Skills, Schools and Synapses · in skills and to long term trends in the demand for labor that favor skilled workers and and have raised their wages both absolutely and relative to

Skills, Schools and Synapses

James HeckmanUniversity of Chicago

Albert A. Michelson Memorial LectureU.S. Naval AcademyAnnapolis, MarylandSeptember 11, 2006

1 / 1

Outline Gaps in Test Scores Explanation Abilities and Outcomes Early Interventions Neuroscience Summary

Skills are an essential ingredient of the success of moderneconomies.

A skilled workforce is a source of productivity growth,innovation and adaptability.

It can better respond to increased uncertainty intechnology and in trade patterns that is characteristic ofour age.

2 / 111





2 / 111





2 / 111


Inequality in the labor market is largely due to inequalityin skills and to long term trends in the demand for laborthat favor skilled workers and and have raised their wagesboth absolutely and relative to unskilled workers becausethey have become more scarce.

The wage premia for skill (college vs. high school andhigh school vs. high school dropout) have increased.

3 / 111


Inequality in the labor market is largely due to inequalityin skills and to long term trends in the demand for laborthat favor skilled workers and and have raised their wagesboth absolutely and relative to unskilled workers becausethey have become more scarce.

The wage premia for skill (college vs. high school andhigh school vs. high school dropout) have increased.

3 / 111


March CPS full-time weekly earnings, 1963–2005.3

5.4

.45

.5.5

5C

OLL

EG

E-G

RA

D/H

SG

all

1960 1970 1980 1990 2000Earnings year

1963-2005.March CPS Full-Time Weekly Earnings.

4 / 111


Percent change in the relative weekly wages of high school dropouts to highschool graduates, CPS March 1963–2003

Figure X. Percentage Change in the Relative Weekly Wages of HS Dropouts to HS

Graduates, CPS March 1963-2003

-0.6-1.4

-5.1

-3.7 -4.0

-14.8-16.0

-14.0

-12.0

-10.0

-8.0

-6.0

-4.0

-2.0

0.0

2.0

4.0

1963-1971 1971-1979 1979-1987 1987-1995 1995-2003 1963-2003Year Range

Perc

enta

ge C

hang

e

Source: Autor, Katz and Kearney (2005) "Trends in U.S. Wage Inequality: Re-Assessing the Revisionists". Based on full time full year workers.

5 / 111


Percent change in the relative weekly wages of high school dropouts tocollege graduates, CPS March 1963–2003Figure X.

Percentage Change in the Relative Weekly Wages of HS Dropouts to College Graduates, CPS March 1963-2003

-8.4

6.1

-15.2-13.0

-7.6

-38.2

-50.0

-40.0

-30.0

-20.0

-10.0

0.0

10.0

1963-1971 1971-1979 1979-1987 1987-1995 1995-2003 1963-2003Year Range

Perc

enta

ge C

hang

e

Source: Autor, Katz and Kearney (2005) "Trends in U.S. Wage Inequality: Re-Assessing the Revisionists". Based on full time full year workers.

6 / 111


Skilled workers are much less likely to participate in crimeor in other social pathologies. Consider the case of crime.

7 / 111


Regression-adjusted probability of incarceration, by years of schooling(whites)

Regression-Adjusted Probability of

Incarceration, by Years of Schooling

Source: Lochner and Moretti (2004)

8 / 111


Regression-adjusted probability of incarceration, by years of schooling(blacks)

Source: Lochner and Moretti (2004)

9 / 111


Men Incarcerated (2000) overall and by educational status

Whites Hispanics Blacks

All men, age 20–40 1.6% 4.6% 11.5%

High school dropouts, age 20–40 6.7% 6.0% 32.4%

Source: Western, Bruce (2006). Punishment and Inequality inAmerica. New York: Russell Sage Foundation.

10 / 111


On many dimensions, participation in productive andpathological behaviors is closely linked to educationalattainment.

Therefore, it is a source of great concern that at a timewhen the demand for skill has increased, the supply ofskill has slowed down.

11 / 111


On many dimensions, participation in productive andpathological behaviors is closely linked to educationalattainment.

Therefore, it is a source of great concern that at a timewhen the demand for skill has increased, the supply ofskill has slowed down.

11 / 111


Educational characteristics of the labor force aged 25 and over, 1980, 2000,2020

�

��

�� !�� !"�!��##�"$� �!%�

�

��

�Figure�$��(��!�13 ��

��Table 20

��

��

'�!!$�,$��-��

��

.��'�!!$�,$��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

�� !��

��/��$��(��!� ��

��$��(��!�#�!*� ��

(�)�(��!��$� *��+�

��$��(��!� ��

'�!!$�,$��-��

��

.��'�!!$�,$��

Educational Characteristics of the Labor ForceAged 25 and Over1980, 2000, 2020

*Assumes that subsequent cohorts have same education at age 25 as the cohort age 25 in 2000.

