Cognitive Ability and Returns to Schooling in Chile
Chris Sakellariou
Division of Economics
Humanities and Social Sciences
Nanyang Technological University
Singapore
e-mail: [email protected]
Abstract: We use the International Adult Literacy Survey (IALS) data, which contain
direct measures of cognitive ability and employ them as additional controls in the
earnings function. On average, inclusion of the direct measure of cognitive ability
reduces the return to schooling by about 25 percent, while one standard deviation
increase in the score increases earnings by 15-20 percent. However, for those in the
lowest earnings/ability quantile, education qualifications at any education level do not
contribute to earnings; rather cognitive ability (quality) is the key to higher earnings. On
the other hand, those in high quantiles (therefore, those of higher cognitive and noncognitive ability) benefit much more from acquiring more schooling, and from the
interaction of additional schooling with the measure of cognitive ability.
1
Introduction
When estimating returns to schooling using a Mincer type earnings function, the
disturbance term will capture individual unobservable attributes and effects which, in
general, tend to influence the schooling decision, hence resulting in a correlation a
between schooling and the error term. Cognitive (as well as non-cognitive) ability is such
an unobservable; if schooling is endogenous then estimation by ordinary least Squares
(OLS) will yield biased estimates of the return to schooling.
When a true measure of ability is an omitted variable in the earnings equation, one of
three different approaches have been used in the empirical literature to capture “true”
return to education. The first approach uses twins, to arrive at a measure of the causal
return to education. For example Ashenfelter & Rouse (1998) and Rouse (1997) using
data from the US, have compared the earnings of twins with different educational levels,
and reported an estimate of the return to education that is about 30 percent smaller than
the OLS estimate.
The second approach uses sources for exogenous variation in educational attainment,
such as institutional changes in the schooling system in the form of changes in
compulsory schooling laws, abolition if fees etc, as well as other “natural variations” (i.e.,
school construction projects) affecting the schooling decision, to estimate a causal return
to education effect using instrumental variable estimation. Most of these estimates of the
return to education based on “natural experiments” report a higher return to education
(rather than a lower one), compared to OLS-based estimates of the return to education
(for example, see Angrist & Krueger, 1991, Kane & Rouse, 1993, Card, 1995, Harmon &
2
Walker, 1995, Meghir & Palme 1999; for developing countries see studies by DuFlo,
1998, Patrinos and Sakellariou, 2005, Sakellariou, 2006). The dominant explanation for
these seemingly counter-intuitive results is that institutional changes in the school system
(such as compulsory schooling laws) affect the schooling decision of a subset of
individuals who, otherwise, would not have pursued a higher level of education and not
the average individual. Furthermore, individuals affected by such reforms tend to have a
higher return to education than the average individual.
The third approach, and also the approach used in this study, uses achievement test scores
measuring cognitive ability, and employ them as additional controls in the earnings
function1. We use the International Adult Literacy Survey (IALS) data, which contain
direct measures of cognitive ability. In particular, the IALS contain labor force
information along with scores from a literacy test. The data set includes three scales to
measure individuals’ literacy levels. These scales relate to prose, document and
quantitative illiteracies. Such data allows the researcher to identify features of earnings
determination that are typically only indirectly observed.
Evidence on the relationship between heterogeneity in ability and returns to education
using a quantile regressions methodology for Chile exists in Montenegro (2001), who
found strong increasing returns by quantile over many years, as well as in the multicountry study by Patrinos et. al. (2006), who reported similar results. Both studies,
without controlling for ability, looked at the pattern of returns by quantile. The concept of
ability utilized in the above two, as well as other studies (for example Arias et. al., 2001),
1
One should, however, keep in mind that both schooling and the test score are generated by the same latent
ability. Therefore, one has to be aware of the joint causality between schooling and test scores (see Hansen,
Heckman & Mullen, 2003, Nordin, 2005).
3
relates to those unobservable, earnings-enhancing, human capital characteristics of an
individual, rather that measures derived from tests. An increasing pattern of returns was
interpreted as evidence in favor of a complementary relationship between ability and
education.
On the other hand, the IALS data have been used in several studies, with little evidence
on Chile, the only country in the survey outside Europe and North America. Blau and
Khan (2001) examined the role of cognitive skills in explaining higher wage inequality in
the U.S. Leuven et. al. (2004) used IALS data for 15 countries (including Chile) and
explored the hypothesis that wage differentials between skill groups across countries are
consistent with a demand and supply framework. Table 7 summarizes the OLS wage
regression estimates, controlling for cognitive achievement. They find that cognitive
achievement is an important determinant of earnings in all countries examined except
Poland and Finland. They also find that about one third of the variation in relative wages
between skill groups across countries is explained by differences in net supply of skill
groups. Green and Riddell (2002) used the measure of literacy in the IALS dataset to
examine the influence of cognitive and unobserved skills on earnings in Canada. They
find that cognitive skills contribute significantly to earnings and that their inclusion in
earnings equations reduces the measured impact of schooling. They also find that the
impact of literacy on earnings does not vary across quantiles of the earnings distribution;
schooling and literacy do not interact in influencing earnings. Their findings suggest that
cognitive and unobserved skills are both productive but that having more of one skill
does not enhance the other's productivity. Devroye and Freeman (2001) used the IALS
survey and found that skill inequality among advanced countries explains only about 7
percent of the cross-country differences in earnings inequality. They also find that the
4
bulk of cross country differences in earnings inequality occur within skill groups, not
between them.
Methodology
In the basic Mincerian human-capital model (Mincer, 1974), schooling is assumed to be
independent of ability, and that the return from schooling investments is equal for all
individuals. However, in the contemporary literature it is acknowledged that the return to
schooling must be different for different ability levels. Intuitively, an estimate of the
average return to schooling probably over-estimates the return for low ability individuals
and under-estimates the return to high ability individuals. One should, therefore allow
ability to affect the rate of return to schooling investments.
We incorporate the literacy score as a measure of cognitive ability to proxy for
unobserved effects. We expect that the inclusion of a direct measure of cognitive ability
will reduce the estimated education coefficient, so that the coefficient on education then
captures the effect of education alone having controlled for ability.
