The effect of cognitive and noncognitive abilities on earnings

The effect of cognitive and noncognitive abilities on earnings: different school systems
The effect of cognitive and noncognitive abilities
on earnings: different school systems
Christoph Weiss
European University Institute
May 2009
The effect of cognitive and noncognitive abilities on earnings: different school systems
Outline
1. Motivation and background
2. Data
3. Methodology
4. Empirical model
5. Basic results
6. Discussion
The effect of cognitive and noncognitive abilities on earnings: different school systems
Motivation
• Cognitive abilities are important determinants of educational
and labor market outcomes
• Personality or noncognitive traits, such as motivation,
attentiveness, self-esteem and perseverance matter as well
• Noncognitive abilities more malleable at later age than
cognitive abilities
• Hypothesis: important implications for the design of the
school system and its effects on labor market outcomes
• Problem: abilities not observable
The effect of cognitive and noncognitive abilities on earnings: different school systems
Background
• Some countries have a comprehensive school system
• Others: system tracking students at the end of primary school
(age 10-11)
• students separated and attend different types of secondary
schools
• Continuous debate among policymakers about the merits of
early tracking as opposed to a comprehensive system
The effect of cognitive and noncognitive abilities on earnings: different school systems
Background
• In the late 60s, the British education system gradually moved
from a system in which tracking was determined by ability
tests at age 11 to a system with no tracking until age 16
• Two different school systems were cohabiting
• old system with grammar and secondary modern schools
• new system with comprehensive schools
• This period of uneven transition is the focus of this study
The effect of cognitive and noncognitive abilities on earnings: different school systems
Literature on tracking
• Galindo-Rueda and Vignoles (2007)
• in the comprehensive system early ability matters less and family
background helps much more in determining educational attainment
• Brunello and Checchi (2007)
• focus on educational performance or attainment may over- or
under-estimate the impact of tracking
• Different countries: Meghir and Palme (2005, Sweden),
Pekkarinen (2005, Finland), Duflo et al. (2008, Kenya)
• Across countries: Hanushek and Woessman (2006), Brunello
and Checchi (2006)
The effect of cognitive and noncognitive abilities on earnings: different school systems
Literature on cognitive and noncognitive skills
and labor market outcomes
• Cunha et al. (2006)
• after age 10 cognitive skills as measured by IQ become stable
• noncognitive skills more maellable at later age
• Heckman and Rubinstein (2001)
• effects of cognitive and noncognitive skills on earnings for high
school dropouts (GED recipients)
• Heckman et al. (2006)
• cognitive and noncognitive abilities explain a variety of labor
market outcomes and risky social behaviors
• Carneiro et al. (2007) confirm evidence for the U.K.
The effect of cognitive and noncognitive abilities on earnings: different school systems
Data
• National Child Development Study (NCDS): a longitudinal
dataset with rich information about all children born in the
UK between 3 and 9 March 1958
• Surveys at age 7, 11, 16, 23, 33, 41 and 46
• I use data on family structure, social environment, educational
attainment and measures of cognitive and noncognitive
abilities
• Some children started secondary school in the system with
early tracking but ended their schooling in the comprehensive
system
• information on which type of schools was attended and year
when the school changed into the comprehensive system
The effect of cognitive and noncognitive abilities on earnings: different school systems
Measure of cognitive abilities
• 4 test scores taken at age 11
• verbal score on a general ability test
• non verbal score on a general ability test
• reading comprehension test
• arithmetic-mathematics test
• I construct four ordinal variables ranking individual’s results by
quartile for each test
• Aim of measure: describe ability to solve abstract problems
The effect of cognitive and noncognitive abilities on earnings: different school systems
2 different measures of noncognitive abilities
• 8 questions (out of 14) read to the mother about the behavior
of her child
• has difficulty in settling to anything for more than a few moments,
is bullied by other children, destroys own or others belongings (e.g.
