Thesis
Reference
Schooling, ability, and wages
PASCHE, Cyril
Abstract
During the 20th century, interest among policy makers and academics increased significantly regarding understanding and improving the relationship between schooling, ability, and wages. As a result, attention has since been focused on the determinants of intellectual proficiency and its effect on labor market outcomes and social behavior. However cases of high-IQ people who fail to succeed in life because of their lack of adaptation to expected social behavior are numerous; similarly, low-IQ people who succeed thanks to their motivation, self-esteem, and trustworthiness are equally numerous. When considering improvements in educational performance, policies generally focus on intelligence tests. The excessive focus on intelligence has likely induced researchers to overlook the impact of behavioral traits. This dissertation contributes to the literature by analyzing the importance of both intelligence and attitude within the return to schooling, and by quantifying the changes of the return to schooling, intelligence, and attitude with labor market experience.
PASCHE, Cyril. Schooling, ability, and wages. Thèse de doctorat : Univ. Genève, 2009, no.
SES 691
URN : urn:nbn:ch:unige-65446
DOI : 10.13097/archive-ouverte/unige:6544
Available at:
http://archive-ouverte.unige.ch/unige:6544
Disclaimer: layout of this document may differ from the published version.
1 / 1
Th`ese pr´esent´ee `a la facult´e des
sciences ´economiques et sociales de l’Universit´e de Gen`eve
par CyrilPasche
pour l’obtention du grade de
Docteur `es sciences ´economiques et sociales mention ´economie politique
Membres du jury de th`ese:
Sandra Black, Professeur, UCLA.
Jean-Marc Falter, Docteur, Universit´e de Gen`eve.
Yves Fl¨uckiger, Professeur, Universit´e de Gen`eve, directeur de th`ese.
Marcelo Olarreaga, Professeur, Universit´e de Gen`eve, pr´esident du jury.
Th`ese no 691 Gen`eve, 2009
Th`ese pr´esent´ee `a la facult´e des
sciences ´economiques et sociales de l’Universit´e de Gen`eve
par CyrilPasche
pour l’obtention du grade de
Docteur `es sciences ´economiques et sociales mention ´economie politique
Membres du jury de th`ese:
Sandra Black, Professeur, UCLA.
Jean-Marc Falter, Docteur, Universit´e de Gen`eve.
Yves Fl¨uckiger, Professeur, Universit´e de Gen`eve, directeur de th`ese.
Marcelo Olarreaga, Professeur, Universit´e de Gen`eve, pr´esident du jury.
Th`ese no 691 Gen`eve, 2009
s’y trouvent énoncées et qui n’engagent que la resposabilité de leur auteur.
Genève, le 12 juin 2009.
Le doyen
Bernard Morard
Impression d’après le manuscrit de l’auteur c 2009 by CyrilPasche
Unknown, but often misattributed to Charles Darwin.
General Introduction . . . . 1
Chapter One . . . . 7
Chapter Two . . . . 35
Chapter Three . . . . 57
General Conclusion . . . . 105
Acknowledgements
I owe thanks to many people with the completion of this dissertation. First, I wish to thank my thesis committee as each member has signi…cantly contributed in attempting to make an economist out of me. Yves Flückiger for accepting to be my thesis director and for giving me the taste for research in economics. Yves has an endless ‡ow of enthusiasm, outstanding guidance, and a fair love for risk. He accepted me as his a Ph.D. student based on my bachelor’s work which consisted in tackling the utmost uninteresting question of estimating the elasticity of men’s labor supply.
I met Sandra Black at the European Science Days, in Austria, during the summer of 2006. Three years later her smile has still not disappeared and her unassuming behavior still inspires me. On one of the numerous occasions when I was ready to quit, Sandra told me I needed perspective on what I was working on and have since done my best to follow her advice. Her input in this dissertation, although we were separated by more than nine thousand kilometers, is tremendous.
Jean-Marc Falter has been a true supply of ideas and optimism during these three years.
If I were to give him a franc for every time I walked into his o¢ ce he could be retired by now. I also thank Jean-Marc for initiating me to David Lodge and for his unique vision of the academic world and international conferences. I can’t possibly imagine what my thesis would look like without his input.
Finally, I am grateful to Marcelo Olarreaga for accepting to be the president of my thesis committee. His knowledge in economics, everlasting enthusiasm, and communication skills still astound me.
Other faculty of both the economics and econometrics departments provided help as well.
I am also grateful to the Swiss Leading House in the Economics of Vocational Education for
…nancial support.
Three friends get a special note since they’ve gone beyond the requirements of friendship.
Magali Bourgeois for listening to my provocative ideas and confronting me on just about everything I ever said for over a decade. Magali Guichard for stressing the importance that noncognitive abilities have not only on wages but also on everything in life. Sylvain Weber, for the Guinnesses we shared on late Friday afternoons in the o¢ ce, for riding the wind with me during our Adirondack bike tour, and for his extreme sense of perfection when it comes to doing economics. Once you’re done with your thesis we could open a bank in Speculator, NY. Many thanks also go to Sibylle Ferreira, Giovanni Ferro Luzzi, Joel Fornerone, Nicolas
Grillet, Robert McLaren, Florence Miguet, Paul Quintas, and Oliver Schmitz.
Coralie, thank you for being at my side for virtually this entire thesis and having su¤ered through this experience with me. I’ve studied the e¤ects of schooling and ability on wages in an economical and rational perspective, but doing so without your love would not have been possible.
Most importantly I’d like to thank my parents for their support and love during all these years. To my mom for reassuring me so often and never doubting on my ability to succeed.
And to my dad for his everlasting patience and trustworthiness. Without their support, there is no question in mind that I wouldn’t have …nished my thesis, and so it is to them I dedicate this dissertation.
General Introduction
Overview
During the 20th century, interest among policy makers and academics increased signi…- cantly regarding understanding and improving the relationship between schooling, ability, and wages. As a result, attention has since been focused on the determinants of intellectual pro…ciency and its e¤ect on labor market outcomes and social behavior. However cases of high-IQ people who fail to succeed in life because of their lack of adaptation to expected so- cial behavior are numerous; similarly, low-IQ people who succeed thanks to their motivation, self-esteem, and trustworthiness are equally numerous.
When considering improvements in educational performance, policies generally focus on intelligence tests. When researchers estimate the social return associated with higher educa- tion, the intergenerational transmission of ability, or a production of achievement function, they generally use IQ tests as a measure of general ability and tend to ignore behavioral traits. Cognitive ability has been favored over noncognitive ability for several reasons: we generally lack reliable measures of noncognitive pro…ciency, data that include measures of behavior are scarce, cultural biases in cognitive testing are small, and there is no single dominant factor than can be used as a general measure of noncognitive ability.
The excessive focus on intelligence has likely induced researchers to overlook the impact of behavioral traits. This dissertation attempts to contribute to the literature by analyzing the importance of cognitive ability and noncognitive ability within the return to schooling, and by quantifying the changes of the return to schooling, cognitive ability, and noncognitive ability with labor market experience.
