• The Productivity Argument for Investing in Young Children

By James J. Heckman and Dimitriy V. Masterov
Review of Agricultural Economics—Volume 29, Number 3—Pages 446–493

This article presents the case for investing more in young American children
who grow up in disadvantaged environments. Figure 1 graphs time series of
alternative measures of the percentage of children in disadvantaged families.
The percentage of children born into, or living in, nontraditional families has
increased greatly in the past thirty years.1,2 Approximately 25% of children are
now born into single parent homes. While the percentages of children living in
poverty and born into poor families have fallen recently, they are still high,
especially among certain subgroups.

Adverse environments place children at risk for social and economic failure.
The accident of birth plays a powerful role in determining adult success.3 Many
have commented on this phenomenon, and most analyses have cast the issue of
assisting children from disadvantaged families as a question of fairness or social
justice.

This article makes a different argument. We argue that, on productivity
grounds, it makes sense to invest in young children from disadvantaged
environments. Substantial evidence shows that these children are more likely to
commit crime, have out-of-wedlock births, and drop out of school. Early
interventions that partially remediate the effects of adverse environments can
reverse some of the harm of disadvantage and have a high economic return.
They benefit not only the children themselves, but also their children, as well as
society at large.

Investing in disadvantaged young children is a rare public policy with no
equity-efficiency tradeoff. It reduces the inequality associated with the accident
of birth and at the same time raises the productivity of society at large.

.
James J. Heckman is the Henry B. Schultz Distinguished Service Professor in the
Department of Economics, University of Chicago.
.
Dimitriy V. Masterov is a graduate student in the Department of Economics,
University of Michigan.
*This lecture was given as the T.W. Schultz Award Lecture at the Allied Social Sciences Association
annual meeting, Chicago, January 5–7, 2007.

This article was not subject to the journal’s standard refereeing process.

Proceedings 447

Figure 1. Percentage of all children born or living in adverse environments
in each year, 1968–2000

Living in a single-Parent Home Living in Poverty Born into Single-Parent Home Born into Poverty
While a more rigorous analysis is necessary to obtain a better understanding
of the effects of early intervention programs, their precise channels of influence,
and their exact benefits and costs, the existing evidence is promising. An
accumulating body of knowledge shows that early childhood interventions for
disadvantaged young children are more effective than interventions that come
later in life. Because of the dynamic nature of the skill formation process,
remediating the effects of early disadvantages at later ages is often prohibitively
costly (Carneiro, Cunha, and Heckman; Cunha and Heckman, 2006). Skill begets
skill; learning begets learning. Early disadvantage, if left untreated, leads to
academic and social difficulties in later years. Advantages accumulate; so do
disadvantages. A large body of evidence shows that postschool remediation
programs like public job training and general educational development (GED)
certification cannot compensate for a childhood of neglect for most people.

This evidence has dramatic consequences for the way we think about policy
toward skill formation. Most current policies directed toward improving the
skills of youth focus on schools as the locus of intervention. The No Child Left
Behind Act uses mandates and punishments to encourage schools to remediate
the educational deficits of disadvantaged children. School accountability
schemes are used to motivate higher levels of achievement for children from
disadvantaged environments.

While these initiatives are well-intentioned, their premise is faulty. Schools
work with what parents give them. The 1966 Coleman Report on inequality in
school achievement clearly documented that the major factor explaining the
variation in the academic performance of children across U.S. schools is the
variation in parental environments—not the variation in per pupil expenditure

Review of Agricultural Economics

across schools or pupil–teacher ratios. Successful schools build on the efforts of
successful families. Failed schools deal in large part with children from
dysfunctional families that do not provide the enriched home environments
enjoyed by middle class and upper middle class children. Since failure in school
is linked to so many social pathologies, each with substantial social and
economic costs, a policy of equality of opportunity in access to home
environments (or their substitutes) is also one that promotes productivity in
schools, the workplace, and society at large.

Rigorous statistical analysis is not needed to show that parents and their
resources matter, although there is a large body of empirical evidence that
supports this claim, as we document below. The issue that has stymied social
policy is how to compensate for adverse family environments in the early years.
One approach has been to reduce the material deprivation suffered by the poor
with transfers from the state, as in Lyndon Johnson’s War on Poverty. Another
approach has been to bolster the family with programs outside the home.
Sometimes, children have been removed from the biological families, as in the
case of the American Indians in the early twentieth century. Policies that have
removed children from homes have had catastrophic consequences.4

An emerging body of evidence suggests that there is a better way to improve
the early years of disadvantaged children. Enriched preschool centers available
to disadvantaged children on a voluntary basis coupled with home visitation
programs have a strong track record of promoting achievement for
disadvantaged children. The economic return to these programs is high,
especially when we consider alternative policies that target children from
disadvantaged environments or the policies targeted to the young adults who
emerge from them. We review the evidence on these programs and suggest that
some version of them be used to supplement the resources of disadvantaged
families with children.

Our logic is simple and compelling. Education and human skill are major
factors determining productivity, both in the workplace and in society. The
family is a major producer of the skills and motivation required for producing
successful students and workers. The most effective policy for improving the
performance of schools is supplementing the childrearing resources of the
disadvantaged families sending children to the schools. The family is a major
determinant of child participation in crime and social deviance. A family
supplementation policy is a successful anticrime policy.

Our emphasis on early childhood interventions does not deny the importance
of schools or firms in producing human skill. Indeed, if proven early
intervention programs are adopted, schools will be more effective, firms will
have better workers to employ and train, and the prison population will decline.
At lower cost to society, bolstered families will produce better educated
students, more trained workers, and better citizens.

This article proceeds in the following way. We first discuss the problem of the
supply of skills to the American economy. Growth in both the quantity and the
quality of the labor force traditionally has been a major source of U.S. output
growth. Given current trends, U.S. growth prospects are poor. Labor force
growth is slowing, especially that of young and skilled workers who are a

Proceedings 449

source of vitality for the entire economy. The composition of the future
workforce will shift toward workers from relatively more dysfunctional families
with commensurately worse skills.

We next discuss the problem of crime in America. Even though the crime rate
has fallen in recent years, the levels and costs of crime are still very high. The
damage to victims and the resources spent on preventing crime and on
incarcerating criminals are large. Early intervention programs targeted toward
disadvantaged families reduce participation in crime. On purely economic
grounds, the case for early childhood intervention is strong.

After describing these two major social problems that impair the productivity
of American society, we summarize trends in adverse child environments. We
summarize a vast literature in social science that establishes that dysfunctional
and disadvantaged families are major producers of cognitive and behavioral
deficits that lead to adverse teenage and adult social and economic outcomes.
The effects of disadvantage appear early and persist. Remediating these
disadvantages at later ages is costly. Human abilities affect lifetime performance
and are shaped early in the life of the child. Early interventions promote
cumulative improvements. Enriched interventions targeted toward children in
disadvantaged environments are cost-effective remedies for reducing crime and
the factors that breed crime, and raising productivity in schools and in the
workplace.

We then move on to summarize the findings of the literature on the economics
of child development that demonstrates the importance of both cognitive and
noncognitive abilities in shaping child educational and economic outcomes.
Both types of abilities are major determinants of the economic return to
education.

Both cognitive and noncognitive abilities are shaped early in life and early
differences in abilities persist. Gaps in college attendance across socioeconomic
groups are largely shaped by abilities formed in the early years. Gaps in child
ability across families of different income levels are associated with parental
environments and parenting practices. Early interventions can partially
remediate these deficits. Later interventions are much less effective. At current
levels of investment, American society over-invests in public job training and
formal education and under-invests in early education for disadvantaged
children.

We summarize the evidence from a variety of early intervention programs
targeted toward disadvantaged children and focus on three early interventions
that followed participants into adulthood. Some of these interventions are
evaluated by the method of random assignment. Early interventions reduce
crime, promote high school graduation and college attendance, reduce grade
repetition and special education costs, and help prevent teenage births. They
raise achievement as measured by test scores. Very early interventions also
appear to raise IQ, especially for girls. Cost–benefit analyses of these programs
reported in the literature show that they are cost-effective. Estimated rates of
return are 16%: 4% for participants and 12% for society at large. The article
concludes with a summary of the argument and some specific policy
recommendations.

Review of Agricultural Economics

Human Capital and Economic Performance

Education and skill are central to the performance of a modern economy. The
emergence of new technologies has raised the demand for highly skilled
workers who are qualified to use them. A wage premium for skilled labor
emerged in many countries in the early 1980s, and wage inequality grew as the
economic return to education (the economic benefit of attending school) rose,
especially in countries like the U.S. where the supply response to the increasing
wage premium was weak.5 Not only did the wages of the skilled rise, but those
with the least ability and education earn less today than comparable workers
would have earned 30 years ago.

Workforce Trends

Table 1, taken from Ellwood, highlights the problems facing the American
labor market in the next two decades. The first column of the table presents the
distribution of the American workforce among age and race-ethnicity categories
in 1980. The second column shows the growth in the categories from 1980 to
2000 and the third column shows the labor force as of 2000. The fourth column
shows the projected growth in the labor force in the next twenty years by

Table 1. Characteristics of the labor force aged twenty-five and over
and components of change: 1980, 2000, and 2020 (millions of
workers)

Labor Growth Labor Growth Labor
Force 1980– Force 2000– Force
Age in 1980 2000 in 2000 2020 in 2020
25–54 65.0 35.1 100.1 3.0 103.1
55–64 11.8 2.2 14.0 12.5 26.5
65+ 3.0 1.4 4.4 4.0 8.4
Total 79.8 38.7 118.5 19.4 137.9
Race/ethnicity/nativity
White non-Hispanic–native 63.0 21.5 84.5 2.6 87.1
Black non-Hispanic–native 7.6 4.6 12.2 2.8 15.0
Hispanic–native 2.5 2.3 4.8 6.8 11.6
Other non-Hispanic–native 0.8 1.0 1.8 1.2 3.0
Hispanic–foreign born 1.8 4.5 6.3 2.8 9.1
Non-Hispanic–foreign born 4.1 4.8 8.9 3.3 12.2
Total 79.8 38.7 118.5 19.4 137.9
Summary
Native white workers 25–54 50.8 19.3 70.1 -7.7 62.4
Native white workers 55 and over 12.2 2.2 14.4 10.3 24.7
Workers of color 25–54 9.4 7.3 16.7 7.7 24.4
Workers of color 55 and over 1.6 0.5 2.1 3.0 5.1
Foreign born workers 5.9 9.4 15.3 6.0 21.3
Total 79.8 38.7 118.5 19.4 137.9

Source: Ellwood (2001).

Proceedings 451

Figure 2. Percent distribution of education among thirty-year-olds
by year

(Ellwood, 2001)

category. With the possible exception of the numbers for immigrants, these are
reliable projections because there is little emigration and the groups being
projected are already alive. The immigration projections come from a carefully
executed U.S. Census study. The labor force is aging and young replacements
for old workers are increasingly in short supply compared to the 1980s.6 The
aging of the American workforce raises serious problems for the future of
American productivity growth.

The workforce of prime-age workers, fueled by the entry of Baby Boomers,
propelled U.S. economic growth in 1980–2000. However, we cannot count on
this source of growth in the next twenty years. Indeed, the largest components of
growth in the workforce will come from older workers as the Baby Boom cohort
ages. Hence, a major source of vitality in the U.S. workforce will be lost. Future
workforce growth will come from older workers and from demographic groups
in which, for a variety of reasons, dysfunctional and disadvantaged families are
more prevalent (see the middle rows of table 1 and the discussion below).

