The Technology of Skill Formation - Home | HCEO · Extract for \The Technology of Skill Formation" by James J. Heckman and Flavio Cunha. Published in the American Economic Review

$Page 1: The Technology of Skill Formation - Home | HCEO · Extract for \The Technology of Skill Formation" by James J. Heckman and Flavio Cunha. Published in the American Economic Review$
Observations Abilities/Outcomes Model

Extract for “The Technology of Skill Formation”

by James J. Heckman and Flavio Cunha. Published in the American Economic Review

(2007). Volume 97, Issue 2, pp.31–47.

James J. HeckmanUniversity of Chicago

AEA Continuing Education ProgramASSA Course: Microeconomics of Life Course Inequality

San Francisco, CA, January 5-7, 2016

by James J. Heckman and Flavio Cunha. Published in the American Economic Review (2007). Volume 97, Issue 2, pp.31–47. [4mm]James J. Heckman University of ChicagoExtract for “The Technology of Skill Formation” 1 / 76


Cunha and Heckman, 2007, and later work

This paper uses simple economic models of skill formation toorganize a large body of evidence on the development of skills inchildren in economics, psychology, education and neuroscience.

The existing theoretical literature on child development ineconomics treats childhood as a single period (see, e.g., Gary S.Becker and Nigel Tomes, 1986; Roland Benabou, 2000; S. RaoAiyagari et al., 2002).

The implicit assumption in this approach is that inputs into theproduction of skills at different stages of childhood are perfectsubstitutes.



To account for a large body of evidence, it is important to buildmodels of skill formation with multiple stages of childhood,where inputs at different stages are complements and wherethere is self-productivity of investment.

In order to rationalize the evidence, it is also important torecognize three distinct credit constraints operating on thefamily and its children.

1 The inability of a child to choose its parents. This is thefundamental constraint imposed by the accident of birth.

2 The inability of parents to borrow against their children’sfuture income to finance investments in them.

3 The inability of parents to borrow against their own income tofinance investments in their children.



A model that is faithful to the evidence must recognize:

a Parental influences are key factors governing childdevelopment;

b Early childhood investments must be distinguished from latechildhood investments;

c An equity-efficiency trade-off exists for late investments, butnot for early investments;

d Abilities are created, not solely inherited, and are multiple invariety;

e The traditional ability-skills dichotomy is misleading. Bothskills and abilities are created; and

f The “nature versus nurture” distinction is obsolete.g Gene-environment or epigenetic interactions are important.



These insights change the way we interpret evidence and designpolicy about investing in children.

Point (a) is emphasized in many papers.

Point (b) is ignored in models that consider only one period ofchildhood investment.

Points (c), (d) and (e) have received scant attention in theformal literature on child investment.

Point (f) is ignored in the literature that partitions the varianceof child outcomes into components due to nature andcomponents due to nurture.



Observations About Human Diversity and HumanDevelopment and Some Facts Our Model Explains



Three Observations

The first observation is that ability matters.

A large number of empirical studies document that cognitiveability is a powerful determinant of wages, schooling,participation in crime and success in many aspects of social andeconomic life.



Second observation: Abilities are multiple in nature.

Noncognitive abilities (perseverance, motivation, timepreference, risk aversion, self-esteem, self-control, preferencesfor leisure) have direct effects on wages (controlling forschooling), schooling, teenage pregnancy, smoking, crime,performance on achievement tests and many other aspects ofsocial and economic life.

See, e.g., Samuel Bowles and Herb Gintis (1976); SamuelBowles et al. (2001); Lex Borghans et al. (2006); Heckman etal. (2006).



Some Evidence on the Importance of Cognitive andNoncognitive Skills

Recent research has shown that earnings, employment, laborforce experience, college attendance, teenage pregnancy,participation in risky activities, compliance with healthprotocols and participation in crime strongly depend oncognitive and noncognitive abilities.

Noncognitive abilities mean socioemotional regulation, timepreference, personality factors and ability to work with others.



Much public policy discussion is devoted to cognitive scores or“smarts.”

Many governments and public policy officials focus onachievement on a test score at a certain age to measure successor failure in schools.

Yet an emerging body of evidence shows that, as is intuitivelyobvious and commonsensical, much more than smarts arerequired.