tears or breaks), is miserable or tearful, is squirmy or fidgety, is
irritable or quick to fly off the handle, fights with other children, is
disobedient
• 9 questions (out of 12) asked to the teacher of the child at
age 11 about behavior problems
• based on the 12 syndromes of the Bristol Social Adjustment Guide
• withdrawal, depression, hostility towards adults, writing off adults
and adults’ standards, anxiety for acceptance by children, hostility
towards children, restlessness, inconsequential behavior,
miscellaneous symptoms
The effect of cognitive and noncognitive abilities on earnings: different school systems
Unobservable abilities
• Items (or questions) provide the basis for the measurement
• Measuring abilities using a single question or a standardized
average of the questions would fail to take into account the
possible measurement uncertainty
• use of variables contaminated by measurement error would
produce biased and inconsistent results
• Item response theory methods for the measurement of latent
variables
• Estimate a probability distribution of unobservable abilities for
each individual
The effect of cognitive and noncognitive abilities on earnings: different school systems
An item response theory method
• Spady (2007): answers to the questions (or item responses)
ordered discrete
• Appropriate notion of monotonicity: first-order stochastic
dominance
• responses of individuals with higher ability first-order
stochastically dominates the responses of individuals with
lower ability
• Observable characteristics, such as the social environment of
the child, also used to measure the latent variables
• measures that do not take into account the effect of family
background “are at best imperfect measures of an individual’s
true noncognitive and cognitive abilities” (Heckman et al.
2006)
The effect of cognitive and noncognitive abilities on earnings: different school systems
An item response theory method
• Probability of a particular responses pattern conditional on
the latent variable θ
p (r1 , r2 , ..., rm |θ ) = p1 (r1 |θ )p2 (r2 |θ )...pm (rm |θ )
(1)
• Probability of a particular response pattern
Z
p (r1 , r2 , ..., rm ) =
p1 (r1 |θ )p2 (r2 |θ )...pm (rm |θ )f (θ )d θ (2)
• Using characteristics to improve the measurement of abilities
p (r1 , r2 , ..., rm |W ) =
=
Z
Z
p1 (r1 |θ, W )p2 (r2 |θ, W )...pm (rm |θ, W )f (θ |W )d θ
(3a)
p1 (r1 |θ )p2 (r2 |θ )...pm (rm |θ )f (θ |W )d θ
(3b)
• Convenient choice: specify f (θ |W ) with N (W β, 1)
The effect of cognitive and noncognitive abilities on earnings: different school systems
Estimation
• By maximum likelihood, subject to the constraint that the
item characteristic curves are monotonic decreasing and not
crossing
• Curves approximated using exponential tilting with two
parameters and shifted Legendre polynomials as basis
functions
• Using Bayes’ Theorem posterior distribution of latent variable
for each individual inferred from the item responses and
characteristics
f ( θ |r , W ) =
f (θ, r |W )
p (r |θ, W )f (θ |W )
p (r | θ )f ( θ |W )
=
=
p (r |W )
p (r |W )
p (r |W )
(4)
The effect of cognitive and noncognitive abilities on earnings: different school systems
Item characteristics curves and
item category response functions
0.8
0.6
0.0
0.2
0.4
F(r|theta)
0.6
0.4
0.0
0.2
F(r|theta)
0.8
1.0
restless
1.0
anxiety acceptance kids
−2
−1
0
1
2
−2
−1
1
2
1
2
1.0
0.8
0.6
0.0
0.2
0.4
f(r|theta)
0.8
0.6
0.4
0.2
0.0
f(r|theta)
0
theta
1.0
theta
−2
−1
0
theta
1
2
−2
−1
0
theta
The effect of cognitive and noncognitive abilities on earnings: different school systems
Posteriors of cognitive abilities for 10 individuals
with two different response patterns
0.0
0.4
0.8
Response pattern 1211
−4
−2
0
2
4
theta
0.0
0.4
0.8
Response pattern 2244
−4
−2
0
theta
2
4
The effect of cognitive and noncognitive abilities on earnings: different school systems
Effect of the characteristics on cognitive abilities, Males (1)
Mother’s age
Father’s age
Mother’s years of education
Father’s years of education
Mother’s interest in child’s education - very
Mother’s interest in child’s education - some
Father’s interest in child’s education - very
Father’s interest in child’s education - some
1 older brother
2 older brothers or more
1 older sister
2 older sisters or more
Note: White robust standard errors in parentheses.