Source: Ellwood (2001).

Table 2

12 / 111


Characteristics of the labor force aged 25 and over and components ofchange, 1980, 2000, 2020 (millions of workers)

Table ��19��

�� !�� !��

��

�

� ��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��"��#$ �� %#&� �� %��

� � � � �

��

!�"��

��

#��

��$��

��$��

'())�"*��

��"��

��"��%�#��

��"��&�'�!��

��"��&�'�!��%�#��

��$�� "��

��

�

��

�

��

��

��

�

��

�

Characteristics of the Labor Force Aged 25 and Overand Components of Change 1980, 2000, 2020

(Millions of Workers)

��

Source: Ellwood (2001).

Table 2

13 / 111



��

� � � � �

��

��

��

��

��

��

� ��

��

��

��

��

��

��

��

��

��

��

��

14 / 111



��

��

��

��

��

��

��

�

��

�

��

��

��

�

��

�

��

� ��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

15 / 111


Percent distribution of education among 30 year olds by year��

�

��

��

��

��

�

!�

"�

#�

$�

��

�$ �$!� �$! �$"� �$" �$#� �$# �$$� �$$ ��

�

��

%�&�'�&��(��

%�&�'�&��)� ��

'��*��'�&��

+��(�� ,��

��

Figure 3

(Ellwood, 2001)

16 / 111


Postsecondary participation rates for adults aged 18-24, by combinations ofsex and race/ethnicity, 1974-2003

17 / 111


College enrollment rates of recent high school completers, by sex35

4045

5055

6065

70C

olle

ge P

artic

ipat

ion

Rat

e (%

)

1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 Year

All Males Females

Source: Authors calculations based on three year moving averages of NCES unpublished sources and, CPS, October 1970 to 2003 data. Table 183Note: Recent high school completers are those who graduated or obtained a GED in the previous 12 months

College Enrollment Rates of Recent High School Completers, by Sex

18 / 111


College enrollment rates of recent high school completers, by race35

4045

5055

6065

70C

olle

ge P

artic

ipat

ion

Rat

e (%

)

1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 Year

All Whites Blacks Hispanics

Source: Authors calculations based on three year moving averages of NCES unpublished sources and, CPS, October 1970 to 2003 data. Table 183Note: Recent high school completers are those who graduated or obtained a GED in the previous 12 months

College Enrollment Rates of Recent High School Completers, by Race

19 / 111


College-going among graduates is increasing.

But when you count GEDs as dropouts, as you should,the dropout is rate increasing.

20 / 111


College-going among graduates is increasing.

But when you count GEDs as dropouts, as you should,the dropout is rate increasing.

20 / 111


The fraction of high school dropouts in the USA, 1971–2000: Twoalternative measures

10.0

15.0

20.0

25.0

30.0

35.0

1971

1972

1973

1974

1975

1976

1977

1978

1979

1980

1981

1982

1983

1984

1985

1986

1987

1988

1989

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

100 - the fraction of graduates as a percent of 17-year-old population

High school dropouts excluding the GEDs - CPS data

Source:

(1) the Department of Education National Center for Education Statistics and (2) American Council on Education, General Educational Development Testing

Service

and (3) U.S. Department of Commerce, Bureau of the Census, Current Population Survey, unpublished tabulations; and U.S. Department of Education, National

Center for Education Statistics, Dropout Rates in the United States.

21 / 111


Cumulative official high school dropout rate by race-ethnicity: persons 14to 20 years old, 1973–1997

22 / 111


The official rate on high school dropouts disguises asubstantial problem.

Many American children taking GEDs.

GEDs are dropouts who pass a test to certify that theyare just as good as high school graduates.

GEDs earn the wages of dropouts.

They do not complete college.

When we adjust the high school dropout rate for GEDs,the high school dropout rate is rising.

23 / 111








23 / 111








23 / 111








23 / 111








23 / 111








23 / 111


GED credentials issued as a percentage of public and private high schoolgraduates, USA, 1960–2001

GED Credentials Issued as a Percentage of Public and Private High School Graduates, USA 1960-2001

0.0

5.0

10.0

15.0

20.0

25.0

1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000

Year

Per

cen

t

Source: (1) The Department of Education National Center for Education Statistics Digest of Educational Statistics, 2001, and (2) American Council on Education, General Educational

Development Testing Service Statistical Report 1989 and 2002

24 / 111


The high rate of dropout leads to a group of people withvery low skills.

Low compared to other advanced nations.

Another way to gauge this is by looking at adult literacyin the U.S. compared to other countries.

