American Economic Journal: Applied Economics 2013, 5(1): 32–64

http://dx.doi.org/10.1257/app.5.1.32

The Trouble with Boys: Social Influences

and the Gender Gap in Disruptive Behavior†
By Marianne Bertrand and Jessica Pan*

This paper explores the importance of the home and school envi-
ronments in explaining the gender gap in disruptive behavior. We
document large differences in the gender gap across key features of
the home environment—boys do especially poorly in broken families.
In contrast, we find little impact of the early school environment on
noncognitive gaps. Differences in endowments explain a small part
of boys’ noncognitive deficit in single-mother families. More impor-
tantly, noncognitive returns to parental inputs differ markedly by
gender. Broken families are associated with worse parental inputs,
and boys’ noncognitive development, unlike that of girls’, appears
extremely responsive to such inputs. (JEL I21, J12, J13, J16, Z13)

T here has been a recent wave of interest in the role that noncognitive skills might
play in explaining educational achievement, labor market success, and other sig-
nificant life outcomes. Jencks (1979) was among the first to document that factors,
such as study habits, industriousness, perseverance, and self-control matter as much
as cognitive skills in explaining occupational achievement. Since then, other studies
have confirmed the central role played by behavioral and socio-emotional factors in
explaining schooling and labor market outcomes (Heckman and Rubinstein 2001;
Heckman, Pinto, and Savelyev forthcoming; Flossmann, Piatek, and Wichert 2006;
Segal forthcoming; Bertrand and Pan 2011).1
Gender has been identified as an important correlate of noncognitive skills. Boys
are known to perform worse than girls on many noncognitive dimensions. Boys have
well documented attention and behavioral difficulties (Beamen, Wheldall, and Kemp
2006; Gilliam 2005; Ready et al. 2005), have lower levels of inhibitory control and
perceptual sensitivity (Else-Quest et al. 2006), are less able to delay gratification
(Silverman 2003), and are more likely to be diagnosed with attention deficit hyper-
activity disorder (see, for example, Szatmari, Offord, and Boyle 1989). In fact, a few
recent papers have claimed that the (reverse) gender gap in college attendance that

* Bertrand: Booth School of Business, University of Chicago, 5807 South Woodlawn Avenue, Chicago, IL
60637, National Bureau of Economic Research, Center for Economic Policy and Research, and Institute for
the Study of Labor (e-mail: [email protected]); Pan: Department of Economics, National
University of Singapore, 1 Arts Link, Singapore 117570 (e-mail: [email protected]). We thank seminar par-
ticipants at the University of Washington St. Louis and the NBER Summer Institute for many helpful comments.
Daniel Tannenbaum, Paul Ho, and Amanda Chuan provided excellent research assistance.
†
To comment on this article in the online discussion forum, or to view additional materials, visit the article page
at http://dx.doi.org/10.1257/app.5.1.32.
1
Using data from both the National Educational Longitudinal Survey (NELS) and the National Longitudinal
Survey of Youth (NLSY 97), Bertrand and Pan (2011) show a negative relationship between suspension in high
school and college attendance or completion, even after controlling for math and reading test scores.

32
Vol. 5 No. 1 bertrand and pan: the trouble with boys 33

currently exists in the United States and many other developed countries might be
best explained by gender differences in noncognitive skills (see, for example, Becker,
Hubbard, and Murphy 2010; and Goldin, Katz, and Kuziemko 2006). In support of
this claim, Jacob (2002) shows that controlling for the greater incidence of school
disciplinary and behavioral problems among boys explains a substantial share of the
female advantage in college enrollment.2
What is less well understood are the sources of the gender gap in noncognitive
skills. Some researchers have stressed biological influences—many of the differ-
ences that exist between male and female brains have been shown to occur in areas
related to mood, emotions, and emotion regulation. The development of the frontal
cortex (which is associated, among other things, with inhibitory control and hence
a decreased risk of externalizing problems) and temporal lobe has been shown to be
considerably faster among girls than boys. Moreover, prior research has established
that variation in in utero exposure to sex hormones, particularly testosterone, is associ-
ated with such structural and functional brain differences, even within gender groups.
Higher levels of prenatal exposure to testosterone have been linked to slower matura-
tion of parts of the temporal cortex, but also lower empathy levels, higher disinhibition
and lower quality of social relationships (see, for example, Baron-Cohen 2002, 2003;
Geary 1998, 2002; McClure 2000; Knickmeyer et al. 2005; Maccoby 1998).
Nevertheless, it remains unclear whether systematic gender differences in child-
rearing inputs or in the production function of noncognitive skills are also contrib-
uting factors. A case for a possible role of social influences is warranted in light
of earlier work showing that noncognitive skills are not fixed but are in fact quite
malleable, and can be shaped by early intervention programs.3 The primary goal of
this paper is to assess the explanatory power of such social or environmental influ-
ences for the gender gap in noncognitive development. In particular, we focus on
externalizing behaviors, which encompass disruptive and acting-out behaviors, such
as aggression and delinquent behavior.
Existing research lacks a systematic look into whether and how the home and
school environments contribute to gender differences in noncognitive skills. Many of
the earlier studies documenting how environmental factors contributed to disruptive
behavior focused exclusively on boys (as this is the “at risk” group), and relied on very
small sample sizes (Zahn-Waxler, Shirtcliff, and Marceau 2008). In contrast, the core
of our analysis is based on a large representative sample of boys and girls in the United
States that started kindergarten in 1998 and was followed until eighth grade.
We start by documenting the gender gap in noncognitive skills among children,
and how that gap evolves over the course of children’s development. What is imme-
diately remarkable is the size of these gender gaps. For example, by fifth grade, girls

2
See also Duckworth and Seligman (2005) and Kenney-Benson et al. (2006).
3
For example, in their study of the Perry preschool program, which was targeted toward disadvantaged inner-
city youth, Heckman et al. (2009) document sizable increases in future employment and earnings among treated
children, and attribute most of these gains to the program’s positive long-run effect on noncognitive development.
In a similar vein, Chetty et al. (2011) study the long-run effects of the Tennessee STAR class size experiment and
document that increases in kindergarten class quality increased earnings, college attendance, and other long-run
outcomes, even though gains in test scores faded over time. They show that the long-run effects of kindergarten
class quality operate primarily through the noncognitive channel by increasing effort, motivating initiative, and
reducing disruptive behavior.
34 American Economic Journal: applied economicsjanuary 2013

score about half a standard deviation below boys in teacher-reported ­externalizing

problems and 0.45 of a standard deviation above boys in teacher-reported self-
control. For comparison, the widely discussed gender gap in math is about 0.15 of
a standard deviation in fifth grade and the (reverse) gender gap in reading is about
0.2 of a standard deviation in fifth grade. The gender gap in all noncognitive skills
widen in the early years of school.
While the data allows us to track the gender gap in many types of noncognitive
skills, our primary interest is in the set of noncognitive skills that map into future edu-
cational and labor market outcomes. We show that externalizing behavior is a crucial
determinant of school suspension, which itself has been shown to directly matter for
long-term educational outcomes. Therefore, our analysis will focus on explaining gen-
der differences in externalizing behavior, which is based on the frequency with which
a child argues, fights, gets angry, acts impulsively, or disturbs ongoing activities.
We test various socialization theories for the gender gap. We consider both home-
based and school-based influences. We fail to uncover any compelling evidence that
any of the early school-based influences that we consider matter. In particular, while
it has been argued that boys might be at a particular disadvantage in more regimented
early schooling environments (because of the slower maturation of key brain areas),
we find no systematic differences in the gender gap based on key features of the kin-
dergarten environment. Also, we find no systematic differences in the gender gap
based on age of kindergarten entry, or teacher gender, or share of disruptive peers in
the school or classroom.
In contrast, our analysis of the home environment uncovers richer patterns.
Family structure is an important correlate of boys’ behavioral deficit. Boys raised
outside of a traditional family (with two biological parents present) fare especially
poorly. For example, the gender gap in externalizing problems when the children are
in fifth grade is nearly twice as large for children raised by single mothers compared
to children raised in traditional families. By eighth grade, the gender gap in school
suspension is close to 25 percentage points among children raised by single moth-
ers, while only 10 percentage points among children in intact families. Boys raised
by teenage mothers also appear to be much more likely to act out.
In the remainder of our analysis, we ask why boys raised by single mothers are
at a higher risk of developing behavioral problems. First, it is possible that single
mothers invest disproportionately less in their sons, or feel less warm toward them.
Indeed, we show that single mothers seem relatively more emotionally distant from
their sons and are also more likely to have reported spanking their sons. Accounting
for these endowment effects explain a small part of the difference in the gender gap
in externalizing behavior (in fifth grade) and grade suspension (in eighth grade)
between children growing up in intact families and families headed by single moth-
ers. Most important though are gender differences in the noncognitive returns to
parental inputs. On average, across children, broken families are associated with
lower levels and lower quality of parental inputs. The noncognitive development
of boys, unlike that of girls, appears extremely responsive to such inputs.4 Overall,

4
A limited literature in psychology has focused on gender differences in the family determinants of externaliz-
ing problems. The findings of this literature have not been consistent. Some research has found that the ­determinants
Vol. 5 No. 1 bertrand and pan: the trouble with boys 35

our findings strongly suggest that boys’ deficit in noncognitive skills is not purely
biological but, instead, subject to very strong environmental influences, particularly
from the home.

I. The Data

The main data source for our analysis is The Early Childhood Longitudinal
Study: Kindergarten Cohort (ECLS-K). ECLS-K is a longitudinal study of a nation-
ally representative sample of over 20,000 children entering kindergarten in 1998.
Information on these children has been gathered until they complete eighth grade.
The full sample was interviewed in the fall and spring of kindergarten, spring of first
grade, spring of third grade, spring of fifth grade, and spring of eighth grade. About
1,000 different schools are included in the sample, and 20 children were interviewed
per school. Information was collected about children’s cognitive, social, emotional,
and physical development from children, families, teachers, and schools. In addi-
tion, information about the children’s home environment, educational practices at
home, environment at school, curriculum, and teacher qualifications was also col-
lected (National Center for Education Statistics 2003).5
We use teacher-reported measures of five behavioral and social-emotional skills:
externalizing problems, self-control, approaches to learning, interpersonal skills, and
internalizing problems. These noncognitive skill measures are adapted from the Social
Skills Rating Scale, a widely used survey technique for detecting social and behav-
ioral problems in the classroom (Gresham and Elliott 1990). As Neidell and Waldfogel
(2010) note, the ECLS-K noncognitive measures appear to have relatively high “valid-
ity based on test-retest reliability, internal consistency, inter-rater reliability, and cor-
relations with other, more advanced behavioral constructs (Elliott et al. 1988) and
are considered the most comprehensive assessment that can be widely administered
in large surveys such as the ECLS-K (Demaray et al. 1995).” Each of the noncogni-
tive measures averages answers to a series of questions that are rated on a scale from
1 (never) to 4 (very often). The measurement of externalizing problems is based on
five questions assessing the frequency with which a child argues, fights, gets angry,
acts impulsively, and disturbs ongoing activities. The measurement of self control is
based on four questions assessing the child’s ability to control behavior by respect-
ing the property rights of others, controlling his or her temper, accepting peer ideas
for group activities, and responding appropriately to pressure from peers. “Approaches
to learning” measures behaviors that affect the ease with which children can benefit
from the learning environment. It averages six items that rate child’s attentiveness, task
persistence, eagerness to learn, learning independence, flexibility, and organization.
Interpersonal skills rate the child’s skill in forming and maintaining friendships; getting

were very similar for boys and girls (e.g., Dishion et al. 1994; Huizinga, Esbensen, and Weiher 1991; Rowe,
Flannery, and Flannery 1995). A few other papers report differences in family correlates of externalizing behaviors
for boys and girls, even though the papers’ conclusions vary as to which gender is more affected by family quality
and family inputs (e.g., Kavanagh and Hops 1994; Lytton and Romney 1991; Rothbaum and Weisz 1994; Emery
and O’Leary 1982; Hetherington 1989; Simons and Chao 1996; Harold and Conger 1997; McFadyen-Ketchum et
al. 1996; Cole, Teti, and Zahn-Waxler 2003; Griffin et al. 2000).
5
See http://nces.ed.gov/ecls for details.
36 American Economic Journal: applied economicsjanuary 2013

along with people who are d­ ifferent; comforting or helping other children; expressing
feelings, ideas, and ­opinions in positive ways; and showing sensitivity to the feelings
of others. Finally, internalizing problems is based on four questions that ask about the
apparent presence of anxiety, loneliness, low self-esteem, and sadness with the child.
Online Appendix Table A1 provides further details of these measures. We complement
these noncognitive variables with a measure of school suspension—a dummy variable
that equals one if the parent reports that the child has experienced an in or out of school
suspension in eighth grade. The sample is restricted to children who have nonmissing
data on key background characteristics, such as gender, race, age at assessment, region,
urbanicity, sibling composition, family structure, mother’s age at first birth, and family
SES in the fall of kindergarten.6 As there is substantial attrition in the ECLS-K panel,
we further restrict the sample to children who have valid teacher ratings of external-
izing behavior in the fall of kindergarten and grade five as well as the parental reported
measure of school suspension in grade eight.7 In all our specifications, we weight the
observations using the eighth grade panel weights provided in ECLS-K.8
Summary statistics for these noncognitive traits are presented in Table 1. We con-
vert each of the teacher ratings into a standardized normal with mean 0 and standard
deviation 1 in the weighted sample after imposing the sample restrictions. Table 1
reports raw mean girl-boy differences in these noncognitive traits. We present these
differences for the fall of kindergarten (the first time they are measured), and the
spring of kindergarten, first, third, and fifth grade (the last time they are measured).
Girls score better on each of the five noncognitive measures at all ages and these
gaps appear to widen as kids age.9
The average boy “acts out” between 0.44 (kindergarten) and 0.52 (fifth grade)
standard deviations more than the average girl on the externalizing problem index.10
The average boy scores between 0.36 and 0.45 standard deviations below the aver-
age girl in terms of self-control or self-regulation. Similar gaps exist for “approaches
to learning” and “interpersonal skills.” The gender gap is smallest for internalizing
problems, a trait that captures anxiety, loneliness, low self-esteem, and sadness.
Finally, girls are less likely to have repeated a grade and to have been suspended

