1
HOW LARGE ARE THE EFFECTS FROM CHANGES IN FAMILY
ENVIRONMENT? A STUDY OF KOREAN AMERICAN
ADOPTEES*
Bruce Sacerdote
Dartmouth College and NBER
bruce.sacerdote@dartmouth.edu
April 6, 2006
Abstract
I analyze a new set of data on Korean-American adoptees who were quasi-randomly assigned to
adoptive families. I find large effects on adoptees' education, income and health from assignment to
parents with more education and assignment to smaller families. Parental education and family size
are significantly more correlated with adoptee outcomes than are parental income or neighborhood
characteristics. Outcomes such as drinking, smoking, and the selectivity of college attended are
more determined by nurture than is educational attainment. Using the standard behavioral genetics
variance decomposition, I find that shared family environment explains fourteen percent of the
variation in educational attainment, thirty five percent of the variation in college selectivity, and
thirty three percent of the variation in drinking behavior.
*
I thank Holt International Children's Services and particularly Laura Hofer and Karla Miller for their help in gathering
data and information on international adoptions. The National Science Foundation provided generous funding for the
entire project including the data collection. I thank Anne Ladenburger, Abigail Ridgeway, Ariel Stern-Markowitz, and
Celia Carmen for tireless research assistance and valuable suggestions. Larry Katz and Edward Glaeser provided
superb editing. I thank Joseph Altonij for very useful comments. Heroic efforts from two anonymous referees greatly
improved the paper. Jonathan Gruber and Kwabena Gyimah-Brempong advised me on the design of the study.
Seminar participants at NBER Summer Institute, Syracuse, Cornell, Case-Western, Brigham Young University,
Vanderbilt, University of South Florida, University of Colorado at Denver and elsewhere contributed very helpful
comments.
2
I. Introduction
Social scientists, policy makers and parents everywhere are interested in the degree to which
children's behavior and outcomes are determined by nature, nurture and the interaction between the
two. This paper uses at adoption in infancy to identify the effects of large scale changes in family
environment on children's outcomes. I compare outcomes for children assigned to smaller families
with highly educated parents to outcomes for children assigned to large families where neither
parent has a college degree. The quasi random assignment of children to families in the data allows
me to give the estimates a causal interpretation. Children assigned to the high education, small
families are twice as likely to graduate from a college ranked by U.S. News & World Report, have
an additional .75 years of education, and are 16 percent more likely to complete four years of
college. My estimated effects may provide a useful context for understanding how more typically
observed changes in neighborhoods, family income, peer quality, or school quality translate into
children's outcomes.
The adoptees in the study are Korean-Americans placed by Holt International Children's
Services during 1964-1985. The adoptees are quasi randomly assigned to families, conditional on
the family being certified by Holt to adopt. Holt uses a queuing (first-come first-served) policy to
assign Korean adoptees to families. As a result, assignment of children to families is effectively
random conditional on the adoptee's cohort and gender. I provide empirical evidence showing that
adoptee's pre-treatment characteristics are uncorrelated with adoptive family characteristics.
The data come from Holt's records and from a survey of adoptees and their families
conducted during 2004-2005. Holt and I originally targeted a sample of adoptees ages 24-34 in
2004, but collected data for all adoptees and nonadoptees in the family. The sample is much larger
3
than most existing adoption studies (with the exception of Bjorkland Lindahl, Plug [2006]) and we
collected a wide range of outcomes. Two chief disadvantages of my data are that the response rate
to the initial survey was low at 34 percent and that I rely on parental reports of adult adoptee
outcomes. To deal with these issues Holt and I resurveyed a sample of the non-respondents and I
show that responses are not significantly correlated with child outcomes. We also surveyed directly
a smaller sample of the adoptees and nonadoptees and I demonstrate the high degree of
correspondence between their responses and their parents.
I apply the standard behavioral genetics framework (see Plomin, DeFries, and Fulker
[1988]) to decompose the variance in child outcomes into variance explained by environment
(nurture), variance explained by genes or initial endowments (nature), and variance explained by
non-shared environment. While educational attainment and income are frequently the focus of
economic studies, these are among the outcomes least affected by differences in family
environment. In contrast, the selectivity of college attended has a much larger nurture component
and the explained variation in "social" outcomes like drinking behavior and marital status appears to
be almost entirely nurture based.
I.A. A Brief History of Holt, Korean-American Adoption and the Assignment Process
Harry and Bertha Holt pioneered international adoption in Seoul, Korea in 1955. The
Holts had built a fortune in lumber and farming in Oregon and were so moved by the plight of
Korean War orphans that they lobbied Congress for a special act to adopt eight of them. When
they returned home with their new children, they discovered that many other Americans also wished
to adopt from Korea.
4
Since 1955 over 100,000 Korean children have been adopted into U.S. families, and the
agency which grew out of the Holt’s initial work, now called Holt International Children’s Services,
has been involved in 30 to 40 percent of these adoptions. Holt currently places about 300 Korean
adoptees per year, and hundreds more from China and from programs in Bulgaria, Ecuador,
Guatemala, Haiti, India, Mongolia, Philippines, Romania, Thailand, Uganda and Vietnam.
The process of adopting through Holt's Korea program takes roughly 12-18 months from
initial application to bringing home the adoptee. The major steps include filing an application,
participating in the home study assessment, attending adoption education classes, passing the
criminal background check, being matched with an adoptee, the adoptee flying to the U.S., and
legally adopting the child in family court. This is an extensive and thorough process requiring
numerous meetings with adoption agency personnel and numerous exchanges of documents.
In part due to U.S. and South Korean law, adoptive parents must meet several criteria
including a minimum family income and being married for three years or longer.1
U.S. law
requires that family income be 125 percent of the poverty level. (The data contain many families
near this threshold.) Korean law requires that the adoptive parents be between the ages of 25 and
45 at the time of application processing and have no more than four children in their current family.
Within the Korea program and conditional upon being qualified to adopt, children are
matched to families on a first come, first served basis. Thus it is the timing of when applications
are completed that creates the matching of parents to children, rather than any matching of parent
and child characteristics. This is what results in quasi random assignment of adoptees to families.
Precisely which adoptee goes to which family is determined by fairly "random" factors such as
5
whether it takes the parents eight weeks versus nine weeks to get their home study completed or to
have their recommendation letters mailed in. Small differences in motivational levels would have
small effects on how quickly the process is completed, but these differences in efficiency are
uncorrelated with any characteristics of the adoptee. In my analysis I include cohort dummies in
case parent and child characteristics co-vary systematically over time.
Parents are not given the opportunity to specify gender or anything else about their future
adoptee. The one exception to this rule is that families with all boys or all girls were allowed to
request a child of the opposite gender.2
In practice, those who were eligible to request girls
frequently did so. This does not present a problem for this study since I condition on adoptee
gender in every specification. The only other opportunity parents have to specify a preference is to
indicate that they would be open to adopting a child with special needs or a disability. I exclude all
such adoptions from the sample.
Holt Children's Services of Korea, a separate organization from Holt International
Children’s Services maintains a network of foster homes in Korea. When Korean mothers (or
families) are forced through life circumstances to give up a child for adoption, the mother (family)
will frequently bring the child to Holt's facility in Seoul or one of the eleven branch offices.3
Holt
places the child in one of its foster homes to await adoption. Currently the majority of children
given up for adoption in Korea are born to unwed mothers. During the 1960s and 1970s when the
adoptees in the study were placed, some of the relinquishments were due to unwed motherhood but
others were due to poverty and the family's inability to care for the child.
1
Information on Holt's policies and assignment process is provided from their website www.holtintl.org and my
interviews with Holt officers.
2
Even this exception was recently eliminated due to the overwhelming excess demand for girls.
3
During the 1970s it was common for the mother to relinquish the child into an orphanage. Holt Korea had a network
of contacts with orphanages and would place children from these orphanages into U.S. families.
6
The physical process of matching of adoptees to families is quite simple and as noted above
uses a queuing system which effectively randomizes children to families.4
Once Holt has a
completed application file and has certified that the parents are qualified to adopt, the file is added
to a small stack of other such applications completed that week. Once per week the stack of
roughly 5-10 completed applications is sent to Holt Children's Services of Korea. Holt Korea
receives the applications and assigns any children in its system to these prospective adoptive
families on a first come first served basis. If no children are available at a given moment, then Holt
Korea holds the file and waits for the next available child. From the point at which the child is
assigned to an adoptive family, it takes about 4.5 months for the child to come home. I provide
evidence below that the child's weight in infancy and other pre-adoption characteristics are
uncorrelated with adoptive parent characteristics such as family income, parental education etc.
I.B. Relation to the Literatures on Adoption and Nature and Nurture
Psychologists and sociologists have long used adoption data as a way to examine the effects
of family environment while (hopefully) controlling for the potentially confounding effects of genes
or other prenatal factors which are likely strongly correlated with environment. The literature has
focused mostly on estimating the heritability of IQ, as in Scarr and Weinberg [1978], and
personality traits as in Loehlin, Horn, and Willerman [1982, 1987, 1994], and Plomin, Defries, and
Fulker [1988, 1994], and health outcomes like alcoholism or depression.5
4
In the 1970s, Holt and Holt Korea had no information about the birth parents which means that any matching based on
birth parent characteristics would have been physically impossible anyway. For a small number of cases (99) Holt
Korea estimated the birth mother's age but did not use that data for matching. I use the data on birth mother's age in one
of several tests for random assignment.
5
See Bouchard and McGue[1981] and Loehlin[1989] summaries of the very large IQ literature.
7
As Jenks [1972] points out, studying IQ and other test scores is potentially very different
than studying economic outcomes. I build on the work of a handful of papers which look at the
effects of adoptive family environment on educational attainment including Teasdale and Owen
[1984], Lichtenstein, Pedersen and McClearn [1992], and Scarr and Weinberg [1994]. My value
added relative to these papers is both the quasi-random assignment of adoptees into families and the
much larger sample sizes in the Holt data.6
A series of recent papers in economics including Björklund, Lindahl and Plug [2006],
Sacerdote [2002], Das and Sjogren [2002], and Plug and Vijverberg [2003] looks at the
transmission of income and education from birth and adoptive parents to adoptees and from parents
to nonadoptees. The goal of this literature is to ask what portion of the transmission of income and
education comes from nature (the birth parents), from nurture (the adoptive parents), and in the case
of Björklund, Lindahl, Plug (BLP) the interaction of the two. The latter paper has by far the largest
data set in that it uses the census of Swedish adoptees placed during 1962-1966. All of the existing
work relies on an implicit or explicit assumption of random assignment of children to adoptive
families. This assumption is necessary in order for the effects from adoptive family environment to
be estimated without bias from the matching of children to parents.
