Birth Order and Delinquency: Evidence from Denmark and

∗
Florida

Sanni Breining † Joseph Doyle‡ David N. Figlio Ÿ Krzysztof Karbownik ¶

Jerey Roth k
July 31, 2017

Abstract
Birth order has been found to have a surprisingly large inuence on educational attainment, yet
much less is known about the role of birth order on delinquency outcomes such as disciplinary
problems in school, juvenile delinquency, and adult crime: outcomes that carry signicant neg-
ative externalities. This paper uses particularly rich datasets from Denmark and the state of
Florida to examine these outcomes and explore potential mechanisms. Despite large dierences
in environments across the two areas, we nd remarkably consistent results: in families with
two or more children, second-born boys are on the order of 20 to 40 percent more likely to be
disciplined in school and enter the criminal justice system compared to rst-born boys even when
we compare siblings. The data allow us to examine a range of potential mechanisms, and the
evidence rules out dierences in health at birth and the quality of schools chosen for children.
We do nd that parental time investment measured by time out of the labor force is higher for
rst-borns at ages 2-4, suggesting that the arrival of a second-born child extends early-childhood
parental investments for rst-borns.

∗
We thank Paul Bingley, Dalton Conley, John Donohue, Jens Ludwig, Justin McCrary, Joe Price, and numerous seminar
participants at the Danish Institute for Local and Regional Government Research, NBER Summer Institute Economics of
Crime Workshop, CESifo Economics of Education Group Conference, University of Essex and University of Notre Dame. We
are grateful to the Florida Departments of Education and Health for providing the de-identied, matched data used in this
analysis. Breining gratefully acknowledges nancial support from CIRRAU and Danish Council for Independent Research.
Figlio appreciates funding from the U.S. Department of Education, the National Institutes of Health, and the Bill and Melinda
Gates Foundation. The conclusions expressed in this paper are those of the authors and do not represent the positions of the
Florida Departments of Education and Health or those of our funders.
†
Aarhus University
‡
MIT, NBER and SFI
Ÿ
Northwestern University and NBER
¶
Northwestern University
k
University of Florida
1 Introduction
Economists and policy makers have long been interested in the production function for human

capital, with increasing attention paid to the development of non-cognitive skills (Cunha et al.

2010; Conti et al. 2016). Delinquency is a welfare-relevant manifestation of lower-level acquisition

of non-cognitive skills. At young ages, disruptive peers have been shown to signicantly impair

learning (Carrell and Hoekstra 2010; Carrell et al. 2016; Kristoersen et al. 2014). At older ages,

crime incurs direct harm on others and leads to signicant investments to deter and punish criminal

activity (Nagin 2013). A better understanding of the causes of crime would inform policies aimed

at preventing delinquency in the rst place.

The family environment plays a major role in the development of these non-cognitive skills and

subsequent delinquency. Widom (1989) discusses a cycle of violence where delinquency among

parents is passed on to children. The sheer amount of time children spend with their family and the

inuences each member has on one another imply that the family must have large eects; however

estimating eects of parental investments and sibling inuences can be confounded by endogeneity

concerns.

This paper studies the eects of family environment on child delinquency outcomes through the

lens of the inuence of birth order across siblings. There is a long history of studying birth order

in social psychology and economics (see, for example, Adler (1928), Behrman and Taubman (1986)

or Sulloway (1996)). The main recent empirical contributions stem from work by Black, Devereux

and Salvanes (BDS). Their 2005 study employed large datasets from Norway for outcomes measured

from 1986-2000, and they nd large birth order eects on child outcomes that appear to dominate

those associated with family size. Using between and within-family variation in birth order, later-

born children achieve a lower level of educational attainment and have worse labor-market outcomes.

They also nd a higher likelihood of teen births among women. Our work complements theirs by

bringing to bear evidence on important outcomes rarely explored in the literature with data from

two very dierent settings.

In particular, this paper oers three main contributions. First, we provide some of the rst

estimates of the eects of birth order on delinquency behavior  disciplinary actions and truancy

at school, juvenile delinquency, and adult crime and imprisonment  using large-scale datasets that

facilitate sibling contrasts. A second innovation is that we study these relationships in two very

dierent environments: Denmark and the state of Florida. Using the same empirical framework,

variation in results (or lack thereof ) across these environments provides insights into whether the

eects stem from particular institutions or are more general in their nature. Third, using especially

rich register data in Denmark and linked administrative data in Florida, we are able to investi-

gate potential mechanisms for such dierences. This information includes measures of infant and

childhood health, parental investments, school quality and sibling composition.

Across both of our locations, and across dierent estimation techniques, we nd that second-born

boys are substantially more likely to exhibit delinquency problems compared to their older sibling. In

particular, involvement with the juvenile justice system is found to be on the order of 30-40 percent

1
higher compared to the mean level of involvement among rst-born boys in both Denmark and

Florida. Incarceration by age 21 is also found to be 40 percent higher in Denmark. These eects are

particularly strong among more severe violent crimes (36 percent). In Florida, similarly large eects

are found for suspensions in school (29 percent) but eects on truancy are much more moderate

and heterogeneous. We nd corroborative evidence when we consider a sample of young adolescents

in Denmark where we can measure behavioral problems directly in the form of hyperactivity and

measures of conduct problems by age 12.

In terms of mechanisms, we can rule out large classes of explanations. These include worse health

at birth (second-born children appear healthier) or in childhood (second-born children have fewer

disabilities), schooling decisions including the age of entry and the quality of schools chosen (second-

born children attend no worse schools and are more likely to attend pre-kindergarten and daycare)

as well as maternal employment (measured by maternity leave) in the rst year of life. We do nd

that maternal employment and the use of daycare is higher for second-borns in years 2-4 compared

to older siblings. While it is well known that rst-borns have undivided attention until the arrival

of the second-born, these results show that the arrival of the second-born child has the potential

to extend the early-childhood parental investment in the rst-born child the may compound the

natural bifurcation of parental attention between rst- and second-born children. We also consider

test score outcomes in both Denmark and Florida, and nd that second-born boys score lower on

reading and math assessments in Denmark, but only on reading tests in Florida, in comparison to

their older siblings. This achievement gap may be a signal and a contributor the delinquency results

that we nd later in life.

The remainder of the paper is organized as follows. Section 2 provides a brief background on

the previous literature with an emphasis on the areas where our results represent a contribution.

Section 3 describes the data. Section 4 presents the empirical framework. Section 5 reports the main

results, mechanisms heterogeneity and robustness checks. Section 6 concludes.

2 Background
2.1 Previous evidence on birth order and delinquency
Previous work focusing on delinquency has employed survey data from the National Longitudinal

Study of Adolescent Health (Argys et al. 2006; Averett et al. 2011; Cundi 2013). These studies tend

to nd modest to no evidence of increased crime associated with birth order, along with substance

use and sexual activity. For example, Cundi (2013) uses Add Health data and nds that self-

reported delinquency - a composite measure including marijuana use, non-violent crime, and binge

drinking - is not related to birth order in regression models that include family xed eects. Silles

(2010) studies the National Child Development Study - a study of children born in one week in

1958 in the UK and followed afterward - and nds that rst-born (and last-born) children have

improved non-cognitive test scores. Unfortunately, it is dicult to estimate family xed-eects

models with precision in these survey datasets. For example, Cundi (2013) includes 655 families,

2
and the odds ratio for delinquency is relatively imprecise (0.86  1.52 for middle-born compared to

rst-born children). The current paper relies on much larger administrative datasets that provide

more precision when using within-family variation in birth order to compare siblings.

2.2 Potential mechanisms

While birth order eects in education and labor market outcomes have been explored, the mecha-

nisms behind these results have not been studied extensively. One potential explanation oered for

such eects is dierent levels of parental investments across children. A standard economic model

that reconciles fertility and parental investment in children is the so-called quantity-quality trade-

o  proposed by Becker (1960), and further developed in Becker and Lewis (1973) and Becker and

Tomes (1976). It is based on the premise of negative correlation between the number of children

and income of the parents originating from rising marginal cost of quality with respect to quantity.

In other words, with each additional child it is more expensive to maintain the same quality of

children, and thus parents are going to invest less in human capital of higher birth order children

(Galor and Weil 2000).

1 More recent literature has found mixed evidence on the quantity-quality

tradeo  (Black et al. 2005; Rosenzweig and Zhang 2009; Angrist et al. 2010; Mogstad and Wiswall

2016).

Empirically, parental time investments are found to be lower for later-born children, as the

rst-born has some undivided attention from parents until a younger sibling arrives (Price 2008).
2

Averett et al. (2011) nd that rst-born children are supervised more and have fewer risky behaviors.

Lehmann et al. (2014) also use NLSY data and show that birth order eects on parental-investment

measures appear to be present at ages 0-2 and grow until ages 5-6 when schooling may mitigate such

eects. Interestingly, the paper reports a broad shift in parenting for later-born children: parents

are more likely to miss prenatal visits, less likely to abstain from alcohol during pregnancy and

breastfeed, and parents provide less early-cognitive stimulation at home. Another direct measure

of parental investment is in the form of nancial transfers, and de Haan (2010) used the Wisconsin

Longitudinal Study to show that such transfers are higher for earlier born children.

A second explanation is the inuence that siblings have on one another: the quintessential peer

eect. Later children, unlike rst-borns, will have older siblings as role models (Zajonc and Markus

1975). In addition, older siblings may also benet from teaching younger siblings and acting as

such a role model (Smith 1993). These peer eects are dicult to isolate due to the simultaneity of

inuences (Manski 1995; Joensen and Nielsen 2015; Black et al. 2016; Qureshi 2017). There is also

a literature on the inuence of sibling composition (brothers vs sisters) on educational attainment.

1
An alternative model of the family that can explain birth-order dierences is oered by Hotz and Pantano (2015)
who argue that parental reputation concerns could lead to more intense monitoring and discipline among older children
as a signal to later born children, and this could lead to direct eects on rst-borns such as superior performance at
school and better behavior. Sulloway (1996) presents a related argument, that rst-born children may be more likely
to identify with their parents and therefore to emulate and obey their parents in order to gain parental attention.
2
Price (2008) investigated American Time Use Survey and found that rst-borns get 20-30 minutes more quality
time each day compared to second-born in a similar family. This appears to be due to equalizing time after the second-
born arrives even thought the rst-born had undivided attention when there was only one child in the household.

3
Butcher and Case (1994) found that the presence of brothers historically increased educational

attainment of women, although there is mixed empirical evidence for these eects with some studies

nding no eects and others nding opposite eects (Hauser and Kuo 1998; Kaestner 1997; Conley

2000; Deschenes 2002; Dayioglu et al. 2009; Chen et al. 2009).

3 Data description
3.1 Denmark
For Denmark, the key data source is the Danish Birth Register, which includes information on all

individuals born in the period 1960-2010, and our main outcomes focus on a subset of children born

between 1981 and 1990. For each child, the dataset includes information on exact date of birth

and various birth outcomes. A unique identication number enables us to link generations, hence

children can be linked to their parents and siblings. Given this structure of the data and access to

date of birth, we can measure each individual's birth order, the completed family size and the sex

composition of children in the family.

3

The unique identiers also allow us to match the birth records to rich data from various admin-

istrative registers. This provides us with demographic characteristics of the parents such as age,

educational attainment, labor market status, earnings and immigrant status.

Our primary outcomes reect delinquency behavior. To characterize such behavior in the Danish

context we exploit access to the Criminal Registers including convictions, suspended prison sentences

and incarceration, date of crime as well as type of crime. In Denmark, from the age of 15 an individual

is considered fully responsible for any criminal act, and we can observe all criminal convictions from

that age onward. Our main measure of risky behavior at the youngest age is an indicator of having

any criminal convictions by age 16, and we also consider having any criminal convictions by age

21. As a more severe measure of crime we also consider having ever been issued suspended or

implemented imprisonment sentence by those ages. All outcome variables are multiplied by 100 to

obtain percent values. Our data also allow us to explore eects at other ages, and across dierent

types of crimes.

In addition, we are able to explore measures of non-cognitive skills during childhood. Delinquency

can be viewed as a behavioral outcome that is the consequence of lower levels of non-cognitive

skill development, and there is evidence of birth order eects on non-cognitive skills (Black et al.

2017); however, these studies evaluate non-cognitive skills later in life (at the time of the military

draft), and are not able to measure behaviors that could be related to later life crime such as

emotional, hyperactivity and conduct problems.

4 The Danish data, on the other hand, provide us

with information gathered in the internationally-recognized Strengths and Diculties Questionnaire

3
We can measure completed family size to the extent that subsequent children in a family with rst and second
borns born from 1981 to 1990 are born before December 2010.
4
On the other hand, Rohrer et al. (2015) nd no evidence for birth order eects on extraversion, agreeableness,
conscientiousness, or imagination.

4
 5
(SDQ) , a behavioral screening questionnaire conducted when children are roughly 12 years old.

These data were collected as part of the Danish two-teachers experiment.

6

The SDQ contains 25 items within ve dimensions. Four of the dimensions concerns diculties

covering conduct problems, emotional symptoms, hyperactivity and peer relationships. One dimen-

sion concerns strengths and covers prosocial behavior.

7 In line with past literature, we can use the

four diculty dimensions to generate the Total Diculty Score reecting the child's socio-emotional

behavior. We follow the SDQ scoring rubric (Goodman et al. 2010) and divide these scores into

four categories: close to average, slightly raised, high and very high.
8 Each of these categories will

constitute an outcome that we will study in the analysis below. Similarly we split the scores on the

fth dimension, prosocial behavior. These measures will allow us to investigate whether there are

dierences in psychological attributes or non-cognitive skills by birth order in early adolescence.

One of the innovations in the paper is our ability to explore potential mechanisms that can explain

the birth order eects. The goal is to explore outcomes across a range of ages in an eort to determine

when dierences across siblings begin to occur. First, we consider conditions surrounding the birth

of the child to measure the health endowed to the child, which is partly due to prenatal investments.

Specically we consider birth weight and whether the mother experienced any complications during

the pregnancy.

Second, we consider maternal investments early in a child's life. This includes information about

days the mother spent on maternity leave along with information on maternal employment status

0-10 years after childbirth. From the Register of Daycare Institutions we obtain information about

the populations' use of daycare facilities. Each year between 1995 and 2008 we have information

about its type and place, and we create a measure of enrollment into out-of-home daycare at ages

1-3.