6
While we have measures of some of these background characteristics at multiple points in time (e.g., family
structure and family SES), throughout the paper, we use the fall-K measures to ensure consistency and to limit
potential endogeneity concerns. Results are generally not sensitive to whether we use the fall-K measures or the
year-specific measures.
7
We have also replicated the results using the sample of children with valid information in each wave separately.
The results are similar to those we obtain from imposing the sample restriction. These results are available from
the authors upon request.
8
Because of the complex sampling procedure utilized by ECLS-K, different weights are suggested depending
on the set of variables used. We use the eighth grade parent panel weight (C1_7FP0). This weight is recommended
to be used for the analysis of parent interview data from six rounds of data collection (fall-kindergarten, spring-
kindergarten, spring-first grade, spring-third grade, spring-fifth grade, and spring-eighth grade), alone or in combi-
nation with (a) child assessment data from any of these six rounds (b) data from any of the six waves of the teacher
questionnaire (teacher-level or child-level) (ECLS-K Combined Eighth Grade and K–8 User’s Manual 2009).
9
Note that the teacher-based gender gaps in at least some of these noncognitive skills might be even larger than
they appear in Table 1 because of gender differences in retention. More boys than girls in a given wave of ECLS-K
are not in their “wave” grade but instead in a lower grade with teachers comparing them to younger, and hence, in
general, more unruly, children.
10
A similar widening of the gender gap in externalizing problems can be found using child-reported ratings
(which are only available for third and fifth grades).
Vol. 5 No. 1 bertrand and pan: the trouble with boys 37

Table 1—Teacher Ratings of Noncognitive Skills,

Suspension, Retention, and Test Scores

Girls Boys Difference (girls-boys)

Externalizing behaviors:
Fall kindergarten −0.222 0.213 −0.435***
(0.841) (1.090) [0.044]
Spring kindergarten −0.232 0.222 −0.454***
(0.874) (1.061) [0.045]
Grade 1 −0.244 0.240 −0.484***
(0.846) (1.079) [0.044]
Grade 3 −0.232 0.231 −0.463***
(0.866) (1.069) [0.049]
Grade 5 −0.263 0.252 −0.515***
(0.826) (1.084) [0.046]
Self control:
Fall kindergarten 0.183 −0.177 0.360***
(0.940) (1.024) [0.044]
Spring kindergarten 0.196 −0.187 0.383***
(0.923) (1.034) [0.045]
Grade 1 0.200 −0.197 0.397***
(0.919) (1.037) [0.045]
Grade 3 0.183 −0.183 0.366***
(0.915) (1.047) [0.048]
Grade 5 0.231 −0.221 0.452***
(0.903) (1.038) [0.045]
Approaches to learning:
Fall kindergarten 0.233 −0.223 0.456***
(0.929) (1.015) [0.043]
Spring kindergarten 0.253 −0.241 0.494***
(0.898) (1.032) [0.043]
Grade 1 0.221 −0.217 0.438***
(0.938) (1.011) [0.045]
Grade 3 0.264 −0.264 0.528***
(0.908) (1.018) [0.045]
Grade 5 0.319 −0.306 0.625***
(0.881) (1.011) [0.042]
Interpersonal skills:
Fall kindergarten 0.194 −0.187 0.381***
(0.961) (1.002) [0.043]
Spring kindergarten 0.202 −0.194 0.396***
(0.958) (1.002) [0.044]
Grade 1 0.218 −0.215 0.433***
(0.945) (1.006) [0.045]
Grade 3 0.232 −0.233 0.465***
(0.933) (1.011) [0.046]
Grade 5 0.284 −0.276 0.560***
(0.894) (1.020) [0.044]

(Continued)
38 American Economic Journal: applied economicsjanuary 2013

Table 1—Teacher Ratings of Noncognitive Skills,

Suspension, Retention, and Test Scores (Continued)

Girls Boys Difference (girls-boys)

Internalizing problems:
Fall kindergarten −0.036 0.035 −0.071*
(0.998) (1.001) [0.042]
Spring kindergarten −0.041 0.039 −0.080*
(0.995) (1.003) [0.045]
Grade 1 −0.048 0.047 −0.096**
(0.947) (1.048) [0.044]
Grade 3 −0.085 0.084 −0.169***
(0.909) (1.076) [0.046]
Grade 5 −0.089 0.086 −0.175***
(0.904) (1.078) [0.044]
Fall-k to grade 8 0.080 0.156 −0.076***
(0.271) (0.363) [0.016]
In/out of school suspension 0.083 0.237 −0.154***
  in grade 8 (0.276) (0.425) [0.017]
Reading scores:
Fall kindergarten 0.095 −0.092 0.188***
(1.019) (0.972) [0.042]
Spring kindergarten 0.107 −0.103 0.209***
(0.985) (1.003) [0.044]
Grade 1 0.137 −0.131 0.268***
(0.943) (1.035) [0.044]
Grade 3 0.118 −0.117 0.235***
(0.943) (1.041) [0.043]
Grade 5 0.104 −0.101 0.205***
(0.927) (1.056) [0.044]
Grade 8 0.112 −0.112 0.224***
(0.943) (1.042) [0.044]
Math scores:
Fall kindergarten 0.006 −0.006 0.012
(0.966) (1.032) [0.041]
Spring kindergarten −0.012 0.012 −0.024
(0.962) (1.035) [0.042]
Grade 1 0.002 −0.002 0.005
(0.953) (1.043) [0.043]
Grade 3 −0.092 0.090 −0.181***
(0.971) (1.020) [0.044]
Grade 5 −0.073 0.070 −0.144***
(0.988) (1.007) [0.044]
Grade 8 −0.026 0.025 −0.051
(0.969) (1.029) [0.045]

Notes: Summary statistics are based on the restricted sample of children with nonmissing
observations for each outcome. Teacher ratings and test scores are standardized to have a mean
of zero and standard deviation one in the weighted sample after imposing the sample restric-
tions. Please refer to the text for sample restrictions. Observations are weighted using eighth
grade parent panel weights (C1_7FP0). Robust standard errors are reported for differences in
the means across genders.
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
  * Significant at the 10 percent level.
Vol. 5 No. 1 bertrand and pan: the trouble with boys 39

from school in eighth grade. Nearly one out of four boys experienced at least one
school suspension in eighth grade, while only one out of ten girls did.
The remaining rows of Table 1 report the well-known gender gaps in both math
and reading skills.11 Girls outperform boys with respect to reading but lag behind in
math. While the (reverse) gender gap in reading is already present in kindergarten,
the gender gap in math only emerges in first grade.12 What is remarkable is how large
the gender gaps in noncognitive skills are compared to the gender gaps in reading and
math scores. The (reverse) gender in reading ranges between 0.2 and 0.27 of a stan-
dard deviation, while the gender gap in math ranges between 0 and 0.18 of a standard
deviation. This is in contrast to the gender gaps in noncognitive skills, which are all
closer to 0.5 standard deviations (with the exception of internalizing problems).
While we do not, as indicated above, directly use parental ratings of noncogni-
tive skills in our analysis below, the top panel of online Appendix Table A3 reports
gender gaps for those ratings as well.13 The questions asked to parents do not per-
fectly match those asked to teachers, but the patterns we observe in online Appendix
Table A3 are qualitatively similar to those in Table 1, even though the magnitude
of the parent-reported gender gaps in noncognitive skills are smaller. Parents rate
daughters as relatively less likely to act impulsively, which maps into the teacher-
reported gender gap in both externalizing and self-control problems.14 Girls are also
more likely to exert self-control and score higher than boys when rated by their
parents on their “approaches to learning” and social skills. Consistent with the small
teacher-reported gender gap in internalizing problems, parents do not perceive any
systematic differences in sadness or loneliness between their sons and daughters.
One might expect that not all of the early childhood noncognitive skills dis-
cussed above feed the same way into future educational and labor market outcomes.
Unfortunately, the ECLS-K data does not go beyond eighth grade, and it is therefore
impossible to directly relate educational achievement or earnings to these various
early childhood noncognitive measures. Instead, we relate these noncognitive mea-
sures to school suspension in eighth grade, which has been shown to be predictive
of college attendance and college completion (Segal forthcoming; Bertrand and Pan
2011). The results are reported in online Appendix Table A4. Controlling for read-
ing and math scores, we find that the strongest and most robust predictor of eighth
grade suspension is externalizing behavior. There appears to be some influence of
fifth grade interpersonal skills, but not kindergarten interpersonal skills. We also see
some influence of self-control when measured in kindergarten, but not when mea-
sured in fifth grade. Finally, there are no statistically significant influences of fifth
grade math and reading scores on the likelihood of eighth grade suspension. A dif-
ferent pattern emerges when we look at the relationship between the different mea-
sures of teaching reported noncognitive skills and reading and math test scores in

11
The reading and math test scores reported are item response theory (IRT) scores provided in ECLS-K, which
we convert into a standardized normal with mean zero and standard deviation one in the weighted sample after
imposing the sample restrictions.
12
See Fryer and Levitt (2010) for an analysis of the gender gap in math scores using the ECLS-K data.
13
Ideally, we would have liked to use the parent reported measures of noncognitive skills in the main analysis
as well. Unfortunately, this data was only collected in kindergarten and first grade.
14
For externalizing behavior/impulsive behavior, the correlations between teacher and parent ratings range
between 0.27 (fall-K) and 0.31 (grade 1).
40 American Economic Journal: applied economicsjanuary 2013

grade 8—a child’s “approach to learning” appears to be the most important p­ redictor
of his or her future cognitive test scores. Internalizing behavior also seems to matter,
although to a much smaller degree.15 Overall, the evidence in online Appendix
Table A4 motivates our focus in most of the analysis on the gender gap in external-
izing problems, or in the tendency to “act out.”

