What pre-college abilities do educational credentials signal?

136 J. Arkes Economics of Education Review 18 1999 133–141 Table 1 Descriptive statistics. Number of observations 5 1064 Mean Standard deviation Average hourly earnings 11.54 5.72 Age 29.52 0.87 Completed the 8th grade 0.994 0.07 High school years completed 3.57 0.98 College years completed 1.20 1.65 Grad years completed 0.16 0.63 GED 0.07 0.25 High school diploma 0.78 0.41 College attendance 0.52 0.50 Associate’s degree 0.05 0.22 Bachelor’s degree 0.20 0.40 Advanced degree 0.04 0.19 Adjusted AFQT score 0.0 28.85 Experience 10.25 2.47 Experience-squared 115.99 48.41 SMSA 0.79 0.41 Northeast 0.15 0.36 South 0.41 0.49 West 0.19 0.39 White 0.66 0.47 8. mathematics knowledge — knowledge of high school mathematics principles; 9. mechanical comprehension — knowledge of mechan- ical and physical principles and ability to visualize how illustrated objects work; and 10. electronics information — knowledge of electricity and electronics. The scores from four of the 10 tests in the ASVAB — arithmetic reasoning, word knowledge, paragraph com- prehension and numerical operations — constitute the Armed Forces Qualification Test AFQT score. 3 In this analysis, I consider the AFQT score as a comprehensive ability measure since it is a commonly used measure of ability and is found to have high correlations — with a median correlation of 0.38 — with job performance in 23 military occupations Widgor and Green, 1991. Since age affects competency in these areas, I use age- corrected test scores, which are the residuals from a regression of the test scores on age dummy variables Neal and Johnson, 1994. The scores are then transfor- med so that each point represents one percentile in the sample. The educational data come from self-reported attain- ment. Included as indicators of the productive skills acquired in school are the number of high school years 3 The AFQT score is the sum of the arithmetic reasoning score, word knowledge score, paragraph comprehension score and one-half the numerical operations score. ranging from 0 to 4, college years and post-graduate years completed. 4,5 The educational credentials are a high school diploma, college attendance without neces- sarily completing a year of college, an associate’s degree, a bachelor’s degree and an advanced degree. If a person has both an associate’s and a bachelor’s degree, the associate’s degree is disregarded because employers would probably be more interested in the abilities that a bachelor’s degree signals than in those marked by an associate’s degree. Also, employers probably would not consider a person with both degrees to be more pro- ductive than one with only a bachelor’s degree. Finally, this analysis may suffer from attrition bias. The 1993 interview for the NLSY includes 78 of the 14-, 15- and 16-year olds from the 1979 survey who took the ASVAB exam in July 1980 and who had not entered college by early 1980. This analysis assumes that attrition is unrelated to the wages respondents earn or the education they attain. To the extent that this is not true, the results may be biased. 6 4. What pre-college abilities do educational credentials signal? To determine what skills credentials signal, I perform regressions of the ability test scores on educational lev- els. The regressions indicate whether employers can infer any information about workers’ pre-college abilities from the credentials they acquire, over and above what employers learn from workers’ years of education com- pleted. The conditional expectation of a worker’s age- adjusted pre-college ability is modeled as a linear func- tion of his education and race as follows: E[AS,C,R] 5 g 1 Sg 1 1 Cg 2 1 Rg 3 1 where A is an ability measure, S is a vector of measures 4 I calculate post-graduate years as the number of years of college beyond 4 years if the person earned a bachelor’s degree. 5 Along with the detailed educational data in the NLSY come instances of inconsistent reporting of educational attainment. There are cases in which a person has reported: earning a high school diploma, as opposed to earning a GED, without complet- ing the 12th grade; attending college without earning a GED or completing the 12th grade; or earning a bachelor’s degree without having completed at least 3 years of college. I exclude from the sample respondents who fall into these categories. These three restrictions exclude 43, 10 and two individuals, respectively. I do, however, include individuals reporting hav- ing earned a GED and having completed less than the 12th grade, and I include those who report completing the 12th grade, but not earning a GED or a high school diploma. 6 The difference in AFQT scores between the male survivors and other males who dropped out of the sample is only 2.2 percentiles, which is not statistically significant. 137 J. Arkes Economics of Education Review 18 1999 133–141 of the years of education completed, C is a vector of educational credentials, and R is a race dummy variable. The estimated relationships are not causal, for we would expect the skills reflected in the ability test scores to affect a person’s schooling decisions. Instead, the regressions show the association between pre-college ability and schooling. 7 While college likely affects cog- nitive ability, the ability that can be inferred in this regression is only pre-college ability and is different from the skills learned in college. Table 2 presents the results of regressions using the AFQT score and each individual score from the 10 tests in the ASVAB. To interpret the results, the coefficient estimate on a credential, say a bachelor’s degree, is the expected percentile difference in ability test scores between two individuals who have the same years of education completed and are observationally equivalent except that only one has a bachelor’s degree. Completing each year of college is associated with higher pre-college AFQT scores. An employer can infer that someone who completes 2 years of college has 2.26 percentiles higher pre-college ability in addition to the skills acquired during college than someone who only completes 1 year of college. This estimate indicates that, among college attendees who do not graduate, those with higher pre-college ability proceed further in college. It would be difficult to reach similar conclusions for high school based on the estimates on years completed because the test was taken at different points in high school for the respondents. The partial associations between the AFQT score and three of the primary educational credentials — a high school diploma, college attendance, and a bachelor’s degree — are statistically significant at the 1 signifi- cance level. A male with a high school diploma and no college is estimated to have an AFQT score that is 8.64 percentiles higher than an observationally equivalent per- son who completed 4 years of high school without com- pleting the requirements for a diploma. The difference in AFQT scores between a high school graduate who attends college for less than 1 year and a high school graduate who does not attend college is estimated to be 10.65 percentiles. This estimate indicates that merely attending college signals to employers greater pre-col- lege abilities than what is signaled by only graduating from high school. An associate’s degree does not appear to be a market signal of higher ability, as measured by the AFQT score. A possible explanation for this result 7 It has been suggested that I regress the credentials on ability. However, I find Eq. 1 to be the most efficient means to determine the partial associations because it holds the years of school completed constant and indicates the actual differ- ences in ability rather than how much one percentile of ability increases the propensity to acquire a credential. is that a year completed at a 2-year college may be less of a signal of ability than a year completed at a 4-year college. Since one cannot distinguish between a 2-year and a 4-year college education in the NLSY, the associ- ate’s degree may be picking up the lower ability associa- ted with a 2-year college education. The average differ- ence in pre-college AFQT scores between someone who earns a bachelor’s degree and a college dropout, holding constant the number of years of college completed, is estimated to be 9.07. Finally, an advanced degree is asso- ciated with 9.32 percentiles higher AFQT scores. The estimate is significant at the 10 level. As for the 10 individual test scores of the ASVAB, college attendance appears to be the greatest signal of ability, while an associate’s degree appears not to signal higher ability at all. A high school diploma signals higher abilities associated with paragraph comprehension and math knowledge scores more than for any of the other scores. College attendance has a statistically sig- nificant coefficient estimate for each test score, and it signals word knowledge and math knowledge more than other skills. An associate’s degree does not signal sig- nificantly higher ability for any skill. A bachelor’s degree signals abilities associated with the three math tests more than any other abilities, and an advanced degree is only a signal of higher verbal abilities. 5. Do employers value educational credentials because they signal pre-college ability?