Young in Class: Implications for Inattentive/Hyperactive Behaviour of Revision

advertisement
 Young in Class: Implications for Inattentive/Hyperactive Behaviour of
Canadian Boys and Girls
Revision
March, 2013
Kelly Chen, Nicole Fortin, Philip Oreopoulos and Shelley Phipps
This research is being conducted as part of the Canadian Institute for Advanced Research
(CIFAR) Programme on Social Interactions, Identity and Well-Being. The NLSCY data were
access through the Atlantic Research Data Centre; we thank Heather Hobson for vetting our
output.
1 Abstract
Are Canadian children who are young relative to their class-mates more likely to exhibit
inattentive/hyperactive behaviours? If so, are there gender differences in the extent to which this
is true? Do the effects on inattentive/hyperactive behaviours of starting school relatively young
persist into adolescence?; and, if so, can this help to explain gender differences in educational
outcomes, behaviours and aspirations of Canadian youth? Using data from the Statistics Canada
National Longitudinal Survey of Children and Youth, we apply two research strategies to address
these questions. A ‘difference in difference’ design compares children who are the same age in
months, but live in provinces and/or time periods with different school start dates. A ‘regression
discontinuity’ design compares scores for children living in the same province who were born
just born and just after the relevant school entry cut-off. Both approaches find more
inattentive/hyperactive behaviour for children who are young in class, especially if the child was
more inattentive/hyperactive prior to school entry. When we control for child
inattentive/hyperactive behaviour at ages 2/3, we find that being young in class exacerbates an
underlying tendency toward inattentive/hyperactive behaviours and thus pushes more boys than
girls into clinical levels. These effects persist into early adolescence and may contribute to
gender differences in other early adolescent school-related behaviours, aspirations and outcomes.
2 Recent research emphasizes the idea that human capital acquisition is a cumulative
process (e.g., Cunha and Heckman, 2009; Currie, 2011; Conti and Heckman, 2012; Heckman,
Stixrud and Urzua, 2006). Both cognitive and non-cognitive capacities developed early in the
educational process can enhance the productivity of later education; moreover, higher capacity in
one dimension is argued to complement the capacity to grow in another (e.g., an attentive child
can learn to read more easily). Thus, experiences at the very start of their school lives can have
long-run repercussions for children’s eventual educational success.
Attention of both scholars and popular media (e.g., Fortin et al., 2012; Gurian, 2009; Sax,
2007) has also recently been drawn to the fact that girls’ academic achievement has now
surpassed boys.’ Young women now comprise 60 percent of the undergraduate population in
most Canadian universities. In 2008, 36.5 percent of young Canadian women aged 25 to 29 had
university degrees compared to 24.1 percent of young men (Drolet, 2011).
In this paper, we explore the possibility that part of the explanation for boys’ lagging
academic motivation and achievement may originate in the early years at school. For example, if
boys come to dislike school a little less at the very beginning, this can snowball over the years
into significantly different educational attitudes/behaviours/outcomes. In particular, we focus
on inattentive/hyperactive behaviour as one important aspect of non-cognitive development that
has become more of a problem in recent decades, matters for educational attainment and has big
gender differences early in life. 1
1
U.S. data (e.g., Akinbami, et al., 2011; Child Trends, 2012; Garfield, 2012) show increased prevalence of Attention
Deficit Hyperactivity Disorder (ADHD), a clinical level of the kind of behavior we study here. AttentionDeficit/Hyperactivity Disorder (ADHD) is among the most commonly diagnosed behavioural disorders for children
in many countries (Elder, 2010; Faraone, et al., 2003; Skounti et al., 2007). ADHD is a developmental,
neurobiological condition defined by the presence of severe and pervasive symptoms of inattention, hyperactivity
and impulsivity which must be exhibited over a period of at least 6 months, before the age of 7 and in at least two
contexts such as home and school (Daley and Birchwood, 2010; Loe and Feldman, 2007). Secular trends are hard to
identify given changes in diagnostic practices, but increases in prevalence are apparent over the past 40 years (Perrin
3 A significant body of research indicates that children who are less attentive or more
hyperactive experience problems with human capital acquisition (see, for example, Daley and
Birchwood; 2010 or Loe and Feldman, 2007 for reviews of the medical/psychological literature).
For example, they score lower on math/reading tests, are more likely to have behavioural
problems at school and/or to repeat a grade (Loe and Feldman, 2007). Most of this research has
focussed on young school-aged children, but, negative implications of hyperactivity on academic
performance have also been found for adolescents (Birchwood, 2010) and even college students
(e.g., Frazier, et al., 2007).
However, not all studies have adequately controlled for the possibility that, for example,
the home environments of children with and without ADHD may differ (e.g., in terms of parental
income, health, etc).2 Thus, important recent economic contributions to the literature include
Currie and Stabile (2006) and Fletcher and Wolfe (2008) who confirm the negative impact of
ADHD on academic achievement in models exploiting sibling differences in samples
representative of the population. From our perspective, an important finding is that that
academic problems are present even for children with only some symptoms of
inattention/hyperactivity (i.e., they are a bit more wiggly/boisterous/distractible than other
children), even if ADHD is not diagnosed or even if hyperactivity is well below clinical levels
(Currie and Stabile, 2006).
Although it was once thought that ADHD symptoms disappeared during adolescence, a
growing body of research indicates that hyperactive/inattentive behaviours continue into
adulthood (Wilens, Biederman and Spencer, 2002). For example, there are follow-up studies of
et al., 2007). Current estimates suggest worldwide ADHD prevalence ranges between 4 and 10 percent (Faraone, et
al., 2003; Skounti et al., 2007; Spencer et al., 2007).
2
This can be a particular problem for studies of clinical samples of children being treated for ADHD (see
Bauermeister et al, 2007, Table 1).
4 clinical populations that demonstrate persistence of symptoms over time (Biederman et al., 1998;
McGee et al., 1991). And, longitudinal research also finds that early childhood
inattention/hyperactivity has negative implications for academic outcomes both in adolescence
(Fletcher and Wolfe, 2007; Currie, et al., 2010) and even in adulthood (e.g., Daley and
Birchwood, 2009; Fletcher, 2013; Frazier et al, 2007).
