AN ECONOMETRIC ANALYSIS OF THE EFFECTS OF
IQ ON PERSONAL INCOME
BY
ABU ISHAQUE MAHBOOB JALAL
SUBMITTED TO PROFESSORS F. HOWLAND & J. BURNETTE
IN PARTIAL COMPLETION OF THE
REQUIREMENTS FOR ECONOMICS 31
26 APRIL 1999
Abstract
This paper mainly explores whether there is any statistically and economically significant relation between IQ and personal income. Using a sample obtained from National Longitudinal Surveys of Youth (NLSY), it finds a significant relation between a person’s percentile score in an intelligence test called Armed Forces Qualification Test (AFQT) and personal income in the year 1993. It also finds that the influence of percentile IQ on personal income increases as the level of education increases. Moreover, the empirical results show that IQ has influences of different magnitudes on personal income depending on the levels of IQ itself. Thus, it suggests a non-linear relationship between percentile IQ and personal income.
Table of Contents
I. / Introduction / 4II. / Literature Review / 6
III. / Theoretical Analysis / 11
IV. / Empirical Results
A. / The Data / 15
B. / Presentation and Interpretation of Empirical Analyses / 21
V. / Conclusion / 40
Appendix A: Sample Questions: Armed Forces Qualification Test
/ 42Bibliography / 44
I. Introduction
For years economists have been constantly trying to decipher the reasons behind the inequality in the distribution of personal income. One of the strongest, and also the most controversial explanations for the variations in people’s earnings is intelligence, which is measured and often interchangeably referred to by Intelligence Quotient (IQ). With an ever-widening gap between the earnings of the rich and the poor despite seemingly equal opportunities offered by the society, the variations in intelligence level have received an unprecedented amount of attention and momentum in the current century. Limited success of the education programs in alleviating this inequality has only strengthened the initiative to find some factors that is genetically determined, to a large extent beyond the control of the mankind, and possibly account for the differences in the ability of people to earn money. Moreover, extensive use of intelligence tests by educational institutions, employers, and entrepreneurs as a measure of academic excellence and skill has given rise to the concern that a cognitive elite based on intelligence level is being created in the modern society[1].
The relation between IQ and personal income has been under intense scrutiny. Though the positive correlation between these two factors has been demonstrated in many studies, IQ as a determinant of personal income is a problematic idea. A number of factors other than intelligence level have been considered and demonstrated to be highly significant in determining the income of an individual. Therefore, it would be an interesting idea to explore whether IQ or intelligence level has any significant relationship with personal income after controlling for some of the most important of those factors. It will also provide us with an insight on how the relation (if any) between IQ and personal income behaves.
This paper is intended mainly to explore statistically and economically significant relationships between IQ and personal income. To this end, I will first provide a discussion on some of the important studies conducted on this topic in the past. Then I will go on to explaining some analytical backgrounds of my topic. Later in this paper I will present the sample under consideration and the results of the empirical analyses. Finally, I will draw a conclusion on the basis of my findings.
II. Literature Review
The relation between the level of intelligence, usually measured in IQ, and income has been an issue of extensive discussion and research to economists as well as other social scientists. Though very few researchers deny the importance of IQ in shaping different aspects of a person’s life, the main debate centers on the types and magnitudes of these effects. A careful study of the available literature on this topic would easily show different kinds of conceptions about the effects of IQ on a person’s income. They range from the viewpoint that IQ is the major determinant of a person’s earnings to the idea that IQ has a minimal, if any, effect on income. There are also researchers who think that the effects of IQ are important but that it works indirectly in determining income.
