How Much Does Childhood Poverty Affect the Life Chances of Children?

Author(s): Greg J. Duncan, W. Jean Yeung, Jeanne Brooks-Gunn and Judith R. Smith
Source: American Sociological Review, Vol. 63, No. 3 (Jun., 1998), pp. 406-423
Published by: American Sociological Association
Stable URL: https://www.jstor.org/stable/2657556
Accessed: 23-01-2019 14:21 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

American Sociological Association is collaborating with JSTOR to digitize, preserve and

extend access to American Sociological Review

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
 HOW MUCH DOES CHILDHOOD POVERTY AFFECT THE
LIFE CHANCES OF CHILDREN?*

Greg J. Duncan W. Jean Yeung

Northwestern University University of Michigan

Jeanne Brooks-Gunn Judith R. Smith

Columbia University Fordham University

Why parental socioeconomic status correlates strongly with various mea-

sures of child and adult achievement is an important and controversial re-
search question. After summarizing findings from recent contributions to this
literature, we conduct two sets of analyses using data from the Panel Study
of Income Dynamics. Completed schooling and nonmarital childbearing are
related to parental income during early and middle childhood, as well as
during adolescence. These analyses suggest that family economic conditions
in early childhood have the greatest impact on achievement, especially
among children in families with low incomes. Estimates from sibling models
support the hypothesis that economic conditions in early childhood are im-
portant determinants of completed schooling.

Poverty rates among U.S. children are The implications of these alarming poverty
one-third higher than they were two de- figures for America's children remain in dis-
cades ago and 1.5 to 4 times as high as the pute. There is little doubt that children raised
rates for children in Canada and Western Eu- in poverty have less enjoyable childhoods.
rope (Rainwater and Smeeding 1995). In But to what extent does poverty adversely
1995, some 15.3 million children lived in affect cognitive and behavioral development
families in which total income failed to ex- and thereby reduce opportunities for success
ceed even the Spartan thresholds (e.g., and happiness in adulthood? Securing an-
$12,158 for a family of three) used to define swers to this important question is difficult
poverty (U.S. Bureau of the Census 1996). for a variety of reasons (Brooks-Gunn and
Duncan 1997; Mayer 1997).
* Direct all correspondence to Greg J. Duncan, First and foremost, past research linking
Institute for Policy Research, Northwestern Uni- economic disadvantage and child develop-
versity, 2040 Sheridan Road, Evanston, IL 60208 ment has rarely incorporated the careful mea-
(greg-duncan@nwu.edu). We thank the National
surement of economic deprivation. Unless
Institute of Child Health and Human Devel-
the data contain reliable measures of both
opment's Family and Child Well-Being Research
family income and correlated aspects of pa-
Network, the Russell Sage Foundation, the W. T.
rental socioeconomic status, it is impossible
Grant Foundation, and the Canadian Institute for
Advanced Research for supporting this research. to estimate the separate contributions of each.
For their many helpful comments we thank Mary Income and social class are far from syn-
Corcoran, Aletha Huston, Robert Michael, Susan onymous. Events like divorce and unemploy-
Mayer, Rob Mare, Christopher Jencks, Terry ment can alter permanently a family's eco-
Adams, George Cave, and Rebecca Blank, and nomic and social position. Because family
participants in the conference "Growing Up incomes are surprisingly volatile (Duncan
Poor" and at seminars at Northwestern Univer-
1988), the relatively modest correlations be-
sity, the University of Michigan, the University
tween economic deprivation and typical
of Montreal, Vanderbilt University, the General
measures of socioeconomic background en-
Accounting Office, the Manpower Demonstration
Research Corporation, the University of Paris able researchers to distinguish statistically
and the University of North Carolina at Chapel between the effects on children's develop-
Hill. ment of income poverty and those of its cor-

406 American Sociological Review, 1998, Vol. 63 (June:406-423)

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
 CHILD POVERTY AND EARLY-ADULT SUCCESS 407

related events and conditions (Hill and the omission of typically unmeasured factors
Duncan 1987; Sewell and Hauser 1975).1 such as parental ability, mental health, or al-
The distinction is crucial, both conceptu- truism in putting the needs of their children's
ally and for public policy reasons. Programs development before their own needs.
that alter family income (e.g., time limits on Third, although this work has provided a
welfare-program benefits, the Earned In- rough guide to the magnitude of the income
come Tax Credit, the minimum wage) are of- effect, it has not revealed the processes by
ten easier to design and administer than pro- which economic conditions affect children.
grams aimed at other family characteristics If, for example, income is important because
(e.g., promoting school completion of the it enables families to provide richer learning
mother or labor-force involvement of men; environments for their children, then policies
reducing out-of-wedlock childbearing). that enrich learning environments directly
Fortunately, several data sets containing might be more efficient in meeting child-de-
reliable longitudinal measures of family in- velopment goals than would a more general
come, socioeconomic status, and children's redistribution of income.
developmental outcomes have become avail- We use whole-childhood data from the
able in the past decade. Much of the work to Panel Study of Income Dynamics (PSID) to
date using these data has estimated "reduced- relate children's completed schooling and
form" models relating outcomes to income nonmarital fertility to parental income dur-
and other components of socioeconomic sta- ing middle childhood, adolescence, and, for
tus and has left unanswered many important the first time, very early childhood. Our
questions. analyses use both individual-based models
First, little is known about the importance and models based on sibling differences in
of the timing of economic deprivation during schooling and parental income.
childhood. Studies of children's early cogni-
tive and physical development suggest that
BACKGROUND
family income in the first five years of life is
a powerful correlate of developmental out- Several recent review articles (Corcoran
comes in early and middle childhood 1995; Haveman and Wolfe 1995) and books
(Duncan, Brooks-Gunn, and Klebanov 1994; (Mayer 1997) summarize the voluminous lit-
Miller and Korenman 1994; Smith, Brooks- erature linking family income and develop-
Gunn, and Klebanov 1997). Similar studies mental outcomes in adolescence and early
focusing on adolescent outcomes such as adulthood. The consensus is that: (1) the ef-
completed schooling and out-of-wedlock fects of parental income vary from one out-
childbearing tend to find much weaker ef- come to another; (2) for achievement-related
fects of income (Haveman and Wolfe 1995). outcomes such as completed schooling and
Yet because the adolescent-based studies early-adult labor market success the esti-
rarely have measures of parental-family in- mated effects of parental income are usually
come from early childhood, it is not known statistically significant, but there is little con-
whether poverty early in childhood has note- sensus regarding the size of these effects;
worthy effects on later outcomes. and (3) by not attending to the confounding
Second, little of this research has employed effects of unmeasured parental and neighbor-
techniques to eliminate biases associated with hood characteristics, even the mostly modest
estimates of the effects of parental income
1 For example, the modest correlations between
may be upwardly biased.
income and other measures of parental socioeco- The comprehensive review by Haveman
nomic status enabled Sewell and Hauser (1975) and Wolfe (1995) illustrates the first two of
to conclude, "There can be little doubt that the
these points:
association of socioeconomic background vari-
ables with son's earnings is due solely to the With but one exception. . . , the family income
intergenerational effect of parents' income, while variable is positively associated with the edu-
the latter cannot to any large extent be explained cational attainment of the child, and the vari-
by the differing abilities, educational attainments, able is statistically significant in more than half
or occupational achievements of the sons of rich of all cases where a positive relationship is es-
and poor families" (p. 84). timated. Simulated changes in family economic