25 / 111





25 / 111





25 / 111


Percentage of each gender who perform at Level 1 on the IALS DocumentLiteracy Scale

��

��

�

��

��

��

��

��

��

��

��

��

!��"�#/��;��;��<��=�>��>��=��;��==�;<��?��/�@�>��=<�;��;�K�@�>�� ;��X��K�#/��;��<��/��X��Z�Z��=��/��=��/��<�>��[\\ =��/��?�\\Z�=��/��]?��\\ �\\��=��^K��;��;<��<��/��/��;��X��_�@��[��^K

��;��=��;/��`/��=��@�>��/��_�@��j�;<��@��;��;��

��<��14

26 / 111


To an economist, the sluggish response of the supply ofskills is an enigma — when the rate of return hasincreased, the supply response should lead to many moreeducated workers, not less.

27 / 111


The increase in education is concentrated among themost affluent American families.

This trend threatens to perpetuate and enhanceinequality across income groups and race groups.

Threatens workforce productivity because more of ourfuture workers are coming from disadvantaged families.

28 / 111





28 / 111





28 / 111


The most common explanation for gaps in collegeattendance is tuition costs.

Tuition has been rising.

To many, this suggests that we should subsidize tuition.

29 / 111





29 / 111





29 / 111


Cameron and Heckman (1998) show that tuition explainslittle of the gap in college going between the affluent andless affluent, between rich and poor, between majoritiesand minorities.

Controlling for ability at the age college decisions aremade, minorities are more likely to enroll in college.

Tuition costs cannot explain the dramatic gaps in highschool dropping out.

30 / 111





30 / 111





30 / 111


Gaps in test scores are substantial between the affluentand the poor; between majority group members andminority group members.

This suggests that perhaps the schools are to blame.

31 / 111


Gaps in test scores are substantial between the affluentand the poor; between majority group members andminority group members.

This suggests that perhaps the schools are to blame.

31 / 111


Gaps in ranks or achievement across race groups andacross income groups open up early, before schools canhave any effect.

Absolute levels of test scores grow. Percentage ranks arestable.

Schools add knowledge.

But gaps open up as early as they are measured.

In a statistical sense, gaps can be closed by controlling forfamily traits.

Consistent with Coleman Report (1966).

32 / 111








32 / 111








32 / 111








32 / 111








32 / 111








32 / 111


Average percentile rank on PIAT-Math score, by income quartileAverage Percentile Rank on PIAT-Math Score, by Income Quartile

35

40

45

50

55

60

65

6 8 10 12

Age

Scor

e p

erce

nti

le

Lowest income quartile Second income quartile Third income quartile Highest income quartile

33 / 111


Average percentile rank on PIAT-Math score, by race

Average Percentile Rank on PIAT-Math Score, by Race

30

35

40

45

50

55

60

65

6 8 10 12

Age

Scor

e p

erce

nti

le

Hispanic Black White

34 / 111


Average percentile rank on anti-social score, by income quartile (whites)H. Average Percentile Rank on Anti-Social Score, by Income Quartile*

Whites Only

25

30

35

40

45

50

55

60

65

4 6 8 10 12

Age

Scor

e p

erce

nti

le


35 / 111


Average percentile rank on anti-social score, by race

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

30

35

40

45

50

55

4 6 8 10 12

Age

Scor

e p

erce

nti

le


36 / 111


Average percentile rank on anti-social score, by income quartile

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

20

25

30

35

40

45

50

55

4 6 8 10 12

Age

Scor

e p

erce

nti

le


37 / 111


Controlling for early family factors eliminates a lot of thegaps in ranks.

38 / 111


Residualized average PIAT-Math score percentiles, by income quartile

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

35

40

45

50

55

60

65

6 8 10 12

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

home at each age

Scor

e p

erce

nti

le


39 / 111


Residualized average PIAT-Math score percentiles, by race

F. Residualized Average PIATM Score Percentile by Race*

30

35

40

45

50

55

60

65

6 8 10 12


home at each age

Scor

e p

erce

nti

le


40 / 111


Residualized average anti-social score percentile, by income quartile

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

20

25

30

35

40

45

50

55

60

4 6 8 10 12


home at each age

Scor

e p

erce

nti

le


41 / 111


Residualized average anti-social score percentile, by race

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

30

35

40

45

50

55

4 6 8 10 12


home at each age

Scor

e p

erce

nti

le


42 / 111


Family factors matter.

A source of concern because, as we have seen, a greaterfraction of American children are being born todisadvantaged families.

While the situation is improving, there is still a majorproblem with child environments.