Motivation,Sociability; ability to work with others,Attention,Self Regulation,Self Esteem,Time Preference,Health and Mental Health.



The importance of noncognitive skills, for example, tends to beunderrated in current policy discussions because they are noteasily measured.

A lot of recent evidence shows that the workplace isincreasingly oriented towards a greater valuation of socialinteraction and sociability.

Evidence from the GED program (Heckman and Rubinstein,2001).



The GED program is a second chance program given tosecondary school dropouts in the U.S. and Canada.

Participation in the GED program is growing. Currently 20% ofU.S. high school “graduates” are dropouts who exam certify.

GEDs are required to pass a test of cognitive abilities.

Level relatively low–at the grade 8 to grade 10 level.

Test is successful in its own terms.



Density of age adjusted AFQT scores, GED recipients and high schoolgraduates with twelve years of schooling

White Males White Females(a) White males

0

5

10

15

20

25

30

-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5

(b) White females

0

5

10

15

20

25

30

35

-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5

HS graduates

GEDs

Source: Heckman, Hsee and Rubinstein (2001)




Black Males Black Females

0

5

10

15

20

25

30

35

-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5

HS graduates

GEDs

0

5

10

15

20

25

30

35

40

-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5





Hispanic Males Hispanic FemalesHS graduates

GEDs

0

5

10

15

20

25

30

-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5

0

5

10

15

20

25

30

35

-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5




Yet GEDs earn at the rate of high school dropouts.

GEDs are as “smart” as ordinary high school graduates.

They lack noncognitive skills.

The GEDs are the wise guys who can’t finish anything.

Most branches of the U.S. military recognize this in theirrecruiting strategies.



There is a lot of evidence that both cognitive and noncognitiveskills are important.

Both cognitive and noncognitive skills matter in a variety ofaspects of life.



Cognitive and noncognitive ability are important determinantsof schooling and socioeconomic success.

In the U.S. and many countries around the world, schoolinggaps across ethnic and income groups have more to do withability deficits than family finances in the school-going years.

Those with higher abilities of both types are more likely to takepost-school company job training, to participate in preventionprograms; less likely be obese; have greater health and mentalhealth.

Look at effects of both cognitive and noncognitive skills onmany measures of social performance.



Ever been in jail by age 30, by ability (males)

Pro

bab

ility

Percentile

.15

.10

.05

.0

0 - 20 21 - 40 41 - 60 61 - 80 81 - 100

Cognitive Noncognitive

Note: This figure plots the probability of a given behavior associated with moving upin one ability distribution for someone after integrating out the other distribution. Forexample, the lines with markers show the effect of increasing noncognitive ability afterintegrating the cognitive ability.

Source: Heckman, Stixrud, and Urzua (2006).



Probability of being single with children (females)

Pro

bab

ility

Percentile

.10

.08

.06

.04

.02

0 - 20 21 - 40 41 - 60 61 - 80 81 - 100

Cognitive Noncognitive

Note: This figure plots the probability of a given behavior associated with moving upin one ability distribution for someone after integrating out the other distribution. Forexample, the lines with markers show the effect of increasing noncognitive ability afterintegrating the cognitive ability.

Source: Heckman, Stixrud, and Urzua (2006).by James J. Heckman and Flavio Cunha. Published in the American Economic Review (2007). Volume 97, Issue 2, pp.31–47. [4mm]James J. Heckman University of ChicagoExtract for “The Technology of Skill Formation” 20 / 76


Probability of being a high school dropout by age 30 (males)

24

68

10

24

68

10

0

0.2

0.4

0.6

0.8

1

Decile of Noncognitive

Figure 1A. Probability of Being a High School Dropout by Age 30 - Malesi. By Decile of Cognitive and Noncognitive Factors

Decile of Cognitive

Pro

bab

ility

1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8

1ii. By Decile of Cognitive Factor

Decile

Pro

bab

ility

an

dC

on

fiden

ce I

nte

rval

(2.

5-97

.5%

)

Notes: The data are simulated from the estimates of the model and our NLSY79 sample. We use the standard convention that higher deciles are associated with higher values of the variable.The confidence intervals are computed using bootstrapping (200 draws).