The effect of introducing ability differences is two-pronged. First, the more able
individuals may be able to ‘convert’ schooling into human capital more efficiently than
the less able, and this raises the return to schooling for the more able. In this case, one
can conclude that ability and education are complementing each other in producing
human capital. On the other hand, the more able may have higher opportunity costs since
they may have been able to earn more in the labor market, if ability to progress in school
is positively correlated with the ability to earn, and this reduces the rate of return to
schooling (Harmon and Walker, 2000). Given a distribution of wages, we assume that the
distribution of wages reflects the distribution of inherent ability. As a result, lower ability
5
individuals predominate in the lower quantiles of the distribution and higher ability
individuals predominate in the upper quantiles of the distribution.
The model in which both the measure of ability and its interaction with schooling affect
earnings and the ability specific return to schooling is outlined below (Griliches, 1977;
Nordin, 2005):
Ln wi = α + βSi + γAi + εi
(1),
where w is the hourly wage rate, S is years of schooling completed, A is ability and ε is an
independently distributed error term. Allowing the return to schooling to depend on ability:
Ln wi = α + β(f(Ai))Si + γAi + εi
(2).
Using the test score as a proxy for cognitive ability:
Ln wi = α + β(f(Ti))Si + γTi + εi
(3),
where T is the IALS test score which is assumed to perfectly measure cognitive ability. Assuming
a linear relationship between test score and cognitive ability:
f(Ti) = t0 + t1Ti
(4),
we obtain the following earnings equation:
Ln wi = α + t0Si + t1TiSi + γTi + εi
(5).
Finally, including experience, its square and other covariates:
6
Ln wi = α + t0Si + t1TiSi + γTi + exp + exp2 + Xi + εi
(6),
where X is a vector of other covariates.
Equation (6) is estimated using both Ordinary Least Squares (OLS) to obtain estimates of
average return to schooling, as well as Quantile regressions. When it is suspected that
various exogenous variables (such as schooling, cognitive ability and experience) influence
parameters of the conditional distribution of the dependent variable other than the mean,
quantile regressions are particularly useful, because they allow the full characterization of
the conditional distribution of the dependent variable, rather than the conditional mean
only.
The Quantile Regressions methodology, therefore, allows an investigator to
differentiate the contribution of regressors along the distribution of the dependent variable.
In particular, the estimation of returns to education entails much more than the fact that, on
average, one more year of education results in a certain percent increase in earnings.
The quantile regression model (Koenker and Bassett, 1978; Buchinsky, 1997) is outlined
below:
ln wi = Xiβθ + uθi,
Xiβθ = (Quantile)θ(lnwi|Xi);
where Xi is a vector of exogenous variables; βθ is the vector of parameters;
(Quantile)θ(lnwi|Xi) is the θth conditional quantile of lnw given X, with 0<θ<1. The θth
quantile is derived by solving the problem (using linear programming):
Min Σρθ(lnwi - Xiβθ),
β∈Rk i
7
where ρθ(ε) is the check function defined as ρθ(ε) = θε if ε≥0, and ρθ(ε) = (θ-1)ε if ε<0.
Standard errors are bootstrap standard errors. The median regression is obtained by setting
θ = 0.5 and similarly for other quantiles. As θ is varied from 0 to 1, the entire distribution
of the dependent variable, conditional on X, is traced.
OLS and Quantile regressions will also be used to estimate the corresponding equations
in which different levels of education replace the years of schooling variable, to gain an
insight as to how the pattern of returns varies with education level.
Data
The International Adult Literacy Survey (IALS) was carried out in 20 countries2 between
1994 and 1998, a project undertaken by the governments of the countries and three
intergovernmental organizations3. It is a carefully designed, innovative survey of adult
populations, and goes beyond just measuring literacy capabilities to assessing how these
capabilities are applied to everyday activities. The IALS was followed by an extensive
quality review (see Murray et al., 1998) which, after comparing the distribution of the
characteristics of the actual and weighted samples, concluded that the actual and
weighted samples were comparable to the overall populations of the IALS countries. The
questionnaire also included questions about labor market status, earnings, education as
well as demographic characteristics. The sample size varies between 1,500 and 6,000 per
country.
2
The countries are: Belgium, Canada, Chile, Czech Republic, Denmark, Finland, Germany, Hungary,
Ireland, Italy, The Netherlands, New Zealand, Norway, Poland, Portugal, Slovenia, Sweden, Switzerland,
the UK and the US.
3
OECD, European Union and UNESCO.
8
The data includes three scales as measures of literacy skills: Prose literacy, Document
literacy and Quantitative literacy, each in the 0-500 range. Prose literacy tests the
understanding and use of texts such as editorials news stories, fiction, and poems.
Document literacy tests skills required to locate and use information contained in a
variety of formats, such as job applications, payroll forms, maps and tables. Quantitative
literacy tests skills required in making calculations after locating numbers embedded in
printed materials; examples of such calculations include determining the interest on a
loan, calculating a tip and balancing a checkbook.
For Chile, the survey covers 98 percent of the population between the ages of 16 and 65
(it excludes residents of institutions and remote areas) and the total number of
respondents was 3,583. A four-stage stratified sample was used and stratification was
performed according to region and type (urban/rural). As is the case for the rest of the
IALS country data, in the Chilean data the three skills are very highly correlated.
Consequently, the average of the three scores will be used in the analysis as an aggregate
IALS score measure (see also Blau and Khan, 2001; Devroye and Freeman, 2001;
Leuven et.al., 2004).
As is the case with the other countries in the IALS dataset, for Chile as well, the relation
between skill level and years of schooling is positive. However, the slope of the skillschooling profile is less steep compared to all but three other countries - Germany, The
Netherlands and Sweden – while the slope is steepest in the Czech Republic, the US,
Slovenia and Canada (see Leuven et. al., 2004).
9
Chile ranks at the bottom in terms of average years of schooling as well as average IALS
scores. The USA ranks highest in educational attainment, while Sweden, along with other
Scandinavian counties rank highest in average IALS test scores, followed by The
Netherlands and Germany.