On top of these trends in the number of workers by age, there is stagnation in
educational attendance rates. Figure 2 shows the distribution of educational
attainment among thirty-year-olds by year. College-going rates have stalled out
for cohorts of Americans born after 1950. This is not a consequence of
immigration of unskilled workers. It is a phenomenon found among native-born
Americans. Currently, 17% of all new high school credentials issued are GEDs.7
Heckman and LaFontaine (2006) document that the high school dropout rate
has increased over time if one counts GEDs as dropouts, as one should, because
GEDs earn the same wages as dropouts, and graduate from college at the same
rate as dropouts.

The growth in the quality of the workforce, which was a mainstay of
economic growth until recently, has diminished. Assuming that these trends
continue, the U.S. economy will add many fewer educated persons to the

Review of Agricultural Economics

Table 2. Educational characteristics of the labor force aged
twenty-five and over: 1980, 2000, and 2020

Labor Growth Labor Growth Labor
Force 1980– Force in 2000– Force
in 1980 2000 in 2000 2020 in 2020

Education
Less than high school 17.3 -5.3 12.00.9 12.9
High school only 31.56.3 37.83.8 41.6
Some schooling beyond high school 13.8 19.1 32.96.2 39.1
College degree or more 17.3 18.5 35.87.7 43.5

Total 79.8 38.7 118.5 18.6 137.1
% with college degree 21.6% 30.2% 31.7%

Note: Assumes that subsequent cohorts have same education at age 25 as the cohort age 25 in 2000.
Source: Ellwood (2001).

workforce in the next two decades than it did in the past two decades (table 2).
Jorgenson, Ho, and Stiroh estimate that the average annual rate of growth of
college labor supply was 4.5% in 1977, but fell to 1.75% in 1990–2000. These
trends are predicted to continue, or possibly worsen.

The slowdown in labor force quality growth has already hurt American
productivity growth. Delong, Katz, and Goldin estimate that increases in
educational attainment boosted the effective quality of the workforce by 0.5% a
year over the years 1915–2000, and thus contributed an average of 0.35
percentage points per year to economic growth over that period.8 The slower
growth in educational attainment of the workforce substantially reduced
productivity growth in recent years compared to its performance in the period
1915–1980. Based on current trends, these authors project that the annual rate of
productivity growth attributable to education—0.35 from 1980–2000—will
decline by half or more (to between 0.17 and 0.06%) in the next two decades.
This will reduce the productivity growth of labor by a substantial 0.18–0.29
percentage points per year and will be a drag on real wage growth and fiscal
revenues.

Literacy and Numeracy

The skills of the U.S. labor force are poor. The U.S. has a thick lower tail of
essentially illiterate and innumerate persons, who are a drag on productivity
and a source of social and economic problems. We use data from the
International Adult Literacy Survey (IALS) to examine literacy and numeracy of
working age adults (ages sixteen to sixty-five).9 Document literacy is defined as
the ability to locate and use information from timetables, graphs, charts, and
forms. Figure 3 presents data on document literacy. Tests for prose literacy and
quantitative literacy produce the same pattern.10

Level 1 performance is essentially functional illiteracy or innumeracy. It
represents the inability to determine the correct amount of medicine from

Proceedings 453

Figure 3. Percentage of each gender who perform at Level 1 on the
IALS document literacy scale

Note: The scale scores were grouped into five levels of increasing difficulty, with Level 1 representing
functional illiteracy. Levels 4 and 5 were combined. The sample is restricted to adults who are between
16–65 years of age at the time of the survey (1994 for the US and Germany, 1996 for the UK, and 1994–
1995 for Sweden). Standard errors are calculated using the methodology described in IALS (2002).

information on a bottle of pills. People who perform at Level 1 can make limited
use of texts that are simple and uncomplicated. They are only able to locate
information in text or data as long as there is no distracting information around
the correct answer. On the quantitative scale, they can only carry out relatively
straightforward operations such as simple addition. Roughly 20% of U.S.
workers fall into this category on each test, a much higher fraction than in some
of the leading European countries. This is a major drag on U.S. competitiveness11
and a source of social problems.

Crime

Crime is a major burden on American society. Anderson (1999) estimates that
the net cost of crime (after netting out transfers) is over $1.3 trillion per year in
2004 dollars. The per capita cost is $4,818, in the same dollars. This figure
includes crime-induced production (of personal protection devices, trafficking
of drugs and operation of correctional facilities), which costs $464 billion per
year; opportunity costs (production foregone by incarcerated offenders, valued
at their estimated wage, time spent locking and installing locks, and so forth) of
$152 billion per year; and the value of risks to life and health (pain, suffering,
and mental distress associated with health losses) of $672 billion annually (table
3). This includes time lost from work by victims as well as value of life lost to
murders. Some of these items like the valuation of life require controversial
judgments. Even ignoring any transfer component or any risks to life and
health, the cost of crime is over $600 billion per year. Although such calculations
are necessarily imprecise and there is disagreement over the exact costs, there is
widespread agreement that the costs of crime are substantial.

Review of Agricultural Economics

Table 3. Aggregate burden of crime

Crime-induced production ($ billion) 464
Opportunity costs ($ billion) 152
Risks to life and health ($ billion) 672
Transfers ($ billion) 706
Gross burden ($ billion) 1,995
Net of transfers ($ billion) 1,289
Per capita ($) 4,818

Source: Anderson (1999). All figures inflated to 2004 dollars using the CPI.

Even though crime rates have recently declined somewhat, their levels remain
high (figure 4(a)). The adult correctional populations (in prison or local jail, on
probation, or on parole) continue to grow despite the drop in measured crime
rates (figure 4(b)). The size of the population under correctional supervision has
continued to increase for all groups, as has the percentage of each group under
supervision. Nine percent of blacks were under the supervision of the criminal
justice system in some form in 1997, although recently this adverse trend has
slowed. Incarceration rates have risen steadily since 1980 and only slowed in the
late 1990s. The inmate population has risen steadily until recently.12 Expenditures
on prisons, police, and the judicial system continue to grow despite the
drop in measured crime rates (figure 4(c)).

These statistics do not convey the full scope of the problem. According to the
Bureau of Justice Statistics (2004), as of the end of 2001, there were an estimated

5.6 million adults who had ever served time in state or federal prison—4.3
million former prisoners and 1.3 million adults in prison. Nearly a third of
former prisoners were still under correctional supervision, including 731,000 on
parole, 437,000 on probation, and 166,000 in local jails. In 2001, an estimated
2.7% of adults in the U.S. had served time in prison, up from 1.8% in 1991 and
1.3% in 1974. The prevalence of imprisonment in 2001 was higher for black
males (16.6%) and Hispanic males (7.7%) than for white males (2.6%). It was also
higher for black females (1.7%) and Hispanic females (0.7%) than white females
(0.3%). Nearly two-thirds of the 3.8 million increase in the number of adults ever
incarcerated between 1974 and 2001 occurred as a result of an increase in first
incarceration rates; one-third occurred as a result of an increase in the number of
residents age eighteen and older. If recent incarceration rates remain unchanged,
an estimated one of every 15 persons (6.6%) will serve time in a prison during
his or her lifetime.
The lifetime chances of a person going to prison are higher for men (11.3%)
than women (1.8%), and for blacks (18.6%) and Hispanics (10%) than whites
(3.4%). Based on current rates of first incarceration, an estimated 32% of black
males will enter state or federal prison during their lifetime, compared to 17% of
Hispanic males and 5.9% of white males.

What Can We Do about This Problem?

It is now well established that education reduces crime. Figure 5, from
Lochner and Moretti, displays this relationship, reported separately for blacks

Proceedings 455

Figure 4. (a) Reported violent and property rates, 1991–2001; (b)
Adult correctional population, 1980–2002; and (c) Total direct expenditures
by criminal justice function, 1982–2001

(a)
6,000
4,000
2,000
0
Note: The murder and nonnegligent homicides that occurred as a result of the events of September
11, 2001 are not included.

(b)
Note: Probation is court ordered community supervision. Prison consists of confinement in a state
or federal correctional facility for more than one year or longer. Jail is confinement in a local facility
while pending trial, awaiting sentencing, serving a sentence less than one year, or awaiting transfer
to another facility after conviction. Parole is community supervision after a period of incarceration.
Data from BJS Justice Expenditure and Employment Extracts.

Continued

and whites. Increasing high school graduation rates is a major crime prevention
strategy. Risk factors promoting crime include poor family backgrounds, which
also promote high school drop out. Poorly educated teenage mothers in
low-income families are much more likely to produce children who participate
in crime.13 We discuss the evidence on the impact of family background on child

Review of Agricultural Economics

Figure 4. Continued

(c)
Source: Justice Expenditure and Employment Extracts

participation in crime in the next section. Although analysts do not agree on
which specific aspects of adverse family environments most affect crime, they
all agree that there is a strong empirical relationship between early adverse
environments and child participation in crime later on in life.

Some of the most convincing estimates of the impact of adverse early
environments on participation in crime come from interventions designed to
remedy those environments. Table 4 presents a summary of the impacts of a
variety of early childhood intervention programs on participation in crime. We
discuss some of these programs in much greater detail below. Here, we
summarize some findings relevant to crime.

Many of these programs were evaluated by the method of random
assignment. Children from disadvantaged populations were randomly
assigned, at early ages, to the enriched child development programs described
in the third column of the table. Most interventions were for children in the
prekindergarten years. Both the experimental treatment group and the controls
were followed over time, often for many years after the intervention. The Perry
Preschool program, which we discuss in detail below, followed the intervention
and control children for more than 30 years after the intervention. Over that
time, the Perry students averaged significantly fewer lifetime arrests than the
comparison group, including arrests for dealing and producing drugs. This
effect was especially pronounced for males. The Abecedarian program appears
to be anomalous, because it did not reduce crime in the treatment group
compared to the control group. It was administered to a population in a
low-crime region in the South. Most studies show dramatic reductions in
criminality and participation in the criminal justice system for treatment group
members. Enriched environments reduce crime. Impoverished environments
promote crime.

Proceedings 457

Figure 5. Regression-adjusted probability of incarceration, by years
of schooling

WHITES

0.018

0.016

0.014

0.012

0.010

0.008

0.006

0.004

0.002
0
2 4 6 8 10 12141618
Years of Schooling

BLACKS

0.07

0.06
0.05
0.04
0.03
0.02
0.01
0

–0.01
Years of Schooling

2 4 6 8 10 12 14 16 18
Source: Lochner and Morelli (2004)

Probability of Imprisonment Probability of Imprisonment

Review of Agricultural Economics

Table 4. Effects of early intervention programs

Program/Study Cost* Program Description Predelinquency Crime
Abecedarian project** N/A Full-time year round No effect
(Ramey et al. 1988) classes for children
from infancy through
Houston PCDC** N/A
preschool
Home visits for parents Rated less aggressive
(Johnson 1988) for two yeas; child and hostile by
nursery care four mothers (ages 8–11)
days per week in year
two (Mexican
Americans)
Perry Preschool
program **
$19,162 Weekly home visits
with parents;
2.3 versus 4.6 lifetime
arrests by age 27; 7%
(Schweinhart, Barnes, intensive, high versus 35% arrested 5
and Weikart 1993) quality preschool or more times
services for one to
two years
Syracuse University $54,483 Weekly home visits for 6% versus 22% had
family development family; day care year probation files;
(Lally, Mangione, and round offenses were less
Honig 1988) severe
Yale experiment $33,319 Family support; home Rated less aggressive
visits and day care as and predelinquent by
needed for thirty
months
teachers and parents
(age 121/2)

Note: All comparisons are for program participants versus nonparticipants. *Costs valued in 2004
dollars. **Studies used a random assignment experimental design to determine program impacts.
Data from Donohue and Siegelman (1998), Schweinhart, Barnes, and Weikart (1993), and Seitz (1990)
for the impacts reported here. N/A indicates not available.
Source: Heckman et al. (1997).