Coef.
.022
.006
.089
.048
.393
.065
.567
.292
-.170
-.314
-.174
-.327
S.E.
(.009)
(.008)
(.024)
(.018)
(.098)
(.085)
(.089)
(.070)
(.062)
(.096)
(.063)
(.096)
The effect of cognitive and noncognitive abilities on earnings: different school systems
Effect of the characteristics on cognitive abilities, Males (1)
English not spoken at home
Mother has a paid job
Serious financial hardship last year
Free school meals
Help mental backwardness
Help mental superiority
Streamed class - higher ability
Streamed class - average ability
Streamed class - lower ability
Class size
Teacher female
Any outstanding ability
Coef.
-.092
-.008
-.250
-.042
-1.236
.565
.767
-.190
-1.087
.004
.034
.141
Note: White robust standard errors in parentheses.
S.E.
(.164)
(.052)
(.103)
(.132)
(.107)
(.230)
(.075)
(.082)
(.101)
(.004)
(.054)
(.062)
The effect of cognitive and noncognitive abilities on earnings: different school systems
Effect of the characteristics on noncognitive abilities,
Males (2)
Mother’s age
Father’s age
Mother’s years of education
Father’s years of education
Mother’s interest - very
Mother’s interest - some
Father’s interest - very
Father’s interest - some
1 older brother
2 older brothers or more
1 older sister
2 older sisters or more
Teacher
Coef.
S.E.
-.010
(.009)
.002
(.008)
-.015
(.022)
-.008
(.018)
1.091
(.113)
.888
(.100)
.456
(.098)
.200
(.075)
.042
(.064)
.117
(.091)
.047
(.066)
-.011
(.104)
Note: White robust standard errors in parentheses.
Mother
Coef.
S.E.
-.002
(.009)
.008
(.008)
-.018
(.023)
-.007
(.018)
.419
(.102)
.253
(.087)
.096
(.093)
.106
(.075)
-.094
(.065)
.194
(.095)
.060
(.070)
.085
(.117)
The effect of cognitive and noncognitive abilities on earnings: different school systems
Effect of the characteristics on noncognitive abilities,
Males (2)
English not spoken at home
Mother has a paid job
Serious financial hardship last year
Free school meals
Help mental backwardness
Help mental superiority
Streamed class - higher ability
Streamed class - average ability
Streamed class - lower ability
Class size
Teacher female
Any outstanding ability
Teacher
Coef.
S.E.
-.040
(.149)
.067
(.056)
-.031
(.106)
.083
(.128)
-.445
(.089)
.156
(.208)
.258
(.083)
.070
(.098)
.017
(.096)
.001
(.004)
.079
(.056)
.159
(.067)
Note: White robust standard errors in parentheses.
Mother
Coef.
S.E.