Chapter One
From a policy perspective, understanding the components of the educational premium is a key factor in improving the educational system. Relevant policies on improving educa- tional standards, such as school lotteries, summer school programs, voucher experiments, inter school competition, class size reductions, increases in teacher salaries, and per-student expenditures are based largely on standardized cognitive tests.
With the single exception of a school of Marxist-economists, little empirical research has focused on determining what actually triggers the educational earning premium.1 When
1See Gintis (1971), Bowles and Gintis (1976), Bowles, Gintis, and Osborne (2001), and Bowles and Gintis
available, such research is based implicitly on the unlikely assumption that …rms have per- fect information on the productive ability of their workers. Previous results suggest that noncognitive traits drive the return to education as the labor market needs a docile, obedi- ent, and motivated workforce. Chapter One looks into separating the educational premium into cognitive and noncognitive components, and contributes to the literature by accounting for asymmetric information on the labor market.
Separating ability into schooling and non-schooling measures is an arduous task. Ideally, the data would reveal which abilities were learned in school and which were acquired away from school. I tackle this research questions using Swiss data from the 2003 Adult Lifeskills and Literacy survey. Based on those data, I …nd that when controlling speci…cally for schooling cognitive ability and not just cognitive ability as a whole, approximately half of the return to schooling is related to cognitive ability. This observation contrasts with previous research that found 20% of the return to schooling to be cognitive and 80% to be noncognitive. My results show that schools are places where one learns, or is sorted, on both knowledge and behavioral criteria. My …ndings also suggest that cognitive ability acquired in school is considerably more likely to be rewarded than its non-schooling counterpart. This e¤ect may be attributed to the signaling value of schooling. Such conclusions give weight to current policies that employ cognitive ability tests to assess schooling quality; however they also address the limits associated with relying exclusively on such ability tests when ranking schools.
Chapter Two
While Chapter One focuses on the components of the educational premium, Chapter Two, in joint work with Jean-Marc Falter, examines the changes in the rate of return to education and ability with labor market experience in Switzerland. The Swiss labor market is composed of people who have participated in a vast vocational system. An apprentice spends one to two days a week in school learning the theory related to his …eld and the rest of his 40 hour week with an employer putting those skills into practice. Approximately half of the Swiss working population has a vocational degree. It takes an apprentice between two to three years to complete basic vocational training and an additional three years to earn an advanced diploma.
Labor market conditions for basic vocational degree holders have deteriorated over the past decades. Understanding the mechanisms involved in determining wages on the labor
(2002).
market is of primary importance to policy makers who wish to prevent any further decline in labor market conditions for vocationally educated workers. The social consequences of di¢ culties related to a vocational education system based on dual training would be more severe than for a publicly provided system. Both the supply and demand for vocational education rely on its capacity to attract talented workers.
Using data from the 2003 Adult Lifeskills and Literacy survey, we …nd that on the Swiss labor market, the way cognitive ability is rewarded depends largely on the type of education and that employer learning takes place gradually among workers with a basic vocational degree.2 This pattern seems to be linked with ability utilization in the workplace and suggests a signi…cant under utilization of general ability for individuals with a vocational diploma when they enter the labor market. Advanced vocational education and university education reveal ability in a direct manner, and employer learning thus becomes immediate.
The results for all types of education demonstrate employer learning is public, which suggests that information on the productive ability of workers is common across all …rms. Gradual employer learning is evidence of job mismatches upon entry into the labor market; therefore policies should not focus exclusively on signaling the job related ability of apprentices but should include general cognitive ability as well.
Chapter Three
In Chapter Three I move to US data to estimate a multiple cognitive and noncognitive ability model of employer learning and statistical discrimination. By looking at the change in the rate of return to schooling and ability with experience, I determine whether …rms discriminate on the basis of schooling, quantify the speed of employer learning, and decompose the gains from schooling into a signaling contribution and a human capital contribution.
The previous literature on employer learning and statistical discrimination relies exclu- sively on cognitive ability tests as the sole proxy of unobserved ability.3 Today, there is a great deal of discussion for using a broader, multi-dimensional approach to ability.4 Noncog- nitive abilities may be as important, if not more so, than cognitive ability in explaining labor market outcomes and social behavior. Using data from the National Longitudinal Survey of Youth 1979, I obtain one measure of cognitive ability, the Armed Forces Quali-
2Bauer and Haisken-DeNew (2001) …nd similar results for blue-collar workers on the German labor market.
3See Farber and Gibbons (1996), Altonji and Pierret (2001), Lange (2007), Schönberg (2007), and Ar- cidiacono, Bayer, and Hizmo (2008).
4See Heckman (1995), Heckman (2000), Heckman and Rubinstein (2001), and Heckman, Stixrud, and Urzua (2006).
…cation Test, and two that assess noncognitive abilities, the Rotter Locus of Control Scale and the Rosenberg Self-Esteem Scale. These behavioral Scales are correlated with more general noncognitive traits such as openness, conscientiousness, extraversion, agreeableness, and neuroticism (emotional stability).
My results suggest that …rms learn about all abilities and statistically discriminate on the basis of schooling. Cognitive based measures underestimate both the speed of employer learning and direct productivity e¤ects of schooling, and consequently overestimate the con- tribution of signaling in the return to schooling. A lower contribution of signaling ensures a higher social return to education. Estimates by level of education reveal that college gradu- ates are o¤ered jobs in-line with their abilities while high school graduates are o¤ered jobs that are not matched to their abilities.
Chapter Three is one of the few studies that provides evidence about the signi…cant role played by noncognitive abilities acquired in school when determining wages. Its …ndings suggest that current policies regarding education are based on a fundamental misconception about the abilities that the labor market demands and rewards. Finally, it also shows that, contrary to the existing literature on signaling, education emits a multi-dimensional signal of ability.
References
[1] Arcidiacono, P., Bayer, P., Hizmo, A., 2008. “Beyond signaling and human capital:
education and the revelation of ability.” NBER Working Paper no. 13951, National Bureau of Economic Research, Cambridge.
[2] Altonji, J.G., Pierret, C.R., 2001. “Employer learning and statistical discrimination.”
Quarterly Journal of Economics 116 (1), 316-350.
[3] Bauer, T.K., Haisken-DeNew, J.P., 2001. “Employer learning and the returns to school- ing.”Labour Economics 8 (2), 161-180.
[4] Bowles, S., Gintis, H., 1976.Schooling in capitalist America. Basic Books. 1st edition.
[5] Bowles, S., Gintis, H., 2002. “Schooling in capitalist America revisited.”Sociology of Education 75 (1), 1-18.
[6] Bowles, S., Gintis, H., Osborne, M., 2001. “The determinants of earnings: a behavioral approach.”Journal of Economic Literature 39 (4), 1137-1176.
[7] Farber, H.S., Gibbons, R., 1996. “Learning and wage dynamics.”Quarterly Journal of Economics 111 (4), 1007–1047.
[8] Gintis, H., 1971. “Education, technology, and the characteristics of worker productivity.”
American Economic Review 61 (2), 266-279.