Next, we add a measure of school starting age to see if parents are treating second-born children

dierently at that stage of the life cycle. Then we use information from the Danish National Patient

Registry and Danish Psychiatric Central Registry. From these data we have information about

psychiatric and behavioral diagnoses of all children diagnosed in Denmark (inpatient, outpatient

or ER). We consider two groups of diagnoses: disorders of psychological development as well as

behavioral and emotional disorders, and we observe whether the child has received such a diagnosis

by age 10.
9

5
The Strengths and Diculties Questionnaire was rst developed by Goodman (1997) and later used in numerous
studies (e.g. Currie and Lin 2007; Kelly et al. 2009; Datta Gupta and Simonsen 2010; Berger and Spiess 2011;
Kristoersen et al. 2014).
6
The experiment was conducted in the 2012-13 school year and took place at the intermediate stage of primary
school (6th grade). The sample consists of around 10,000 pupils from 231 public schools across 14 municipalities. If a
school is part of the sample all its 6th graders were included. For more information about this data see 2L Rapport -
Undersøgelse af eekten af tolærerordninger by Andersen et al. (2014). The survey we are using was conducted before
the two-teacher experiment began. Approximately 10,000 sixth grade pupils answered the questions in September
2012. Here they were around 12 years old.
7
For a detailed list of questions see http://www.sdqinfo.com/.
8
Each of the scales of the SDQ is scored from 0-10, and we follow the psychology literature in forming the index
and discretizing it: we add up the four scores to calculate a total diculty score (range 0-40), and then categorize
them such that 80% of children score 'close to average', 10% are 'slightly raised', 5% are 'high' and 5% are 'very high'.
9
Psychological development disorders cover diagnoses classied within the ICD10 codes F80-F89. Behavioral and

5
 To obtain a measure of performance in school we study data from the School Registers. In

Denmark the typical child enters rst grade of compulsory schooling education the year they turn

seven.
10 Compulsory school is completed with a ninth grade exit exam.
11 We measure performance

in school as ninth grade exit exam grades in reading and math. These exam results are based on

assessments by the student's teacher and by an external reviewer. All grades are standardized to

have zero mean and unit standard deviation within each cohort.

3.2 Florida
The Florida Departments of Education and Health merged birth records for all children born in

Florida between 1992 and 2002 with school records for the academic years 1995-96 through 2012-13

for the purpose of this research. The Florida agencies matched children along three dimensions:

rst and last names, date of birth, and social security number. Rather than conducting probabilistic

matching, the match was performed such that a child would be considered matched so long as

(1) there were no more than two instances of modest inconsistencies, and (2) there were no other

children who could plausibly be matched using the same criteria. Common variables excluded from

the match were used as checks of match quality. These checks conrmed a very high and clean match

rate: In the overall match on the entire population, the sex recorded on birth records disagreed with

the sex recorded in school records in about one-one thousandth of one percent of cases, suggesting

that these dierences are almost surely due to typos in the birth or school records.

Between 1992 and 2002, 2,047,663 births were recorded by the Florida Bureau of Vital Statistics;

however, in the current project we can utilize only the 1994 to 2002 births as we are unable to match

siblings for the 1992 and 1993 birth cohorts. There were 1,609,470 singleton births in Florida be-

tween 1994 and 2002, and of these 1,290,077 children were subsequently observed in Florida public

school data, representing an 80.2 percent match rate. The match rate is almost identical to the

percentage of children who are born in Florida, reside there until schooling age, and attend public

school, as computed using data from the decennial Census and American Community Survey for

years 2000 through 2009.

12 In the school districts representing the vast majority of Florida house-

holds, children are matched to households, and we link these anonymized school records back to the

relevant students' birth records (via the birth-school link conducted by the Florida Departments of

Education and Health) to determine birth order; this backward-matching also allows us to check the

likelihood that students we believe to be siblings are actually siblings (e.g., by comparing maternal

characteristics such as date of birth).

13

emotional disorders cover diagnoses classied within the ICD10 codes F90-F98; one of the most frequent diagnoses
within this category is ADHD.
10
A few children may be one year younger and a few may be one year older; we also consider school starting age as
an outcome.
11
The Danish educational system is publicly funded from primary school throughout college. No tuition is charged
and once the student turns 18 he or she is oered monthly nancial aid from the government.
12
Figlio et al. (2013) provide extensive validity checks on the matching procedure.
13
We can only observe births and match siblings among those born between 1994 and 2002. This means that we
do not observe siblings born outside of this data window, and thus cannot compute completed family size in the case
of Florida. However, because of the match to birth records we know the exact birth order as enumerated on the

6
 Florida birth certicates enumerate demographic characteristics of the mother (including educa-

tion, marital status and place of residence), demographic characteristics of the father if he appears

on the birth certicate, and health and demographic characteristics of the newborn.
14 In particular,

we observe birth weight and indicators for any maternal health problems, whether or not they are

related to the pregnancy.

Our primary in-school delinquency outcomes are suspensions and truancy. We obtain these

measures from administrative school records that are available for the school years from 2002-03

through 2011-12. We measure suspensions as an indicator variable whether the child was ever

temporarily removed from school in a given school year, while we measure truancy based on the

reported absence rate, dened as number of days absent (net of suspended days) in school to total

number of days that the child was enrolled in a given school year.
15 Both variables are multiplied

by 100 to obtain percent values.

16 This results in repeated observations for each student, which will

allow us to trace eects across dierent grades.

17

In addition, the Florida Education Data Warehouse provides us with a wide array of additional

outcomes that we use to investigate the mechanisms beyond the identied birth order eects. Our

measures of academic performance are based on Florida Comprehensive Assessment Test (FCAT)

in mathematics and reading, a state-wide standardized yearly assessment of all students in Florida

conducted in grades three through ten. In this paper we focus on test scores in grades three through

eight, because curriculum dierences make interpersonal test score comparisons relatively dicult in

high school (e.g., one tenth grader is taking algebra while the other is enrolled in calculus). Therefore,

each child in the sample can contribute up to six observations, one for each grade observation.

Unlike in Denmark, we cannot match children to their medical records in Florida. However, the

Florida school records also provide an opportunity to observe poor health outcomes in childhood

and adolescence. Specically, we observe whether each child receives special education services for

physical disabilities, such as orthopedic impairments, sensory impairments, and speech pathologies,

as well as intellectual and emotional disabilities. These measures are high-quality indicators of some

birth certicate. Based on this information we select rst births to the mother which means that no rst born in our
empirical sample can have an older sibling, however, both children can have younger siblings born after 2002 that
we do not observe in the data. Given that our results are insensitive to controlling or not controlling for completed
family size (we can observe this information in Denmark and in large Florida county), which is also consistent with
the prior literature, we do not believe that this constraint in Florida state-wide data is a major limitation.
14
In a very small number of cases (fewer than 100 sibling pairs) where the race or ethnicity of the mother on sibling's
birth certicates do not match, we assign the race and ethnicity associated with the birth certicate of the rst-born
child.
15
Florida law denes a habitual truant as a student who has 15 or more unexcused absences within 90 calendar
days with or without the knowledge or consent of the student's parent or guardian. We focus on days absent from
school as an indicator of a proclivity toward truancy.
16
The Florida records also include information on disciplinary referrals that do not result in suspensions, but the
statewide referral data only include the most serious disciplinary infractions and therefore include just a subset of
all disciplinary incidents; 93 percent of the disciplinary incidents reported in the state records result in suspension.
Because we are not entirely certain about the rules governing which non-suspension referrals are included in the data,
we concentrate exclusively on disciplinary referrals that result in suspension.
17
We base our analysis on longitudinal imputed grades in which we assign students to grade they would have been
in had they progressed one grade per year for the rst time we observe them in rst grade, regardless of whether they
were promoted to the next grade. Our results are similar if we focus on actual grades rather than imputed grades.

7
elements of child health as they are linked to school nances and monitored by Florida Department

of Education.

While we have limited ability to measure investment in human capital, we can measure one of

its components, namely school quality as dened by the state of Florida via its school accountability

system. Since 1999, the Florida Department of Education has awarded each of its public schools a

letter grade ranging from A (best) to F (worst). Initially, the grading system was based mainly on

average prociency rates on the FCAT standardized exam. Beginning in 2002, grades were based

on a combination of average FCAT prociency rates and average student level FCAT test score

gains from year to year; other quality indicators, such as competency in science, were subsequently

added.
18 We utilize this information to construct contemporaneous school quality measure. For

ease of implementation, we measure the school letter grades numerically, with A schools receiving 4

points and F schools receiving 0 points. As additional measures of parental investment in children,

we also investigate if the second-born is less likely to attend pre-kindergarten and whether parents

hold back the child a year at the start of schooling.

To measure interactions with the juvenile justice system, one large anonymous Florida school

district reports whether children have been incarcerated in juvenile detention centers.
19 Between

the 1989-90 academic year and the 2004-05 academic year, the school district retained a code in

its student master records for whether the student was in a juvenile detention center, and main-

tained records for that student as long as the student remained registered for school in the district.

Florida children are required to remain in school until their sixteenth birthday, so this variable is

most complete for children under the age of sixteen. Given the ending date of the records for the

juvenile detention code, we can measure juvenile detention through the end of the mandatory school

attendance period for children born on or before August 31, 1989.

20 For the purpose of this juvenile-

delinquency analysis, we measure whether children are siblings based on whether they are co-resident

and have the same last name

21 ; since we can begin to measure residential addresses in this county

in the 1989-90 academic year, we feel most comfortable about our ability to measure birth order for

children born on or after September 1, 1979.

22 Limiting ourselves to rst- and second-borns who

were born between September 1, 1979 and August 31, 1989 leaves us with 24,928 children in 12,464

families. In this school district, we cannot measure variables observed on birth certicates, but we

are able to observe whether a child is limited English procient, receives subsidized meals, or lives

in a low-income micro-neighborhood.

18
See Rouse et al. (2013) for a detailed description of the over-time changes to Florida's school accountability
system.
19
School districts in Florida are county-level. We are unable to link the Florida statewide school records to juvenile
crime data, and the children in the matched birth-schools data are too young to measure juvenile crime for many
sibling pairs in any event.
20
September 1st is the typical cuto for school attendance in any given year.
21
We have also relaxed this denition of sibship to look at consistently coresident children with dierent last names,
and continue to nd similar results, but we prefer the same-name denition because we are most condent there that
the children are full siblings.
22
With this restriction, we will only miscode birth order for families with more than a six year gap between older
siblings and the oldest-observed children enrolled in school in 1989-90.

8
3.3 Sample construction
We construct the analysis samples for both locations in the same way, to the extent possible. As in

the previous literature, we exclude twins from the sample because all of the mechanisms for birth

order eects that we postulate involve the children being the product of successive pregnancies. Our

main analysis employs a sample of the rst two-born children in families with two or more children,

as the previous literature has found that birth order eects are particularly salient for the rst two

children in a family. Further, given our emphasis on delinquency outcomes we focus on families

where the second-born child is a boy, that is boy-boy and girl-boy pairs. The main analysis sample

is also restricted to full biological siblings meaning that all the children have the same mother and

father so that we can exploit variation within these stable families by comparing outcomes across

siblings.
23 We present results for second-born girls, and for other types of families, as separate

investigations later in the paper.

Table 1 reports summary statistics for all families, all families with two or more children where the

rst two children are full biological siblings, and our main analysis sample. In Denmark, the family

characteristics are very similar across these samples, with mother's age at rst birth moderately

lower for families with 2 or more children compared to all families. In Florida, families with 2 or

more children have higher educated mothers, come from higher income ZIP codes, and are more

likely to be white. These dierences are largely due to the restriction that siblings have the same

father. For example, if we study all mothers with two or more children irrespective of the father

status then the fraction of African American mothers increases to 22 percent.

One of the main contributions of the paper is that we can compare birth order eects using

the same types of model specications and similar outcome measures across two very dierent

environments. Table 1 shows that there are clear dierences in the observables between Denmark

and Florida. The average age at rst birth in Denmark is lower than in Florida by about one year.

Within the sample of two or more children the spacing between rst- and second-born is bigger

in Denmark than in Florida. The education systems are dierent, but in Denmark, 32 percent of

mothers have a basic education that ends after 9 years and 23 percent have a college degree, whereas

in Florida 10 percent of mothers in our analysis sample are high school dropouts and 31 percent

have a college degree.

24

Florida has much more racial-ethnic variation than Denmark, and although race or ethnicity are

not reported in the Danish Registers, national statistics reveal that 8.9 percent of people living in

Denmark are immigrants, and of these 58 percent originate from non-western countries. There is

a negligible fraction of immigrants of Hispanic origin. In contrast 22 percent of mothers in Florida

are African American and 24 percent are of Hispanic origin. Finally, the fraction of immigrants is

23
In Denmark we observe complete information on fathers in 98 percent of sibling population and the father is the
same in 92 percent of families across the rst two births. In Florida we observe complete information on fathers in 79
percent of sibling population and the father is the same in 70 percent of families across the rst two births.
24
In about 10 percent of cases we do not observe maternal and paternal education in Denmark. Similarly we do not
observe maternal and paternal employment in 9 percent of cases. In OLS regressions where we use these variables as
controls we impute mean values and create an indicator variable for missing education and/or employment.

9
twice as high in Florida as it is in Denmark, and furthermore the sources of immigration dier. In

Denmark immigrants come mostly from Turkey, Iraq, Bosnia-Herzegovina, Iran, and Pakistan. In

Florida the most frequently-occurring countries are Cuba, Mexico, Colombia, Venezuela, and the

Dominican Republic, but in addition, 44 percent of immigrants to Florida come from non-Hispanic

countries in the Caribbean, Europe, Asia, and Africa.

The last panel of Table 1 reports means for the outcome variables. Interactions with justice

system are not uncommon, and in Denmark by age 21 nearly 11 percent of individuals have a con-

viction, 4.2 percent are given suspended prison sentences and 1.4 percent are actually incarcerated.
25

When we focus on boy-boy and girl-boy families these rates are somewhat higher. In Florida, 5.6

percent of our analysis sample is suspended in a given year and the absence rate is 3.8 percent.