Results
The working sample includes males employed for wages between the ages of 18 and 65.
The dependent variable in the earnings regressions is the logarithm of the hourly wage,
calculated using information on yearly earnings from wages, hours worked per week and
weeks worked per year.
Table 1 reports the mean values of the logarithm of hourly wage, years of schooling,
cognitive achievement score and years of experience, as well as the mean values of years
of schooling, cognitive achievement score and years of experience by earnings quantile.
As expected, years of schooling and cognitive achievement scores increase by earnings
quantile, while higher paid employees have less average years of experience (a younger
lot).
In the estimation of earnings functions, cognitive achievement scores have been
standardized; therefore, the estimated coefficient of cognitive achievement measures the
approximate percentage change in the hourly wage arising from a one standard deviation
increase in the score. Likewise, the coefficient of cognitive achievement-years of
schooling interaction variable measures the approximate percentage change in the hourly
10
wage arising from a one standard deviation increase in cognitive achievement, interacted
with one additional year of schooling.
Regressing earnings on the standardized cognitive achievement score (Table 2), after
omitting schooling (and other covariates) from the wage equation, solves the problem of
endogeneity of schooling, resulting in an estimate of the net effect of ability on earnings;
that is, the combination of the direct effect as well as the effect through schooling. It is
found that cognitive ability explains about 15 percent of the variance of log wages and
that one standard deviation increase in the score increases earnings by about one-third.
Table 3a presents the OLS regression results for equation (6). Inclusion of the direct
measure of cognitive ability (column 3) reduces the return to schooling by about 25
percent, while one standard deviation increase in cognitive achievement increases
earnings by a significant 15 percent. This is approximately equivalent to the effect on
earnings of two additional years of schooling. Once both the cognitive achievement and
its interaction with years of schooling are included (column 4), the interaction term is
positive and significant on average, while the effect of cognitive achivement becomes
insignificant. This is also the case when other covariates (marital status, urban residence
and father’s and mother’s education) are included in the specification (column 5).
Columns 6 and 7 report results when the years of schooling variable is replaced by years
of schooling dummies. In column 7, the specification includes cognitive achievement and
its interaction with years of schooling, while in column 6 it does not. Without controlling
for cognitive achievement, one additional year of schooling, including years of basic
schooling (8 years), results in significant earnings increases. However, after controlling
11
for cognitive achievement, the effect of one additional year of basic schooling has an
insignificant effect on earnings; the effect of additional secondary and tertiary schooling,
on the other hand has a significant independent effect on earnings. Comparing columns 4
and 7, with the inclusion of years of schooling dummies, the independent effect of
cognitive achievement is now significant on average, while the interaction term turns
insignificant.
Table 3b reports results similar to those in table 3a, after replacing years of schooling
with education levels. In Chile, General Basic education is compulsory and lasts for eight
years (from age 6 to age 14), divided into two cycles of four years each. Secondary
education (Educación media) lasts for four years (from 14 to 18). Higher education is
provided by universities, professional institutes and technical training centers.
The results are generally in line with those in Table 3a. The return to Basic education is
insignificant, except for column 1 which does not control for cognitive achievement or
other covariates. Returns per year of education level are, however, significant for upper
secondary and higher education levels. The highest returns are observed for higher nonuniversity qualifications, followed by returns to university qualifications. The
independent effect of cognitive achievement (column 2) is positive and significant (at
about 20 percent increase in earnings, in response to one standard deviation increase in
the IALS score).
When the interaction effects of cognitive achievement with each education level is added
in the specification, the independent effect of cognitive achievement decreases
12
significantly, however, the interaction terms are small (column 3). The effects of
cognitive achivement and its interaction with each education level further decrease once
the additional covariates are added in (column 4).
Tables 2, 4a, 4b, 5a, 5b, 6a and 6b present the quantile regression results. These results
are vastly more illuminating, compared to the OLS results. Table 4a presents the
estimated Mincerian equations by earnings quantile, which are useful for purposes of
comparison with other quantile regression estimates for Chile, as well as with the
cognitive achievement augmented quantile regression estimates in Tables 5a, 5b, 6a, 6b.
Without controlling for cognitive achievement, quantile returns to one additional year of
schooling exhibit a U-shaped pattern, increasing after the 10th quantile. The 90th-10th
inter-quantile difference is 0.02 while the 90th-25th inter-quantile difference is 0.045.
These estimates are qualitatively similar to those obtained by Patrinos et. al. (2006) and
Montenegro (2001). The only difference is that in these two studies returns increase
monotonically.
Table 2 gives the quantile regression results for the specification in column 1 of Table 3a.
The combined effect (direct and the effect through schooling) shows little variation
across quantiles. One standard deviation increase in cognitive achievement increases
earnings by between one-quarter and one-third. In lower quantiles (10th and 25th),
cognitive achievement alone explains about the same amount of the variance in earnings
as the specification in Table 4a (standard Mincerian specification).
Table A5a presents results after controlling for the independent effect of cognitive
achievement. Schooling returns by quantile now exhibit a sharp and strictly increasing
13
pattern, with a 90th-10th inter-quantile difference of 0.075. The independent effect of
cognitive achievement is positive and highly significant up to the 75th quantile of
earnings, suggesting that one standard deviation increase in achievement increases
earnings by about 20 percent. However, this effect all but disappears at the 90th quantile.
Comparing the effect of an increase in achievement scores to the effect of one additional
year of schooling, the effect of one standard deviation increase in the score is equivalent
to the effect of 5, 3.5, 3, and 2 additional years of schooling for quantiles 10, 25, 50 and
75, respectively.
Comparing schooling return estimates in Tables 4a and 5a, when
cognitive achievement is not controlled for the estimates of returns to schooling are
upward biased, except for the 90th quantile of earnings, where the estimate remains
approximately unchanged after controlling for cognitive achievement.