[9] Heckman, J., 2000. “Policies to foster human capital.”Research in Economics 54 (3), 3-56.
[10] Heckman, J., Rubinstein, Y., 2001. “The importance of noncognitive skills: lessons from the GED testing program.”American Economic Review 91 (2), 145-149.
[11] Heckman, J., Stixrud, J., Urzua, S., 2006. “The e¤ects of cognitive and noncognitive abilities on labor market outcomes and social behavior.”Journal of Labor Economics 24 (3), 411-482.
[12] Lange, F., 2007. “The speed of employer learning.”Journal of Labor Economics 25 (1), 1-35.
[13] Schönberg, U., 2007. “Testing for asymmetric employer learning.”Journal of Labor Economics 25 (4), 651-691.
Chapter One
What is it About Schooling that the Labor Market Rewards? The
Components of the Return to Schooling
Abstract
This paper focuses on understanding the link between what schools teach and what the labor market rewards. Research on this topic is scarce and is conducted by a single school of economists. Previous results …nd that schools provide the labor market with a docile, obedient, and motivated workforce. This paper contributes to the literature by removing the assumption of symmetric information on the productive ability of workers from the baseline model. Estimates show the labor market rewards cognitive and noncognitive abilities related to schooling in similar shares. This contrasts with the previous literature and suggests that policies should not only use cognitive tests to rank schools as they do currently, but also incorporate noncognitive tests.
1 Introduction
The majority of the working population spent a considerable amount of their life in school.
Understanding what the labor market pays for in education is valuable for policy makers as it allows a more e¢ cient allocation of educational resources. Matching the ability output of schooling with the labor market demand for ability will potentially increase both the private and public returns to schooling.
Both the human capital and signaling theories hypothesize that schooling and ability are positively related. On the one hand, the human capital theory predicts that “[s]ome schools, like those for barbers, specialize in [the production of] one skill, while others, like universities, o¤er a large and diverse set” (Becker 1994). On the other hand, the signaling theory assumes “[schooling] is productive for the individual, but, it does not increase his real marginal product at all” (Spence 1973). Neither the human capital nor signaling theories specify explicitly which type of ability is acquired or signaled by schooling. However, the human capital literature is based implicitly on the notion that schooling improves general cognitive ability, but it ignores noncognitive ability entirely. In a similar trend, the signaling model assumes that smarter people …nd it less costly to undertake further education and that education emits a signal of one-dimensional ability.
With the notable exception of Gintis (1971), Bowles and Gintis (1976), Bowles, Gintis, and Osborne (2001), and Bowles and Gintis (2002), little research has sought to discover what actually triggers the educational earning premium. Their estimates suggest that less than 20% of the return to schooling can be assigned formally to cognitive ability, and they attribute the balance to noncognitive abilities. If one accepts these results, then schools function primarily as a place where one acquires, or signals, personality traits. Such a …nding is clearly at odds with our educational selection system, which is based predominantly on cognitive tests, such as the SAT, the ACT, and the GRE, our international educational ranking system, such as PISA, and governmental decisions regarding education, such as the No Child Left Behind Act.1
1The Scholastic Assessment Test, the American College Test and the Graduate Record Examination are standardized cognitive achievement examinations in math, reading, writing, and science reasoning that are used for virtually every US college and graduate school admission. Seven out of eight Ivy League colleges currently require cognitive tests in their admission process. The Programme for International Student Assessment is conducted by the OECD and measures ability in reading, math, and problem solving of 15 year old children. The No Child Left Behind Act requires all US public schools to administer ability tests to their students on an annual basis. If a school fails to make adequate yearly progress, it is placed on a list of failing schools.
With respect to the literature, this paper relaxes the assumption of symmetric information on the productive ability of workers. Doing so yields two contributions. First, it shows why previous estimates of schooling’s cognitive component are downward biased and those for the noncognitive component are upward biased. Second, by comparing the schooling cognitive ability coe¢ cient with the non-schooling cognitive ability coe¢ cient, a signaling measure that corroborates the …ndings in the employer learning literature is obtained.
The methodological approach in this paper di¤ers from methods used previously to quan- tify schooling’s components by separating schooling cognitive ability from non-schooling cog- nitive ability. Such a distinction is needed because formal education may not be a perfect screening device or the sole learning environment of cognitive ability, and because the return to cognitive ability may then consequently depend on its origins. The model, therefore, dis- tinguishes people who are highly able, both in and out of school, from those who are solely able according to schooling standards, from those who are able but lack su¢ cient schooling to prove it, and from those who have a low ability in any regard. This paper does not seek to disentangle the signaling and human capital puzzle; instead, it strives to decompose the in …ne components of the return to schooling.
The paper unfolds as follows. Section 2 summarizes brie‡y the relevant literature, Section 3 presents the theoretical model, Section 4 describes the data, Section 5 displays the results, and …nally, Section 6 concludes.
2 Relevant Literature
2.1 Determining the Components of the Return to Schooling
Measuring the cognitive and noncognitive components of the return to schooling requires two wage regressions. The …rst is the basic Mincerian wage equation that regresses earnings on years of education, an expression of labor market experience, and a set of control vari- ables. The second is an augmented expression of the Mincerian wage equation that includes a measure of cognitive ability. Computing the ratio of the years of schooling coe¢ cient, when controlling for cognitive ability, to the years of schooling coe¢ cient, when omitting cognitive ability, yields what Bowles, Gintis, and Osborne (2001) refer to as the "noncognitive" com- ponent of the return to schooling. Using 25 US studies, the authors …nd that on average, controlling for cognitive ability reduces the years of schooling coe¢ cient by 18% and suggest cognitive ability represents less than a …fth of the return to schooling. The remaining 82%
of the return to schooling could be associated with more advanced cognitive ability that is
not captured by basic measures, measurement errors, or noncognitive ability. The authors favor the noncognitive ability hypothesis: “The most obvious potential problem - that the cognitive score might be measured with considerably more error than the schooling variable and hence [the noncognitive component of the return to schooling] is upwards biased- is almost certainly not the case” (Bowles, Gintis, and Osborne 2001), and “these studies pro- vide strong support for the A¤ective Model [noncognitive ability hypothesis], and indicate that cognitive development is not the central means by which education enhances worker success”(Gintis 1971). Bowles and Gintis have a limited audience among economists; despite the controversy concerning their results, they tackle a key research question in education and labor economics.
2.2 Splitting Cognitive Ability per Origin
Ishikawa and Ryan (2002) study the relationship between schooling, schooling cognitive ability, non-schooling cognitive ability, and wages by using the 1992 National Adult Literacy Survey. In a …rst step, they split the total cognitive ability measure between schooling cognitive ability and non-schooling cognitive ability. To do so, they regress the total cognitive measure over the number of years of schooling and schooling type dummies (e.g., primary or high school) and obtain a predicted measure of schooling cognitive ability, where non- schooling cognitive ability is equal to the residual. In a second step, they estimate wages when controlling for schooling type, schooling cognitive ability, and non-schooling cognitive ability. Their results show, for the most part, that it is schooling cognitive ability that a¤ects wages in the US.