There is surprisingly little ADHD research discussing gender differences. Indeed, since
boys are much more likely to be treated for ADHD, clinical studies have a particularly male
focus. However, population estimates indicate that many more girls exhibit symptoms of ADHD
than are diagnosed (Gerson and Gerson, 2002); and, ‘ADHD females share with their male
counterparts prototypical features of the disorder (e.g., inattention, impulsivity, and
hyperactivity), [and] high rates of school failure’ (Wilens, et al., 2002). This is of particular
interest for our study with its focus on gender and on inattentive/hyperactive behaviours that fall
short of clinical ADHD.
Another branch of research upon which we build presents evidence that children who are
young within grade at school are more likely to be diagnosed with ADHD (e.g., Elder, 2010;
Evans, 2010; Morrow, et al, 2012). Using U.S. data., Elder, for example, finds that 8.4 percent
of children born in the month prior to the state cut-off for kindergarten eligibility are diagnosed
with ADHD compared to 5.1 percent for those born in the month after. Morrow et al., 2012
report similar findings for B.C. Although these authors focus on the implications of being young
in class for diagnosis of ADHD, we argue that it is also plausible that being young in class
actually increases inattentive/hyperactive behaviour (though not necessarily to clinical levels).
Since young in class children can be almost one full year younger than some of their peers,
5 expectations for paying attention, sitting still, etc might be hard to achieve, leading them to ‘tune
out’ or ‘burst out’ with more boisterous behaviour both at school and at home.
Although the effects of hyperactivity are known to persist over the long-term, there is
some dispute in the literature about how long-lasting we might expect effects of being relatively
young in class to be. Bedard and Dhuey (2008) present evidence for a variety of countries (and
the Canadian province of BC) that relative school start age effects persist into the adult years;
Smith (2007), using the same BC data set, also finds evidence of effects persisting into the high
school years. On the other hand, Bertrand and Pan (2011) argue, using U.S. data, that effects
dissipate; Dobkin and Ferreira (2010), again in the U.S. context, argue that there are no longterm implications for adult labour market outcomes since younger children, on the one hand,
have poorer academic outcomes during their school years, but, on the other hand, are more likely
to pursue further education.
Building on these different strands of literature, we pose three basic research questions::
1) Are Canadian children in kindergarten through grade 4 who are young relative to their classmates more likely to exhibit symptoms of inattentive/hyperactive behaviour at home as reported
by their parents,3 and if so, are there gender differences in the extent to which this is true? 2)
Longitudinally, do children who exhibit higher levels of inattention/hyperactive behaviour at age
2/3, before they enter public school have a particularly difficult time adjusting if they are young
in class and is this particularly an issue for little boys?; 3) Do effects of being
inattentive/hyperactive early in a child’s schooling career, perhaps exacerbated by being young
at school, have negative effects on academic outcomes in early adolescence, and if so, do such
3
Note that this is not the same thing as asking if the child has ever been diagnosed with ADHD. We are also
interested in inattentive/hyperactive behaviours well below clinical thresholds.
6 differences help to explain lagging school outcomes and disinterested behaviours/attitudes of
adolescent boys compared to girls (Fortin, et al., 2012). 4
Using microdata from the National Longitudinal Survey of Children and Youth (1994
through 2008), we use, first, a ‘difference-in-difference’ strategy to compare children of exactly
the same age (in months) living in different provinces and/or time periods with different school
start cut-offs so that some of the children are ‘young in class’ while others of the same age are
‘old in class.’ Second, we use a ‘regression discontinuity’ (RD) design (see Lee and Lemieux,
2010 for a ‘user’s guide’ to the RD approach) in which we compare hyperactivity scores for
children living in the same province who were born just before and just after the relevant school
start cut-off date. This is possible since the NLSCY provides exact day of birth for each child.
Third, we take advantage of the longitudinal structure of the NLSCY to control for
inattention/hyperactive behaviour observed at ages 2/3 and ask if being young in class
exacerbates a pre-existing tendency to be inattentive/hyperactive? Finally, we further exploit the
longitudinal nature of the NLSCY to test whether hyperactivity observed in the early years at
school (e.g., at ages six/seven) helps to explain differences between boys and girls at age 13/14
in terms of both parent reports of school achievement and young adolescent self reports of
educational attitudes/behaviours and aspirations.
1. Kindergarten/Elementary School Legislation in Ten Provinces
Provincial authority over education policy in Canada provides a significant level of variation
across time and place in rules about when children start school. Kindergarten (grade primary in
Nova Scotia) is available for five-year olds in all provinces, but what we principally exploit for
4
Fortin et al, 2012, demonstrate that while there has been growth in the high achievement of girls in the U.S. (i.e.,
probability of getting A’s), there has been growth in the likelihood of boys getting C’s.
7 our analyses is that the date by which the child must have turned five varies from September 1 to
March 1. Table 1 shows the nine different school entry cut-off dates used during our study
period.
Since a typical school year runs from September to the end of June in all provinces, children
are admitted to school at one time each year. Whether they qualify for entry depends on their
exact date of birth. Because of the single entry age cut-off, some children will be admitted to
school a full year earlier than others. For example, if the cut-off date is December 31 in a
calendar year when a child turns five, it means that children who are born on or before December
31 will start school when they are 4 years and 8 months and become the youngest student in the
class. On the other hand, children who are born on January 1 will have to wait till next year to
enter kindergarten, which means they actually start when they are 5 years and 8 months, and
become the oldest student in the class.
With the exceptions of Alberta and Saskatchewan, school entry age eligibility in Canada is
determined by the provincial Ministry of Education and is specified in provincial statutes that are
usually contained in provincial Education Acts. Within the provinces of Alberta and
Saskatchewan, school boards may set their own age requirements for entering school (see notes
to Table 1). For the purpose of this study, we utilize information on school entry age cut-off in
eight provinces and school boards/districts of Alberta and six school boards in Saskatchewan that
can be identified by the Census Metropolitan Area (CMA) code available in the NLSCY. Data
are directly compiled from provincial Education Acts and Department of Education or school
board websites. We also use additional sources to help verify the compilation, including a survey
8 on school start age across Canada by the PEI Department of Education,5 and publications by
Statistics Canada that summarize school eligibility legislation.