In their book The Bell Curve: Intelligence and Class Structure in American Life (1994), psychologist Richard J. Herrnstein and political scientist Charles Murray provide an extensive discussion on the effects of IQ on different aspects of American life. They consider IQ to be mostly an inherited trait and connect differences in intelligence levels to differences in wealth, income, education, unemployment, idleness, injury, family structure, crime and such other issues. In their analyses Herrnstein and Murray use data accumulated through National Longitudinal Survey of Youth (NLSY) – an ongoing survey (starting in 1979) of a nationally representative sample of 12686 people who were between 14 and 23 years of age in 1979. As a measure of cognitive ability[2] they apply percentile scores obtained from an intelligence test taken by all participants in NLSY called Armed Forces Qualification Test (AFQT). It is worth mentioning that AFQT is assembled from an average of four of the ten achievement tests designed to measure proficiency in vocabulary, basic science, arithmetic operations, etc. in an armed forces training program named Armed Services Vocational Aptitude Battery (ASVAB). From their analyses Herrnstein and Murray obtain empirical results showing strong association between low scores in AFQT and being in poverty. Results show that “whites with IQs in the bottom 5 percent of the distribution of cognitive ability are fifteen times more likely to be poor that those with IQs in the top 5 percent.” (Herrnstein and Murray, p. 127) The authors also find poverty, unemployment, and welfare dependency to be more strongly associated with IQ than socioeconomic status (which includes information about education, occupation, and income of the parents of the participants). Using linear logistic model of the form:
logit (p) = Log (p/(1-p)) = + x
in the analysis of NLSY data, they decide that low intelligence translates into a comparatively high risk of poverty. Moreover, Herrnstein and Murray believe that ethnic inequalities could be attributed, to a large extent, to the differences in their levels of intelligence.
The viewpoints expressed by Herrnstein and Murray as well as their methods of analyzing data have been under constant criticism by many researchers. Such a critique is James J. Heckman’s article “Lessons from the Bell Curve” published in Journal of Political Economy (1995). Though the author does not deny the important role of IQ in determining the earnings of a person, he is not ready to accept it to be the most important factor. He provides five main reasons that might disprove the claims of The Bell Curve. First, he finds that AFQT fails to explain a significant portion of the variability in low wages. The highest R2 explained by AFQT is less than 22 percent in log wages. Hence, there must be factors other than IQ that explain a significant portion of the variability across persons’ earnings. Secondly, AFQT scores are confounded by years of schooling. In this context the author mentions the findings of Neal and Johnson (1994)[3] that one more year of schooling can raise AFQT scores by 0.22 standard deviations for men and by 0.25 standard deviations for women. It brings forth the concern that maybe AFQT is not an effective measure of the intelligence of a person. Moreover, the gap of AFQT scores between whites and blacks can almost be eliminated through four additional years of education for blacks. Thirdly, Herrnstein and Murray attribute inadequate importance to the role of education in explaining the differences in income. Here the author quotes the findings of Taber (1994)[4]: “on average, an extra year of schooling … increases earnings by at least a substantial 6-8 percent.” (Heckman, 1111) Fourthly, Heckman doubts the precision of Herrnstein and Murray’s use of the variable that describes socioeconomic status of the participants of NLSY. The AFQT was conducted to persons who were between 14 and 23 years of age in 1979. On the other hand, the variable ‘Socioeconomic Status’ (SES) includes education, occupation, and family income measured in one year. Therefore, it is not likely that one year’s situation will describe 14 to 23 years of socioeconomic status of the participants. Finally, Herrnstein and Murray misunderstand the ability of improvements in technology and management skills. Through the use of better entrepreneurs and technology, even the low-skilled persons could be utilized and included in the labor force. These kinds of changes would make the effects of IQ on income comparatively smaller.