3 Model
3.1 Wages and the Components of the Return to Schooling
Quantifying the components of the return to schooling requires two estimates: one that omits a measure of cognitive ability and one that includes such a measure. I begin with the Mincerian wage equation:
wi = 0+ 1Si+ 2Xi+"i: (1)
where w is the natural logarithm of wages,S is the number of years of schooling, while X is an expression for labor market experience and other control variables. The Mincerian return to schooling 1 measures the return to both cognitive and noncognitive abilities,
acquired or signaled by schooling, on log wages w. For ease of exposition, subscript i is removed.
Wage estimates controlling for cognitive ability are usually expressed as follows:
w= 0+ 1S+ 2T CA+ 3X+": (2)
T CA is a measure of total cognitive ability and 2 is its return. T CA measures the cognitive ability an individual possesses without regard to where it was acquired or how it is signaled. As cognitive ability is captured by T CA, 1 becomes the "noncognitive return to schooling" according to Bowles, Gintis, and Osborne (2001).2
Following the line of Bowles, Gintis, and Osborne (2001) the components of the return to schooling are measured as follows:
= 1
1
and = 1 : (3)
is the "noncognitive" component of the return to schooling and its cognitive com- ponent. If schooling in‡uences wages solely by increasing one’s cognitive ability, would be zero. In this case, the years of schooling coe¢ cient, 1 in equation (2) drops to zero when one controls for cognitive ability because the e¤ect of schooling is captured entirely by the cognitive ability variable (i.e., schooling noncognitive abilities are not rewarded). Con- versely, if the e¤ect of schooling on cognitive ability explains none of schooling’s contribution to wages, then is equal to one because the inclusion of the cognitive ability measure does not a¤ect the return to schooling (i.e., 1 = 1).
Anticipating further development, the ratios of equation (3) appear unbiased in only three extreme cases: cognitive ability is exclusively acquired or signaled by schooling and the return to non-schooling cognitive ability is consequently nil; employers have immediate and perfect information on workers’abilities and reward both schooling cognitive ability and non-schooling cognitive ability at the same rate; or schooling does not enhance cognitive ability and its return is consequently zero.
To obtain an unbiased measure of the components of the return to schooling, Bowles, Gintis, and Osborne (2001) suggest “to ask what schools teach and to consider the economic return to the resulting curricular outcomes.” This is what equation (4) does by measuring explicitly the economic return to schooling cognitive and noncognitive abilities:
w= 0+ 1S+ 2SCA+ 3X+": (4)
2There is a broad range of literature on the measurement error of ability. I purposely do not include this literature, as the aim of this paper is to review the previous …ndings of Bowles, Gintis, and Osborne (2001) while removing the assumption of symmetric information on the productive ability of workers.
The years of schooling coe¢ cient 1 now quanti…es the noncognitive return to schooling (the cognitive component of schooling is captured by 2). SCAis the cognitive ability score based on schooling. The noncognitive and cognitive components of the return to schooling, when controlling for schooling cognitive abilities, are approximated as follows:
0 = 1
1
and 0 = 1 0: (5)
The speci…cations of equations (3) and (5), which measure the components of the return to schooling, assume that the average contributions of schooling cognitive ability and schooling noncognitive abilities are quantitatively constant across years of schooling. This assumption holds on the following two arguments. First “schools continually maintain their hold on students. As they "master" one type of behavioral regulation, they are either allowed to progress to the next or channeled into the corresponding level in the hierarchy of production”
(Bowles and Gintis 1976). Across all levels of schooling, individuals acquire noncognitive abilities, from rule-following in primary school to norm internalization in graduate school.
Second, studies with data containing noncognitive ability measures are scarce and generally fail to identify the schooling or non-schooling origin of noncognitive pro…ciency.
3.2 Signaling Measure
The …nal wage equation includes a measure of non-schooling cognitive ability:
w= 0+ 1S+ 2SCA+ 3N SCA+ 4X+": (6)
One interesting feature in equation (6) lies in comparing the schooling cognitive ability coe¢ cient, 2, with the non-schooling cognitive ability coe¢ cient, 3, thus leading to a signaling measure. Schooling cognitive ability and non-schooling cognitive ability are similar in nature as they initiate from the same, original cognitive ability measure included in the data. Their sole dissimilarity is how they are signaled to employers. Di¤erences between 2
and 3 should not be interpreted as a di¤erence in the rate of return, but as a disparity in the odds of them being rewarded by the labor market. Equation (7) informs us on the odds schooling cognitive ability has in being rewarded over non-schooling cognitive ability:
= 2
3
: (7)
is presumably larger than 1 because information is neither free nor obtained immedi- ately. Schooling cognitive ability, 2, is immediately rewarded as it is observed by employers upon hiring. Assuming employers learn about the true productive ability of each worker over
time, non-schooling cognitive ability will gradually be rewarded by the labor market and will decrease over time.
3.3 Cognitive Ability per Origin
The challenge when estimating equation (4) is that a measure of schooling cognitive ability is not available in the data. Progress in understanding what the labor market pays for in schooling is possible, but there is no free lunch, and the identi…cation is not trivial. To obtain such a measure, I borrow from Ishikawa and Ryan (2002):
T CA= 0+ 1ST + : (8)
Equation (8) asks explicitly what schools teach, or signal, in terms of cognitive ability.
Using the coe¢ cients in equation (8), I can obtain a predicted measure of schooling cognitive ability in equation (9). In order to know what cognitive ability is acquired or signaled in school, I do not need to establish a causal relation but solely a correlation.3 Equation (9) informs us that people with a given schooling degree have a given cognitive ability level, but it does not disentangle the signaling and human capital puzzle:
SCA=E(T CAjST): (9)
Non-schooling cognitive ability is equal to the total measure of cognitive ability minus the predicted schooling cognitive ability measure:
N SCA=T CA E(T CAjST) = : (10)
The years of schooling covariate is omitted from equations (9) and (10), and is substituted by the schooling type dummies for two reasons. First, di¤erent curricula that require the
3As mentioned in Ishikawa and Ryan (2002), this estimate is not straightforward if one seeks to obtain a causal relationship because of an endogeneity problem. On the one hand, pursuing further schooling may be a screening process in which only those with higher abilities move on to. On the other hand, those with higher levels of ability may be discouraged from to pursuing further schooling due to the high wages they are o¤ered at their present level of schooling. The two-way relationship between cognitive skills and schooling could bias, either upwards or downwards, the estimate depending on the relative sizes of these counter forces.
Farber and Gibbons (1996) estimate non-schooling cognitive ability using an OLS estimate. The results of Charette and Meng (1998), in which instruments’exogeneity is debatable, suggest the impact of schooling on cognitive ability is underestimated in an OLS framework. Conversely, the results of Glick and Sahn (2009), based on panel data, suggest the OLS and IV schooling estimates are consistent if not identical in magnitude when estimating cognitive ability.
same number of years of schooling may yield di¤erent cognitive pro…ciency. Regressing cognitive ability on the number of years of schooling would omit the di¤erence in the cognitive intensity of educational programs of similar length. Second, using the number of years of schooling as an independent variable for estimating both schooling cognitive ability and wages would lead us to a perfect multicollinearity issue in equations (4) and (6).