43 / 111





43 / 111





43 / 111


Trends in unhealthy child environmentsFigure 17

Trends in Unhealthy Child Environments

0

5

10

15

20

25

30

35

40

45

50

1940

1942

1944

1946

1948

1950

1952

1954

1956

1958

1960

1962

1964

1966

1968

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

Year

% of All Births to Unmarried Women 15-44 Birth rate for 1000 Unmarried Women 15-44Birth Rate per 1000 Unmarried Women 15-19 % of All Children Under 18 Living W/ 1 Parent% of All Children Under 18 Living in Poverty

Data for births and birth rates are from Ventura and Bachrach (2000). Data for children living with one parent are available at the census bureau at http://www.census.gov/population/socdemo/hh-fam/tabCH-1.txt. Data for children living in poverty is available at www.childtrendsdatabank.org/

44 / 111


Abilities Are Multiple

To understand the sources of difference in schooling, andparticipation in a variety of social pathologies, and eventhe development of animal species, we need to appreciatethat there are multiple abilities.

45 / 111


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

NCLB focuses on achievement on a test score at a certainage to measure success or failure in schools.

Yet an emerging body of evidence shows that, as isintuitively obvious and commonsensical, much more thansmarts are required.

Motivation,Attention,Self Esteem,Time Preference.

46 / 111






46 / 111






46 / 111





Motivation,

Attention,Self Esteem,Time Preference.

46 / 111





Motivation,Attention,

Self Esteem,Time Preference.

46 / 111





Motivation,Attention,Self Esteem,

Time Preference.

46 / 111






46 / 111


Their importance tends to be underrated in current policydiscussions because they are not easily measured.

Evidence from the GED program (Heckman andRubinstein, 2001).

47 / 111


Their importance tends to be underrated in current policydiscussions because they are not easily measured.

Evidence from the GED program (Heckman andRubinstein, 2001).

47 / 111


The GED program is a second chance program given tosecondary school dropouts.

We have already seen that participation in the GEDprogram is growing.

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.

48 / 111







48 / 111







48 / 111







48 / 111







48 / 111


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

GED recipients and high school graduates with twelve years of schooling

Figure 2.13Density of age adjusted AFQT scores,

(c) Black males

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

(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

(e) Hispanic 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

(f) Hispanic 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

(d) Black females

0

5

10

15

20

25

30

35

40

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

Source: Heckman, Hsee and Rubinstein (2001).

Source: Heckman, Hsee and Rubinstein (2001)

49 / 111





(c) Black males

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

(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

(e) Hispanic 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


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

(d) Black females

0

5

10

15

20

25

30

35

40

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



50 / 111





(c) Black males

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

(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

(e) Hispanic 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


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

(d) Black females

0

5

10

15

20

25

30

35

40

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



51 / 111


Yet GEDs earn at the rate of high school dropouts.

52 / 111


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 military recognize this in theirrecruiting strategies.

53 / 111






53 / 111






53 / 111






53 / 111


There is a lot of evidence that both cognitive andnoncognitive skills are important.

Both cognitive and noncognitive skills matter in a varietyof aspects of life.

54 / 111


There is a lot of evidence that both cognitive andnoncognitive skills are important.

Both cognitive and noncognitive skills matter in a varietyof aspects of life.

54 / 111


Cognitive and noncognitive ability are importantdeterminants of schooling and socioeconomic success.

In the U.S., schooling gaps have more to do with abilitydeficits than family finances in the school-going years.

Those with higher abilities of both types are more likelyto take post-school company job training.

Look at effects of both cognitive and noncognitive skillson many measures of social performance.

55 / 111






55 / 111






55 / 111






55 / 111


Probability of being a 4-year college graduate, by ability

0

.2

.4

.6

Pro

bab

ility

0 20 21 40 41 60 61 80 81 100Percentile

Males Cognitive Males Non Cognitive Females Cognitive Females Non Cognitive

Note: This figure plots the probability of a given behavior associated with moving up in one ability distribution forsomeone after integrating out the other distribution. For example, the lines with markers show the effect ofincreasing noncognitive ability after integrating the cognitive ability. Source: Heckman, Stixrud, and Urzua (2006).

Probability of Being a 4 yr College Graduate, by Ability

56 / 111


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

0

.05

.1

.15

Pro

bab

ility

0 20 21 40 41 60 61 80 81 100Percentile

Cognitive Non Cognitive


Ever in Jail by Age 30, by Ability Males

57 / 111


Probability of being single with children (females)

.02

.04

.06

.08

.1

Pro

bab

ility

0 20 21 40 41 60 61 80 81 100Percentile

Cognitive Non Cognitive


Probabilty of Being Single with Children Females

58 / 111


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

59 / 111


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

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

ilit

y

1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8


Decile

Pro

bab

ilit

y a

nd

Co

nfid

en

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

60 / 111


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

61 / 111


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

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

ilit

y

1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8


Decile

Pro

bab

ilit

y a

nd

Co

nfid

en

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

62 / 111


Probability of incarceration by age 30 (males)

24

68

10

24

68

10

0

0.2

0.4

0.6

0.8

1


Figure 1E. Probability of Incarceration by Age 30 - 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

63 / 111


Probability of incarceration by age 30 (males) (cont.)