II. What Drives the Gender Gap in Noncognitive Skills?

In an attempt to understand the forces driving the gender gap in noncognitive

skills, we examine how the gap varies across various subsamples of the data. To the
extent that the magnitude of the gap varies along observable dimensions, this may
shed light on the sources of the gender gap. For this analysis, we restrict ourselves to
the following measures of noncognitive skills: externalizing behavior (both in fall-K
and fifth grade) and the likelihood of suspension in eighth grade.16

A. The Home Environment

Summary statistics by gender for the home environment variables and child’s back-
ground characteristics that we exploit in the analysis are presented in online Appendix
Tables A5 and A6. There are only very small differences in family structure, parental
SES and sibling composition across genders, consistent with the expectation that gen-
der is essentially randomly assigned across families.17 In other words, it is not the case
that boys are disproportionately likely to grow up in disadvantaged families, and thus
differences in family background cannot be a direct factor in explaining the overall
gender gap in noncognitive skills. We also see no systematic differences across gender
in region of residence or rate of urban living.18
We do, however, observe some differences with respect to the level of inputs
these families are investing in their children. For most of the analysis that follows,
we measure these parental inputs in kindergarten. Parents are significantly more
likely to read to their girls, and there are generally more books around girls. They are
also more likely to take their girls to a concert and to sign them up for some extra-
curricular activity. Overall, the parental input composite that averages these vari-
ous components of parental investments (see online Appendix Table A2 for details)
shows an advantage for girls. Parents are slightly more likely to have reported spank-
ing their boys as compared to their girls in the past week (in kindergarten), although
this difference is not statistically significant. Finally, there is a small but statistically

15
Duncan et al. (2007) also fail to find an association between early externalizing behavior and later test scores.
16
In analysis not reported here, we also considered the likelihood of school retention (in grade 5 or grade 8) as
an alternative outcome variable. The patterns we uncover throughout this paper on school suspension essentially
extend to school retention.
17
There is some evidence that boys and girls may be raised in different family environments due to a small
degree of son preference in the United States. For example, Dahl and Moretti (2008) find that girls are more likely
to have absent fathers and tend to have more siblings. Similar evidence is also found by Fortin, Oreopoulos, and
Phipps (2011). The gender differences are generally quite small and our failure to find gender differences in the
family environment in our sample is likely due to the relatively small sample size in ECLS-K.
18
We find some small gender differences in family SES. In this sample, boys are slightly more likely to be in
the lowest two SES quintiles as compared to girls. However, the magnitude of these differences are very small and
are only marginally significant at the 10 percent level.
Vol. 5 No. 1 bertrand and pan: the trouble with boys 41

significant gender gap in the emotional distance composite we constructed based

on questions relating to how close parents feel to their kindergarteners (see online
Appendix Table A2). In the kindergarten survey, parents of daughters are less likely
to report being too busy to play with their child relative to parents of sons. They also
feel more loved by their daughters than their sons.19
Directly controlling for the differences in the home environment we observe
between boys and girls has a minimal effect on the gender gaps in externalizing
behavior and school suspension. This is demonstrated in Table 2. The first panel
of Table 2 presents the estimated gender gap (female dummy) in externalizing
behavior in fall of kindergarten (column 1), externalizing behavior in fifth grade
(column 2), and school suspension in eighth grade (column 3), controlling for race
dummies (black, Hispanic, Asian, other), age at assessment at fall-K, age-squared,
birth weight, number of older brothers, number of younger brothers, number of
older sisters, number of younger sisters, and dummies for region and urbanicity.
The second panel (panel B) replicates the same analysis but restricts the sample to
those children for which we have nonmissing information on family background
and other parental inputs. Panel C adds controls for family type and socio-economic
background, all measured at fall-kindergarten: dummies for family structure (single
mother, both biological parents, and other family structures), a dummy for teen
motherhood, and five dummy variables indicating the family’s SES quintiles. As
expected based on the evidence in online Appendix Table A5, adding these controls
pretty much leaves the estimated gender gaps in externalizing behavior and school
suspension unaffected. Panel D further controls for the parental input measures pre-
sented in online Appendix Table A5, again all measured at fall-kindergarten: the
HOME index, the emotional supportiveness index, and parent’s disciplinary style.
Overall, the addition of controls for family quality and parental inputs (e.g., from
panel B to panel D) increase R2 values by between 0.03 and 0.07. The parental input
controls do contribute to reducing the estimated gender gap on both externalizing
behavior in fifth grade and suspension in eighth grade, but this effect is quantitatively
very small. The gender gap in externalizing behavior in fifth grade goes from −0.50
to −0.48, and the gender gap in school suspension goes from −0.15 to −0.14. The
estimated gender gap remains virtually the same when we control for these parental
input measures in a more flexible manner (results not reported here). Hence, we see
little systematic evidence of differences in family background or parental inputs
between boys and girls that would directly contribute to boys’ conduct problems.
Perhaps not surprisingly, to a first degree approximation, boys and girls are raised in
fairly similar families and receive fairly similar parental inputs, at least to the extent
that these variables are adequately captured in the data.
Yet, similar backgrounds for boys and girls may hide important home influences
for the gender gap in noncognitive development. For example, some have raised
the possibility that boys raised by single mothers are at a particular disadvantage
in terms of behavioral development. This could arise, for example, because single

19
Similar patterns emerge when we consider summary measures of home inputs, disciplinary style, and emotional
distance as an average from kindergarten to third grade (see Bertrand and Pan 2011). Baker and Milligan (2011) also
find girl/boy differences in parental inputs in the United States and Canada, especially for inputs promoting literacy.
42 American Economic Journal: applied economicsjanuary 2013

Table 2—Gender Gap Controlling for Various Characteristics

Gender gap (female-male)

Externalizing behavior Externalizing behavior Suspension in
fall-K grade 5 grade 8
Panel A.
Female −0.435*** −0.496*** −0.148***
(0.042) (0.043) (0.016)
5,741 5,741 5,741
0.09 0.13 0.10

Panel B. Excluding observations with missing observations for parental inputs

Female −0.444*** −0.492*** −0.150***
(0.043) (0.044) (0.016)
5,573 5,573 5,573
0.09 0.13 0.10

Panel C. Including additional controls for family quality (at fall-kindergarten)

Female −0.439*** −0.481*** −0.144***
(0.043) (0.042) (0.015)
5,573 5,573 5,573
0.11 0.17 0.15

Panel D. Including additional controls for parental inputs (at kindergarten)

Female −0.425*** −0.475*** −0.141***
(0.041) (0.041) (0.015)
5,573 5,573 5,573
0.12 0.18 0.17

Panel E. Including additional controls for school environment

Female −0.432*** −0.481*** −0.140***
(0.040) (0.040) (0.015)
5,573 5,573 5,573
0.17 0.20 0.18

Notes: Each cell corresponds to a different regression of the outcome (indicated in each col-
umn) on a female dummy and background covariates that include race dummies (black,
Hispanic, Asian, other), age at assessment at fall-K, age-squared, birthweight, number of older
brothers, younger brothers, older sisters, younger sisters, and dummies for region and urbanic-
ity. Each row includes a different set of controls. Panel A reports the raw gender gap control-
ling for the above list of background covariates. Panel B reports the raw gender gap excluding
observations with missing observations for parental inputs. Panel C reports the gender gap after
controlling for measures of family quality (all measured at fall-kindergarten) that include dum-
mies for family structure (single mother, both biological parents, and other family structures),
a dummy indicating whether the mother is younger than 20 at first birth, and five dummy vari-
ables indicating the family’s SES quintiles. Panel D includes additional controls for paren-
tal inputs which include the HOME index, the emotional supportiveness index, and parent’s
disciplinary style (see online Appendix A2 for details on the construction of these indexes).
Panel E further includes controls for school environment (variables listed in online Appendix
Table A7). Observations are weighted using eighth grade parent panel weights (C1_7FP0).
Robust standard errors reported.
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
  * Significant at the 10 percent level.

mothers have a preference for girls and disproportionately invest in them.20 Another
possibility is that even if single mothers invest equally in the development of their

20
For example, Lundberg, Pabilonia, and Ward-Batts (2007) find that single moms spend more time with
daughters relative to sons than married moms. We report similar evidence in Section III.
Vol. 5 No. 1 bertrand and pan: the trouble with boys 43

boys and girls, maternal input could be a poorer substitute for the lack of pater-
nal inputs when it comes to raising a boy. Similar arguments may apply to young
and immature mothers who could be less invested in their sons or less effective at
controlling their behavior. Another possible argument is that boys, because they
are born at a greater risk of developing behavioral and socio-emotional problems
(as suggested by the medical literature discussed in the introduction), benefit more
from the greater average level of inputs that is typically provided in more educated,
richer, or intact families.21 In other words, boys’ noncognitive development may be
more responsive to parental inputs as compared to girls.’ Hence, while we may not
observe any differences in the home environment between boys and girls on aver-
age, this does not necessarily imply that the home environment is not a contributing
factor to the gender gap in noncognitive skills.
To explore this issue, Table 3 focuses on variation in the gender gap in noncogni-
tive skills across family types and other key features of the home environment. Each
row corresponds to a different subset of the data and each cell corresponds to a dif-
ferent regression. Reported in each cell is the estimated female dummy based on a
separate subsample of the data. Background covariates include race dummies, age at
assessment at fall-K, age-squared, birth weight, number of older brothers, number of
younger brothers, number of older sisters, number of younger sisters, and dummies for
region and urbanicity. We report robust standard errors in brackets. Column 1 reports
the estimated gender gap in externalizing behavior in fall-K, column 2 reports the
estimated gender gap in externalizing behavior in fifth grade, and column 3 reports
the estimated gender gap in the likelihood of any suspension during eighth grade. The
various aspects of the home environment are all measured in the fall of kindergarten.
Panel A focuses on family structure. We compare the gender gap in noncogni-
tive skills across three family structures: intact families (two biological parents),
single mothers, and other family structures.22 Looking across family structures in
column 1, the size of the gender gap in externalizing behavior in the fall of kinder-
garten appears relatively similar across intact and nonintact families. This is con-
firmed by a formal test of the equality of the gaps, which yields a p-value of 0.269.
Strikingly, differential patterns by family structure appear to emerge in grade 5 and
grade 8 for externalizing behavior and grade suspension, respectively. In grade 5
(column 2), the gender gap in externalizing behavior more than doubles among
children in single mother families and remains constant among children in intact
families and those in the “other” family structure. The gender gaps in externalizing
behavior in fifth grade and suspension in grade 8 (column 3) is smallest in intact
families. All other family structures appear detrimental to boys. For example, the
gender gap in externalizing behavior in fifth grade goes from −0.41 standard devia-
tions among children in intact families to −0.77 standard deviations among children
raised by a single mother. Also, while the gender gap in eighth grade suspension is
only about −0.10 among children raised by two biological parents, this gap grows to

21
Guryan, Hurst, and Kearney (2008) show that college-educated and noncollege educated parents allocate
different amounts of time to their children, with highly educated parents devoting much more time to their kids
compared to lower educated parents.
22
The “other” family structure includes a range of family types, such as children with a biological father only,
one biological parent and one nonbiological parent, adoptive parents, related guardians, and nonrelated guardians.
44 American Economic Journal: applied economicsjanuary 2013

Table 3—Gender Gap in Noncognitive Skills—The Role

of the Home Environment

Gender gap (female-male)

Externalizing behavior Suspension in
Fall-K Grade 5 grade 8
Panel A. By family structure:
Single mother −0.322*** −0.766*** −0.249***
[0.112] [0.119] [0.047]
760 760 760
Two biological parents −0.444*** −0.405*** −0.103***
[0.048] [0.042] [0.015]
4,502 4,502 4,502
Other family structure −0.584*** −0.554*** −0.271***
[0.119] [0.155] [0.057]
479 479 479

p-value for difference 0.269 0.013 0.000

Panel B. By SES:
1st quartile (lowest) −0.435*** −0.631*** −0.180***
[0.120] [0.114] [0.044]
613 613 613
2nd quartile −0.364*** −0.527*** −0.161***
[0.090] [0.115] [0.040]
946 946 946
3rd quartile −0.544*** −0.532*** −0.194***
[0.099] [0.082] [0.037]
1,109 1,109 1,109
4th quartile −0.411*** −0.422*** −0.119***
[0.073] [0.069] [0.026]
1,394 1,394 1,394
5th quartile (highest) −0.357*** −0.346*** −0.074***
[0.065] [0.067] [0.020]
1,679 1,679 1,679

p-value for difference 0.595 0.160 0.016

Panel C. By mother’s age at first birth

Less than 20 years old −0.495*** −0.776*** −0.259***
[0.102] [0.113] [0.042]
966 966 966
More than 20 years old −0.420*** −0.403*** −0.112***
[0.045] [0.040] [0.015]
4,775 4,775 4,775

p-value for difference 0.504 0.002 0.001

Notes: Each row corresponds to a different subset of the data. Each cell corresponds to a dif-
ferent regression of the outcome (indicated in each column) on a female dummy and back-
ground covariates that include race dummies (black, Hispanic, Asian, other), age at assessment
at ­fall-K, age-squared, birthweight, number of older brothers, younger brothers, older sisters,
younger sisters, and dummies for region and urbanicity. The p-value for difference at the bottom
of each panel tests whether the gender gaps for each subset of the data is statistically different.
Sample is restricted to those with nonmissing observations on family structure, mother’s age at
first birth, family SES, gender, the background covariates, fall-K, and grade 5 teacher ratings of
externalizing behavior and parental reports of school suspension in eighth grade. Observations
are weighted using eighth grade parent panel weights. Robust standard errors reported.
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
  * Significant at the 10 percent level.
Vol. 5 No. 1 bertrand and pan: the trouble with boys 45