1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8

1iii. By Decile of Noncognitive Factor

Decile

24

68

10

24

68

10

0

0.2

0.4

0.6

0.8

1


Figure 1A. Probability of Being a High School Dropout by Age 30 - Malesi. By Decile of Cognitive and Noncognitive Factors

Decile of Cognitive

Pro

bab

ility

1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8


Decile

Pro

bab

ility

an

dC

on

fiden

ce I

nte

rval

(2.

5-97

.5%

)


1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8


Decile



Probability of being a 4-year college graduate by age 30 (males)

24

68

10

123456789100

0.2

0.4

0.6

0.8

1


Figure 1C. Probability of Being a 4-yr College Graduate by Age 30 - Malesi. By Decile of Cognitive and Noncognitive Factors

Decile of Cognitive

Pro

bab

ility

1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8


Decile

Pro

bab

ility

an

dC

on

fiden

ce I

nte

rval

(2.

5-97

.5%

)


1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8


Decile

24

68

10

123456789100

0.2

0.4

0.6

0.8

1


Figure 1C. Probability of Being a 4-yr College Graduate by Age 30 - Malesi. By Decile of Cognitive and Noncognitive Factors

Decile of Cognitive

Pro

bab

ility

1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8


Decile

Pro

bab

ility

an

dC

on

fiden

ce I

nte

rval

(2.

5-97

.5%

)


1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8


Decile



Probability of daily smoking by age 18 (males)

24

68

10

24

68

10

0

0.2

0.4

0.6

0.8

1


Figure 1F. Probability Of Daily Smoking By Age 18 - Malesi. By Decile of Cognitive and Noncognitive Factor

Decile of Cognitive

Pro

bab

ility

1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8


Decile

Pro

bab

ility

an

dC

on

fiden

ce I

nte

rval

(2.

5-97

.5%

)


1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8


Decile

24

68

10

24

68

10

0

0.2

0.4

0.6

0.8

1


Figure 1F. Probability Of Daily Smoking By Age 18 - Malesi. By Decile of Cognitive and Noncognitive Factor

Decile of Cognitive

Pro

bab

ility

1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8


Decile

Pro

bab

ility

an

dC

on

fiden

ce I

nte

rval

(2.

5-97

.5%

)


1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8


Decile



Mean log wages by age 30 (males)

24

68

10

123456789102

2.2

2.4

2.6

2.8

3

Decile of Non-Cognitive

Mean Log Wages by Age 30 - Males i. By Decile of Cognitive and Non-Cognitive Factors

Decile of Cognitive

Lo

g W

ages

2 4 6 8 102

2.5

3i. By Decile of Cognitive Factor

Decile

Lo

g W

ages

an

dC

on

fid

ence

In

terv

al (

2.75

-97.

5%)

2 4 6 8 102

2.5

3ii. By Decile of Non-Cognitive Factor

Decile

Notes: The data are simulated from the estimates of the model and our NLSY79 sample. We use the standard convention that higher deciles are associated withhigher values of the variable. The confidence intervals are computed using bootstrapping (50 draws).

Mean Log Wages by Age 30 - Males



Controlling for ability, minorities are more likely to attendcollege than others despite their lower family incomes(Cameron and Heckman, 2001).



A third observation is that the “nature versus nurture”distinction is obsolete.

The modern literature on epigenetic expression teaches us thatthe sharp distinction between acquired skills and ability featuredin the early human capital literature is not tenable.

Additive “nature” and “nurture” models, while traditional andstill used in many studies of heritability and family influence,mischaracterize how ability is manifested.



Abilities are produced, and gene expression is governed byenvironmental conditions (Eric Turkheimer et al., 2003).

Measured abilities are susceptible to environmental influences,including in utero experiences, and also have geneticcomponents.

These factors interact to produce abilities that have both agenetic and an acquired character and the modified genes areheritable. Genes and environment cannot be meaningfullyparsed by traditional linear models that assign variance to eachcomponent, even though it is traditional to do so.



Six Facts We Explain

First, ability gaps between individuals and across socioeconomicgroups open up at early ages, for both cognitive andnoncognitive skills.

Adjusting for family background by regression analysis reducesthese gaps.

Experimental manipulations of early environments (Perry,Abecedarian et al.) show that these effects are causal.