Table 6a presents the results from the preferred specification, which controls for both the
independent effect of the cognitive ability measure and its interaction with years of
schooling. In this case, at the lowest quantile, the independent effect of cognitive
achievement is very strong, indicating that one standard deviation increase in the score
increases earnings by about one-third, while the interaction term is negative and
significant. For quantiles 25-90 the picture is drastically different. The independent effect
of cognitive achievement
is insignificant in quantiles 25 to 75 and negative and
significant in the 90th quantile. Instead, what matters is the earnings enhancing,
complimentary relationship between cognitive achievement and schooling. One standard
deviation increase in the score combined with one additional year of schooling increases
earnings by between 1.5 percent (25th quantile) and 4.5 percent (90th quantile), while the
14
return to schooling estimate at the highest (90th) quantile is now about 15 percent lower
compared to the one reported in Table 5a (the specification without the interaction term).
Tables 5b and 6b contain the results of the cognitive achievement augmented quantile
regressions, using education level dummies. Returns exhibit an increasing pattern, with
the exception of Upper Secondary education for which returns are neither increasing nor
decreasing. Returns to University education sharply increase with quantile (with or
without controlling for cognitive achievement), from zero in the lowest to about 30
percent in the highest quantile, and even more so in the case of Higher (non-university)
education (from 5 to about 50 percent). Returns to various education levels are generally
significant only in higher quantiles of the earnings distribution. In particular, the
annualized return to Basic as well as Upper Secondary education is significant at
quantiles 75 and 90, the return to Higher (non-university) education is significant in
quantiles 50-90 and the return to University education is significant in all but the lowest
quantile. On the other hand, the effect of cognitive achievement and its interaction with
various levels of education are similar to those found in Table 6a, with positive
interaction terms found mostly in the higher quantiles (but insignificant, except for
quantile 75) and negative but insignificant in the lowest quantile.
Summarizing, for individuals in the lowest earnings quantile, education qualifications at
any level (quantity of schooling) do not contribute to earnings; rather cognitive
achievement (quality) is the key to higher earnings. On the other hand, those in high
quantiles (therefore, those of higher ability) benefit much more from acquiring more
schooling, and from the interaction of additional schooling with ability.
15
Cognitive skills are, therefore, important in determining earnings. However, cognitive
skills are not the only skills that are relevant, and this has been frequently stressed in the
literature (see for example, Heckman et. al., 2006; Green, 2001; Bowles and Gintis, 2000;
Bowles et. al., 2001). One can, for example, observe that the amount of additional
variation in earnings that is typically explained (including this study) by the inclusion of
the cognitive achievement measure is rather modest. Furthermore, the inclusion of other
explanatory variables, along with education and the measure of cognitive achievement,
such as experience, parental background and other individual characteristics, typically
explains less than half of the variation in earnings. Non-cognitive ability may be
productive on its own right, and the effect may be large. Non-cognitive traits include
attitude, communication skills, motivation, persistence, as well as dependability and
docility (especially in low skill labor markets) etc. Bowles and Gintis (2000) argue that
non-cognitive skills may account for more than half of measured returns to schooling,
and that most of the schooling impact is through endowing individuals with such noncognitive traits. Heckman et. al., (2006) agree, and further suggest that latent noncognitive skills, corrected for schooling and family background raise wages through their
direct effects on productivity and through their indirect effects on schooling and work
experience.
One would expect that higher non-cognitive skills (as well as cognitive skills) result in
more school acquisition and that they complement schooling, resulting in higher returns
to schooling as measured from earnings functions. When ability measures (cognitive or
non-cognitive) are absent in earnings functions, the effect of ability is captured in the
16
residuals. Then, an increasing pattern of returns by earnings/ability quantile is an
indication of a complementary relationship between ability and schooling. This is the
case for Chile as well as most other countries (for example in Europe, USA and Latin
America, but not Africa and East Asia – see Patrinos et. at., 2006). Once cognitive ability
is controlled for using a proper measure (for example, IALS literacy scores in this and
other studies), it is of interest to re-look at the pattern of returns across the earnings
distribution. From tables 5a and 6a, measured returns to schooling exhibit an increasing
pattern which is even stronger compared to when the measure of cognitive achievement
are absent (Table 4a). When cognitive achievement is controlled for (and conditioning for
schooling and experience) the 90th-10th inter-quantile difference in schooling returns is
0.075, while when cognitive achievement as well as its interaction with years of
schooling is included the 90th-10th inter-quantile difference is 0.04, compared to 0.02 in
Table 4a. This suggests that, along with cognitive ability, non-cognitive ability interacts
with schooling in a positive way, increasing the benefits from schooling acquisition.
Future studies could explore whether the findings for Chile, the only Latin American
country in the IALS surveys (for which schooling attainment as well as test score are the
lowest among the IALS group of countries) differ from other countries. Furthermore, one
could review Chile’s education reforms during the 1980s and 1990s and identify a
suitable instrument based on a significant education reform (one such reform could be the
introduction of vouchers), in order to analyze treatment effects, with and without
conditioning for cognitive ability.
Conclusions
17
The return to investing in education, based on past empirical studies, is known to differ
between individuals in different parts of the earnings/ability distribution. For example,
evidence from the US (Ingram and Neumann, 2005) shows that over the past decades,
individuals with college education but without specific skills experienced the lowest
benefits from investing in education. Individuals with low ability, therefore may not
benefit as much from investing in education, compared to individuals in the upper part of
the ability distribution; for the latter, ability is expected to interact positively with
education resulting in higher benefits from education investments.
This paper estimates the return to investments in education and the return to cognitive
skills across the ability distribution. We use data from the International Adult Literacy
Survey (IALS), a carefully designed, innovative survey of adult populations, which goes
beyond just measuring literacy capabilities, to assessing how these capabilities are
applied to everyday activities. Such data allows the researcher to identify features of
earnings determination that are typically only indirectly observed.
It is found that, inclusion of cognitive achievement, after conditioning for years of
schooling and experience, reduces the return to schooling by about 25 percent on average
and that one standard deviation increase in the score increases earnings by 15-20 percent,
compared to about 35 percent when schooling and other covariates are eliminated from
the equation. Therefore, there is an additional, indirect effect of cognitive achievement on
earnings through schooling.