−0.25 and −0.27 for boys raised by a single mother or in other family arrangements,
respectively. The p-values reported at the bottom of panel A in columns 2 and 3
strongly reject that the gender gaps are equal across family structures.
Of course, family structure correlates strongly with other family background
characteristics. One obvious candidate is the family’s socio-economic status. For
example, the share of children raised by a single mother is about 40 percent in
the lowest SES quintile, while that share is only 9 percent in the highest SES
quintile. Similarly, only about 44 percent of children in the lowest SES quin-
tile are raised by two biological parents, while more than 80 percent are in the
highest SES quintile. In other words, what might be interpreted as the (negative)
influence of a missing biological father on boys’ noncognitive development might
in fact reflect differential returns to other socio-economic inputs by gender. We
directly examine the potential role of socio-economic background in the remain-
ing subsamples presented in panel B of Table 3. We see some evidence of a nega-
tive ­socio-economic gradient in the gender gap in noncognitive skills in the later
grades. For example, the gender gap in externalizing behavior in fifth grade is
about −0.63, −0.53, −0.53, −0.42, and −0.35 among children in the lowest, sec-
ond, third, fourth, and fifth SES quintiles, respectively. Similar to the comparison
across family structures, these differential patterns by family SES are not evident
in the fall of kindergarten.
Panel C of Table 3 compares the gender gap in noncognitive skills based on moth-
er’s age at first birth. This is another variable that is likely to be strongly correlated to
family structure. Indeed, we find that about 47 percent of children raised by a single
mother are also children of a mother who was less than 20 years old at the time of
a first birth. In contrast, only 19 percent of children in intact families have teenage
mothers. Panel C confirms that mother’s age at first birth is another strong correlate
of the gender gap in noncognitive skills. For example, the gender gap in externalizing
behavior in fifth grade is −0.78 among children raised by mothers who first experi-
enced motherhood in their teens. It is only −0.40 among children raised by mothers
that did not experience motherhood until their twenties. The equivalent figures for
the gender gap in eighth grade school suspension are −0.26 and −0.11, respectively.
Figure 1 provides some graphical evidence on externalizing problems among boys
and girls based on these three family types. We include here all the years in which
externalizing behavior is measured in ECLS-K. Panel A contrasts single mothers
to intact families. Panel B contrasts families in the lower two and higher three SES
quintiles. Finally, panel C contrasts families with mothers that first gave birth in
their teens and those that first gave birth in their twenties or later. While females
have a much lower incidence of externalizing patterns in all grades, in each panel,
the size of the gender gap is similar across family types as kids enter kindergarten.
The only group that appears to “trend” negatively over time is that of boys raised in
“lower quality” families (single mothers, lower SES, teenage mothers). Across the
other three sets of children, the gaps in externalizing problems remain stable over
time. These figures, in combination with the analysis in Table 3, are consistent with
the idea that some features of the home environment are particularly detrimental
to boys’ noncognitive development. Particularly striking is the widening deficit we
observe for boys raised in disadvantaged families over time. In Section III we will
46 American Economic Journal: applied economicsjanuary 2013

Panel A. By family structure Panel B. By SES

0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
−0.2 −0.2
−0.4 −0.4
K1 K2 1 3 5 K1 K2 1 3 5
Grade Grade
Single mom, female Single mom, male Low SES, female Low SES, male
2 bio parents, female 2 bio parents, male High SES, female High SES, male

Panel C. By age first birth

0.8
0.6
0.4
0.2
0
< 20 years, female < 20 years, male
−0.2
>= 20 years, female >=20 years, male
−0.4
K1 K2 1 3 5
Grade

Figure 1. Teacher Reported Externalizing Behavior from Fall-K to Grade 5

by Gender and Family Structure, SES, and Age at First Birth

Notes: The basic sample restrictions are described in the text. In addition, for each grade, the sample is further
restricted to children with nonmissing teacher reported externalizing behavior ratings in that year. The teacher-
reported externalizing behavior index is standardized with mean 0 and standard deviation 1 in the full, weighted
sample in each grade. Panel A plots the standardized teacher ratings in each grade separately by gender and fam-
ily structure (dashed lines indicate single mom families and solid lines indicate intact families where both bio-
logical parents are present). Panel B plots the teacher ratings separately by gender and family SES (dashed lines
indicate families in the lowest two quintiles of SES while the solid lines indicate families in the highest three quin-
tiles). Panel C plots the teacher ratings by gender and mother’s age at first birth (dashed lines indicate children with
mothers who were aged less than 20 years at first birth, while the solid lines indicate children whose mothers were
20 years or older at first birth). Observations are weighted by eighth grade parent panel weights (C1_7FP0).

explore what drives these relationships. In particular, we will focus on why boys
raised by single mothers seem to be at such a disadvantage.

B. The School Environment

Both the home and school environments affect children’s early socialization.
Therefore, we now turn our attention to possible influences of the early school envi-
ronment on the incidence of behavior problems among boys. In particular, some
psychologists have discussed the possibility that boys are at risk of developing behav-
ioral problems because schools expect too much from them from a very young age,
when their brain maturity is not quite on par with that of girls. Parents might be some-
what aware of this, or at least believe in the relevance of this argument in that, as
others have pointed out before, they are more likely to hold their sons back when it
comes to starting school (a pattern we confirm below). In a popular but controversial
book, Sax (2007) argues that many boys develop negative attitudes toward school
and exhibit conduct problems because of challenging experiences in kindergarten.
Vol. 5 No. 1 bertrand and pan: the trouble with boys 47

He further argues that recent changes in early education are placing more demands
on children, with the unintended consequence of disadvantaging many boys. Others
have pointed out that the early school environment might be particularly problem-
atic for boys because of the overwhelming share of female teachers, who may either
discriminate against boys or be less able to adapt their teaching or disciplining styles
to the specific needs of boys (Dee 2006). Finally, it is possible that boys’ conduct is
particularly responsive to negative peer effects. Boys might be particularly at risk of
developing behavioral problems if placed in a classroom with more disruptive kids.
We assess the relevance of these various theories by comparing the gender gap
in externalizing behavior across these features of the kindergarten and early school
environments. In online Appendix Table A7, we compare average characteristics of
the early school environment by gender. There is little difference in the early school
environment by gender. Parents are slightly more likely to hold their sons back, but
the difference is not very large. The average girl is about 0.25 months younger than
the average boy when she starts kindergarten. By and large, it appears boys and
girls start their education at schools that are very similar in terms of their academic
demands, teaching style, and teacher gender. We also see no differences in average
peer characteristics, in particular the share of peers with externalizing problems.
Not surprisingly, simply controlling for these school-related variables does little
to the estimated gender gap in noncognitive skills. This is shown in the last panel
of Table 2, where we present the estimated gender gap in noncognitive skills after
controlling for both the home environment and the variables reported in online
Appendix Table A7. The estimated gender gaps in externalizing problems in kinder-
garten, externalizing problems in fifth grade, and the likelihood of school suspen-
sion in eighth grade are essentially unchanged after controlling for these features of
the early school environment.23
Table 4 assesses whether there are any systematic differences in the gender gap in
noncognitive skills based on these features of the early school environment. Again,
while these elements of the early childhood environment are essentially balanced
across gender, it is still possible that boys are particularly at risk of developing con-
duct problems in some early school settings.
We first compare the gender gap in externalizing behavior and grade suspension
across two groups of children based on whether their actual kindergarten entry age
was above or below the median age of kindergarten entry among all the children in
the sample.24 The gender gaps in externalizing behavior and grade suspension are

23
Due to the large number of missing observations for some of the school environment variables, in the bottom
panel of Table 2, we use all the observations (with nonmissing values for parental input measures) and include
indicator variables for children with missing values on each of the school environment variables.
24
The child’s age of entry in kindergarten is a potentially endogenous variable as parents may be more likely to
hold their boys back if they judge them to be less ready for school, which is likely to be related to their behavioral
maturity. In results available upon request, we also consider a breakdown of the gender gap in noncognitive skills
by predicted age of kindergarten entry. The child’s predicted age of kindergarten entry is computed based on the
child’s month of birth and the state cut-off age for kindergarten entry (Elder and Lubotsky 2009). Children in states
where the cut-off is set by local school districts are excluded from the calculation. This measure tells us the age at
which a child should enter kindergarten if he/she fully complied with the state cut-off based on his/her month of
birth. We have also estimated specifications where we compare children based on whether their actual or predicted
kindergarten entrance age is above or below the median kindergarten entrance age in their state of residence. The
results are very similar and are available upon request.
48 American Economic Journal: applied economicsjanuary 2013

Table 4—The Early School Environment

Gender gap (female-male)

Externalizing behavior Suspension in
Fall-K Grade 5 grade 8
Panel A. Average age at kindergarten entry
Above median −0.491*** −0.503*** −0.133***
[0.064] [0.067] [0.024]
2,465 2,465 2,465
Below median −0.390*** −0.488*** −0.155***
[0.055] [0.054] [0.020]
3,275 3,275 3,275

p-value of difference 0.233 0.859 0.472

Panel B. Kindergarten type

Private −0.326*** −0.279*** −0.088***
[0.080] [0.079] [0.022]
1,355 1,355 1,355
Public −0.457*** −0.532*** −0.156***
[0.047] [0.047] [0.018]
4,386 4,386 4,386

p-value of difference 0.157 0.006 0.017

Panel C. Emphasis on reading and math

Above median −0.478*** −0.477*** −0.136***
[0.074] [0.065] [0.026]
2,263 2,263 2,263
Below median −0.401*** −0.508*** −0.152***
[0.050] [0.053]] [0.019]
3,046 3,046 3,046

p-value of difference 0.387 0.719 0.627

Panel D. Emphasis on homework

Above median −0.259*** −0.458*** −0.135***
[0.071] [0.067] [0.025]
1,965 1,965 1,965
Below median −0.553*** −0.515*** −0.156***
[0.053] [0.056] [0.020]
3,596 3,596 3,596

p-value of difference 0.001 0.515 0.519

Panel E. Emphasis on achievement/behavior/cooperation/following directions

Above median −0.440*** −0.536*** −0.156***
[0.063] [0.059] [0.024]
2,545 2,545 2,545
Below median −0.412*** −0.458*** −0.133***
[0.058] [0.065] [0.022]
2,928 2,928 2,928

p-value of difference 0.739 0.374 0.476

Panel F. Time spent on physical education

Above median −0.460*** −0.489*** −0.177***
[0.070] [0.067] [0.025]
2,185 2,185 2,185
(Continued)
Vol. 5 No. 1 bertrand and pan: the trouble with boys 49

Table 4—The Early School Environment (Continued)

Gender gap (female-male)

Externalizing behavior Suspension in
Fall-K Grade 5 grade 8
Below median −0.402*** −0.478*** −0.128***
[0.053] [0.052] [0.020]
3,158 3,158 3,158

p-value of difference 0.506 0.901 0.127

Panel G. Time spent on recess

Above median −0.384*** −0.314*** −0.113***
[0.094] [0.080] [0.030]
1,075 1,075 1,075
Below median −0.433*** −0.498*** −0.134***
[0.063] [0.059] [0.022]
2,645 2,645 2,645

p-value of difference 0.660 0.064 0.561

Panel H. School has formal retention policy

Yes −0.422*** −0.518*** −0.136***
[0.066] [0.070] [0.025]
2,355 2,355 2,355
No −0.426*** −0.413*** −0.151***
[0.055] [0.056] [0.021]
2,674 2,674 2,674

p-value of difference 0.963 0.244 0.642

Panel I. Overall kindergarten environment index (emphasis on reading, homework, retention

policy)
Above median −0.378*** −0.422*** −0.104***
[0.078] [0.071] [0.029]
1,675 1,675 1,675
Below median −0.465*** −0.510*** −0.165***
[0.053] [0.062] [0.021]
2,927 2,927 2,927

p-value of difference 0.358 0.348 0.084

Panel J. Kindergarten peers

Average classroom peer externalizing score
Above median −0.477*** −0.496*** −0.141***
[0.062] [0.063] [0.023]
2,811 2,811 2,811
Below median −0.432*** −0.505*** −0.158***
[0.056] [0.054] [0.022]
2,875 2,875 2,875

p-value of difference 0.587 0.909 0.600

Panel K. Teacher gender

Female teacher in kindergarten −0.434*** −0.502*** −0.143***
[0.044] [0.043] [0.016]
5,494 5,494 5,494
(Continued)
50 American Economic Journal: applied economicsjanuary 2013