There is also a literature in economics which uses twins data to separate out nature and
nurture effects on educational attainment, income and obesity. A series of papers including
Taubman [1989], Behrman and Taubman [1989] and Behrman, Rosenzweig and Taubman [1994]
use comparisons of identical and fraternal twins and their offspring along with the behavioral
6
For instance, the numbers of adoptive sibling pairs in these three existing papers are 27, 61, and 59 respectively. Other
existing papers which examine adoptee's educational attainment using a different methodology are Teasdale and Owen
[1986] and Maughan, Collishaw and Pickles[1998].
8
genetics (BG) model to identify the nature and nurture components of these outcomes. However,
Goldberger [1978] points out a number of severe limitations to the BG approach.
I depart from the existing economics literature on adoption in four important ways. First,
the mechanism by which Holt assigns children to families is known and is effectively random. I
provide empirical evidence as well as institutional detail on this point. Second, I use the traditional
behavioral genetics model to arrive at explicit estimates of the relative importance of nature and
nurture in explaining variance in a variety of outcomes. Third I have a wider range of outcomes
than was available in previous economic studies and thus can provide a comparison between
nurture's effect on "social" versus labor market outcomes.
Finally, I calculate treatment effects on child outcomes from being assigned to a small, high
education family.7
The treatment effects framework requires significantly fewer assumptions than a
more structural approach like the BG framework. I focus on small, high education families because
these two family characteristics appear to be the most correlated with adoptee outcomes and may
indicate something about the quantity and quality of parental attention the adoptees receive. For
certain outcomes, assignment to a small high education family has an effect similar in size to that of
a one standard deviation change in the index of family environment implied by the BG model.
II. Empirical Frameworks
II.A. Variance Decomposition
My first empirical approach is a nature-nurture variance decomposition via the standard
behavioral genetics (BG) model. This exercise provides an estimate of the importance of nature and
7
By "treatment effect" I mean the casual difference in outcomes that results from a child being assigned to one type of
family versus another. Holt is of course assigning children to a range of families which I aggregate into discrete
treatment groups for this part of my empirical work.
9
nurture in explaining the variation in child outcomes. Suppose that child outcomes (Y) are
produced by a linear and additive combination of genetic inputs (G), shared (common) family
environment (F) and unexplained factors, which the BG literature often calls non-shared or separate
environment, (S). This implies that child's educational attainment can be expressed as follows
(1) Child's years of education (Y)= G + F + S
The strong assumptions here are that nature (G) and shared family environment (F) enter
linearly and additively. To get a variance breakdown one generally further assumes that G and F
are not correlated for both the adoptees and non-adoptees.8
Taking the variance of both sides
yields:
(2) σ2
Y = σ2
G + σ2
F + σ2
S
Dividing both sides by the variance in the outcome (σ2
y ) and defining h2
=σ2
g /σ2
y, c2
=σ2
F
/σ2
y , and e2
= σ2
S /σ2
Y yields the standard BG relationship:
(2a) 1 = h2
+ c2
+ e2
The variance of child outcomes is the sum of the variance from the genetic inputs (h2
or
heritability), the variance from family environment (c2
) and the variance from non-shared
environment (e2
), i.e. the residual. From this starting point, a variety of variances and covariances
8
This assumption can be justified for the Holt adoptees on the basis of quasi random assignment of children to families.
Environment and genes are surely correlated for the non-adoptees, so on the surface this would seem like an
indefensible assumption. One of the referees informs me that behavioral geneticists think of G as representing both the
direct effects of genes and the effects of gene environment correlation. This is important for interpreting the variance
breakdown.
10
of outcomes can be expressed as functions of h, c, and e. The sample moments can then be used to
identify these underlying parameters. For example, if one standardizes Y,F,G,S to be mean zero
variance one, the correlation in outcomes between two adoptive siblings equals Corr (Y1, Y2) =
Cov(Y1,Y2)=Cov(F1,F1)=Var(F1)=c2
.
The correlation in outcomes between two nonadoptive siblings equals Corr (Y1, Y2) = Cov
(G1+F1+S1, G2+F2+S2) = Cov (G1+F1, 1/2G1+F1) = ½ h2
+ c2
. This assumes that non-adoptive
siblings share half of the same genetic endowment and the same common environment (See Plomin
et. al [2001] for a discussion). Thus one can recover the full variance breakdown (h2
,c2
,e2
) from
just the correlation among adoptive and biological siblings. Notice that h2
= twice the difference in
correlations in the outcome between the adoptive and biological siblings. Given the decomposition
I calculate how much I would expect a child's outcome to change given a one standard deviation
change in the index of family environment, F. This quantity is c* σy.9
It is possible that my data exclude some critical (low) level of family environment which
could be altered to deliver a larger percent of variance apportioned to shared family environment or
larger treatment effects than those found here (See Turkheimer et. al. [2003]). Stoolmiller [1999]
points out that c2
could significantly underestimate the fraction of variation explained by nurture if
there is some restriction in the range of family environments observed. In practice the variance of
family environment in the Holt data is as large as the variance of family environment observed in
the U.S. population. For example in the Holt data the standard deviations for mother's years of
education, college status, and family income are 2.45 years, 49.9 percent, and $23,600 respectively.
Using the National Longitudinal Survey of Youth 1979 (NLSY79) data, weighted to represent the
U.S. population in 1979, the standard deviations for mother's education, college status and family
9
By definition the r-squared from a regression of Y on F = c2
= σ2
yhat /σ2
y, so c* σy = σyhat.
11
income are 2.50 years, 31.7 percent, and $14,216. The Holt data do contain low income and low
education families, even though the means of these two variables are significantly above U.S.
averages.
I also use Census data to compare the distribution of Holt adoptive family income at the
time of adoption to that of all married, two parent families with children in the U.S. in 1980.
Family income for the latter group is calculated from the Individual Public Use Micro Sample
(IPUMS) Census data. Contrary to common perception (likely based on current adoptive families),
the adoptive families in the sample are not universally high income families. Families with less
than $10,000 of annual income are significantly underrepresented in the Holt data, but such families
still represent 25 percent of the Holt sample, versus 45 percent for US families with children in
1980. Sixty percent of the Holt families have income of $25,000 or less versus 73 percent for U.S.
families with children.
II.B. Treatment Effects
I also estimate the treatment effect on adoptee outcomes from being assigned to a particular
family type. Interpreting these treatment effects as causal requires only that assignment to treatment
group is quasi random. I define three different types of adoptive families based on their observables.
Type one are high education, small families, meaning there are three or fewer children total and
both parents have four years of college. Adoptees in such families comprise 27 percent of the
sample. Type three families (12 percent of the adoptees) are those in which neither parent has four
years of college and there are four or more children in the family. Type two families are the set of
all other families not in either of the extreme groups. I calculate the treatment effects from
assignment to type one versus three (high education, small versus lower education large) and type
two versus three.
12
To do this I take the set of adoptees in my sample and run regressions of the following form:
(3) Ei= α + β1*T1i + β2*T2i + Malei + Ai + Ci + εi
Where Ei is educational attainment for child i, T1i is a dummy for being assigned to a family
of type 1, T2i is a dummy for being assigned to a family of type 2, Ai is full set of single year of age
dummies, and Ci are a full set of cohort dummies.
The less educated larger adoptive families are the omitted category. The cohort dummies
are included since assignment to treatment group is quasi random within the time period during
which the family applied to adopt through Holt. In other words, child and family characteristics
might vary over time in a non-random way. Cohort is defined as the year in which the child
initially entered the Holt system in Korea. The age dummies are an additional control for the fact
that outcomes like educational attainment vary with child age. Age and cohort are not perfectly
collinear since there is some modest variation in age at adoption. (See summary statistics below.)
The gender dummy is included because in a limited number of cases adoptive families are able to
request the adoptee's gender. And like age dummies, the gender dummy improves precision on the
estimated treatment effects by removing additional variation that would otherwise end up in the
error term.
Due to quasi random assignment, β1 can be interpreted as the causal effect of assignment to
a high education small family, relative to assignment to a less educated large family.10
I report
results for both comparisons. Of course, the causal effect need not work directly via parental
education or family size since other important environmental factors vary across family types.
13
These factors could include income, parental attention, school quality neighborhood quality etc. I
defined family types using parental education and family size because these are the two observables
that are most strongly correlated with child outcomes. And defining treatment groups in this way
delivers treatment effects on educational outcomes that are similar in magnitude to a one standard
deviation move in the index of shared family environment.
II.C. Estimation of Transmission Coefficients
Most studies of intergenerational correlations in economics focus on transmission
coefficients in which the child's outcome is regressed on the parent's outcome. See Solon [1999] or
Bowles, Gintis and Osborne Groves [2005] for a review. In order to provide comparability between
my results and those in the existing literature I also calculate transmission coefficients for a variety
of outcomes. For the adoptees this means running regressions of the following form:
(4) Ei= α + δ1*EMi + Malei + Ai + Ci + εi
where EMi is adoptive mother's years of education and the other variables are as above.
Again, the quasi random assignment ensures that there is no correlation between the child's initial
health or genetic endowments and adoptive mother's education. This allows me to obtain an
estimate of δ1 that is not biased by selection of adoptees into families. δ1 is the transmission that
takes place purely through nurture and not through genes or pre-natal environment. For the
nonadoptees I run analogous transmission regressions of the form:
(4A) Ej= α + δ2*EMj + Malej + Aj + Cj + εj
10
Without random assignment, β1 represents an unknown mix of treatment effects plus selection into the family
(treatment group). This is why previous adoption studies by economists (including my own) generally avoid using
causal or treatment effects language, and instead focus on estimating transmission coefficients.