18
The quantile regression results are, however, much more informative. Before
conditioning for years of schooling and experience, the combined direct and indirect
effect of the cognitive achievement on earnings is consistently high (25 to 35 percent
increase in earnings in response to one standard deviation increase in the score).
However, once we condition for schooling and experience, the independent effect of
cognitive achievement is positive and highly significant only up to the 75th quantile of
earnings, but this effect all but disappears at the 90th quantile. When cognitive
achievement is not controlled for, therefore, the OLS estimates of returns to schooling are
upward biased, except for the 90th quantile of earnings, where the estimate remain
approximately unchanged. When controlling for the independent effect of the cognitive
achievement as well as its interaction with years of schooling, the independent effect of
cognitive achievement is positive and particularly strong at the lowest earnings quantile.
For quantiles 25-90 the picture is drastically different. The independent effect of the
cognitive achievement at higher quantiles is insignificant. Instead, what matters is the
earnings enhancing, complimentary relationship between the measure of cognitive ability
and schooling.
We can therefore conclude that for those at the lowest earnings quantile, education
qualifications alone do not contribute to earnings; rather cognitive ability (quality of skill)
is the main contributor to higher earnings. On the other hand, those in high quantiles
(therefore, those of higher ability) benefit much more from acquiring more schooling, and
from the interaction of additional schooling with ability.
19
Finally there is, mostly indirect, but convincing evidence that non-cognitive ability may
be productive on its own right and that the effect may be large. Non-cognitive traits may
account for most of the schooling impact, through their indirect effects on schooling and
work experience.
In this study, having controlled for cognitive ability using IALS scores, the pattern of
measured returns to schooling provides an indication for the relationship between noncognitive ability (captured in the residuals) and schooling. We find a strong increasing
pattern. This suggests that, along with cognitive ability, non-cognitive traits complement
schooling, thus enhancing the benefits of education investments.
20
References
Angrist, J. & A. Krueger (1991). “Does Compulsory School Attendance Affect Schooling
and Earnings?” The Quarterly Journal of Economics, Volume 106 (4): 979-1014.
Arias, O. K., F. Hallock and W. Sosa (2001), “Individual Heterogeneity in the Returns to
Schooling: Instrumental Variables Quantile Regression Using Twins Data.” Empirical
Economics 26:7-40.
Ashenfelter, O. and C. E. Rouse (1998). “Income, Schooling and Ability, Evidence from
a New Sample of Identical Twins”. The Quarterly Journal of Economics, Volume 113
(1): 253-284.
Bowles, S. and H. Gintis (2000). “Does Schooling Raise Earnings by Making People
Smarter?” in K. Arrow, S. Bowles and S. Durlauf (eds) Meritocracy and Economic
Inequality. Princeton University Press. Princeton.
Bowles, S., Gintis, H. and M. Osborne (2001). “Incentive Enhancing Preferences:
Personality Behavior and Earnings.” American Economic Review, Volume 91 (2): 155158.
Blau, F. and L. Khan (2001). “International Differences in Male Wage Inequality:
Institutions Versus Market Forces.” Journal of Political Economy, Volume 104: 791–
837.
Buchinsky, M. (1997). “Recent Advances in Quantile Regression Models: A Practical
Guideline for Empirical Research.” The Journal of Human Resources, Volume 33(1): 88126.
Card, D. (1995). “Earnings, Schooling, and Ability Revisited.” in S. Polacheck (ed.),
Research in Labor Economics, Volume: 23-48, JAI Press: Greenwich, Connecticut.
Devroye, D. and R. Freeman (2001). ” Does inequality in skills explain inequality of
earnings across advanced countries?” NBER Working Paper 8140.
DuFlo, E. (1998). “Schooling and labor market consequences of school construction in
Indonesia” Evidence from an unusual policy experiment.” American Economic Review
Volume 91 (4): 795-813.
Green, D. (2001). “Literacy Skills, Non-Cognitive Skills and Earnings: An Economist’s
Perspective.” Department of Economics, University of British Columbia, Canada.
Green, D. and C. Riddell (2002). “Literacy and Earnings: An Investigation of the
Interaction of Cognitive and Non-Cognitive Attributes in Earnings Generation.”
Discussion Paper: 02-11, Department of Economics, University of British Columbia,
Canada.
21
Griliches, Z. (1977). “Estimating the Returns to Schooling: Some Econometric
Problems.” Econometrica, Volume 45 (1): 1-22.
Hansen K., Heckman, J. and K. Mullen (2003). “The effect of Schooling and Ability on
Achievement Test Scores.” Journal of Econometrics, Volume 121 )1-2): 39-98
Harmon, C. and I. Walker (1995). “Estimates of the Economic Returns to Schooling for
the United Kingdom.” American Economic Review, Volume 85 (5): 1278-1286.
_____________ (2000). “Returns to the Quantity and Quality of Education: Evidence
for Men in England and Wales.” Economica, Volume 67: 19-35.
Hecknam, J., Stixrud, J. and S. Urzua (2006). “The Effects of Cognitive and NonCognitive Abilities on Labor Market Outcomes and Social Behavior.” NBER Working
Paper 12006, Cambridge, MA.
Ingram, B. and G. Neumann (2005). “The Returns to Skill.” Forthcoming in Labour
Economics.
Kane, T and T. Rouse (1993). “Labor Market Returns to two- and Four-Year Colleges: is
a Credit a Credit and do Degrees Matter.” NBER Working paper No: 4268.
Koenker,R. and G. Bassett (1978). “Regression Quantiles.” Econometrica Volume 50:4361.
Leuven, E., H. Oosterbeek and H. van Ophem (2004). “Explaining International
Differences in Male Skill Wage Differentials by Differences in Demand and Supply of
Skills.” The Economic Journal, Volume 114 (April): 466-486.
Meghir, C. and M. Palme (1999). “Assessing the Effect of Schooling on Earnings Using a
Social Experiment.” Working Paper Series No: W99/10, The Institute for Fiscal Studies.
Mincer, J. (1974). “Schooling, Experience and Earnings, Columbia University Press:
New York.
