Higher Learning, Greater Good: The Private and Social Benefits of Higher Education

Appendixes

331 appendix a Correcting for Ability Bias in Returns to Higher Education In estimating both the market and the non-market returns to investment in higher education, some of the returns are due to innate ability and related family factors. These largely average out for large numbers of families or in nationwide data with exceptions to be discussed. But they are not part of the value-added by a particular campus or part of the pure return to higher education for certain individuals or curricula. These are biased due to ability differences and need to be removed from the increments to earnings and to non-market private and social benefits to remove the bias that would overstate both types of returns to education. In addition to this ability bias, there is also measurement error, especially as schooling levels are self-reported or reported by departments of education, particularly in developing countries. This measurement error due to overstating education quality or enrollment levels operates to increase the estimated return to education and therefore is normally in an offsetting direction from ability bias. The downward correction to the effect of education on earnings or on non-market outcomes for ability must therefore be reduced by an upward correction for measurement error especially to get the net ability bias needed to obtain the true value-added by education. For nationwide or statewide data, these two biases largely cancel out. However, some campuses and some curricula are above the median in the ability of their entering students, and others are below. There is some self-selection by individuals of higher ability in choosing further education, even at the median campus, which is reinforced by some institutions that are very selective in admitting students. The result is that some campuses are above and others below the median ACT or SAT test scores. Some individual students also learn more at college than others. So the size of the correction for net ability bias needs to be related to whether a particular campus, or curricula, or student, is above or below this median. In what follows the issue is addressed of how much the correction for net ability bias must be for those situations that are at, above, or below the median achievement test scores of entering students. 332 appendix a Ability Bias for Individuals, Curricula, or Campuses at the Median All estimates of net ability bias at, above, or below the median ability level are best made using large samples of identical twins. This provides rigorous experimental laboratory-like controls for ability and for other family factors as between genetically identical twins who have grown up in the same household, given that the innate intelligence and family background for within twin pair comparisons is identical. It permits isolation of differences in market and non-market returns to education that are due to differences in their formal education and not to differences in their ability or family background. A significant advance was made recently in studies of identical twins using reports by others to control for measurement error in self-reported schooling to estimate the net ability bias in the returns to schooling. For clarity, we will refer to net ability bias as any (normally upward) bias in earnings due to innate ability as distinguished from earnings after any offsetting measurement error has been netted out. These types of computations were introduced into the literature by Ashenfelter and Krueger (1994) and Behrman et al. (1994). There is wide agreement that identical twins studies offer probably the best basis for estimating the pure returns to education since they provide highly controlled conditions for the identical abilities and family backgrounds between monozygotic twins. This recent evidence to be discussed indicates that ability bias is significant (Behrman and Rosenzweig, 1999, pp. 165–67), but there is also wide agreement that measurement error in an offsetting direction is significant. In practice, it is the net ability bias obtained from identical twins studies in relation to ordinary least-squares applied to raw data without controls for either ability or measurement error that is most important and useful. That is because it is seldom possible to fully control for ability. Even where test scores, such as the widely used SAT and ACT scores, are available, they measure achievement and not innate ability. Other such micro-studies often are not replicated annually or are not representative of the education system. Furthermore, most existing national-, state-, or community-wide data that...