would present a very easy test for whether education impacts (for example) average age at marriage: if yes, we?d expect to find a similar step-like profile in a graph for that. In either case, while we?d have a difficult time attributing different marital ages for two males born in 1954 to their different educational attainments (rather than for example reflect different family backgrounds), we would feel much more confident in attributing any difference in average marital ages between all males born in 1955 versus 1954 (cohorts who should otherwise be identical on average). This is the essence of the econometric technique of ?instrumental variables? (IV): in this case, we ignore data for individuals? actual years of education (which might conceal factors like family background) and instead only throw into the regression their year of birth as a proxy or instrument for education.
Before departing from hypothetical simplicity, it should be noted that the logic wouldn?t change if average age at marriage were actually increasing across birth cohorts along a constant underlying trend: if education matters, we would expect to find the same step-up but with lines leading into and out of the step sloped according to that trend. Likewise, we?d expect a similar upward shift if average years of education by cohort were subject to constant underlying trend on both sides of the policy shock. Finally, if we wanted to look for the impact of education on other outcomes like wages, we?d have to find a way to untangle factors like respondents? age at the date surveyed. The econometrics gets complicated, but the fundamental focus remains very simple: looking for corresponding jumps in schooling and the outcome variable of interest.
Interestingly, the profile of average years of education for males born in Taipei,China around the 1950s looks remarkably similar to the sloped step conjectured above. While not quite as sharp as the jump in that idealized profile, a cohort-by-cohort plot of average years of schooling for boys shows an upward shift vis-à-vis an underlying trend line starting with the 1955 cohort?precisely those who turned age 13 (the norm for lower secondary school entry) in or just after 1968, the year in which Taipei,China extended compulsory education from 6 to 9 years. After a short transition period, average years of education across birth cohorts return to remarkably similar trend rate.
Did the reform work??Impact on education levels. While uptake was not universal (i.e., some non-compliance), data from household surveys confirm the increases shown in enrolment figures. Interestingly, while girls? schooling had been catching up with boys (rising along a faster underlying trend), the reform appears to have had less impact on girls? schooling: I estimate upward shifts relative to pre-existing trends of more than 0.4 years of education for males and 0.25 years for females in the first six cohorts affected by compulsory junior high schooling. Virtually all of this increase is explained by rising years of lower secondary schooling, which again suggests that it was the reform (rather than, for example, the effects of hidden labor market shifts on returns to education) that drove the increase. While this is somewhat surprising given that lower initial enrolment rates meant that the reform had a larger potential impact on girls? access to lower secondary schooling, but it may be explained by social and cultural factors. Finally, in terms of policy implications for other contexts, it important to note that the reform?s impact on raising enrolments in part traces back to a reinforcing supply-side factor: aggressive investment in school construction in the two years prior to the reform.
So does education really matter? Findings from this policy experiment in Taipei,China suggest a resounding ?yes?. Moreover, while the finding that the reform caused a more muted upward shift in girls? schooling may be somewhat ?disappointing?, the estimates confirm broader evidence and intuition in finding that expanded education had a bigger impact on females, conditional on having received education thanks to the reform (see also below).
First, I find increased wages (economists? typical focus in estimating returns to education), but moreover an increased likelihood that the individual is even engaged in formal wage-based employment [2]. An additional year of schooling is estimated to increase males? likelihood of reporting positive earnings by roughly 2% and, controlling for this selection effect, to raise annual earnings from work by 5.8%. Corresponding effects are stronger for females: the preferred specification indicates a 5.2% increase in the likelihood of reporting earnings, and a rise in earnings of 16.7% per year of additional schooling. Part of this is explained by the reform?s apparent role in shifting men and women who would most likely otherwise have worked in agriculture into industry and service sectors.