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
 408 AMERICAN SOCIOLOGICAL REVIEW

resources, however, are associated with rela- ment than were economic conditions during
tively small changes in educational attain- adolescence. In fact, none of the achieve-
ments. The range of elasticities is wide-about ment studies using exclusively adolescence-
.02 to .2. (P. 1856)
based income measures found large effects.
In contrast, all of the studies of ability with
With respect to its relationship to out-of- income measured during early childhood
wedlock childbearing, found large effects.2 Left unanswered in
... parental income is negative and usually, these and all other analyses is the importance
but not always, significant .... The few re- for adolescent and early-adult outcomes of
ports of the quantitative effects of simulated family economic conditions in the earliest
changes in variables suggest that decreases in stages of childhood.3
parental income. .. will lead to small increases Smith et al. (1997) provide a useful set of
in the probability that teen girls will experience benchmarks for the sizes of the effects of in-
a nonmarital birth. (Haveman and Wolfe 1995:
come on ability and achievement in early
1863)
childhood. They draw data on parental socio-
economic status and ability and achievement
measures from the National Longitudinal
Recent Research
Survey of Youth and the Infant Health and
More recent contributions to this literature Development Program. All of the tests were
include a coordinated analysis by 12 groups independently normed with means of 100
of researchers working with 10 different de- and standard deviations of around 15. To al-
velopmental data sets, most of which offer low for a nonlinear relationship between in-
longitudinal measurement of parental family come and achievement, Smith et al. use re-
income as well as measurements of the gressions in which family income between
achievement, behavior, or health of individu- birth and the time of the test (adjusted for
als at various points in life (Duncan and family size) is represented as a series of
Brooks-Gunn 1997). Some outcomes, such dummy variables and that also control for
as IQ at age 2 and motor development be- differences in the child's race, birth weight,
tween birth and age 3, were measured in the age, and gender, as well as for the mother's
first years of a child's life. Others, such as education and family structure.
career attainment and mortality, were mea- When compared with children in families
sured as late as the sixth decade of life. with incomes between 1.5 and 2.0 times the
A common element across these studies is poverty line, children in families with in-
a "replication" analysis in which the same comes less than one-half of the poverty line
measures-family income, maternal school- were found to score between 6 and 13 points
ing, family structure-were included in a re- lower on the various standardized tests. In all
gression model predicting child and adult cases, these differences were statistically sig-
outcomes. Taken as a whole, the results sug-
gest that family income at times had large
2 Income effects were considered to be "large"
but rather selective effects on children's at-
if the regression-adjusted changes in the depen-
tainments. Most noteworthy was the impor-
dent variable associated with substantial income
tance of the type of outcome being consid-
changes-(1) an additional $10,000 of income,
ered. Family income had its largest correla- (2) an increase in family income from below the
tions with children's ability and achievement poverty line to between the poverty line and twice
measures. In contrast, virtually none of the the poverty line, and/or (3) a change from persis-
behavior, mental health, or physical health tent poverty to no poverty-amounted to at least
measures represented by the 12 developmen- one-quarter of a standard deviation for most of
tal studies were predicted strongly by family the dependent variables used in a particular
analysis.
income.
3 Haveman, Wolfe, and Spaulding (1991) esti-
Second, the childhood stage at which in-
mated the effects of a combined poverty and wel-
come was measured was clearly significant.
fare measure averaged over ages 4 through 7.
Family economic conditions in early and Haveman and Wolfe (1994) estimated a stage-
middle childhood appeared to be far more specific model of the effects of poverty alone, but
important for shaping ability and achieve- the earliest measurement of it is at child's age 6.

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
 CHILD POVERTY AND EARLY-ADULT SUCCESS 409

nificant. Children in families with incomes and one-third of the effects of income on the
closer to but still below the poverty line also achievement scores of elementary school
did worse than children in the higher income children. Thus, in the case of the cognitive
reference group; the differences were development of preschoolers, income mat-
smaller, although usually, but not always, ters to a substantial degree because it is as-
statistically significant. The smallest differ- sociated with a richer learning environment
ences appeared for the earliest (age 2) mea- for the children.
sure of cognitive ability, although there was Other studies have found evidence that low
no monotonic increase across the data in the income produces economic pressures that
estimated effect of poverty with the age of lead to conflict between parents over finan-
the child. Also noteworthy is the fact that the cial matters (Conger, Conger, and Elder
associations between family poverty and 1997; Conger et al. 1992, 1993). This, in
cognitive ability appear to be just as large for turn, increases the harshness of the mother's
full-scale IQ measures as for the reading and parenting and undermines the adolescent's
math achievement tests. These findings are self-confidence and achievement. Specifi-
consistent with the hypothesis that increas- cally, a family's income level is a powerful
ing the incomes of children whose family in- predictor of the reported economic pressure
comes are below or near the poverty line will felt by family members. Economic pressure
have a larger impact on early-childhood abil- has both direct and indirect effects on ado-
ity and achievement than would increasing lescent achievement. Parental financial con-
the incomes of children in middle-class and flicts were particularly detrimental to the
affluent families.4 self-confidence and achievement of boys.
Some research has attempted to explain
why economic conditions appear to affect
Are the Income "Effects" Causal?
achievement. Consistent with a number of
other studies, Smith et al. (1997) find that Much of the existing empirical literature
the quality of the home environment-its consists of regressions relating developmen-
opportunities for learning, the warmth of tal outcomes to parental income and a mod-
mother-child interactions, and the physical est set of socioeconomic and demographic
condition of the home-accounts for a sub- control variables. As such, they show the as-
stantial portion of the powerful effects of sociations between parental income and vari-
family income on cognitive outcomes. Spe- ous outcomes for children, after the regres-
cifically, differences in the home environ- sion techniques adjust statistically for mea-
ments of high- and low-income children ex- sured socioeconomic and demographic dif-
plained close to one-half of the effects of in- ferences between high- and low-income
come on the cognitive development of pre- families.
school children and between one-quarter A persistent concern with these kinds of
analyses is that the estimated effect of in-
4 Smith et al.(1997) show somewhat larger ef- come might be spurious, caused by the mu-
fects than those found in some of the other stud-
tual association that parental income and the
ies using the National Longitudinal Survey of
outcomes for children share with some un-
Youth. Blau (1995) summarizes much of this lit-
erature with calculations of cognitive test score
measured "true" causal factor. Suppose, for
changes associated with a $10,000 increase in example, that the mental health of parents is
family income. Typical of the estimated effects the key ingredient for children's success and
are those of Korenman and Miller (1994), who that measures of parental mental health
report that a $10,000 increase in permanent in- were not included in the models. Because
come is associated with one-fifth of a standard positive mental health in parents is likely to
deviation in outcomes when income is initially
make parents more successful in the labor
less than one-half the poverty line, but less than
market as well as to lead to fewer problems
one-tenth of a standard deviation when initial in-
with their children, the absence of adjust-
come is well above the poverty line. Blau's analy-
ments for differences in parental mental
sis shows how much more responsive the test
scores are to long-run income than to income health may produce a serious overstatement
measured in a single year near the time the test of the role income plays in causing
was administered. children's success.