Table 4—The Early School Environment (Continued)

Gender gap (female-male)

Externalizing behavior Suspension in
Fall-K Grade 5 grade 8
Male teacher in kindergarten −0.226 −0.483* −0.048
[0.175] [0.280] [0.093]
81 81 81

p-value of difference 0.195 0.937 0.255

All female teachers from K to grade 5 −0.404*** −0.382*** −0.110***

[0.059] [0.053] [0.021]
3,514 3,514 3,514
At least one male teacher from K to grade 5 −0.482*** −0.598*** −0.160***
[0.088] [0.078] [0.031]
947 947 947

p-value of difference 0.457 0.021 0.185

Notes: Each row corresponds to a different subset of the data. Each cell corresponds to a differ-
ent regression of the outcome (indicated in each column) on a female dummy and background
covariates that include race dummies (black, Hispanic, Asian, other), age at assessment at fall-
K, age-squared, birthweight, number of older brothers, younger brothers, older sisters, younger
sisters, and dummies for region and urbanicity. The p-value for difference at the bottom of each
panel tests whether the gender gaps for each subset of the data is statistically different. Sample
is restricted to those with nonmissing observations on family structure, mother’s age at first-
birth, family SES, gender, the background covariates, fall-K, and grade 5 teacher ratings of
externalizing behavior and parental reports of school suspension in eighth grade. Observations
are weighted using eighth grade parent panel weights. Robust standard errors reported.
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
  * Significant at the 10 percent level.

broadly similar across the two groups of children regardless of whether we use the
child’s actual or predicted age. This suggests that kindergarten entry age has a lim-
ited influence on the gender gap in noncognitive skills.
Panel B compares the gender gap in noncognitive skills between public and pri-
vate kindergartens. Boys’ noncognitive deficit appears to be somewhat larger in a
public school environment. Of course, this raises the question of the extent to which
this has to do with the school or the home environment, as we expect children from
more disadvantaged families to be more likely to attend public schools.
The next rows of Table 4 focus on specific features of the kindergarten envi-
ronment. We are particularly interested in comparing how boys are faring behav-
iorally in more and less regulated kindergarten environments. Again, one of the
arguments that has been brought forward by “pop psychologists” and relayed in
the media is that boys might be unable to cope in kindergartens that force them
to pay attention for long periods of time and demand more regulated learning, as
their brain might not be mature enough to deal with such structured learning at an
early age. To proxy for this, we separate kindergartens based on whether they put
above or below median emphasis on reading and math in the full sample, put above
or below median emphasis on homework, put above or below median emphasis
on achievement and behavior, how much time they spent on physical activity and,
how much time they spent on recess. We see strikingly little ­difference in the
Vol. 5 No. 1 bertrand and pan: the trouble with boys 51

gender gap in noncognitive skills across these dimensions of how regulated the
kindergarten environment is. Similarly, when we separate kindergartens based on
whether or not they have a formal retention policy (with the view that a formal
retention policy proxies for a more regulated environment), we see little system-
atic differences in the gender gap. Not surprisingly, a breakdown of the data based
on a summary index of how regulated the kindergarten is (which loads positively
on emphasis on math and reading, emphasis on homework, and having a formal
retention policy) also fails to show large differences.25 In contrast to the popular
claims discussed above, boys appear to be doing relatively better behaviorally and
face a smaller risk of school suspension in higher grades when placed in a more
regulated kindergarten environment.26
The next characteristic we consider in Table 4 is peer quality. In particular, we
compute the gender gap in externalizing behavior and school suspension for chil-
dren whose kindergarten classroom scored above or below the median in terms of
the average level of externalizing problems.27 Again, we fail to find any evidence
that boys are particularly at risk of behavioral problems if surrounded by more dis-
ruptive peers in early school settings.
Finally, in the remaining rows of Table 4, we consider another feature of the
kindergarten and early school environment, teacher gender. Indeed, as we discussed
above, some have also raised the possibility that kindergarten teachers, a large share
of them being women, are simply less able or willing to attend to the behavioral and
emotional needs of boys. Such a possibility might be a further concern as our mea-
sures of noncognitive development are based on teachers’ evaluation. So, two things
are possible: female teachers do a poor job at emotionally guiding boys (e.g., real
effects on boy’s behavior), or female teachers are unduly harsh in judging boys
(e.g., the gender gap in noncognitive skills we observe is in part the result of biased
teacher assessments).
We first compare the gender gap in noncognitive skills based on the gender of
the kindergarten teacher. Note that the sample of children with a male kindergar-
ten teacher is very small and therefore standard errors are quite large. The gender
gap in externalizing behavior in fall-K and school suspension in grade 8 appears
smaller among children with a male kindergarten teacher relative to those with a
female teacher, although these differences are not significant at conventional lev-
els. Nevertheless, we do not find systematic differences in externalizing behavior in
grade 5. If we separate children based on whether all their teachers from kindergar-
ten to grade 5 are female or whether they had at least one male teacher over those
grades, we find no systematic difference in the gender gaps. In these tabulations, if
anything, the gender gaps appear to be larger among those with at least one male
teacher from kindergarten to grade 5 relative to those with all female teachers.

25
We find some evidence that kindergartens that spend more time on recess are associated with smaller gender
gaps in externalizing behavior in grade 5, however, this is only marginally significant at the 10 percent level, and we
do not see similar patterns for suspension in grade 8.
26
Bertrand and Pan (2011) also considered the possibility that a more regulated kindergarten environment
might be only deleterious to boys that enter such an environment at a very young age, but found no evidence for this.
27
We exclude the focal child when computing the average externalizing behavior of a child’s peers in the classroom.
52 American Economic Journal: applied economicsjanuary 2013

Overall, the evidence in Table 4 is in sharp contrast to the evidence in Table 3.

While we find large differences in the gender gap in noncognitive skills across key
features of the home environment, we do not observe systematic patterns across
features of the kindergarten environment. These findings motivate the deeper inves-
tigation into the parent-children dynamics that we perform in the remaining sections
of the paper.

III. Why Are Boys Raised by Single Mothers Particularly at Risk?

One of most striking patterns we have documented so far relates to the espe-
cially large gender gap in noncognitive skills in nonintact families that emerge by
middle school. Why is this happening? One possibility is that boys raised with-
out a biological father receive especially low levels of parental inputs, parental
warmth and emotional supportiveness, or parental expectations, compared to girls
raised in similar families. While we already established that there are only small
differences in the home environment between boys and girls, it is still possible that
this aggregate analysis hides more systematic differences when we hold family
structure constant.
In online Appendix Table A8, we therefore reproduce the summary statistics
reported in online Appendix Table A5 separately by family structure (single mothers,
intact families, other). Again, we prefer to measure these parental inputs in kinder-
garten to limit reverse causality concerns. For completeness though, we also report
averages over the kindergarten to third grade interviews. When it comes to the home
environment composite, which loads on parental inputs, such as reading to children
or engaging children in extra-curricular activities, we find no evidence that single
mothers invest disproportionately more in their girls, while daughters tend to receive
slightly more inputs than sons; but this is true both in intact families and in families
headed by a single mother.
However, we do observe somewhat larger gender gaps in emotional supportive-
ness and disciplining style in families headed by single mothers. Specifically, while
parents in both intact and broken families report more emotional distance with their
sons as compared to their daughters, the gap is larger in broken families. While there
is no evidence of a gender difference in the likelihood that parents in intact families
spanked their child, boys in broken families are about 13 percentage points more
likely to have been spanked in the last week as compared to girls. We see somewhat
similar patterns when we measure average parental inputs from kindergarten and
third grade. The most robust difference across family structures appears to be with
respect to emotional distance. Single mothers appear especially distant from their
sons.28 One concern with these results is that the incidence of spanking and the
lack of emotional support could be a consequence of the child’s misbehavior. Note,
however, that both these measures were asked in fall-K, and we did not find any
evidence of gender differences in externalizing behavior across family structures in

28
Of course, one might be particularly concerned when we use input over this longer time period that the child’s
misbehavior is the cause rather than the consequence of this relative lack of emotional support. Therefore, for most
of the analysis, we use the parental input measures in fall-K to minimize endogeneity concerns.
Vol. 5 No. 1 bertrand and pan: the trouble with boys 53

the fall of kindergarten. The differences only emerge in the later grades. Therefore,
it is unlikely that the gender differences in parental inputs across family structures
are simply a response to the child’s behavior in fall-K. We will address this concern
more formally by including the child’s prior externalizing behavior in fall-K as a
control in some of our later specifications.
Complementary evidence of a larger gender gap in investment in families headed
by a single mother can be found in other surveys. Using data from the 2003–2005
American Time Use Survey (ATUS) and the Child Development Supplement of
the Panel Study of Income Dynamics (PSID-CDS), Lundberg, Pabilonia, and
Ward-Batts (2007) find that single mothers spend more time with daughters rela-
tive to sons compared to married mothers. We extend the results of Lundberg,
Pabilonia, and Ward-Batts (2007) to the longer time period (2003–2010) that is
now available in the ATUS (see online Appendix Table A9). We find that mothers
in stable family structures do not appear to spend significantly less time with their
sons than they do with their daughters. In contrast, single mothers spend between
1.2 and 1.4 hours less per week with their sons than their daughters in the sample
of kids under 5. In the sample of children under 3 (where endogeneity concerns
about parent time allocation are more limited), single mothers spend between 2.1
and 2.3 fewer hours of total childcare with their sons. In summary, the evidence
in online Appendix Tables A8 and A9 suggests that boys raised by single mothers
might be particularly disadvantaged as they receive lower levels of parental inputs
compared to all other groups of children in the sample.
But how much of the disparity in the gender gap in noncognitive skills across
family structures can these differences in the gender gap in parental inputs explain?
Tables 5A and 5B address this question. Specifically, columns 1 and 4 report, by
family structure, the estimated female dummy in a regression of externalizing
behavior in fifth grade (Table 5A) or suspension in grade 8 (Table 5B), where we
only control for background variables (race dummies, age at assessment at fall-
K, age-squared, birth weight, number of older brothers, number of younger broth-
ers, number of older sisters, number of younger sisters, and number of dummies
for region and urbanicity). These gaps are very similar to the raw gaps reported
in Table 3.29 Among children raised by single mothers, boys score about 0.77 of
a standard deviation below girls in terms of externalizing behavior in fifth grade.
They are about 0.25 percentage points more likely than girls to be suspended in
eighth grade (see column 1 in Table 5A and 5B, respectively). The equivalent fig-
ures among children raised in intact families are 0.40 and 0.10, respectively (see
column 4). Columns 2 and 5 show how this estimated gender gap by family struc-
ture is affected by controlling for the parental inputs listed in online Appendix
Table A8. Among children raised by single mothers, the gap in externalizing behav-
ior goes down to 0.71 ­(column 2, Table 5A), while the gap in school suspension
goes down to 0.23 (column 2, Table 5B). Not surprisingly, given our findings in
online Appendix Table A8, controlling for parental inputs has a limited impact on
the estimated gender gaps in intact families (column 5). The gap in externalizing

29
The slight difference in the coefficient and number of observations is attributable to the fact that in Tables 5A
and 5B, the sample is restricted to children with no missing values for the parental input measures.
54 American Economic Journal: applied economicsjanuary 2013

Table 5A—Gender Gap in Externalizing Behavior in Broken Families:

Role of Family Quality and Parental Inputs

Outcome: Externalizing behavior in grade 5

Single mom Both biological parents
(1) (2) (3) (4) (5) (6)
Female −0.769*** −0.709*** −0.578*** −0.402*** −0.394*** −0.255***
[0.123] [0.107] [0.096] [0.043] [0.042] [0.041]
HOME index (below median) −0.170 −0.222 0.023 0.009
[0.149] [0.138] [0.045] [0.042]
Warmth index (below median) 0.150 0.124 0.117*** 0.067
[0.115] [0.103] [0.043] [0.041]
Spanked child last week 0.260* 0.195 0.146*** 0.078
[0.137] [0.123] [0.051] [0.049]
Age first birth < 20 0.468*** 0.471*** 0.210*** 0.196***
[0.139] [0.118] [0.077] [0.072]
1st SES quintile 0.043 −0.008 0.141 0.070
[0.182] [0.169] [0.100] [0.102]
2nd SES quintile −0.024 −0.017 0.165** 0.175**
[0.170] [0.149] [0.075] [0.073]
3rd SES quintile −0.059 −0.189 0.120* 0.106*
[0.156] [0.141] [0.062] [0.059]
4th SES quintile −0.011 −0.004 0.072 0.055
[0.196] [0.184] [0.055] [0.054]
Externalizing behavior 0.392*** 0.318***
in Fall-K [0.053] [0.025]