Despite the vehement criticisms of the studies of Herrnstein and Murray, there are many other studies that show a positive relation between the level of intelligence and earnings. Such a study is illustrated in the article “Higher Education, Mental Ability, and Screening” by Paul J. Taubman and Terence J. Wales published in The Journal of Political Economy (1973). Here the authors operationally defined mental ability (or intelligence level) to represent mathematical ability, coordination, verbal ability, and spatial perception. In their analysis, the authors used regression analysis allowing for non-linear effects of intelligence and included only the top half of the mental ability distribution. They used scores from an intelligence test named Aviation Cadet Qualifying Test (ACQT) as a measure of the IQ of the participants. ACQT is composed of seventeen tests that measure abilities such as mathematical and reasoning skills, physical coordination, reaction to stress, and spatial perception. Taubman and Wales (1973) found that of the abilities mentioned above, only the mathematical ability is a statistically significant determinant of a person’s income. It suggests limited overall influence of IQ. They also found that though mental ability has very little influence on earnings in the initial level, the influence grows over time. The rate of growth is higher for persons with graduate training and higher mental ability.
One of the popular explanations offered to account for income inequality is that additional years of schooling (up to a certain level) increase a person’s earnings. However, there are debates whether this education – income relation is overestimated for not including ability differences in the analysis. An attempt to explore this question after controlling for ability is a central topic of the article “Education, Income, and Ability” by Zvi Griliches and William M. Mason published in The Journal of Political Economy (1972). They apply linear regression model on a 1964 sample of U. S. military veterans accumulated through Current Population Survey (CPS). IQ scores from AFQT were used as a measure of intelligence. The authors found that the coefficient of the variable measuring education in the regression equation that did not include ability was 0.0528. After including ability into the equation, the coefficient turned out to be 0.062. The authors considered the difference (only 12%) to be not very significant. Moreover, the results obtained by the authors show very little significance of intelligence level in determining income. It suggests that leaving out ability (or IQ) does not necessarily lead education – income relation to be overestimated. Thus, totally contrary to Herrnstein and Murray's viewpoints, the authors decide,
“If AFQT is a good measure of IQ and if IQ is largely inherited, then the direct contribution of heredity to current income is minute.” (Griliches and Mason, p. S99)
III. Theoretical Analysis
Intelligence generally refers to the ability of a person to adapt effectively to his surroundings and to exploit the available opportunities for his well being. In doing so an intellectual individual brings about changes in himself, tries to change his environment and/or shifts to a new setting. Social scientists agree that this kind of successful adaptation essentially necessitates superiority in a number of cognitive processes – perception, memory, reasoning, learning, creativity, faculty, problem solving etc. However, intelligence is not necessarily considered an excellence in a single ability but an effective combination of the abilities. Similarly, an individual’s income significantly depends on his ability to demonstrate expertise in performing a job. It is theoretically plausible to assume that a person with superiority in those cognitive processes would have a better chance of performing the job efficiently. Thus, an employer would get better return from employing a person with higher intelligence and would be ready to pay more for his service. In other words, since Wage = Marginal Product of Labor, a person with higher cognitive abilities will have higher productivity and thus higher earnings. Therefore, we can assume that if it is possible to measure the intelligence numerically, we will find a positive correlation between intelligence level and personal income.
The most prevalent means of measuring a person’s intelligence level is through Intelligence Quotient or IQ. Most of the intelligence tests today measure abilities such as problem solving, judgment, comprehension, and reasoning. The scores obtained in these tests are computed on the basis of certain statistical distributions (usually Normal Distribution). However, a large quantity of debates has centered on the measurability of intelligence. Intelligence is mostly an abstract idea. It is also considered to be, in large part, genetically determined[5]. Though a statistical measure of a person’s correct responses to an intelligence test is attainable, it is hard to determine conclusively which of the cognitive processes shape intelligence. Thus, no intelligence test can give a definitive picture of a person’s intelligence level. However, most social scientists believe that a well-designed intelligence test can give a good numerical measurement of the intelligence level (obviously with a certain degree of error) of an individual.
