Previous work and a new interpretation of wage premiums for credentials

134 J. Arkes Economics of Education Review 18 1999 133–141 of the policies that make it difficult for employers to administer ability tests to prospective employees. One other contribution of this paper is that it identifies another educational credential not ordinarily considered as one, namely college attendance. By attending college, a person arguably displays the motivation to learn and to improve himself, thereby distinguishing himself from someone who only acquires a high school diploma. In addition, it may show that some institution that may have access to his cognitive ability test scores has deemed this person capable of completing its academic program. For an ability measure, I use the Armed Forces Quali- fication Test, which was administered to the respondents in the National Longitudinal Survey of Youth NLSY. The test measures a wide range of abilities. It is important in sorting models to consider pre-existing abilities or pre-college abilities in this case. Otherwise, it would be difficult to separate the abilities signaled by college from the productive skills acquired in college. Thus, I consider a sample from the NLSY of respondents who were no older than 18 and had not attended college at the time they took the ability test. The results show statistically significant partial associations between pre- college ability and a high school diploma, college attend- ance, a bachelor’s degree and an advanced degree, but not an associate’s degree. Given that credentials signal certain abilities, employers should value the attainment of credentials based on how much new information is conveyed by cre- dentials about workers’ abilities and how much they value the abilities. For instance, suppose that a bach- elor’s degree signals three traits: competent mathemat- ical abilities, high motivation and good memory. Employers may value only mathematical abilities and motivation. Furthermore, employers may already have an adequate sense of the worker’s mathematical abilities by observing school transcripts. Thus, even though a bach- elor’s degree may signal good memory and high math- ematical abilities, employers would value the degree because it signals motivation, and not because it signals the other traits. To determine the extent to which employers value the attainment of credentials for signaling pre-college abili- ties, I take it as given that firms reward educational attainment in part because a higher education signals qualities that are initially unobservable to employers and that indicate greater productivity. Thus, I employ a sort- ing model of wage determination, which allows edu- cation to have an effect on productivity, as well as serv- ing as a signal of pre-existing ability Weiss, 1995. The results suggest that employers value and reward a bach- elor’s degree in part because it signals abilities reflected in the pre-college ability scores. Results for the other credentials are inconclusive due to large standard errors.

2. Previous work and a new interpretation of wage premiums for credentials

In the Spence 1974 signaling model, individuals enter the labor force with different levels of productivity. Since employers cannot initially discern the true pro- ductivity of workers and since acquiring this information may be costly, employers rely on market signals of pro- ductivity, such as a bachelor’s degree, to determine workers’ wages. The more-productive individuals can distinguish themselves from less-productive individuals by completing a bachelor’s degree or obtaining some other credential, thereby signaling their higher pro- ductivity. In a sorting equilibrium, employers’ prior beliefs on the ability or productivity differences across educational levels are confirmed by the ability and pro- ductivity distributions of new workers. In light of the signaling or sorting model, several authors have attempted to estimate the wage premiums associated with certain degrees: in particular a high school diploma, an associate’s degree, a bachelor’s degree and an advanced degree Hungerford and Solon, 1987; Belman and Heywood, 1991; Frazis, 1993; Hey- wood, 1994; Grubb, 1993; Kane and Rouse, 1995. They include years of education completed and credentials earned in wage regressions. The coefficient estimate on a credential represents the wage difference between cre- dentialled and noncredentialled workers, holding years of education and other factors constant. If wages are pro- portional to expected productivity, then the coefficient estimate on a credential indicates the productivity differ- ences between credentialled and noncredentialled work- ers. It should be noted that the return to acquiring a cre- dential — as opposed to the wage premium for a creden- tial — should only last until the employer determines the worker’s true productivity. At that point, the worker’s wage would be set independently of the credentials he has acquired. Therefore, it would be incorrect to call a wage premium associated with acquiring a credential a return to that credential. The wage premium would only gauge the true return since we do not know how quickly employers learn workers’ true productivities. While workers’ wages would eventually depend solely on their true productivities, in a sorting equilibrium the difference in wages between credentialled and noncredentialled workers would be the same after employers learn the workers’ true productivities as before their productivities are made known, assuming that productivity grows at the same rate for credentialled and noncredentialled workers. A thorny issue in the estimation of wage premiums for credentials is whether to include ability measures in the wage regression. The issue depends on whether one is estimating a sorting model or a human capital model of wage determination. Weiss 1995 notes that the dif- ference between these two models is in how one should 135 J. Arkes Economics of Education Review 18 1999 133–141 interpret the presence of attributes that are unobservable to the firm and correlated with productivity and edu- cation. In a human capital model, if the researcher cannot measure or does not include some ability-type measure that is correlated with educational attainment and pro- ductivity, then the estimated returns to education would be biased, but a sorting model should intentionally omit attributes that are unobservable to the firm and correlated with education and productivity. The premise is that employers cannot observe abilities in potential and new employees, but can only infer workers’ skills by observ- ing their credentials. Thus, part of the return to education is attributable to education signaling the attributes that employers find worthwhile, some of which are unobserv- able to the employers. To include measures of these attri- butes in a wage regression would result in a downwardly biased measure of the wage premium for credentials. Therefore, when estimating the wage premiums associa- ted with credentials, it is important to indicate whether one is assuming that a sorting model or a human capital model applies in order to properly interpret the estimates. An important point from the discussion above is that in a sorting framework, controlling for abilities that are unobservable to employers would result in an estimated wage premium for credentials that is meaningless by itself since it no longer represents the difference in pro- ductivity between the credentialled and noncredentialled workers. In fact, if a researcher were to control for the complete set of abilities that the credentials signal, then the estimated wage premiums for the credentials should be zero. However, the differences in the estimates when controlling and not controlling for abilities are informa- tive since they identify the portions of perceived pro- ductivity differences between credentialled and noncre- dentialled workers attributable to credentials signaling these abilities. Frazis 1993 controls for ability — observed ability and unobserved ability calculated from an ordered pro- bit — to determine if selection bias explains the college degree effect. He finds that including ability measures has little effect on the college degree coefficient estimate. In addition, he interacts the ability measures with years of college to address a hypothesis of Chiswick 1973, that the degree effect results from the fact that those who are receiving more benefits from education are more likely to stay in school and receive their degrees. Again, Frazis finds no evidence supporting this hypothesis. This analysis advances Frazis’ work in several ways. First, Frazis uses mean values to proxy for measurable abilities for respondents who have missing ability data, who comprise 27 of the sample. His method could dampen the explanatory power of these ability measures. In contrast, I exclude these workers since my interest is in whether credentials signal these abilities. Second, whereas Frazis considers a sample of 25-year olds, many of whom had not completed their schooling, this analysis uses a sample of workers aged 28–30. Thus, it is more likely that the respondents in my sample are finished with school. Third, while Frazis only focuses on the bachelor’s degree, I examine the high school diploma, college attendance, the associate’s degree and the advanced degree credentials as well. Finally, I consider the initial stage of the sorting model, analyzing what abilities credentials signal.

3. Data