Estimating the returns to education for Australian youth via rank-order instrumental variables
Introduction
Ashenfelter et al. (this issue) discuss how the estimation of the schooling effect on wages is frequently contaminated by the endogeneity of the schooling level. They also discuss the various approaches, and their respective weaknesses, adopted to overcome the endogeneity of schooling choice in wage regressions. As noted by Ashenfelter et al. each of these methodologies has various economic, and/or statistical, assumptions associated with it which are frequently contentious, in terms of their economic or statistical implications, but are crucial to the identification of the schooling effect. Accordingly it is useful to think of an alternative strategy which can be implemented to control for the endogeneity of schooling level in wage equations.
In this paper we illustrate how the rank-order instrumental variables (IV) estimator proposed by Vella and Verbeek (1997)can be applied to estimate the impact of schooling on wages while controlling for the endogeneity of education. We illustrate this using data for Australian youth taken from the 1985 wave of the Australian Longitudinal Survey. The basic idea underlying the rank-order IV estimator is that, like other IV based estimators, the endogeneity of schooling can be accounted for by identifying and subsequently exploiting the similarity of observations. The procedure first requires that the cross-sectional data set can be allocated into various subsets where the choice of subsets must satisfy some conditions which we outline below. The observations are then ordered within each of these subsets on the basis of some measure of the unobserved heterogeneity responsible for the endogeneity of schooling. Moreover, we index observations by their rank-order, in terms of their endowment of unobserved heterogeneity, in their subset of the data and we assume that observations are similar to observations of the same rank-order in other subsets of the data. The effect of education on wages is then identified by comparing individuals in one subset with the individuals in similar areas of the distribution of the unobserved heterogeneity in other subsets. An advantage of this procedure is that while it relies on some alternative identifying restriction, that we discuss below, it can be implemented when the requirements for the methodologies discussed in Ashenfelter et al. above are not satisfied. More explicitly, it does not have special data requirements, such as panel or twins data which require either multiple observations on the same individual or information on each of a set of twins, respectively, and does not rely upon the use of potentially contentious exclusion restrictions. Accordingly, it is potentially of use in this area of research in which it is often difficult to disentangle the direct effects of background characteristics on wages and their indirect effects operating through educational attainment.
Section 2briefly outlines the model and describes the rank-order IV procedure. Section 3describes the data and presents some of the empirical results. Section 4contains the rank-order IV estimates of the impact of schooling on education. Some concluding comments are offered in Section 5.
Section snippets
Rank-order instrumental variables
Consider the following model of wages and educational attainment:where wi and Educationi, respectively denote the log hourly wage and number of years of education of individual i; xi and zi are vectors of individual and work related characteristics; β, δ, θ and γ are parameters to be estimated; and ui and vi are zero mean error terms with a potentially non-zero correlation. This correlation between the disturbances invalidates the use of ordinary
Data
The past two decades have seen substantial changes in the average level of educational attainment of cohorts entering the Australian labor market. This, in part, reflects the changing needs of the Australian economy. However, it also is the product of a concerted effort on the behalf of the Australian Federal Government to increase the high school retention rates of youth, combined with substantial changes in the structure and objectives of the tertiary education system. The most recent of
Estimating the returns to education for Australian youth by rank-order instrumental variables
We now focus on estimation of this model by rank-order IV. As we noted above, a reasonable candidate as our basis for allocating the data into the various subsets is the state in which the individual attended school. To ensure that none of the subsets has a relatively small number of observations we restrict our focus to those individuals who acquired education in one of the states of Australia. This is the reason that our analysis of the education equation excluded individuals who reported
Conclusion
This paper employs the rank-order instrumental variable procedure of Vella and Verbeek (1997)to estimate the returns to education for Australian youth. The attraction of the approach employed is that it can account for the endogeneity of schooling in the wage equation via the use of instrumental variables without the use of exclusion restrictions. Naturally this is accomplished via the use of alternative restrictions although the restrictions employed are likely to be satisfied in this
Acknowledgements
We are grateful to seminar participants at New York University for helpful discussion. We would also like to thank the editors for comments on an earlier draft.
References (9)
- et al.
Selection bias and human capital investment: estimating the rates of return to education for young males
Labour Economics
(1996) - et al.
Estimates of the economic return to schooling from a new sample of twins
American Economic Review
(1994) - et al.
Income, schooling and ability: evidence from a new sample of identical twins
Quarterly Journal of Economics
(1998) - Card, D., 1995. Using geographic variation in college proximity to estimate the return to schooling. In: Christofides,...
Cited by (20)
Estimating a class of triangular simultaneous equations models without exclusion restrictions
2010, Journal of EconometricsEstimating returns to education using different natural experiment techniques
2008, Economics of Education ReviewMicroinsurance and Economic Growth in Africa
2023, RisksEffects of haze pollution on pesticide use by rice farmers: fresh evidence from rural areas of China
2021, Environmental Science and Pollution Research