One could argue that employers can use tests during the hiring process to measure ability.
In a 20 country international survey Ryan, Mc Farland, and Page (1999) show that the average employee selection is conducted using cognitive ability tests in one out of 10 job selections. The works of Farber and Gibbons (1996), Altonji and Pierret (2001), Lange (2007), and Schönberg (2007) are all based on the assumption that …rms do not observe individual productive ability upon hire. Their results provide strong empirical evidence for this assumption.
3.4 The Omitted Variable Bias - What Can Be Expected?
The aim of this subsection is to show that my results on determining the components of the return to schooling are neither mechanic nor arbitrary but rely on the omitted variable bias properties. I give both a formal mathematical demonstration and a more intuitive graphical explanation as to why previous estimates of the components of the return to schooling are biased in virtually all cases.
Using the omitted variable formula, see Greene (2007), and the schooling and cognitive ability coe¢ cients of equations (1), (2), and (4) I obtain the following equations:
E[ 1j ] = 1+ 2Cov(S; T CA)
V ar(S) + 3Cov(S; X) V ar(S)
= 1+ 2 Cov(S; SCA)
V ar(S) +Cov(S; N SCA)
V ar(S) +
= 1+ 2[ + ] + ; (11)
E[ 1j ] = 1+ 2Cov(S; SCA)
V ar(S) + 3Cov(S; N SCA)
V ar(S) + 4Cov(S; X) V ar(S)
= 1+ 2 + 3 + : (12)
I make three assumptions based on the variables in equations (11) and (12):
Schooling and cognitive ability are assumed to be non-negatively correlated; conse- quently,Cov(S; SCA)=V ar(S) (de…ned as ) is larger than, or equal to, zero.
Years of schooling and non-schooling cognitive ability are uncorrelated; consequently, Cov(S; N SCA)=V ar(S) (de…ned as ) is nil. This assumption is nevertheless not straightforward as four situations may occur: (1) advanced schooling and high non- schooling cognitive ability ( > 0), (2) advanced schooling, yet low non-schooling cognitive ability ( < 0), (3) little schooling, yet high non-schooling cognitive ability ( < 0), and (4) little schooling and low non-schooling cognitive ability ( > 0).
Relaxing this assumption yields some inde…nite solutions; therefore, I prefer to restrain myself to the case where = 0.
3
Cov(S;X)
V ar(S) (de…ned as ) is equal to 4Cov(S;X)
V ar(S) (de…ned as ). Controlling for total cognitive ability or jointly for schooling cognitive ability and non-schooling cognitive ability does not in‡uence the control variable coe¢ cients (e.g., years of experience, country of birth, etc.).
Making use of the previous assumptions and the equality of equations (11) and (12):
E[ 1j ] = 1+ 2 + 2 + = 1+ 2 + 3 +
=) 1+ 2 = 1+ 2
=) 1 1 = ( 2 2) : (13)
The return to total cognitive ability, 2, is equal to the weighted return to schooling cognitive ability, 2, and to non-schooling cognitive ability, 3.4 Depending on the di¤erent rate of return to schooling cognitive ability and non-schooling cognitive ability, …ve situations may occur as shown in Table 1.
<Insert Table 1>
In Table 1, the cognitive and noncognitive components of the return to schooling, mea- sured in Bowles, Gintis, and Osborne (2001), are accurate in just three cases (i.e., 1 = 1).
In the …rst case, situation A, cognitive ability originates solely from schools and people have no non-schooling cognitive ability. Consequently SCA = T CA, 3 = 0 and 2 = 2; therefore, 1 = 1. In the second case, situation C, employers have perfect and immediate information on non-schooling cognitive ability and therefore the return to schooling, non- schooling, and total cognitive abilities are identical. Consequently, 2 = 3 = 2and 1 = 1.
4 2= 2T CASCA + 3N SCAT CA , withT CA=SCA+N SCAand SCA; N SCA 0.
Finally, in situation E, schooling yields no cognitive ability whatsoever and = 0. As cog- nitive ability is orthogonal to years of schooling, the inclusion of the former does not a¤ect the return to the latter. Consequently,N SCA=T CA, 2 = 0and 3 = 2, leaving 1 = 1. In a more general and realistic setting, situation B occurs when the return to schooling cognitive ability is larger than the return to non-schooling cognitive ability ( 2 > 2 >
3 > 0 and therefore 1 > 1). This may occur when schooling has a positive signaling value and when employers learn on non-schooling cognitive ability with experience. Were situation B to occur, the noncognitive component of the return to schooling is overestimated when one simply controls for total cognitive ability; conversely, the cognitive component is underestimated. This happens because the e¤ect of schooling cognitive ability on wages is underestimated due to the relatively lower return of non-schooling cognitive ability.
Situation D is the improbable case where the return to non-schooling cognitive ability is larger than the return to schooling cognitive ability ( 3 > 2 > 2 > 0 and 1 > 1). This may arise if the schooling system is an ine¢ cient place to acquire or signal cognitive ability and the labor market trusts non-schooling cognitive ability over schooling cognitive ability.
In this situation, the noncognitive component of the return to schooling is underestimated when one controls for total cognitive ability; conversely, the cognitive component is overes- timated. This occurs because the role of schooling cognitive ability is overestimated due to the relatively higher return of non-schooling cognitive ability.
<Insert Figure 1>
In Figure 1 the curve plots the relative mis-measurement of the components of the return to schooling for the di¤erent relative returns in schooling cognitive ability over non-schooling cognitive ability. Situations above the horizontal continuous line represent cases where the return to schooling cognitive ability is larger than the return to non-schooling cognitive ability while situations below the line represent reverse cases. To the left of the dashed line, the noncognitive component of the return to schooling is underestimated according to the method used in Bowles, Gintis, and Osborne (2001) while to the right of the dashed line, the noncognitive component of the return to schooling is overestimated.
Anticipating empirical results and making use of the signi…cant evidence of employer learning on unobserved productive ability found in Farber and Gibbons (1996), Altonji and Pierret (2001), Lange (2007), and Schönberg (2007), I favor situation B in which schooling cognitive ability is better rewarded than non-schooling cognitive ability, yet with both re- turns being positive. Consequently, 2 > 2 > 3 > 0 and 1 < 1. Because the researcher
has access to better information on the productive ability of workers than …rms, he under- estimates the monetary return of schooling cognitive ability on wages and biases measures of the noncognitive component of the return to school upwards.
4 Data
The Adult Literacy and Lifeskill Survey (ALL) is a cross-section international comparative survey designed to assess the literacy (prose and document) and numeracy of the adult pop- ulation. The 2003 survey was conducted in the Bermudas, Canada, Italy, Norway, Switzer- land, the United States, and the Mexican state of Nuevo Leon. The full Swiss sample is comprised of 5’120 individuals. The data was collected in a two part, face-to-face interview.
