Abstract
A nationally representative dataset from the Philippines is used to derive returns to schooling estimates from instrumental variables, utilizing a supply-side intervention in the education market capable of generating significant changes in schooling. These estimates apply to a subgroup of, mainly, liquidity constrained individuals, in the spirit of the Local Average Treatment Effect (LATE) literature. Returns to schooling estimates which apply to a subgroup of individuals affected by policy intervention may be more interesting from a policy perspective than the return to the ‘average’ individual. The findings are consistent with other recent evidence suggesting that the causal effect of education, at least for certain subgroups of individuals, is as big or bigger than what is suggested by OLS estimates.
Acknowledgements
I would like to thank the Department of Economics, De La Salle University in Manila for making the data available.
Notes
1 Estimates derived using the 1999 Annual Poverty Indicator Survey suggest that the incidence of labour force participation without schooling in this age group exceeds 10%.
2 This statistic is numerically identical to the J statistic derived from a two-step GMM for that equation (see, Baum et al., Citation2003).
3 Under the null hypothesis both estimates are consistent, but the OLS estimate is more efficient. Furthermore, the chosen flavour of the test has the additional advantage of performing better when the instruments are weak (Staiger and Stock, Citation1997; Baum et al., Citation2003).
4 Jointly undertaken by the National Statistics Office of the Philippines, the World Bank mission and the UNDP.
5 Using age 29 as a cutoff, results in a return to schooling estimate of 0.137 and a t-value 2.0, compared to an estimate of 0.151 and a t-value of 2.0 when the age 28 is the cutoff (see section on results).
6 For example, using age 30 as the cutoff results in a point estimate of 0.058 and a t-value of 0.9.
7 Pereira and Martins (Citation2004) use a meta-analysis and provide support for the use of the simple Mincerian function as opposed to an extended specification which includes covariates (such as occupation) whose value can depend on education. Palme and Wright (Citation1998) discuss the suitability of the quadratic approximation versus alternatives such as cubic spline functions.
8 The point estimate of the ‘reform’ dummy did not change when, in addition to all other regressors, 15 region dummies were included.
9 Hausman tests are reported, as is usually the case in other such studies; however, one is cautioned to the fact that a Hausman test may not be very useful, since different instruments estimate different parameters and, therefore, will never result in the same estimate even if the instruments used are valid.
10 Denny and Harmon (Citation2000), in their study for Ireland used the same instrument as in this study (elimination of school fees), along with its interaction with family background but, presumably, did not obtain results based on the sole use of the ‘no fees’ instrument worth reporting. This may be due to a lower variability introduced in the Irish data as a result of the reform, compared to a low income country such as the Philippines.
11 Results are available upon request.