Lochner and Moretti present convincing nonexperimental evidence that
increasing educational attainment reduces crime and that the inverse
relationship between crime and education in figure 5 is not a correlational
artifact arising from unobserved variables that are common to both crime and
education. Using Census data, they show that one more year of schooling
reduces the probability of incarceration by 0.37 percentage points for blacks, and

0.1 for whites.14 To put this evidence in perspective, 23% of the black–white
difference in average incarceration rates can be explained by the differences in
education between these groups. Using the FBI’s Uniform Crime Reports, they
find that the greatest impacts of education are associated with reducing arrests
for murder, assault, and motor vehicle theft.
Lochner and Moretti also calculate the social savings from crime reduction
associated with completing secondary education. They show that a 1% increase
in the high school graduation rate would yield $1.8 billion dollars in social
benefits in 2004 dollars. This increase would reduce the number of crimes by

Proceedings 459

Table 5. Estimated social benefits of increasing high school
completion rates by 1%

Estimated Change
in Crime Social Benefits

Violent crimes
Murder -373 $1,457,179,565
Rape 1,559 -$179,450,969
Robbery 918 -$11,116,176
Assault -37,135 $475,045,373

Property crimes
Burglary -9,467 $12,052,009
Larceny/theft -35,105 $8,958,962
Motor vehicle theft -14,238 $22,869,192
Arson -469 $23,637,635

Total -94,310 $1,809,175,590

Notes: Victim costs and property losses taken from table 2 of Miller, Cohen, and Wiersema (1996).
Incarceration costs per crime equal the incarceration cost per inmate, $17,027 (U.S. Department of
Justice 1999), multiplied by the incarceration rate (U.S. Department of Justice 1994). Total costs are
calculated as the sum of victim costs and incarceration costs less 80% of the property loss (already
included in victim costs) for all crimes except arson. Total costs for arson are the sum of victim costs
and incarceration costs, since there is no transfer of property between victim and criminal. Estimated
changes in crimes adjust the arrest effect by the number of crimes per arrest. The social benefit is the
estimated change in crimes times the total cost per crime. All dollar figures are adjusted to 2004 dollars
using the CPI.
Source: Lochner and Moretti (2004).

more than 94,000 each year (table 5). The social benefits include reduced losses
in productivity and wages, lower medical costs, and smaller quality-of-life
reductions stemming from crime.15 They also include reductions in costs of
incarceration.16

High school graduation confers an extra benefitof14–26% beyond private
returns captured by the high school graduate wages that are pocketed by
graduates. This is an important benefit of education beyond its private return
that suggests overall under-investment in the population of disadvantaged
children at risk for committing crime. Completing high school raises a student’s
wages by about $10,372 per year (in 2004 dollars), and the direct cost of
completing one year of secondary school is approximately $8,000 per student in
1997 (in 2004 dollars). Looking only at the savings from reduced crime, the
return is $1,638–$2,967 per year, so that expenditure is cost effective even if we
ignore the direct benefits in earnings and even if we assume that the benefits
decline as the youths grow older.

Comparing the effect of educational expenditure with the effect of hiring an
additional police officer suggests that promoting education may be a better
strategy. Using a somewhat different framework, Levitt claims that an
additional sworn police officer in a large U.S. city would reduce annual costs
from crime by about $200,000 dollars at a public cost of $80,000 per year. These
are recurrent annual costs.

Review of Agricultural Economics

Lochner and Moretti estimate that in steady state it would cost $15,000 per
year in terms of direct costs to produce enough high school graduates to reduce
crime by the same amount. This cost ignores foregone earnings in high school
but it also ignores all of the benefits from high school graduation documented in
Heckman, Lochner, and Todd. If Levitt’s estimate is correct, educational policy is
far more effective per dollar spent than expenditure on police.17,18

Trends in Children’s Home Environments and the
Consequences of Adverse Environments

Demographers have documented that over the past forty years, the aggregate
birth rate has declined, but in the past few decades relatively more of all
American children born are born into adverse environments. The definition of
adversity varies among studies, but the measures used are strongly interrelated.
Most scholars recognize that absence of a father, low levels of financial
resources, low parental education and ability, a lack of cognitive and emotional
stimulation, and poor parenting skills are characteristics of adverse
environments. Determining the relative importance of these factors is an
ongoing debate. Each seems to play a factor in affecting child outcomes.

Family Structure

Fewer children are living with two parents who are married. In 2003, 68% of
children under age eighteen lived with two married parents, down from 77% in
1980.19 This percentage has remained stable since 1995, after trending
downward for many years. The percentage of children who live with only one
parent, or in a home where the parents are not married, increased by 8% since
1980 to 28%. The percentage of children who live with no parents remained
roughly constant around 3–4% during this period. The source of single
parenthood has also changed. Relatively more children are living with a single
parent who has never been married (figure 6(a)).

The aggregate trends conceal a great deal of variation across demographic
groups. In 2003, 77% of non-Hispanic white children lived with two married
parents, while 20% lived with only one parent or with unmarried parents. The
corresponding percentages for blacks were 36% and 56%. For Hispanics, it was
65% and 31%.20 Among blacks, the percentage of children living with a
never-married parent has increased dramatically over time.21

Nonmarital Childbearing

Since the 1965 Moynihan Report, many analysts have focused on family
structure—the absence of a parent and the attendant decline in financial,
emotional, and cognitive resources—as an important source of social
problems.22 Over time, while the birth rate has fallen, births to unmarried
women have risen until very recently.

After rising dramatically since 1940, out-of-wedlock childbearing leveled off
in the 1990s but remains at a very high level.23 The number of births to
unmarried women increased from 1.17 to 1.3 million between 1990 and 1999. The

Proceedings 461

Figure 6. (a) Percent of all children living with one parent by marital
status of single parent; (b) Percent of children in single mother homes
by education of the mother; (c) Percent of women with children who
had never been married by education of mother; and (d) Births to
unmarried women under age nineteen as a percentage of total births
in a given year by race

(b)
45%

40%
35%
30%
25%
20%
15%
10%
5%

loohcShgiHnahTsseLylnOdarGloohcShgiHSome Post-Secondary
eroMroeergeDsrolehcaB
0%
1962 1967 1972 1977 1982 1987 1992 1997 2 002
Source: Jencks and Ellwood (2004), using March Current Population Survey.

Continued

Review of Agricultural Economics

Figure 6. Continued

(c)
40%
35%
30%
25%
20%
15%
10%

Source: Jencks and Ellwood (2004), using March Current Population Survey.

%0%51nahtsseL2 sraey12 sraey13–15 years
sraeyeromro61
1961967 197 1977 1982 1987 1992 1997 2 0

22 02

(d)
Whites (Inc. Hispanics), under 19 All Others, under 19 All Races, under 19
birthrate for unmarried women increased from 43.8 births per 1,000 unmarried
women aged fifteen to forty-four years in 1990 to 46.9 in 1994, before falling back
to 43.9 in 1999.24 The percentage of all births to unmarried women has risen
from 28% in 1990 to 33% in 1999, though it has been roughly constant at 32–33%
since 1994. To put these numbers in perspective, in 1940, this number was 3.8%.

The birth rate for unmarried black women has been higher than that of white
unmarried women (including Hispanic women), but this gap has narrowed in

Proceedings 463

recent years because the birth rate has grown at a faster pace for unmarried
white women.25 In 1970, the rate for unmarried black women was roughly seven
times the rate for unmarried white women—96 per 1,000 versus 14 per 1,000. By
1998, the gap was reduced by 70%; it became 73 versus 38 per 1,000.

Unfortunately, the birthrate for unmarried Hispanic women is only available
for the 1990s, but it is the highest among the three demographic groups. In 1990,
the birthrate for unmarried Hispanic women was 89.6 per 1,000, peaked at 101.2
per 1,000, and fell to 90.1 per 1,000 in 1998.26

The same trend holds for the percentage of births to unmarried mothers
within each race.27 In 1969, 5.5% of white children were born to unmarried
mothers. The corresponding percentage for blacks was 34.9%. By 1999, these
numbers were 26.7% and 68.8%, respectively. The percentage for Hispanics in
1999 was 42.1% versus 36.7% in 1990. Until recently, unmarried births have been
increasing overall, although the percentage due to minority mothers has
stabilized.28

Single parenthood is much more prevalent for high school dropouts (see
figure 6(b) and the discussion in Ellwood and Jencks 2002). Although the media
has focused on celebrities who choose single parenthood, the bulk of single
mothers have high school education or less and the majority of this group
consists of high school dropouts (figure 6(c)). The incidence of divorce is greater
for this group as well.29 The percentage of children born to unmarried teenagers
has trended up dramatically over the past fifty years. Close to 10% of all
children were born to unmarried teenage mothers in 2000 (figure 6(d)).

Many pathologies are associated with less educated mothers and teenage
mothers. They are less likely to marry when they have children and they are
more likely to divorce. Their abilities (see Armor 2003), family incomes, and the
emotional and intellectual support accorded children are low. Figures 7(a) and

(b) show that younger mothers provide less emotional and cognitive stimulation
for their children, as do mothers with less schooling (figures 7(c) and (d)). While
the debate is not settled as to which features of adverse family environments are
most harmful to the success of children, there is uniform agreement that poor
environments adversely affect child outcomes.
Other studies show the same pattern. Mayer analyzed child outcomes
classified by a long run measure of parental income.30 Low family income is
associated with single parenthood, divorce, reduced education, and low
parental ability. Child test scores are greater for children from higher income
families. Teenage pregnancy and high school dropout rates are strongly
negatively correlated with family income. Young adult education, earnings,
wage rates, and participation in social pathologies are much greater for children
from poor families. Mayer does not isolate which factors in the constellation of
poverty are the main causes of poor child outcomes; but the constellation has a
clear association with adverse child outcomes.

McLanahan and Sandefur focus on another aspect of childhood disadvantage:
one-parent versus two-parent families. For a variety of datasets, and controlling
for parental education, and family size, they show that: (1) attrition from high
school is higher,31 while test scores and school expectations are lower for
children from one-parent families32; (2) college enrollment is lower33; (3) labor

Review of Agricultural Economics

Figure 7. (a) Average cognitive stimulation score by mother’s age at
birth; (b) Average emotional stimulation score by mother’s age at
birth; (c) Average cognitive stimulation score by mother’s final years
of schooling; and (d) Average emotional stimulation score by mother’s
final years of schooling

(a)
Note: Overall stimulation is a measure of the quality of the child’s home environment. It comprises
emotional and cognitive stimulation subscores. It is based on measures of resources, such as books,
and on interactions with parents. The score is measured in percentiles.

(b)
Note: Overall stimulation is a measure of the quality of the child’s home environment. It comprises
emotional and cognitive stimulation subscores. It is based on measures of resources, such as books,
and on interactions with parents. The score is measured in percentiles.

Continued

Proceedings 465

Figure 7. Continued

(c)
Note: Overall stimulation is a measure of the quality of the child’s home environment. It comprises
emotional and cognitive stimulation subscores. It is based on measures of resources, such as books,
and on interactions with parents. The score is measured in percentiles.