Background controls Yes Yes Yes Yes Yes Yes

Observations 723 723 723 4,389 4,389 4,389
R2 0.24 0.29 0.40 0.08 0.10 0.21

Notes: Each column is a separate regression of the female dummy on externalizing behavior in grade 5 (Table 5A)
and suspension in grade 8 (Table 5B) for children in two biological parent families and single mom families, con-
trolling for various subsets of variables. Columns 1 and 4 control for background characteristics that include race
dummies (black, Hispanic, Asian, other), age at assessment at fall-K, age-squared, birthweight, number of older
brothers, younger brothers, older sisters, younger sisters, and dummies for region and urbanicity. Columns 2 and 5
include controls for parental inputs (home index, warmth index and discipline harshness) and family quality (moth-
er’s age at first birth and family SES). Columns 3 and 6 control for externalizing behavior in the fall of kindergarten.
Sample is restricted to those with nonmissing observations on family structure, mother’s age at first birth, family
SES, gender, the background covariates, fall-K, and grade 5 teacher ratings of externalizing behavior and parental
reports of school suspension in eighth grade. Observations are weighted using eighth grade parent panel weights.
Robust standard errors reported.
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
  * Significant at the 10 percent level.

b­ehavior in fifth grade goes down to 0.39, and the gap in school suspension remains
at approximately 0.10. Hence, what started as a (0.77 − 0.40)/0.40 = 93 percent
larger gap in externalizing behavior among children raised by single mothers has
been reduced to a (0.71 − 0.39)/0.39 = 82 percent larger gap. Similarly, the origi-
nal (0.25 − 0.10)/0.10 = 150 percent larger gap in suspension has been reduced
to a (0.23 − 0.10)/0.10 = 130 percent larger gap. In other words, our results are
consistent with the view that a small but nontrivial share of boys’ higher rate of
behavioral problems in single mother families might be due to differences in inputs
by child gender. Nevertheless, these findings remain speculative due to the relatively
large standard errors of the estimates.
Vol. 5 No. 1 bertrand and pan: the trouble with boys 55

Table 5B—Gender Gap in Suspension in Broken Families:

Role of Family Quality and Parental Inputs

Outcome: Suspension in grade 8

Single mom Both biological parents
(1) (2) (3) (4) (5) (6)
Female −0.253*** −0.228*** −0.205*** −0.104*** −0.098*** −0.079***
[0.049] [0.044] [0.045] [0.015] [0.015] [0.015]
HOME index (below median) 0.073 0.064 0.017 0.015
[0.049] [0.047] [0.015] [0.015]
Warmth index (below median) 0.058 0.054 0.053*** 0.046***
[0.047] [0.046] [0.015] [0.015]
Spanked child last week 0.090* 0.079 0.032* 0.023
[0.049] [0.049] [0.019] [0.019]
Age first birth < 20 0.209*** 0.209*** 0.092*** 0.090***
[0.053] [0.052] [0.030] [0.029]
1st SES quintile −0.069 −0.078 0.102*** 0.092**
[0.086] [0.084] [0.039] [0.037]
2nd SES quintile −0.062 −0.060 0.085*** 0.086***
[0.085] [0.079] [0.024] [0.024]
3rd SES quintile −0.002 −0.025 0.081*** 0.079***
[0.079] [0.076] [0.022] [0.022]
4th SES quintile −0.142* −0.141* 0.019 0.017
[0.081] [0.077] [0.016] [0.016]
Externalizing behavior 0.069*** 0.044***
in Fall-K [0.024] [0.012]

Background controls Yes Yes Yes Yes Yes Yes

Observations 723 723 723 4,389 4,389 4,389
R2 0.20 0.27 0.30 0.05 0.09 0.11

Notes: Each column is a separate regression of the female dummy on externalizing behavior in grade 5 (Table 5A)
and suspension in grade 8 (Table 5B) for children in two biological parent families and single mom families, con-
trolling for various subsets of variables. Columns 1 and 4 control for background characteristics that include race
dummies (black, Hispanic, Asian, other), age at assessment at fall-K, age-squared, birthweight, number of older
brothers, younger brothers, older sisters, younger sisters, and dummies for region and urbanicity. Columns 2 and 5
include controls for parental inputs (home index, warmth index and discipline harshness) and family quality (moth-
er’s age at first birth and family SES). Columns 3 and 6 control for externalizing behavior in the fall of kindergarten.
Sample is restricted to those with nonmissing observations on family structure, mother’s age at first birth, family
SES, gender, the background covariates, fall-K, and grade 5 teacher ratings of externalizing behavior and parental
reports of school suspension in eighth grade. Observations are weighted using eighth grade parent panel weights.
Robust standard errors reported.
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
  * Significant at the 10 percent level.

Next, we explore the possibility that the remaining (large) unexplained difference
in the gender gap by family structure may be due to gender differences in the returns
to parental inputs and parental quality. We begin by exploring the ­relationship
between externalizing behavior (Table 6A, columns 1–4) and suspension (Table 6A,
columns 5–8), and parental inputs across both single mom and intact families, sepa-
rately by gender. Columns 1 (girls), 2 (boys), 5 (girls), and 6 (boys) of Table 6A
include three measures of parental inputs—the HOME index, parental warmth
index, and whether the child was spanked last week. The most striking finding is that
there is a stronger relationship between parental inputs (measured in k­ indergarten)
56 American Economic Journal: applied economicsjanuary 2013

Table 6A—Gender Gap in Noncognitive Skills:

Role of Returns to Family Quality and Parental Inputs

Outcome: Externalizing behavior in grade 5 Outcome: Suspension in grade 8

Overall (single mom + both biological parents)
Girls Boys Girls Boys Girls Boys Girls Boys
(1) (2) (3) (4) (5) (6) (7) (8)
HOME index 0.009 0.081 −0.022 −0.020 0.039*** 0.073** 0.022 0.030
(below median) [0.050] [0.086] [0.052] [0.083] [0.015] [0.028] [0.015] [0.027]
Warmth index 0.034 0.215*** 0.035 0.217*** 0.026 0.080*** 0.027 0.081***
(below median) [0.049] [0.074] [0.049] [0.068] [0.017] [0.027] [0.017] [0.025]
Spanked child 0.127** 0.268*** 0.119* 0.245*** 0.016 0.098*** 0.011 0.088***
last week [0.062] [0.093] [0.062] [0.083] [0.021] [0.034] [0.021] [0.031]
Age first birth < 20 0.046 0.554*** 0.052* 0.209***
[0.079] [0.114] [0.029] [0.042]
1st SES quintile 0.148 0.238* 0.071* 0.100**
[0.101] [0.137] [0.039] [0.050]
2nd SES quintile 0.115 0.225** 0.048 0.103***
[0.087] [0.113] [0.030] [0.038]
3rd SES quintile 0.068 0.154 0.046* 0.098***
[0.064] [0.098] [0.024] [0.037]
4th SES quintile 0.053 0.102 −0.010 0.023
[0.064] [0.083] [0.017] [0.030]

Background controls Yes Yes Yes Yes Yes Yes Yes Yes
Observations 2,540 2,572 2,540 2,572 2,540 2,572 2,540 2,572
R2 0.077 0.103 0.080 0.154 0.061 0.123 0.079 0.177

Notes: Each column is a different regression with teacher reported externalizing behavior in grade 5 (columns 1 to 4)
or suspension in grade 8 (columns 5 to 8) as the outcome separately by child gender. The sample is restricted to
single mom and two biological parent families. All regressions control for background covariates that include race
dummies (black, Hispanic, Asian, other), age at assessment at fall-K, age-squared, birthweight, number of older
brothers, younger brothers, older sisters, younger sisters, and dummies for region and urbanicity. Columns 1, 2,
5, and 6 includes three variables for parental inputs measured in the fall of kindergarten—the HOME and Warmth
Indexes are dummy variables that indicate 1 if the child has a HOME or Warmth score that is below that of the median
child in the sample. Spanked child last week is a dummy variable indicating that the parent reported spanking the
child at least once in the past week. Column 3, 4, 7, and 8 includes two additional variables that proxy for family
quality. The first is a dummy variable that indicates that the child’s biological mother was less than 20 at first birth,
the second variable is a measure of family SES which comprises five dummy variables that indicate the family SES
quintile (the fifth quintile (highest SES), is the reference category. Observations are weighted using eighth grade
parent panel weights. Robust standard errors reported.
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
  * Significant at the 10 percent level.

and further noncognitive development for boys than there is for girls. For exam-
ple, for boys, above median levels of the HOME index and parental warmth index
decrease the likelihood of externalizing behavior in fifth grade by 0.08 and 0.2 stan-
dard deviations, respectively. In contrast, we see virtually no relationship between
these input variables and girls’ externalizing problems in fifth grade. The reliance on
harsher disciplining in kindergarten feeds into more conduct problems and a higher
likelihood of school suspension in eighth grade for boys. Again, in contrast, there
is a much weaker relationship between spanking and noncognitive problems for
girls. These findings are particularly relevant in that, as we saw in online Appendix
Table A8, intact families score, on average (across children), higher than broken
families on both the HOME index and the warmth index.
Vol. 5 No. 1 bertrand and pan: the trouble with boys 57

Table 6B—Gender Gap in Noncognitive Skills:

Returns to Family Quality and Parental Inputs by Family Structure

Outcome: Externalizing behavior in grade 5

Single mom Both biological parents
Girls Boys Girls Boys Girls Boys Girls Boys
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A.
HOME index −0.215 −0.013 −0.181 −0.176 0.049 0.094 0.017 0.030
(below median) [0.139] [0.281] [0.144] [0.255] [0.051] [0.072] [0.052] [0.072]
Warmth index 0.118 0.135 0.124 0.255 0.015 0.202*** 0.025 0.200***
(below median) [0.123] [0.219] [0.123] [0.183] [0.051] [0.066] [0.051] [0.065]
Spanked child last week 0.131 0.311 0.104 0.227 0.094 0.217*** 0.082 0.207***
[0.144] [0.233] [0.139] [0.191] [0.061] [0.078] [0.062] [0.074]
Age first birth < 20 0.154 0.771*** 0.007 0.362***
[0.132] [0.235] [0.087] [0.117]
1st SES quintile −0.001 0.168 0.112 0.167
[0.200] [0.269] [0.119] [0.135]
2nd SES quintile −0.229 0.293 0.205** 0.163
[0.195] [0.266] [0.098] [0.108]
3rd SES quintile −0.027 −0.042 0.079 0.178*
[0.186] [0.238] [0.065] [0.100]
4th SES quintile 0.170 −0.106 0.005 0.134
[0.244] [0.270] [0.063] [0.086]

Observations 385 338 385 338 2,155 2,234 2,155 2,234

R2 0.14 0.19 0.17 0.28 0.08 0.05 0.09 0.07

(Continued)

Columns 3, 4, 7, and 8 further control for family SES and teenage motherhood.
We view these two variables as additional proxies for the quantity and quality of
parental inputs. Family SES and teenage motherhood appear to have a larger effect
on boys’ noncognitive development compared to girls’. Particularly striking is the
effect of teenage motherhood. Holding all else constant, boys raised by teenage
mothers have externalizing behavior scores (suspension rates) that are 0.55 stan-
dard deviations (21 percentage points) higher than boys raised by a nonteenage
mother. The equivalent figure for girls is about 0.05 standard deviations (5 percent-
age points). The fact that boys do especially poorly behaviorally when raised by
teenage mothers is particularly relevant to us given the lower share of such mothers
in intact families.
Table 6B replicates the analysis in Table 6A separately for single mother fami-
lies (columns 1–4) and families in which both biological parents are present (col-
umns 5–8). The differential returns by gender documented in the first four columns
are broadly similar across family types. If anything, they are somewhat larger among
children raised by single mothers, especially for teenage motherhood. Although for
a number of the other parental input measures, the estimates are often noisier (likely
in part reflecting the smaller sample sizes in this subgroup of the data).
Online Appendix Table A10 presents the results of the Oaxaca-decompositions
associated with the estimation of these regressions. Model 1 considers columns 1,
2, 5, and 6 in Table 6B; and according to this, it appears that boys’ higher returns to
58 American Economic Journal: applied economicsjanuary 2013

