A r c h i v e d  I n f o r m a t i o n

Appendix D:
Research on Human Capital and Screening Theories

Accepting the positive relationship between education and subsequent earnings, there is a voluminous literature on why this relationship exists. This appendix reports briefly on that literature. The older explanation, human capital theory, first developed by Becker (1962) and by Schultz (1962), states that skills acquired in school contribute to an individual?s subsequent productivity. In turn, profit maximizing firms pay higher wages to more productive individuals. Alternative explanations, screening and signaling theories, appear in the economics literature in the early 1970?s as economists began to explore the consequences of imperfect information (Arrow 1973, Spence 1973). The screening argument is that employers have imperfect information about the productivity of potential employees. One readily available piece of information is years of education. If potential employees who have traits and abilities that make them productive also get high levels of education, employers can use education as a "screen," increasing the probability of hiring productive workers by hiring only those with high levels of education. Particularly able workers, for their part, realize that employers have imperfect information about which workers will be productive and which will not. Accordingly, they will acquire high levels of education to "signal" their ability to employers. These high-ability individuals also have ability to succeed in school. Low ability individuals might also wish to acquire high levels of education, so that they could get through the screen, but their lack of ability makes it difficult for them to succeed in school. Thus, the argument goes, it is not that schooling imparts any productivity-enhancing skills, but rather that, in a world of imperfect information, schooling identifies those who are inherently more productive. Note that the screening/signaling argument is different from a credentialism argument that says employers hire people with credentials but that the required credentials have little or nothing to do with actual productivity on the job. The screening/signaling approach, like the human capital approach, says that those with credentials are in fact more productive than those without credentials. The difference between the two is that human capital theory says that schooling adds to productivity, while screening theory says that schooling serves to identify productive people.

Because human capital theory and screening theory both explain the positive relationship between education and earnings as resulting from a positive relationship between education and productive ability, it is very difficult to separate the two explanations empirically, and thus to determine which is more correct. There have been a great many studies attempting to do so, using a wide variety of approaches. These cannot be systematically surveyed here. Instead we shall note a limited number of recent papers. An excellent, although not comprehensive, bibliography appears in Weiss (1995). One approach to separating the 2 explanations relates to employer learning. While employers may initially lack information on the productive abilities of their hires, they will over time acquire information. If education does not enhance productivity but only identifies productive people, then education should become less correlated with earnings as job experience accumulates. Pascarella and Terenzini (1991) cite a number of studies taking this approach. The conclusion from this work is fairly clear?education does not appear to decline in importance as experience increases. This approach does not really separate the two explanations, however, because, if education is an effective screen, the better educated are the more productive, and the will thus earn more on average even after employers have more information. Altonji and Pierret (1996) seek to overcome that objection by exploring empirically how quickly employers learn about productivity of workers. They do this by examining how coefficients on proxies for ability, presumably unobserved by employers at initial hire, change over time. If employers are found to learn quickly, the return paid to a signal will be small, hence whatever return is paid to education must be due to accumulated human capital. The authors find that the coefficients on ability proxies rise with experience, suggesting that employers do in fact learn about worker productivity. Using these estimates, they calculate a range of estimates of the return to education assuming education had no productivity enhancing effect. They conclude from their calculations that the signaling component of the return to education is small.

Another recent approach is to examine samples of twins who are raised in the same household. The argument is that abilities and family background are similar, so that any unobserved abilities relating to productivity would be the same for each pair. The relationship between education and earnings thus must relate to the productivity-enhancing effect of the education. Ashenfelter and Krueger (1994) and Miller, Mulvey and Martin (1995) studied large samples of twins in the United Sates and Australia, respectively. Both studies use an instrumental variables approach to minimize omitted variables bias. Both find that the relationship between education and earnings is dominated by the education itself ? i.e. education is productivity enhancing. Ability and family background play relatively little role. Weiss (1995), however, argues that twin studies are not definitive. He notes that a screening model would produce the same result as long as employers do not observe the education choices of both twins in a pair. A screening employers uses an observable variable-years of schooling ? to proxy for an unobserved variable ? capacity to be productive. Not knowing the education level of the twin of the prospective employee, the employer will hire and pay a wage based on the observed variable, schooling, for the individual. The twin with more schooling will get higher wages in the marketplace.

The third recent approach, by Kroch and Sjoblom (1994) seeks to separate the explanations by examining both the absolute amount of education a working has and also the amount of education that workers has relative to his/her peers. They argue that the relative education level, not the absolute education level, of a potential worker is in fact the signaling variable. Assuming that the underlying distribution of unobserved productive ability is unchanged from cohort to cohort, employers infer ability from education level relative to others in the same age cohort, not from the absolute level of education. Thus the authors include both the absolute and the relative education level in their model, reasoning that a strong positive coefficient on the absolute level is indicative of the effect of accumulated human capital, while a strong positive coefficient on the relative level is indicative of the importance of screening. Applying their method to eight different samples, using several specifications, they find that the relative level of education is rarely significant, leading them to conclude that the signaling effect is weak relative to the productivity enhancing effect.

The preponderance of empirical evidence, as represented by the three approaches noted above and other recent work, suggests that the human capital model explains more of the relationship between education and earnings that the screen model. This conclusion must be tempered with caution however. Most of the attempts to separate the explanations are not fully successful, because human capital and signaling "stories" can be told that are consistent with the results. As Weiss (1995) points out, "the [signaling] approach has gained broad acceptance among microeconomic theorists, but many labor economists remain skeptical." He suggests that screening/signaling models are especially good at explaining certain facts?for example, why the variance of wages increases with education, or why there is often a discontinuously large return to completion of high school or college (the so-called "sheepskin" effect). He also cites studies, quite different from any reported in this synthesis, regarding returns to specific cognitive skills and to courses taken in secondary school, which seem to favor a screening explanation. In all of these cases of course, there are at least ad hoc explanations that are consistent with a human capital story as well. Our reading of the evidence at this point is that both processes are at work, but it is our impression that human capital theory explains more of the variance in outcomes such as wages and earning than does screening theory.

[Appendix C: Adjusting Post-1989 ACT Scores]

[ Title Page ]