24

68

10

24

68

10

0

0.2

0.4

0.6

0.8

1


Figure 1E. Probability of Incarceration by Age 30 - Malesi. By Decile of Cognitive and Noncognitive Factor

Decile of Cognitive

Pro

bab

ilit

y

1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8


Decile

Pro

bab

ilit

y a

nd

Co

nfid

en

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

64 / 111


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

65 / 111


Probability of daily smoking by age 18 (males) (cont.)

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

ilit

y

1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8


Decile

Pro

bab

ilit

y a

nd

Co

nfid

en

ce I

nte

rval (2

.5-9

7.5

%)


1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8


Decile

66 / 111


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

67 / 111


Mean log wages by age 30 (males) (cont.)

2 4 6 8 102

2.5

3i. By Decile of Cognitive Factor

Decile

Lo

g W

ages

and

Con

fid

en

ce In

terv

al (

2.7

5-9

7.5

%)

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

68 / 111


How do these early differences in abilities arise?

Is the difference due to genetics?

Herrnstein and Murray claimed so in The Bell Curve.

They used an achievement test score measurement at 14to show that genes were important.

They also implicitly claimed that compensation for earlydeficits was not possible.

This score has been shown to be affected by schoolingand family environments.

But possibly, with better measures, they are right.

69 / 111









69 / 111









69 / 111









69 / 111









69 / 111









69 / 111









69 / 111


Evidence on epigenetics suggests that the genes vs.environment distinction is obsolete.

A large body of recent work suggests thatgene-environmental interactions are central to explaininghuman (and animal) development.MOA gene expression is modified by the environment.Suomi: short allele / long allele species are affecteddifferently by the environment.Turkheimer (2003) on behavioral genetics.

70 / 111



A large body of recent work suggests thatgene-environmental interactions are central to explaininghuman (and animal) development.

MOA gene expression is modified by the environment.Suomi: short allele / long allele species are affecteddifferently by the environment.Turkheimer (2003) on behavioral genetics.

70 / 111



A large body of recent work suggests thatgene-environmental interactions are central to explaininghuman (and animal) development.MOA gene expression is modified by the environment.

Suomi: short allele / long allele species are affecteddifferently by the environment.Turkheimer (2003) on behavioral genetics.

70 / 111



A large body of recent work suggests thatgene-environmental interactions are central to explaininghuman (and animal) development.MOA gene expression is modified by the environment.Suomi: short allele / long allele species are affecteddifferently by the environment.

Turkheimer (2003) on behavioral genetics.

70 / 111



A large body of recent work suggests thatgene-environmental interactions are central to explaininghuman (and animal) development.MOA gene expression is modified by the environment.Suomi: short allele / long allele species are affecteddifferently by the environment.Turkheimer (2003) on behavioral genetics.

70 / 111


Evidence that shows that family backgrounds eliminatetest score gaps is correlational in nature.

Correlational evidence is always suspect.

We cannot know true causal effect without some furtherassumption because of some possible unmeasured thirdcommon cause.

There is evidence from experimental interventions thatavoids some of the interpretive problems with thecorrelational evidence.

71 / 111






71 / 111






71 / 111






71 / 111


The most reliable data come from experiments thatprovided substantial enrichment of the early environmentsof children living in low-income families.

Two of these investigations, the Perry Preschool Programand the Abecedarian Program, are the most informativefor the purposes of this discussion because they employeda random assignment design and collected long-termfollow-up data.

72 / 111


The most reliable data come from experiments thatprovided substantial enrichment of the early environmentsof children living in low-income families.

Two of these investigations, the Perry Preschool Programand the Abecedarian Program, are the most informativefor the purposes of this discussion because they employeda random assignment design and collected long-termfollow-up data.

72 / 111


These longitudinal studies demonstrated substantial,positive effects of early environmental enrichment on arange of cognitive and “non-cognitive” skills, schoolingachievement, job performance, and social behaviors, longafter the intervention ended.

Data from non-controlled assessments of Head Start andthe Chicago Child-Parent Centers programs suggestsimilar conclusions. The data from Head Start representonly short-term effects.

73 / 111


These longitudinal studies demonstrated substantial,positive effects of early environmental enrichment on arange of cognitive and “non-cognitive” skills, schoolingachievement, job performance, and social behaviors, longafter the intervention ended.