4 Empirical model
The main estimating equations based on boy-boy and girl-boy pairs for person i born in cohort c(i)
in family f (i) are of the form:

Yi = β0 + β1 Secondi + β2 Girli + β3 Xi + δc(i) + µf (i) + εi (1)

where β1 compares second-born boys to their older sibling, β2 is an intercept shift for girls in

girl-boy families, X contains controls described below, δc are cohort eects included to control for the
possibility that secular trends in population-level student performance might confound birth order

estimates, and µf are family xed eects to exploit within-family variation. The xed eects are

appealing because they control for factors that are invariant within families, and we are particularly

interested in comparing siblings. In practice, previous work suggests that family xed eects do

not have a large impact on estimates (Black et al. 2005; Bagger et al. 2013; Härkönen 2014). A

cost of within-family estimation is that standard errors can increase, but with our large samples

precision is not particularly problematic. We report results with and without family xed eects,

and when we exclude family xed eects our vector X contains mother and father age at birth and

education, marital status and immigrant status indicators as well as regional controls (ZIP codes

 1,472 in total  for Florida and municipalities - 275 in total - for Denmark).
26 In Denmark we

further include characteristics at the rst birth such as maternal and paternal employment, family

income and number of children until 2011 as a measure of completed family size, while in Florida

25
We exclude all trac oenses to focus on more serious oenses. Our data on convictions are very similar to
national statistics (Statistik). Our means are between 12 and 20 percent below the national rate of suspended and
implemented incarceration sentences measured in this cumulative way (by age 16 and by age 21). This undercounting
is likely driven by the selected sample of families that we include in our nal analysis.
26
In these models, the introduction of birth cohort indicators helps avoiding omitted variable bias due to dierences
in outcomes across cohorts, however it also introduces an imbalance in maternal age at rst birth across rst and
second born children in the same cohort. We control for mother's age at rst birth and mother's age at the child's
birth, so the model relies on dierent spacing across families to identify the dierent eects. In the end we emphasize
the family xed eects models which avoid the discussed issue, and birth order embeds eects of being reared by
parents who are slightly older for second borns compared to rst borns.

10
we include race and ethnicity indicators.
27 We cluster standard errors at the family level.

In the Florida school records we observe individual students multiple times between grades

one (three) and eight. Since we want to observe children longitudinally as they age, we impose a

longitudinal component into the observed grades. Namely, we start with observations in grade one

and then assign children to subsequent imputed grade levels with each school year, which allows

us to account for grade repeaters. In the main results we pool all the student observations together

and in Figures 2, 4 and A1 we present estimates as children progress through schooling, scaling the

estimates, except for test scores, by the baseline rst-born boy mean for each imputed grade.
28

In the Florida juvenile-delinquency data, we can include an indicator for (rst observed) limited

English prociency, subsidized school meal status (a common measure of socio-economic status), and

average school meal status in student micro-neighborhood. This latter variable is possible because

the school district divides the county into over 1,000 micro-neighborhoods for the purposes of school

assignment and school bus routing and scheduling. While the neighborhoods vary in size, on average

between 50 and 200 students live in a micro-neighborhood at any given time.

29

One estimation concern is that family structure can have its own eects on a family other than

direct birth order eects. Bagger et al. (2013) note that birth order and family size are jointly

determined. That is, one cannot manipulate family size, holding within-family distribution of birth

orders constant. Another example is that families with same-sex children among their rst-borns

are more likely to have additional births - a measure that is often used as an instrument for family

size (Angrist et al. 2010; Conley and Glauber 2006). Black et al. (2005) also use this instrument in

their analysis of family size, but they note that (1) same-sex siblings may have an independent eect

on child outcomes and (2) that the birth order eects are stable when instrumenting for completed

family size. To address these issues, we will show results with and without controls for completed

family size. We will also report results for dierent sex compositions that drive the instrument for

family size used in the literature. This serves as both a robustness check and an exploration of

sibling inuences under the assumption that same-sex siblings have more inuence on one another

compared to mixed-sex siblings.

Another issue is that last-born children may have worse outcomes when families decide not

to continue having children if they have one that is particularly costly e.g. has health problems

(Ejrnæs and Portner 2004). Our main analysis includes families with two or more children, and we

also examine results in families with three or more children so that the second-born is not the last

born.
27
In Denmark maternal and paternal employment as well as family income are measured in the year before birth.
Our xed eects results hold also when we include time varying family controls such as mother's age, education and
marital status as well as zip code/municipality of residence.
28
In Florida we observe test scores for grades three through eight, and thus in Table A3 we restrict our attention
to these grades only. We observe suspensions and truancy in grades one through eight for birth cohorts 1994 to 2000
while three through eight for birth cohorts 2001 and 2002. We construct the imputed longitudinal grade starting in
the rst observed grade in these samples. The results are robust to restricting the delinquency measures to grades
three through eight.
29
Specic numbers are not provided here in order to preserve the anonymity of the school district in question.

11
5 Main results
This section rst reports results for delinquency outcomes (Table 2), which is then followed by the

analysis of dierent types of criminal activity in Denmark (Table 3). Table 4 presents short-run non-

cognitive outcomes at age 12 in Denmark while Table 5 investigates early childhood mechanisms.

We then present short-run cognitive outcomes, namely test scores in Table A3. Subsequent tables

further examine potential mechanisms and heterogeneity focusing on spacing between siblings (Table

6) and maternal education (Table 7). Finally, appendix tables document contrasts between native

born and immigrant parents (Table A6) as well as for Florida racial and ethnic dierences (Table

A7).

5.1 Delinquency
Table 2 reports our main results for delinquency. Panels A, B and C show results for Denmark, where

second-born boys are more likely to have criminal-justice involvement and serve time in prison both

by age 16 and 21. In models with family xed eects, second-born boys are 1.6 percentage-points

more likely to have been found convicted of a crime when aged 15 or 16 compared to rst-borns.

Recall that all second-borns in the analysis sample are boys, and thus in the tables we include the

mean of the outcome for rst born boys as a useful comparison, for example 1.6 percentage points

is 35 percent higher than rst-born boys. By age 21, the estimate suggests a 3.6 percentage-point

increase, or 22 percent higher than the mean for rst-born boys. These results are fairly stable

comparing OLS models with controls and in models with family xed eects. The early-age result

grows somewhat in magnitude when we include family xed eects. The larger eect (compared to

the mean) at younger ages when crime is less common is consistent with birth order eects having

particularly strong inuences for the extremes of the behavioral distribution. We document this

phenomenon in Figure 1, which depicts elevated eects on criminal activity for ages 16 to 18 that

subsequently stabilize for all outcomes.

When we look at more serious crimes - those that yield a prison sentence - the means of the

dependent variables for implemented and suspended sentences among rst-born boys are 0.1 percent

and 1.0 percent by age 16, and 2.7 and 7.2 percent by age 21, respectively. While the outcomes

are exceedingly rare by age 16, we nd that second-born boys are much more likely to have these

outcomes by that age compared to their older siblings. By age 21 when the outcomes are observed

for more young adults, Table 2 shows that second-born boys are found to be 1.1 and 2.4 percentage-

points more likely to have spent time or be sentenced to spending time in prison, respectively, which

implies 40 and 33 percent eects. These are very large eects given that delinquency outcomes are

associated with signicant externalities.

Panel D of Table 2 provides analogous results for juvenile-delinquency by age 16 in a single,

large anonymous county in Florida. In the OLS model without family xed eects but with a wider

set of controls, second-born boys are 0.5 percentage-points more likely to become involved with the

juvenile justice system, which is approximately 21 percent higher than the mean for rst-born boys.

12
When we focus on within-family comparisons by adding family xed eects, the estimate increases

to 1.1 percentage-points, or 42 percent. Notably, even though as documented in Table 1 the state

of Florida is very dierent from Denmark in terms of observable socio-demographic characteristics

as well as the judicial systems, the estimated eects of being born second are remarkably similar

across the two locations.

Another measure of behavioral issues can be found in the Florida schooling data in the form of

suspensions and absences. We dene these as an indicator for being ever suspended and the absence

rate in a given school year multiplied by 100. Remarkably, we nd similar results for suspensions as

they are 1.7-4.3 percentage-points higher (or 29-72 percent higher than the mean for rst-born boys).

For absences, we nd point estimates of 0.01 to 0.35 percentage-points more absences from school,

or 0.2 to 9.4 percent higher. For suspensions, similar to the criminal activity results, we document

moderately larger second-born eects earlier in life, namely in elementary school in comparison

to middle school (left panel of Figure 2). This pattern is not conrmed for absence rates (right

panel of Figure 2), however, where we observe fewer absences for the youngest grades, and more

absences among second-borns in grades 4, 5 and 8.

30 This could be explained by the fact that in

early schooling years absenteeism is much more related to health and acclimatization to education

processes than intentionally missing classes: a sign of misconduct in later grades that corroborates

the delinquency ndings.

31

For the delinquency outcomes measured in Florida schooling data, adding family xed eects

reduces the point estimates. One explanation is that the set of controls varies across panels D and

E due to data availability. Given that there are meaningful controls at our disposal in both settings,

however, the fact that they increase point estimates for some outcomes but decrease point estimates

for others helps to rule out straightforward explanations that the results are due to omitted variable

bias that operates in a similar way across outcomes. In any event, we favor the within-family

comparison including family xed eects to absorb time-invariant characteristics of families and we

proceed with it throughout the subsequent sections of the paper.

32

The uniqueness of Danish registry data allows us to investigate dierent types of crime. Given

the larger eects found here for relatively rare delinquency outcomes compared to the smaller eects

found for more common labor-market outcomes in the literature, it is possible that decreased non-

cognitive abilities aect particularly extreme parts of the distribution. If this working hypothesis

is correct, we should see larger birth order eects for more severe rather than petty crimes. Table

30
Approximately zero eects documented in Table 2 are driven by this heterogeneity over time as the negative early
schooling coecients cancel out with positive coecients is grades 4, 5 and 8. Thus in subsequent tables we mostly
focus on truancy in grades 4 to 8.
31
The early grades results are less negative and the later grades results are more positive for absences when we
estimate OLS regressions with or without controls, and thus in Table 2 in columns (4) and (5) of panel E we estimate
positive and signicant coecients. This discrepancy supports the need for within and across families comparison
rather than just across families comparison.
32
Even though at each grade we compare individuals from the same family, across grades we have a repeated
cross-section of families i.e. not all pairs utilized in grade one comparisons are old enough to make it to grade eight
comparisons. Due to data restriction we are underpowered to run this panel analysis, however when we restrict the
sample to grades one to six then our results are qualitatively consistent with the repeated-cross section estimates for
these grades.

13
3 reports the results for violent crime (e.g. assault, manslaughter, rape or death threats), property

crime (e.g. forgery, arson or burglary) and special crime (e.g. drug oenses, illegal weapon possession

or human tracking) convictions. First, as expected, the conviction rate is the highest for property

crimes, followed by special crimes and lowest for violent crimes. However, the second-born eects

as a proportion of the sample mean are the largest among violent crimes, which is suggestive that

birth order has particularly strong inuences for more extreme outcomes.

5.2 Non-cognitive measures

In addition to measures of delinquency, we are able to study the relationships between birth order

and measures of non-cognitive skills observed as part of the Strengths and Diculties Questionnaire

(SDQ), measured at approximately age twelve for a subset of the Danish population of children.

One limitation of this data set is that we cannot implement the family xed eects strategy due

to observations on only three cohorts in selected schools. Therefore, we present OLS estimates

with a rich set of controls including child birth year dummies, mother's age at rst birth, mother's

age, father's age, municipality dummies, maternal and paternal education, maternal and paternal

employment, family income, immigrant dummy, and number of children in family.

Tables 4 and A1 present the results. Panel A of Table 4 presents total diculties score, and we

show its sub-components - emotional, conducts, hyperactivity and peer problems - in panels A to D

of Table A1. In panel B of Table 4 we present a positive outcome, namely prosociality score.

Second-born boys are about 5 percent less likely to be categorized as close to average in com-

parison to their older siblings, and although the estimates are statistically insignicant, we nd

positive coecients on all the remaining categories of total diculty score. Investigation of the

sub-components of the total score reveals that second-born boys are particularly adversely aected

in terms of hyperactivity. They are 4.9 percentage-points less likely to score close to average on the

hyperactivity component. At the same time, we do not nd any statistically signicant eects for

emotional problems, conduct problems, peer problems and prosocial behavior. Given that we cannot

use family xed eects and only some of the relationships are statistically signicant, some caution

is warranted in the interpretation. Still, we view these results as consistent with Heckman (2000),

who nds non-cognitive skills to be important determinants of success. The suggestive increase in

hyperactivity problems may also be a precursor to criminal-justice involvement later in life (Heller

et al. 2015).

5.3 Mechanisms: Early-life measures

Our rst set of potential mechanisms investigates what happens to these children prior to attending

school, and we present the eects of birth order on maternal, infant and childhood health as well

as probe into several parental-investment channels including dierential labor supply responses or

school choices. If we nd meaningful eects at these younger ages, this provides some insights into

the timing of when dierences begin to appear.

14
 Table 5 presents the results. Columns (1) and (2) consider birth weight and complications in

pregnancy as measures of the health endowment at birth and proxy for maternal health during

pregnancy. Here we nd that second-borns weigh 3.7 percent more on average in Denmark and

2.6 percent more in Florida. Similarly, second-borns have fewer complications during pregnancy

compared to rst-borns (20 percent lower in Denmark, 12 percent lower in Florida). These dierences

are not explained by the fact that mothers are older for second-born births.
33 This result suggests

that second-borns are not starting out in worse health compared to their older sibling, ruling out

one potential mechanism. This is consistent with work by Brenoe and Molitor (2017) and Lundberg

and Svaleryd (2016) who nd that at-birth and early-life health measures such as birth weight or

hospitalizations are higher among rst-borns in both Denmark and Sweden.