.279
(.159)
-.059
(.058)
-.191
(.103)
-.159
(.126)
-.287
(.108)
-.014
(.254)
.016
(.090)
.198
(.088)
-.111
(.101)
.000 (.004)
.045
(.056)
.042
(.067)
The effect of cognitive and noncognitive abilities on earnings: different school systems
Empirical model
Y = hY (C , N, T , S, X , U )
(5)
• Y log earnings (continuous), C and N cognitive and noncognitive
abilities (continuous), T school system (dummy), S educational
attainment (discrete), X ≡ {Xk }K
k =1 vector of covariates
• Assume unobserved errors U i.i.d. with U ∼ Fu and function hY
differentiable and strictly monotonic increasing in U
• Consider a linear additive form of the model
Y = C α1 + Nα2 + T α3 + S α4 + X > α5 + U
(6)
• Interested in estimating parameters α1 , α2 given values of T , S, X
The effect of cognitive and noncognitive abilities on earnings: different school systems
Correlation between abilities and educational levels
• I compute the expectation of the posterior distribution of the
latent
variables for each individual
R
θf (θ |r , W )d θ = E (θ |r , W )
• New variables standardized to have mean 0 and variance 1
Cognitive abilities at age 7
Noncognitive (mother) at age 7
Noncognitive (teacher) at age 7
Cognitive abilities at age 11
Noncognitive (mother) at age 11
Noncognitive (teacher) at age 11
Educational levels
C7
M7
T7
C11
M11
T11
.140
.300
.514
.140
.242
.300
.191
.120
.345
.158
.103
.271
.150
.310
.150
.146
.290
.394
.170
.118
.192
Note: Kendall τ rank correlation coefficient. Educational levels are coded 0 to 5. Sample size is 1622.
Edu
The effect of cognitive and noncognitive abilities on earnings: different school systems
Kernel density estimation of abilities, by education level
0.6
Cognitive abilities
0.3
0.0
0.1
0.2
Density
0.4
0.5
Level 0
Level 1
Level 2
Level 3
Level 4
Level 5
−3
−2
−1
0
1
0.5
Noncognitive abilities (teacher)
0.3
0.4
Level 0
Level 1
Level 2
Level 3
Level 4
Level 5
0.1
0.0
0.1
0.2
Density
0.3
0.4
Level 0
Level 1
Level 2
Level 3
Level 4
Level 5
0.0
Density
3
0.2
0.5
Noncognitive abilities (mother)
2
−3
−2
−1
0
1
2
3
−3
−2
−1
0
1
2
3
The effect of cognitive and noncognitive abilities on earnings: different school systems
Basic results: Males (N = 1168)
• Dependent variable: log hourly earnings
• Noncognitive abilities as assessed by the mother
(A)
(B)
(C)
(D)
Coef.
S.E.
Coef.
S.E.
Coef.
S.E.
Cognitive abilities
.132
(.023)
.094
(.030)
.101
(.028)
.032
(.017)
Noncognitive (mother)
.043
(.017)
.041
(.019)
Compr. school system
-.077
(.067)
-.067
(.067)
-.068
(.067)
Years school compr.
.009
(.015)
.009
(.015)
.010
(.015)
Educational level 1
.310
(.111)
.344
(.114)
.342
(.113)
Educational level 2
.281
(.099)
.281
(.100)
.273
(.100)
Educational level 3
.301
(.104)
.294
(.108)
.284
(.108)
Educational level 4
.465
(.107)
.447
(.110)
.430
(.110)
Educational level 5
.539
(.131)
.508
(.134)
.482
(.135)
2.182
(.118)
Constant
2.477
(.030)
2.114
(.099)
2.177
(.118)
Note: MacKinnon and White (1985) HC3 standard errors in parentheses. Model (D) uses a standardized average of
the items. Models (C) and (D) include all the characteristics.
Coef.
.187
.043
-.079
.008
S.E.
(.018)
(.017)
(.068)
(.015)
The effect of cognitive and noncognitive abilities on earnings: different school systems
Basic results: Males (N = 1168)
• Dependent variable: log hourly earnings
• Noncognitive abilities as assessed by the teacher
(A)
(B)
(C)
(D)
Coef.
S.E.
Coef.
S.E.
Coef.
S.E.
Cognitive abilities
.122
(.023)
.085
(.030)
.096
(.028)
.033
(.023)
Noncognitive (teacher)
.043
(.021)
.042
(.022)
Compr. school system
-.065
(.068)
-.056
(.067)
-.060
(.067)
Years school compr.