Figure 1: Children of NLSY Average Standardized Score PIAT Math byPermanent Income Quartile Figure 1

Children of NLSYAverage Standardized Score

PIAT Math b Permanent Income uartile

-0.8000

-0.6000

-0.4000

-0.2000

0.0000

0.2000

0.4000

0.6000

0.8000

5 6 7 8 9 10 11 12 13 14

Age

Bottom Quartile Second Quartile Third Quartile Top Quartile

Source: Full Sample of CNLSY



Figure D0: Trend in Mean Cognitive Score by Maternal Education

Source: Brooks-Gunn et al., (2006).by James J. Heckman and Flavio Cunha. Published in the American Economic Review (2007). Volume 97, Issue 2, pp.31–47. [4mm]James J. Heckman University of ChicagoExtract for “The Technology of Skill Formation” 30 / 76


The dramatic results on the importance of the early years increating differences among children shown in the previous grapharise if “Bayley scores” are used as a measure of cognition atage 1.

As Michael Lewis and Harry McGurk (1972) point out, this isillegitimate since the Bayley score tests other aspects of childdevelopment in addition to cognition.



Figure D00: Children of NLSY Average Standardized ScorePeabody Picture Vocabulary Test by Permanent Income QuartileFigure 1

Children of NLSYAverage Standardized Score

^

-0.8000

-0.6000

-0.4000

-0.2000

0.0000

0.2000

0.4000

0.6000

0.8000

3 4 5 6 7 8 9 10 11 12 13 14

Age

Bottom Quartile Second Quartile Third Quartile Top Quartile

Source: Full Sample of CNLSY



Figure D1a. Average percentile rank on PIAT-Math score, by incomequartile

Average Percentile Rank on PIAT-Math Score, by Income Quartile

Age

Scor

e pe

rcen

tile

Highest income quartileThird income quartileSecond income quartileLowest income quartile

10 126 8

65

60

55

50

45

40

35



Figure D1b. Adjusted average PIAT-Math score percentiles, by incomequartile

D . Residualized Average PIAT-Math Score Percentiles by Income Quartile*

10 12

Age* Residualized on maternal education, maternal AFQT (corrected for the effect of schooling) and broken

home at each age

Scor

e pe

rcen

tile


65

60

55

50

45

40

356 8



Figure D2a. Average percentile rank on PIAT-Math score, by race

Average Percentile Rank on PIAT-Math Score, by Race

Age

Scor

e pe

rcen

tile

Hispanic Black White

10 126 8

65

60

55

50

45

40

35



Figure D2b. Adjusted average PIAT-Math score percentiles, by race

F . Residualized Average PIATM Score Percentile by Race*


home at each age


10 126 8

65

60

55

50

45

40

35



Figure D3a. Average percentile rank on anti-social behavior score, byincome quartile

G . Average Percentile Rank on Anti-Social Score, by Income Quartile*

Age

Scor

e pe

rcen

tile


10 126 84

55

50

45

40

35

25

20

30



Figure D3b. Adjusted average anti-social behavior score percentile, byincome quartile

J. Residualized Average Anti-Social Score Percentile by Income Quartile*


home at each age

Scor

e pe

rcen

tile

10 126 84


55

50

45

40

35

25

20

30

60



Figure D4a. Average percentile rank on anti-social behavior score, by race

I . Average Percentile Rank on Anti-Social Score, by Race

Age

Scor

e pe

rcen

tile


10 126 84

55

50

45

40

35

30



Figure D4b. Adjusted average anti-social behavior score percentile, by race

L . Residualized Average Anti-Social Score Percentile by Race*


home at each age

Scor

e pe

rcen

tile

10 126 84


55

50

45

40

35

30



Figure D5a. Early Childhood Longitudinal Study (ECLS)ReadingEarly Childhood Longitudinal Study (ECLS)

(a) Reading

10

20

30

40

50

60

70

55 65 75 85

Age (months)

Rea

ding

Sco

re

Low povertyHigh povery

Figure D5a

Source: Raudenbush (2006)Source: Raudenbush (2006)



Figure D5b. Mean trajectories, high and low priority schools (ECLS)Math Mean trajectories, high and low poverty schools (ECLS)