Next we consider parental time investment in the form of daycare/preschool and maternal em-

ployment. We nd that daycare at age 1 is 17 percentage-points lower in Denmark for second-born

boys (Table A2). However by age 2 and 3 second-born boys are 3.4 and 10.8 percentage-points more

likely to be enrolled in out-of-home daycare (5 and 14 percent higher than the mean for rst-born

boys). This nding corresponds well with maternal employment eects, which we depict in Figure

3 for 0 to 10 years after child birth. It shows that 2-4 years after birth mothers of second-born boys

are signicantly more likely to be employed compared to mothers of rst-borns. The net eect of

substituting maternal care with institutional care on a range of child outcomes is ambiguous, and in

fact earlier return to the labor market may be harmful for boys (Fan et al. 2015). We additionally

document that maternity leave is modestly shorter for second-born boys compared to the rst-borns

 8.6 fewer days of maternity leave compared to a mean of 150 days for rst-born boys yields about

6 percent eect. It thus appears that the addition of a second-born child results in mothers spending

more of a rst-born's early childhood out of the of labor force: maternity leave upon the arrival of

a second-born adds time that mothers spend with rst-borns in addition to the undivided attention

received by rst-borns in their rst year of life. When we include the endogenous maternal employ-

ment measures in the main specications as a control, we continue to nd sizable birth order eects,

however. This suggests that it is not maternal employment per se that results in the delinquency

dierences, but this choice may be a proxy that is correlated with dierent levels of parental at-

tention for rst-borns not only prior to when the second born arrives but into the rst years of the

second-born's life.

In Florida we cannot observe maternal employment or maternity leave uptake but we do observe

pre-kindergarten participation, which can begin as early as age two. We dene our outcome variable

as either parents declaring pre-kindergarten attendance of their children at the time of kindergarten

enrollment or actually observing children being enrolled in public pre-kindergarten. We nd a

nearly 7 percentage-point higher participation rate among second-born boys, which is equivalent to

33
There is a positive relationship between maternal age and birth weight when we regress log birth weight on
maternal age even if we include family xed eects, however the main results do not change when we add controls
for maternal age at birth. We have also investigated several other birth outcomes, and we nd positive eects for
birth weight in levels, one and ve minutes Apgar scores; negative eects for probability of being low birth weight
and being preterm; and no signicant eect for abnormal conditions at birth or congenital anomalies. These results
are qualitatively similar in both Denmark and Florida.

15
12 percent eect. This nding could indicate a reduction in direct parental investment in favor of

institutionalized care but pre-kindergarten participation is generally viewed in the literature as a

positive investment in human capital (Havnes and Mogstad 2011). If so, then our ndings suggest

that this channel does not explain the dierence in delinquency.

In terms of school starting age, we see a 3.5 percent decrease in age at rst grade in Denmark;

however, we do not document any elevated probability of being held back in rst grade in Florida.

Given that the evidence on school starting age and age at testing are mixed, we view our estimates as

rather not supporting this particular channel of the lower investment in second-born children.
34 In

fact, it seems that second-born boys in Florida are initially attending higher quality primary schools

in comparison to their older siblings, although the school quality gradient changes as children progress

through schooling. In subsequent grades we observe a mix of positive and negative coecients

generally oscillating around zero (Figure 4). Thus, attending lower quality schools located in less-

auent neighborhoods should not explain a large part of our delinquency results.

Both data sets allow us to also investigate childhood health as measured by various disabilities.

About 18 percent of school age children in Florida are diagnosed with some form of disability,

and in our sibling sample we observe 6.8 percent and 6.9 percent of rst-born boys having been

diagnosed with cognitive and behavioral disability, respectively. The incidence of disability diagnoses

in Denmark in far lower with about 2.4 percent on average. Columns (7) and (8) in Table 5 document

no birth order gradient in either developmental or behavioral disabilities for both locations, and if

anything, we actually see a decreased probability of second-born being diagnosed with cognitive

disability in Florida.
35

One innovation in the current paper is that we are able to explore a wide range of potential

alternative mechanisms, including health endowments at birth, parental investment in the form

of maternal employment and the use of daycare, as well as school choice. While we do not nd

the primary mechanism that results in the higher delinquency outcomes later in life, we do nd

relatively precise zero results on some of these measures. On others we nd decreased direct parental

investment in the form of time spent with parents in favor of indirect investment in the form of formal

childcare arrangements. Whether and how these dierences could aect observed later life gaps is

an open research question. This leaves us with parental investments within the home (as opposed

to time out of the home and in the labor force) and sibling inuences as our leading explanations

for the birth order results.

34
Black et al. (2011) document small positive eects of starting school younger but much larger negative eects of
age at test. On the other hand, they also document that boys starting schooling younger have higher incidence of
mental health problems at age 18. The school starting age eects are generally relatively small and thus should not
be the main drivers of our birth order eects in Denmark.
35
In the case of Denmark we have also investigated addiction-related hospitalizations by either age 16 or 21, which
include poisoning by narcotics or psychodysleptics or toxic eects of alcohol. We do not nd any birth order gaps for
these hospitalization outcomes. Similarly we do not nd dierences between second-born boys and their older siblings
in the probability of dying by either 16 or 21. Finally, we looked at the number of emergency room visits by age 5
and 10. For both of these outcomes we nd that second-born boys have about 10 percent more visits than their older
siblings; however, we are unable to distinguish if these are due to poorer underlying health or more accidents related
to e.g. increased hyperactivity documented in Table A1.

16
5.4 Test scores
Our next set of results considers childhood test scores. These are of interest in their own right

as educational outcomes, and indeed have been studied in the previous literature. We believe the

similarity in the results with the previous literature, to the extent it occurs, bolsters the credibility

of our empirical approach and corroborates the quality of the data used in both settings. We are

also interested in test scores to investigate whether the behavioral dierences identied in our main

results are also evident in these measures that are more traditionally thought of as measures of

cognitive skills, and these could themselves have eects on delinquency. In Denmark, we observe

them for 9th grade, so slightly younger than our crime by age 16 results, and in Florida we have

results for younger children - grades 3-8 - so as early as age 8. If we nd eects at these younger

ages, this may provide some insights into the timing of when dierences begin to appear, and also

whether they grow or shrink as children age.

Table A3 shows that in Denmark, second-born boys have lower test scores: 0.11 and 0.14 standard

deviations for math and reading, respectively. In Florida, we do not nd a birth-order eect for math

scores, but we do nd that reading scores are 0.08 standard deviations lower for second-born boys.

When we look at each grade separately (Figure A1), we nd that the positive but insignicant

estimate for mathematics is driven by early grades. In fact, for grades three and four we nd

positive and statistically signicant eects of being the second-born. This result would be consistent

with the higher early investment into cognitive skills of second-born boys that fades out over time,

allowing the non-cognitive aspects to dominate. For reading we observe negative estimates in all

studied grades that, except for grade 8, are relatively constant over time.

5.5 Heterogeneous delinquency eects

We have documented very similar eects of birth order on delinquency in both Denmark and Florida;

however, the richness of the data in both locations allows us to investigate heterogeneity of these

eects in much more detail. Given the precision of the results, we tend not to nd statistically

signicant dierences across subgroups, but an exploration of the changing point estimates can

point toward areas for future research and begin to separate dierent explanations. We divide this

analysis into two parts. First, we continue to consider potential mechanisms through spacing and sex

composition of the sibship. Investigation of distance between siblings in terms of age and sex could

provide us with some clues about implicit dierential investments related to the nature of family

arrangements. Second, we investigate if the birth order eects are dierent in more versus less

advantaged families  this is proxied by education, maternal employment (only Denmark), nativity

as well as race and ethnicity (only Florida).

We begin by presenting in Table 6 the dierential eects of birth order by spacing between rst-

and second-born children in our data to investigate whether the results vary by family structure.
36

To preserve sample sizes, we divide families into those that had the second birth up to two and a half

36
Buckles and Munnich (2012) document that a one-year increase in spacing increases test scores for older siblings
by about 0.17 SD, which is comparable to their birth order eect.

17
years after the rst one and those where the distance in age between siblings is greater than two and

a half years. We nd that in both locations delinquency is generally higher among families where

the children are closer in age. That said, there are dierences in the relative magnitudes of the birth

order eects by spacing groups in Denmark versus Florida. In Denmark we nd larger birth order

eects for the closer spaced siblings while in Florida we nd that sibling pairs with longer spacing

duration have relatively larger birth order eects. We are underpowered to detect statistically

signicant birth order eects by spacing in the unnamed Florida county where we observe juvenile

justice system interactions, but the results are larger for siblings spaced further apart akin to what

we found for suspensions. In Denmark the results on convictions at age 16 are weaker and we do

not have precise estimates, yet the point estimate for prison sentences by age 16 is substantially

larger for families where the siblings are closely spaced. Our results are more precise at age 21 where

we see both higher convictions and incarceration sentencing for the second-borns born up to two

and a half years after their older sibling. Unfortunately, these spacing comparisons cannot separate

the leading explanations of the sources of birth order eects, as spacing aects not only parental

capacity for investment in children but also simultaneously changes the inuence siblings have on

one another.
37

The results for Denmark are broadly consistent with the idea that resources and parental atten-

tion are diluted more in families where the children are born closer to one another. If boys require

more attention from parents (Bertrand and Pan 2013; Autor et al. 2016) then this nding could

help to explain why more closely-spaced sibling pairs have greater incidence of delinquency. It's not

entirely clear why the birth order eect is greater in Denmark for more closely spaced siblings while

the opposite is true for Florida. One possibility could involve dierent practices regarding maternity

leave following pregnancy. As we observed in Section 5.3, a potential mechanism for birth order

eects, at least in Denmark, where we observe maternal labor market outcomes, is that the rst-

born child experiences extra maternal attention when very young and again when the second-born

child is born; if maternal attention is dierentially important during very early childhood, then this

dierence could generate the nding that the birth order eect is particularly pronounced in Den-

mark when children are closely spaced. In Florida, maternity leave policies are not as generous, so

one might imagine that the double-dose of extra maternal attention would be attenuated, thereby

putting more weight on the time at home without siblings explanation, which privileges rst-born

children the longer the birth spacing. These explanations, of course, are speculative.

In addition to birth spacing, another family-structure measure of interest is the sibling sex

composition as described in the discussion of the empirical model. We chose to include all families

that have a second-born boy in the main specication because restricting the analysis to boy-boy

families could introduce a sample selection bias. For example, two children of the same gender can

impact family size, which can have its own eects on our outcomes. Nevertheless, if there is gender

antagonism between siblings then it is plausible that the second-born eects would be larger among

the girl-boy families. On the other hand, if there is gender complementarity in delinquency behavior

37
We also explored how results vary by spacing and by sibling composition, however the results were both imprecisely
estimated and the point estimates did not point to a consistent explanation for the ndings.

18
(e.g., it is higher among individuals of the same sex due to, for example, peer pressure) we might

expect to observe higher estimates among the boy-boy pairs. The results along these dimensions are

largely mixed (Table A4). In the case of both types of incarceration sentences in Denmark, boys fare

somewhat worse in families with an older sister as compared to with an older brother. For example,

in the case of incarceration by age 21, second-born boys in girl-boy families have an expected rate of

4.4 percent whereas second-born boys in boy-boy families have an expected rate of 3.3 percent, both

of which are higher than the mean for rst-born boys of 2.7 percent. On the other hand, the results

are similar in the two family compositions for convictions in Denmark and truancy in Florida, while

the pattern is actually reversed for juvenile incarceration in Florida.

One explanation for birth order eects is that parental investment is diluted with the arrival of

more children, which motivates us to investigate heterogeneity by socio-economic status. In Table 7

we divide families into those with mothers having less than 12 years of formal education (equivalent

of high school dropout in the United States) and those who have 12 or more years of schooling. More

educated families tend to have more resources, live in more auent and safer neighborhoods, and

are able to send their children to higher-rated schools. It is also the case that delinquency rates are

more than twice as high in the less educated sample for virtually all our measures. That said, the

estimates for Denmark suggest very similar second-born eects for convictions among both high and

low educated parents. The estimates are larger among the families where mother has less education,

but this is oset by the higher baseline delinquency rate among children raised in these households.

This pattern is somewhat dierent for crimes that involve either form of prison sentence. Specically,

by age 16 the dierences in rates are much larger among households with lower educated mothers;

relative to the sample means, this pattern is reversed by age 21. In Florida, the estimated eect

sizes are similar for both suspensions and truancy, albeit much larger for suspensions from school.

A similar resource availability story can be examined by looking at maternal employment, al-

though this also impacts time spent with children as discussed above. Our results, reported in ap-

pendix Table A5, document that delinquency is on average higher in households where the mother

is not employed one year after birth in case of both children. The point estimates and eect-sizes

are also higher for the non-employed samples with the exception of convictions by the age of 21

where after scaling they are comparable. Thus, especially for the teenage crime it appears that

the benecial eects of maternal employment such as higher income may dominate the potential

negative impacts for these more extreme outcomes.

There exists ample evidence that children with immigrant backgrounds fare worse in terms of

many later life outcomes (Borjas 2006; Abramitzky and Boustan 2016) although the dierences vary

for schooling outcomes, with Aslund et al. (2011) documenting relative disadvantage in Sweden while

Figlio and Ozek (2016) showing superior educational outcomes for the children of immigrants in the

United States. Very little is known about the birth order eects in these families. We document

the gradient between second-born boys and their older siblings for immigrant and native families

in Table A6. First, it is worth noting that the children of immigrants in Denmark have higher

convictions and prison sentencing rates, but the children of immigrants in Florida have lower rates

19
of suspensions and absences in comparison to the children of native-born mothers. In both locations,

however, we nd larger second-born eect sizes among the children of immigrants relative to the

children of native-born residents with a notable exception of teen incarceration in Denmark where

the pattern is reversed.

Although Denmark is fairly racially and ethnically homogeneous, in Florida the student popula-

tion we investigate is 20 percent African American and another 18 percent Hispanic. It is also the

case that African Americans and Hispanics have on average lower socio-economic status and their

children fare worse in schooling outcomes. Thus, we present results separately for African Ameri-

cans, non-African Americans, Hispanics and non-Hispanics in Table A7. Absenteeism is lower among

these two minority groups in comparison to the rest of the population, but the suspension rate for

African Americans is more than twice as high as for non-African Americans. The second-born eect

is similar among Hispanics and non-Hispanics for suspensions but larger for the former group for

truancy. When comparing African Americans to non-African Americans, however, the eect sizes

for both outcomes are larger among non-African Americans relative to the means of the subsamples.