Table 6B—Gender Gap in Noncognitive Skills:

Returns to Family Quality and Parental Inputs by Family Structure (Continued)

Outcome: Suspension in grade 8

Single mom Both biological parents
Girls Boys Girls Boys Girls Boys Girls Boys
(1) (2) (3) (4) (5) (6) (7) (8)
Panel B.
HOME index 0.092* 0.065 0.097* 0.033 0.028** 0.068** 0.008 0.034
(below median) [0.053] [0.081] [0.058] [0.069] [0.013] [0.027] [0.013] [0.026]
Warmth index 0.045 0.047 0.032 0.114* 0.021 0.079*** 0.026* 0.077***
(below median) [0.050] [0.072] [0.052] [0.064] [0.016] [0.027] [0.015] [0.026]
Spanked child last week −0.052 0.241*** −0.054 0.232*** 0.033 0.044 0.028 0.039
[0.052] [0.076] [0.050] [0.066] [0.021] [0.033] [0.020] [0.031]
Age first birth < 20 0.096 0.328*** 0.057* 0.123**
[0.059] [0.073] [0.031] [0.048]
1st SES quintile −0.066 −0.011 0.073* 0.116**
[0.117] [0.103] [0.040] [0.058]
2nd SES quintile −0.108 0.081 0.069** 0.094**
[0.121] [0.103] [0.030] [0.037]
3rd SES quintile −0.042 0.144 0.066*** 0.092**
[0.105] [0.099] [0.024] [0.037]
4th SES quintile −0.166 0.039 0.010 0.031
[0.103] [0.108] [0.012] [0.030]

Background controls Yes Yes Yes Yes Yes Yes Yes Yes
Observations 385 338 385 338 2,155 2,234 2,155 2,234
R2 0.14 0.24 0.17 0.34 0.05 0.06 0.07 0.08

Notes: Each column is a different regression with externalizing behavior in grade 5 (panel A) or suspension in grade
8 (panel B) as the outcome separately by child gender. Columns 1 to 4 restrict the sample to single mom families
and columns 5 to 8 restrict the sample to families with both biological parents present. Background covariates are
identical to Table 6A. Columns 1, 2, 5, and 6 includes three variables for parental inputs measured in the fall of
kindergarten—the HOME and Warmth Indexes are dummy variables that indicate 1 if the child has a HOME or
Warmth score that is below that of the median child in the sample. Spanked child last week is a dummy variable
indicating that the parent reported spanking the child at least once in the past week. Columns 3, 4, 7, and 8 includes
two additional variables that proxy for family quality. The first is a dummy variable that indicates that the child’s
biological mother was less than 20 at first birth, the second variable is a measure of family SES which comprises
five dummy variables that indicate the family SES quintile (the fifth quintile (highest SES)) is the reference cate-
gory. Observations are weighted using eighth grade parent panel weights. Robust standard errors reported.
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
  * Significant at the 10 percent level.

the HOME index, parental warmth index, and discipline can explain up to 13 per-
cent of the larger gender gap in externalizing behavior in fifth grade in broken fami-
lies, and 15 percent of the larger gender gap in eighth grade suspension. Model 2,
corresponding to columns 3, 4, 7, and 8 in Table 6B, includes family SES and teen-
age motherhood; and together, these additional factors explain up to 60 (51) per-
cent of the larger gender gap in externalizing behavior (grade suspension) among
broken families.
In Table 7, we present the results of a final model (Model 3), where we replicate
the analysis from Model 2 but include one additional control: the child’s level of
externalizing problems as measured in the fall of kindergarten. In other words, we
now hold children’s behavior in kindergarten constant and estimate the returns to
Vol. 5 No. 1 bertrand and pan: the trouble with boys 59

Table 7—Gender Gap in Noncognitive Skills in Broken Families:

Role of Family Quality and Parental Inputs, Controlling for Behavior in Fall-K

Outcome: Externalizing behavior in grade 5 Outcome: Suspension in grade 8

Single Both biological Single Both biological
mom parents mom parents
Girls Boys Girls Boys Girls Boys Girls Boys
(1) (2) (3) (4) (5) (6) (7) (8)
HOME index −0.229* −0.221 0.015 0.005 0.086 0.026 0.007 0.030
(below median) (0.124) (0.238) (0.051) (0.067) (0.053) (0.068) (0.013) (0.026)
Warmth index 0.092 0.205 −0.017 0.143** 0.025 0.107* 0.022 0.068***
(below median) (0.106) (0.164) (0.049) (0.061) (0.047) (0.064) (0.015) (0.025)
Spanked child −0.017 0.241 0.039 0.119* −0.079 0.234*** 0.024 0.025
last week (0.118) (0.173) (0.057) (0.072) (0.050) (0.067) (0.021) (0.031)
Age first birth < 20 0.203* 0.728*** −0.003 0.351*** 0.107* 0.322*** 0.056* 0.122***
(0.115) (0.195) (0.081) (0.105) (0.061) (0.070) (0.031) (0.047)
1st SES quintile 0.021 0.038 0.083 0.057 −0.062 −0.031 0.070* 0.099*
(0.168) (0.277) (0.117) (0.138) (0.108) (0.104) (0.039) (0.055)
2nd SES quintile −0.150 0.195 0.203** 0.184* −0.091 0.066 0.069** 0.097***
(0.162) (0.260) (0.095) (0.106) (0.102) (0.104) (0.029) (0.038)
3rd SES quintile −0.108 −0.186 0.081 0.148 −0.058 0.122 0.066*** 0.088**
(0.157) (0.232) (0.064) (0.095) (0.096) (0.097) (0.024) (0.036)
4th SES quintile 0.179 −0.117 −0.003 0.105 −0.164* 0.037 0.009 0.026
(0.205) (0.283) (0.061) (0.083) (0.090) (0.107) (0.012) (0.030)
Externalizing 0.417*** 0.373*** 0.299*** 0.324*** 0.087** 0.057* 0.030** 0.050***
behavior in fall-K (0.063) (0.071) (0.033) (0.031) (0.039) (0.029) (0.014) (0.015)

Background controls Yes Yes Yes Yes Yes Yes Yes Yes
Observations 385 338 2,155 2,234 385 338 2,155 2,234
R2 0.33 0.38 0.17 0.19 0.21 0.35 0.08 0.10

Notes: Each column is a different regression with teacher reported externalizing behavior in grade 5 (columns 1–4)
and suspension in grade 8 (columns 5–8) as the outcome separately by child gender and family structure (single
mom and two biological parent families). All regressions control for background covariates that include race dum-
mies (black, Hispanic, Asian, other), age at assessment at fall-K, age-squared, birthweight, number of older broth-
ers, younger brothers, older sisters, younger sisters, and dummies for region and urbanicity. Sample is restricted to
those with nonmissing observations on family structure, mother’s age at first birth, family SES, gender, the back-
ground covariates, fall-K, and grade 5 teacher ratings of externalizing behavior and parental reports of school sus-
pension in eighth grade. Observations are weighted using eighth grade parent panel weights. Robust standard errors
reported.
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
  * Significant at the 10 percent level.

kindergarten parental inputs on future noncognitive outcomes. This specification

addresses the possibility that differences in the levels and returns to parental inputs
could be merely picking up gender differences in initial externalizing behavior. For
example, mothers could be spending less time and be less warm toward their boys
because boys are more ill behaved than girls, especially in single-parent families.
Another possibility is that gender differences in returns could arise because noncog-
nitive problems are more persistent among boys than among girls. By controlling
for the child’s initial behavior (in fall-K), this gets us closer to being able to interpret
our estimated coefficients in Table 7 as evidence of gender differences in the returns
to parental inputs. The results in Table 7 and the Oaxaca decompositions in Model 3
in online Appendix Table A10 indicate that controlling for the child’s behavior in
60 American Economic Journal: applied economicsjanuary 2013

the fall of kindergarten leaves our main findings virtually unchanged. We continue
to observe larger returns to higher levels of parental inputs and parental quality for
boys compared to girls.
Finally, one potential reverse causation story for the differential returns we observe
is that parents may systematically adjust their input in response to the behavior of
their boys and girls. For example, one possibility is that parents spend more time with
boys who are well behaved but do not discriminate when it comes to spending time
with girls. Online Appendix Table A11 sheds some light on this issue by estimating
dynamic specifications that relate future parental inputs (in grade 3 and grade 5)
on past behavior (externalizing behavior in the fall of kindergarten).30 We estimate
these regressions separately for boys and girls in single mother and intact f­ amilies.
We find little evidence that parental inputs, as measured by the HOME Index, are
affected significantly by a child’s externalizing behavior in kindergarten. This is
true for both boys and girls and across family types. When the outcome variable
is replaced by the parental warmth index in each grade, unlike the HOME Index, it
appears that parental warmth in kindergarten and grade 3 is negatively affected by
the child’s past behavior. The overall patterns, however, run counter to the idea that
parents respond more negatively to boys’ misbehavior. For intact families, parental
warmth in grade 3 is similarly negatively affected by past behavior for both girls and
boys, while for single mom families, an opposite pattern emerges—single moms
appear to be more likely to be less warm to their girls who misbehave, but appear
unaffected by boys’ past behavior. When we look at whether the child was spanked
last week as the outcome, we do not find any systematic evidence that the effect of
past behavior on spanking is larger for boys than for girls. Overall, the findings from
the dynamic regressions suggest that the differential returns documented cannot be
fully explained by reverse causation.

IV. Conclusion

While a study of the gender gap in noncognitive skills is of intellectual interest

per se, our primary motivation for undertaking this study is the accumulating evi-
dence suggesting that boys’ noncognitive deficit might be a primary factor holding
them back from completing the higher levels of education that are demanded in the
skill-biased economies that now characterize most developed countries. We suggest
that boys’ higher tendency to act out and develop conduct problems might be par-
ticularly relevant to their relative absence in colleges.
The biological and medical literatures have rather convincingly established
nature-based explanations for boys’ noncognitive deficit. The fact that we fail to
isolate any subsample of the data where there is no gender gap in noncognitive skills
certainly reinforces our belief that biological forces are at play. However, our find-
ings suggest that social, or nurture-based, influences are also important. In particu-
lar, we find that boys’ higher likelihood to act out and eventually experience a school

30
We have also estimated fully dynamic specification where we look at the cumulative returns of the entire
stream of past behavior (up till t − 1), controlling for previous parental inputs (up till t − 1). The results are broadly
similar and are available upon request.
Vol. 5 No. 1 bertrand and pan: the trouble with boys 61

suspension is about twice as large in the sample of children raised by single mothers,
as well as in the sample of children raised by teenage mothers. On the other hand,
we fail to find any large or systematic variation in the gender gap in noncognitive
skills across some key features of the early school environment, such as age of entry
in kindergarten, how regimented or intellectually demanding the school environ-
ment is, teacher gender, or peers’ noncognitive skills.
When we look deeper into the reasons as to why boys are doing especially poorly
when raised by single mothers, we find evidence suggesting that a small but non-
trivial part of their disadvantage might be related to differential inputs, with sin-
gle mothers investing more in their girls and feeling emotionally closer to them.
Nevertheless, these findings are imprecise due in part to the small sample sizes, and
imperfect input measures available in this dataset.
Most striking though are our findings regarding gender differences in the noncog-
nitive returns to parental inputs. Across all family structures, we observe that boys’
likelihood to act out is sharply reduced when faced with larger and better parental
inputs. For girls, the relationship between parental inputs and behavioral outcomes
appears to be much weaker. As these parental inputs are typically higher and of bet-
ter quality in intact families, this largely contributes to why boys with single moth-
ers are so much more disruptive and eventually face school suspension.
One noteworthy finding is the widening of the gender gap over time in disadvan-
taged families. Boys raised in such families appear to be losing more ground after
the beginning of formal schooling. This is surprising from the perspective that chil-
dren spend more time in the home prior to the beginning of formal schooling, and
hence, one may have expected home influences to be strongest in those early years.
One possible explanation is that the returns to early parental investments are not
fully reflected in boys’ behavior until later ages. Also, high levels of early parental
investments may improve the noncognitive returns to schooling. In any case, this
finding seems to warrant further research.
From a broader perspective, our findings are particularly relevant in light of a
recent literature documenting a growing socio-economic gradient in the amount of
time parents spend with children, with highly educated parents spending increasing
amounts of time in childcare-related activities (Guryan, Hurst, and Kearney 2008).
If one takes our findings at face value, higher amounts of parental time might be
extremely beneficial to the noncognitive development of boys. Assuming that a cer-
tain share of boys are born at the risk of developing behavioral problems, higher
levels of parental investment may prevent more of these at-risk boys from devel-
oping conduct problems. The fact that the growth in parental time is concentrated
among more educated or higher SES families suggests we might see a growing
socio-economic gradient in the gender gap in noncognitive skills and, consequently,
college completion, and future outcomes.