14
where the nonadoptees are indexed by j and Mj simply represents mother's education (not adoptive
mother's education.) This yields an estimate of the transmission of education (outcomes) from
parents to children when there is a genetic connection between the parent and child. A comparison
between δ1 and δ2 is an estimate of how much of the transmission of education (outcomes) works
through nurture, as opposed to through nature and nurture combined.
III. Data Description
Holt and I collected data on adoptive parents and their children using Holt records and a
mail in survey.11
A copy of the survey is included as Appendix III in Sacerdote [2005].12
The
survey asked parents for information on their education, occupation, income and health, where the
latter includes height, weight, smoking and drinking status. We also asked the parents questions on
the children's health, education, and income. We collected basic demographic outcomes for the
children including marital status and number of children. We sent a pilot survey to 1,000 of the
families. We then sent a main mailing to an additional 2,500 families. We then sent surveys to a
subset of 653 of the children (both adoptees and nonadoptees) to measure the degree to which
parents and children gave the same answer when asked about the child's outcome. Finally we sent
400 follow up surveys to a random subset of the parents who did not respond in the main mailing.
The purpose of the follow-up was to allow us to ask whether non-response is correlated with either
child outcomes or family background.
Parents were eligible for inclusion in the survey if they adopted a child through Holt's Korea
program during 1970-1980, making the children ages 24-34 in 2004 when the survey was run.
11
The effort required extensive work from Holt officers and employees and from a team of research assistants at
Dartmouth.
12
This working paper is available at www.dartmouth.edu/~bsacerdo.
15
There were roughly 7,700 such families who met this criterion and as mentioned above we sent the
survey to a random sample of 3,500 of these families.
Holt maintains electronic records with some basic pieces of information such as name,
address, and adoptees' names and ages for each of their adoptive families. Whenever Holt has
contact with a family they update the relevant address in their database. Contact may occur due to a
family's use of Holt's post-adoption services, a family's subscribing to Holt's monthly magazine, or
because of a donation to Holt. In addition, Holt contracts with a direct mail company to keep the
addresses as accurate and up to date as possible. This is done in part by matching on exact names
using U.S. phone directories.
In the pilot and main mailing, our cover letter promised respondents a check for $50. This
was paid immediately upon receipt of a completed survey. The survey of the children was
conducted in a similar manner and also had an incentive payment of $50. For the follow up survey
to nonrespondents we used U.S. Priority Mail envelopes (to make the envelope more noticeable)
and we offered $75.
Appendix Table I shows the sample sizes and response rates for the various mailings. The
main mailing went to 2500 adoptive families. We received back 851 completed surveys for a
response rate of 34 percent. We resurveyed 400 of the nonrespondents and had a 35 percent
response on the resurvey. While neither response rate is terribly high, the resurvey data plus the
complete data in Holt records allow me to examine the possible severity of nonresponse bias. In
Sacerdote [2005] I provide a detailed analysis of nonresponse bias. In this analysis I show that
adoptee outcomes are not statistically significant predictors of whether the parents responded to the
16
original mailing with the follow-up mailing. This is some evidence against a story in which only
parents of "successful" adoptees respond.
I use the administrative data in Holt records to ask whether parental characteristics predict
non-response and report the results below in Appendix Table II. Parental characteristics do have
modest power to predict nonresponse, though the estimated coefficients are small and in opposing
directions. An additional year of father's education raises the probability of response by 1.4 percent,
but a 10 percent increase in family income would decrease the probability of response by 1 percent.
Since there is some evidence that response is correlated with family background, in all of the
estimates below I attempt to correct for nonresponse bias by using Wooldridge's [1999] inverse
probability weighting.13
III.B. Evidence of Random Assignment
The description of the adoption process in section II explains why assignment to families is
effectively random conditional on the adoptee's cohort. Here I provide statistical evidence that
adoptee and parent pre-treatment characteristics are indeed uncorrelated. Table I is calculated using
data from Holt's records and includes all families to whom we sent surveys, whether or not they
responded. Each column is a separate regression. I regress pre-treatment characteristics of the
adoptee on pre-treatment characteristics of the adoptive family. The dependent variables are the
adoptee's age at arrival in the U.S., weight upon entering the Holt system, height upon entering, and
a dummy for male.14
All regressions include dummies for adoptee age and cohort.
13
Wooldridge demonstrates that one way to correct for nonresponse bias is to weight the observations by 1/[1P(response)].
To estimate the probability of response, I use the fitted values from the probit in Appendix Table II.
Where it is impossible to calculate the fitted value (due to missing x's) I assign the observation the average response rate
as the probability of response. The correction for nonresponse bias makes almost no difference in the estimates.
14
Male is usually a right hand side control to ensure quasi-randomization but I include it in column 4 as a dependent
variable just to make the point that even the male dummy is uncorrelated with parent characteristics.
17
The right hand side variables are the log of family income, father's years of education,
mother's years of education, and median income in adoptive family's zipcode. None of the family
background characteristics are statistically significant predictors of adoptee age at arrival, height,
weight or gender. The last row of the table shows the p-value for the joint significance of family
background in predicting adoptee characteristics. As a group the family background variables are
not significant in any of the regressions. In a very small number of cases Holt records contain data
on the birth mother's marital status, age at adoptee's birth, and years of education. I use these as the
dependent variable in columns (5), (6), and (7) respectively. And again I find no statistically
significant relationship between birth mother characteristics and adoptive family characteristics.
Table II performs the same exercise as Table I, but within my analysis sample of Holt
adoptees whose parents responded to the survey. The dependent variables are weight and height at
the time of entry into the Holt system, the child's age at arrival in the U.S. and a dummy for the
adoptee being male. I include age and cohort dummies. There are more right hand side variables
relative to the previous table since in the analysis sample I have all of the survey measures of
adoptive family pre-treatment characteristics. For example, I now include mother's and father's
heights, body mass indices and drinking status. Again, there is no statistically significant
relationship between adoptee pre-treatment characteristics and adoptive family background
characteristics. This can be seen in the last two rows of the table which report the F-statistic for the
joint significance of the parental characteristics and the corresponding p-value.
III.C. Summary Statistics For The Analysis Sample
The survey collects outcomes for up to 5 children in the family. Fortunately for the
purposes of sample size, most families had more than one child, and in many cases families had
18
more than one Holt adoptee from Korea. Of the 1197 families, 359 have two children, 329 have 3
children, and 230 have four children. Ninety two families have six or seven children, but
unfortunately we only collected information on 5 of the children in these large families. Only 68
families have a single child, and that child is of course a Holt adoptee. In single child families,
where there is exactly one Holt adoptee, 78 percent of the adoptees are girls. In families of two
children, 80 percent of the children are adoptees and 63 percent are girls. In the larger families, 55-
60 percent of the children are adoptees and about 55 percent are girls.
For the analysis sample, I limit the data to children who are currently ages 19-40. I dropped
adoptees who were not adopted through Holt's Korea program and adoptees who were listed as
special needs children (since the latter are not randomly assigned). For the purposes of analyzing
educational attainment, college status, and the child's family income, I further limit the sample to
children ages 25-40. Of course to calculate data items like family size and percent girls in the
family, I included all children in the family before limiting the sample on age, country of origin,
Holt adoptee status, or special needs adoptee status.
Table III displays summary statistics for both the adoptees and nonadoptees. There are 1650
adoptees and 1196 nonadoptees. The adoptees are 30 percent male with an average age of 28. Fifty
eight percent of them have four years of college. Thirty seven percent of the adoptees have four
years of college from a college which is ranked by U.S. News. This is a dummy variable which
equals zero if the adoptee did not graduate from college or if the adoptee's college was not listed in
the U.S. News Rankings. Conditional on the adoptee attending (not necessarily graduating from) a
U.S. News Ranked College, the mean acceptance rate of the college was 70 percent. Reported
annual family income for the adoptees is $49,000. Twenty three percent of the adoptees smoke and
19
59 percent drink. These latter outcomes do not measure intensity of drinking and smoking but
rather are dummy variables.
Since the survey was filled out by the parents, a natural question to ask is whether the
parents report accurate answers for their children. I address this question in Appendix Table III.
Holt and I sent surveys to 653 of the children (two thirds of whom were adoptees) and received
back surveys from 55 percent of those contacted. I was able to successfully match parent and child
responses for about 229 adoptees and 93 nonadoptees.15
Appendix Table III shows the correlations
between parent and child responses for these observations. For adoptee's years of education and
college status, parent and adoptee responses have a correlation of .89 and .85 respectively. For
adoptees height and BMI the correlations are .90 and .74 respectively. The only outcome with a
correlation of less than .50 is the child's drinking status. This might mean that my estimate of
family environment's ability to explain drinking behavior is biased upward or downward depending
on what sort of errors (or intentional misreporting) is involved. If parents are biased toward
reporting that their adoptee has the same drinking behavior as their own, this would lead to an
upward bias in the estimate of nurture's explanatory power.
IV. Results
Figures I, II, and III display some of the raw means graphically and foreshadow some of the
key results. Figure I shows the probability of graduating from college by family size, separately for
the adoptees and nonadoptees. Both groups show a steep decline in college graduation rates
associated with each additional child added to the family. This fact survives all of the additional
controls I can add (see discussion of Table VI below). Either there is a direct impact of family size
15
Because I don't have child names in the database, I had to match parent and child surveys on family id, gender, age
and adoption status. I matched 90 percent of the surveys.
20
on educational attainment, or as Black et. al. [2005b] suggest, family size proxies for something
important and unobserved about the family.
Figure II shows the mean of child's years of education for both the adoptees and
nonadoptees for each level of mother's education. Here one can see strong transmission of
education from mothers to children. The upward sloping line is steeper for the nonadoptees relative
to the adoptees. For both groups, the largest jump in average educational attainment is associated
with the mother moving from 11 years of schooling to 12.
Figure III shows that income follows a rather different pattern. The graph displays the mean
of child's family income (in thousands of dollars per year) by eight categories of adoptive family
income at the time of adoption (survey measure). Income transmission is almost non-existent for
the adoptees but strongly positive for the nonadoptees.