Thus far we have focused our analysis on a more positively selected sample of stable families

where the mother and father are the same for both children. However, in Florida 30 percent of

children in families with two or more children have dierent or unknown fathers for the rst two

births, while in Denmark this number is lower (9.5 percent) but still non-trivial. Since family

instability is likely correlated with behavioral problems, and thus delinquency (Autor et al. 2016),

we did not want to pool together these two types of families. In Table A8 we present separately

the results for boy-boy and girl-boy families with dierent fathers across the two births.
38 First,

with the exception of truancy, the delinquency rates are much higher in these families in comparison

to the baseline sample - from about two times for convictions and suspensions to ve times for

teenage incarceration. Second, the samples are much smaller resulting in much larger standard

errors. Nevertheless, for convictions and suspensions, we nd very similar eect sizes relative to the

sample mean compared to Table 2. In cases where the mean values increase the most across samples,

suspended or implemented imprisonment sentences, we do not estimate any signicant birth order

gaps. The point estimates are also substantially smaller than these documented for more stable

families. A hypothesis consistent with this set of results may suggest that resources and parental

attention that could prevent the most extreme behavior do not matter so much in families with

changing male role models. It is also possible that changing family structure might alter the ways

in which male role models treat rst-born versus second-born children.

5.6 Robustness checks

Given the very large magnitude of the dierences between rst- and second-born children, we check

the robustness of the results in a number of ways. In the main set of results our xed eects

specication controls only for cohort eects but it is possible that our estimates are still biased

38
The set of families where the identity of the father is unknown for at least one birth is 20.6 percent in Florida
and only 2.3 percent in Denmark.

20
because we do not account for mother time varying characteristics like marital status, education

or place of residence. These characteristics could change across births and inuence the second-

born either positively (e.g. mother improving her education) or negatively (e.g. family moving

to a poorer neighborhood due to reduced resources). When we control for maternal age at birth,

maternal education, maternal marital status and location at the time of birth in the xed eects

models our estimates are similar (panel A in Table A9).

In Table 5 we have documented that second-borns are moderately healthier at the time of birth

in comparison to their older siblings, and thus, if there is a causal eect of health at birth on

delinquency then the main results could be mediated through improved health. When we control

for birth outcomes in our preferred specication (panel B of Table A9) the results do not change

which suggests that better health in infancy does not moderate the eects on delinquency. At the

same time, because health at birth has a causal eect on cognitive development (Figlio et al. 2014)

our second-born estimates for test scores increase in magnitude when we control for birth weight.

One explanation for our main results pertains to potential stopping rules that parents use when

determining family size. In our preferred sample we focus on the rst two births among families

with two or more children, but it is possible that issues associated with the second-born child might

induce some families to cease their fertility following the second-born child. To address this issue,

we drop from the sample all families for whom we observe exactly two children (panel C in Table

A9). This cuts our sample by about 60 and 85 percent in Denmark and Florida, respectively. In

Denmark this is almost entirely due to families stopping their fertility at parity two, while in Florida

we lose even more observations because of the nature of the data  since we only observe births

over 9 years, it is possible that the families which we drop now as having exactly two children will

continue their fertility beyond the year where we can observe them. In these larger families the

average delinquency rates for the rst-born boys are higher than in the preferred sample which may

be related to the lower socioeconomic status of families with higher fertility. Nonetheless, our results

are very stable whether we include exact parity two families or not.

5.7 Eects for girls

We primarily focus on boys in this paper because criminal justice involvement and other forms of

delinquency are much less common among girls. For example, less than 8 per ten thousand rst-born

girls are incarcerated by age 21 in Denmark, as compared to 271 rst-born boys per ten thousand

in our analysis sample. The results are imprecise for the sample of girls only, but suggest that

second-born girls have higher delinquency compared to rst-born sisters (all compared to a very

small base). In the sample of boy-girl pairs, however, second-born girls have lower delinquency

compared to rst-born girls in general in four out of nine outcomes as documented in Table A10.

School suspensions are somewhat higher for second born girls, and truancy occurs at a similar rate

across genders and across sibling sex compositions. Overall, we view these results for girls as mixed,

but also less policy relevant given our focus on delinquency outcomes that are much more prevalent

among boys.

21
 We also investigate the degree to which any dierences in birth order eects by second-born

gender are possibly due to dierent mechanisms aecting girls versus boys. For instance, it may be

the case that second-born girls are in even better health, or that parents invest more in early human

capital if the second-born child is female. As can be seen in appendix Table A11, compared to the

parallel second-born boy mechanisms presented in Table 5, most of the potential mechanisms that we

explored (e.g., neonatal health, pre-kindergarten participation, and school quality) are remarkably

similar for second-born boys and second-born girls. This general pattern of ndings leads us to

believe that the mechanisms we explored do not account for a substantial part of boy-girl dierences

in birth order eects on delinquency.

6 Conclusions
Previous research has paid considerable attention to identifying the eects of birth order on a range

of cognitive outcomes such as test scores, educational attainment, and wages. We contribute to this

literature by studying a set of more extreme outcomes relating to delinquency. In order to gauge the

degree to which birth order is determining delinquent behavior, we perform, to the extent practicable,

parallel analyses in two considerably dierent societies  Denmark and the United States (Florida).

These societies have much dierent approaches to crime and punishment and are demographically

very dierent, but our sets of ndings are fundamentally quite similar.

We nd, consistently across both locations, that second-born boys are substantially more likely to

exhibit delinquency problems compared to their older sibling. Second-born boys in Denmark have

substantially higher rates of juvenile crime, particularly severe violent crime, and imprisonment

than do their older siblings. We observe antecedents for these patterns in terms of early delinquency

problems (especially disciplinary behavior that results in suspensions but also in middle school

truancy) as well as in terms of direct measures of conduct problems and hyperactivity at age 12.

Our data allow us to explore a wide variety of early-life outcomes in order to study potential

mechanisms through which birth order eects are occurring. We are able to rule out broad classes

of explanations. We nd no evidence that second-born children are less healthy, and indeed second-

born children appear to be healthier at birth and have lower rates of disability in childhood. We also

nd no evidence that parents invest less in second-born children's education. These children attend

no-worse schools and are more likely to attend pre-kindergarten. We consider dierences in parental

attention as a potential contributing factor to the gaps in delinquency across the birth order. In

our administrative data the measures of direct parental investment include maternal employment

and use of daycare. We discovered that the arrival of the second-born child has the potential to

extend the early-childhood parental investment in the rst-born child, in addition to the fact that

rst-borns experience undivided attention until the arrival of the second-born.

These new results have important implications for social policy. Crime, delinquency, and incar-

ceration have enormous social costs and are associated with major losses in human potential. Our

ndings that birth order appears to inuence the likelihood of delinquency among boys, and that

22
dierences begin to appear early, suggests potentially fruitful avenues for monitoring and interven-

tions. Our ndings regarding systematically dierent dosages of early-childhood parental attention

as a plausible mechanism also provide a call for future research on this topic and engender further

discussion of parental leave as a long-run social benet.

23
References
Abramitzky, Ran and Leah Boustan,  Immigration in American Economic History, Journal
of Economic Literature, 2016, forthcoming.
Adler, Alfred,  Characteristics of the rst, second, and third child, Children, 1928, 3 (5), 14.
Andersen, Simon, Louise Beuchert-Pedersen, Skyt Helena Nielsen, and Mette Thom-
sen,  2L Rapport: Undersogelse af eekten af tolaererordninger, 2014.
Angrist, Joshua D, Victor Lavy, and Analia Schlosser,  Multiple Experiments for the Causal
Link between the Quantity and Quality of Children, Journal of Labor Economics, 2010.
Argys, Laura, Susan Averett, and Daniel Rees,  The Eect of Older Siblings on Adolescent
Behavior: Is it Siblings or Parents, 2006.

Aslund, Olof, Per-Anders Edin, Peter Fredriksson, and Hans Gronqvist,  Peers, Neigh-
borhoods, and Immigrant Student Achievement: Evidence from a Placement Policy, American
Economic Journal: Applied Economics, 2011.
Autor, David, David Figlio, Krzysztof Karbownik, Jerey Roth, and Melanie Wasser-
man,  Family Disadvantage and the Gender Gap in Behavioral and Educational Outcomes,
2016.

Averett, Susan, Laura Argys, and Daniel Rees,  Older Siblings and Adolescent Risky Behav-
ior: Does Parenting Play a Role?, Journal of Population Economics, 2011, 24 (3), 957978.
Bagger, Jesper, Javier Birchenall, Hani Mansour, and Sergio Urzua,  Education, Birth

Order, and Family Size, 2013.

Becker, Gary S.,  An Economic Analysis of Fertility, in Demographic and Economic Change in

Developed Countries, Columbia University Press, 1960, pp. 209240.

and Gregg H. Lewis,  On the Interaction between the Quantity and Quality of Children,

Journal of Political Economy, 1973, 81 (2), S279S288.

and Nigel Tomes,  Child Endowments and the Quantity and Quality of Children, Journal of
Political Economy, 1976, 84 (4), S143S162.
Behrman, Jere R. and Paul Taubman,  Birth Order, Schooling, and Earnings, Journal of
Labor Economics, 1986, 4 (3), 121145.
Berger, Eva M. and Katharina C. Spiess,  Maternal Life Satisfaction and Child Outcomes:

Are They Related?, Journal of Economic Psychology, 2011, 32 (1), 142158.

Bertrand, Marianne and Jessica Pan,  The Trouble with Boys: Social Inuence and the Gender

Gap in Disruptive Behavior, American Economic Journal: Applied Economics, 2013, 5 (1), 3264.

24
Black, Sandra, Erik Gronqvist, and Bjorn Ockert,  Born to Lead? The Eect of Birth Order

on Non-Cognitive Skills, Review of Economics and Statistics, 2017, forthcoming.

, Paul Devereux, and Kjell Salvanes,  The More The Merrier? The Eects of Family Size

and Birth Order on Children's Education, Quarterly Journal of Economics, 2005, pp. 669700.
, , and ,  Too Young to Leave the Nest? The Eects of School Starting Age, Review of
Economics and Statistics, 2011, 93 (2), 455467.
, Sanni Breining, David Figlio, Jonathan Guryan, Krzysztof Karbownik, Skyt Helena
Nielsen, Jerey Roth, and Marianne Simonsen,  Sibling Spillovers, 2016.
Borjas, George,  Making it in America: Social Mobility in the Immigrant Population, Future of
Children, 2006, 16 (2), 5571.
Brenoe, Anne and Ramona Molitor,  Birth order and health of newborns: What can we learn

from Danish registry data, Journal of Population Economics, 2017, forthcoming.

Buckles, Kasey and Elizabeth Munnich,  Birth Spacing and Sibling Outcomes, Journal of
Human Resources, 2012, 47 (3), 613642.
Butcher, Kristin F. and Anne Case,  The Eect of Sibling Sex Composition on Women's

Education and Earnings, Quarterly Journal of Economics, 1994, 109 (3), 531563.
Carrell, Scott E. and Mark Hoekstra,  Externalities in the Classroom: How Children Exposed

to Domestic Violence Aect Everyone's Kids, American Economic Journal: Applied Economics,
2010, 2 (1), 211228.
, , and Elira Kuka,  The Long-Run Eects of Disruptive Peers, 2016.

Chen, Yen-Chien, Stacey H. Chen, and Jin-Tau Liu,  Separate Eects of Sibling Gender and
Family Size on Educational Achievements - Methods and First Evidence from Population Birth

Registry, 2009.

Conley, Dalton,  Sibship Sex Composition: Eects on Educational Attainment, Social Science
Research, 2000, 29 (3), 441457.
and Rebecca Glauber,  Parental Educational Investment and Children's Academic Risk: Es-

timates of the Impact of Sibship Size and Birth Order from Exogenous Variation in Fertility,

Journal of Human Resources, 2006, 41, 722737.

Conti, Gabriella, James J Heckman, and Rodrigo Pinto,  The Eects of Two Inuential

Early Childhood Interventions on Health and Healthy Behaviors, Economic Journal, 2016, 126
(596), F28F65.

Cundi, Patrick,  Ordered Delinquency: The "Eects" of Birth Order on Delinquency, Person-
ality and Social Psychology Bulletin, 2013, 39 (8), 10171029.

25
Cunha, Flavio, James J Heckman, and Schennach Susanne M.,  Estimating the Technology
of Cognitive and Noncognitive Skill Formation, Econometrica, 2010, 78 (3), 883931.
Currie, Janet and Wanchuan Lin,  Chipping away at health: More on the relationship between

income and child health, Health Aairs, 2007, 26, 331344.

Dayioglu, Meltem, Murat G. Kirdar, and Aysit Tansel,  Impact of Sibship Size, Birth Order
and Sex Composition on School Enrolment in Urban Turkey, Oxford Bulletin of Economics and
Statistics, 2009, 71 (3), 399426.
de Haan, Monique,  Birth order, family size and educational attainment, Economics of Education
Review, 2010, 29, 576588.
Deschenes, Olivier,  Estimating the Eects of Family Background on the Return to Schooling,
2002.

Ejrnæs, Mette and Claus C. Portner,  Birth Order and the Intrahousehold Allocation of Time
and Education, Review of Economics and Statistics, 2004, 86 (4), 10081019.
Fan, Xiaodong, Hanming Fang, and Simen Markussen,  Mother's Employment and Chil-

dren's Educational Gender Gap, 2015.

Figlio, David and Umut Ozek,  Cross-Generational Dierences in Educational Outcomes in the
Second Great Wave of Immigration, 2016.

, Jonathan Guryan, Krzysztof Karbownik, and Jerey Roth, The Eects of Poor Neona-
tal Health on Children's Cognitive Development, 2013.

, , , and ,  The Eects of Poor Neonatal Health on Children's Cognitive Development,

American Economic Review, 2014, 104 (12), 392155.

Galor, Oded and David N. Weil,  Population, Technology, and Growth: From Malthusian

Stagnation to the Demographic Transition and Beyond, American Economic Review, 2000, 90
(4), 806828.

Goodman, Anna, Richard Rowe, and Ye Gan,  http://www.sdqinfo.com/c3.html, 2010.