As shown in Table 1, the most common cut-off date for school eligibility across provinces is
December 31. During our study period, six provinces (i.e. British Columbia, Manitoba (19872000), Ontario (2001-2004), New Brunswick, Newfoundland and Labrador, six school districts
in Saskatchewan and 4 school districts in Alberta) used this cut-off date to determine eligibility
for kindergarten. The second most common school-entry cut-off date is September 30, used in
Quebec and Nova Scotia. Otherwise, start date varies widely: for example, Calgary uses
September 1 while Edmonton uses March 1. Also, PEI and Manitoba have both made several
changes in cut-off date during our study period.
Of note is the fact that kindergarten is only mandatory in Quebec.6 Compulsory schooling
outside Quebec begins in grade one (see Oreopoulos, 2005 and 2006). Nevertheless, as we will
show, in all provinces nearly all families comply with the norms established through legislation
and begin public school with kindergarten at the ‘appropriate’ age.
2. Data
Statistics Canada’s National Longitudinal Survey of Children and Youth (NSLCY) is a
nationally representative dataset of Canadian children that tracks their development and wellbeing from birth to early adulthood, with data collection occurring at two year intervals. The first
survey round took place in 1994/95 with a nationally representative sample of 22,831 children
5
2003. http://www.ed.gov.nl.ca/edu/k12/kindergarten.html. Ontario and Quebec offer two grades of kindergarten: junior kindergarten and senior kindergarten. Junior
kindergarten, which is attended by four-year-olds, is optional in both provinces, but senior kindergarten is
mandatory for five-year-olds in Quebec. Other provinces have only one year of kindergarten (called ‘primary’ in
Nova Scotia). Kindergarten usually runs on a half-day or every-other-day schedule, while starting in 2007 (?) and
2010 (?), Ontario and British Colombia introduced full-day Monday to Friday kindergarten. Starting in 2010, after
our study period, kindergarten became compulsory in PEI. For more detailed description of kindergarten entry age
in Canada, see Lefebvre, Merrigan and Verstraete (2009). 6
9 aged 0-11. In addition to following the original longitudinal children, a new cohort aged 0-1 was
added at each new cycle allowing the construction of a sizable repeated cross-sectional dataset.
Most information for children under the age of 10 is reported by a parent, specifically the parent
selected as the person ‘most knowledgeable about the child (or, pmk – most typically the mother)
during a personal interview in the home.
In the first section of the paper, we select children between 4 and 9 years of age7 and who
are attending public school or publically funded Catholic schools. Since junior kindergarten is
only available in some provinces, our basic sample excludes children currently attending junior
kindergarten (though we later test the sensitivity of our results to having attended junior
kindergarten).8 To maximize our sample size, observations from cycles 1 through 8 (1994-2008)
are pooled.
A nice feature of the NLSCY compared to data sets used in some other studies of the
implications of being young in class is that we are able to match each child in the NSLCY to the
province-mandated elementary school eligibility cut-off when he/she was 4/5 years old.9 That is,
we use the longitudinal nature of the data to select children with parent reports of province of
residence at kindergarten start age (i.e. 4/5 years old) who have been living in the same province
ever since. With these restrictions, we have a sample of 34,500 children.10
7
For the purpose of sample selection, we use the NLSCY ‘effective age’ since this determines the set of questions
that will be asked about the child. Effective age is calculated as cycle year minus year of birth.
8
We also exclude a small number of children who did not live in either one- or two-parent families (e.g., those in
foster care or institutions). For Alberta and Saskatchewan, children living outside of the CMA’s for which we know
school start age rules are also excluded. 9
For example, Evans et al., (2010) do not have this information for any of the 3 data sets they employ.
10
Note that a child may appear more than once in the sample (e.g., at 6 and 8) since we are pooling observations
across cycles. We have also estimated all models using a sample in which each child is randomly selected to appear
just once with no substantive differences in our conclusions.
10 The measure of inattention/hyperactivity we use is based on parent11 reports. In all survey
years, parents of young children were asked to assess how often, in the home setting, their
child:12
□ “Can’t sit still or is restless?”
□ “Is easily distracted, has trouble sticking to any activity?”
□ “Can’t concentrate, can’t pay attention for long?”
□ “Is impulsive, acts without thinking?”
□ “Has difficulty waiting for his turn in games or groups?”
□ “Can’t settle to anything for more than a few minutes?”
□ “Is inattentive?” 13
For each behavior, the parent can choose: ‘never or not true’ (=0); ‘sometimes or somewhat true’
(=1); or, ‘often or very true’ (=2). Responses are summed to construct a scale ranging in value
from 0 to 14, with a high score indicating the highest level of inattentive/hyperactive behaviour.
The mean hyperactivity score for our sample is 3.9, though boys have higher scores than girls
(4.6 compared to 3.6). To put these scores in perspective, children in the NLSCY reported by
their parents to be taking the drug Ritalin, commonly prescribed for children diagnosed with
ADHD, have a mean score of 9.4. Distributions of the hyperactivity scores for 4 to 9 year old
boys and girls are presented in Figure 1.
11
Although teacher reports are also available for three cycles, teacher response rates are extremely low, so that we
do not regard them as a reliable source of information. Also, previous research (Elder, 2010) emphasizes that
teacher perceptions may be particularly likely to be systematically correlated with a child’s age relative age to
classmates. 12
Specifically, pmk’s are asked: “How often would you say [child’s name] …” 13
Earlier cycles also contained an additional question (‘fidgets’); we re-constructed the hyperactivity score to obtain
cross-cycle consistently. 11 3. Basic Research Methods
As noted above, we use two sources of variation to identify the effects of relative age on
parent-reported inattentive/hyperactive behavior of Canadian children. First, across provinces
and time14 we exploit differences in school start age policies which mean some children of
exactly the same age (in months) are relatively young in class in one province but not another
(difference in difference design). Second, within province, we exploit the randomness in exact
date of birth for children born on either side of the school entry age cut-off and compare
outcomes of otherwise similar children who received distinct treatment due to the enrolment
eligibility legislation (regression discontinuity design).
Difference in Difference Design
Differences across the ten provinces in the cut-off dates for school entry as well as
changes in cut-off dates during our study period for some provinces provide sufficient variation
to allow us to compare children of exactly the same age in provinces with different cut-offs, so
that in one province, the child is ‘young in class’ while in the other, he/she is ‘old in class.’