To look at broader impacts of girls? schooling associated with the 1968 reform, I used a much smaller, single-year Taiwan Women and Family Survey (1989) and a more complex IV estimation approach. While subject to large standard errors, I find the following effects, which are plausible and align with broader empirical evidence:
(i) age at marriage increases roughly 1-to-1 with additional years of schooling;
(ii) maternity behaviour and investment in children?s human capital?an additional year of mother?s schooling (as a girl) appears to raise the likelihood that her own children will be enrolled in kindergarten by at least 8%;
(iii) effects (albeit imprecisely estimated) on health and old age security: e.g., a year of schooling appears associated with at least an 8% rise in the likelihood a woman will have accumulated any retirement savings by the date surveyed; and
(iv) a statistically significant impact of schooling on a multi-dimensional proxy of empowerment (including women?s autonomous control of their own income).
Looking beyond the numbers: are IV estimates really ?better??
There is much discussion in the economics literature on use of IV or other techniques to avoid biases such as due to the noted problems of endogeneity or unmeasured factors. In this case, by comparing across birth cohorts (in the absence of any other obvious birth year-specific shock) rather than individuals, we can be more confident that our estimates are ?cleaner? and that correlation reflects causality. However, while much less recognized in studies using IV estimation, an equally important subtlety concerns what Imbens and Angrist (1994) call ?local average treatment effects?. Namely, strictly speaking, what IV in this case is measuring is the effect of education on only those persons for whom the reform caused a specific behavioural shift: i.e., we are measuring the impact of lower secondary schooling on precisely those individuals who would not likely have received that schooling in the absence of the policy shift (reflecting, for example, their parents? low income and/or low value placed on education). I would thus argue that IV-based estimation provides a more compelling picture for thinking about policies (in this case or other contexts) on extending compulsory education to the extent these are justified as enhancing equity: i.e., expanding enrolment and providing longer-term benefits to more marginalized groups.
III. Broader Lessons
While the 1968 reform was not pursued in isolation from other policies and shifts, looking back nearly four decades later, Taipei,China?s experience appears to provide strong evidence that universalizing access to good quality and affordable basic education remains fundamentally important to promoting inclusive growth and equitable development.
But what about sustainability? Arguably, the latter needs to be assessed in a different light for public goods like basic education. Whereas cost-recovery may be the best yardstick for services with benefits accruing mostly to individuals, for those with considerable (and often non-monetized) benefits for society as a whole, I would argue that sustainability often boils down, first and foremost, to government commitment. Alongside the noted pre-reform school construction boom, government expenditure on education nearly tripled (in nominal terms) from school years 1965-66 to 1970-71, comprising a growing majority of total (public plus private) expenditures on education, which reached the 4 percent of GDP threshold in the first year of the reform.
Looking across the Taiwan Strait, the People?s Republic in China (PRC) has enacted a reform to phase out fees for nine-year compulsory education in rural areas starting in the poorer Western Region in 2006, an historic move to which the Asian Development Bank is very proud to have contributed via dialogue with Government partners [3]. More generally, for developing countries, it appears clear that adopting (including fully operationalizing and adequately funding) such policies, particularly in poorer rural areas, represents a key step in leveling the playing field and shifting from vicious to virtuous cycles of human capital accumulation and overall wellbeing.
Notes
[1] This piece summarizes the author?s dissertation research, cited in Spohr, C. ?Formal Schooling and Workforce Participation in a Rapidly Developing Economy: Evidence from ?Compulsory? Junior High School in Taiwan?, Journal of Development Economics, Vol. 70/2 (April 2003), pp. 291?327. Content herein reflects the views of the author and not necessarily those of the Asian Development Bank, its Board of Directors, or the countries they represent.
[2] For those affected by the reform, I estimate that a year of additional schooling increased the likelihood of being in wage-based employment by roughly 4% for males and 6% for females.
[3] In March 2008, the PRC subsequently announced the move to eliminate fees for urban compulsory education.
Back to full contents of NORRAG NEWS 40.
Download the full issue of NORRAG NEWS 40 in pdf.