(b) Math

10

20

30

40

50

60

55 65 75 85

Age (months)

Mat

h Sc

ore

Low povertyHigh poverty

Figure D5b




Figure D6a. Average Trajectories, Grades 1–3, high and low poverty schools(Sustaining Effects Study)ReadingAverage trajectories, Grades 1-3, high and low poverty schools (Sustaining Effects Study)

(a) Reading

350

400

450

500

550

70 75 80 85 90 95

Age (months)

Rea

ding

Sco

re

Low povertyHigh Poverty

Figure D6a




Figure D6b. Average Trajectories, Grades 1–3, high and low povertyschools (Sustaining Effects Study)Math

Average trajectories, Grades 1-3, high and low poverty schools (Sustaining Effects Study)

(b) Math

350

400

450

500

550

70 75 80 85 90 95

Age (months)

Mat

h Sc

ore


Figure D6b




Figure D7a. Average achievement trajectories, grades 8–12, (NELS 88)Science

Average achievement trajectories, Grades 8-12 (NELS 88).(a) Science

15

20

25

30

8 10 12

Grade

Scie

nce

Scor

e


Figure D7a

Source: Raudenbush (2006)

Source: Raudenbush (2006)



Figure D7b. Average achievement trajectories, grades 8–12, (NELS 88)MathAverage achievement trajectories, Grades 8-12 (NELS 88).

(b) Math

30

35

40

45

50

55

60

8 10 12

Grade

Mat

h Sc

ore


Figure D7b




Figure D8a. Growth as a function of student social background: ECLSReading

Growth as a function of student social background: ECLS

(a) Reading

10

20

30

40

50

60

70

55 65 75 85

Age (months)

Rea

ding

Sco

re

Hi SESLow SES

Figure D8a




Figure D8b. Growth as a function of student social background: ECLSMath

Growth as a function of student social background: ECLS

(b) Math

10

20

30

40

50

60

70

55 65 75 85

Age (months)

Mat

h Sc

ore

Hi SESLow SES

Figure D8b




Figure D9a. Growth as a function of school poverty for poor children:sustaining effects dataReading

Growth as a Function of School Poverty for Poor Children: Sustaining Effects Data

(a) Reading

350

400

450

500

550

70 75 80 85 90 95

Age (months)

Rea

ding

Sco

re


Figure D9a




Figure D9b. Growth as a function of school poverty for poor children:sustaining effects dataMath

Growth as a Function of School Poverty for Poor Children: Sustaining Effects Data

(b) Math

350

400

450

500

550

70 75 80 85 90 95

Age (months)

Mat

h Sc

ore

Low PovertyHigh poverty

Figure D9b




Schooling quality and school resources have relatively smalleffects on ability deficits and only marginally account for anydivergence by age across children from different socioeconomicgroups in test scores.

See Heckman, Larenas et al. (2004) and Raudenbush (2006).



Gaps also emerge in health. These appear to be divergentwith age, at least in the U.S.



Health and income for children and adults, U.S. National Health Interview Survey

1986-1995. From Case, A., Lubotsky, D. & Paxson, C. (2002), American

Economic Review, Vol. 92, 1308-1334.

)emocniylimaf(nl110198

5.1

57.1

2

52.2

3-0sega

8-4sega

21-9sega

71-31sega

heal

thsta

tus(

1 =

exce

llent

to5

= po

or)



Second, in both animal and human species, there is compellingevidence of critical and sensitive periods in the development ofthe child.



Second language learning

200

220

240

260

280

Native 3-7 8-19 11-17 17 -22 23-28 31-39

Second Language Learning�� 6



The later remediation is given to a disadvantaged child, the lesseffective it is.

A substantial body of evidence suggests that returns toadolescent education for the most disadvantaged and less ableare lower than the returns for the more advantaged.



The economic returns to adolescent intervensions — jobtraining, high school graduation, and college attendance — arelower for less able persons.