IV.A. Variance Decomposition
Table IV shows the correlations in outcomes among sibling pairs after removing age, cohort
and gender effects.16
For educational attainment, biological siblings have a correlation of .34 which
is 2.4 times larger than the correlation of .14 for adoptive siblings. In contrast drinking behavior is
almost as correlated for adoptive siblings as for biological siblings. Marital status is actually
slightly more correlated for the adoptive siblings than the biological siblings with correlations of .08
and .05 respectively.
In Table V, I use the behavioral genetics framework to translate these correlations into the
percent explained by nature, shared family environment and the residual (i.e. non-shared
21
environment). Of the variation in college graduation status, 13.5 percent can be explained by
family environment, 41 percent by nature, and 46 percent by non-shared environment. Variation in
years of education has a similar breakdown at 16 percent family environment, 44 percent nature,
and 40 percent non-shared environment.
Whether the percent of variation in educational attainment that is explained by family
environment is high or low depends on one's priors. Following Duncan, Boisjoly, Mullan Harris
[2001], Jencks and Brown [1979] and others, I show in the next section that a family environment
share of 13 percent can lead to large effects on children's outcomes from changes in children's
family environment.
My variance breakdown for years of education differs somewhat from the existing BG
literature on IQ scores. Reviews of the adoption literature by Bouchard and McGue [1981] or
Plomin et al. [2001] suggest that for adults roughly half the variation in IQ can be explained by
genetic factors and that family environment explains almost none of the variation. I find a
significantly larger role for family environment in explaining educational attainment and this might
be because I am examining a different outcome. In comparison to Behrman and Taubman's [1989]
work on educational attainment in twins and their offspring which finds heritability of 81 percent, I
find much smaller heritability and a larger percent explained by family environment. My estimates
are much closer to those of Jencks and Brown [1979] which also uses twins data.
The few existing adoption behavioral genetics studies on educational attainment find a large
range in percent of variation that is explained by family environment. My results are similar to
those of Lichtenstein, Pedersen, and McClearn [1992] who find that family environment explains 21
16
All correlations in the table except for adoptive siblings' height and weight and nonadoptive siblings' marital status
22
percent of the variance, and Scarr and Weinberg [1994] who find an adoptive sibling correlation of
.13. Teasdale and Owen [1984] find an adoptive sibling correlation of .43. All three of these
studies use completely different samples. Teasdale and Owen are examining a small sample of
Danish siblings reared apart and Lichtenstein et al are examining a small sample of Swedish twins
reared apart. Differential selection of adoptees into families could explain the differences in results,
or perhaps there is something fundamentally different about outcomes for siblings reared apart.
Perhaps the more interesting fact is how much the percent of variance attributable to nurture
varies across different outcomes. When I consider graduating from a U.S. News ranked college,
family environment explains 25 percent of the variation, instead of the 14 percent of variation in
graduating from any college. Variation in the selectivity of college attended is explained 34 percent
by family environment and 25 percent by genetics. For drinking behavior, 34 percent of the
variation is explained by family environment and almost none of the variation is explained by
genetics.17
My results on body mass index are consistent with several existing papers that
demonstrate the high heritability of weight, notably Grilo and Pogue-Geile [1991] and Cardon
[1994]. Variation in marital status appears to have a modest family environment component (8
percent) and no genetic component. My preferred interpretation of these results as a whole is that
as one looks at more "social" outcomes such as drinking behavior or choice of college, family
environment plays a larger role than it does in say the determination of years of education.
IV.B. Treatment Effects and Multiple Regression Results
are statistically different from zero at the one percent level and most have p-values even smaller than one percent.
17
The existing literature finds that alcoholism is highly heritable (Cloninger, Bohman, and Sigvardsson [1981]). A
plausible explanation for the difference in findings is that different processes are at work in the creation of social
drinking versus alcoholism.
23
Naturally understanding the causal mechanisms underlying the effects of family
environment is at least as interesting as measuring the percent of variation explained. Because
income, parental education, neighborhood quality, and many unobservables all covary, it is
impossible to definitively separate out root causes. Nonetheless in Table VI, I use multiple
regression to attempt to identify which aspects of family environment are the most important for the
adoptees' outcomes. I regress various outcomes for the adoptees on mother's years of education, the
number of children in the family, the log of family income at the time of adoption, log of median
income in the zip code, a dummy for being the only adoptee, the fraction of girls in the family, and
mother's body mass index, and drinking status.
A relatively clear pattern emerges. Even controlling for income and other aspects of family
background, mother's education and the number of children in the family have a large effect on the
adoptees' years of education and college status. Each additional child in the family reduces
adoptees' educational attainment by .12 years. Mother's drinking status has a large effect on
adoptees' drinking.18
In contrast none of the income measures matter for adoptee's education or
income. In addition to the specifications shown here, I tried all possible combinations of income
from the survey (both at adoption and now), income from Holt records, and various measures of zip
code income. One response to this negative result might be that the income measures are too noisy.
However, this is not entirely persuasive because the family income measures are strongly correlated
with each other and with outcomes for the nonadoptees.
Furthermore there is a literature associated with Mayer [1997] and Blau [1999] which argues
that in the U.S., large changes in income result in only small increases in child test scores and
educational attainment. These authors suggest that other aspects of family environment are much
24
more important in determining child outcomes than is income per se. My results from the Holt data
lead to the same conclusion.
One reasonable hypothesis to explain the results in Table VI is that quantity and quality of
parental attention matters a tremendous amount for the adoptees outcomes. Each additional child in
the family reduces the amount of parental attention available per child. Increases in birth order also
have a negative but statistically insignificant effect on educational attainment for the adoptees.19
Harris [1998] famously developed the thesis that parental attention is not a key input into child
outcomes. One part of her evidence is the result that by adulthood, adoptee IQ is largely unaffected
by adoptive family environment and she infers from this that effects on education and occupation
may also be small. In contrast, this paper shows that adoptive family environment has large effects
on economic outcomes.
Harris argues that peer and neighborhood influences are the primary determinant of why
children turn out the way they do. However, Duncan, Boisjoly and Mullan Harris [2001] point out
that within family resemblance on outcomes (for achievement and delinquency) is much stronger
than within school or within neighborhood resemblance. Their upper bound on the potential scope
for peer and school influences is modest relative to the upper bound for the influence of genes and
shared family environment. My results are consistent with Duncan et al in that I do not find that zip
code income, zip code education measures, urban density, or percent black have statistically
significant impacts on adoptee outcomes.20
In contrast characteristics about the adoptive family
itself (namely parental education and family size) are strongly correlated with adoptee outcomes.
18
Using father's education and father's drinking status instead delivers similar coefficients to using the reports for the
mother. (Results not shown).
19
Not shown. I tried defining adoptee birth order in the adoptive family in various ways including using actual birth
years for all children in the family, and using the order of arrival in the household.
25
To claim that the effects of family environment found here work strictly through peer effects
(instead of parents) would require peer effects that are orders of magnitude larger than any of the
modern peer effects studies like Hoxby [2000], Sacerdote [2001], or Angrist and Lang [2004] that
focus on isolating the causal impact of peers.
In Table VII I proceed to explicit estimates of treatment effects from assignment to a high
education, small family via equation (3). I also present the effect from a one standard deviation
change in the index of family environment using the behavioral genetics framework. In each row,
Columns (1) and (2) show coefficients on dummy variables from a single regression. I regress the
adoptee's outcome on a dummy for being assigned to a family in group 2 (neither small, high
education nor large less education) or to group 1 (a high education family with three or fewer
children). The omitted category consists of adoptees in group 3, i.e. those with four or more
children and in which neither parent has a college degree. These regressions include controls for
adoptee gender, adoptee age, and adoptee cohort.
Column (3) reproduces the "effect" of assignment to a high education small family for the
nonadoptees in the sample. This is to provide a basis for comparison and to show how effect sizes
change when genes are covarying with family environment and parents are raising their biological
child. Column (4) is the effect from a one standard deviation change in the index of family
environment.
Assignment to a small, high education family relative to a lesser educated, large family
increases educational attainment by .75 years and raises the probability of graduating from college
20
In the regressions reported in Table VI, I only include zip code income. But all zip code level measures have small
and insignificant effects. Admittedly unlike the Duncan et al paper, this may not be a particularly powerful test of
whether neighborhoods matter.
26
by 16.1 percentage points. The probability of graduating from a U.S. News Ranked college is
increased by 23.1 percentage points relative to a mean of 37.3. These effects are similar to the
effects of a one standard deviation increase in the index of family environment, using the variance
decomposition implied by the behavioral genetics model. The latter approach yields effects of .85
additional years of education, a 17.9 percentage point increase in college graduation probability,
and a 22.4 percentage point increase in the probability of graduating from a U.S. News Ranked
college. These effects from family environment strike me as quite large and I provide some context
for this statement in the discussion section below.
The point estimates for the effect of a high education small family on child's family income
(an increase of 11.3 percent) is also relatively large but the coefficient is not statistically significant.
A one standard deviation change in the index of family environment is associated with a 26.3
percent increase in adoptee's family income.
There are also statistically significant treatment effects on drinking behavior. Assignment to
a small, high education family raises the drinking rate by 17.8 percentage points. A one standard
deviation change in the index of family environment is associated with a 28 percentage point
change in the drinking rate. The literature on the relationship between drinking and income
suggests a generally positive but non-linear relationship between own drinking and own income.
Moderate levels of drinking are strongly positively associated with income, education and socioeconomic
status. Auld [2005] finds a 10 percent wage premium for moderate drinking relative to
abstinence. My positive treatment effects on drinking from assignment to a small, high education
family seem to be in the same spirit as the existing literature.
27
Assignment to a high education, small family reduces the adoptees' number of children (at
the time of the survey). Adoptees of high education, small families have .20 fewer children relative
to adoptees assigned to less educated large families. Some of the observed treatment effect might
be delayed fertility as opposed to reduced fertility. Given that the average age of the adoptees is 28,
I do not explore effects on completed fertility.
IV.C. Transmission Coefficients and Comparison to Other Adoption Studies
Table VIII shows results from my third empirical approach, namely calculating transmission
coefficients from parents to children as in equation (4) and (4A). I include these since the
economics literature generally measures parent-child connections using transmission coefficients.