.006
(.015)
.005
(.015)
.007
(.014)
Educational level 1
.302
(.113)
.336
(.115)
.332
(.114)
Educational level 2
.277
(.101)
.276
(.102)
.267
(.101)
Educational level 3
.293
(.106)
.285
(.110)
.275
(.109)
Educational level 4
.456
(.110)
.438
(.111)
.419
(.112)
Educational level 5
.530
(.134)
.500
(.136)
.474
(.137)
2.205
(.120)
Constant
2.477
(.030)
2.130
(.108)
2.203
(.122)
Note: MacKinnon and White (1985) HC3 standard errors in parentheses. Model (D) uses a standardized average of
the items. Models (C) and (D) include all the characteristics.
Coef.
.174
.048
-.067
.005
S.E.
(.019)
(.021)
(.069)
(.015)
The effect of cognitive and noncognitive abilities on earnings: different school systems
Basic results using a Tobit 2 selection model
• Dependent variable: log hourly earnings
• Noncognitive abilities as assessed by the teacher
Cognitive abilities
Noncognitive (teacher)
Comprehensive school system
Years school comprehensive
Educational level 1
Educational level 2
Educational level 3
Educational level 4
Educational level 5
Constant
σ
ρ
Male
Coef.
S.E.
.127 (.025)
.044 (.021)
-.065 (.064)
.007 (.014)
.317 (.120)
.285 (.106)
.303 (.109)
.487 (.124)
.558 (.136)
2.054 (.173)
.612 (.019)
.150 (.307)
Female
Coef.
S.E.
.070 (.022)
.005 (.020)
-.104 (.063)
.021 (.014)
.055 (.084)
.168 (.075)
.191 (.084)
.552 (.080)
.799 (.108)
1.605 (.112)
.595 (.030)
.227 (.140)
Note: For males, 550 observations are censored and 1168 uncensored. For females, 569 observations are censored
and 1204 uncensored.
The effect of cognitive and noncognitive abilities on earnings: different school systems
Basic results from quantile regression: Males (N = 1168)
−0.010
0.00
0.10
0.02
0.000
0.04
0.18
0.14
yearscompr
0.010
noncognitive_mother
0.06
cognitive
0.2
0.4
0.6
0.8
0.2
0.6
0.8
0.2
educ[, 1]
0.6
0.8
educ[, 2]
0.30
0.4
0.3
0.05
0.10
0.1
0.20
0.2
0.00
−0.10 −0.05
0.4
0.40
comprehensive
0.4
0.2
0.4
0.6
0.8
0.2
0.6
0.8
0.2
0.4
0.6
0.8
educ[, 5]
0.5
0.4
0.3
0.3
0.20
0.4
0.30
0.5
0.6
0.6
0.7
educ[, 4]
0.40
educ[, 3]
0.4
0.2
0.4
0.6
0.8
0.2
0.4
0.6
0.8
0.2
0.4
0.6
0.8
The effect of cognitive and noncognitive abilities on earnings: different school systems
Basic results from quantile regression: Males (N = 1168)
• Dependent variable: log hourly earnings
• Noncognitive abilities as assessed by the mother
τ
Cognitive abilities
Noncognitive (mother)
Compr. school system
Years school compr.
Educational level 1
Educational level 2
Educational level 3
Educational level 4
Educational level 5
Constant
0.25
Coef.
S.E.
.114 (.017)
.031 (.015)
-.005 (.011)
.010 (.048)
.361 (.083)
.329 (.072)
.409 (.075)
.622 (.076)
.717 (.084)
1.730 (.070)
Note: White robust standard errors in parentheses.
0.5
Coef.
.138
.021
-.003
.003
.164
.205
.235
.472
.518
2.130
S.E.
(.016)
(.014)
(.011)
(.047)
(.164)
(.163)
(.162)
(.163)
(.176)
(.161)
0.75
Coef.
S.E.
.167 (.017)
.027 (.014)
.003 (.011)
.027 (.054)
.098 (.098)
.122 (.102)
.220 (.106)
.351 (.100)
.434 (.104)
2.459 (.100)
The effect of cognitive and noncognitive abilities on earnings: different school systems
Marginal effects from an ordered probit model evaluated at
the mean, for each educational level (coded 0 to 5),
Males (N = 1718)
0
Cognitive abilities
Noncognitive (teacher)
Compr. school system
Years school compr.