The …rst part is a 45 minute, nine theme questionnaire on schooling and citizenship, lin- guistic information, parental information, labor force information, literacy and numeracy, adult schooling and training, numeracy practices, information and communication technol- ogy literacy, and household information. The second part is a written cognitive ability test in prose, document, and numeracy ability. The test is graded on a continuous scale, and questions re‡ect the daily challenges individuals must confront. All three cognitive ability measures are highly correlated and introducing them jointly yields inconsistent results. The use of a single cognitive ability factor is traditional in the literature. The total cognitive ability measure used throughout this paper is the arithmetic average of prose, document, and numeracy ability.
The …nal sample, restricted to individuals having worked without interruption during the 12 months preceding the interview, is comprised of 1’146 men and 984 women. For part-time workers (less than 40 hours a week) a full-time (40 hours a week) standardized wage is computed. The descriptive statistics in Table 2 show that men undergo more years of schooling and have a higher total cognitive ability score. As predicted by the human capital and signaling theories, schooling and cognitive ability are positively related as shown in Table 3.
<Insert Tables 2 and 3>
As stated by Ishikawa and Ryan (2002), and Green and Riddell (2003), cross-section data, such as the one I use, generally lack variables to instrument the potential endogeneity of the years of schooling variable. Therefore, I prefer not to use instruments rather than to force results out of bad ones.
5 Empirical Results
5.1 Separating Cognitive Ability by Sources
The results of regressing schooling variables on total cognitive ability are displayed in Table 4. Ten schooling dummy variables are included in the regression: general culture school (GCS), basic vocational training (BVT), high school (HS), teaching program (TP), advanced vocational training (AVT), applied science school (ASS), bachelor degree (BAC), master degree (MAS), Ph.D., and other. The base category is composed of people not having pursued further schooling beyond compulsory education.
<Insert Table 4>
As predicted by both the human capital and signaling theories, Table 4 shows a strong and positive correlation between cognitive ability and schooling diplomas. Yet schooling covariates explain just a …fth of the variance in cognitive ability. This …nding is consistent with previous research based both on children and adults. For example, “[a]cross almost all the speci…cations considered, we found that mother’s accumulated ability, as measured by the AFQT [Armed Forces Quali…cation Test], and home inputs (contemporaneous and lagged) are substantive determinants of children’s test scores in math and reading”(Todd and Wolpin 2007). Further, “[t]he picture that emerges suggests a powerful role for environment in shaping individual IQ”(Dickens and Flynn 2001). Controlling for schooling cognitive ability when quantifying the cognitive and noncognitive components of the return to schooling (and not just total cognitive ability) is closer to reality because it accounts for what schooling truly yields and allows asymmetric information to exist on the labor market.
5.2 The Components of the Return to Schooling and Signaling
To compare both estimates of schooling’s components and thus obtain a signaling measure, I run four wage regressions. Model I, equation (1), the baseline model, is the standard Mincerian wage equation. Model II, equation (2), adds the total cognitive ability measure.
The schooling coe¢ cients of Models I and II allow me to measure the components of the return to schooling according to the model of Bowles, Gintis, and Osborne (2001). Model III, equation (4), controls for the same variables as Model I as well as for schooling cognitive ability. Comparing the schooling coe¢ cients of Models I and III yields the components of the return to schooling according to the method developed in this paper. Model IV, equation
(6), includes a measure of non-schooling cognitive ability. This allows me to compare the schooling cognitive ability coe¢ cient with the non-schooling cognitive ability coe¢ cient.5 5.2.1 Male Estimates
In Model I of Table 5, an additional year of schooling increases wages by 7.8%. This return encompasses the return to both the cognitive and noncognitive abilities yielded by an ad- ditional year of schooling. All things being equal, maximum wage is reached after 32 years of labor market experience. Similar results are found when using data representative of the full population.
<Insert Table 5>
As predicted by Bowles, Gintis, and Osborne (2001), there is a small, yet statistically signi…cant drop in the years of schooling coe¢ cient between Models I and II. If one assumes all the cognitive ability people possess is acquired or signaled by schooling, then the noncognitive curriculum of an additional year of schooling enhances wages by 6.7%. Consequently, the cognitive ability related to an additional year of schooling only increases wages by a mere 1.1% (7.8%-6.7%). Model III drops the assumption that cognitive ability is acquired entirely in school and controls for cognitive ability that originates only from a schooling environment.
The years of schooling coe¢ cient is now a mere 3.9%, half its initial value. The schooling cognitive ability coe¢ cient is considerably larger than the total cognitive ability coe¢ cient.
This is because schooling cognitive ability is better rewarded, due to the educational signal, than similar non-schooling cognitive ability.
<Insert Table 6>
Table 6 reports the cognitive and noncognitive components of the return to schooling, which is measured using both the method of Bowles, Gintis, and Osborne (2001) and the one developed in this paper. According to the Bowles, Gintis, and Osborne (2001) model, 86% of what the labor market rewards in schooling is noncognitive. If one accepts this result, then schools are a place where people acquire noncognitive abilities or are sorted according to noncognitive ability criteria. Such an important noncognitive component also suggests that using cognitive ability tests as the sole measure of schooling quality is somewhat careless
5Robustness checks were conducted for all estimations by including, both separately and jointly, a blue- collar dummy, nine activity dummies and …fteen industry dummies. Schooling and ability coe¢ cients remain signi…cant but are smaller in size. The conclusion of this paper remains unchanged.
as cognitive ability represents less than 15% of the private return to schooling. Policies that focus on improving cognitive pro…ciency would have an elusive e¤ect on labor market outcomes.
The values of 0 and 0 in Table 6 show that 50% of the return to schooling is cognitive.
The average marginal return on wages of an extra year of schooling is 8%; half is due to cognitive ability, and the other half is due to noncognitive ability. This result o¤ers an intermediate answer to the provocative results of Bowles, Gintis, and Osborne (2001) on the fact that schools provide a docile workforce, and the (over)focus of policies, such as PISA and the No Child Left Behind Act, which center almost exclusively on cognitive pro…ciency.
The inclusion of non-schooling cognitive ability in Model IV has little e¤ect on the years of schooling and schooling cognitive ability coe¢ cients. When comparing the cognitive abil- ity coe¢ cients, one sees that schooling cognitive ability is three times more likely to be rewarded on average than non-schooling cognitive ability (=.1826/.0670). Cognitive ability that is identical in nature but originates from di¤erent areas is rewarded at totally di¤erent rates. Unreported estimates show that the interaction term of non-schooling cognitive abil- ity with years of experience is positive and statistically di¤erent from zero. Conversely, the interaction of schooling and schooling cognitive ability with years of experience is negative and statistically di¤erent from zero. This suggests that employer learning takes place on the labor market as predicted by Farber and Gibbons (1996), Altonji and Pierret (2001), Lange (2007), and Schönberg (2007). Initial wages are set on the basis of the predicted ability of workers (via the educational signal). As …rms learn about their workers through experience, wages are set by the true productive ability of each employee, and the e¤ect of schooling in determining wages declines.