Goodman, Robert,  The Strengths and Diculties Questionnaire: A Research Note, Journal of
Child Psychology and Psychiatry, 1997, 38, 581586.
Gupta, Nabanita Datta and Marianne Simonsen,  Eects of Universal Child Care Participa-
tion on Pre-teen Skills and Risky Behaviors, 2010.

Härkönen, Juho,  Birth Order Eects on Educational Attainment and Educational Transitions in
West Germany, European Sociological Review, 2014, 30 (2), 166179.

26
Hauser, Robert M. and Hsiang-Hui D. Kuo,  Does the Gender Composition of Sibships Aect
Women's Educational Attainment?, Journal of Human Resources, 1998, 33 (3), 644657.
Havnes, Tarjei and Magne Mogstad,  No Child Left Behind: Subsidized Child Care and Chil-

dren's Long-Run Outcomes, American Economic Journal: Economic Policy, 2011, 3 (2), 97129.
Heckman, James J,  Policies to foster human capital, Research in Economics, 2000, 54, 356.
Heller, Sara B., Anuj K. Shah, Jonathan Guryan, Jens Ludwig, Sendhil Mullainathan,
and Harold A. Pollack,  Thinking Fast and Slow? Some Field Experiments to Reduce Crime
and Dropout in Chicago, 2015.

Hotz, Joseph V. and Juan Pantano,  Strategic Parenting, Birth Order and School Performance,
Journal of Population Economics, 2015.
Joensen, Juanna and Skyt Helena Nielsen,  Spillovers in Educational Choice, 2015.

Kaestner, Robert,  Are Brothers Really Better? Sibling Sex Composition and Educational

Achievement Revisited, Journal of Human Resources, 1997, 32 (2), 250284.

Kelly, Yvonne, Amanda Sacker, Ron Gray, Jon Kelly, Dieter Wolke, and Maria A.
Quigley,  Light drinking in pregnancy, a risk for behavioural problems and cognitive decits at
3 years of age?, International Journal of Epidemiology, 2009, 38, 129140.
Kristoersen, Jannie, Morten Kraegpoth, Skyt Helena Nielsen, and Marianne Simon-
sen,  Disruptive School Peers and Student Outcomes, 2014.
Lehmann, Jee-Yeon K., Ana Nuevo-Chiquero, and Marian Vidal-Fernandez,  The early
origins of birth order dierences in children's outcomes and parental behavior, 2014.

Lundberg, Evelina and Helena Svaleryd,  Birth Order and Child Health, 2016.

Manski, Charles F.,  Identication problems in social sciences, 1995.

Mogstad, Magne and Matthew Wiswall,  Testing the Quantity-Quality Model of Fertility:

Linearity, Marginal Eects, and Total Eects, Quantitative Economics, 2016.

Nagin, Daniel,  Deterrence: A Review of the Evidence by a Criminologist for Economists, Annual
Review of Economics, 2013, 5, 83105.
Price, Joseph,  Parent-Child Quality Time: Does Birth Order Matter?, Journal of Human Re-
sources, 2008, 43 (1), 240265.
Qureshi, Javaeria,  Siblings, Teachers and Spillovers on Academic Achievement, Journal of Hu-
man Resources, 2017, forthcoming.
Rohrer, Julia, Boris Eglo, and Stefan Schmukle,  Examining the eects of birth order on
personality, Proceedings of the National Academy of Sciences, 2015, 112 (46), 1422414229.

27
Rosenzweig, Mark and Junsen Zhang,  Do Population Control Policies Induce More Human
Capital Investment? Twins, Birth Weight and China's One-Child Policy, Review of Economic
Studies, 2009, 76 (3), 11491174.
Rouse, Cecilia, David Figlio, Dan Goldhaber, and Jane Hannaway,  Feeling the Florida

Heat? How Low-Performing Schools Respond to Voucher and Accountability Pressure, American
Economic Journal: Economic Policy, 2013, 5 (2), 251281.
Silles, Mary,  The Implications of Family Size and Birth Order for Test Scores and Behavioral

Development, Economics of Education Review, 2010, 29 (5), 795803.

Smith, Thomas E.,  Academic Achievement and Teaching Younger Siblings, Social Psychology
Quarterly, 1993, 56 (1), 7785.
Statistik, Danmarks, Fodselsargange efter kon, alder ved forste dom og lovovertraedelse.
Sulloway, Frank, Born to Rebel: Birth Order, Family Dynamics, and Creative Lives, New York:
Pantheon, 1996.

Widom, C. S.,  The cycle of violence, Science, 1989, 244 (4901), 160166.
Zajonc, Robert B. and Gregory B. Markus,  Birth order and intellectual development, Psy-
chological Review, 1975, 82 (1), 7488.

28
Figures and Tables

Table 1: Descriptive Statistics

Denmark Florida
2+ Boy-Boy and 2+ Boy-Boy and
All All 2+ families All All 2+ families
Girl-Boy families Girl-Boy families
(1) (2) (3) (4) (5) (6)
Parental controls - Mother
Age mother – first-born 25.23 25.04 25.04 - 25.82 25.88
Education
Basic 32.52 31.86 31.69 20.30 10.45 10.15
High school or vocational 34.98 34.42 34.42 58.60 59.25 59.27
College 22.08 22.89 23.02 21.10 30.30 30.58
Employed 75.36 74.38 74.37 - - -
Parental controls - Father
Age father – first-born 28.26 28.14 28.11 - 28.55 28.61
Education
Basic 23.85 23.70 23.54 14.83 10.86 10.69
High school or vocational 44.57 43.85 43.61 59.59 60.90 60.85
College 20.22 21.02 21.26 25.59 28.23 28.47
Employed 83.99 84.19 84.12 - - -
Family controls
Income
Low 22.54 22.84 22.92 25.02 15.79 15.53
Low-Medium 27.61 28.79 28.84 25.06 22.75 22.81
Medium-High 26.25 26.27 26.17 25.19 28.72 28.67
High 23.60 22.10 22.08 24.73 32.74 32.99
Mother immigrant 11.94 12.78 12.87 24.03 20.58 20.37
Father immigrant 10.62 11.52 11.66 - - -
Mother African American - - - 21.89 11.24 10.89
Mother Hispanic - - - 23.94 22.94 22.74
Spacing (first-born to second-born) - 3.03 3.03 - 2.65 2.65
Completed family size 2.34 2.59 2.59 - - -
Outcomes
Convicted by age 16 3.26 3.27 4.29 - - -
Convicted by age 21 10.93 11.01 14.39 - - -
Sentenced to prison by age 16 0.78 0.83 1.24 - - -
Sentenced to prison by age 21 4.22 4.36 6.12 - - -
Incarcerated by age 16 0.11 0.12 0.18 - - -
Incarcerated by age 21 1.40 1.48 2.18 - - -
Suspended in school - - - - 4.51 5.61
Absence rate in school - - - - 3.83 3.84
Number of children 448,979 235,720 121,100 1,592,342 148,788 75,914

Note: Sample in column (1) is based on all singleton births between 1981 and 1990 in Denmark. Sample in column
(4) is based on all singleton births between 1994 and 2002 in the state of Florida. We are unable to provide maternal
age at rst birth for this sample since we do not know older and younger siblings for the full population of births. We
also do not observe education for fathers who do not appear on the birth certicate and the sample size for paternal
education is 1,277,106. Columns (2) and (5) are based on the sample of families with two or more children and the
same father. We can link siblings only in about 70 percent of counties of Florida and only if both births occurred in
the state of Florida between 1994 and 2002. We can link all siblings in Denmark. Columns (3) and (6) present the
subsample of interest with second-born boys. Employment and income in Denmark are measured in the year before
birth. All other controls are measured at the time of child's birth. Individual level income in Denmark and ZIP code
of birth median income based on 2000 Census for Florida. Spacing refers to distance in years between rst and second
birth. Binary variables are multiplied by 100 giving percent values.

29
 Table 2: Main results

(1) (2) (3) (4) (5) (6)

OLS with OLS with
OLS Fixed effects OLS Fixed effects
controls controls
Panel A. Denmark: Convictions
By age 16 By age 21
Second-born 0.647*** 1.370*** 1.616*** 2.034*** 3.866*** 3.618***
(0.167) (0.259) (0.307) (0.284) (0.427) (0.483)
Girl -3.156*** -3.116*** -2.740*** -12.114*** -12.017*** -11.923***
(0.137) (0.137) (0.218) (0.240) (0.236) (0.356)
Mean of Y for first-born boy 4.571 16.330
Number of children/families 121,100/60,550 121,100/60,550
Panel B. Denmark: Sentenced to prison
By age 16 By age 21
Second-born 0.444*** 0.732*** 0.917*** 0.944*** 2.221*** 2.393***
(0.090) (0.152) (0.174) (0.201) (0.310) (0.348)
Girl -0.916*** -0.898*** -0.655*** -6.360*** -6.312*** -6.242***
(0.061) (0.061) (0.121) (0.156) (0.155) (0.248)
Mean of Y for first-born boy 1.038 7.217
Number of children/families 121,100/60,550 121,100/60,550
Panel C. Denmark: Incarcerated
By age 16 By age 21
Second-born 0.065* 0.246*** 0.311*** 0.271** 0.980*** 1.086***
(0.034) (0.059) (0.068) (0.129) (0.199) (0.227)
Girl -0.127*** -0.123*** -0.080* -2.586*** -2.566*** -2.619***
(0.022) (0.021) (0.046) (0.094) (0.094) (0.155)
Mean of Y for first-born boy 0.138 2.710
Number of children/families 121,100/60,550 121,100/60,550
Panel D. Large FL County: In juvenile justice system
By age 16
Second-born 0.547** 0.514* 1.146**
(0.267) (0.266) (0.561)
Girl -1.511*** -1.522*** -1.229**
(0.239) (0.237) (0.509)
Mean of Y for first-born boy 2.582
Number of children/families 24,928/12,464
Panel E. Florida: Behavioral outcomes in school
Suspensions Absences
Second-born 3.664*** 4.272*** 1.726*** 0.351*** 0.230*** 0.008
(0.151) (0.210) (0.224) (0.030) (0.040) (0.030)
Girl -3.935*** -3.974*** -4.279*** 0.051 0.035 -0.034
(0.126) (0.120) (0.177) (0.034) (0.032) (0.024)
Mean of Y for first-born boy 5.942 3.725
Number of observations/children/families 349,184/76,036/38,018 347,736/75,914/37,957

Note: Standard errors clustered at mother level. All dependent variables, except for absence rate, are binary indicators
multiplied by 100. These are: probability of being convicted (panel A), probability of being sentenced to suspended
imprisonment (panel B), probability of being incarcerated (panel C), probability of being in juvenile detention center
(panel D) and probability of being suspended in school (columns (1) to (3) in panel E). OLS and FE regressions in
columns (1), (3), (4) and (6) include only child birth year dummies. Controls in columns (2) and (5) in Panels A, B
and C in addition to birth year dummies include mother's age at rst birth, mother's age, father's age, municipality
dummies, maternal and paternal education, maternal and paternal employment, family income, immigrant dummy,
number of children in family. Controls in column (2) in Panel D in addition to birth year dummies include rst
observed limited English prociency, subsidized school meal status, and average school meal status in student micro-
neighborhood. Controls in columns (2) and (5) in Panel E in addition to birth year dummies include mother age
at rst birth, mother age, father age, ZIP code of birth dummies, mother and father education dummies as well as
indicator variables for race, ethnicity, immigrant status and marital status at the time of birth. Panels A, B and C
are based on 1981 to 1990 birth cohorts; panel D is based on September 1979 to August 1989 births while panel D is
based on 1994 to 2002 birth cohorts. Sample is based on boy-boy and girl-boy sibling pairs from families with two or
more siblings. 30
 Table 3: Types of crime by age 21: Denmark birth cohorts 1981-1990

(1) (2) (3)

Violent Crime Property Crime Special Crime
Second-born 1.606*** 2.163*** 2.099***
(0.292) (0.416) (0.356)
Girl -3.925*** -7.406*** -5.877***
(0.204) (0.300) (0.257)
Mean of Y for first-born boy 4.526 10.885 6.551
Number of children/families 121,100/60,550
Note: Standard errors clustered at mother level. All estimates come from xed eects regressions controlling for
child birth year dummies. All crime outcomes are measured based on convictions for violent, property and special
oenses. Special crime includes for example drug oenses and illegal weapon possession. Sample is based on boy-boy
and girl-boy sibling pairs from families with two or more siblings.

Table 4: Short-run non-cognitive outcomes: Denmark birth cohorts 1999-2000

(1) (2) (3) (4)

Close to average Slightly raised High Very high
Panel A: Total difficulties score
Second-born -0.041* 0.021 0.006 0.014
(0.023) (0.018) (0.011) (0.013)
Girl 0.009 -0.033** 0.009 0.015
(0.019) (0.014) (0.010) (0.011)
Mean of Y for first-born boy 0.829 0.094 0.034 0.043
Panel B: Prosociality score
Second-born -0.024 0.020 0.002 0.003
(0.027) (0.021) (0.017) (0.014)
Girl 0.138*** -0.062*** -0.041*** -0.035***
(0.021) (0.017) (0.013) (0.010)
Mean of Y for first-born boy 0.695 0.155 0.090 0.059
Number of children 3,370
Note: Heteroskedasticity robust standard errors. All outcome variables are binary indicators. OLS regressions in
columns (1)-(4) include the following controls: child birth year dummies, mother age at rst birth, mother age, father
age, municipality dummies, maternal and paternal education, maternal and paternal employment, family income,
immigrant dummy, number of children in family. The questionnaire was answered by 6th grade students from birth
cohorts 1998-2000. Panel A is a combination of emotional problems score, conduct problems score, hyperactivity score
and peer eects problems score. We present the results for these scores separately in Table A1.