(1)
I/H i = α+ τ Young6mosip + β1 Ageinmonths i + β2 Grade i + β3 Cycle i + + λ Xi + ε i
where I/H i is the inattentive/hyperactivity score (based on parent report) for child i;
Young6mosipt is a dummy variable indicating that the child was born in the 6 months prior to the
school-entry cut-off in that year and province; Ageinmonths i is the child’s age in months at the
14
However, the number of changes across time in school start age policies are more limited than the differences
across provinces, thus, we rely more heavily on the cross-province policy variation. 12 time of the survey;15 Cycle i refers to the NLSCY cycle from which the observation is drawn
(beginning with Cycle 1 collected in 1994 through Cycle 8 collected in 2008); Grade i refers to
the child’s current grade at school; Xi includes child gender (in pooled boy/girl models), parental
education, log family equivalent income16, family structure and parent immigrant status. We
estimate the DID model with and without the additional covariates for the combined sample of
boys and girls as well as separately for boys and girls. Longitudinal weights are used for all
estimates; standard errors are clustered by province.
Regression Discontinuity Design
Following Elder (2010), our second approach to estimating the effect of a child being
relatively young in his/her class at school on parent reports of child hyperactivity is to compare
children living in the same province who were born shortly before compared to shortly after the
relevant school cut-off date. The RD approach has also been used to study a implications of
being relatively young in class for a variety of child outcomes (e.g., Dhuey and Lipscomb, 2010;
Dobkin and Ferreira, 2010) though no studies of which we are aware exploit the variation that
exists in school-entry policies across Canada (though see Bedard and Dhuey, 2006; Morrow, et
al., 2012; Smith, 2007 for studies using data within the province of British Columbia).
Intuitively, the idea of the RD strategy is that it is hard to think of any plausible reason
why, for example, a Nova Scotian child born in September would be more hyperactive than a
Nova Scotian child born in October of the same year except that the September child would be
the youngest in his or her class while the October child would be among the oldest. But, Figure
15
This is important, since inattentive/hyperactive behaviours change, generally diminishing with the child’s age
(e.g., Spencer et al., 2007). 16
‘Equivalent’ income is family income adjusted for the differing needs of families of different size, using a
‘square-root of family size’ equivalence scale.
13 1 shows that, in fact, Canadian children born just prior to the school cut-off date have higher
levels of inattention/hyperactivity than those born just after. This is evident both in the pooled
boy/girl sample and in separate estimates for boys and for girls (see Figure 2).
More formally, we estimate the following RD model:
(2)
I/H i = α+ τ Youngi + γ f(bdi-ci) + β1 Province i + β2 Cycle i + β3 Grade i + λ Xi + ε i
where I/ H i is the inattentive/hyperactive score (based on parent report) for child i; Youngi is a
dummy variable indicating that the child has a birth-date in the period immediately prior to the
school-entry cut-off;17 f(bdi-c) is a function of the number of days between the child’s exact
birthdate, bdi, and his or her relevant school-entry cut-off date;18 Province i is the child’s
province of residence; Cycle i refers to the NLSCY cycle from which the observation is drawn
(beginning with Cycle 1 collected in 1994 through Cycle 8 collected in 2008); Grade i refers to
the child’s current grade at school; Xi includes child gender (in pooled boy/girl models), parental
education, log family equivalent income,19 family structure, and parent immigrant status. Again,
we first estimate for a combined sample of boys and girls, adding a dummy for ‘boy’ to the set of
covariates; we then estimate separately for boys and girls. Longitudinal weights are employed
for all analyses; standard errors for the RD estimates are clustered at birth dates.
17
We have used windows around the school cut-off day ranging from one month before and one month after to six
months before and six months after.
18
We have tried specifying f(bdi-c) as a linear, quadratic and cubic function.
19
Family incomes are in real 2006 Canadian dollars, using CPI with 2001 basket content. ‘Equivalent’ income
adjusts dollar income to reflect differences in needs of families of different size. We use the Luxembourg Income
Study equivalence scale (square root of family size) to make this adjustment.
14 4. Basic Estimation Results20
Difference in Difference Results
Difference-in-difference (DID) results are reported in Table 2.21 For both boys and girls,
we find that children who are young in class have more parent-reported symptoms of
inattention/hyperactivity than their peers in other provinces/time periods who are the same age
but who are positioned differently relatively to their class-room peers as a result of differences
across place/time in school start cut-offs. The size of the ‘young in class’ effect of about 0.4
points is very similar for boys and girls (0.13 of a standard deviation for boys; 0.14 of a standard
deviation for girls).22 But boys have significantly higher inattentive/hyperactive scores than girls
(about 1 point relative to the over-all mean of about 4 points), so that this effect is more likely to
push them into clinical levels than it does for girls. Relative to other key covariates, the size of
effect for being young in class is roughly 60 percent as large as having a parent with a university
level education, for example, though in the opposite direction. The young in class effect is larger
than the lone-mother association for girls (about 80 percent as large for boys).23
Robustness Checks for the DID Estimates
Table 3 presents a series of robustness checks. Currie and Stabile (2006) have noted that
there are strong negative associations between inattentive/hyperactive behaviour and human
capital acquisition at well below normal ‘clinical’ levels of these behaviours (i.e., for children
who are inattentive or boisterous but whose symptoms are far from being diagnosed as
20
OLS results for the linear inattentive/hyperactive score are reported in the paper. We have also estimated ordered
probit models as well as an OLS model for log(hyperactivity score +1). The general nature of results is unchanged. 21
We have tried a variety of specifications for age in months. Higher order terms were generally not statistically
significant and other conclusions were not affected. Hence, we report only the quadratic ‘age in months’ variable.
22
Formal statistical tests reject the hypothesis of a significant difference in the size of the ‘young’ effect for girls and
boys (i.e., the interaction of ‘boy’ and ‘young’ is never statistically significant in pooled models). Evans et al.,
(2010) also found no significant difference for boys compared to girls.
23
Bertrand and Pan (2011) also find that boys have a harder time in lone-mother families.