-.032
-.003
.003
-.001
Cognitive abilities
Noncognitive (teacher)
Compr. school system
Years school compr.
.008
.001
-.001
.000
1
(.004)
(.002)
(.005)
(.001)
-.066
-.007
.006
.002
(.004)
(.001)
(.001)
(.000)
.171
.019
-.017
.005
3
2
(.006)
(.003)
(.010)
-(.900)
-.133
-.015
.013
-.004
(.011)
(.008)
(.026)
(.006)
.052
.006
-.005
.002
4
(.009)
(.006)
(.020)
(.004)
5
(.005)
(.002)
(.008)
(.002)
The effect of cognitive and noncognitive abilities on earnings: different school systems
Marginal effects from an ordered probit model evaluated at
the mean, for each educational level (coded 0 to 5),
Females (N = 1773)
0
Cognitive abilities
Noncognitive (teacher)
Compr. school system
Years school compr.
-.051
-.011
.020
-.004
Cognitive abilities
Noncognitive (teacher)
Compr. school system
Years school compr.
.029
.004
-.010
.002
1
(.005)
(.002)
(.007)
(.002)
-.086
-.013
.035
-.007
(.004)
(.001)
(.004)
(.001)
.175
.026
-.073
.014
3
2
(.007)
(.004)
(.012)
(.003)
-.104
-.016
.045
-.008
(.010)
(.008)
(.026)
(.006)
.038
.006
-.017
.003
4
(.008)
(.005)
(.017)
(.003)
5
(.004)
(.002)
(.007)
(.001)
The effect of cognitive and noncognitive abilities on earnings: different school systems
IV model
• Suppose C11 endogenous
Y = C11 α1 + N11 α2 + T α3 + S α4 + X > α5 + U + V
>
>
C11 = Z β 1 + X β 2 + V .
(7a)
(7b)
• Instruments: cognitive and noncognitive abilities at age 7, father’s
(mother’s) education, father’s (mother’s) interest in the education
of his child at age 7, older siblings
• Suppose S endogenous
Y = C α1 + Nα2 + T α3 + S α4 + X > α5 + U + V
>
>
S = C β1 + N β2 + T β3 + Z β4 + X β5 + V
• Instruments: father’s (mother’s) education, father’s (mother’s)
interest in the education of his child, older siblings
(8a)
(8b)
The effect of cognitive and noncognitive abilities on earnings: different school systems
OLS and IV results: Males (N = 543)
• Dependent variable: log hourly earnings
• Endogenous variable: cognitive abilities at age 11
• Noncognitive abilities as assessed by the mother
OLS
Coef.
.070
S.E.
(.033)
GMM
(C)
Coef.
S.E.
.160
(.052)
.058
-.169
.012
.450
.379
.354
.595
.632
2.083
(.030)
(.095)
(.021)
(.171)
(.147)
(.153)
(.152)
(.172)
(.145)
.033
-.170
.016
.446
.348
.284
.485
.491
2.083
(A)
Cognitive abilities
Noncognitive at age 7
Noncognitive at age 11
Compr. school system
Years school compr.
Educational level 1
Educational level 2
Educational level 3
Educational level 4
Educational level 5
Constant
Coef.
.095
.013
S.E.
(.032)
(.029)
-.168
.013
.422
.377
.364
.614
.637
2.078
(.095)
(.021)
(.172)
(.148)
(.152)
(.149)
(.170)
(.145)
(B)
Note: White robust standard errors in parentheses. Model C is an instrumental variable model.