Data from non-controlled assessments of Head Start andthe Chicago Child-Parent Centers programs suggestsimilar conclusions. The data from Head Start representonly short-term effects.

73 / 111


Early Intervention Programs for Disadvantaged Children

The Perry Program was an intensive preschool programthat was administered to 58 disadvantaged, black childrenin Ypsilanti, Michigan between 1962 and 1967.

The treatment consisted of a daily 2.5 hour classroomsession on weekday mornings and a weekly 90 minutehome visit by the teacher on weekday afternoons. Thelength of each preschool year was 30 weeks.

The control and treatment groups have been followedthrough age 40.

74 / 111






74 / 111






74 / 111


The Abecedarian Program involved 111 disadvantagedchildren, born between 1972 and 1977, whose familiesscored high on a risk index. The mean age at entry was4.4 months.

The program was a year-round, full-day intervention thatcontinued through age 8. The children were followed upuntil age 21, and the project is ongoing.

75 / 111


The Abecedarian Program involved 111 disadvantagedchildren, born between 1972 and 1977, whose familiesscored high on a risk index. The mean age at entry was4.4 months.

The program was a year-round, full-day intervention thatcontinued through age 8. The children were followed upuntil age 21, and the project is ongoing.

75 / 111


In both the Perry and Abecedarian Programs there was aconsistent pattern of successful outcomes for treatmentgroup members compared with control group members.

For the Perry Program, an initial increase in IQdisappeared gradually over 4 years following theintervention, as has been observed in other studies.

76 / 111


In both the Perry and Abecedarian Programs there was aconsistent pattern of successful outcomes for treatmentgroup members compared with control group members.

For the Perry Program, an initial increase in IQdisappeared gradually over 4 years following theintervention, as has been observed in other studies.

76 / 111


However, in the more intensive Abecedarian Program,which intervened earlier (starting at age 4 months) andlasted longer (until age 8), the gain in IQ persisted intoadulthood (21 years old).

This early and persistent increase in IQ is importantbecause IQ is a strong predictor of socioeconomic success.

Positive effects of these interventions were alsodocumented for a wide range of social behaviors.

At the oldest ages tested (Perry: 40 yrs; Abecedarian: 21yrs), individuals scored higher on achievement tests,reached higher levels of education, required less specialeducation, earned higher wages, were more likely to own ahome, and were less likely to go on welfare or beincarcerated than individuals from the control groups.

77 / 111






77 / 111






77 / 111






77 / 111


Perry preschool program: educational effects, by treatment group

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 Group

Figure 7B

Treatment Control

78 / 111


Perry preschool program: economic effects at age 27, by treatment 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 Group

Figure 7C

Treatment Control

79 / 111


Perry preschool program: arrests per person before age 40, by treatmentgroup

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 Group

Figure 7D

Felony Misdemeanor Juvenile

80 / 111


Perry did not raise IQ.

It raised noncognitive skills.

81 / 111


Perry did not raise IQ.

It raised noncognitive skills.

81 / 111


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

86.387.1 86.9 86.8

84.6

75

80

85

90

95

100

IQ

4 5 6 7 8 9 10Entry

Age

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 Group

Figure 7A

82 / 111


Other similar programs

83 / 111


Academic and social benefits at school exit for CPC participants6

84 / 111


Abecedarian academic outcomes 8D

85 / 111


Other benefits of Abecedarian8E

86 / 111


Intervening at an early enough age can actually raise IQof the participants.

87 / 111


Abecedarian program: IQ, by age and treatment group

98.4 97.8

94.795.8

89.7

93.893.3

88.65

90.1

85.2

80

85

90

95

100

IQ

6.5 8 12 15 21

Age

Treatment Group Control Group

Source: Barnett (2004).

Abecedarian Program: IQ, by Age and Treatment Group

Figure 8A

88 / 111


Many studies have shown that these aspects of behaviortranslate directly or indirectly into high economic return.An estimated rate of return (the return per dollar of cost)to the Perry Program is in excess of 17%.

This high rate of return is much higher than standardreturns on a stock market equity and suggests thatsociety at large can benefit substantially from these kindsof interventions.

89 / 111


Many studies have shown that these aspects of behaviortranslate directly or indirectly into high economic return.An estimated rate of return (the return per dollar of cost)to the Perry Program is in excess of 17%.

This high rate of return is much higher than standardreturns on a stock market equity and suggests thatsociety at large can benefit substantially from these kindsof interventions.

89 / 111


The evidence from these intervention studies shows thefollowing.

First, skills beget skills.

That is, all capabilities are built on a foundation ofcapacities that are developed earlier.