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
 410 AMERICAN SOCIOLOGICAL REVIEW

Randomized experiments constitute one well as PARSCHOOL) and gives the following
solution to this omitted-variables problem. equation relating change in cognitive ability
The negative-income-tax experiments con- to the total income between t and t + 5:
ducted in the 1970s provide inconclusive
evidence on the effects of experimental in- AACHtUt + 5 = P1I AlINCOMEt, + 5

creases in income on children's outcomes

+ A Et~t + 5, (3)
(Currie 1995). Substantial income effects where A Xtj + 5 indicates the chang
were found on children's nutrition, early able from year t to year t + 5.
school achievement, and high school In their analysis of the effects of persistent
completion in some sites, but not in others. poverty on IQ at age 5 and behavior prob-
Because the site with the largest effects for lems, Duncan et al. (1994) estimate such an
younger children (North Carolina) was also equation based on change data between ages
the poorest, one interpretation of the results 3 and 5 and find highly significant effects of
is that income effects are largest for the poor- parental income between children's ages 3
est families. and 5 on changes in IQ between ages 3 and
To illustrate nonexperimental solutions to 5. Results for the estimated effects of income
the problem of omitted-variable bias, con- on changes in behavior problems were in the
sider a simple model in which achievement expected direction, but were not significant
at time t (AcHt) is a function of lifetimeatin-
conventional levels.
come up to point t (YINcoMEt), a permanent Change models estimated on nonexperi-
and observed component of other aspects of mental data are not without their problems,
parental background (PARSCHOOL), an unob- as one still must worry about the source of
served permanent family-specific component the changes in the right-hand-side variables
(FAM), an unobserved permanent individual (Heckman and Robb 1985). In the context
component (IND), and a random error term of developmental changes, one needs to
(-Et): make sure that the motivations, conditions,
and events causing the income change either
ACHt = a + PI I INCOMEt + J32PARSCHOOL did not affect development directly or are
somehow controlled for in the statistical
+ 03FAM + 4IND + E. (1)
analysis.
Time-varying measures of family conditions Another model-based approach is to esti-
(e.g., maternal employment) could be added mate a level equation like equation 1, but to
to this model as well, although this raises is- attempt to remove the spurious correlation
sues of whether such conditions are jointly between income and development through an
determined as part of the process by which instrumental-variables procedure. This pro-
families develop a strategy for having and cedure amounts to replacing the lifetime in-
raising their children (Blau 1995). come variable (JINCOME) with an instrumen-
Much of the recent work relating family tal variable that is purged of YINCOME'S SpU-
income to developmental outcomes is based rious correlation with unobserved factors
on estimating a version of this equation that such as family (FAM) and child's achieve-
omits and fails to adjust otherwise for the ef- ment (ACH). The trick is to find a variable
fects of the unmeasured family and indi- that is highly correlated with 2INCOME but is
vidual variables. One way around this omit- not highly correlated with the unobservable
ted-variables problem is to estimate change components of family (FAM) and individual
models. If the relationship in equation 1 (IND). This task is difficult because almost all
holds, say, five years later, at t + 5, then we correlates of JINCoME are arguably correlates
have: of unobserved determinants of children's de-
velopment as well.
ACHt + 5= a+ PlI IINCOMEt + 5 Mayer (1997) provides a set of tests for
+ 032PARSCHOOL + J 3FAM omitted-variable bias, including the addition
+f 4IND + Et + 5 (2) of measures of parental income after the oc-
currence of the outcome as well as only those
Differencing these two equations eliminates components of parental income that are
the confounding effects of FAM and IND (as fairly independent of the actions of the fam-

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
CHILD POVERTY AND EARLY-ADULT SUCCESS 411

ily. In the first case, she argues that future Key to the estimation of this formulation
income cannot have caused the prior out- is sufficient variability between siblings in
come, so that its inclusion adjusts for unmea- their family income histories between birth
sured characteristics of the parents. The ad- and the point of measurement of AcH-a con-
dition of future income almost always pro- dition that is obviously not met in the case
duces a large reduction in the estimated ef- of twins, but is met in the case of nontwin
fect of prior parental income; thus she con- siblings. If, as seems reasonable, sibling dif-
cludes that much of the estimated effect of ferences in the unobserved individual com-
income from replication models is spurious. ponent (IND) are largely independent of in-
In the second case, the argument is that the come differences, then estimating equation 6
level of income components such as welfare with sibling data produces estimates of in-
and earnings (as well as the children's out- come effects that are largely free from the
comes under study) may reflect the effects of confounding effects of unobserved family
important unmeasured parental characteris- characteristics.
tics. If components such as asset income are
less affected by these unmeasured parental
DATA
characteristics, their coefficients ought to
provide a better gauge of "true" income ef- Data for our analysis of these issues come
fects. Following this procedure, Mayer finds from the Panel Study of Income Dynamics
small and often nonsignificant coefficients (PSID), a longitudinal survey of U.S. house-
for these income components. holds. Since 1968 the PSID has followed, in-
As Mayer points out, these procedures are terviewed annually, processed, analyzed, and
not without their problems. If families antici- disseminated information from a representa-
pate future income changes and adjust their tive sample of about 5,000 families (Hill
consumption accordingly, and the consump- 1992). Splitoff families are formed when
tion changes benefit or hurt children, then children leave home, when couples divorce,
future income does indeed play a causal role. and when more complicated changes break
The likely measurement error in income families apart. This procedure produces an
sources such as dividends and interest will unbiased sample of families each year as
impart a downward bias in their coefficients. well as a continuously representative sample
Moreover, interest and dividends are almost of children born into families each year.
universally absent from the income packages The PSID's original design focused on
of families at or below the poverty line. poverty by oversampling low-income and
Another approach to eliminating bias is to minority households. Weights have been cre-
use sibling differences. Suppose the relation- ated and are used here to adjust for the origi-
ship in equation 1 holds for two siblings, A nal oversampling of the poor and for differ-
and B, within the same family, and that the ential attrition.5
measured and unmeasured characteristics of Our individual-based analyses use the
the family do not change from one sibling to sample of 1,323 children born between 1967
the next and: and 1973 and present in the PSID between
birth and age 20. Our sibling analyses are
ACHA= ao+ PIlSINCOMEA + /32PARSCHOOL based on the 328 sibling pairs drawn from
+ /33FAM + J,4INDA + EA; (4) the individual-based sample. Given the co-
hort range chosen, siblings cannot be more
than six years apart in age. Barring non-
ACHB = a + PIE Y2INCOMEB + /2 PARSCHOOL
+ /33FAM + J4INDb + Eb; (5) response problems that are not corrected by

Differencing across sibling pairs eliminates

5 For completed schooling, we use the indi-
the FAM and PARSCHOOL components and
vidual weight associated with the interview year
leaves:
in which the schooling was reported. For

AcHA-ACHB = nonmarital births, we use the individual weight

associated with the interview year in which the
,I3 (E INCOMEA - Y2INCOMEB) + most recent marital and fertility histories were re-
134 (INDA - INDB) + (eA - SB)- (6) ported.

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
 412 AMERICAN SOCIOLOGICAL REVIEW

Table 1. Intercorrelations, Means, and Standard Deviations for the Family Income Variables: Panel
Study of Income Dynamics

Correlation Coefficients (Zero-Order)

Income Variable (1) (2) (3) (4) Mean S.D.