15 ADHD24). Thus, a first test of robustness is to see if being young at school increases parent
reports of child inattentive/hyperactive symptoms at home for ‘non-clinical’ cases. This is done
by excluding: 1) children who are reported to be on Ritalin; and, 2) children who are in the top
decile of the inattentive/hyperactive score. Relatively few children in our sample are reported to
be taking Ritalin (2.8 percent of boys; 0.8 percent of girls), so results are little affected by this
exclusion.25 The estimated coefficient for ‘young’ is only about half as large when we exclude
all children in the top decile of the inattentive/hyperactive behaviour score, but it remains
statistically significant.
Since our analysis defines children as young in class by comparing their birth date with
the legislated school start date for their province/time period, a concern might be that parents do
not always comply with the legislation. For example, if some parents are aware of the literature
on being ‘young in class,’ they may choose to hold their children back in order to provide an
advantage relative to peers (academic ‘redshirting’). This seems particularly likely if the child is
perceived as ‘not ready’ for school. If this is so, then we would have more ‘less able’ children in
the ‘old’ group, and we would underestimate the implications of being ‘young.’ However,
Appendix Table 2 indicates that, in Canada, only a very small number of children are not in
compliance with the school entry regulations in their year/province (at most 3 percent). Not
24
Attention-Deficit/Hyperactivity Disorder (ADHD) is among the most commonly diagnosed behavioural disorders
for children in many countries (Elder, 2010; Faraone, et al., 2003; Skounti et al., 2007). ADHD is a developmental,
neurobiological condition defined by the presence of severe and pervasive symptoms of inattention, hyperactivity
and impulsivity which must be exhibited over a period of at least 6 months, before the age of 7 and in at least two
contexts such as home and school (Daley and Birchwood, 2010; Loe and Feldman, 2007).
25
Perhaps surprisingly, given the availability of public health insurance in Canada, rates of treatment with Ritalin
seems to be lower in Canada than in the U.S. (e.g., Currie and Stabile (2006) estimated that 1.4 percent of Canadian
children aged 4 to 11 in 1994 took Ritalin, compared to 3.3 percent in of the same age and in the same year in the
U.S.
16 surprisingly, then, we see almost no change in estimates if these children are excluded from our
sample.26
Another concern could be that some parents, aware of the ‘young in class’ literature,
attempt to time the month of their child’s birth to ensure that he/she is old in class. However,
Bedard and Dhuey (2006) find no evidence of such behavior in their study of 20 countries, nor
do we find a jump in birth frequency just after the school entry threshold in our data (see Figure
7, that uses pooled cycles of data for provinces without a change in school entry date).
It is also possible that some children born just after the school entry age cut-off, yet who
were perceived as ‘ready’ to start school were sent to private schools that might be more flexible
with respect to parental preferences. Again, this would suggest that we under-estimate the true
effect of birth date on inattentive hyperactive behaviour of children, if parents are more likely to
start particularly able children early. Again, to make sure this is not a significant problem, we
repeat our analyses for a sample that includes all children in the NLSCY regardless of their
schooling status (i.e., home-schooled or not) or type of school attended (public or private). Since
less than 4 percent of children in grades K through 4 attend private school in Canada, results are
also robust to adding children who are home-schooled or attending private schools back into the
sample (see Table 3).
A further issue is that some regions offer public junior kindergarten and some do not and
the availability of daycare varies across the country.27 Since having attended junior kindergarten
or structured daycare may later influence the child’s behaviour in kindergarten or grade one, we
control for whether the child attended a structured daycare or a junior kindergarten when he/she
26
Indeed, these results cannot be released from the Research Data Centre in order to avoid possible residual
disclosure, given the very small numbers of children involved. This contrasts with U.S. findings of growing
numbers of ‘redshirted’ children (e.g., Deming and Dynarski, 2008; Evans et al., 2010).
27
See, for example, Lefebvre and Merrigan, 2008 or Lefebvre, Merrigan, and Verstraete, 2009.
17 was 3/4. To do this, it is necessary to further limit our sample to children observed at age 3 or 4
and with answers to questions about attendance in junior kindergarten or formal daycare at that
time. Again, our main causal findings about increased inattentive/hyperactive behaviour for
children who are relatively young at school are unaffected by the inclusion of this control.28
Regression Discontinuity Results
RD results for the pooled boy/girl sample are presented in Table 4 which reports only the
estimated coefficient for ‘boy’ and for the ‘young’ variable. Again, we find that children who
are relatively young in class are reported by their parents to exhibit more inattentive/hyperactive
in the home setting. The RD results suggest that children who are young in school have scores
that are about 0.5 points higher than their peers on the ‘other side of the window.’ These
estimates are similar in magnitude to those obtained using DID methods. Again, while we find
boys to have significantly higher inattentive/hyperactive scores than girls (about 1 point), we
never find a statistically different impact of being young in class for boys than for girls. In the
estimates reported here, we include all covariates and use a linear specification for the distance
from cut-off function, f(bdi-c). We have also used quadratic and cubic specifications for the
running variable. Higher order terms are nearly always insignificant; main conclusions are not
affected, so results are not reported here in the interests of space. We report results for window
lengths of two and three months before/after the school cut-off date. The same series of
robustness checks are again carried out. Perhaps due to the smaller sample sizes, the RD
estimates are slightly less robust than the DID estimates. Nonetheless, the basic conclusion that
children who are relatively young in class exhibit more inattentive/hyperactive behaviour than
their relatively older peers generally holds.
28
While there is no association between having attended junior kindergarten at 3/4 and current
inattentive/hyperactive score, we do find a correlation between having structured daycare at 3/4 and higher
inattentive/hyperactive scores at 5/9. This is consistent the findings of Baker, Gruber and Milligan (2008). 18 In order for our interpretation that higher levels of inattentive/hyperactive behaviours are
caused by being young relative to class-mates, we do not want it to be the case that other
potential reasons for hyperactivity ‘jump’ near the school start-date cut-off. One very nice
feature of the RD strategy is that it is easily possible to test for discontinuities in observable
variables by estimating a set of regressions of the form of equation (2) for each of the covariates
in our data set (Dobkin and Ferreira, 2009 and Lee and Lemieux, 2010). We do this, for one,
two and three-month windows and find no evidence of discontinuities around the cut-off for any
of the covariates except, as we would expect, that grade is statistically significant (and positive).
These results are available on request.