5.2.2 Female Estimates
Results on female wage estimations are always subject to selectivity biases and years of ex- perience mis-measurements. Despite these caveats, the comments expressed for men remain valid for women and con…rm previous results on the components of the return to schooling.
When a measure of total cognitive ability is included, the years of schooling coe¢ cient drops by less than 10%. That decrease is 56% when I control for schooling cognitive ability, which suggests that less than half the return to schooling is noncognitive.
The return rate of schooling cognitive ability relative to non-schooling cognitive ability is considerably higher than for men, at 4.9 versus 2.7. A possible explanation may be that on average, women undergo shorter spells of employment and that the learning on their
non-schooling productive ability is consequently limited. Therefore, the educational signal and the abilities related to schooling play a far more important role in setting wages.
<Insert Table 7>
The cognitive component of the return to schooling shown in Table 8 is considerably higher than what is found when using the method advanced by Bowles, Gintis, and Osborne (2001). These results suggest that more than half of what the labor market rewards in schooling is cognitive.
<Insert Table 8>
5.2.3 Multicollinearity Measures
A potential drawback when one controls for years of schooling, total cognitive ability, and schooling cognitive ability is whether the variables are multicollinear. To be on the safe side, I measure the variance in‡ation factors (VIF) in all four estimations. A measure of VIF involves examining the R2 by regressing each independent variable against all the others.
The rule of the thumb, suggested by Chatterjee and Hadi (2006), is that the VIF value for each variable should remain below 10. In the absence of any linear relation between the independent variable, the VIF is equal to one.
<Insert Table 9>
Models I and II are canonical wage estimates, and their mean VIF is between 3.8 and 4.1; the mean VIF of Models III and IV is within the same range. The inclusion of the schooling cognitive ability and non-schooling cognitive ability measures does not "load" the model with multicollinearity. The VIF of years of schooling and schooling cognitive ability remains well below the critical threshold of 10.
6 Conclusion
Little research has addressed the essential question of knowing what it is about schooling that the labor market rewards. The answer given by a sole group of economists has been noncognitive abilities: schools help produce a well behaved workforce for the labor market.
The question matters politically because it bears on the consideration of what schools should teach and the content of schooling quality tests.
This paper provides theoretical proof that previous measures of the components of the return to schooling were biased in favor of noncognitive abilities. My empirical estimates show that the return to schooling is composed both of cognitive ability (e.g.,the capacity to process information and apply knowledge) and noncognitive abilities (e.g., behavioral and personality traits) in equal shares. My results consequently challenge previous research, such as Gintis (1971), Bowles and Gintis (1976), Bowles, Gintis, and Osborne (2001), and Bowles and Gintis (2002), which suggests the return to schooling is predominantly noncognitive.
Stated di¤erently, my estimates ensure that the capacity to process information and apply knowledge that originates from schooling is largely rewarded by the labor market. Con- sequently, less space is dedicated to personality traits. Measures also show that cognitive ability acquired or signaled via schooling diplomas is several times more likely to be rewarded than similar cognitive ability acquired elsewhere.
The ability measure used in this paper is one of basic cognitive ability, suggesting the cog- nitive component of the return to schooling measured here may be a lower bound. Advanced cognitive ability, largely job dependent, such as rapid matrix ‡ipping for econometricians or neat snipping for barbers, are bound to increase the cognitive component of the return to schooling.
My …ndings have direct policy implications, as they both validate the use of cognitive ability tests as a measure of schooling quality and promote cognitive ability to take a conse- quent share of schooling curricula. Simultaneously, these results suggest that our educational system may place an excessive focus on the importance of cognitive test scores, such as the SAT, the ACT, the GRE, or PISA. Noncognitive abilities are important on the labor mar- ket, and the development of pro-cognitive policies should not be detrimental to noncognitive abilities.
References
[1] Altonji, J.G., Pierret, C., 2001. “Employer learning and statistical discrimination.”Quar- terly Journal of Economics 116 (1), 313-350.
[2] Becker, G., 1994.Human capital: a theoretical and empirical analysis, with special refer- ence to education. University of Chicago Press. 3rd edition.
[3] Bowles, S., Gintis, H., 1976.Schooling in capitalist America. Basic Books. 1st edition.
[4] Bowles, S., Gintis, H., 2002. “Schooling in capitalist America revisited.”Sociology of Education 75 (1), 1-18.
[5] Bowles, S., Gintis, H., Osborne, M., 2001. “The determinants of earnings: a behavioral approach.”Journal of Economic Literature 39 (4), 1137-1176.
[6] Charette, M., Meng, R., 1998. “The determinants of literacy and numeracy, and the e¤ect of literacy and numeracy on labour market outcomes.”The Canadian Journal of Economics 31 (3), 495-517.
[7] Chatterjee, S., Hadi, A., 2006. Regression analysis by example. Wiley-Interscience. 3rd edition.
[8] Dickens, W., Flynn, J., 2001. “Heritability estimates versus large environmental e¤ects:
the IQ paradox resolved.”Psychological Review 108 (2), 346-369.
[9] Farber, H., Gibbons, R., 1996. “Learning and wage dynamics.”Quarterly Journal of Economics 111 (4), 1007-1047.
[10] Gintis, H., 1971. “Education, technology, and the characteristics of worker productivity.”
American Economic Review 61 (2), 266-279.
[11] Glick, P., Sahn, D., 2009. “Cognitive skills among children in Senegal: disentangling the roles of schooling and family background.”Economics of Education Review 28 (2), 178-188.
[12] Green, D., Riddell, C., 2003. “Literacy and earnings: an investigation of the interaction of cognitive and unobserved skills in earnings generation.”Labour Economics 10 (2), 165- 184.
[13] Greene, W., 2007. Econometric analysis. Pearson. 6th edition.
[14] Ishikawa, M., Ryan, D., 2002. “Schooling, basic skills and economic outcomes.”Eco- nomics of Education Review 21 (3), 231-243.
[15] Lange, F., 2007. “The speed of employer learning.”Journal of Labour Economics 25 (1), 1-35.
[16] Ryan, A.M., Mc Farland L., Page, R., 1999. “An international look at selection practices:
nation and culture as explanations for variability in practice.”Personnel Psychology 52 (2), 359-391.
[17] Schönberg, U., 2007. “Testing for asymmetric employer learning.”Journal of Labor Economics 25 (4), 651-691.
[18] Spence, M., 1973. “Job market signaling.”Quarterly Journal of Economics 87 (3), 355- 374.
[19] Todd, P., Wolpin, K., 2007. “The production of cognitive achievement in children: home, school, and racial test score gaps.”The Journal of Human Capital 1 (1), 91-136.
Table 1: The Mis-Measurement of the Noncognitive Component of the Return to Schooling.