(d)
Note: Overall stimulation is a measure of the quality of the child’s home environment. It comprises
emotional and cognitive stimulation subscores. It is based on measures of resources, such as books,
and on interactions with parents. The score is measured in percentiles.

force and school withdrawal is greater for disadvantaged children34; and (4)
teenage pregnancy is greater.35 Ginther and Pollak extend their analysis to note
that the real dichotomy is that between children living with both biological
parents versus other family structures. Being raised in an intact, two-parent
family benefits child outcomes, relative to other family statuses.

Review of Agricultural Economics

Armor presents evidence on a variety of home environmental factors and uses
test scores of children as the outcomes for his analysis. Test scores, taken at early
ages, predict schooling and many other outcomes (Cameron and Heckman
2001). Armor shows the gap in ability and knowledge of math between children
of teenage mothers and children of older mothers.36 The gaps are 20 points
when he does not control for maternal ability and are smaller but still important
when he controls for parental ability (6 points higher ability leads a person to
complete two more years of school). His book demonstrates the importance of
parental ability as well as the additional negative effect of teenage pregnancy on
child outcomes.

Armor studies the effects of cognitive stimulation on child ability and math
scores.37 He goes part way toward isolating the factors characterizing adverse
environments. Armor studies the effects of various environmental factors on the
ability and math achievement of children.38 Mothers’ ability plays an important
role but even controlling for that effect, family environmental factors play a
substantial role in raising child test scores. Controlling for maternal ability,
never-wed mothers who provide above average cognitive stimulation to their
children can largely offset the circumstance of single parenthood in terms of
their child’s cognitive outcomes. This evidence is consistent with a large body of
research reported in the National Research Council Report Neurons to
Neighborhoods (Shonkoff and Phillips; Carneiro, Heckman, and Masterov; Cunha
et al.).

The growth of adverse childhood environments explains a substantial part of
the problems of schools, skills and crime in American society. It is especially
problematic that poor environments are more common in the minority
populations on which America must depend for the growth in its labor force
(recall the data in table 1). Unless these environments are improved, one cannot
rely on a growth in the skills of these groups to propel growth in workforce
quality at the rate we have experienced in the past.

The Importance of Cognitive and Noncognitive Ability
in Economic Life

A large literature has established the importance of both cognitive and
noncognitive ability in social and economic life. Basic intelligence, acquired
skills, social skills, self-control, and persistence matter for success in life (see
Heckman, Stixrud, and Urzua [2006] for recent evidence). The full implications
of this body of evidence have not yet made their way into the design of
economic and social policy.

Cameron and Heckman (1999, 2001) document that substantial gaps in the
college-going rates of different racial and ethnic groups, which are nominally
due to gaps in parental family income in the college-going years, are actually
due to ability differences—that is, child college readiness. Adjusting for ability,
family income and tuition play only minor roles in accounting for disparity in
college attendance rates. This evidence explains why so many poor or
disadvantaged children fail to utilize the programs that subsidize the college
tuitions of the disadvantaged.

Proceedings 467

Figure 8. (a) Fraction of women who gave birth by their eighteenth
birthday; (b) Fraction of male respondents in jail at age thirty or below;
(c) Average cognitive stimulation score by mother’s AFQT decile;
and (d) Average emotional stimulation score by mother’s AFQT decile

Data from NLSY

(a)
.2

0 2 4 6 810
AFQT Decile

Fraction

.15
.1
.05

0

Note: Uses the AFQT calculation procedure as defined by the Department of Defense in 1989. Data
used 1979–2000

Data from NLSY

(b)
.2

.15

0 2 4 6 810
AFQT Decile

Fraction

.1

.05

0

Note: Uses the AFQT calculation procedure as defined by the Department of Defense in 1989. Data
used 1979–2000

Continued

Review of Agricultural Economics

Figure 8. Continued

(c)
Note: Overall stimulation is a measure of the quality of the child’s home environment. It comprises
emotional and cognitive stimulation subscores. It is based on measures of resources, such as books,
and on interactions with parents. The score is measured in percentiles.

(d)
Note: Overall stimulation is a measure of the quality of the child’s home environment. It comprises
emotional and cognitive stimulation subscores. It is based on measures of resources, such as books,
and on interactions with parents. The score is measured in percentiles.

Proceedings 469

In the next section, we show that the ability gaps that explain college
attendance gaps open up early, before schooling begins. A school-based policy
for eliminating these gaps is less effective. Ability formed in the early years is
also important in explaining crime, teenage pregnancy, and a variety of social
pathologies. Figure 8(a) shows that women with low cognitive ability are more
likely to bear children when they are young. Figure 8(b) shows that low
cognitive ability is associated with a higher probability of incarceration. Ability
also affects the economic return to each year of schooling. Figures 8(c) and (d)
show that mothers with low cognitive ability provide less cognitive and
emotional stimulation for their children. Finally, in their research, Carneiro and
Heckman (2003) show that the economic returns to one year of college for
people of different ability differ greatly.39 Those at the bottom 5% of the ability
distribution get half of the return to education of those at the top 5% of the
ability distribution. Ability also affects wages independently of schooling, as
shown in Carneiro, Heckman, and Masterov.

Heckman, Stixrud, and Urzua analyze the changes in the probability of
various outcomes that are brought about by altering cognitive or noncognitive
ability, holding the other constant. Figure 9(a), taken from their study, clearly
shows that higher levels of both cognitive and noncognitive skills are associated
with lower rates of attrition from high school. For many outcome measures in
their study, increasing noncognitive ability by the same percentile has a higher
effect on outcomes than cognitive ability.

Increasing noncognitive ability to the highest level reduces the probability of
being a high school dropout to virtually zero for females with average cognitive
ability.40 The same argument holds for other behavioral outcomes. Both types of
ability have the same effect on reducing the likelihood of spending time in jail
by age thirty (figure 9(b)). Figure 9(c) shows the same effect for smoking.
Figure 9(d) show this for teenage pregnancy. For this outcome, noncognitive
ability is as important as cognitive ability. 41

Human Ability and Its Determinants

The recent synthesis of neuroscience and social science has produced a much
deeper understanding of the processes by which skills are formed over the life
cycle, although much remains to be known (Shonkoff and Phillips; Knudsen
et al.; Cunha and Heckman forthcoming). The social science literature
establishes that both cognitive and noncognitive abilities affect schooling
attainment, participation in welfare, teenage pregnancy, and crime (see
Heckman and LaFontaine [2006] for a comprehensive analysis). More able and
engaged parents produce more able children.

The recent literature distinguishes between IQ and achievement tests. IQ
approximates intellectual capacity. Achievement tests capture knowledge in
specific areas. IQ spurs achievement. At the same time, persons more motivated
to learn and more persistent, and those who plan ahead—important aspects of
noncognitive skills—also score higher on achievement tests at the same level of
IQ. Families produce both cognitive and noncognitive skills, and both matter for
the social and economic success of the child. Gaps among income and race
groups open up early and persist.

470 Review of Agricultural Economics
Figure 9. (a) Probability of being a high school dropout by age
thirty—males; (b) Probability of incarceration by age thirty—males;
(c) Probability of daily smoking by age eighteen—males; and (d) Probability
of being single with child at age eighteen—females
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). Source: Heckman, Stixrud, and
Urzua (2006).
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). Source: Heckman, Stixrud, and
Urzua (2006).
Continued

Proceedings 471
Figure 9. Continued
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). Source: Heckman, Stixrud, and
Urzua (2006).
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). Source: Heckman, Stixrud, and
Urzua (2006).

Review of Agricultural Economics

Figure 10(a) presents the average percentile ranks on a math test administered
at ages six, eight, ten, and twelve for children from different income groups. The
test measures a composite of raw IQ and achievement.42 Gaps in ranks by family
income are substantial overall. Figure 10(b) shows that these differentials are
greatly reduced when the scores are adjusted by mother’s IQ, education, and
intact family status. Similar adjustments appear when the mother’s status is
controlled for, and when other test scores are used. Enriched environments
produce higher ability children.43

Figures 11(a) and (b) present parallel analyses for noncognitive skills. A high
value of an antisocial score stands for a range of behavioral problems. High
scores are associated with low-income environments; low scores with
high-income environments. Again, gaps open up early among income groups,
and again, gaps can largely be eliminated by accounting for the quality of the
early environments facing the child.44 A large body of literature, surveyed in
Carneiro and Heckman (2003) and Cunha et al., demonstrates that skill gaps
open up early, before schooling begins, and that these gaps are major
determinants of social and economic success. The strong association between
family characteristics and child performance measured by cognitive and
noncognitive skills also demonstrates the value of a strategy targeted toward
disadvantaged families.

Implications of the Evidence on Ability for Skill Formation Policy

The policy implications of the emerging body of evidence on the technology
of human skill formation are substantial. Conventional school-based policies
start too late to effectively remedy early deficits, although they can do some
good. The best way to improve the schools is to improve the early environments
of the children sent to them.

At current levels of funding, incremental expenditures on schooling quality
are unlikely to be effective. Table 6 is based on estimates of the effect of
schooling on earnings from a article by Card and Krueger that greatly
influenced recent California efforts to reduce class size. It shows the discounted
economic returns (i.e., effects on discounted lifetime income) to decreasing
pupil–teacher ratios by five, but keeping the quality of students the same.
Reducing pupil–teacher ratios is frequently advocated to raise the performance
of schools. Taking the most favorable estimates reported by these advocates of
schooling programs produces a net negative return, even if the social cost of
taxation used to fund schooling is ignored and optimistic estimates of aggregate
productivity growth are used. The cost of reducing class size would be better
spent on giving children a savings account. These calculations are too optimistic
because they understate the full costs of the policy, which would entail
substantial increases in teacher salaries to hire the new teachers, or lower the
quality of teachers hired into the school system.45

The celebrated Tennessee Star experiment produced, at best, marginal gains to
participants that did not survive a rigorous cost benefit analysis (see the
discussions in Hanushek; Krueger). The widely discussed policy of improving
schools by reducing pupil–teacher ratios is unlikely to have substantial benefits
unless the quality of the input going to school is improved (see Carneiro and

Proceedings 473

Figure 10. Children of NLSY (a) (Average percentile rank on PIAT-
Math score, by income quartile*; and (b) Adjusted average PIAT-Math
percentiles by income quartile*

(a)
*Income quartiles are computed from average family income between the ages of six and ten.

(b)
*Adjusted by maternal education, maternal AFQT (corrected for the effect of schooling) and
broken home at each age

Heckman 2003). The importance of family to the success in schools has been
known since the Coleman Report, but this wisdom has not yet found its way
into policy.

Tuition and family income support for families of children in the college-going
years are often proposed. The basis for this policy recommendation is the
empirical regularity that child college-going rates are inversely related to family
income in the college-going years. This empirical association is treated as a
causal relationship which should guide policy. Politicians around the world
campaign on this issue. The recent literature, surveyed in Carneiro and

Review of Agricultural Economics

Figure 11. Children of NLSY (a) Average percentile rank on antisocial
score, by income quartile*, and (b) Adjusted average anti-social
score percentile, by income quartile*

(a)
(b)
*Adjusted by maternal education, maternal AFQT (corrected for the effect of schooling) and
broken home at each age

Heckman (2002, 2003), documents that at most 8% of American children are
cash-constrained in the college-going years. While a policy targeted to the
cash-constrained has a high economic return, it will not go far in promoting
college attendance or reducing schooling among racial and ethnic groups.