References

Baker, Michael, and Kevin Milligan. 2011. “Sex Differences in the Care of Young Children in Canada
and the U.S.” Unpublished.
Baron-Cohen, Simon. 2002. “The Extreme Male Brain Theory of Autism.” Trends in Cognitive
­Sciences 6 (6): 248–54.
62 American Economic Journal: applied economicsjanuary 2013

Baron-Cohen, Simon. 2003. The Essential Difference: Men, Women, and the Extreme Male Brain. Lon-
don: Allan Lane.
Beaman, Robyn, Kevin Wheldall, and Coral Kemp. 2006. “Differential Teacher Attention to Boys and
Girls in the Classroom.” Educational Review 58 (3): 339–66.
Becker, Gary S., William H. J. Hubbard, and Kevin M. Murphy. 2010. “The Market for College Grad-
uates and the Worldwide Boom in Higher Education of Women.” American Economic Review 100
(2): 229–33.
Bertrand, Marianne, and Jessica Pan. 2011. “The Trouble with Boys: Social Influences and the Gender
Gap in Disruptive Behavior.” National Bureau of Economic Research (NBER) Working Paper 17541.
Bertrand, Marianne, and Jessica Pan. 2013. “The Trouble with Boys: Social Influences and the Gen-
der Gap in Disruptive Behavior: Dataset.” American Economic Journal: Applied Economics. http://
dx.doi.org/10.1257/app.5.1.32.
Chetty, Raj, John N. Friedman, Nathaniel Hilger, Emmanual Saez, Diane Whitmore Schanzenbach,
and Danny Yagan. 2011. “How Does Your Kindergarten Class Affect Your Earnings? Evidence
from Project Star.” Quarterly Journal of Economics 126 (4): 1593–1660.
Cole, Pamela M., Laureen O. Teti, and Caroline Zahn-Waxler. 2003. “Mutual Emotion Regulation and
the Stability of Conduct Problems between Preschool and Early School Age.” Development and
Psychopathology 15 (1): 1–18.
Dahl, Gordon B., and Enrico Moretti. 2008. “The Demand for Sons.” Review of Economic Studies 75
(4): 1085–1120.
Dee, Thomas S. 2006. “The Why Chromosome: How a Teacher’s Gender Affects Boys and Girls.” Edu-
cation Next 6 (4): 68–75.
Demaray, Michelle, Stacey Ruffalo, John Carlson, R. T. Busse, Amy Olson, Susan McManus, and Amy
Levanthal. 1995. “Social Skills Assessment: A Comparative Evaluation of Six Published Rating
Scales.” School Psychology Review 24 (4): 648–71.
Dishion, Thomas J., Terry E. Duncan, J. Mark Eddy, Beverly I. Fagot, and Rebecca Fetrow. 1994.
“The World of Parents and Peers: Coercive Exchanges and Children’s Social Adaptation.” Social
Development 3 (3): 255–68.
Duckworth, Angela L., and Martin E. P. Seligman. 2005. “Self-Discipline Outdoes IQ in Predicting
Academic Performance of Adolescents.” Psychological Science 16 (12): 939–44.
Duncan, Greg J., Chantelle J. Dowsett, Amy Claessens, Katherine Magnusen, Aletha C. Huston,
Pamela Klebanov, Linda S. Pagani, et al. 2007. “School Readiness and Later Achievement.” Devel-
opment Psychology 43 (6): 1428–46.
Elder, Todd E., and Darren H. Lubotsky. 2009. “Kindergarten Entrance Age and Children’s Achieve-
ment: Impacts of State Policies, Family Background, and Peers.” Journal of Human Resources 44
(3): 641–83.
Elliott, Stephen, Frank Gresham, Terry Freeman, and George McCloskey.  1988.  “Teacher and
Observed Rating of Children’s Social Skills: Validation of the Social Skills Rating Scales.” Journal
of Psychoeducational Assessment 6 (2): 152–61.
Else-Quest, Nicole M., Janet Shilbey Hyde, Hill H. Goldsmith, and Carol A. Van Hulle. 2006. “Gender
Differences in Temperament: A Meta-Analysis.” Psychological Bulletin 132 (1): 33–72.
Emery, Robert E., and K. Daniel O’Leary. 1982. “Children’s Perceptions of Marital Discord and
Behavior Problems of Boys and Girls.” Journal of Abnormal Child Psychology 10 (1): 11–24.
Flossman, Anton L., Remi Piatek, and Laura Wichert. 2006. “Going Beyond Returns to Education:
The Role of Noncognitive Skills on Wages in Germany.” Unpublished.
Fortin, Nicole M., Philip Oreopoulos, and Shelley Phipps. 2011. “Leaving Boys Behind: Gender Dis-
parities in High Academic Achievement.” Unpublished.
Fryer, Roland G., Jr., and Steven D. Levitt. 2010. “An Empirical Analysis of the Gender Gap in Math-
ematics.” American Economic Journal: Applied Economics 2 (2): 210–40.
Geary, David C. 1998. Male, Female: The Evolution of Human Sex Differences. Washington, DC:
American Psychological Association.
Geary, David C. 2002. “Sexual Selection and Human Life History.” In Advances in Child Development
and Behavior, Vol. 30, edited by Rivert V. Kail, 41–100. San Diego: Academic Press.
Gilliam, Walter S. 2005. Prekindergartners Left behind: Expulsion Rates in State Prekindergarten Pro-
grams. Yale University Foundation for Child Development Policy Brief Series 3. New Haven, May.
Goldin, Claudia, Lawrence F. Katz, and Ilyana Kuziemko. 2006. “The Homecoming of American Col-
lege Women: The Reversal of the College Gender Gap.” Journal of Economic Perspectives 20 (4):
133–56.
Gresham, Frank M., and Stephen N. Elliot. 1990. Social Skills Rating System. Circle Pines, MN:
American Guidance Service.
Vol. 5 No. 1 bertrand and pan: the trouble with boys 63

Griffin, Kenneth W., Gilbert J. Botvin, Lawrence M. Scheier, Tracy Diaz, and Nicole L. Miller. 2000.
“Parenting Practices as Predictors of Substance Use, Delinquency, and Aggression among Urban
Minority Youth: Moderating Effects of Family Structure and Gender.” Psychology of Addictive
Behaviors 14 (2): 174–84.
Guryan, Jonathan, Erik Hurst, and Melissa S. Kearney. 2008. “Parental Education and Parental Time
with Children.” Journal of Economic Perspectives 22 (3): 23–46.
Harold, Gordon T., and Rand D. Conger. 1997. “Marital Conflict and Adolescent Distress: The Role of
Adolescent Awareness.” Child Development 68 (2): 333–50.
Heckman, James J., Rodrigo R. Pinto, and Peter Savelyev. Forthcoming. “Understanding the Mecha-
nisms Through Which an Influential Early Childhood Program Boosted Adult Outcomes.” Ameri-
can Economic Review.
Heckman, James J., and Yona Rubinstein. 2001. “The Importance of Noncognitive Skills: Lessons
from the GED Testing Program.” American Economic Review 91 (2): 145–49.
Hetherington, E. Mavis. 1989. “Coping with Family Transitions: Winners, Losers, and Survivors.”
Child Development 60 (1): 1–14.
Huizinga, David, Finn-Aage Esbensen, and Anne Wylie Weiher. 1991. “Are There Multiple Paths to
Delinquency?” Journal of Criminal Law and Criminology 82 (1): 83–118.
Jacob, Brian A. 2002. “Where the Boys Aren’t: Non-Cognitive Skills, Returns to School and the Gen-
der Gap in Higher Education.” Economics of Education Review 21 (6): 589–98.
Jencks, Christopher. 1979. Who Gets Ahead? The Determinants of Economic Success in America. New
York: Basic Books.
Kavanagh, Kate, and Hyman Hops. 1994. “Good Girls? Bad Boys? Gender and Development as Con-
text for Diagnosis and Treatment.” In Advances in Clinical Child Psychology, Vol. 16, edited by
Thomas H. Ollendick and Ronald J. Prinz, 45–79. New York: Plenum Press.
Kenney-Benson, Gwen A., Eva M. Pomerantz, Allison M. Ryan, and Helen Patrick. 2006. “Sex Dif-
ferences in Math Performance: The Role of Children’s Approach to Schoolwork.” Developmental
Psychology 42 (1): 11–26.
Knickmeyer, Rebecca, Simon Baron-Cohen, Peter Raggatt, and Kevin Taylor. 2005. “Foetal Testoster-
one, Social Relationships, and Restricted Interests in Children.” Journal of Child Psychology and
Psychiatry 46 (2): 198–210.
Lundberg, Shelly, Sabrina Wulff Pabilonia, and Jennifer Ward-Batts. 2007. “Time Allocation of Par-
ents and Investments in Sons and Daughters.” Unpublished.
Lytton, Hugh, and David M. Romney. 1991. “Parents’ Differential Socialization of Boys and Girls: A
Meta-Analysis.” Psychological Bulletin 109 (2): 267–96.
Maccoby, Eleanor E. 1998. The Two Sexes: Growing Up Apart, Coming Together. Cambridge, MA:
Harvard University Press.
McClure, Erin B. 2000. “A Meta-Analytic Review of Sex Differences in Facial Expression Process-
ing and Their Development in Infants, Children and Adolescents.” Psychological Bulletin 126 (3):
424–53.
McFadyen-Ketchum, Steven A., John E. Bates, Kenneth A. Dodge, and Gregory S. Pettit. 1996. “Pat-
tern of Change in Early Childhood Aggressive-Disruptive Behavior: Gender Differences in Predic-
tions from Early Coercive and Affectionate Mother-Child Interactions.” Child Development 67 (5):
2417–33.
Neidell, Matthew, and Jane Waldfogel. 2010. “Cognitive and Noncognitive Peer Effects in Early Edu-
cation.” Review of Economics and Statistics 92 (3): 562–76.
Ready, Douglas D., Laura F. LoGerfo, David T. Burkam, and Valerie E. Lee. 2005. “Explaining Girls’
Advantage in Kindergarten Literacy Learning: Do Classroom Behaviors Make a Difference?” Ele-
mentary School Journal 106 (1): 21–38.
Rothbaum, Fred, and John R. Weisz. 1994. “Parental Caregiving and Child Externalizing Behavior in
Nonclinical Samples: A Meta-Analysis.” Psychological Bulletin 116 (1): 55–74.
Rowe, David C., Alexander T. Flannery, and David J. Flannery. 1995. “Sex Differences in Crime: Do
Means and Within-Sex Variation Have Similar Causes?” Journal of Research in Crime and Delin-
quency 32 (1): 84–101.
Sax, Leonard. 2007. Boys Adrift: The Five Factors Driving the Growing Epidemic of Unmotivated
Boys and Underachieving Young Men. New York: Basic Books.
Segal, Carmit. Forthcoming. “Misbehavior, Education and Labor Market Outcomes.” Journal of the
European Economic Association.
Silverman, Irwin W. 2003. “Gender Differences in Delay of Gratification: A Meta-Analysis.” Sex Roles
49 (9–10): 451–63.
64 American Economic Journal: applied economicsjanuary 2013

Simons, Ronald L., and Wei Chao. 1996. “Conduct Problems.” In Understanding Differences Between
Divorced and Intact Families: Stress, Interaction, and Child Outcomes, edited by R. L. Simons and
Associates, 125–43. Thousand Oaks: Sage Publications.
Szatmari, Peter, David R. Offord, and Michael H. Boyle. 1989. “Ontario Child Health Study: Preva-
lence of Attention Deficit Disorder with Hyperactivity.” Journal of Child Psychology and Psychia-
try 30 (2): 219–30.
Zahn-Waxler, Caroline, Elizabeth A. Shirtcliff, and Kristine Marceau. 2008. “Disorders of Child-
hood and Adolescence: Gender and Psychopathology.” Annual Review of Clinical Psychology 4
(1): 275–303.