Transmission coefficients do not necessarily have a direct causal interpretation, but rather are a
convenient and standard way to measure how changes in the child's outcome are associated with
changes in the parental characteristic.
For educational attainment, each additional year of mother's education is associated with .09
years of education for adoptees and .32 years for the nonadoptees.21
The ratio of these two numbers
may be a useful summary statistic: For the measured transmission of education from mothers to
children, roughly 28 percent of this is working directly through nurture.
Body mass index and height exhibit strong transmission from mothers to children for the
nonadoptees but exhibit no transmission for the adoptees. Again, drinking has the appearance of a
21
The income transmission coefficients are actually from the second stage of instrumental variables regressions. This is
done in an attempt to clean up measurement error in income. I instrument for parental income as measured by the
survey using parental income from Holt records and median income in the parents' zip code. I do not devote space to
discussing the findings on income transmission for several reasons. First, I have only self reports of single years of
income. This can greatly bias the estimated coefficients as discussed by Solon [1999] , Mazumder[2002] and others.
Second, age of the parents and children matters a great deal too (Solon and Haider [2004]). The adoptees and
nonadoptees both have different average ages AND are significantly younger than the subjects in careful studies of
income transmission.
28
social outcome which is transmitted equally well to adoptees and nonadoptees. The coefficients are
.21 and .24 respectively. In Figure V, I plot the adoptee transmission coefficients against the
nonadoptee transmission coefficients. As with the sibling correlations in Figure IV, certain
outcomes such as height, body mass index and years of education are significantly above the 45
degree line while drinking is almost directly on the 45 degree line.
Comparing results for the Holt data to results from other adoption data sets may tell us
something about the degree to which sorting of adoptees into families biases the estimates of
nurture's impact in existing studies. The four adoption datasets I consider are the Holt data, the
Swedish data as analyzed by BLP[2006], the Wisconsin Longitudinal Survey as analyzed by Plug
[2004], and the NLSY79 data as analyzed by Sacerdote [2002]. The adoptees in BLP's study are
ages 35-37 in 1999. The adoptees in Plug's study are age 23 and older. For the NLSY data I use
adoptees ages 28-36 in 1993.
For the adoptees, the Holt study and the Swedish (BLP) study yield similar transmission
coefficients. The transmission coefficient for years of education is .089 for Holt and .074 for BLP.
BLP's estimate of transmission of college status is somewhat larger than my estimate, but the
estimates are within 1.2 standard deviations. A reasonable interpretation of the similarity of
educational transmission coefficients in the Holt and Swedish samples is that any selection bias in
the Swedish data is not particularly severe.22
In contrast the studies using the NLSY and WLS data find much larger coefficients for
transmission of education from mothers to adoptees. These studies find coefficients of roughly .28.
22
Obviously fundamental differences between the U.S. and Sweden in the transmission process and differences in
"restriction of range" among the SES of adoptive parents in the two samples could be offsetting some of the selection
effects.
29
The most natural explanation for this difference is that there is strong positive (and unknown)
selection of adoptees into adoptive families in those data sets. This would tend to bias the
coefficient for the adoptees upward, and probably towards the coefficients for nonadoptees.
V. Discussion and Conclusion
Analysis of the Holt data reveals several useful facts. First there are large treatment effects
from adoptees being assigned to high education, small families. Adoptees in such families have an
additional .75 years of education and are 16.1 percentage points more likely to graduate from
college, relative to adoptees in larger, less educated families. The treatment effect on graduating
from a U.S. News ranked college is 23.1 percentage points. A one standard deviation change in the
index of family environment causes an increase in years of education of .85, an increase in the
college graduation rate of 17.9 percentage points, and a 26.3 percentage point increase in the rate of
graduating from a U.S. News ranked college.
Shared family environment can explain roughly 16 percent of the variation in educational
attainment and 14 percent of the variation in the adoptee's family income. Genetic factors explain
44 percent of the variation in educational attainment and 33 percent of the variation in income. In
contrast, social outcomes like drinking, smoking, and marital status appear to be very nurture based.
For drinking, family environment explains 34 percent of the variation while genes explains only 6
percent. The selectivity of the college attended is also strongly influenced by nurture.
Consistent with existing work on adoptees, shifts in family environment do not have a large
influence on body mass index or probability of being overweight. Family environment explains
almost none of the variation in weight (pounds) while genetics explains 46 percent. Mother's BMI
30
is transmitted to adoptees BMI with a coefficient of .002, but to nonadoptees with a coefficient of
.221.
The study design does not allow me to make definitive statements about the causal
mechanisms underlying the treatment effects found. However, I show that family size and parental
education are much more strongly correlated with the adoptees' outcomes than are any of the
measures of family income, zip code income, or other zip code characteristics. This finding
suggests that the quality and quantity of parental attention may be two of the factors underlying the
influence of family environment on outcomes, and that these factors matter more than income.
Parental drinking is the factor most strongly associated with adoptee drinking, which of course
suggests that adoptees pick up the behaviors they see modeled at home.
As Harris notes23
, assignment of an adoptee to one family type versus another is one of the
largest environmental interventions one might imagine, since family income, parental education,
neighborhood peer quality and school quality are all shifted simultaneously from very early
childhood onward. My estimated treatment effects provide a context for the possible effects from a
large class of other environmental interventions or policy changes of interest. For example,
Dynarski [2005] finds that large college tuition subsidies raise the rate of college graduation by 3
percent. This is roughly one fifth the effect of sending an adoptee to a high education small family,
or roughly equivalent to a .17 standard deviation increase in an index of family environment.
Neal [1997], Evans and Schwab [1995], Altonji, Elder and Taber [2000] find that attending
a Catholic school raises the probability of college going by 12-13 percent. This translates into
perhaps a 6 percent effect on the college graduation rate if about half of these college entrants
31
obtain a four year degree. Thus the Catholic schooling effect is roughly equivalent to a .34 standard
deviation increase in the index of family environment.
The effects from a one standard deviation move in family environment may also provide a
comparison basis for effects observed in other commonly observed environmental interventions
such as altering a child's school peer group (Hoxby [2000], Angrist and Lang [2004]), switching the
child's school (Rouse [1998], Nechyba and Vigdor [2003], Cullen, Jacobs and Levitt [2005]), or
switching the entire neighborhood and school as in the Moving to Opportunity experiment (Katz,
Kling and Liebman [2001], Kling, Ludwig and Katz [2005], Ludwig, Duncan Hirschfield [2001]).24
The effect measured here from having a more educated mother would seem to place on upper bound
on interventions that exogenously raise parental education as in Black, Devereux and Salvanes
[2005] and Currie and Moretti [2003]. I say this because the exogenous shock in the Holt data
changes not only parental education but many other covariates.
The effects measured in the Holt data could useful for understanding black-white education
(and possibly wage) differentials. The black-white gap in years of education is .78 years and the
gap in the college completion rate is 15.4 percentage points. These numbers are equal to the effects
of assigning an adoptee to a high education, small family relative to a large, less educated family.
Or put another way, the black-white gap in education is equivalent to about a one standard deviation
move in the index of family environment. Do we believe that the average environments faced by
black and white children differ by as much one standard deviations of the family environment index
23
1998, pp. 260-261. This is similar in spirit to how Hernnstein and Murray [1994] attempt to tie the BG literature to
possible effects from environmental interventions.
24
I have in mind here a very specific set of family environment and schooling interventions. Namely ones that seek to
influence child outcomes by raising family income, neighborhood quality, peer quality or school quality. I'm also
thinking of those environmental interventions within the context of a rich country like the U.S..
32
in the Holt data? It certainly seems at least possible given the mountains of evidence on the
environmental disadvantages faced by black children in the U.S.
In conclusion, the study extends our understanding of how differences in family
environment lead to differences in outcomes for children. Rather than contradicting existing
findings on the effects of environment, these results contribute to a consistent picture as to which
outcomes are most affected by family environment, how much they are affected, and what some of
the underlying mechanisms might be. Hopefully researchers will use the results here and the raw
data to further our understanding of why children turn out the way they do.
33
References
Altonji, Joseph G., Todd E. Elder and Christopher R. Taber, “Selection of Observed and
Unobserved Variables: Assessing the Effectiveness of Catholic Schools,” National Bureau
of Economic Research Working Paper No. 7831, 2000.
Angrist, Joshua D. and Kevin Lang, "Does School Integration Generate Peer Effects? Evidence
from Boston's Metco Program," American Economic Review, XCIV (2004), 1613-34.
Auld, M Christopher, "Smoking, Drinking, and Income," Journal of Human Resources, XC (2005),
505-18.
Behrman, Jere R. and Paul Taubman, “Is Schooling 'Mostly in the Genes'? Nature Nurture
Decomposition Using Data on Relatives,” Journal of Political Economy, XCVII (1989),
1425-1446.
Behrman, Jere R., Mark R. Rosenzweig and Paul Taubman, “Endowments and the Allocation of
Schooling in the Family and in the Marriage Market,” Journal of Political Economy, CII
(1994), 1131-1174.
Björklund, Anders, Mikael Lindahl, Erik Plug, "Intergenerational Effects in Sweden: What Can
We Learn from Adoption Data?" The Quarterly Journal of Economics, forthcoming (August
2006).
Black, Sandra, Paul J. Devereux and Kjell G. Salvanes, "Why the Apple Doesn't Fall Far:
Understanding Intergenerational Transmission of Human Capital," The American Economic
Review, XCV (2005), 437-449.
Black, Sandra, Paul J. Devereux and Kjell G. Salvanes, "The More the Merrier? The Effect of
Family Size and Birth Order on Children’s Education," The Quarterly Journal of
Economics, CXX (2005b), 669-700.
Blau, David M., "The Effect Of Income On Child Development," The Review of Economics and
Statistics, vol. LXXXI (1999), 261-276.
Bowles, Samuel, Herbert Gintis, Melissa Osborne Groves, eds., Unequal Chances: Family
Background And Economic Success, (Princeton and Oxford: Princeton University Press,
2005).
Bouchard, T.J., and M. McGue, "Familial Studies of Intelligence: A Review," Science, CCXII,
1055-1059.
Cardon, Lon R., "Height, Weight and Obesity," in DeFries, John C., Robert Plomin, and David W.