31
 Table 5: Mechanisms: neonatal and childhood health, parental investment and school quality

(1) (2) (3) (4) (5) (6) (7) (8)

Panel A. Denmark: Birth cohorts 1981-1990
Maternal Disorders of Behavioral and
Complications in Days of maternity Out of home School starting age
Ln(birth weight) employment one psychological emotional
pregnancy leave daycare at age 2 (1st grade)
year after birth development disorders
Second-born 0.037*** -0.053*** -8.574*** -0.005 0.039*** -0.235*** -0.000 -0.002
(0.002) (0.005) (0.692) (0.004) (0.006) (0.006) (0.003) (0.003)
Girl -0.040*** -0.002 -0.232 -0.006* -0.006 -0.107 -0.015*** -0.021***
(0.001) (0.004) (0.538) (0.003) (0.004) (0.005) (0.002) (0.002)
Mean of Y for first-born boy 8.141 0.264 150 0.767 0.667 6.760 0.018 0.030
Number of children 172,642 172,934 98,576 175,686 89,262 174,724 85,258 85,258
Number of families 86,467 86,467 49,288 87,843 44,631 87,362 42,629 42,629

32
Panel B. Florida: Birth cohorts 1994-2002
Attending A
Complications in Pre-kindergarten School quality Behavioral
Ln(birth weight) Held back a year quality school in Cognitive disability
pregnancy participation (0-4) in third grade disability
third grade
Second-born 0.026*** -0.027*** 0.066*** -0.006 -0.014** -0.019* -0.008* -0.000
(0.003) (0.007) (0.008) (0.005) (0.006) (0.010) (0.004) (0.004)
Girl -0.040*** 0.005 -0.030*** -0.041*** 0.001 0.006 -0.036*** -0.032***
(0.002) (0.005) (0.006) (0.004) (0.005) (0.008) (0.003) (0.003)
Mean of Y for first-born boy 8.114 0.220 0.546 0.067 0.761 3.648 0.068 0.069
Number of children 76,036 76,036 76,036 62,753 71,902 71,902 76,036 76,036
Number of families 38,018 38,018 38,018 37,237 35,951 35,951 38,018 38,018

Note: Standard errors clustered at mother level. All estimates come from xed eects regressions controlling for child birth year dummies. Sample is based on
boy-boy and girl-boy sibling pairs from families with two or more siblings.
 Table 6: Heterogeneity eects by spacing between rst- and second-born child

(1) (2) (3) (4)

Spaced up to 2.5 Spaced above 2.5 Spaced up to 2.5 Spaced above 2.5
years apart years apart years apart years apart
Panel A. Denmark: Convictions
By age 16 By age 21
Second-born 0.143 0.332 4.119*** 1.982**
(0.704) (0.487) (1.073) (0.796)
Girl -2.926*** -2.625*** -13.194*** -11.081***
(0.365) (0.270) (0.577) (0.450)
Mean of Y for first-born boy 5.213 4.137 18.112 15.127
Number of children 48,552 72,548 48,552 72,548
Number of families 24,276 36,274 24,276 36,274
Panel B. Denmark: Sentenced to prison
By age 16 By age 21
Second-born 0.884** 0.203 4.198*** 0.935*
(0.412) (0.263) (0.813) (0.546)
Girl -0.863*** -0.519*** -6.977*** -5.762***
(0.208) (0.146) (0.417) (0.307)
Mean of Y for first-born boy 1.407 0.789 8.202 6.552
Number of children 48,552 72,548 48,552 72,548
Number of families 24,276 36,274 24,276 36,274
Panel C. Denmark: Incarcerated
By age 16 By age 21
Second-born 0.435** 0.115* 1.896*** 0.108
(0.180) (0.070) (0.552) (0.335)
Girl -0.191** -0.006 -3.080*** -2.315***
(0.091) (0.047) (0.272) (0.185)
Mean of Y for first-born boy 0.230 0.075 3.203 2.377
Number of children 48,552 72,548 48,552 72,548
Number of families 24,276 36,274 24,276 36,274
Panel D. Large FL County: In juvenile justice system
By age 16
Second-born 0.449 0.911
(0.886) (1.516)
Girl -1.377** -0.925
(0.623) (0.886)
Mean of Y for first-born boy 2.540 2.674
Number of children 16,958 7,970
Number of families 8,479 3,985
Panel E. Florida: Behavioral outcomes in school
Suspensions (G1 to G8) Absences (G4 to G8)
Second-born 1.748*** 2.911*** 0.144** 0.416***
(0.426) (0.433) (0.071) (0.096)
Girl -4.770*** -3.620*** -0.144*** -0.088*
(0.249) (0.246) (0.042) (0.050)
Mean of Y for first-born boy 6.764 4.842 3.761 3.485
Number of observations 198,872 150,312 113,774 86,382
Number of children 40,064 35,972 35,076 31,360
Number of families 20,032 17,986 17,538 15,680

Note: Standard errors clustered at mother level. All estimates come from xed eects regressions controlling for child
birth year dummies. Sample is based on boy-boy and girl-boy sibling pairs from families with two or more siblings.

33
 Table 7: Heterogeneity eects by maternal education at birth

(1) (2) (3) (4)

< 12 years 12+ years < 12 years 12+ years
Panel A. Denmark: Convictions
By age 16 By age 21
Second-born 1.651*** 0.642* 4.468*** 2.301***
(0.629) (0.339) (0.936) (0.599)
Girl -3.511*** -1.629*** -16.208*** -8.580***
(0.451) (0.234) (0.698) (0.425)
Mean of Y for first-born boy 6.559 2.741 22.594 11.158
Number of children 39,386 68,234 39,386 68,234
Number of families 19,693 34,117 19,693 34,117
Panel B. Denmark: Sentenced to prison
By age 16 By age 21
Second-born 1.259*** 0.245 2.845*** 1.271***
(0.365) (0.167) (0.722) (0.375)
Girl -0.664** -0.424*** -8.757*** -3.610***
(0.261) (0.115) (0.521) (0.263)
Mean of Y for first-born boy 1.459 0.514 10.855 3.974
Number of children 39,386 68,234 39,386 68,234
Number of families 19,693 34,117 19,693 34,117
Panel C. Denmark: Incarcerated
By age 16 By age 21
Second-born 0.454*** 0.075 0.862* 0.515**
(0.127) (0.056) (0.478) (0.216)
Girl 0.035 -0.004 -4.125*** -1.106***
(0.088) (0.036) (0.336) (0.144)
Mean of Y for first-born boy 0.099 0.051 4.158 1.187
Number of children 39,386 68,234 39,386 68,234
Number of families 19,693 34,117 19,693 34,117
Panel D. Florida: Behavioral outcomes in school
Suspensions (G1 to G8) Absences (G4 to G8)
Second-born 2.883*** 1.421*** 0.084 0.158***
(0.826) (0.225) (0.161) (0.040)
Girl -7.478*** -3.836*** -0.154 -0.117***
(0.696) (0.177) (0.134) (0.032)
Mean of Y for first-born boy 12.439 5.096 5.349 5.558
Number of observations 40,980 308,204 23,186 176,970
Number of children 8,760 67,276 7,506 58,930
Number of families 4,380 33,638 3,753 29,465
Note: Standard errors clustered at mother level. All estimates come from xed eects regressions controlling for child
birth year dummies. Sample is based on boy-boy and girl-boy sibling pairs from families with two or more siblings.

34
 Figure 1: Birth order eects on criminal activity in early adulthood: Estimates for second-born
boys by age in Denmark

A. Convicted B. Sentenced to prison

60

120
100
40

Sentenced to prison
80
Conviction

60
20

40
20
0

16 17 18 19 20 21 22 23 24 25 16 17 18 19 20 21 22 23 24 25
Age Age

C. Incarcerated
300200
Incarcerated
100
0

16 17 18 19 20 21 22 23 24 25
Age

Note: All estimates come from xed eects regressions controlling for child birth year dummies and are scaled by means of
dependent variable for rst-born boy in a given group. Dependent variables are delinquency measures by given age. Standard
errors clustered at mother level with reported 95% condence intervals. Sample is based on boy-boy and girl-boy sibling
pairs from families with two or more siblings.

35
Figure 2: Birth order eects on delinquency in childhood: Estimates for second-born boys by grade
in Florida

A. Suspensions B. Absence rate

100

20
Second born effects (%) absence rate
Second born effects (%) suspensions
80

10
60

0
40

-10
20 0

-20
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
Longitudinal grade Longitudinal grade

Note: Sample is based on boy-boy and girl-boy sibling pairs from families with two or more siblings. Dependent variables
are delinquency measures in a given grade. For further details see notes in Figure 1.

Figure 3: Maternal labor market activity after birth: Estimates for second-born boys in Denmark

A. Any employment B. Full time employment

.4
.05

Employed >= 200 days in a year

.2
0
Any employment
-.05

0
-.1

-.2
-.15

0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
Years after birth Years after birth

Note: Sample is based on boy-boy and girl-boy sibling pairs from families with two or more siblings. Dependent variables
are maternal employment measures in a given year after child birth i.e. being employed 1st year after birth, 2nd year after birth
etc. For further details see notes in Figure 1.

36
Figure 4: Birth order eects on attended school quality: Estimates for second-born boys by grade
in Florida

A. Attended A rated school B. School quality (0 to 4 scale)

3
Second born effects (%) attending A quality school
6

Second born effects (%) school quality

2
4

1
2

0
0

-1
-4 -2

-2
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
Longitudinal grade Longitudinal grade

Note: Sample is based on boy-boy and girl-boy sibling pairs from families with two or more siblings. Dependent variables
are school quality measures in a given grade. For further details see notes in Figure 1.

37
Appendix Tables

Table A1: Short-run non-cognitive outcomes: Denmark birth cohorts 1999-2000

(1) (2) (3) (4)

Close to average Slightly raised High Very high
Panel A: Emotional problems score
Second-born -0.002 0.008 -0.013 0.006
(0.020) (0.014) (0.010) (0.011)
Girl -0.114*** 0.031** 0.011 0.071***
(0.019) (0.014) (0.011) (0.012)
Mean of Y for first-born boy 0.878 0.059 0.041 0.022
Panel B: Conduct problems score
Second-born -0.013 -0.002 -0.000 0.016
(0.021) (0.016) (0.012) (0.010)
Girl 0.051*** -0.016 -0.029*** -0.006
(0.017) (0.013) (0.010) (0.008)
Mean of Y for first-born boy 0.847 0.079 0.051 0.023
Panel C: Hyperactivity score
Second-born -0.049** -0.005 0.029** 0.024
(0.023) (0.017) (0.012) (0.015)
Girl 0.041** -0.016 -0.007 -0.019
(0.019) (0.013) (0.010) (0.011)
Mean of Y for first-born boy 0.820 0.080 0.040 0.061
Panel D: Peer problems score
Second-born -0.013 -0.024 0.023 0.013
(0.029) (0.022) (0.019) (0.018)
Girl 0.079*** -0.052*** -0.024* -0.003
(0.024) (0.018) (0.014) (0.015)
Mean of Y for first-born boy 0.638 0.170 0.097 0.095
Number of children 3,370
Note: Heteroskedasticity robust standard errors. All outcome variables are binary indicators. OLS regressions in
columns (1)-(4) include the following controls: child birth year dummies, mother's age at rst birth, mother's age,
father's age, municipality dummies, maternal and paternal education, maternal and paternal employment, family
income, immigrant dummy, number of children in family. The questionnaire was answered by 6th grade students from
birth cohorts 1998-2000. These are sub-scores for the results reported in panel A of Table 4.

38
 Table A2: Daycare by age of the child: Denmark birth cohorts 1993-2000

(1) (2) (3)

At age 1 At age 2 At age 3
Second-born -0.172*** 0.034*** 0.108***
(0.007) (0.006) (0.006)
Girl -0.002 -0.007 -0.011***
(0.005) (0.004) (0.004)
Mean of Y for first-born boy 0.160 0.667 0.764
Number of children/families 87,442/43,721
Note: Standard errors clustered at mother level. All estimates come from xed eects regressions controlling for child
birth year dummies. Sample is based on boy-boy and girl-boy sibling pairs from families with two or more siblings.

Table A3: Short-run cognitive outcomes

(1) (2) (3) (4)

Denmark: Birth cohorts 1986-1997 Florida: Birth cohorts 1994-2002
Math Reading Math Reading
Second-born -0.109*** -0.137*** 0.013 -0.076***
(0.010) (0.012) (0.011) (0.011)
Girl -0.084*** 0.347*** -0.039*** 0.081***
(0.007) (0.009) (0.009) (0.009)
Mean of Y for first-born boy 0.278 -0.044 0.459 0.386
Number of observations 144,906 144,906 307,402 310,650
Number of children 144,906 144,906 72,888 72,998
Number of families 72,453 72,453 36,444 36,499
Note: Standard errors clustered at mother level. All estimates come from xed eects regressions controlling for child
birth year dummies. Sample is based on boy-boy and girl-boy sibling pairs from families with two or more siblings.

39
 Table A4: Main results by gender composition: Boy-boy vs. girl-boy pairs

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Denmark Florida
Convictions Sentenced to prison Incarcerated Juvenile
justice system Suspensions Absence rate
By age 16 By age 21 By age 16 By age 21 By age 16 By age 21
by age 16
Panel A: Boy-Boy families
Second-born 1.607*** 2.748*** 0.555** 1.491*** 0.201** 0.543* 1.326* 1.193*** 0.124**
(0.419) (0.648) (0.243) (0.480) (0.094) (0.317) (0.768) (0.295) (0.052)
Mean of Y for first-born boy 4.566 16.325 1.038 7.216 0.138 2.711 2.567 5.942 3.643
E[Y|Second born = boy, BB family] 6.173 19.073 1.593 8.707 0.339 3.254 3.893 7.135 3.767

40
Number of children 62,420 13,050 39,476 34,528
Number of families 31,210 6,525 19,738 17,264

Panel B: Girl-Boy families

Second-born 4.362*** 16.414*** 1.948*** 9.552*** 0.505*** 4.257*** 2.186*** 6.566*** 0.297***
(0.374) (0.612) (0.204) (0.421) (0.080) (0.266) (0.655) (0.298) (0.055)
Mean of Y for first-born girl 1.418 4.224 0.119 0.859 0.010 0.126 1.091 2.010 3.589
E[Y|Second born = boy, BG family] 5.780 20.638 2.067 10.411 0.515 4.383 3.277 8.576 3.886

Number of children 58,680 11,916 36,560 31,908

Number of families 29,340 5,958 18,280 15,954
Note: Standard errors clustered at mother level. All estimates come from xed eects regressions controlling for child birth year dummies. Birth cohorts for
Denmark: 1981-1990; birth cohorts unnamed county in Florida (column (7)): September 1979 to August 1989; birth cohorts for the state of Florida: 1994-2002.
 Table A5: Heterogeneity eects by maternal employment: Denmark birth cohorts 1981-1990

(1) (2) (3) (4)

Non-employed Employed Non-employed Employed
Panel A. Convictions
By age 16 By age 21
Second-born 2.849** 0.716** 4.018** 3.023***
(1.158) (0.365) (1.690) (0.603)
Girl -3.018*** -2.161*** -14.566*** -11.088***
(0.826) (0.248) (1.259) (0.424)
Mean of Y for first-born boy 5.998 3.805 21.923 14.451
Panel B. Sentenced to prison
By age 16 By age 21
Second-born 1.910*** 0.201 3.522*** 1.574***
(0.630) (0.203) (1.272) (0.414)
Girl -0.457 -0.589*** -7.330*** -5.211***
(0.469) (0.135) (0.930) (0.284)
Mean of Y for first-born boy 1.413 0.847 10.514 5.999
Panel C. Incarcerated
By age 16 By age 21
Second-born 0.646** 0.097* 1.907** 0.544**
(0.278) (0.054) (0.860) (0.253)
Girl 0.088 -0.038 -2.878*** -1.960***
(0.172) (0.041) (0.610) (0.166)
Mean of Y for first-born boy 0.138 0.048 3.895 2.040
Number of children 11,426 80,286 11,426 80,286
Number of families 5,713 40,143 5,713 40,143
Note: Maternal employment is dened as a mother working full or part time one year after birth. Standard errors
clustered at mother level. All estimates come from xed eects regressions controlling for child birth year dummies.
Sample is based on boy-boy and girl-boy sibling pairs from families with two or more siblings.