As Carneiro and Heckman (2003), Cunha et al., and Cunha and Heckman
(forthcoming) document, the real credit constraint facing children is not the lack
of access to funds for tuition and room and board in the college-going years.
Rather, it is the inability of children to borrow against future income to buy a
parental environment that will allow them to fulfill their potential. It is the
accident of birth.

Proceedings 475

Table 6. Evaluating school quality policies: Discounted net returns
to decreasing pupil–teacher ratio by five pupils per teacher for
people with twelve years of schooling in 1990

Annual Rate of Return
to Earnings from School
Productivity Includes 50% Social
Quality Change
Growth Rate (%) Cost of Funds 1% 2%
7% Discount rate
0 Yes -9,056 -8,092
0 No -5,716 -4,752
1 Yes -8,878 -7,736
1 No -5,538 -4,396
5% Discount rate
0 Yes -9,255 -7,537
0 No -5,597 -3,880
1 Yes -8,887 -6,802
1 No -5,230 -3,145
3% Discount Rate
0 Yes -8,840 -5,591
0 No -4,810 -1,562
1 Yes -8,036 -3,984
1 No -4,007 45

Note: All values, in 1990 dollars, are given as net present values at age eight of an individual; costs
of schooling improvements are incurred between ages six and eighteen; and benefits from increased
earnings occur between ages nineteen and sixty-five. Data for costs are from NCES 1993. Costs of
adding new teachers include salaries and capital, administrative, and maintenance expenditures.
Estimates of increases in earnings resulting from decreasing the pupil–teacher ratio by five pupils per
teacher come from Card and Krueger (1992, table 3), which produces a range of estimated earnings
increase from about 1 to 4%, whereas most of the estimates are in the 1 to 2% range, which we use
in this article. To capture the benefits of smaller class sizes, students must attend twelve years of
higher-quality schooling. We calculate the costs for one year of improvements and then calculate the
present value of the costs over the twelve years of school attendance.

The empirical regularity that drives policy discussions has been
misinterpreted. The widely discussed correlation between parental income in
the child’s college-going years and child college participation arises only
because it is lifetime resources that affect college readiness and college-going, and
family lifetime resources are strongly positively related to family resources
available to the adolescent in the college-going years.

Government job training programs and GED programs are second chance
efforts designed to remedy the deficits caused by early childhood and schooling
neglect. The GED program does not confer benefits to very many of its
participants (Heckman and LaFontaine 2006). Job training programs targeted at
the disadvantaged do not produce high rates of return and fail to lift
participants out of poverty (see the evidence in Heckman, LaLonde, and Smith;
and in Martin and Grubb 2001). At current levels of funding, these programs are
largely ineffective and cannot remedy the skill deficits accumulated over a
lifetime of neglect.

Review of Agricultural Economics

Figure 12. Rates of return to human capital investment in disadvantaged
children

Rate of
Return to Preschool Programs
nestment

Ivin Human
Capital

Schooling

Job Training

Preschool School Post School

0

g

Ae

Cunha and Heckman (forthcoming), and Cunha, Heckman, and Schennach
formalize the technology of skill formation by families and estimate empirical
models of dynamic skill formation. They show that investments in children are
complementary and that early investments improve the return on later
investments. The self productivity of early investment warrants more
investment in the young.

Their analysis shows that the young receive highest returns to a dollar of
investment. Early skills breed later skills because early learning begets later
learning. Both on theoretical and empirical grounds, at current levels of funding,
investment in the young is warranted. Returns are highest for investments made
at younger ages and remedial investments are often prohibitively costly.
Figure 12 summarizes their model and the findings of an entire literature.
Returns for disadvantaged children are highest for investments made at young
ages. The optimal investment profile declines with age. This pattern is true for
all children. But more advantaged children receive massive early investments
from their parents that disadvantaged children do not receive. Figure 12 shows
the returns for human capital programs for the disadvantaged at current levels
of investment.

This literature does not suggest that no investments should be made in
schooling or post-school on-the-job training. They are major sources of skill
formation. Indeed, the complementarity or synergism between investments at
early and later ages suggests that early investment has to be complemented by
later investment to be successful. Currie and Thomas suggest that the effects of
early investment will dissipate unless it is followed by later investment. If early
investments are made, the returns to later investments will rise. Investment in

Proceedings 477

the preschool years raises the productivity of schooling and post-school job
training. Cunha and Heckman (2006) show that adolescent remediation for the
effects of adverse early environments is very costly and Cunha and Heckman
(forthcoming) present an analytical synthesis of the literature.

However, the self-productivity of investment suggests that an optimal
investment strategy should focus investments in the early years compared to the
later years. Carneiro and Heckman (2003) argue as an empirical proposition in
the U.S. that there is currently under-investment in the young, especially in
disadvantaged populations.46

Two matters of concern arise in using this evidence to guide policy. First, it is
associational or correlational. It establishes empirical relationships that may or
may not be causal. Second, while family factors matter, it is far from obvious
how to improve families. We cannot easily raise the education of parents, nor
can we improve their IQs.

The evidence presented in Armor, in figures 10 and 11, and in the other
studies reviewed here suggests that early investment is productive. But
traditionally, the early years of a child’s life are the exclusive province of the
family. The tough question is how to enrich the family and at the same time
preserve the benefits of parents? An accumulating body of evidence on
voluntary interventions points the way. We now turn to a review of the evidence
on the benefits of these voluntary interventions.

In the past forty years, many voluntary interventions have been devised to
improve the early years of children by supplementing the resources of
disadvantaged families. These family supplements do not actively intrude on
family life, yet they enrich the early years of the child.

Some of these interventions have been implemented using random
assignment. Packages of enriched environments are randomly assigned to
children in disadvantaged environments, while children in comparable families
are randomly denied access to the enriched treatment. When successfully
implemented, randomization allows analysts to be more confident that the
empirical associations produced by the interventions are causal. The findings
from this experimental literature bolster the evidence from the associational
literature that we have just discussed.

Evidence from Enriched Preschool Programs

Currie, and Blau and Currie present comprehensive surveys of numerous
preschool programs and their measured effects.47 The programs they analyze
vary, both in terms of age of enrollment and age of exit. The effects, however, are
generally consistent, although in some cases only weak effects are found.
Generally, performance of children in school is improved in terms of less grade
repetition, more graduation, and higher test scores. Unfortunately, many of
these programs are not evaluated by following children into late adolescence or
adulthood and looking at their outcomes.

Three programs have long-term follow-ups, and we focus on them here. They
all target high-risk children from disadvantaged families. The Chicago
Child-Parent Centers (CPC) is a half-day program on a large scale in the

Review of Agricultural Economics

Chicago public schools. It is evaluated by a nonexperimental method (matching)
and has a sample of about 1,500 children. The second program is the
Abecedarian program, a full-day, year-round educational child care program in
Chapel Hill, NC. It was evaluated by randomization and has 111 participants.
Students are followed to age twenty-one. Finally, the High/Scope Perry
Preschool is a small-scale half-day program in the Ypsilanti, MI public schools. It
was evaluated by experimental methods. Sample size is 123, and follow-up is to
age twenty-seven. CPC and Perry had a parental involvement component—
Abecedarian did not.

The programs differ by duration and child age of entry. Abecedarian started
with young children in the first months of life. Perry and the CPC program start
with older children, three or, four to five years old. The programs differ in
intensity.48 It is also important to point out that the comparison made in all of
the studies is between children with enriched preschool environments and
children with ordinary early environments, some of whom may attend
preschool and kindergarten, albeit of a less intense variety.49

Program Descriptions

Perry Preschool experiment

The Perry Preschool experiment was an intensive preschool program
administered to randomly selected black children enrolled in the program over
five different waves between 1962 and 1967. All the children came from
Ypsilanti, MI. A control group provides researchers with an appropriate
benchmark to evaluate the effects of the preschool program.

The assignment to the experimental group was performed in the following
way. Candidate families were identified from a census of the families of the
students attending the Perry school at the date of operation of the program,
neighborhood group referrals and door-to-door canvassing. Poor children who
scored between 75 and 85 on the Stanford–Binet IQ test were randomly divided
into two undesignated groups.50 The children were then transferred across
groups to equalize the socioeconomic status, cognitive ability (as measured by
the IQ test), and gender composition of the samples. Finally, a coin was tossed to
determine which group received the treatment and which did not. Initially the
treatment and control groups included 64 children each, but the actual treatment
and control groups contained 58 and 65 children, respectively.51

Children entered the Perry School in five waves, starting with wave zero (of
four-year-olds) and wave one (of three-year-olds) in 1962, and then waves two,
three, and four (of three-year-olds) entered in each subsequent year through
1965. The average age at entry was 42.3 months. With the exception of wave
zero, treatment children spent two years attending the program. In the final year
of the program, 11 three-year-olds who were not included in the data attended
the program with the 12 four-year-olds who were. About half of the children
were living with two parents. The average mother was twenty-nine years old
and completed 9.4 years of school.

The treatment consisted of a daily two-and-a-half hour classroom session on
weekday mornings and a weekly ninety-minute home visit by the teacher on
weekday afternoons to involve the mother in the educational process. The length

Proceedings 479

of each preschool year was 30 weeks, beginning in mid-October and ending in
May. Ten female teachers filled the four teaching positions over the course of the
study, resulting in the average child–teacher ratio of 5.7 for the duration of the
program.52 All teachers were certified to teach in elementary, early childhood, or
special education.53 If it were administered today, the Perry Preschool program
would cost approximately $9,785 per participant per year in 2004 dollars.

Abecedarian Project

The Abecedarian Project recruited 111 children born between 1972 and 1977
whose 109 families scored high on the High Risk Index.54 It enrolls and
intervenes in the lives of children beginning a few months after birth.
Enrollment is based on the characteristics of the families more than on those of
the children, as in the Perry program. Virtually all of the children were black,
and their parents had low levels of education, income, cognitive ability, and
high levels of pathological behavior. The children were screened for mental
retardation. 76% of the children lived in a single parent or multigenerational
household. The average mother in this group was less than twenty years old,
completed ten years of schooling and had an IQ of 85. There were four cohorts
of about 28 students each. By the time they were six weeks old, the children
were assigned randomly to either a preschool intervention or a control group.
The mean age of entry was 4.4 months. At age five—just as they were about to
enter kindergarten—all of the children were reassigned to either a school age
intervention through age eight or to a control group. This produced four distinct
groups: children who experienced no intervention at all, those who experienced
an intervention when they were young, those who experienced an intervention
when they were older; and finally, those who enjoyed a high-quality
intervention throughout their whole childhood. The children were followed up
until age twenty-one.

The Abecedarian program was more intensive than the Perry program. Its
preschool program was a year-round, full-day intervention. The initial
infant-to-teacher ratio was 3:1, though it grew to a child-to-teacher ratio of 6:1 as
the kids progressed through the program. Infants in the control group received
an iron-fortified formula for fifteen months and diapers as needed to create an
incentive for participation. Many of the control children were enrolled in
preschool and/or kindergarten.

During the first three primary school years, a home-school teacher would
meet with the parents and help them provide supplemental educational
activities at home. The teacher provided a curriculum tailored specifically for
each child. The target set for the parents was at least fifteen minutes per day of
supplementary activities. This home-school teacher would also serve as a liaison
between the teachers and the family, and she would interact with the parents
and the teachers about every two weeks. She would also help the family deal
with other issues that might improve their ability to care for the child, such as
finding employment, navigating the bureaucracy of social services agencies, and
transporting children to appointments. Data were collected regularly up to age
twenty-one. In terms of 2004 dollars, it cost roughly $15,000 per year.