Fulker, Nature and Nurture During Middle Childhood, (Oxford, UK: Blackwell Press,
1994).
Cloninger C.R., M. Bohman , and S. Sigvardson , "Inheritance Of Alcohol-Abuse - Cross-Fostering
Analysis Of Adopted Men," Archives Of General Psychiatry XXXVIII (1981), 861-868.
34
Cullen, Julie Berry, Brian Jacob, and Steven Levitt, “The Impact of School Choice on Student
Outcomes: An Analysis of the Chicago Public Schools.” Journal of Public Economics,
LXXXIX (2005), 729-760.
Currie, Janet and Enrico Moretti, "Mother's Education and the Intergenerational Transmission of
Human Capital: Evidence from College Openings," Quarterly Journal of Economics,
CXVIII (2003), 1495 – 1532.
Darwin, Charles, On the Origin of Species by Means of Natural Selection, (New York: Appleton,
1859).
Das, Mitali and Tanja Sjogren, "The Intergenerational Link in Income Mobility: Evidence From
Adoptions," Economics Letters, LXXV, 55-60.
DeFries, John C., Robert Plomin, and David W. Fulker, Nature and Nurture During Middle
Childhood, (Oxford, UK: Blackwell Press, 1994).
Dickens, William and James R. Flynn, "Heritability Estimates versus Large Environmental Effects
The IQ Paradox Resolved, " Psychological Review, CVIII (2001), 346-369.
Duncan, Greg J, Johanne Boisjoly, Kathleen Mullan Harris, "Sibling, Peer, Neighbor, and
Schoolmate Correlations as Indicators of the Importance of Context for Adolescent
Development," Demography, XXXVIII (2001),437-447.
Dynarksi, Susan, "Building the Stock of College Educated Labor," National Bureau of Economic
Research Working Paper No. 11604, 2005.
Evans, William N. and Robert M Schwab, “Finishing High School and Starting College: Do
Catholic Schools Make a Difference?” The Quarterly Journal of Economics, 1995, 941-974.
Goldberger, Arthur S, "The Genetic Determination of Income: Comment," American Economic
Review, CXVIII (1978), 960-69.
Grilo, Carlos M. and Michael F. Pogue-Geile, "The Nature Of Environmental Influences On Weight
And Obesity: A Behavior Genetic Analysis," Psychological Bulletin, CX (1991), 520-537.
Harris, Judith Rich, The Nurture Assumption: Why Children Turn out the Way They Do, (New
York: The Free Press, 1998).
Herrnstein, Richard J. and Charles Murray, The Bell Curve: Intelligence and Class Structure in
American Life, (New York: The Free Press, 1994).
Hoxby, Caroline M., "The Effects of Class Size on Student Achievement: New Evidence from
Population Variation," Quarterly Journal of Economics, CXV (2000), 1239-1285.
Jacob, Brian A, "Public Housing, Housing Vouchers, and Student Achievement: Evidence from
Public Housing Demolitions in Chicago," American Economic Review, XCIV(2004), 233-
58.
35
Jencks, Cristopher, Inequality: A Reassessment of the Effects of Family and Schooling in America,
(New York: Basic Books, 1972.)
Jencks, Cristopher. and Marsha Brown, “Genes and Social Stratification;” in P. Taubman (ed.),
Kinometrics: The Determinants of Economic Success Within and Between Families, (New
York: North Holland Elsevier, 1979).
Katz, Lawrence F., Jeffrey R. Kling, and Jeffrey B. Liebman, "Moving to Opportunity in Boston:
Early Results of a Randomized Mobility Experiment," Quarterly Journal of Economics,
CXVI (2001), 607-654.
Kling, Jeffrey R, Jens Ludwig and Lawrence F. Katz, "Neighborhhod Effects on Crime for Female
and Male Youth: Evidence from a Randomized Housing Voucher Experiment," The
Quarterly Journal of Economics, CXX (2005), 87-130.
Lichtenstein, Paul, Nancy L. Pedersen, and G.E. McClearn, "The Origins Of Individual-Differences
In Occupational-Status And Educational-Level - A Study Of Twins Reared Apart And Together,"
Acta Sociologica XXXV (Mar 1992),13-31.
Loehlin, John C., “Partitioning Environmental and Genetic Contributions to Behavioral
Development,” American Psychologist, XCIV (1989), 1285-1292.
Loehlin, John C., Joseph M. Horn, and Lee Willerman, “Differential Inheritance of Mental Abilities
in the Texas Adoption Project,” Intelligence, XIX (1994), 325-336.
Loehlin, John C., Joseph M. Horn, and Lee Willerman, “Personality Resemblances Between Unwed
Mothers and Their Adopted-Away Offspring,” Journal of Personality and Social
Psychology, XLII (1982), 1089-1099.
Loehlin, John C., Lee Willerman, and Joseph M. Horn, “Personality Resemblance in Adoptive
Families: A 10-Year Followup,” Journal of Personality and Social Psychology, LIII (1987),
961-969.
Ludwig, Jens, Greg J. Duncan, and Paul Hirschfield, "Urban Poverty and Juvenile Crime: Evidence
from a Randomized Housing Mobility Experiment," Quarterly Journal of Economics, CXVI
(2001), 655-680.
Maughan, Barbara, Stephen Collishaw, and Andrew Pickles, “School Achievement and Adult
Qualification Among Adoptees: A Longitudinal Study,” Journal of Child Psychology and
Psychiatry, XXXIX (1998), 669-685.
Mazumder, Bhashkar, "The Apple Falls Even Closer to the Tree than We Thought," in Bowles,
Samuel, Herbert Gintis, Melissa Osborne Groves, eds., Unequal Chances: Family
Background And Economic Success, (Princeton and Oxford: Princeton University Press,
2005.)
Mayer, Susan E, "What Money Can't Buy: Family Income And Children's Life Chances,"
(Cambridge and London: Harvard University Press, 1997).
36
Neal, Derek, “The Effects of Catholic Secondary Schooling on Educational Attainment,” Journal of
Labor Economics, XV (1997), 98-123.
Nechyba, Thomas and Jacob Vigdor, "Peer Effects in North Carolina Public Schools," Mimeo,
Duke University, 2003.
Plomin, Robert, John C. DeFries, David W. Fulker, Nature And Nurture During Infancy And Early
Childhood, (Cambridge, England; New York: Cambridge University Press, 1988).
Plomin, Robert, John C. DeFries, Gerald E. McClearn, Peter McGuffin, Behavioral Genetics 4th
Edition, (New York: Worth Publishers, 2001).
Plug, Erik, "Estimating the Effect of Mother's Schooling on Children's Schooling Using a Sample of
Adoptees," American Economic Review, XCIV (2004), 358-368.
Plug, Erik and Wim Vijverberg, "Schooling, Family Background and Adoption: Is It Nature or Is It
Nurture?," Journal of Political Economy, CXI (2003), 611-641.
Ridley, Matt, Nature Via Nurture, New York: Harper Collins, 2003.
Rouse, Cecilia E., "Private School Vouchers and Student Achievement: An Evaluation of the
Milwaukee Parental Choice Program," Quarterly Journal of Economics, CXIII (1998), 553-
602.
Sacerdote, Bruce, "The Nature and Nurture of Economic Outcomes," American Economic Review,
XCII (2002), 344-48.
Sacerdote, Bruce I., "Peer Effects With Random Assignment, " Quarterly Journal of Economics,
CXVI (2001), 681-704.
Sacerdote, Bruce I., "How Large Are the Treatment Effects From Family Environment (Working
Paper), " mimeo, Dartmouth College, (2005).
Scarr, Sandra and Richard Weinberg “The Influence of Family Background on Intellectual
Attainment,” American Sociological Review, XVIII (1978), 674-692.
Scarr, Sandra, and Richard A. Weinberg, "Educational And Occupational Achievements Of
Brothers And Sisters In Adoptive And Biologically Related Families," Behavior Genetics XXIV
(July 1994), 301-325.
Solon, Gary, “Intergenerational Mobility in the Labor Market,” in Orley Ashenfelter and David
Card eds., Handbook of Labor Economics, Vol 3. Elsevier Science B.V., 1999.
Solon, Gary and Steven Haider, "Life-Cycle Variation in the Association between Current and
Lifetime Earnings," mimeo 2004.
Stoolmiller, Michael, "Implications of the Restricted Range of Family Environments for Estimates
of Heritability and Nonshared Environment in Behavior-Genetic Adoption Studies,"
Psychological Bulletin, CXV (1999), 392-409.
37
Taubman, Paul “Role of Parental Income in Educational Attainment,” American Economic Review,
LXXIX (1989), 57-61.
Teasdale T.W and David R. Owen, "Heredity And Familial Environment In Intelligence And
Educational Level—A Sibling Study," Nature CCCIX (June 1984), 620-622.
Teasdale T.W and David R. Owen, "The Influence Of Paternal Social-Class On Intelligence And
Educational-Level In Male Adoptees And Non-Adoptees," British Journal Of Educational
Psychology CVI (Feb 1986) 3-12.
Turkheimer, Eric, Andreana Haley, Mary Waldron, Brian D'Onofrio, and Irving I. Gottesman
"Socioeconomic status modifies heritability of IQ in young children," Psychological Science, XIV
(November 2003), 623-628.
Vogler, G. P., T. I. A. Sorensen, A. J. Stunkard, M. R. Srinivasan, and D. C. Rao, “Influences of
Genes and Shared Family Environment on Adult Body Mass Index Assessed in an Adoption
Study by a Comprehensive Path Model,” International Journal of Obesity XIX(1995),40–
45.
Wooldridge, Jeffrey M. "Asymptotic Properties of Weighted M-Estimators for Variable Probability
Samples," Econometrica, CXVII (1999), 1385-1406.
38
Table I
Evidence of Random Assignment Using Administrative Records
(1) (2) (3) (4) (5) (6) (7)
Adoptee's
Age at
Arrival in
U.S.