Montenegro, C. (2001). "Wage Distribution in Chile: Does Gender Matter? A Quantile
Regression Approach." World Bank, Washington, D.C.
Murray, T, I. Kirsch and L. Jenkins (1998). “Adult Literacy in OECD Countries:
Technical Report on the First International Adult Literacy Survey.” No. NCES 98-053,
Washington D.C.: U.S. Department of Education, National Center of Education Statistics.
Nordin (2005). “Ability and Rates of Return to Schooling – Making Use of the Swedish
Enlistment Battery Test.” Department of Economics, Lund University, Sweden.
Patrinos, H. and C. Sakellariou (2005). “Schooling and Labor Market Impacts of a
Natural Policy Experiment.” LABOUR, Volume 19 (4): 705-719.
22
Patrinos H, C. Ridao Cano and C. Sakellariou (2006). “Heterogeneity in Ability and
Returns to Education: Multi-country Evidence from Latin America and East Asia.”
World Bank, Washington, D.C.
Rouse, C. (1997). “Further Estimates of the Economic Return to Schooling from a New
Sample of Twins.” Unpublished discussion paper, Princeton University.
Sakellariou, C. (2006). “Education Policy Reform, Local Average Treatment Effect and
Returns to Schooling from Instrumental Variables in the Philippines.” Applied
Economics, Volume 38: 473-481.
23
Appendix
Table 1: Means of Variables by Earnings Quantile, Male Employees
Variable
Overall 10th
25th
50th
75th
90th
99th
mean
Log (hourly wage)
6.41
5.56
5.95
6.17
6.87
7.46
8.84
(0.88) (0.63) (0.12)
(0.13)
(0.16) (0.16) (0.36)
Years of schooling
9.3
6.9
7.3
8.4
9.6
11.5
13.4
(4.3)
(3.9)
(3.9)
(3.7)
(3.8)
(3.6)
(4.4)
IALS score
209.3 167.3 181.5
198.8
221.1 237.3 259.3
(62.8) (65.1) (61.4)
(54.5)
(55.7) (51.7) (54.7)
Years of experience
21.5
25.7
22.9
21.9
20.4
20.2
17.9
(13.4) (14.5) (14.8)
(13.5)
(11.9) (12.7) (12.5)
N
892*
88
159
217
211
125
82
Note: standard deviation in parentheses.
*Includes ten observations for those with earnings which exceed the 99th percentile.
Table 2: Returns to Schooling Using Quantile Regressions (Male employees)
Dependent Variable:
Q10
Q25
Q50
Q75
Q90
log of hourly wage
Stad. ScoreIALS
0.255
0.250
0.330
0.354
0.328
(5.2)
(8.0)
(11.9)
(7.6)
(7.1)
Constant
5.69
6.05
6.45
6.86
7.33
(124.2)
(182.7)
(221.1)
(150.3)
(171.4)
PseudoR2
0.061
0.068
0.101
0.098
0.127
Note: t-values in parentheses.
24
Table 3a: Returns to Schooling (Male employees)
Dependent Variable: log of hourly
wage
Years of schooling
(1)
(2)
(3)
(4)
(5)
(6)
(7)
-
0.113
(14.9)
0.087
(8.9)
0.082
(8.3)
0.089
(7.3)
-
-
-
-
-
-
-
2
-
-
-
-
-
3
-
-
-
-
-
4
-
-
-
-
-
5
-
-
-
-
-
6
-
-
-
-
-
7
-
-
-
-
-
8
-
-
-
-
-
9
-
-
-
-
-
10
-
-
-
-
-
11
-
-
-
-
-
12
-
-
-
-
-
13
-
-
-
-
-
14
-
-
-
-
-
15
-
-
-
-
-
16
-
-
-
-
-
17
-
-
-
-
-
>17
-
-
-
-
-
0.728
(2.0)
0.396
(1.6)
0.483
(2.1)
0.738
(3.1)
0.661
(2.9)
0.416
(1.9)
0.844
(3.5)
0.836
(3.8)
0.940
(4.0)
1.158
(5.1)
0.709
(3.0)
1.268
(5.9)
1.316
(5.6)
1.606
(6.3)
1.840
(7.2)
1.939
(7.6)
2.094
(8.5)
2.403
(9.2)
0.660
(1.8)
0.344
(1.3)
0.328
(1.3)
0.568
(2.2)
0.484
(1.9)
0.204
(0.8)
0.610
(2.2)
0.548
(2.1)
0.600
(2.2)
0.821
(3.1)
0.374
(1.4)
0.886
(3.5)
0.877
(3.2)
1.155
(4.1)
1.370
(4.9)
1.466
(5.2)
1.604
(5.9)
1.939
(6.9)
Experience
-
Exp. Squared
-
0.013
(2.0)
-0.0000
(0.1)
-
0.015
(2.2)
-0.0000
(0.2)
0.153
(4.2)
-
0.020
(2.9)
-0.0002
(1.2)
0.004
(0.1)
0.018
(3.0)
0.013
(1.4)
-0.0001
(0.4)
0.019
(0.2)
0.014
(1.9)
0.020
(2.9)
-0.0002
(1.2)
-
0.021
(3.1)
-0.0002
(1.5)
0.182
(2.0)
0.003
(0.3)
Years of schooling dummies
1
Standardized ScoreIALS
Standardized ScoreIALS*years of
schooling
0.342
(12.8)
-
-
-
25
Other controls
Married
-
-
-
-
Urban
-
-
-
-
Father upper secondary or higher
-
-
-
-
Mother upper secondary or higher
-
-
-
-
6.45
(243.9)
0.158
861
5.11
(41.0)
0.215
861
5.33
(39.7)
0.230
861
5.30
(39.6)
0.237
861
Constant
R2-adj.
N
0.072
(1.0)
0.075
(0.9)
0.201
(2.0)
0.086
(0.8)
-
-
-
-
-
-
-
-
5.15
(32.3)
0.276
664
5.10
(23.2)
0.247
861
5.40
(20.5)
0.261
861
Note: t-values in parentheses.