How intelligence level influences a person’s income is also subject to an extensive debate. Herrnstein and Murray (1994) offer the idea of the formation of a “cognitive elite” though screening of people on the basis of their intelligence level, who finally end up being highly paid in their jobs. Through anecdotal descriptions of the development of the American society in the second half of the current century, they observe that America is too much dependent on IQ in making its decisions. This screening based on IQ seriously starts at end of the high-school level when students apply to Colleges. Since the number of institutions that offer quality education is remarkably limited, a large number of students compete to get into these few ‘elite’ colleges. These prestigious institutions pick the best and intelligent students depending on IQ scores (such as SAT scores) and interview. When these students graduate, they get into prestigious jobs and earn more money than students of normal intelligence level. Thus, a cognitive class based on intelligence level is formed. On the other hand, the employers always try to employ the best persons they can find for a job. Naturally, a person with higher intelligence level will show better ability to master the job, to adapt to the new settings, to climb up the corporate ladder, and thus to be highly paid. Therefore, it is possible to find a positive relation between IQ and personal income.
The relationship between IQ and personal income is often discounted through the argument that it is almost impossible to unscramble the effects of education and intelligence level on personal income. A person with a higher intelligence level has a better chance of completing higher level of academic education. Moreover, it is an established fact that up to a certain point, one additional year of education increases a person’s earning by a statistically significant amount. Furthermore, education provides an individual with vital knowledge and skill necessary to adapt to the environment. It also trains a person to employ his cognitive processes more effectively.
It is, therefore, necessary to perform an empirical analysis to find out whether intelligence level or IQ has a ‘significant’ effect on personal income after controlling for Education and other confounding variables. A regression analysis is most appropriate in such kind of analysis. A typical multi-variable regression model may be of the form:
Y = 0 + 1 * X1 + 2 * X2 + 3 * X3 + … … … … + n * Xn +
Where, i = Co-efficient parameters of the independent variables
= an error term
Here, a box model is necessary to model the error term . We can use the Standard Econometric Gaussian Error Box Model. However, some assumptions are indispensable for the use of the GEB. We have to assume that the average of the box is zero, the errors are identically distributed, independent of each other, and not correlated with any of the independent variables. It is probable that there are violations of these assumptions in the sample. However, we can easily find out the violations during the empirical analysis and correct as much as possible by using different statistical tools.
Aside from the regression analysis, we can use other statistical methods, such as correlation, elasticity, graphs, etc. to explore the statistical and economic significance of the effects of IQ on personal income.
IV. Empirical Results
A. THE DATA:
The data I will use in my empirical analysis is obtained from National Longitudinal Surveys: Youth 1979 - 1994 Public Codebook: Version 7.0.4. The National Longitudinal Surveys of Youth (NLSY) is conducted by the U. S. Bureau of the Census in cooperation with some other institutions such as U. S. Bureau of Labor Statistics, NORC – University of Chicago, U. S. Department of Health and Human Services, U. S. Department of Defense and Armed Services, U. S. Department of Education, etc. Now the data is gathered by National Opinion Research Council (NORC) under the supervision of the Center for Human Resources Research, Ohio State University. It is an ongoing survey of nationally representative youths who were between 14 and 22 years old in 1979 – the starting year of the survey. The number of participants was initially 12686. The sample includes significant number of participants (more than their national percentile representation) from minority groups such as Blacks, Hispanics, and low-income Whites. This database is particularly interesting in a sense that it is longitudinal and thus helps us follow the changes in the same participants over time. It also allows us explore information about a sample that combines a number of elements that otherwise have to be studied separately.
The dependent variable in my empirical analysis is Personal Income. It shows the amount in dollars each participant received from wages, salary, commissions, or tips from all jobs (except for money received from military service), before deductions for taxes or anything else in the year 1993. It also includes incomes from agriculture, non-firm business, partnership, and professional practice. My sample excludes all the participants who had income of 0 dollars in 1993. It is done mainly to eliminate the effect of a large number of 0 dollars from my analysis. Here it is noticeable that the number of participants with 0 dollars of income in 1993 is 412. Among them a significant portion (186 participants) has completed 12 years of education. Thus, one of the possible reasons for 0 dollars income is that the participants just have finished high school (or dropped out of school or college) and have not got any job yet or doing something else. Moreover, if a person is not in the labor force, we cannot tell what amount he might have earned if working.