Weight
When
Entered
Holt
System
(Lbs)
Height
When
Entered
Holt
System
(Inches)
Adoptee
is Male
Birth
Mother
Was
Married
Birth
Mother's
Age at
Adoptee's
Birth
Birth
Mother's
Highest
Grade
Completed
Log 0.001 0.310 0.188 0.017 0.060 0.556 0.171
Family
Income
(0.127) (0.258) (0.225) (0.022) (0.081) (1.784) (0.968)
Father's -0.006 0.009 -0.019 0.006 -0.007 0.358 -0.094
Years of
Education
(0.010) (0.043) (0.036) (0.004) (0.015) (0.376) (0.158)
Mother's -0.018 -0.037 0.014 0.003 0.001 -0.128 0.159
Years of
Education
(0.015) (0.067) (0.040) (0.005) (0.019) (0.545) (0.172)
Log Median 0.145 0.201 0.149 -0.041 0.061 -3.422 -0.119
Income in
Zipcode in
1980
(0.203) (0.285) (0.232) (0.029) (0.111) (2.523) (1.106)
Observations 2158 2156 2157 2161 126 99 81
R-squared 0.143 0.704 0.640 0.232 0.345
F or Chi2
test for
Parental
Coeffs=0
1.10 0.72 0.48 6.47 1.48 0.88 0.23
Prob > F,
Prob > Chi2
0.354 0.580 0.753 0.166 0.830 0.479 0.920
Robust standard errors in parentheses: standard errors are clustered at the family level.
* significant at 5%; ** significant at 1%
All data are from Holt Records and include children of the families who responded and did not respond to the survey. I
regress pre-adoption characteristics for the child on adoptive family characteristics. Regressions include dummies for
the child's age and dummies for the year in which the child entered the Holt system. Information on birth mothers
(columns (5)-(7)) is only available in a limited number of cases. The final rows show the F statistic for the joint
significance of the four parental background variables. Columns (4)-(5) are probit regressions. For these, ∂y/ ∂x is
reported and I use a chi square test for the joint significance.
39
Table II
Evidence of Random Assignment Within Analysis Sample
(1) (2) (3) (4)
Weight at initial
social history lbs
Height at initial
social history
inches
Child's Age at
Arrival
Child is Male
Mother's Years of 0.009 -0.045 -0.013 0.009
Education (0.088) (0.078) (0.012) (0.007)
Father's Years of -0.047 0.004 -0.000 0.001
Education (0.073) (0.069) (0.010) (0.005)
Log Parent's -0.113 0.044 0.013 0.011
Household
Income
(0.276) (0.243) (0.041) (0.022)
Mother's BMI -0.032 -0.034 0.002 0.001
(0.047) (0.047) (0.005) (0.003)
Father's BMI -0.046 -0.061 -0.009 -0.002
(0.037) (0.034) (0.006) (0.003)
Mother Drinks -0.000 0.043 -0.031 -0.008
(0.456) (0.417) (0.059) (0.033)
Father Drinks 0.298 -0.248 -0.050 -0.000
(0.472) (0.426) (0.063) (0.035)
Mother's Height
Inches
0.082 0.062 0.001 -0.001
(0.070) (0.061) (0.008) (0.006)
Father's Height
Inches
0.053 0.071 0.011 0.004
(0.063) (0.056) (0.007) (0.004)
Constant 2.138 18.040 0.507 0.636
(5.993) (5.437)** (0.700) (0.609)
Observations 989 1038 1040 1056
R-squared 0.188 0.320 0.266 0.062
F test, Parental
Coeffs=0
0.78 1.00 0.60 0.55
Prob > F 0.635 0.441 0.800 0.838
Robust standard errors in parentheses: standard errors are clustered at the family level.
* significant at 5%; ** significant at 1%
I regress child pre-treatment characteristics on adoptive family characteristics. This is the sample of adoptees whose
families responded to the survey and for whom we have the relevant variables. All of the right hand side measures are
taken from the survey data, but similar results obtain if we use parental education and income as reported in Holt
Records. All columns include dummies for child's age and for year of admission to Holt. The final rows show the F
test for the hypothesis that all the coefficients on the parental characteristics are zero.
40
Table III
Summary Statistics
Adoptees Nonadoptees
Variable Obs Mean Std. Dev. Obs Mean
FOR THE CHILDREN
Child is Male 1650 0.295 0.456 1196 0.622
Child's Current Age 1650 28.215 4.557 1196 32.292
Child's Age at Arrival in U.S. 1640 1.369 0.826
Child's Years of Education 1256 15.088 2.153 1051 15.914
Child Has 4+ Years College? 1256 0.576 0.494 1051 0.713
Child Graduated from a U.S.
News Ranked College?
1256 0.373 0.484 1051 0.469
Acceptance Rate of College 725 0.697 0.174 598 0.672
Child's Family Income 1209 49.268 35.141 1025 64.239
Log(Child's Family Income) 1209 3.648 0.740 1025 3.932
Child is Married 1642 0.386 0.487 1156 0.633
Child's Number of Children 1562 0.520 0.905 1125 1.083
Child's BMI 1590 23.113 3.733 1130 24.007
Child is Overweight? 1590 0.240 0.427 1130 0.343
Child is Obese? 1590 0.061 0.239 1130 0.061
Child Smokes? 1649 0.230 0.421 1161 0.115
Child Drinks? 1635 0.593 0.491 1149 0.687
Child Ever Had Asthma? 1650 0.089 0.285 1188 0.089
FOR THE PARENTS
Mother's Years of Education 1650 15.122 2.456 1180 15.285
Mother Has 4+ Years College 1650 0.528 0.499 1188 0.547
Father's Years of Education 1635 15.908 2.879 1171 16.272
Father Has 4+ Years College 1650 0.618 0.486 1179 0.673
Income at Time of Adoption
(Survey)
1624 32.472 23.646 1166 33.649
Income at Time of Adoption
(Holt Records)
1218 16817.780 9893.756 939 16675.520
Log Family Income (Holt
Records)
1216 9.591 0.534 935 9.581
Mother is Overweight? 1574 0.463 0.499 1132 0.436
Mother Smokes? 1629 0.033 0.177 1162 0.023
Mother Drinks? 1624 0.526 0.499 1161 0.571
Both Parents Have College
Degrees And Family Has
Three Or Fewer Children
1627 0.273 0.446 1170 0.232
Neither Parent Has A College
Degree and Family Has Four
Or More Children
1627 0.128 0.334 1170 0.170
These are the means and standard deviations for the sample. Children are ages 19-40 in 2004. All adoptees are Korean American
Adoptees placed by Holt. Child's family income, years of education, college status are reported for children ages 25-40.
Graduation from a U.S. News Ranked College and Acceptance Rate are determined by matching the child's reported undergraduate
institution with the 2004 U.S. News Rankings. (Non college graduates and graduates from unranked colleges are assigned a zero
for U.S. News Ranked Dummy.) Sample sizes vary due to differential reporting on the surveys and in Holt Records.
41
Table IV
Correlations In Outcomes Among Pairs of Adoptive Siblings
And Pairs of Biological Siblings
Outcome Adoptive
Sibling
correlation
Biological
Sibling
correlation
N
adoptive
N
Biological
Has 4 Years of College 0.135 0.338 1360 578
Highest Grade Completed 0.157 0.378 1360 578
Family Income 0.110 0.277 1314 554
Log (Family Income) 0.139 0.301 1314 554
Drinks 0.336 0.363 1903 640
Smokes 0.152 0.289 1938 654
Height 0.014 0.443 1910 646
Weight 0.044 0.273 1822 629
BMI 0.115 0.269 1821 629
Overweight 0.087 0.173 1821 629
Attended US News Ranked
School 0.249 0.416 1360 578
Acceptance Rate of School 0.337 0.460 560 245
Married? 0.076 0.048 1917 650
Number of Children 0.105 0.203 1802 633
I form all possible pairs of siblings within the data set. I purge the outcome variables of variation due to age dummies,
cohort dummies and gender. I report the correlation in outcomes for adoptive sibling pairs and biological sibling pairs.
Adoptive sibling pairs occur when either one or both of the siblings in a family are adoptees (and the adoptees do not
share a biological mother or father). Biological sibling pairs are those that share a biological mother and father who are
also the "nurturing" parents. All of the correlations are statistically different from zero at the 1 percent level except for
height and weight among adoptive siblings.
42
Table V
Proportion of Outcome Variance Explained By Heritability, Shared Family
Environment and Non-shared Environment Using A Simple Behavioral Genetics
Model
Outcome Proportion
Explained By
Nurture
(Shared
Family
Environment)
Proportion
Explained By
Nature
(Heritability)
Unexplained
Portion
(Non-
Shared
Evironment)
Has 4 Years of College 0.135 0.406 0.459
Highest Grade Completed 0.157 0.443 0.400
Family Income 0.110 0.334 0.556
Log (Family Income) 0.139 0.324 0.537
Drinks 0.336 0.055 0.609
Smokes 0.152 0.273 0.575
Height 0.014 0.858 0.128
Weight 0.044 0.458 0.498
BMI 0.115 0.308 0.577
Overweight 0.087 0.172 0.741
Attended US News Ranked
School 0.249 0.335 0.417
Acceptance Rate of School 0.337 0.245 0.418
Married? 0.076 -0.056 0.979
Number of Children 0.105 0.196 0.699
I use the simple BG model described in the text to decompose the variance in each outcome into the portions
attributable to genes (heritability), shared family environment, and non-shared family environment (i.e. the unexplained
portion). See equations (2), (2A) and the paragraph that follows.