Review of Agricultural Economics

Chicago Child-Parent Center and expansion program

The Chicago Child-Parent Center was not evaluated by the method of random
assignment but by matching treated children to comparable non-treated children
on the basis of age, eligibility for intervention, and family socioeconomic status.
It was started in 1967, in eleven public schools serving impoverished
neighborhoods of Chicago. Using federal funds, the center provided half-day
preschool program for three-and four-year-olds during the nine months that
they were in school. The program provided an array of services, including
health and social services, and free meals. It also sought to include the parents,
including helping the parents complete school, home visits, and field trips.

In 1978, state funding became available, and the program was extended
through third grade and included full-day kindergarten. Eventually, 24 centers
provided preschool and after-school activities, up to second or third grade. This
is the period during which the sample analyzed by Reynolds et al. was enrolled
in the program. The preschool program ran three hours per day during the week
for the nine months that school was in session, and usually included a six-week
summer program.

During the kindergarten years, more services were provided at the affiliated
school. Teacher–child ratios were 17:2 for the preschool component and 25:2 for
kindergarten. Participation during the primary years was open to any child in
the school. Program participants experienced reduced class sizes of 25 rather
than 35 or more. Teachers’ aides, extra instructional materials, and enrichment
activities were also available. Some children continued to participate in CPC
through age nine, for a maximum of six years.55 Ninety-three percent of the
children were black and 7% were Hispanic. Costs were considerably less, but
intensity was correspondingly lower (Cunha et al.).

Lessons from Early Interventions

These and other studies of interventions for children from low-income
families find that participants experienced higher achievement test scores,
decreased grade retention, reduced time in special education, less crime and
delinquency, and increased high school graduation. The gains vary with quality
and age at which the program is started, and there are important differences by
the sex of the child.

Programs differ in the measures they use to evaluate the outcomes. As a result,
it is hard to compare the programs using a standard basket of benefits. The CPC
program had significant effects on high school graduation rates, reductions in
special (remedial) education, grade repetition, and juvenile arrest (figure 13).

The Perry Preschool program is the flagship intervention. Children are
followed through age forty, with data collected annually from ages three to
eleven, and again at ages fourteen, fifteen, nineteen, twenty-seven, and forty.56
The boost in IQ faded by the time the children were in second grade
(figure 14(a)), but the program had substantial effects on educational
achievement. Test scores for the treatment group were consistently and
significantly higher through age fourteen, and as were literacy scores at nineteen
and twenty-seven. Participants had higher grades and were more likely to
graduate from high school. Substantially less time was spent in special

Proceedings 481

Figure 13. Academic and social benefits at school exit for CPC
participants

education or in repeating grades, and high school graduation rates of
participants improved (figure 14(b)).

Participants were more likely to be employed, to earn more (figure 14(c)), and
they were less dependent on welfare.57 There was substantially less crime among
participants (figure 14(d))—both in terms of incidence and severity, a recurrent
finding of early intervention programs (recall the evidence summarized in
table 5). However, there was no significant difference in grade retention by age
twenty-seven between the two groups. Teenage pregnancy was lower, and
marriage rates were higher by age twenty-seven for program participants.

The Abecedarian Program appears to have had an effect on IQ, but it is
concentrated primarily among girls.58 Figure 15(a) shows the overall IQ gap
between treatments and controls. It is persistent over ages. The Abecedarian
program intervenes in the very early years, and it is known that IQ is malleable
when children are very young (see, e.g., Armor; and the references in Cunha
and Heckman forthcoming). This message is reinforced by the fact that the IQ
boost was not found among children who only experienced the later
intervention. Comparable effects are found for reading (figure 15(b)) and math
achievement scores (figure 15(c)). The test score effects persist through age
twenty-one, which is the last age analyzed.

Figure 15(d) shows that there were substantial academic benefits. Treatment
group members participated less in remedial special education at age fifteen and
repeated fewer grades at all ages. High school graduation and four-year college
participation rates were high. Participants were less likely to smoke and had
better jobs (figure 15(e)).

Review of Agricultural Economics

Figure 14. (a) Perry Preschool: IQ over time; (b) Perry Preschool: Educational
effects; (c) Perry Preschool: Economic outcomes; and (d)
Perry Preschool: Arrests per person by age twenty-seven

(a)
(b)
Graduated fromHigh School on Time
Age 14Achievement at10th Percentile
Continued

Table 7 presents estimated costs and benefits of the Perry and Chicago
programs with benefits discounted at a 3% rate. All figures are in 2004 dollars.
The benefits vary among programs.59 Perry produced some gain to parents in
terms of reduced child care costs, and earnings gains for participants were
substantial. The K-12 benefit arises from the increment in student quality and is
a reduction in special education costs. This benefit is substantial across all

Proceedings 483

Figure 14. Continued

(c)
Own HomeNever on Welfareas Adult
(d)
programs. The college/adult category represents the extra tuition paid by
students who go to college. Crime represents the reduction in direct costs
(incarceration and criminal justice system) as well as damage done to victims.
This excludes transfers. Welfare effects are modest. Future Generation (FG)
Earnings represents the improvement in the earnings of the children of the
program participants. Smoking and health benefits were not measured in the
Perry and Chicago data. For Abecedarian, there were substantial effects,

Review of Agricultural Economics

Figure 15. (a) Abecedarian IQ scores over time; (b) Abecedarian reading
achievement over time; (c) Abecedarian math achievement over
time; (d) Abecedarian academic outcomes; and (e) Other benefits of
Abecedarian

(a)
(b)
Continued

including major differences in smoking rates. CPC documents a decline in child
abuse and the costs of treating abused children. The costs of Perry are
substantial but per year are about the average cost of expenditure on public
school students. CPC per year costs about $6,796 for the preschool and $3,428 for
the school-age component (in 2004 dollars). The reported benefit–cost ratios are

Proceedings 485

Figure 15. Continued

(c)
(d)
Continued

substantial: 9 to 1 for Perry; 8 to 1 for Chicago CPC. Rolnick and Grunewald
claim that the annual rate of return for Perry is 4% for participants and 12% for
society at large, for a total of 16%.

Much more research is needed on Perry, CPC, and a wide variety of other
early childhood program results. Results from these programs need to be put on
a common footing to understand better the differences in samples, treatments,
and effects.60 A much more careful analysis of the effects of scaling up the model

Review of Agricultural Economics

Figure 15. Continued

(e)
Smoker at Age 21
Skilled Job or Higher Education atAge 21
Table 7. Economic benefits and costs

Perry Preschool 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
Future generation 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 52,038
Benefits-to-costs ratio 9.11 7.77

Notes: All values discounted at 3% and are in 2004 dollars. Numbers differ slightly from earlier
estimates because Future Generations (FG) Earnings for Perry and CPC were estimated using the
ratio of FG Earnings Effect to Earnings Effect (about 15%) that was found in Abecedarian.
Source: Barnett (2004).

programs to the target population, and its effects on costs, has to be undertaken
before these estimates can be considered definitive.

The gain from the pilot programs is a lower bound on the potential benefitof
intervening in the early years: although the costs are well established, many of
the benefits cannot be precisely monetized. For instance, we do not yet have a
full accounting of how the children of the participants will respond to the

Proceedings 487

intervention, and neglecting this likely understates its effect. Extrapolating from
old, small, and local programs to large, national ones in the future is precarious
business—a fact often neglected in the early childhood literature. The benefits of
these interventions appear to be sufficiently large that the actual or potential
program may remain cost-effective even after a large reduction in its efficacy.

The Case for Early Intervention

Without claiming to offer a monolithic explanation for the origins of the major
social problems discussed in this article, we nonetheless point out the important
role of disadvantaged families in producing less educated and less motivated
persons and in producing persons more prone to participate in crime. A large
literature establishes that children from disadvantaged homes are less educated
and more likely to participate in social pathologies, including crime. In the past
forty years or so, the American family has come under stress. Relatively more
American children are being raised in the adverse environments that produce
less educated and less skilled individuals and persons more likely to commit
crime and participate in socially deviant behavior.

American society has traditionally appealed to the schools to remedy what
failed families produce. Current policies such as the No Child Left Behind Act are
premised on using schools to remedy the consequences of disadvantaged
families. Schools can only work with what families give them. Successful
schools are those that teach children from functioning families.

In addition, the current emphasis in American schools is on test scores, and
tests ignore crucial noncognitive components of motivation, persistence, and
self-control that successful families foster in their children. Both cognitive and
noncognitive skills are important for success in school and in life.61 The enriched
early childhood interventions have had their greatest impacts on creating
motivation and successful attitudes among participants—traits usually ignored
in discussions of educational policy.

A large body of empirical work at the interface of neuroscience and social
science has established that fundamental cognitive and noncognitive skills are
produced in the early years of childhood, long before children start
kindergarten. The technology of skill formation developed by economists shows
that learning and motivation are dynamic, cumulative processes.62 Schooling
comes too late in the life cycle of the child to be the main locus of remediation
for the disadvantaged. Public schools focus only on tested academic knowledge
and not the noncognitive behavioral components that are needed for success in
life. Schools cannot be expected to duplicate what a successfully functioning
family gives its children. Parental environments play a crucial part in shaping
the lives of children.

Later remediation of early deficits is costly, and often prohibitively so.63
Remedial schooling, public sector job training programs, and second chance
GED programs are largely ineffective at current levels of funding. While these
programs can be improved, and do help a few, they are not cost-effective when
compared with alternative policies.

Families matter. But most Americans are justifiably reluctant to intervene in
the early years and prefer to respect the sanctity of the family. In the past forty

Review of Agricultural Economics

years, American society has experimented with voluntary enriched family
supplementation programs, which offer children from disadvantaged
environments some of the cognitive and emotional stimulation and enrichment
given by more advantaged families.

Children who received these enriched environments were followed into
adulthood. Comparing their social and economic outcomes to those of similar
children denied access to these environments by randomization, one finds that
the treated children perform better at school, are less likely to drop out of school,
and are more likely to graduate high school and to attend college. The treated
children are less likely to be teenage mothers and foster a new generation of
deprived children. They are less likely to be on welfare and less likely to smoke
or use drugs. Treated students have higher test scores. A principal benefitof
early childhood intervention is in shaping the noncognitive skills—behavior,
motivation, and self control—that are not considered an important outcome of
the schooling curriculum in current policy discussions.

The estimated rate of return to the Perry Preschool program is about 16%.
This includes benefits from reduced remediation and reduced crime, as well as
the increased earnings of the participant. All of the children targeted for
intervention are of low ability. While much work remains to be done to bolster
the case for wide-scale application of these programs to disadvantaged families,
the current evidence is powerfully suggestive, if not yet definitive, that
large-scale programs will be effective. None of this evidence supports universal
preschool programs.

It is important to note what we are not saying. We do not claim that all skills
and motivations are formed in the early years, nor that schools and firms do not
matter in producing effective people. We are also not offering any claims that the
early years are the sole determinants of later success, or that persons who are
raised in disadvantaged families should be absolved of any guilt when they
participate in crime. We are simply arguing that early environments play a large
role in shaping later outcomes and that their importance is neglected in current
policy. The recent evidence on the technology of human skill formation
establishes that enriched early environments need to be followed up by good
schooling and workplace learning environments. Complementarity of
investments at different ages is an intrinsic feature of the human skill formation
process. Enriching the early years will promote the productivity of schools by
giving teachers better-quality students. Improving the schools will in turn,
improve the quality of the workforce.