5. Is Being Young at School Harder for Children with Greater Inattention/Hyperactivity
Prior to School Entry?
The consensus in the medical literature appears to be that while ADHD diagnosis most
often occurs during the elementary school years, symptoms of inattentive/hyperactive behaviour
are typically already evident during the pre-school years (Loe et al., 2008). In this section we
make further use of the longitudinal nature of the NLSCY to ask whether increases in
inattentive/hyperactive behaviour associated with being young in class are larger for children
who arrive at school entry age with higher levels of inattention/hyperactivity? That is, do
children who have troubles sitting still or paying attention find it harder to cope in a classroom
when they are among the youngest members of their class? To the best of our knowledge, this
question has not previously been studied in the literature, and may be particularly important for a
gender comparison, given that pre-school boys are more inattentive/hyperactive than girls (see
Figure 4).
19 Restricting the sample to children for whom we have parent reports of hyperactivity both
at ages 2/3 and at ages 4 through 9, we estimate the following difference-in-difference29
specification:
(3)
I/H i = α+ τ1 Young6mosip + τ2 I/H2/3i + τ2 Hyper2/3i X Young6mosip +
β1 Ageinmonths i + β2 Cycle i + β3 Grade i + λ Xi + ε i
where I/H i is again parent-reported inattentive/hyperactive behaviour for the child at ages 4
through 9, and I/H2/3 i is the child’s inattentive/hyperactive score at age 2 or 3. We also include
an interaction between inattention/hyperactivity at age 2 or 3 and the ‘young at school’ variable,
to test whether symptoms of inattention/hyperactivity are exacerbated by being young at school,
using both a linear specification for pre-school inattention/hyperactivity as well as categorical
variables indicating the child’s pre-school I/H percentile (e.g., greater than 75th). Other
explanatory variables are as described for the difference-in-difference model (1) above. Note
that a further advantage of controlling for pre-school inattentive/hyperactive behavior is that if
there is any gender bias in pmk reports (e.g., parents think ‘boys will be boys’ and so more
boisterous), adding the score at age 2/3 means we are now effectively estimating a first
difference model for the child’s inattention/hyperactivity score so that we need not be so
concerned about this form of reporting bias, provided it is consistent over time.
Results for these models are reported in Table 5. Key findings are that: 1) there is strong
persistence in reported inattentive/hyperactive behaviour; 2) being young in class nevertheless
increases parent reports of inattention/hyperactivity, even controlling for scores reported prior to
school entry; 3) children who are more inattentive/hyperactive pre-schoolers have the hardest
time being young in class. Indeed, the total increase in inattentive/hyperactive score for a child
who was in the top quantile of the pre-school distribution and young in class is 0.712 points
29
Our focus from here on is on the DID models where we have a larger sample size.
20 (0.264 + 0.448, or, about one quarter of a standard deviation). Since boys comprise 57 percent of
the toddlers in the highest quartile of the hyperactivity distribution, in this sense, being ‘young in
class’ is more of a boys’ issue.
6. Are Inattentive/Hyperactive Behaviours During Early School Years Predictive of
Educational Outcomes in Early Adolescence?
Our final question is whether higher levels of inattention/hyperactivity during early years
at school matter for educational behaviours, aspirations and/or outcomes during adolescence;
and, if so, whether gender differences in inattention/hyperactivity help explain observed
differences in adolescent educational outcomes? Certainly, the literature arguing that the
acquisition of both cognitive and non-cognitive skills is a cumulative process suggests
‘snowball’ effects from what happens early in life. And, if this is the case, gender differences in
inattention/hyperactivity in early childhood, perhaps exacerbated by school entry laws, could be
a factor in explaining gender differences in educational outcomes, attitudes and aspirations by
early adolescence. We are particularly interested in aspirations, since, in other research, we have
found that differences in self-reported post-secondary aspirations have been the most important
factor in explaining historical changes in the relative educational outcomes of boys/girls in the
U.S. (Fortin, et al., 2012).
In order to address these questions, we now focus on a sample of 14 and 15 year old
children for whom we also have complete data at 6/7. By this age, adolescents are themselves
asked, with the consent of the parent, to fill out a paper and pencil survey which the Statistics
Canada interviewer promises will not be shown to the parent. Pmk’s also continue to answer
21 surveys about the child. We make use of both child-reported and pmk-reported data to measure
school-related outcomes.
First, we consider the issue of persistence of inattentive/hyperactive behaviour from
parent reports when the child is 6/7 to child self-reports at the age of 14/15. Specifically, we
estimate by OLS:
(4)
I/H14/15 i = α+ β I/H6/7 i + λ Xi + ε i
where Xi, as before, includes child gender, parental education, log family equivalent
income, family structure, parent immigrant status and cycle. We report both OLS estimates and
IV estimates, in which we use Young6mosipt , indicating that the child was born in the 6 months
prior to the school-entry cut-off in that year and province, as an instrument for his/her
hyperactivity score at age 6/7.
All estimates use longitudinal weights, with standard errors clustered at the province
level. As is clear from Table 6, the parent-reported score for the child at age 6/7 is a strong
predictor of the child’s own reports of his/her score at age 14/15, in both the OLS and IV
estimates. A child whose pmk-reported inattentive/hyperactive score was one point higher at age
6/7 is estimated to have a self-reported score that ranges from 0.20 (OLS) to 0.9 (IV) points
higher than his/her otherwise similar peers.
We next ask whether inattentive/hyperactive behaviour reported at age 6/7 is predictive
of other outcomes for adolescents at ages 14/15. We have chosen outcomes to reflect overall
performance at school, attitudes/behaviour, self-confidence and aspirations: 1) the pmk’s report
of the child’s over-all success at school. Specifically, the pmk is asked “Based on your
knowledge of his schoolwork, including report cards,” is [your child] doing overall?” Responses
possibilities are: Very well, well, average, poorly, very poorly; 2) The pmk’s report of whether
22 the child has ever repeated a grade at school; 3) The child’s report of whether he/she “does
homework when assigned:” all of the time, most of the time, some of the time, rarely or never;
4) The child assessment of whether he or she can ‘understand hard questions,’ rarely, sometimes,
often or very often; 5) “How important is it to you to get good grades?” Response categories
include: Very important, somewhat important, not very important, not important at all; 6) ‘How
far do you hope to go in school,?’ with responses ranging from middle/school or junior high to
more than one university degree. Figures 8 through 13 illustrate, first, that adolescent girls have
better school-related outcomes than boys, with the exception that boys have more selfconfidence.