(1) Family income at ages 0 to 5 1.00 3.70 2.10

(2) Family income at ages 6 to 10 .82 1.00 4.56 3.07

(3) Family income at ages 11 to 15 .72 .87 1.00 5.20 4.14

(4) Family income at ages 0 to 15 .87 .96 .95 1.00 4.49 2.91

Note: Family income is in $10,000s (1993 dollars).

our weighting adjustments, the experiences Our analyses include control variables for
of this group of children ought to be nation- race, gender, number of siblings, the com-
ally representative of the cohorts from which pleted schooling of the mother, the age of
they were sampled. The sibling sample rep- the mother at the time of the child's birth,
resents sibling pairs drawn from these co- whether the family ever lived in the South,
horts, but not the more general set of chil- family structure, maternal employment, and
dren in these cohorts. These data enable us residential mobility. Our family structure
to test for the relative importance of family measures are a series of dummy variables
income in early and middle childhood as indicating whether the child was born into a
well as adolescence in explaining two impor- nonintact family, and stage-specific mea-
tant outcomes-years of completed school- sures of whether the child's parents experi-
ing and the timing of a first nonmarital birth. enced a divorce or remarriage. Maternal em-
Most of our analyses use measures of ployment is captured by stage-specific mea-
schooling and fertility ascertained as recently sures of the number of years in which the
as possible in the PSID. This is typically at mother worked 1,000 or more hours. Resi-
age 25 or later-earlier only in the cases of dential mobility is measured with stage-spe-
individuals who were lost to attrition be- cific counts of the number of years in which
tween the year they turned 20 and the 1995 the family reported a residential move. In
interviewing wave. Our event-history analy- the case of stage-specific analyses, the vari-
sis of nonmarital fertility begins at age 16 ables are measured over three age ranges:
and is censored by attrition from the study, a birth to age 5, ages 6 to 10, and ages 11 to
marital birth, or the termination of a first 15.
marriage into which no children were born.
Our income measure is the total pretax in-
come of all family members, inflated to 1993 RESULTS
price levels using the Consumer Price Index
Income Correlations across Childhood
(CPI-UX1) and averaged over all the years
of childhood or over all the years within the We began our analysis with an investigation
given childhood stage under consideration. A of the nature of family income across all
common practice in studies like ours is to use childhood stages. Table 1 shows that the av-
a size-adjusted measure of family income, erage family income increases substantially
typically the "income-to-needs" ratio, ob- across childhood. Over the entire sample, in-
tained by dividing total household income by come averaged across ages 11 to 15 is some
the official U.S. poverty threshold corre- 40 percent higher than income averaged
sponding to the size of the given household. across ages 0 to 5. Zero-order correlations of
A disadvantage of this formulation is that the five-year average family incomes across the
ratio imposes restrictions on the size of the childhood stages are high-.82 and .87 for
separate effects of income and family size. adjacent stages and .72 for average family
In our analyses, we include income and fam- incomes between child's birth and age 5 and
ily size as separate variables. between ages 11 to 15.

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
 CHILD POVERTY AND EARLY-ADULT SUCCESS 413

Table 2. Percentage Distribution of Family Income at Child's Ages 11 to 15 by Family Income at

Child's Ages 0 to 5: Panel Study of Income Dynamics

Family Income at Ages 0 to 5

Percent Percent Percent Percent

Less than $15,000 to $25,000 to $35,000
Family Income at Ages 11 to 15 $15,000 $24,999 $34,999 or more

Less than $15,000 39.0 15.1 2.2 1.4

$15,000 to $24,999 33.2 31.4 10.7 3.2

$25,000 to $34,999 15.7 17.7 21.7 7.7

$35,000 or more 12.1 35.9 65.4 87.8

Total 100.0 100.0 100.0 100.0

Note: Family income is in 1993 dollars.

Despite what appear to be high correla- schooling, age of the mother at the time of
tions, there was considerable movement of the child's birth, whether the child ever lived
families across income classes. Table 2 cross- in the South, family structure, maternal em-
classifies family incomes averaged across ployment, and residential mobility. Descrip-
child's ages 0 to 5 and 11 to 15. Only a mi- tive statistics and estimated coefficients for
nority (39.0 percent) of children with family these control variables from a subset of the
incomes below $15,000 in early childhood models are presented in Appendix A.
still had incomes that low in adolescence, and Model 1 for each dependent variable re-
more than one-quarter (27.8 percent) of the ports the coefficient and standard error on
initially low-income children had incomes in average annual family income in linear form
adolescence that were $25,000 or more. and scaled in $10,000s, 1993 dollars. As with
Year-to-year income changes also produce past studies, income has a statistically sig-
considerable differences in the income expe- nificant but substantively modest impact on
riences of siblings (data not shown). Roughly the outcome variables. An additional $10,000
one-fifth of the sibling pairs in our sibling of family income is associated with a .14-
sample had average family incomes between year increase in years of schooling com-
birth and age 5 that differed by more than pleted, a 26-percent (i.e., e 23) increase in the
$5,000, while roughly one-quarter experi- odds of completing high school, and a 35-
enced income differences that large in the percent (i.e., 1 - e-43) drop in the relative
second and third childhood stages. When risk of a first nonmarital birth.
taking childhood as a whole, nearly one-half In Model 2 we allowed the effect of income
of the siblings had 15-year average incomes to vary with the level of income by fitting a
that differed by more than $5,000. two-segment spline function, with separate
slopes for children in families with average
total incomes under and over $20,000. The
Whole-Childhood Income Effects
first coefficient represents the estimated
Table 3 presents results from regressions fit- slope (with income scaled in $10,000s) for
ting various functional forms for average to- the under-$20,000 group, and the second co-
tal family income at child's ages 0 to 15: (1) efficient represents the difference in slope be-
OLS models predicting years of completed tween the over-$20,000 and under-$20,000
schooling; (2) logistic models predicting the groups. This nonlinear form clearly fits the
successful completion of high school; and schooling data better, with much bigger esti-
(3) Cox models of the timing of a first mated impacts for income increments for
nonmarital birth. Control variables common low-income than middle-income and high-
to all the regressions are: the child's race and income families. In the case of the nonmarital
sex, total number of siblings, whether the fertility model, the log-likelihoods for the lin-
family head was black, maternal years of ear and spline models are identical. For chil-

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
 414 AMERICAN SOCIOLOGICAL REVIEW

Table 3. Coefficients from the Regression of Child's Outcome Variables on Family Income at Ages 0
to 15: Panel Study of Income Dynamics

Dependent Variables/Models

Years of High School Hazard of

Completed Schoolinga Completion b Nonmarital Birth c

Independent Variable (1) (2) (3) (4) (1) (2) (3) (4) (1) (2) (3) (4)

Family Income at Child's Ages 0 to 15

Linear function .14* - - .23* - -.43* -
(.02) (.07) (.10)

Spline function
Income < $20,000 - 1.30* - 1.97* - - -.50 -
(.29) (.44) (.41)
Difference between - -1.17* - -1.84* - - -.08 -
income < $20,000 (.30) (.46) (.44)
and > $20,000

Natural logarithm 1.16* 1.35* - -1.18* -

(.11) (.26) (.26)

Dummy Variables for Family Income

$15,000 to $24,999 .82* 1.41* -.54
(.27) (.38) (.35)
$25,000 to $34,999 1.41* 1.83* -.94
(.28) (.43) (.41)
$35,000 to $49,999 1.69* 2.48* - -1.44*
(.28) (.45) (.43)
$50,000 and over 2.35* 2.64* - -2.40*
(.29) (.49) (.54)

Adjusted R2 .192 .201 .219 .216

-2 Log likelihood 718.9 702.6 701.1 694.6 1,266.1 1,266.1 1,271.1 1,267.3

Note: Numbers in parentheses are standard errors. In Model 4, the omitted category for family income is
"less than $15,000." The mean years of schooling completed was 13.5 (S.D. = 2.1); the mean rate of high
school completion was .90 (S.D. = .30).
a OLS models; N = 1,323.
bLogistic models; N = 1,323.
Cox models; N = 620.