41
 Table A6: Heterogeneity eects by immigration origin of the parents

(1) (2) (3) (4)

Native-born Immigrant Native-born Immigrant
Panel A. Denmark: Convictions
By age 16 By age 21
Second-born 0.895*** 3.629*** 2.705*** 6.375***
(0.321) (0.856) (0.524) (1.214)
Girl -2.101*** -6.277*** -11.072*** -16.572***
(0.222) (0.725) (0.376) (1.029)
Mean of Y for first-born boy 3.860 8.046 14.847 22.100
Number of children 102,898 18,202 102,898 18,202
Number of families 51,449 9,101 51,449 9,101
Panel B. Denmark: Sentenced to prison
By age 16 By age 21
Second-born 0.404** 2.475*** 1.515*** 4.469***
(0.174) (0.529) (0.363) (0.978)
Girl -0.391*** -2.101*** -5.136*** -12.332***
(0.118) (0.446) (0.254) (0.807)
Mean of Y for first-born boy 0.795 2.807 6.236 11.204
Number of children 102,898 18,202 102,898 18,202
Number of families 51,449 9,101 51,449 9,101
Panel C. Denmark: Incarcerated
By age 16 By age 21
Second-born 0.129*** 0.889*** 0.527** 2.371***
(0.049) (0.281) (0.225) (0.707)
Girl -0.003 -0.500** -1.958*** -6.284***
(0.035) (0.233) (0.151) (0.577)
Mean of Y for first-born boy 0.045 0.764 2.112 5.251
Number of children 102,898 18,202 102,898 18,202
Number of families 51,449 9,101 51,449 9,101
Panel D. Florida: Behavioral outcomes in school
Suspensions (G1 to G8) Absences (G4 to G8)
Second-born 1.737*** 1.655*** 0.117** 0.264***
(0.265) (0.396) (0.048) (0.072)
Girl -4.573*** -3.137*** -0.121*** -0.122**
(0.204) (0.341) (0.038) (0.062)
Mean of Y for first-born boy 6.294 4.542 3.832 2.901
Number of observations 278,674 70,510 159,454 40,702
Number of children 60,546 15,490 52,840 13,596
Number of families 30,273 7,745 26,420 6,798
Note: Standard errors clustered at mother level. All estimates come from xed eects regressions controlling for child
birth year dummies. Sample is based on boy-boy and girl-boy sibling pairs from families with two or more siblings.
We dene immigrants in Florida based on the fact that mother was born outside of the US while in Denmark if either
of the parents were born outside of Denmark.

42
 Table A7: Heterogeneity eects by race and ethnicity: Florida birth cohorts 1994-2002

(1) (2) (3) (4)

Suspensions (G1-G8) Absences (G4-G8)
Panel A. Race
African Non-African African Non-African
American American American American
Second-born 2.215** 1.584*** -0.014 0.172***
(0.871) (0.224) (0.107) (0.043)
Girl -8.928*** -3.731*** -0.272*** -0.103***
(0.786) (0.174) (0.094) (0.035)
Mean of Y for first-born boy 12.384 5.183 3.223 3.691
Number of observations 36,696 312,488 20,826 179,330
Number of children 8,284 67,752 7,138 59,298
Number of families 4,142 33,876 3,569 29,649
Panel B. Ethnicity
Hispanic Non-Hispanic Hispanic Non-Hispanic
Second-born 1.501*** 1.809*** 0.247*** 0.117**
(0.399) (0.268) (0.073) (0.048)
Girl -3.870*** -4.389*** -0.209*** -0.094**
(0.350) (0.204) (0.061) (0.038)
Mean of Y for first-born boy 5.264 6.142 3.368 3.725
Number of observations 78,578 270,606 44,998 155,158
Number of children 17,296 58,740 15,116 51,320
Number of families 8,648 29,370 7,558 25,660
Note: Standard errors clustered at mother level. All estimates come from xed eects regressions controlling for child
birth year dummies. Sample is based on boy-boy and girl-boy sibling pairs from families with two or more siblings.
We dene race and ethnicity based on maternal information at birth.

43
 Table A8: Main results: Sample of families with dierent fathers

(1) (2) (3) (4) (5) (6) (7) (8)

Denmark Florida
Convictions Sentenced to prison Incarcerated
Suspensions Absence rate
By age 16 By age 21 By age 16 By age 21 By age 16 By age 21
Second-born 3.612** 6.261*** 0.179 0.954 -0.090 1.092 2.952*** 0.131
(1.563) (2.338) (0.904) (1.831) (0.397) (1.252) (0.830) (0.111)

44
Girl -4.278*** -21.560*** -1.903*** -15.979*** -0.716*** -7.661*** -6.044*** -0.015
(1.076) (1.620) (0.627) (1.246) (0.255) (0.845) (0.660) (0.097)
Mean of Y for first-born boy 9.483 32.367 2.978 17.712 0.705 7.641 11.385 4.510
Number of children 10,106 10,466 10,426
Number of families 5,053 5,233 5,213
Note: Sample is based on all families where father is known and dierent irrespective of the marital status. Standard errors clustered at mother level. All estimates
come from xed eects regressions controlling for child birth year dummies. Sample is based on boy-boy and girl-boy sibling pairs from families with two or more
siblings.
 Table A9: Main results - robustness of xed eects models

(1) (2) (3) (4) (5) (6) (7) (8)

Denmark Florida
Convictions Sentenced to prison Incarcerated
Suspensions Absence rate
By age 16 By age 21 By age 16 By age 21 By age 16 By age 21
Panel A: Additional maternal/household controls
Second-born 1.626*** 3.463*** 0.940*** 2.311*** 0.300*** 1.025*** 1.584*** 0.007
(0.309) (0.487) (0.175) (0.351) (0.067) (0.229) (0.227) (0.031)
Girl -2.741*** -11.931*** -0.646*** -6.246*** -0.084* -2.623*** -4.278*** -0.034
(0.218) (0.356) (0.121) (0.249) (0.046) (0.155) (0.176) (0.024)
Panel B: Controlling for birth outcomes
Second-born 1.613*** 3.549*** 0.910*** 2.316*** 0.307*** 1.059*** 1.701*** 0.010
(0.308) (0.485) (0.175) (0.350) (0.068) (0.228) (0.226) (0.030)
Girl -2.721*** -11.822*** -0.643*** -6.144*** -0.073 -2.580*** -4.242*** -0.041
(0.220) (0.358) (0.122) (0.251) (0.047) (0.158) (0.179) (0.025)

45
Mean of Y for first-born boy 4.571 16.330 1.038 7.217 0.138 2.710 5.942 3.725
Number of children 121,100 76,036 75,914
Number of families 60,550 38,018 37,957
Panel C: Dropping families of size two
Second-born 1.729*** 4.694*** 1.035*** 2.928*** 0.270** 1.382*** 2.175*** -0.066
(0.517) (0.793) (0.314) (0.598) (0.131) (0.407) (0.626) (0.078)
Girl -3.199*** -14.050*** -0.754*** -8.225*** -0.250*** -3.829*** -5.948*** 0.109*
(0.373) (0.586) (0.216) (0.430) (0.092) (0.282) (0.507) (0.064)
Mean of Y for first-born boy 5.483 19.175 1.396 9.182 0.249 3.706 8.010 3.952
Number of children 49,688 10,480 10,478
Number of families 24,844 5,240 5,239
Note: Standard errors clustered at mother level. All estimates come from xed eects regressions controlling for child birth year dummies. Additional controls
in panel A include mother's age at birth, education, marital status and ZIP code of residence in the case of Florida. In Denmark controls include municipality
dummies, mother's age at birth, maternal and paternal education, maternal and paternal employment as well as family income. Additional controls in panel
B include log birth weight, maternal pregnancy complications and gestational length. Panel C drops families that have exactly two children. Birth cohorts for
Denmark: 1981-1990; birth cohorts for the state of Florida: 1994-2002. Sample is based on boy-boy and girl-boy sibling pairs from families with two or more
siblings.
 Table A10: Main results by gender composition: Girl-girl vs. boy-girl pairs

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Denmark Florida
Convictions Sentenced to prison Incarcerated Juvenile
justice system Suspensions Absence rate
By age 16 By age 21 By age 16 By age 21 By age 16 By age 21
by age 16
Panel A: Girl-Girl families
Second-born 0.030 0.593 0.004 0.715*** -0.001 0.086 -0.237 0.818*** 0.101**
(0.252) (0.413) (0.101) (0.225) (0.026) (0.091) (0.554) (0.187) (0.050)
Mean of Y for first-born girl 1.428 4.290 0.136 0.854 0.004 0.079 1.214 1.921 3.585
E[Y|Second born = girl, GG family] 1.458 4.883 0.140 1.569 0.003 0.165 0.977 2.739 3.686

46
Number of children 55,756 11,696 35,166 31,118
Number of families 27,878 5,848 17,583 15,559

Panel B: Boy-Girl families

Second-born -3.179*** -12.820*** -1.098*** -7.243*** -0.184*** -3.231*** -1.525** -4.108*** -0.153***
(0.342) (0.589) (0.160) (0.395) (0.062) (0.247) (0.661) (0.275) (0.051)
Mean of Y for first-born boy 4.073 15.680 0.935 6.555 0.139 2.438 2.485 6.276 3.738
E[Y|Second born = girl, BG family] 0.894 2.860 -0.163 -0.688 -0.045 -0.793 0.960 2.168 3.585
Number of children 58,864 11,992 37,828 33,474
Number of families 29,432 5,996 18,914 16,737

Note: Standard errors clustered at mother level. All estimates come from xed eects regressions controlling for child birth year dummies. Birth cohorts for
Denmark: 1981-1990; birth cohorts unnamed county in Florida (column (7)): September 1979 to August 1989; birth cohorts for the state of Florida: 1994-2002.
 Table A11: Mechanisms: neonatal and childhood health, parental investment and school quality. Second-born girl eects

(1) (2) (3) (4) (5) (6) (7) (8)

Panel A. Denmark: Birth cohorts 1981-1990
Maternal Disorders of Behavioral and
Complications in Days of maternity Out of home School starting age
Ln(birth weight) employment one psychological emotional
pregnancy leave daycare at age 2 (1st grade)
year after birth development disorders
Second-born 0.036*** -0.051*** -8.341*** -0.002 0.042*** -0.257*** -0.004* -0.001
(0.002) (0.005) (0.720) (0.004) (0.007) (0.007) (0.002) (0.003)
Girl -0.036*** 0.001 0.395 -0.004 0.007 -0.095*** -0.017*** -0.021***
(0.001) (0.004) (0.551) (0.003) (0.005) (0.005) (0.001) (0.002)
Mean of Y for first-born girl 8.107 0.259 149 0.762 0.653 6.665 0.004 0.010
Number of children 161,814 161,814 92,992 164,476 84,072 165,012 80,558 80,558
Number of families 80,907 80,907 46,496 82,238 42,036 82,506 40,279 40,279

47
Panel B. Florida: Birth cohorts 1994-2002
Attending A
Complications in Pre-kindergarten School quality Behavioral
Ln(birth weight) Held back a year quality school in Cognitive disability
pregnancy participation (0-4) in third grade disability
third grade
Second-born 0.027*** -0.034*** 0.057*** -0.007* 0.004 0.006 -0.009*** 0.007*
(0.003) (0.007) (0.008) (0.004) (0.007) (0.011) (0.003) (0.004)
Girl -0.035*** 0.004 -0.031*** -0.033*** 0.001 -0.001 -0.037*** -0.036***
(0.002) (0.006) (0.006) (0.003) (0.005) (0.008) (0.003) (0.003)
Mean of Y for first-born girl 8.080 0.222 0.516 0.032 0.750 3.631 0.031 0.036
Number of children 72,994 72,993 72,994 59,986 69,230 69,230 72,994 72,994
Number of families 36,497 36,497 36,497 35,698 34,615 34,615 36,497 36,497
Note: Standard errors clustered at mother level. All estimates come from xed eects regressions controlling for child birth year dummies. Sample is based on
girl-girl and boy-girl sibling pairs from families with two or more siblings.
Appendix Figures

Figure A1: Birth order eects on cognitive development in childhood: Estimates for second-born
boys by grade in Florida

A. Mathematics B. Reading

.02
.06

0
Second born effects (SD) mathematics
.04

Second born effects (SD) reading

-.1 -.08 -.06 -.04 -.02
-.04 -.02 0 .02

-.12
-.06

-.14
3 4 5 6 7 8 3 4 5 6 7 8
Longitudinal grade Longitudinal grade

Note: Sample is based on boy-boy and girl-boy sibling pairs from families with two or more siblings. Dependent variables
are mathematics or reading test score in a given grade. For further details see notes in Figure 1.

48