For each of these adolescent outcomes, we estimate:
(5) Yij = α+ β I/H6/7 i + λ Xi + ε i
where Yij is adolescent outcome j for child i and, as above, I/H6/7 i is his/her hyperactivity score
at age 6/7. We estimate specification (5) using both OLS and IV, again, using Young6mosipt as
an instrument for the child’s inattentive/hyperactive behaviour at age 6/7.
Table 7 reports that inattentive/hyperactive behaviour at age six/seven is predictive of worse
outcomes at ages 14/15 for all of the above outcomes, and using both OLS and IV estimation.
Since boys are more likely to have been inattentive/hyperactive before school entry, they would
be more affected by being ‘young in class’ than girls. This has the potential to exacerbate the
pre-existing differences between the genders early in their school careers. Given that child
development is understood as a cumulative process, it is perhaps not surprising that we see
negative links between early boisterous or inattentive behaviour and later attitudes toward school
and post-secondary aspirations.
23 7. Conclusions
We find strong causal links between being young at school for Canadian children aged 4
through 9 and parent reports of inattentive/hyperactive behaviour at home and. Ours is the first
Canadian study to exploit variation across time and place in school entry cut-offs using
nationally representative data. In carrying out our analyses, we use both difference in difference
and regression discontinuity research designs. Obtaining the same results using these two
different research strategies is novel in the literature and strengthens our comfort in the
plausibility of our conclusions. Importantly, increases are observed both for children with
clinical levels of hyperactive behaviour and for those whose behaviour would classify them as
well below clinical levels and, associations between being young in class and these behaviours is
the same for boys and girls.
Given the longitudinal structure of our data, the National Longitudinal Survey of
Children and Youth, a contribution of our research is that we are able to control for parent
reports of child inattentive/hyperactive behaviour at ages 2/3, before the children start school.
We find that being young in class exacerbates an underlying tendency toward
inattentive/hyperactive behaviours, perhaps even pushing some children to clinical levels. Since
boys are reported to have higher levels of inattentive/hyperactive behaviour prior to starting
school, in this sense, the problem we identify is more of a ‘boy’s’ than a ‘girl’s’ issue.
Finally, we examine the idea that higher levels of inattentive/hyperactive behaviour at
age 6/7 are associated with lower academic achievements and aspirations at age 14/15; and, in
particular, that differences in inattentive/hyperactive behaviour helps to explain subsequent
gender differences in academic outcomes. Our data show that boys have significantly worse
24 academic performance than girls for all measures studied except self confidence. However, the
size of the ‘boy’ coefficient falls considerably once we control for inattention/hyperactivity at
age 6/7 (in both OLS and IV estimates. Researchers and policy makers concerned about the
‘boys’ problem’ in Canadian schools, might direct further research attention to understanding
how to help very young boys make the transition into school.
25 References
Almond, Douglas and Currie, Janet. 2011a. "Human Capital Development Before Age Five."
Handbook of Labor Economics. Elsevier.
Almond, Doug and Currie, Janet. 2011b. “Killing Me Softly: The Fetal Origins Hypothesis.”
Journal of Economic Perspective, 25(3): 153-72.
Michael Baker, Jonathan Gruber, Kevin Milligan, 2008 "Universal childcare, maternal labor
supply, and family well-being," Journal of Political Economy, 116, pp. 709–745.
Bauermeister, Jose J., Shrout, Patrick E., Ramirez, Rafael, Bravo, Milagors, Alegria, Margarita,
Martinez-Tabloas, Alfonso, Chavez, Ligia, Rubio-Stipc, Maritza, Garcia, Pedro, Ribera,
Julio, Canino, Glorisa. “ADHD Correlates, Comorbidity, and Impairment in Community
and Treated Samples of Children and Adolescents. Journal of Abnormal Child
Psychology, 35, 883-898.
Bedard, Kelly, and Dhuey, Elizabeth. 2006. “The Persistence of Early Childhood Maturity:
International Evidence of Long-Run Age Effects.” The Quarterly Journal of Economics.
1437-1572.
Bertrand, Marianne and Pan, Jessica. 2011. “The Trouble with Boys: Social Influences and the
Gender Gap in Disruptive Behaviour.” National Bureau of Economic Research, Working
Paper No. 17541. Forthcoming in American Economic Journal: Applied Economics.
Conti, Gabriella and Heckman, James. 2012. “The Economics of Child Well-Being.” National
Bureau of Economic Research. Working paper 18466.
Cornelissen, Thomas, Dustmann, Christian and Trentini, Claudia. 2013. “Early School
Exposure, Test Scores, and Noncognitive Outcomes.” Draft.
Cunha, Flavio and Heckman, James. 2009. "The Economics and Psychology of Inequality and
Human Development." Journal of the European Economics Association. 7(2-3), pp. 32064.
Currie, Janet. 2011. "Inequality at Birth: Some Causes and Consequences." American
Economic Review,” 101:3, pp. 1-22.
Currie, Janet. 2009. "Health, Wealthy, and Wise: Socioeconomic Status, Poor Health in
Childhood, and Human Capital Development." Journal of Economic Literature,. 47:1,
pp. 87-122.
Currie, Janet and Stabile, Mark. 2003. “Socioeconomic Status and Child health: Why is the
Relationship Stronger for Older Children?” American Economic Review, 93: 1813-1823.
26 Currie, Janet and Stabile, Mark. 2006. “Child mental health and human capital accumulation:
The case of ADHD.” Journal of Health Economics, 25, 1094-1118.
Currie, Janet, Stabile, Mark, Manivong, Phongsack and Roos, Leslie L. 2010. “Child Health
and Young Adult Outcomes.” The Journal of Human Resources, 45:3, 517-548.
Daley, D and Birchwood, J. 2010. “ADHD and academic performance: why does ADHD
impact on academic performance and what can be done to support ADHD children in the
classroom?” Child: care, health and development, 35:4, 455-464.
Deming, David and Dynarski, Susan. 2008. “The Lengthening of Childhood.” National Bureau
of Economics Research. Working Paper 14124.