26
Table 3b: Returns by Education Qualification (Annualized Coefficient estimates) - Male
employees
Dependent Variable: log of hourly wage
Basic (per year)
Standardized ScoreIALS
(1)
0.070
(2.3)
0.107
(6.4)
0.219
(8.3)
0.164
(10.8)
0.021
(3.0)
-0.0003
(2.1)
-
Stand. ScoreIALS *Basic
-
(2)
0.014
(0.4)
0.075
(3.1)
0.175
(4.9)
0.125
(6.0)
0.023
(3.3)
-0.0003
(2.2)
0.207
(5.9)
-
Stand. ScoreIALS *Upper Secondary
-
-
Stand. ScoreIALS *Higher Non-University
-
-
Stand. ScoreIALS *University
-
-
Other controls
Married
-
-
-
Urban
-
-
-
Father upper secondary or higher
-
-
-
Mother upper secondary or higher
-
-
-
5.77
(49.2)
0.190
861
5.94
(50.0)
0.220
861
5.88
(42.6)
0.221
861
Upper Secondary (per year)
(vs. basic)
Higher Non-University (per year)
(vs. upper secondary)
University (per year)
(vs. upper secondary)
Experience
Exp. Squared
Constant
R2-adj.
N
(3)
0.039
(1.0)
0.067
(3.0)
0.288
(4.7)
0.166
(4.7)
0.022
(3.2)
-0.0003
(2.3)
0.121
(1.5)
0.140
(1.5)
0.131
(1.2)
-0.125
(0.7)
0.024
(0.2)
(4)
0.041
(0.8)
0.063
(2.2)
0.271
(3.8)
0.179
(4.2)
0.014
(1.5)
-0.0002
(1.2)
0.082
(0.7)
0.137
(1.0)
0.139
(1.0)
-0.102
(0.5)
-0.098
(0.5)
0.116
(1.5)
0.127
(1.5)
0.172
(1.7)
0.133
(1.2)
5.75
(30.2)
0.256
664
Note: t-values in parentheses.
27
Table 4a: Returns to Schooling Using Quantile Regressions (Male employees)
Dependent Variable:
Q10
Q25
Q50
Q75
log of hourly wage
Years of schooling
0.102
0.075
0.112
0.114
(6.0)
(9.0)
(14.0)
(8.7)
Experience
0.023
0.017
0.027
0.002
(1.7)
(2.1)
(3.6)
(0.2)
Exp. Squared
-0.0002
-0.0002
-0.0002
0.0003
(0.9)
(1.2)
(1.6)
(1.6)
Constant
4.39
5.12
5.01
5.65
(17.1)
(36.3)
(37.4)
(27.8)
PseudoR2
0.065
0.078
0.120
0.167
Note: t-values in parentheses.
Q90
0.121
(6.3)
0.004
(0.4)
0.0002
(1.2)
5.86
(20.4)
0.213
Table 4b: Returns to Education Qualifications Using Quantile Regressions (Annualized
Coefficient Estimates) - Male employees
Dependent Variable: log of
Q10
Q25
Q50
Q75
Q90
hourly wage
0.116
0.102
0.072
0.045
0.072
Basic (per year)
(4.0)
(3.2)
(2.4)
(1.4)
(1.1)
0.081
0.079
0.081
0.079
0.125
Upper Secondary (per year)
(6.6)
(5.6)
(5.3)
(4.1)
(3.2)
(vs. basic)
0.463
0.217
0.261
0.197
0.054
Higher Non-Univ. (per year)
(12.3)
(6.6)
(7.5)
(5.8)
(2.8)
(vs. upper secondary)
0.287
0.209
0.217
0.142
-0.013
University (per year)
(14.3)
(10.6)
(11.7)
(7.8)
(1.4)
(vs. upper secondary)
Experience
0.021
0.031
0.042
0.005
0.004
(1.3)
(3.7)
(5.4)
(0.5)
(0.4)
Exp. Squared
-0.0003
-0.0005
-0.0007
0.0001
0.0001
(0.0)
(3.1)
(4.5)
(0.7)
(0.6)
Constant
5.02
5.44
5.65
6.27
6.54
(19.3)
(41.4)
(46.1)
(43.3)
(44.2)
PseudoR2
0.044
0.067
0.118
0.167
0.254
Note: t-values in parentheses.
28
Table 5a: Returns to Schooling Using Quantile Regressions (Male employees)
Dependent Variable:
Q10
Q25
Q50
Q75
log of hourly wage
Years of schooling
0.043
0.057
0.069
0.087
(3.0)
(5.5)
(4.8)
(6.8)
Experience
0.026
0.026
0.022
0.001
(2.6)
(3.4)
(2.2)
(0.1)
Exp. Squared
-0.0003
-0.0003
-0.0002
0.0003
(1.7)
(1.8)
(1.1)
(1.8)
Stand. ScoreIALS
0.205
0.197
0.214
0.174
(4.0)
(4.9)
(4.0)
(4.0)
Constant
4.98
5.17
5.49
5.83
(26.9)
(35.9)
(27.3)
(33.9)
PseudoR2
0.082
0.099
0.139
0.178
Note: t-values in parentheses.
Q90
0.118
(4.6)
0.003
(0.3)
0.0003
(1.2)
0.020
(0.3)
5.88
(17.7)
0.213
Table 5b: Annualized Returns to Education Qualifications Using Quantile Regressions
(Male employees)
Dependent Variable: log of
Q10
Q25
Q50
Q75
Q90
hourly wage
0.056
0.033
0.014
-0.022
0.018
Basic (per year)
(2.1)
(1.4)
(0.4)
(0.7)
(0.3)
0.076
0.074
0.039
0.051
0.053
Upper Secondary (per year)
(5.3)
(4.4)
(1.6)
(1.2)
(1.5)
(vs. basic)
0.457
0.174
0.202
0.085
0.155
Higher Non-Univ. (per year)
(11.7)
(5.8)
(3.7)
(2.1)
(1.5)
(vs. upper secondary)
0.281
0.158
0.139
0.095
0.048
University (per year)
(13.6)
(9.0)
(5.0)
(3.7)
(1.1)
(vs. upper secondary)
Experience
0.027
0.033
0.034
0.006
0.006
(2.4)
(4.6)
(4.3)
(0.8)
(1.2)
Exp. Squared
-0.0004
-0.0005
-0.0005
0.0001
0.0001
(0.7)
(3.8)
(3.6)
(0.6)
(0.6)
Stand. ScoreIALS
0.245
0.247
0.243
0.211
0.083
(3.9)
(6.8)
(6.2)
(7.1)
(2.5)
Constant
5.23
5.66
5.92
6.45
6.71
(28.0)
(48.1)
(45.1)
(60.2)
(66.9)
PseudoR2
0.076
0.097
0.146
0.187
0.262
Note: t-values in parentheses.