*p < .05 (two-tailed tests)

dren in low-income families, a $10,000 in- The spline for high school completion also
crease in family-income is associated with indicates a much larger incremental effect-
1.3 years of additional schooling, an effect a seven-fold increase in the odds of graduat-
that is nearly 10 times as large as the esti-
mated impact from the linear form. Income presented in Model 1. To investigate the bias in
increments for children in high-income fami- studies based on family income measured only in

lies have a significantly smaller impact- adolescence, we estimate completed-schooling

models using the linear and spline functions and
only .13 (1.30 - 1.17) additional years of
family income averaged between ages 11 and 15,
schooling per $10,000 income increment.6
the same demographic controls but no other in-
come-related measures. We found that the linear
6 Although the .13 difference is small relative
effect was 64 percent (.09/.14) as large for the
to the standard errors of the spline coefficients, 11-15 age period versus the 0-15 age period. The
its significance is better judged relative to the coefficient on the first spline segment was 65 per-
standard error (.02) of the linear income measure cent as large (.85/1.30).

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
 CHILD POVERTY AND EARLY-ADULT SUCCESS 415

ing per $10,000 increment-for low-income set of regressions that included measures of
as compared to high-income children. In family income averaged over the first, sec-
contrast to the school-related outcomes, the ond, and third five-year segments of the
hypothesis of a linear effect of income can- children's lives (Table 4). In all other re-
not be rejected for nonmarital childbearing. spects, the regression models are identical to
Model 3 indicates that the pattern of di- the whole-childhood regressions presented in
minishing returns to increments in family in- Table 3. Because a given five-year average
come is approximated reasonably well with income level produces one-third the total
a logarithmic transformation of family in- childhood income of that same income level
come. In fact, the fit of the schooling models averaged over 15 years, a stage-specific
(but not the nonmartial fertility model) is model in which income was constant and tim-
better with the log form of income than with ing did not matter will produce stage-specific
the spline function. income coefficients that are roughly one-third
A disadvantage of the log form is that it the size of a whole-childhood model.
does not isolate the portion of the income Taken as a whole, the results show that
distribution producing the biggest impact on timing matters a great deal for the schooling
the dependent variable. For this reason, we outcomes; income increments early in life
also estimated a more flexible parameteriza- for children in low-income families are as-
tion of the income-outcome relationships-a sociated with large increments to completed
series of dummy variables, the results of schooling. For example, the spline model
which are presented in Model 4. Children in suggests that, controlling for income in other
families with annual incomes that averaged stages, a $10,000 increment to income aver-
less than $15,000 constitute the omitted aged over the first five years of life for chil-
group in these regressions. In contrast to dren in low-income families is associated
these low-income children, children in fami- with an increment of .81 years in completed
lies with incomes between $15,000 and schooling and an increase of 2.9 times in the
$25,000 completed .82 years more schooling odds of finishing high school. These esti-
and enjoyed 4.1 times greater odds of com- mated effects are much larger than the corre-
pleting high school, but had an insignificant sponding estimated effects of income mea-
lower risk of a nonmarital birth.7 Schooling sured between child's ages 6 to 10 and 11 to
differences between the $15,000-$24,999 15.8 The logarithmic version of the model
and $25,000-$34,999 groups were more than shows that income during adolescence has an
one-half of a year and were statistically sig- effect as powerful as income in early child-
nificant at the p < .01 level. In the case of hood for years of completed schooling. In the
completing high school, there were much case of high school graduation, parental in-
smaller improvements in the odds of gradu- come during adolescence is much less impor-
ating associated with income increases other tant, suggesting that adolescent-based paren-
than those at the very bottom of the income tal income is more important for college-re-
distribution. lated decisions (see below).
The more flexible dummy-variable version
of the model (Model 4) confirms the greater
Stage-Specific Income Effects
importance of economic conditions during
To allow for the differential impact of income the first five years of life for completed
by childhood stage, we estimated a second
8 Although larger, the coefficients for the
7 The mean incomes of children in the dummy variables for family income at child's age
<$15,000, $15,000-$24,999, $25,000-$34,999, 0 to 5 were never significantly larger than the co-
$35,000-$49,999, and >$50,000 income groups efficients for the corresponding dummy variables
were $11,403, $19,996, $30,553, $41,906, and for ages 6 to 10 and 11 to 15. However, a model
$74,739, respectively. Thus, the increment in av- that includes dummy variables for ages 0 to 5
erage income associated with membership in the family income categories and dummy variables
first two income groups was about $8,600, while for family income averaged over the 10-year pe-
the increment associated with membership in the riod between ages 6 and 15 produces significant
highest two income groups was much larger- differences between all corresponding sets of co-
about $32,800. efficients.

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
 416 AMERICAN SOCIOLOGICAL REVIEW

Table 4. Coefficients from the Regression of Child's Outcome Variables on Childhood-Stage-Specific

Family Income: Panel Study of Income Dynamics

Dependent Variables/Models

Years of High School Hazard of

Completed Schoolinga Completionb Nonmarital BirthC

Independent Variable (1) (2) (3) (4) (1) (2) (3) (4) (1) (2) (3) (4)

Family Income at Child's Ages 0 to 5

Linear function .12* .38* - -.16 -
(.05) (.13) (.14)

Spline function

Under $20,000 - .81* - 1.05* - - -.03 -

(.28) (.43) (.37)

Difference between - -.72* - -.85 - - -.16 -

income < $20,000 (.29) (.48) (.42)
and > $20,000

Natural logarithm - .54* 1.07* - - -.37 -

(.18) (.35) (.31)

Dummy variables for family income

$15,000 to $24,999 - - - .66* - .56 - - .10

(.25) (.36) (.32)
$25,000 to $34,999 - - - .73* - 1.15* - - -.26
(.28) (.44) (.40)

$35,000 to $49,999 - - - .78* - 1.58* - - -.97

(.30) (.52) (.47)

$50,000 and over - - - 1.41* - 1.53* - - -1.13

(.33) (.67) (.67)

Family Income at Child's Ages 6 to 10

Linear function -.01 - -.07 - .06 -

(.04) (.10) (.13)

Spline function

Under $20,000 - .45 .22 - - -.21

(.36) (.28) (.45)

Difference between - -.47 -.30 - - .30 -

income < $20,000 (.36) (.30) (.48)
and > $20,000

Natural logarithm - -.06 -.18 - .20 -

(.12) (.40) (.36)

Dummy variables for family income

$15,000 to $24,999 - .16 .80* -.21

(.30) (.44) (.36)

$25,000 to $34,999 - - - .24 .32 -.09

(.35) (.53) (.45)

$35,000 to $49,999 - - - .44 .36 .22

(.38) (.62) (.35)

$50,000 and over - - - .33 .32 .89

(.40) (.72) (.62)

(Table 4 continued on next page)

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
 CHILD POVERTY AND EARLY-ADULT SUCCESS 417

(Table 4 continued from previous page)

Dependent Variables/Models

Years of High School Hazard of

Completed Schoolinga Completionb Nonmarital Birthc

Independent Variable (1) (2) (3) (4) (1) (2) (3) (4) (1) (2) (3) (4)

Family Income at Child's Ages 11 to 15

Linear function .05* - - .06 - -.29* -
(.02) (.08) (.09)
Spline function

Under $20,000 .32 - .42 - -.05 -

(.27) (.24) (.38)
Difference between -.26 - -.40 - -.27 -
income < $20,000 (.27) (.26) (.41)
and > $20,000

Natural logarithm - .57* .58 - -.89* -

(.14) (.29) (.26)
Dummy variables for family income

$15,000 to $24,999 - - - .34 - - .38 - - .22

(.27) (.41) (.54)

$25,000 to $34,999 - - - .41 - - .96 - - -.16

(.29) (.49) (.40)
$35,000 to $49,999 - - - .36 - - .62 - - --1.02
(.31) (.52) (.48)