The available evidence on the technology of skill formation shows the
self-productivity of early investment. Figure 12 summarizes the argument. At
current levels of public support, America under-invests in the early years of its
disadvantaged children. Redirecting funds toward the early years is a sound
investment in the productivity and safety of American society, and also removes
a powerful source of inequality.

Acknowledgments

This research was supported by a grant from NICHD (NIH R01-HD043411) and a grant from the
Pew Charitable Trust and the Partnership for America’s Economic Success. The web appendix for
this paper can be downloaded from http://jenni.uchicago.edu/Invest/.

Proceedings 489

Endnotes

1Nontraditional families include single-parent families and families where the parents are not
married. The evidence summarized below shows that children raised in nontraditional families fare
worse in many aspects of social and economic life.

2Ventura and Bachrach, who use data from birth certificates, estimate that nonmarital
childbearing is considerably higher than the number reported in this paper. In recent years, their
estimate is approximately 10 percentage points higher than what we report here. However, their
data do not contain much background information on the mothers, so they are less useful for the
type of analysis that we want to perform. Hence we will use the more conservative estimate.

3See, e.g., Mazumder and the other essays in Bowles, Gintis, and Osborne Groves.

4See Trennert on The Phoenix Indian School, and Mayer on the oscillation of American policy
between improving the material condition of the poor family and replacing it with surrogate
institutions like orphanages and foster care.

5See Katz and Autor for a review of the evidence on skill-biased technological change. For
international evidence, see Machin and Van Reenen.

6See figures A1 and A2 in our web appendix. Figures and tables that have a prefix “A” in the
numbering are from the web appendix, which is available from http://jenni.uchicago.edu/Invest/.

7The GED is an exam-certified, alternative high school degree.

8The share of labor is 0.7 so 0.7 × 0.5 = 0.35 is the contribution of workforce quality to economic
growth.

9The International Adult Literacy Survey (IALS) was conducted by 13 countries to collect
information on adult literacy. In this survey, large samples of adults (ranging from 1,500 to 6,000 per
country) were given the same broad test of their literacy skills between 1994 and 1996. Australia,
Belgium (Flanders), Canada, Germany, Great Britain, Ireland, Netherlands, Northern Ireland, New
Zealand, Poland, Sweden, Switzerland, and the United States participated in the IALS. More
information on the IALS is available in documents located at http://www.nald.ca/nls/
ials/introduc.htm and IALS.

10Data on these two scales appear in figures A3a and A3b on the web. Prose literacy is defined as
the knowledge and skills required to understand and use information from texts such as newspaper
articles and fictional passages. Quantitative literacy (numeracy) is defined as the ability to perform
arithmetic operations (either alone or sequentially) to numbers embedded in printed materials, such
as calculating savings from an advertisement or the interest earned on an investment.

11These cross-country differences are not driven by illiterate immigrants coming to the U.S. While
immigrants perform worse on the three tests relative to natives, including immigrants in the analysis
only raises the proportion of U.S. females in Level 1 significantly for prose, quantitative, and
document literacy. The difference is not significant for any other group or level.

12These trends are documented at our website. See figures A4a–44c.

13See table A1.

14The extra year of school is assumed to take place during high school years. The effect of an extra
year of kindergarten or college is likely to be rather different.

15Lochner and Moretti use estimates of victim costs and property losses taken from Miller, Cohen,
and Wiersema, which are based on jury awards in civil suits. Some costs cannot be quantified
accurately or are unobservable. These include costs of precautionary behavior, private security
expenditures, some law enforcement and judicial costs (i.e., costs that are not related to dealing with
particular crimes) and the cost of drug offenses. Some crimes are also omitted from the analysis.

16Incarceration cost per crime is equal to the incarceration cost per inmate multiplied by
incarceration rate for that crime (approximately $17,000).

17It is important to note that this is a steady state calculation. The payoff to pre-K interventions
shows up ten to fifteen years later, whereas the effects of increasing police on crime are more
immediately realized. The discounted returns from the two policies are less different, but a 5:1 gap
can tolerate a lot of discounting and still be substantial.

18Lochner and Moretti actually present a comparison of flow costs ($80,000 per year on a police
officer) with a one time stock cost ($600,000 to educate 100 new high school students at a cost of
$6,000 per year assuming that dropouts get eleven years of school). Cameron and Heckman (2001)
estimate 10.6 years. Assuming a 40-year working life (including criminal career life) the annual
replacement flow cost is $15,000 a year ($6,000×2.5). Even cutting the career life in half produces a
flow cost that is less than hiring a policeman. Spending $9,000 per year (to account for the 1.5 year
gap between high school dropouts and graduates) still makes education cost effective. The evidence
from the Perry Preschool Program suggests that our calculation is conservative. At a cost of $9,000
(2004) per participant, the high school graduation rate was raised by 0.17 from 0.60. To get 2.5 more
students to graduate requires that we spend only $5,300 per pupil. Foregone earnings in high school
are small and are offset by the rise.

19See Figure A5a.

Review of Agricultural Economics

20See Federal Interagency Forum on Child and Family Statistics for more details.

21See Figure A5b.

22Ginther and Pollak summarize the evidence succinctly and present a more nuanced analysis of
family types on adverse outcomes.

23See Ventura and Bachrach. See figure A5c.

24The corresponding birthrate for married women in these three years was 93.2, 83.8, and 87.3.

25See figure A5d.

26Birthrates by age within race/ethnic groups show essentially the same pattern as the overall
rated by race/ethnicity.

27See figure A5e.

28See Figure A5f.

29See Figure A5g.

30See Table A2, where we reproduce her results.

31See Table A3a.

32See Table A3b.

33See Table A3c.

34See Table A3d.

35See Table A3e.

36See Table A4a.

37See Table A4b.

38See Table A4c.

39See Table A5.

40Figures A6a–c in the web appendix show the same pattern for other levels of educational
attainment like high school graduation and college attendance.

41Figures A8a–c in the web appendix show the same pattern for other reproductive outcomes.

42The test measures age-appropriate math knowledge.

43Figures A9a–d repeat this analysis for different race and income groups.

44Figures A10a–d repeat this analysis for different race and income groups.

45The recent California initiative to reduce pupil–teacher ratios ended in widely acknowledged
failure (Stecher and Bohrnstedt).

46See Figure A11 for a diagram of the investment profile.

47Table A6, from Currie, describes some of the main programs, evaluated by randomized
assignment, and their consequences. Table A7 shows the effects of large-scale public early
childhood programs which were not evaluated by randomized assignment.

48See Table A8.

49Arguably the experimental studies understate the value of early childhood interventions
against “no intervention” because some of the control group children received treatment. See
Heckman, LaLonde, and Smith for an additional discussion of randomization.

50Poverty status was determined by a formula that considered rooms per person in the child’s
household, parental schooling, and occupational level. The IQ range was labeled as “borderline
educable mentally retarded” by the state of Michigan at the time of the experiment. Only children
without an organic mental handicap were included in the study.

51Some aspect of the assignment was clearly nonrandom. First, younger children were assigned
to the same group as their older siblings. Two treatment children were transferred to the control
group because their mothers were not able to participate in any classes or home visits because they
were employed far from home. Four treatment children left the program before completing the
second year of preschool when their families relocated and one control child died. Thus, the final
sample consisted of 123 children coming from 100 families. In the control group, 41 families
contributed 1 child each, and 12 families contributed 2 children each. In the treatment group, 39
families contributed 1 child apiece, 6 families contributed 2 children apiece, 1 family contributed 3
and another 4 children. Assigning younger siblings to the same group effectively made the family,
rather than the individual, the unit of analysis. Still, it is difficult to argue that assigning siblings at
random would have been a better strategy. So-called spillovers to the control siblings from home
visits would have been one possible source of bias since mothers cannot be expected to treat siblings
in accordance with their experimental status. Another potential source of bias is spillover from one
sibling to another. In any case, differences in background characteristics between the two
experimental groups are virtually nonexistent, with the exception of much higher rates of maternal
employment at program entry in the treatment group.

52This number is low relative to other early education experiments. For instance, the
student–teacher ratio for the Chicago Child-Parent Center and Expansion Program ranged from 8 to
12 (see Fuerst and Fuerst).

53Schweinhart, Barnes, and Weikart argue that the certification of the teachers is an important
component in the success of the Perry Preschool.

Proceedings 491

54The factors that were considered consisted of weighted measures of maternal and paternal

education levels, family income, absence of the father from the home, poor social or family support

for the mother, indication that older siblings have academic problems, the use of welfare, unskilled

employment, low parent IQ, family members who sought counseling or support from various

community agencies. Parental income and education were considered most important in calculating

the index.
55These costs depend on the stage of the program and are presented in detail in the next section.
56See Schweinhart et al. for a summary of results up through age forty.
57The difference in employment rates was only significant at age nineteen.
58Heckman notes that the Perry program tends to show stronger effects for girls than boys.
59There is a cost benefit study of the Abecedarian program (Barnett and Masse), but it is highly

speculative, so we did not include it here.
60This task is being undertaken by a consortium housed at the Harris School, University of

Chicago.
61See the evidence in Heckman, Stixrud, and Urzua.
62See Cunha et al. (2006) for a summary of this evidence and Knudsen et al.
63See the evidence in Cunha and Heckman (2006).

References

Anderson, D.A. “The Aggregate Burden of Crime.” J. Law and Econ. 42, 2(1999):611–42.

Armor, D.J. Maximizing Intelligence. New Brunswick, NJ: Transaction Publishers, 2003.

Barnett, W.S. “Benefit-Cost Analysis of Preschool Education.” PowerPoint presentation. Available at
http://nieer.org/resources/files/BarnettBenefits.ppt, accessed Nov 2004.

Barnett, W.S., and L.N. Masse. “A Benefit-Cost Analysis of the Abecedarian Early Childhood
Intervention.” Technical report, National Institute for Early Education Research (NIEER), New
Brunswick, New Jersey, 2002.

Blau, D., and J. Currie. “Preschool, Daycare, and Afterschool Care: Who’s Minding the Kids?” In
Handbook of the Economics of Education, Volume 1 of Handbooks in Economics, E. Hanushek and F.
Welch, eds. Amsterdam: North-Holland, 2006.

Bowles, S., H. Gintis, and M. Osborne Groves, eds. Unequal Chances: Family Background and Economic
Success. Princeton, NJ: Princeton University Press, 2005.

Bureau of Justice. “Bureau of Justice Statistics Criminal Offenders Statistics.” Available at
http://www.ojp.usdoj.gov/bjs/ accessed 2004.

Cameron, S.V., and J.J. Heckman. “Can Tuition Policy Combat Rising Wage Inequality?” In Financing
College Tuition: Government Policies and Educational Priorities, Chapter 5, M. Kosters, ed., p. 125.
Washington DC: AEI Press, 1999.

——. “The Dynamics of Educational Attainment for Black, Hispanic, and White Males.” J. Pol. Econ.
109, 3(2001):455–99.

Card, D., and A.B. Krueger. “Does School Quality Matter? Returns to Education and the
Characteristics of Public Schools in the United States.” J. Pol. Econ. 100, 1(1992):1–40.

Carneiro, P., F. Cunha, and J.J. Heckman. “Interpreting the Evidence of Family Influence on Child
Development.” In The Economics of Early Childhood Development: Lessons for Economic Policy,
Minneapolis, Minnesota: The Federal Reserve Bank, 2003.

Carneiro, P., and J.J. Heckman. “The Evidence on Credit Constraints in Post-Secondary Schooling.”
Econ. J. 112, 482(2002):705–34.