(.032)
(.095)
(.021)
(.167)
(.150)
(.166)
(.166)
(.204)
(.145)
The effect of cognitive and noncognitive abilities on earnings: different school systems
IV results: Males (N = 1168)
• Dependent variable: log hourly earnings
• Endogenous variable: educational levels
• Instruments: father’s (mother’s) education, father’s (mother’s)
interest in the education of his child, older siblings
(A)
Cognitive abilities
Noncognitive abilities (mother)
Noncognitive abilities (teacher)
Comprehensive school system
Years school comprehensive
Educational levels
Constant
Coef.
.076
.039
S.E.
(.052)
(.017)
-.111
.015
.173
1.969
(.065)
(.014)
(.076)
(.225)
(B)
Coef.
.068
S.E.
(.051)
.041
-.097
.012
.168
2.054
(.021)
(.065)
(.014)
(.076)
(.173)
Note: White robust standard errors in parentheses. Educational levels are coded 0 to 5.
The effect of cognitive and noncognitive abilities on earnings: different school systems
Conclusion
• Cognitive and noncognitive abilities at age 11 have an effect
on the earnings of men at age 41
• School system does not appear to have an effect
• Results different for women
• Overall results not inconsistent with education signaling in the
labor market
The effect of cognitive and noncognitive abilities on earnings: different school systems
Some further thoughts
• Make use of the panel structure of the data (abilities and
characteristics at age 7, 11, 16)
• Control for characteristics at age 41
• Look at different outcomes (at age 33 and 46)... and present
them
• Exploit the posteriors rather than only their expectations
The effect of cognitive and noncognitive abilities on earnings: different school systems
Characteristics: descriptive statistics, by gender
Mother age
Father age
Mother years of education
Father years of education
Mother interest in education - very
Mother interest in child education - some
Father interest in child education - very
Father interest in child education - some
English not spoken at home
Mother has a paid job
Serious financial hardship last year
Free school meals
Help mental backwardness
Help mental superiority
Streamed class - higher ability
Streamed class - average ability
Streamed class - lower ability
Class size
Teacher female
Any outstanding ability
1 older brother
2 older brothers or more
1 older sister
2 older sisters or more
Sample size
Male
Mean
S.D.
27.33
5.41
30.20
5.96
9.94
1.40
9.95
1.85
0.45
0.42
0.32
0.33
0.03
0.62
0.08
0.06
0.08
0.01
0.14
0.10
0.08
34.78
7.30
0.41
0.26
0.27
0.12
0.28
0.10
1718
Female
Mean
S.D.
27.54
5.59
30.51
6.25
10.04
1.62
10.01
1.92
0.49
0.39
0.36
0.29
0.02
0.63
0.09
0.06
0.04
0.02
0.16
0.10
0.06
35.00
6.94
0.46
0.24
0.28
0.11
0.28
0.11
1773
The effect of cognitive and noncognitive abilities on earnings: different school systems
Outcomes: descriptive statistics, by gender
Hourly gross pay
Weekly gross pay
Sample size
Mean
14.68
589.39
Male
S.D.
13.50
573.42
1168
Median
11.77
475.00
Mean
9.34
284.62
Female
S.D. Median
9.72
7.25
339.95 225.00
1204
The effect of cognitive and noncognitive abilities on earnings: different school systems
Education levels, by gender and school system
No qualification
Low level q.
O-level and equivalent q.
A-level and equivalent q.
First degree and equivalent q.
Higher degree and equivalent q.
Sample size
No qualification
Low level q.
O-level and equivalent q.
A-level and equivalent q.
First degree and equivalent q.
Higher degree and equivalent q.
Sample size
Years
of school
9
10
11
13
16
17.5
Years
of school
9
10
11
13
16
17.5
Males
0.05
0.08
0.27
0.23
0.31
0.07
1718
Females
0.08
0.11
0.32
0.15
0.29
0.05
1773
Comprehensive
M
F
0.06
0.09
0.08
0.12
0.28
0.35
0.23
0.15
0.27
0.25
0.07
0.04
554
543
Early tracking
M
F
0.03
0.06
0.09
0.10
0.25
0.27
0.21
0.15
0.36
0.35
0.07
0.07
413
443