$50,000 and over - - - 1.08* - - 1.08 - - - -1.67*

(.32) (.59) (.54)

Adjusted R2 .192 .215 .220 .232

-2 Log likelihood 713.1 695.8 697.6 688.0 1,262.6 1,262.1 1,268.1 1,255.5

Note: Numbers in parentheses are standard errors. In Model 4, the omitted

"less than $15,000."
a OLS models; N = 1,323.
b Logistic models; N = 1,323.
c Cox models; N = 620.
*p < .05 (two-tailed tests)

schooling. Children with family incomes in during adolescence, we estimated logistic re-
early childhood in the $15,000-$24,999 gressions for college attendance and college
range average .66 years more schooling than completion (results not shown). The coeffi-
children in the lowest income group. In the cient on the high-income-during-adolescence
case of high school graduation, income in- dummy variable was highly significant (and
crements have similar effects across the three positive) in the college attendance model, but
lowest income categories. With the exception not in the college completion model. Thus,
of high-income adolescents, there was little the primary way in which well-to-do parents
consistent evidence of income effects on of adolescents appear to affect completed
completed schooling in other stages of child- schooling is by enabling their children to en-
hood. And with the exception of high paren- ter college.
tal income during early childhood and ado- We investigated whether the effects of fam-
lescence, stage-specific income failed to pre- ily income varied across important demo-
dict nonmarital childbearing. graphic subgroups and found little evidence
To better understand the apparent effect on that this was the case for whole-childhood
completed schooling of high parental income income. For example, the coefficient on the

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
 418 AMERICAN SOCIOLOGICAL REVIEW

log of whole-childhood family income in the tively, with standard errors in the .4 to .6
completed-schooling model was 1.16 (Table range. The increments in schooling associ-
3, Model 3). The corresponding coefficient ated with income increases from less than
for whites was 1.20; for blacks, .89; for fe- $15,000 to more than $50,000 for the three
males, 1.30; and for males, 1.04. In the stage- stages were 1.39, .01, and 1.66, respectively.
specific models of completed schooling, the
coefficients on log income associated with
Sibling Models
the three childhood stages were .54, -.06 and
.57, respectively (Table 4, Model 3). Corre- Last, we estimated a series of sibling models
sponding coefficients for whites were .40, by drawing the 328 sibling pairs from the
.00, and .77; for blacks, .96, .08, and .19; for 1,323 children used in the individual-based
females, .33, .20, and .71; and for males, .71, models (Table 5). Model 1 includes only sib-
-.18, .61. Standard errors for these coeffi- ling differences in ages 0 to 15 average fam-
cients were in the .2 to .3 range, so one should ily income and sex (same-sex siblings were
not overinterpret these differences. coded 0, female/male pairs were coded +1
In light of the fact that some of our con- and male/female pairs were coded -1). To
trol variables could be viewed as endog- adjust for important events that might have
enous, we estimated a version of our com- produced the income changes, Model 2 adds
pleted-schooling model that included a more sibling differences in age of the mother at the
limited set of predictors-the child's sex time of the birth, and stage-specific differ-
and total number of siblings, whether the ences in family structure, years of full-time
family head was black, maternal years of maternal work, and the number of residen-
schooling, age of the mother at the time of tial moves. Models 3 and 4 repeat these
the child's birth, and whether the child ever analyses but allow for differences in child-
lived in the South. The key coefficients on hood-stage-specific average income.
the stage-specific dummy variables differed Note the assumptions implicit in these sib-
only slightly from those presented in Table ling models. In particular, these models as-
4. For example, the new coefficients (com- sume that the effects of nonconstant family
pared with Table 4 coefficients in parenthe- variables are the same for each sibling, re-
ses) for the child's age 0 to 5 family income gardless of sex and position in the birth or-
categories were .70 (.66), .80 (.73), .86 der. In addition, we restrict our analysis to
(.78), and 1.52 (1.41). As with the estimates linear effects of income. Our relatively small
in Table 4, none of the coefficients for the sample sizes precluded a more complete
child's ages 6 to 10 family income catego- analysis; extensions along these lines are
ries was statistically significant at a conven- clearly warranted.9
tional level. For the child's ages 11 to 15 in- In the whole-childhood income models
come categories, only the coefficient associ- (Models 1 and 2), the estimated effect of sib-
ated with the highest income dummy vari- ling differences in income (coefficient = .22)
able was significant, with a magnitude of on differences in schooling completed was
1.18 (versus 1.08 in Table 4).
9 We attempted to fit a spline function to these
Given the complications associated with
sibling data to allow for different effects of posi-
the low-income portion of the PSID sample,
tive and negative income differences. The coeffi-
we investigated the robustness of the find- cients and standard errors on the first segments
ings by estimating (without weighting) the were similar to those presented in Table 5. The
stage-specific completed-schooling models standard errors on the second segments were too
on the 681 observations from the cross-sec- large (around .80) to provide any precision in the
tion portion of the sample. Not surprisingly, estimated coefficients. We also fitted a log model

standard errors were considerably larger, but to these sibling data. The results were similar, al-
though not as significant. In the case of the age 0
the pattern of coefficients, particularly for
to 15 family income model with covariates, the
the first stage of childhood, was similar. The
coefficient and standard error on the income vari-
increments in schooling associated with in-
able were .96 and .70, respectively. In the case of
come increases from less than $15,000 to be- the stage-specific log model, coefficients and
tween $15,000 and $25,000 for the three standard errors were .50 (.33), .16 (.36), and -.30
stages were .88, -.35, and .99 years, respec- (.37).

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
 CHILD POVERTY AND EARLY-ADULT SUCCESS 419

Table 5. Coefficients from the Regression of Years of Schooling Completed on Selected Independent
Variables: Siblings from the Panel Study of Income Dynamics

Difference between Siblings in: Model 1 Model 2 Model 3 Model 4

Family income at ages 0 to 15 .22* .20*

(.08) (.09)
Family income at ages 0 to 5 .15* .18*
(.07) (.09)
Family income at ages 6 to 10 .01 -.01
(.06) (.07)
Family income at ages 11 to 15 .06 .04
(.06) (.08)
Sex .34* .36* .33* .33*
(.11) (.11) (.11) (.11)

Age of mother at child's birth .03 .02

(.06) (.05)
Born into a nonintact family .05 .03
(.27) (.24)
Ever divorced, ages 0 to 5 -.25 -.56
(.44) (.44)
Ever divorced, ages 6 to 10 -.35 -.52
(.41) (.36)
Ever divorced, ages 11 to 15 -.48 -.68*
(.34) (.30)
Ever remarriede, ages 0 to 5 .57 .64*
(.48) (.32)
Ever remarriede, ages 6 to 10 .61 -.07
(.64) (.53)
Ever remarriede, ages 11 to 15 .36 -.53
(.38) (.33)
Years moved, ages 0 to 5 -.06 -.07
(.10) (.10)
Years moved, ages 6 to 10 -.06 .00
(.12) (.11)
Years moved, ages 11 to 15 -.15 .23*
(.12) (.12)
Years mother worked 1,000 hours or .01 -.02
more, ages 0 to 5 (.12) (.09)
Years mother worked 1,000 hours or -.08 -.09
more, ages 6 to 10 (.08) (.09)
Years mother worked 1,000 hours or -.06 -.13
more, ages 11 to 15 (.08) (.09)

