Unobserved Ability and the Return to Schooling

Abstract

IN THE EMPIRICAL LITERATURE, it is customary to estimate the return to schooling by ordinary least squares (OLS) or Instrumental Variable (IV) techniques.2 The choice of OLS is justified only if realized schooling and unobserved labor market ability are uncorrelated. If not, OLS estimates suffer the Ability Bias and other estimation methods, such as IV, may be used. The volume of work devoted to the return to schooling is a good indication of the importance of this topic as perceived by economists.3 Frequently, instrumental variables (IV) techniques are applied in a context where the instrument is only weakly correlated with schooling attainments (Staiger and Stock (1997)). As a consequence, the validity of very high returns to schooling, reported in a simple regression framework, should be seriously questioned; see Manski and Pepper (2000).4 When IV techniques are chosen, the log wage regression is usually assumed to be linear in schooling. However, there is no obvious reason to presume that the local returns to schooling are independent of grade level. Indeed, heterogeneity in any component of schooling choices (subjective discount rates, ability or specific taste for schooling) will lead to improper inference if the local returns are erroneously assumed to be constant. As individuals with lower taste for schooling tend to stop school earlier, OLS (or IV) estimates of the return to schooling, which impose equality between local and average returns at all levels of schooling, will be strongly affected by the relative frequencies of individuals with high and low taste for schooling. More precisely, if there are large differences in local returns between various grade levels, the OLS estimate (measuring an average log wage increment per year of schooling) will tend to be biased toward the co-editor and three anonymous referees for comments and suggestions on this version or earlier versions. Belzil thanks the Social Sciences and Humanities Research Council of Canada for generous funding. The usual disclaimer applies. 2 To use the same terminology as in the reduced-form literature, the return (local) to schooling refers to the percentage wage increase per additional year of schooling. In the paper, the terms local and marginal returns may be used interchangeably. The average return refers to the slope of the straight line between the intercept and the expected log wage at a given number of years of schooling. 3 A World Wide Web survey of the most recent literature indicates that, since 1970, more than 200 published articles or working papers (set in a reduced-form) have been devoted to the estimation of the return to schooling or surrounding issues. Very often, studies based on IV methods conclude that the returns to schooling can be between 20% and 40% above the OLS estimates. Reported estimates around 15% per year (for the US) are not uncommon although standard errors are typically very high. See Card (1999) for a survey. 4 Manski and Pepper (2000) obtain upper bound estimates of the return to schooling from a sample taken from the NLSY. Their results cast doubts on the high returns reported in the literature.