90 / 111





90 / 111





90 / 111


This principle stems from two characteristics that areintrinsic to the nature of learning:

Early learning confers value on acquired skills, whichleads to self-reinforcing motivation to learn more;Early mastery of a range of cognitive, social, andemotional competencies makes learning at later agesmore efficient and therefore easier and more likely tocontinue.

91 / 111



Early learning confers value on acquired skills, whichleads to self-reinforcing motivation to learn more;

Early mastery of a range of cognitive, social, andemotional competencies makes learning at later agesmore efficient and therefore easier and more likely tocontinue.

91 / 111



Early learning confers value on acquired skills, whichleads to self-reinforcing motivation to learn more;Early mastery of a range of cognitive, social, andemotional competencies makes learning at later agesmore efficient and therefore easier and more likely tocontinue.

91 / 111


Technology of skill formation

St+1 = f (St , It)↑ ↑ ↑

Skill attime t + 1

Skill attime t

Investment

92 / 111


Two polar cases

“Leontief cases” and perfect substitution.

St+1 = min (St , It)

A polar opposite is the case of perfect substitutes

St+1 = a St + b It

Leads to ideas of critical and sensitive periods.

Language in humans is an example of sensitive periods.

93 / 111


Two polar cases




St+1 = a St + b It



93 / 111


Two polar cases




St+1 = a St + b It



93 / 111


Two polar cases




St+1 = a St + b It



93 / 111


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

94 / 111


Legend

Sensitive period for second language acquisition. Englishlanguage proficiency scores as a function of age of arrival inthe United States for a group of Chinese and Korean adultimmigrants (n = 46). All subjects were students or faculty atthe University of Illinois and had been in the U.S. for at least10 years prior to testing. The test measured a variety ofgrammatic judgements. From Johnson and Newport (1989).

95 / 111


Romanian adoption study supports this

Within-UK Adoptees Romanian Orphans

Age of Adoption (Months): 6 Before 6 Age 6-24 Age 24-42

Weight at Adoption - -2.1 -2.3 -- (1.7) (1.7) -

Height at Adoption - -1.8 -2.2 -- (1.6) (2.4) -

Denver Developmental - 76.5 48.1 -Scale at Adoption - (48.1) (25.4) -

See Rutter et al. (1998) and O’Connor et al. (2000) for more details on the analysis.

96 / 111


Romanian adoption study supports this (cont.)



Weight at Age 4 0.45 -0.02 0.04 -(0.79) (0.92) (0.94) -

Height at Age 4 0.25 -0.29 -0.36 -(0.91) (0.89) (1.02) -

Denver Developmental 117.7 115.7 96.7 -Scale at Age 4 (24.3) (23.4) (21.3) -

McCarthy GCI at 109.4 105.9 91.7 -Age 4 (14.8) (17.9) (18.0) -


97 / 111


Romanian adoption study supports this (cont.)



Weight at Age 6 0.30 0.02 -0.25 -0.85(0.90) (0.97) (0.96) (0.98)

Percentage with Denver 2 0 5 18Developmental Scale at (1) (0) (2) (7)Age 6 Below 70

McCarthy GCI at 117 114 99 90Age 6 (17.8) (18.3) (19.2) (23.8)


98 / 111


Similar results found in animal studies

Experimental evidence on macaque monkey populationsupports this conclusion.

Draw on the work of Judy Cameron, Oregon NationalPrimate Research Center.

99 / 111


Similar results found in animal studies

Experimental evidence on macaque monkey populationsupports this conclusion.

Draw on the work of Judy Cameron, Oregon NationalPrimate Research Center.

99 / 111


Studies of Monkeys

1 monthsurrogate mother provided at:

2 monthnone

0

10

20

30

40

50

0 2 4 6 8 10 12 14

1.5 month

�� 5

100 / 111


A model of critical and sensitive periods

Consider a simple model–two periods of childhood (takenfrom Cunha and Heckman, 2006).

Human Capital S2 in Period 2:

S2 =[γI φ

1 + (1− γ) I φ2

] 1φ

I1 is early investment; I2 is late investment.