Constant .37* .41* .37* .26

(.37) (.14) (.08) (.13)
Adjusted R2 .044 .038 .044 .044

Note: Numbers in parentheses are standard errors. All income measures are scaled in $10,000s (1993
dollars). N = 328.
*p < .05 (two-tailed tests)

somewhat larger than the corresponding co- Estimates from the childhood-stage-spe-
efficient (.14) in the individual-based model cific income difference models were some-
presented in Table 3. The standard error was what sensitive to the treatment of the few
also larger, although still less than one-half sibling pairs with large income differences.
of the coefficient estimate. Adjustments for The results for Models 3 and 4 impose no
differences in family conditions had almost truncation on the outliers and show, once
no effect on the income coefficient. again, that economic conditions are most im-

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
 420 AMERICAN SOCIOLOGICAL REVIEW

portant in early childhood. Controls for cor- from our individual-based models; however,
related family conditions increase slightly the imprecision of the estimates left the re-
the coefficient estimate for early-childhood sults far from definitive.
income. 10 That high parental income during adoles-
cence facilitates entry into college is not sur-
prising. Why income early in childhood ap-
SUMMARY AND DISCUSSION
pears to matter more for achievement than
In our examination of links between family for behavior may be due to the importance
income and child development, we have of school readiness in determining the course
summarized recent contributions to the lit- of schooling for children. Income poverty
erature and conducted new empirical work. has a strong association with a low level of
Striking consistencies have emerged. preschool ability, which is associated with
An important "stylized fact" in the recent low test scores later in childhood as well as
literature is that family income has much grade failure, school disengagement, and
stronger associations with achievement and dropping out of school, even when controls
ability-related outcomes for children than for family characteristics such as maternal
with measures of health and behavior. A sec- schooling, household structure, and welfare
ond noteworthy result is that early childhood receipt are included (Brooks-Gunn, Guo, and
appears to be the stage in which family eco- Furstenberg 1993; Guo, Brooks-Gunn, and
nomic conditions matter the most. And third, Harris 1996).
the estimated impact of family income on Why might this be the case? Preschool
completed schooling appears to be larger for ability sets the stage for children's transition
children in low-income families than those into the formal school system. Children who
in high-income families. have not learned skills such as color naming,
Our PSID-based analyses of the effects of sorting, counting, letters, and the names of
family income during childhood on com- everyday objects are at a disadvantage com-
pleted schooling and nonmartial fertility pared with children who have mastered these
were consistent with each of these points. We skills. Schools tend to classify children very
found that family income had a stronger as- early-language arts groups are often formed
sociation with completed schooling than with in kindergarten or first grade. Teachers also
nonmarital fertility. A second result was clear tend to identify children as having potential
evidence that family income in early child- school problems in the first years, with these
hood had a bigger impact on completed ratings being at least as predictive as read-
schooling than did income during middle ing- and math-readiness test scores
childhood. At the high end of the socioeco- (Entwistle and Alexander 1989).
nomic scale, our evidence suggests that en- The same is not as true for behavior prob-
try into college is facilitated if parental in- lems. The correlations between preschool
come during adolescence is high. And third, behavior problems and elementary school
the impact of family income on completed behavior problems are not as strong as those
schooling was largest for children in low-in- found for achievement (Guo et al. 1996).
come families. Moreover, behavior problems seem to be
Our attempt to use sibling differences to more strongly influenced than is school
eliminate the influence of unmeasured per- achievement by other family events
sistent family characteristics from our esti- (Campbell 1995; Links 1983; Sameroff et al.
mated effects of income was only partially 1993). Other contextual factors gain in im-
successful. Results from our sibling-based portance as children age-peers have a ma-
models were not inconsistent with results jor impact on juvenile delinquency, for ex-
ample. Thus, it may be possible for a child
with moderate levels of behavior problems in
10 Truncating income changes to be no more
the early years to have no such problems at
than $20,000 in absolute value produced coeffi-
cients and standard errors associated with income the end of elementary school, while children
during the three childhood stages of .20 (.13), .16 with moderate readiness problems are less
(.13), and -.02 (.1 1), respectively, in models con- likely to be able to catch up in the academic
trolling for the full set of life events. sphere.

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
 CHILD POVERTY AND EARLY-ADULT SUCCESS 421

Taken as a whole, our data are consistent welfare receipt, family structure, parental in-
with the hypothesis that raising the incomes volvement on children's educational outcome,

of poor families will enhance the abilities economic success, andfamilyformation behavior.

and attainments of their children. Most im- Jeanne Brooks-Gunn is the Virginia and
portant appears to be the elimination of deep Leonard Marx Professor of Child Development at
and persistent poverty during a child's early Columbia University's Teachers College. She is
years. Income increments to nonpoor fami- also Director of the Center for Children and
Families, an academic center dedicated to policy
lies or to families with older children may be
research on children. Her most recent books are
desirable on other grounds, but do not appear
Consequences of Growing Up Poor (with Greg
particularly effective in enhancing children's
Duncan, Russell Sage, 1997) and Neighborhood
achievement or changing their behavior. Poverty (Russell Sage, 1997).

Judith R. Smith is Associate Professor at the

Greg J. Duncan is Professor of Education and
Social Policy and a Faculty Associate in the In- Graduate School of Social Services at Fordham
University at Lincoln Center, as well as Research
stitute for Policy Research at Northwestern Uni-
Associate at the Center for Young Children and
versity. Much of Duncan's career has been spent
Families at Teachers College, Columbia Univer-
at the University of Michigan on the Panel Study
of Income Dynamics data collection project. sity. Her research interests focus on the effects of

More recently, his research has focused on how poverty, welfare receipt and maternal employ-
ment on young children. She is currently involved
economic conditions in families and neighbor-
hoods affect child development and on how wel- with Jeanne Brooks-Gunn in a research project

fare reform affects families and children. titled "Making Ends Meet," which combines
qualitative and quantitative data collection and
W. Jean Yeung is Assistant Research Scientist at investigates the effects of receiving an economic
the Institute for Social Research, University of sanction for noncompliance in a welfare-to-work
Michigan. She is a co-principal investigator for program in a sample of women with young chil-
the Panel Study of Income Dynamics. Her re- dren under age 6. The outcomes of interest are
search focuses on the effects offamily and paren- family budgeting, family functioning and motiva-
tal characteristics on children's well-being. She tion, and obstacles to participation in the man-
has examined the long-term effects of poverty, dated work program.

Appendix A. Means, Standard Deviations, and Unstandardized Coefficients for Control Variables In-
cluded in Model 4 of Table 4

Coefficient

Mean Years of Completed High School Hazard of

Independent Variable (S.D.) Schooling Completion Nonmarital Birth

Child's Race/Gender a

Nonblack male .46 -.33* -.53*

(.51) (.1I1) (.25)
Black male .07 .18 .06
(.27) (.23) (.42)
Black female .07 .07 -.23 .51*
(.26) (.22) (.40) (.28)

Total number of siblings 2.31 -.12* -.05 .28 *

(1.89) (.03) (.06) (.05)

Mother's years of schooling 12.65 .18* .16* -.06

(2.06) (.02) (.06) (.05)

Age of mother at child's birth 24.56 .00 -.05 -.04*

(7.87) (.01) (.06) (.02)

Nonmissing data on age of .96 .68 1.46* -.09

mother at child's birth (.20) (.38) (.74) (.66)

Ever lived in South .37 -.05 -.47* -.33

(.49) (.11) (.22) (.22)

(Appendix A continued on next page)

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
 422 AMERICAN SOCIOLOGICAL REVIEW

(Appendix A continuedfrom previous page)