29
Table 6a: Returns to Schooling Using Quantile Regressions (Male employees)
Dependent Variable: log of hourly wage
Years of schooling
Experience
Exp. Squared
Standardized ScoreIALS
Stand. ScoreIALS*years of schooling
Constant
PseudoR2
Q10
0.060
(3.7)
0.022
(2.0)
-0.0002
(1.2)
0.370
(4.1)
-0.022
(2.5)
4.89
(24.4)
0.086
Q25
0.055
(5.5)
0.034
(4.6)
-0.0005
(3.2)
0.065
(1.0)
0.015
(2.5)
5.11
(38.3)
0.102
Q50
0.065
(4.5)
0.030
(2.9)
-0.0004
(2.0)
-0.030
(0.3)
0.026
(3.0)
5.40
(27.3)
0.149
Q75
0.082
(5.7)
0.002
(0.2)
0.0002
(1.1)
-0.119
(1.5)
0.026
(3.5)
5.87
(30.1)
0.193
Q90
0.100
(5.1)
0.014
(1.3)
-0.0000
(0.1)
-0.309
(2.8)
0.045
(5.3)
5.87
(20.3)
0.247
Note: t-values in parentheses.
Table 6b: Returns by Education Qualification Using Quantile Regressions (Annualized
Coefficient Estimates) - Male employees
Dependent Variable: log of hourly wage
Basic (per year)
Upper Secondary (per year)
(vs. basic)
Higher Non-University (per year)
(vs. upper secondary)
University (per year)
(vs. upper secondary)
Experience
Exp. Squared
Standardized ScoreIALS
Stand. ScoreIALS *Basic
Stand. ScoreIALS *Upper Secondary
Stand. ScoreIALS *Higher Non-University
Stand. ScoreIALS *University
Constant
PseudoR2
Q10
-0.009
(0.1)
0.055
(0.9)
0.054
(0.3)
0.019
(1.0)
0.021
(1.9)
-0.0003
(1.4)
0.316
(2.4)
-0.017
(0.1)
-0.170
(0.9)
-0.016
(0.1)
Q25
0.003
(0.1)
0.060
(1.8)
0.095
(1.8)
0.147
(3.4)
0.034
(4.3)
-0.0005
(3.6)
0.172
(2.0)
0.17
(1.2)
0.035
(0.3)
0.001
(0.0)
Q50
0.013
(0.3)
0.041
(1.3)
0.338
(3.7)
0.179
(3.5)
0.033
(3.7)
-0.0006
(3.1)
0.228
(2.4)
0.007
(0.1)
0.060
(0.4)
-0.278
(1.4)
Q75
0.080
(3.4)
0.066
(6.1)
0.203
(5.8)
0.164
(8.8)
0.001
(0.2)
0.0002
(1.9)
0.014
(0.3)
0.249
(3.9)
0.241
(3.3)
0.113
(0.9)
Q90
0.098
(2.6)
0.066
(4.3)
0.502
(11.9)
0.289
(9.9)
0.005
(0.5)
0.0001
(0.5)
0.016
(0.2)
0.044
(0.5)
0.133
(1.3)
0.021
(0.1)
-0.314
(1.3)
5.41
(24.1)
0.081
-0.097
(0.6)
5.58
(37.0)
0.099
-0.061
(0.3)
5.93
(34.2)
0.149
0.189
(2.2)
6.38
(65.3)
0.192
0.110
(1.0)
6.57
(37.0)
0.264
Note: t-values in parentheses.
30
Table 7: Returns to Education and Cognitive Achievement from OLS Wage Regressions
IALS score/100
N
R2(a)
Education
Country
Education
With IALS
Without IALS
score
score
Canada
0.064 (0.008)** 0.052 (0.008)** 0.138 (0.044)**
855
0.30
Chile
0.103 (0.012)** 0.074 (0.014)** 0.285 (0.074)**
715
0.22
Czech Rep.
0.052 (0.008)** 0.046 (0.009)** 0.097 (0.046)*
519
0.15
Denmark
0.046 (0.004)** 0.042 (0.004)** 0.091 (0.031)**
986
0.25
Finland
0.050 (0.006)** 0.045 (0.006)**
0.095 (0.054)
651
0.14
Germany
0.047 (0.012)** 0.040 (0.012)** 0.152 (0.053)**
396
0.26
Hungary
0.090 (0.014)** 0.075 (0.016)** 0.253 (0.096)**
382
0.18
Italy
0.072 (0.005)** 0.065 (0.005)** 0.090 (0.041)*
529
0.42
Netherlands
0.044 (0.005)** 0.035 (0.006)** 0.233 (0.054)**
759
0.26
Norway
0.044 (0.006)** 0.039 (0.007)** 0.098 (0.046)*
966
0.15
Poland
0.099 (0.010)** 0.098 (0.011)**
0.012 (0.046)
543
0.16
Slovenia
0.080 (0.008)** 0.067 (0.009)** 0.116 (0.041)**
470
0.22
Sweden
0.034 (0.005)** 0.028 (0.005)** 0.129 (0.034)**
743
0.16
Switzerland
0.049 (0.008)** 0.032 (0.008)** 0.236 (0.048)**
607
0.27
United States
0.092 (0.008)** 0.063 (0.010)** 0.238 (0.047)**
686
0.35
Source: Leuven et.al. (2004)
Note: Standard errors in parentheses. * significant at the 5% level; ** significant at the 1%
level.
a R2 is from regression with IALS score.
31