Table 1. Return to one year of college for individuals at different percentilesof the math test score distributionWhite males from high school and beyond

Table 2.4Return to one year of college for individuals

at different percentiles of the math test score distributionWhite males from High School and Beyond

5% 25% 50% 75% 95%Average return in the population 0.1121 0.1374 0.1606 0.1831 0.2101

(0.0400) (0.0328) (0.0357) (0.0458) (0.0622)Return for those who attend college 0.1640 0.1893 0.2125 0.2350 0.2621

(0.0503) (0.0582) (0.0676) (0.0801) (0.0962)Return for those who do not attend college 0.0702 0.0954 0.1187 0.1411 0.1682

(0.0536) (0.0385) (0.0298) (0.0305) (0.0425)Return for those at the margin 0.1203 0.1456 0.1689 0.1913 0.2184

(0.0364) (0.0300) (0.0345) (0.0453) (0.0631)Wages are measured in 1991 by dividing annual earnings by hours worked per weekmultiplied by 52. The math test score is and average of two 10th grade math test scores.There are no dropouts in the sample and the schooling variable is binary (high school - college).The gross returns to college are divided by 3.5 (average difference in years of schoolingbetween high school graduates that go to college and high school graduates that do not in asample of white males in the NLSY). To construct the numbers in the table we proceed in twosteps. First we compute the marginal treatment effect using the method of local instrumentalvariables as in Carneiro, Heckman and Vytlacil (2001). The parameters in the table aredifferent weighted averages of the marginal treatment effect. Therefore, in the second stepwe compute the appropriate weight for each parameter and use it to construct a weightedaverage of the m arginal treatment effect ( see also Carneiro, 2002). Individuals at t he marginare indifferent between attending college or not.

Source: Carneiro and Heckman (2003)



Third, despite the low returns to interventions targeted towarddisadvantaged adolescents, the empirical literature shows higheconomic returns for remedial investments in youngdisadvantaged children.



Fourth, if early investment in disadvantaged children is notfollowed up by later investment, its effect tends to weaken atlater ages.

Currie and Thomas (1995) document a decline in theperformance of minority Head Start participants after theyleave the program.



Fifth, the effects of credit constraints on the child’s adultoutcomes depend on the age at which they bind for the child’sfamily.

Controlling for cognitive ability, under meritocratic policiescurrently in place in American society, family income during thechild’s college-going years plays only a minor role indetermining child college participation.

Holding ability fixed, minorities are more likely to attend collegethan others despite their lower family incomes.



Carneiro and Heckman present evidence for the United Statesthat only a small fraction (at most 8%) of the families ofadolescents are credit constrained in making their collegedecisions.

This evidence is supported in research by Cameron and Taber(2004) and Stinebrickner and Stinebrickner (2006).



There is some evidence that credit constraints operating in theearly years have effects on adult ability and schooling outcomes,but there is not full agreement in the literature on themagnitude of the effect (Duncan and Brooks-Gunn, 1997; Dahland Lochner, 2004; Duncan and Ariel Kalil, 2006; Carneiro andHeckman, 2003).



The empirically important market failures in the life cycle ofskill formation in contemporary American society are theinability of children to buy their parents or the lifetimeresources that parents provide.

It is not credit constraints facing families seeking to secureloans for a child’s education when the child is an adolescent.



Sixth, socioemotional (noncognitive) skills foster cognitive skillsand are an important product of successful families andsuccessful interventions in disadvantaged families.

The Perry Preschool Program, which was evaluated by randomassignment, did not boost participant adult IQ but enhancedperformance of participants in a number of dimensions,including elevated scores on achievement tests, employmentand reduced participation in a variety of social pathologies.



Figure D10a. Perry Preschool Program: IQ, by age and treatment group

79.6

95.5 94.9

91.3 91.7

88.1 87.7

85

78.5

83.3 83.5

86.3 87.1 86.9 86.884.6

75

80

85

90

95

100

IQ

4 5 6 7 8 9 10EntryAge

Treatment Group Control Group

Source: Perry Preschool Program. IQ measured on the Stanford Binet Intelligence Scale (Terman & Merrill, 1960).Test was administered at program entry and each of the ages indicated.

Perry Preschool Program: IQ, by Age and Treatment GroupFigure 2a

16by James J. Heckman and Flavio Cunha. Published in the American Economic Review (2007). Volume 97, Issue 2, pp.31–47. [4mm]James J. Heckman University of ChicagoExtract for “The Technology of Skill Formation” 66 / 76


Figure D10b. Perry Preschool Program: educational effects, by treatmentgroup

45%

66%

15%

49%

34%

15%

0% 10% 20% 30% 40% 50% 60% 70%

On Time Grad.from HS

High Achievementat Age 14*

SpecialEducation

Source: Barnett (2004).Notes: *High achievement defined as performance at or above the lowest 10th percentile on the California AchievementTest (1970).