Coefficient

Mean Years of Completed High School Hazard of

Independent Variable (S.D.) Schooling Completion Nonmarital Birth

Child's Family Structure

Born into a nonintact family .14 -.24 -.08 .29

(.36) (.18) (.33) (.29)

Ever divorced, ages 0 to 5 .09 .14 .16 .87*

(.29) (.22) (.43) (.39)
Ever divorced, ages 6 to 10 .08 .15 .17 .42
(.28) (.23) (.43) (.36)
Ever divorced, ages 11 to 15 .07 .16 .59 .70
(.27) (.22) (.46) (.38)

Ever (re)married, ages 0 to 5 .06 -.37 -.26 -.40

(.24) (.26) (.47) (.47)

Ever (re)married, ages 6 to 10 .06 .10 .18 -1.14*

(.24) (.26) (.51) (.47)

Ever (re)married, ages 11 to 15 .07 .01 -.74 -.01

(.26) (.24) (.45) (.41)

Residential Mobility
Years moved, ages 0 to 5 1.30 -.05 -.24* -.01
(1.28) (.05) (.09) (.09)

Years moved, ages 6 to 10 .87 -.05 -.12 .02

(1.14) (.05) (.10) (.09)

Years moved, ages 11 to 15 .69 -.13 -.16 .16

(1.08) (.06) (.10) (.09)

Maternal Employment
Years mother worked 1,000 hours 1.04 .00 .13 -.09
or more, ages 0 to 5 (1.55) (.12) (.10) (.09)

Years mother worked 1,000 hours 1.59 -.29* -.01 .01

or more, ages 6 to 10 (1.90) (.12) (.09) (.09)

Years mother worked 1,000 hours 2.38 .20 .07 .08

or more, ages 11 to 15 (2.08) (.13) (.07) (.07)

Note: For coefficients, numbers in parentheses are standard errors. N = 1,323.

a For child's race/gender, the omitted category is "nonblack female."

:p < .05 (two-tailed tests)

REFERENCES

Blau, David. 1995. "The Effect of Income on 1997. "Family Economic Hardship and Adoles-
Child Development." Department of Econom- cent Adjustment: Mediating and Moderating
ics, University of North Carolina, Chapel Hill. Processes." Pp. 288-310 in Consequences of
Unpublished manuscript. Growing Up Poor, edited by G. Duncan and J.
Brooks-Gunn, Jeanne, Guang Guo, and Frank F. Brooks-Gunn. New York: Russell Sage.
Furstenberg, Jr. 1993. "Who Drops Out and Conger, Rand D., Katherine J. Conger, Glen H.
Who Continues Beyond High School? A 20- Elder, Jr., Frederick 0. Lorenz, Ronald L.
year Follow-up of Black Urban Youth." Jour- Simons, and Les B. Whitbeck. 1992. "A Fam-
nal of Research on Adolescence 3:271-94. ily Process Model of Economic Hardship and
Campbell, S. B. 1995. "Behavior Problems in Adjustment of Early Adolescent Boys." Child
Preschool Children: A Review of Recent Re- Development 63:526-41.
search." Journal of Child Psychology and Psy- . 1993. "Family Economic Stress and Ad-
chiatry 36:113-49. justment of Early Adolescent Girls." Develop-
Conger, Rand, Katherine Conger and Glen Elder. mental Psychology 29:206-19.

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms
CHILD POVERTY AND EARLY-ADULT SUCCESS 423

Corcoran, Mary. 1995. "Rags to Rags: Poverty CA: Sage.

and Mobility in the United States" Annual Re- Hill, Martha and Greg Duncan. 1987. "Parental
view of Sociology 21:237-67. Family Income and the Socioeconomic Attain-
Currie, Janet. 1995. Welfare and the Well-Being ment of Children." Social Science Research
of Children. Chur, Switzerland: Harwood Aca- 16:39-73.
demic Publishers. Korenman, Sanders and Jane Miller. 1994.
Duncan, Greg. 1988. "The Volatility of Family "Long-Term Poverty and Child Development
Income Over the Life Course." Pp. 317-58 in in the United States: Results from the NLSY."
Life-Span Development and Behavior, vol. 9, Discussion Paper No. 1044-94, Institute for Re-
edited by P. Baltes, D. Featherman, and R. M. search on Poverty, University of Wisconsin,
Lerner. Hillsdale, NJ: Lawrence Erlbaum As- Madison, WI.
sociates. Links, Paul. 1983. "Community Surveys of the
Duncan, Greg J. and Jeanne Brooks-Gunn, eds. Prevalence of Childhood Psychiatric Disorders:
1997. Consequences of Growing Up Poor. New A Review." Child Development 54:531-48.
York: Russell Sage. Mayer, Susan. 1997. What Money Can't Buy: The
Duncan, Greg, Jeanne Brooks-Gunn, and Pamela Effect of Parental Income on Children's Out-
Klebanov. 1994. "Economic Deprivation and comes. Cambridge, MA: Harvard University
Early-Childhood Development." Child Devel- Press.
opment 65:296-318. Miller, Jane and Sanders Korenman. 1994. "Pov-
Entwistle, Doris and Karl Alexander. 1989. erty and Children's Nutritional Status in the
"Early Schooling as a 'Critical Period' Phe- United States." American Journal of Epidemi-
nomenon." Research in Sociology of Education ology 140:233-43.
and Socialization 8:27-55. Rainwater, Lee and Timothy Smeeding. 1995.
Guo, Greg, Jeanne Brooks-Gunn, and Kathleen "Doing Poorly: The Real Income of American
M. Harris. 1996. "Parents' Labor-Force At- Children in a Comparative Perspective." Work-
tachment and Grade Retention among Urban ing Paper No. 127, Maxwell School, Syracuse
Black Children." Sociology of Education 69: University, Syracuse, NY.
217-36. Sameroff, Arnold J., Ronald Seifer, Alfred
Haveman, Robert and Barbara Wolfe. 1994. Suc- Baldwin, and Clara Baldwin. 1993. "Stability
ceeding Generations: On the Effects of Invest- of Intelligence from Preschool to Adolescence:
ments in Children. New York: Russell Sage. The Influence of Social and Family Risk Fac-
. 1995. "The Determinants of Children's tors." Child Development 64:80-97.
Attainments: A Review of Methods and Find- Sewell, William and Robert Hauser. 1975. Edu-
ings." Journal of Economic Literature 33: cation, Occupation and Earnings: Achievement
1829-78. in the Early Career. New York: Academic
Haveman, Robert, Barbara Wolfe, and James Press.
Spaulding. 1991. "Educational Achievement Smith, Judith, Jeanne Brooks-Gunn, and Pamela
and Childhood Events and Circumstances." Klebanov. 1997. "Consequences of Growing
Demography 28: 133-58. Up Poor for Young Children." In Conse-
Heckman, James J. and Richard Robb. 1985. "Al- quences of Growing Up Poor, edited by G.
ternative Methods for Evaluating the Impact of Duncan and J. Brooks-Gunn. New York:
Interventions." Pp. 512-38 in Longitudinal Russell Sage.
Analysis of Labor Market Data, edited by J. U.S. Bureau of the Census. 1996. "The Official
Heckman and B. Singer. New York: Cam- Statistics." Washington, DC: U.S. Bureau of
bridge University Press. the Census. Retrieved December 4, 1996 (http:/
Hill, Martha. 1992. The Panel Study of Income /www.census.gov/ftp/pub/hhes/poverty/pov95/
Dynamics: A User's Guide. Newbury Park, povest .html).

This content downloaded from 155.48.255.218 on Wed, 23 Jan 2019 14:21:45 UTC
All use subject to https://about.jstor.org/terms