Carneiro, P., and J. J. Heckman. “Human Capital Policy.”,In Inequality in America: What Role for
Human Capital Policies?, J.J. Heckman, A.B. Krueger, and B.M. Friedman, eds. Cambridge, MA:
MIT Press, 2003.

Carneiro, P., J. J. Heckman, and D. V. Masterov. “Labor Market Discrimination and Racial
Differences in Pre-market Factors.” J. Law and Econ. 48, 1(2005):1–39.

Coleman, J.S. Equality of Educational Opportunity. Washington DC: U.S. Dept. of Health, Education,
and Welfare, Office of Education, 1966.

Cunha, F., and J.J. Heckman. “Investing in Our Young People.” Unpublished manuscript, University
of Chicago, Department of Economics, 2006.

——. “The Technology of Skill Formation.” Amer. Econ. Rev. 97, 2, 2007.

Cunha, F., J. J. Heckman, L. J. Lochner, and D. V. Masterov. “Interpreting the Evidence on Life Cycle
Skill Formation.” In Handbook of the Economics of Education, E.A. Hanushek and F. Welch, eds.,
pp. 697–812. Amsterdam: North-Holland, 2006.

Cunha, F., J. J. Heckman, and S. M. Schennach. “Estimating the Technology of Cognitive and
Noncognitive Skill Formation.” Unpublished manuscript, University of Chicago, Department
of Economics. Presented at the Yale Conference on Macro and Labor Economics, May 5–7, 2006.
Under revision, Econometrica, 2007.

Review of Agricultural Economics

Currie, J. “Early Childhood Education Programs.” J. Econ. Perspect. 15, 2(2001):213–38.

Currie, J., and D. Thomas. “School Quality and the Longer-term Effects of Head Start.” J. Hum.
Resources 35, 4(2000):755–74.

Delong, J.B., L. Katz, and C. Goldin. “Sustaining American Economic Growth.” In Agenda for the
Nation, H. Aaron, J. Lindsay, and P. Nivola, eds., Washington DC: Brookings Institution Press,
2003.

Donohue, J.J., and P. Siegelman. “Allocating Resources among Prisons and Social Programs in the
Battle against Crime.” J. of Legal Studies 27, 1(1998):1–43.

Ellwood, D.T. “The Sputtering Labor Force of the Twenty-First Century: Can Social Policy Help?” In
The Roaring Nineties: Can Full Employment Be Sustained?, A. Krueger and R. Solow, eds.,
pp. 421–489. New York: Russell Sage Foundation, 2001.

Ellwood, D.T., and C. Jencks. “The Growing Differences in Family Structure: What Do We Know?
Where Do We Look for Answers?” Unpublished manuscript, Kennedy School of Government
Harvard University, 2002.

——. “The Spread of Single-parent Families in the United States Since 1960.” In The Future of the
Family, Chapter 2, D. P. Moynihan, L. Rainwater, and T. Smeeding, eds. New York: Russell Sage
Foundation, 2004.

Federal Interagency Forum on Child and Family Statistics. America’s Children in Brief: Key National
Indicators of Well-Being, 2004. Washington DC: Federal Interagency Forum on Child and Family
Statistics, 2004.

Fuerst, J., and D. Fuerst. “Chicago Experience with an Early Childhood Program: The Special Case of
the Child Parent Center Program.” Urban Educ. 28, 1(1993):69–96.

Ginther, D.K., and R.A. Pollak. “Family Structure and Children’s Educational Outcomes: Blended
Families, Stylized Facts, and Descriptive Regressions.” Demography 41, 4(2004):671–96.

Hanushek, E. “The Failure of Input-based Schooling Policies.” Econ. J. 113, 485(2003):F64–F98.

Heckman, J. J. “Invited Comments.” In Lifetime Effects: The High/Scope Perry Preschool Study through
Age 40, L. J. Schweinhart, J. Montie, Z. Xiang, W.S. Barnett, C.R. Belfield, and M. Nores, eds.,
pp. 229–33. Ypsilanti, MI: High/Scope Press. Monographs of the High/Scope Educational
Research Foundation, 14.

Heckman, J.J., and P. LaFontaine. “Bias-Corrected Estimates of GED Returns.” J. Labor Econ. 24,
3(2006):661–700.

——. America’s Dropout Problem: The GED and the Importance of Social and Emotional Skills. Chicago:
University of Chicago Press, Forthcoming.

Heckman, J.J., R.J. LaLonde, and J.A. Smith. “The Economics and Econometrics of Active Labor
Market Programs.” In Handbook of Labor Economics, Volume 3A, Chapter 31, O. Ashenfelter and

D. Card, eds., pp. 1865–2097. New York: North-Holland, 1999.
Heckman, J. J., L.J. Lochner, J. Smith, and C. Taber. “The Effects of Government Policy on Human
Capital Investment and Wage Inequality.” Chicago Policy Rev. 1, 2(1997):1–40.
Heckman, J.J., L.J. Lochner, and P.E. Todd. “Earnings Equations and Rates of Return: The Mincer
Equation and Beyond.” In Handbook of the Economics of Education, E.A. Hanushek and F. Welch,
eds., pp. 307–458. Amsterdam: North-Holland, 2006.

Heckman, J.J., J. Stixrud, and S. Urzua. “The Effects of Cognitive and Noncognitive Abilities on
Labor Market Outcomes and Social Behavior.” J. of Labor Econ. 24, 3(2006):411–82.

International Adult Literacy Survey. International Adult Literacy Survey Microdata User’s Guide.
Ottowa: Statistics Canada, 2002.

Johnson, D.L. “Primary Prevention of Behavior Problems in Young Children: The Houston
Parent-Child Development Center”.In 14 Ounces of Prevention: A Casebook for Practitioners,
Chapter 4, E. Cowen, R. Price, J. Ramos-McKay, and R. Lorion, eds., pp. 44–52. Washington DC:
American Psychological Association, 1988.

Jorgenson, D.W., M.S. Ho, and K.J. Stiroh. “Growth of US Industries and Investments in Information
Technology and Higher Education.” In Measuring Capital in the New Economy, C. Corrado, J. C.
Haltiwanger, and D. E. Sichel, eds.. Chicago: University of Chicago Press, 2003.

Katz, L. F., and D. H. Autor. “Changes in the Wage Structure and Earnings Inequality.” In Handbook
of Labor Economics, Volume 3, Chapter 25, O. Ashenfelter and D. Card, eds., pp. 1463–1555. New
York: North-Holland, 1999.

Knudsen, E.I., J.J. Heckman, J. Cameron, and J.P. Shonkoff. “Economic, Neurobiological, and
Behavioral Perspectives on Building America’s Future Workforce.” Proc. Natl. Acad. Sci. 103,
27(2006):10155–62.

Krueger, A.B. “Economic Considerations and Class Size.” Econ. J. 113, 485(2003):F34–F63.

Lally, J.R., P.L. Manggione, and A.S. Honig. “Syracuse University Family Development Research
Program: Long-range Impact of an Early Intervention with Low-income Children and Their
Families.” In Parent Education as Early Childhood Intervention: Emerging Directions in Theory,
Research, and Practice, D. R. Powell, ed., pp. 79–104. Norwood, NJ: Ablex Publishing Company,
1988.

Proceedings 493

Levitt, S. “Using Electoral Cycles in Police Hiring to Estimate the Effect of Police on Crime.” Amer.
Econ. Rev. 87, 3(1997):270–90.

Lochner, L.J., and E. Moretti. “The Effect of Education on Crime: Evidence from Prison Inmates,
Arrests, and Self-reports.” Amer. Econ. Rev. 94, 1(2004):155–89.

Lynch, J.P. Profile of Inmates in the United States and in England and Wales, 1991. Washington DC: U.S.
Department of Justice, Office of Justice Programs, Bureau of Justice Statistics, 1994.

Machin, S., and J. Van Reenen. “Technology and Changes in Skill Structure: Evidence from Seven
OECD Countries.” Quart. J. Econ. 113, 4(1998):1215–44.

Martin, J.P., and D. Grubb. “What Works and for Whom: A Review of OECD Countries’ Experiences
with Active Labour Market Policies.” Swedish Econ. Policy Rev. 8, 2(2001):9–56.

Mayer, S.E. What Money Can’t Buy: Family Income and Children’s Life Chances. Cambridge, MA:
Harvard University Press, 1997.

Mazumder, B. “The Apple Falls Even Closer to the Tree Than We Thought: New and Revised
Estimates of the Intergenerational Inheritance of Earnings.” In Unequal Chances: Family
Background and Economic Success, S. Bowles, H. Gintis, and M. Osbornedr Groves, eds., pp.
80–99. Princeton, NJ: Princeton University Press, 2005.

McLanahan, S., and G.D. Sandefur. Growing Up with a Single Parent: What Hurts, What Helps.
Cambridge, MA: Harvard University Press, 1994.

Miller, T.R., M.A. Cohen, and B. Wiersema. Victim Costs and Consequences: A New Look. Washington
DC: U.S. Dept. of Justice, Office of Justice Programs, National Institute of Justice, 1996.

Moynihan, D.P. Employment, Income, and the Ordeal of the Negro Family. Indianapolis: Bobbs-Merrill
Company, 1965.

Ramey, C.T., D.M. Bryant, B.H. Wasik, J.J. Sparling, and F.A. Campbell. “Early Intervention for
High-risk Children: The Carolina Early Intervention Program.” In 14 Ounces of Prevention: A
Casebook for Practitioners, R.H. Price, ed. Washington DC: American Psychological Association,
1988.

Reynolds, A.J., J.A. Temple, D.L. Robertson, and E.A. Mann. “Long-term Effects of an Early
Childhood Intervention on Educational Achievement and Juvenile Arrest—A 15-year
Follow-up of Low-income Children in Public Schools.” J. Amer. Med. Assoc. 285,
18(2001):2339–46.

Rolnick, A., and R. Grunewald. “Early Childhood Development: Economic Development with a
High Public Return.” Technical report, Federal Reserve Bank of Minneapolis, Minneapolis, MN,
2003.

Schweinhart, L.J., H.V. Barnes, and D. Weikart. Significant Benefits: The High-Scope Perry Preschool
Study Through Age 27. Ypsilanti, MI: High/Scope Press, 1993.

Schweinhart, L.J., J. Montie, Z. Xiang, W.S. Barnett, C.R. Belfield, and M. Nores. Lifetime Effects: The
High/Scope Perry Preschool Study through Age 40. Ypsilanti, MI: High/Scope Press, 2005.

Seitz, V. “Intervention Programs for Impoverished Children: A Comparison of Educational and
Family Support Models. In Annals of Child Development: A Research Annual, R. Vasta, ed., Volume
7, pp. 73–103. London: Jessica Kingsley, 1990.

Shonkoff, J.P., and D. Phillips. From Neurons to Neighborhoods: The Science of Early Child Development.
Washington DC: National Academy Press, 2000.

Stecher, B.M., and G.W. Bohrnstedt. Class Size Reduction in California: Summary of the 1998–99
Evaluation Findings. Sacramento, CA: California Department of Education, 2000.

Stephan, J.J. State Prison Expenditures, 1996. Washington DC: U.S. Dept. of Justice, Office of Justice
Programs, Bureau of Justice Statistics, 1999.

Trennert, R.A. The Phoenix Indian School: Forced Assimilation in Arizona, 1891–1935. Norman, OK:
University of Oklahoma Press, 1988.

Ventura, S.J., and C.A. Bachrach. “Nonmarital Childbearing in the United States, 1940–99.” National
Vital Statistics Reports 48, 16(2000).

Leave a Reply

You must be logged in to post a comment.