Perry Preschool Program: Educational Effects, by Treatment GroupFigure 7B

Treatment Control



Figure D10c. Perry Preschool Program: economic effects at age 27, bytreatment group

14%

29%

13%

36%

7%

29%

0% 10% 20% 30% 40%

Never on Welfareas Adult*

Own Home

Earn +$2,000Monthly

Source: Barnett (2004). *Updated through Age 40 using recent Perry Preschool Program data, derived from self reportand all available state records.

Perry Preschool Program: Economic Effects at Age 27, by Treatment GroupFigure 7C

Treatment Control



Figure D10d. Perry Preschool Program: arrests per person before age 40,by treatment group

1.2 3.9 .4

2.1 6.7 .6

0 2 4 6 8 10

Treatment

Control

Source: Perry Preschool Program. Juvenile arrests are defined as arrests prior to age 19.

Perry Preschool Program: Arrests per Person before Age 40, by Treatment GroupFigure 7D

Felony Misdemeanor Juvenile



Table D1. Economic benefits and costs

Perry Chicago CPC

Child Care 986 1,916

Earnings 40,537 32,099

K-12 9,184 5,634

College/Adult -782 -644

Crime 94,065 15,329

Welfare 355 546

FG Earnings 6,181 4,894

Abuse/Neglect 0 344

Total Benefits 150,525 60,117

Total Costs 16,514 7,738

Net Present Value 134,011 52,380

Benefits-To-Costs Ratio 9.11 7.77

Table D1 Economic Benefits And Costs

Notes: All values discounted at 3% and are in $2004. Numbers differ

slightly from earlier estimates because FG Earnings for Perry and

Chicago were estimated using the ratio of FG Earnings Effect to

Earnings Effect (about 15%) that was found in Abecedarian

Source: Barnett, 2004.Source: Barnett (2004)



A Model of Skill Formation

Agents possess a vector of abilities at each age.

These abilities (or skills) are multiple in nature and range frompure cognitive abilities (e.g. IQ) to noncognitive abilities(patience, self control, temperament, risk aversion, timepreference).

These abilities are used with different weights in different tasksin the labor market and in social life more generally.



The human skill formation process is governed by a multistagetechnology.

Each stage corresponds to a period in the life cycle of a child.

Like earlier work by Ben-Porath (1967), we use a productionfunction to determine the relationship between inputs and theoutput of skill.

Unlike Ben-Porath, in our models qualitatively different inputscan be used at different stages and the technologies can bedifferent at different stages of child development.



Ben-Porath focuses on adult investments where time and itsopportunity cost play important roles.

For child investments, parents make decisions and childopportunity costs are less relevant.



The outputs at each stage in our technology are the levels ofeach skill achieved at that stage.

Some stages of the lifecycle may be more productive inproducing some skills than other stages, and some inputs maybe more productive at some stages than at other stages.

Those stages that are more effective in producing certain skillsare called “sensitive periods” for the acquisition of those skills.

If one stage alone is effective in producing a skill (or ability), itis called a “critical period” for that skill.



The skills produced at one stage augment the skills attained atlater stages. This effect is termed self-productivity.

Skills are self-reinforcing and cross-fertilizing.

A second key feature of skill formation is dynamiccomplementarity.

Skills produced at one stage raise the productivity ofinvestment at subsequent stages. Complementarity implies thatlevels of skill investments at different ages bolster each other.They are synergistic.



Together, dynamic complementarity and self-productivityproduce multiplier effects which are the mechanisms throughwhich skills beget skills and abilities beget abilities.

Dynamic complementarity, self-productivity of human capital,and multiplier effects imply an equity-efficiency trade-off forlate child investments but not for early investments.

These concepts, embedded in alternative market settings,explain the six facts previously listed.


Documents

The Technology of Skill Formation - Home | HCEO · Extract for \The Technology of Skill Formation" by James J. Heckman and Flavio Cunha. Published in the American Economic Review