43
Table VI
Regression of Adoptee Outcomes on Family Characteristics
(1) (2) (3) (4) (5) (6) (7)
Child's
Years of
Education
Child
Has 4+
Years
College
Log
Child's
Household
Income
Child's
BMI
Child
Overweight
Child
Drinks
(yes/no)
Number
of
Children
Mother's 0.097 0.023 0.003 -0.074 -0.007 0.010 -0.017
Years of
Education
(0.027)** (0.007)** (0.010) (0.055) (0.006) (0.006) (0.009)
Number of -0.120 -0.026 -0.044 0.106 0.011 0.001 0.016
Children (0.050)* (0.012)* (0.017)* (0.093) (0.010) (0.011) (0.018)
Log Parent's -0.057 -0.001 0.027 -0.229 -0.031 0.008 0.085
Household
Income
(0.098) (0.025) (0.038) (0.197) (0.021) (0.023) (0.035)*
Log (Zip Code -0.133 0.045 -0.015 -0.232 -0.044 0.045 -0.179
Income) (0.286) (0.069) (0.104) (0.502) (0.060) (0.070) (0.118)
Child is Only -0.058 -0.010 -0.026 -0.222 -0.042 0.026 -0.030
Adoptee in
Family
(0.153) (0.037) (0.054) (0.301) (0.031) (0.034) (0.058)
Fraction Girls 0.078 0.042 -0.241 -0.095 -0.060 0.013 -0.225
in Family (0.296) (0.073) (0.104)* (0.626) (0.064) (0.065) (0.105)*
Mother Drinks 0.097 -0.009 0.016 -0.248 -0.016 0.188 -0.085
(0.138) (0.034) (0.047) (0.267) (0.029) (0.030)** (0.046)
Mother's BMI -0.028 -0.005 -0.008 0.002 -0.001 0.001 0.003
(0.014)* (0.003) (0.005) (0.024) (0.003) (0.003) (0.005)
Child is Male -0.633 -0.145 -0.285 1.704 0.192 0.090 -0.247
(0.163)** (0.039)** (0.055)** (0.283)** (0.035)** (0.032)** (0.049)**
Constant 15.412 4.082 26.717 1.446
(1.169)** (0.450)** (2.414)** (0.478)**
Observations 1173 1173 1136 1138 1138 1532 1463
R-squared 0.081 0.124 0.080 0.220
I regress adoptee's outcome on a set of the adoptive family characteristics. Each column is a separate regression.
Columns (2), (5), and (6) are probits and ∂y/∂x is reported. A full set of age dummies and dummies for year of
admission to Holt are included in all columns.
Robust standard errors in parentheses: I cluster at the family level.
* significant at 5%; ** significant at 1%
44
Table VII
Treatment Effects From Assignment to High Education, Small Family
Treatment
Effect "Middle
Group" of
Families Vs.
Large, Less
Educated
Treatment
Effect High
Education
Small
Family Vs
Large, Less
Educated
NON
ADOPTEES
High Ed Small
Family Vs Large,
Less Educated
Effect from a 1
Standard
Deviation
Change in
Family
Environment
Index
Child's Years of Education 0.314
(0.226)
0.749
(0.245)**
2.157
(0.264)**
0.845
Child Has 4+ Years College 0.060
(0.056)
0.161
(0.057)**
0.317
(0.031)**
0.179
Log Child's Household Income 0.071
(0.081)
0.113
(0.089)
0.210
(0.089)*
0.263
Child Four Year College Ranked by U.S.
News
0.082
(0.052)
0.231
(0.060)**
0.365
(0.052)**
0.224
Acceptance Rate of Child's College -0.007
(0.035)
0.016
(0.036)
-0.053
(0.032)
0.098
Child Drinks (yes/no) 0.099
(0.050)*
0.178
(0.049)**
0.229
(0.041)**
0.280
Child Smokes (yes/no) 0.013
(0.044)
-0.006
(0.048)
-0.075
(0.024)**
0.162
Child's BMI -0.509
(0.460)
-0.941
(0.468)*
-0.929
(0.498)
1.224
Child Overweight -0.030
(0.047)
-0.077
(0.045)
-0.088
(0.048)
0.121
Child Obese -0.020
(0.023)
-0.044
(0.018)*
-0.037
(0.018)*
0.047
Child Has Asthma -0.005
(0.028)
0.013
(0.031)
-0.005
(0.034)
0.085
Number of Children -0.070
(0.099)
-0.199
(0.103)*
-0.580
(0.132)**
0.267
Child is Married 0.014
(0.050)
0.000
(0.056)
-0.092
(0.053)
0.123
I split the sample into three groups: High education small families are defined as those with three or fewer children in
which both the mother and father have a college degree (Type 1). Twenty seven percent of adoptees are assigned to
such a family. Large lesser educated families are defined as those with four or more children and where neither parent
has a college degree (Type 3). Thirteen percent of adoptees are assigned to such a family. The remaining families
(which are either small or have a parent with a college degree) are Type 2. Column (1) shows the coefficient on the
dummy for assignment to Type 2 relative to Group 3. Column (2) shows the coefficient on the dummy for assignment
to Type 1 (small high education) relative to Type 3 (large less educated).
Column (3) shows this Type 1 versus 3 "effect" for the non-adoptees. In a each row, the effects in columns (1) and (2)
are estimated together with a single regression while column (3) uses a separate regression. Column (4) shows the effect
for the adoptees from a one standard deviation move in an index of shared family environment. This is calculated by
taking the square root of the variance share explained by shared family environment in the previous table and
multiplying by the standard deviation of the outcome variable. I.e., R* σy =σyhat = predicted effect on the outcome from
a one standard deviation change in an index of family environment. Standard errors are corrected for within family
correlation (I cluster by family).
45
Table VIII
Transmission Coefficients from Parents to Children
For Adoptees and Non Adoptees
(1) (2)
Adoptees:
Transmission
Coefficient
Nonadoptees
Transmission
Coefficient
Years of Education (Mother to Child) 0.089 0.315
(0.029)** (0.038)**
Has 4+ Years College (Mother to Child) 0.102 0.302
(0.034)** (0.037)**
Log Household Income (Parents to 0.186 0.246
Child) (0.111) (0.080)**
Height Inches (Mother to Child) -0.004 0.491
(0.034) (0.049)**
Is Obese (Mother to Child) 0.003 0.108
(0.020) (0.034)**
Is Overweight (Mother to Child) -0.026 0.174
(0.029) (0.037)**
BMI (Mother to Child) 0.002 0.221
(0.025) (0.045)**
Smokes (0-1) (Mother to Child) 0.132 0.108
(0.088) (0.115)
Drinks (0-1) (Mother to Child) 0.210 0.244
(0.033)** (0.038)**
I regress the child's outcome on the corresponding outcome for the mother (or in the case of income, the parents). Each
cell is from a separate regression which also includes age dummies, dummies for year of admission to Holt, and a
dummy for the child being male. For income and education regressions I restrict the sample to children ages 25+. For
log (income), I attempt to correct for measurement error in parent's income by instrumenting for the survey measure of
parents' income using the parents' income measure reported in Holt records.
Robust standard errors in parentheses: I cluster at the family level
* significant at 5%; ** significant at 1%
46
Appendix Table I
Response Rates for The Mailings
Sent Responded Response Rate
Pilot 1000 252 0.252
Main mailing 2500 851 0.340
Re-survey 400 141 0.353
Mailing to
Children
653 359 0.550
This shows the response rates for the three mailings to adoptive parents and the one mailing to adoptees and their
siblings.
47
Appendix Table II
Probit for Response To Survey
Family Responded W/o Followup
Log Family Income -0.093
(0.026)**
Father's Years of Education 0.014
(0.005)**
Mother's Years of Education 0.017
(0.006)**
Log Median Income in Zipcode in 1980 -0.016
(0.042)
Percentage in Zip Code with 4+ Years College 1980 -0.056
(0.161)
Percent Black in Families' Zip Code 1980 -0.038
(0.111)
Observations 2138
F test, Parental Coeffs=0 40.43
Prob > chi2 0.000
Standard errors in parentheses
* significant at 5%; ** significant at 1%
I regress a dummy for responding to the survey on adoptive family characteristics as measured in the Holt Records. The
dummy is for responding in the original round of the survey, regardless of what the family did in the follow up round. I
run a probit and report ∂y/∂x. Sample consists of all families who were sent a survey. Regression includes dummies
for child's age and year of admission to Holt.
48
Appendix Table III
Correlation Between Parent and Child Responses for Child Outcomes
Child responses are obtained from a separate (smaller) sample of the children. Sample sizes are shown beneath each
correlation.
Parents
Adoptees
Parents
Nonadoptees
Child's Years of Education 0.8946 0.7843
228 92
Child's College Status 0.8547 0.8321
228 92
Child's Family Income 0.6554 0.862
210 89
Child's Height (Inches) 0.897 0.9538
228 92
Child's BMI 0.7411 0.7618
220 90
Child Drinks? (0-1) 0.4824 0.3868
229 91
Child Smokes (0-1) 0.5951 0.7694
229 92
Child is Married? (0-1) 0.816 0.7251
229 90
49
Figure I: Mean (College Attendance) By Family Size
dashed line is for nonadoptees (higher line), solid line is for adoptees
Nonadoptees
Adoptees
.4
.5
.6
.7
.8
MeanofChildHas4YearsCollege
1 2 3 4 5
Number of Children in the Family
50
13
14
15
16
17
11 12 13 14 15 16 17 18 19 20
Mother's Years of Education
MeanofChild'sYearsofEducationFigure II: Mean Child's Years of Education Vs Mothers
dashed line is for nonadoptees, solid line is for adoptees
Nonadoptees
Adoptees
51
30
40
50
60
70
80
90
100
110
120
130
140
10 25 40 60 85 125 175 200
Parent's Income
MeanofChild'sFamilyIncome'000
Figure III: Mean of Child's Family Income By Parent's Income At Adoption
dashed line is for nonadoptees (higher line) , adoptees
Nonadoptees
Adoptees
52
Figure IV: Comparison of Adoptive and Nonadoptive Sibling Correlations For
Various Outcomes
This graph displays the results in Table IV.
Has 4 Years of College
Highest Grade Completed
Family Income
Log (Family Income)
Drinks
Smokes
Height
Weight
BMI
Overweight
Graduate US News Ranked School
School Acceptance Rate
Married?
Number of Children
Obese
Asthma
0
.1
.2
.3
.4
.5
BiologicalSiblingCorrelation
0 .1 .2 .3 .4 .5
Adoptive Sibling Correlation
53
Figure V: Comparison of Coefficient of Transmission from Parent to Child
Graph shows coefficient from a regression of child's outcome on mother's outcome for adoptees and
nonadoptees in the sample.
Years of Education
Has 4+ Years College
Log Household Income
Height Inches
Is Obese
Is Overweight
BMI
Smokes (0-1)
Drinks (0-1)
0.1.2.3.4.5
NonadopteesTransmissionCoefficient
0 .1 .2 .3 .4 .5
Adoptees Transmission Coefficient