Situation SCA NSCA
A 2 > 3 = 0 2 = 2 1 = 1 >0 =0
B 2 > 3 >0 2 > 2 1 < 1 >0 >0
C 2 = 3 2 = 2 1 = 1 >0 >0
D 3 > 2 >0 2 < 2 1 > 1 >0 >0
E 3 > 2 = 0 2 < 2 1 = 1 =0 >0
Note: The Situations refer to Figure 1. 2 and 3 are the return rates to schooling cognitive ability (SCA) and non-schooling cognitive ability (NSCA). 1 captures the noncognitive return to schooling according to Bowles, Gintis, and Osborne (2001) and 1 according to the method developed in this paper. 2 measures the return to total cognitive ability.
Situation E requires cognitive ability to be unrelated to schooling ( = 0).
Table 2: Summary Statistics.
Male Female
Variable Mean SD Mean SD
Years of schooling 14.56 3.28 13.78 3.30
ln(annual wage) 11.31 .56 11.05 .52
Total cognitive ability 290.21 37.43 282.19 34.35 Years of potential experience 21.46 11.43 21.55 11.44
French speaking (%) 29.4 - 35.7 -
Italian speaking (%) 24.5 - 22.1 -
Born in Switzerland (%) 81.8 - 81.1 -
Father born in Switzerland (%) 70.9 - 69.0 - Mother born in Switzerland (%) 68.8 - 65.7 - Father university degree (%) 21.9 - 22.1 - Mother university degree (%) 5.9 - 7.0 -
Note: Total cognitive ability is the average score in prose, document, and numeracy. The score ranges from 0 to 500. Statistics are based on the unweighted Swiss sample of the Adult Literacy and Lifeskill Survey (ALL).
The sample size is 1’146 males and 984 females.
Table 3: The Number of Years of Schooling and Total Cognitive Ability per Schooling Type, Summary Statistics.
Years of Schooling Total Cognitive Ability
Male Female Male Female
Variable Mean SD Mean SD Mean SD Mean SD
Junior high school 10.00 1.56 9.62 1.46 253.73 36.68 252.47 33.44 General culture school 12.28 1.27 12.33 1.33 284.72 40.94 268.11 29.49 Basic vocational training 12.75 1.73 12.23 1.70 276.14 33.64 274.45 29.50 High school 13.79 1.73 14.44 2.64 294.13 42.36 288.51 36.18 Teaching program 15.72 2.26 14.48 1.38 301.58 34.54 297.07 29.75 Advanced voc. training 14.79 2.36 14.67 2.32 299.33 29.33 292.65 31.15 Advanced science school 16.19 2.11 15.52 2.24 309.97 34.45 285.64 38.83
Bachelor 17.24 1.75 16.88 1.39 305.42 35.51 289.05 32.56
Master 18.43 2.09 18.36 2.19 311.34 31.85 305.45 26.82
Ph.D. 20.78 3.14 21.10 2.78 318.85 25.21 307.60 38.53
Other 14.92 2.68 14.22 3.67 272.89 40.20 288.7 27.31
Note: Total cognitive ability is the average score in prose, document, and numeracy. The score ranges from 0 to 500. Statistics are based on the unweighted Swiss sample of the Adult Literacy and Lifeskill Survey (ALL). The sample size is 1’146 males and 984 females.
Table 4: The E¤ects of Schooling Type on Total Cognitive Ability, OLS estimates.
Male Female
General culture school 30.9899 15.6347 (10.5205) (6.0147) Basic vocational training 22.4093 21.9757 (4.9380) (3.8249)
High school 40.3974 36.0363
(7.2887) (5.2247)
Teaching program 47.8528 44.6005
(8.2562) (5.1600) Advanced vocational training 45.6020 40.1780 (5.1111) (4.6110) Applied science school 56.2407 33.1708 (6.1377) (8.1729)
Bachelor 51.6878 36.5796
(7.7023) (7.4313)
Master 57.6129 52.9731
(5.3777) (4.3495)
Ph.D. 65.1208 55.1236
(5.7758) (7.8992)
Other 19.1594 36.2119
(12.1068) (9.3240)
Constant 253.7311 252.4739
(4.6803) (3.5248)
R2 .2078 .1817
Note: The base category is junior high school. The dependant variable is is the average score in prose, document, and numeracy. The score ranges from 0 to 500. The standard errors in parentheses are robust. The sample size is 1’146 males and 984 females.
Table 5: The E¤ects of Schooling and Cognitive Ability on Log Wages, Male Sample, OLS estimates.
Model: I II III IV
Years of schooling .0779 .0671 .0392 .0380
(.0055) (.0057) (.0073) (.0072)
Total Cognitive Ability - .1082 - -
(.0142)
Schooling Cognitive Ability - - .1762 .1826 (.0238) (.0236)
Non-Schooling Cog. Ability - - - .0670
(.0130)
Experience/10 .6742 .6683 .6038 .6111
(.0637) (.0633) (.0573) (.0574) (Experience/10)2 -.1059 -.1015 -.0943 -.0931
(.0122) (.0122) (.0112) (.0113)
Constant 9.2904 9.4798 9.9587 9.9825
(.1294) (.1331) (.1375) (.1372)
R2 .3794 .4054 .4199 .4312
Note: The coe¢ cients of regressions of log annual wages on years of schooling, cognitive ability, and a quadratic expression of potential experience are shown.
All speci…cations allow for dummy variables for place of birth, place of residence, father’s place of birth, mother’s place of birth, father’s education, and mother’s education. Cognitive ability variables are standardized with a mean of zero and a standard deviation of one. The standard errors in parentheses are robust.
The sample size is 1’146 males.
Table 6: The Cognitive and Noncognitive Components of the Return to Schololing, Male Sample.
Components: 0 0
86.14% 13.86% 50.32% 49.68%
Note: and measure the noncognitive and cognitive components of the return to schooling according to Bowles, Gintis, and Osborne (2001); 0 and 0 according to the method developed in this paper.
is obtained using the estimated return to schooling in models I and II; = 1 . 0 is obtained using the return to schooling in models I and III; 0 = 1 0.
Table 7: The E¤ects of Schooling and Cognitive Ability on Log Wages, Female Sample, OLS estimates.
Model: I II III IV
Years of schooling .0706 .0641 .0310 .0299
(.0056) (.0059) (.0067) (.0067)
Total Cognitive Ability - .0714 - -
(.0183)
Schooling Cognitive Ability - - .1808 .1869 (.0217) (.0219)
Non-Schooling Cog. Ability - - - .0385
(.0159)
Experience/10 .4740 .4767 .4228 .4280
(.0652) (.0641) (.0059) (.0594) (Experience/10)2 -.0778 -.0754 -.0672 -.0665
(.0129) (.0126) (.0120) (.0119)
Constant 9.4297 9.5319 10.0624 10.0797
(.1232) (.1236) (.1214) (.1210)
R2 .2524 .2650 .3016 .3054
Note: The coe¢ cients of regressions of log annual wages on years of schooling, cognitive ability, and a quadratic expression of potential experience are shown.
All speci…cations allow for dummy variables for place of birth, place of residence, father’s place of birth, mother’s place of birth, father’s education, and mother’s education. Cognitive ability variables are standardized with a mean of zero and a standard deviation of one. The standard errors in parentheses are robust.
The sample size is 984 females.