101 / 111





S2 =[γI φ

1 + (1− γ) I φ2

] 1φ


101 / 111





S2 =[γI φ

1 + (1− γ) I φ2

] 1φ


101 / 111


The ratio of early to late investment in human capital as a function of theskill multiplier for different values of complementarity

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.90

0.5

1

1.5

2

2.5

3

3.5

4

Figure 10

Leontief= - 0.5

CobbDouglas= 0.5

Skill Multiplier ( )

The Ratio of Early to Late Investment in Human Capital as a

Function of the Skill Multiplier for Different Values of Complementarity

102 / 111


Legend

This figure shows the optimal ratio of early to late investments, 1

2

as a function of the skill multiplierparameter for di erent values of the complementarity parameter assuming that the interest rate is zero.The optimal ratio 1

2

is the solution of the parental problem of maximizing the present value of the child’s wealththrough investments in human capital, and transfers of risk-free bonds, In order to do that, parents have todecide how to allocate a total of dollars into early and late investments in human capital, 1 and 2 respectively,and risk-free bonds. Let denote the present value as of period “3” of the future prices of one e ciency unit ofhuman capital: =

P=3 (1+ ) 3 The parents solve

max

μ1

1 +

¶2[ + ]

subject to the budget constraint

1 +2

(1 + )+(1 + )2

=

and the technology of skill formation:

=h

1 + (1 ) 2

i

for 0 1 0 1 and 1 From the first-order conditions it follows that 1

2

=h(1 )(1+ )

i 1

1

This

ratio is plotted in this figure when (Leontief), = 0 5 = 0 (Cobb-Douglas) and = 0 5 and forvalues of the skill multiplier between 0 1 and 0 9

103 / 111


Recent work by Cunha and Heckman estimates this technologyand shows the relative effectiveness of early and lateinterventions.

104 / 111


Baseline

Changing Initial Conditions -

Moving Children to the 4th

Adolescent Intervention:

Moving Investments at Last

Transition from 1st to 9th

Decile

High School Graduation 0.4109 0.6579 0.6391

Enrollment in College 0.0448 0.1264 0.1165

Conviction 0.2276 0.1710 0.1773

Probation 0.2152 0.1487 0.1562

Welfare 0.1767 0.0905 0.0968

Disadvantaged Children: First Decile in the Distribution of Cognitive and Non-Cognitive Skills at Age 6

Comparison of Different Investment Strategies

Mothers are in First Decile in the Distribution of Cognitive and Non-Cognitive Skills at Ages 14-21

Decile of Distribution of Skills

only through Early Investment

105 / 111


Baseline

Changing Initial

Conditions - Moving

Children to the 4th

Decile of Distribution of

Skills

Changing Initial

Conditions and

Performing Adolescent

Intervention

Changing Initial

Conditions and

Performing a Balanced

Intervention

High School Graduation 0.4109 0.6579 0.8477 0.9135

Enrollment in College 0.0448 0.1264 0.2724 0.3755

Conviction 0.2276 0.1710 0.1272 0.1083

Probation 0.2152 0.1487 0.1009 0.0815

Welfare 0.1767 0.0905 0.0415 0.0259

Disadvantaged Children: First Decile in the Distribution of Cognitive and Non-Cognitive Skills at Age 6

Comparison of Different Investment Strategies

Mothers are in First Decile in the Distribution of Cognitive and Non-Cognitive Skills at Ages 14-21

106 / 111


Using the estimated technology to simulate balanced interventions

2 4 6 8 10 12 14 16 1878

80

82

84

86

88

90

92

94

96

Figure 18

Using the Estimated Technology to Simulate Balanced Interventions

Age

Simulated Perry Contro l

Simulated Perry Treatment

Simulated Perry Plus Follow Up

Sco

re i

n C

og

nit

ive S

kills

107 / 111


Using the estimated technology to simulate Perry preschool with follow up

2 4 6 8 10 12 14 16 1875

80

85

90

95

100

105

110

Figure 19

Using the Estimated Technology to

Simulate Perry Preschool with Follow Up

Age

Simulated Perry ControlSimulated Perry TreatmentSimulated Lifetime Perry Treatment

Cognit

ive

Sco

re

108 / 111


The evidence strongly supports early initial investmentthat is sustained.

Optimal distribution of investment:

Invest early? Yes.But must be followed up to be effective.

This is a consequence of complementarity.

109 / 111






109 / 111




Invest early? Yes.

But must be followed up to be effective.


109 / 111






109 / 111






109 / 111


Summary

Skills matter.

America has a skills problem. Rising inequality is a signalof this problem.

More than smarts is required.

NCLB and a lot of American social policy focus only onsmarts.

Skill gaps emerge early and can be traced in part toadverse early environments.

110 / 111


Summary

Skills matter.





110 / 111


Summary

Skills matter.





110 / 111


Summary

Skills matter.





110 / 111


Summary

Skills matter.





110 / 111


Summary

Schools and tuition do not matter as much as is oftenthought.

Late remediation not very effective.

Remediation can work, but is costly.

Social policy should be directed toward the malleableearly years.

Evidence from human and animal species supports thisconclusion.

111 / 111


Summary






111 / 111


Summary






111 / 111


Summary






111 / 111


Summary






111 / 111

Documents

Skills, Schools and Synapses · in skills and to long term trends in the demand for labor that favor skilled workers and and have raised their wages both absolutely and relative to