Dhuey, Elizabeth and Lipscomb, Stephen. 2010. “Disabled or young? Relative age and special
education diagnoses in schools.” Economics of Education Review, 29: 857-872.
Dobkin, Carlos and Ferreira, Fernando. 2009. “Do school entry laws affect educational
attainment and labor market outcomes?” Economics of Education Review, (29), 40-54.
Drolet, Marie. 2011. “Why has the gender wage gap narrowed?” Perspectives on Labour and
Income. (Spring) Pp. 3-14.
Elder, Todd E. 2010. “The importance of relative standards in ADHD diagnosis: Evidence
based on exact birth dates.” Journal of Health Economics, 29, 641-656.
Evans, William N. Morrill, Melinda S. And Parente, Stephen T. 2010. “Measuring
inappropriate medical diagnosis and treatment in survey data: The case of ADHD among
school-age children.” The Journal of Health Economics. 29: 657-673.
Faraone, Stephen V. Sergeant, Joseph. Gillberg, Christopher. Biederman, Joseph. 2003. “The
worldwide prevalence of ADHD: is it an American condition?” World Psychiatry 2(2):
104-113.
Fletcher, Jason. 2013. “The Effects of Childhood ADHD on Adult Labor Market Outcomes.”
NBER, Working Paper 18689.
Fletcher, Jason and Wolfe, Barbara. 2008. “Child mental health and human capital
accumulation: The case of ADHD revisited.” Journal of Health Economics, 27, 794800.
Fortin, Nicole. Oreopoulos, Philip and Phipps, Shelley. 2012. “Leaving Boys Behind: Gender
Disparities in High Academic Achievement.”
Frazier, Thomas W., Youngstrom, Eric A., Glutting, Joseph J. and Watkins, Marley W. 2007.
“ADHD and Achievement: Meta-Analysis of the Child, Adolescent and Adult
27 Literatures and a Concomitant Study With College Students.” Journal of Learning
Disabilities, 40:1, 49-65.
Gerson, J. And Gerson, Jonathan. 2002. “A meta-analytic review of gender differences in
ADHD.” 5:143-154.
Gurian, Michael. 2009. The Purpose of Boys: Helping Our Sons Find Meaning, Significance,
and Direction in Their Lives. San Francisco: Jossey-Bass.
Halperin, J.M. and Healthy, D.M. 2011. “The influences of environmental enrichment, cognitive
enhancement, and physical exercise on brain development: Can we alter the
developmental trajectory of ADHD?” Nueroscience and Biobehavioral Reviews, 35: 621634.
Ivis, Frank J. And Adlaf, Edward M. 1999. “Prevalence of methylphenidate use among
adolescents in Ontario,” Canadian Journal of Public Health, 90:5, 309-312.
Lee, David S. and Lemieux, Thomas. 2010. Journal of Economic Literature, 28:281-355.
Lefebvre, Pierre and Merrigan, Philip, 2008, “Child-Care Policy and the Labor Supply of
Mothers with Young Children: A Natural Experiment from Canada,” Journal of Labor
Economics, 26:3, 519-547.
Lefebvre, Pierre, Merrigan, Philip and Verstraete, Matthieu, (2009), “Dynamic labour supply
effects of childcare subsidies: Evidence from a Canadian natural experiment on low-fee
universal child care,” Labour Economics, 16, issue 5, p. 490-502.
Loe, Irene M and Feldman, Heidi M. 2007. “Academic and Educational Outcomes of Children
with ADHD.” Journal of Pediatric Psychology, 32:6, 643-654.
Loe, Irene M. Balestrino, Maria D., Phelps, Randall A., Kurs-Lasky, Marcia, Chaves-Gnecco,
Diego, Paradise, Jack L., and Feldman, Heidi M. 2008. “Early histories of school-Aged
children with attention-deficit/hyperactivity disorder.” Child Development, 79:6, 18531868.
Mahone, E Mark and Wodka, Erick L. 2008. “The neurobiological profile of girls with ADHD.”
Developmental Disabilities Research Reviews, 14: 276-284.
Morrow, Richard L., Garland, Jane. Wright, James. Maclure, Malcolm. Tayler, Suzanne. And
Dormuth, Colin. 2012. “Influence of relative age on diagnosis and treatment of
attention-deficit/hyperactivity disorder in children.” Canadian Medical Association
Journal. 184(7): 755-761.
Oreopoulos, Phil, (2005), “Canadian compulsory school laws and their impact on educational
attainment and future earnings,” Analytical Studies Branch Research Paper Series,
2005251e, Statistics Canada, Analytical Studies Branch. CJE?
28 Oreopoulos, Philip, (2006), “Estimating average and local average treatment effects of education
when compulsory schooling laws really matter,” American Economic Review, 96, issue 1,
p. 152-175.
Perrin, James M. Bloom, Sheila R. Gortmaker, Steven L. 2007. “The Increase of childhood
Chronic Conditions in the United States.” Journal of the American Medical Association,
297(24): 2755-2759.
Romano, Elisa. Tremblay, Richard E. Farhat, Abdeljelil and Cote, Sylvana. 2006.
“Development and Prediction of Hyperactive Symptoms From 2 to 7 Years in a
Population-Based Sample.” Pediatrics, 227:2101-2110.
Sax, Leonard. 2007. Boys Adrift: The Five Factors Driving the Growing Epidemic of
Unmotivated Boys and Underachieving Young Men. New York: Basic Books.
Skounti, Maria. Philalithis, Anastas. Galanakis, Emmanouil. 2007. “Variations in prevalence
of attention deficit hyperactivity disorder worldwide.” European Journal of Pediatrics,
166:177-123.
Smith, Justin. 2007. “Can regression discontinuity help answer an age-old question in
education? The effect of age on elementary and secondary school achievement.” Draft.
Spencer, Thomas J. Biederman, Joseph and Mick, Eric. 2007. “Attention-Deficit/Hyperactivity
Disorder: Diagnosis, Lifespan, Comorbidities, and Neurobiology.” Journal of Pediatric
Psychology, 32:6, 631-642.
Wilens, Timothy E., Biederman, Joseph and Spencer, Thomas J. 2002, “Attention
Deficit/Hyperactivity Disorder Across the Lifespan.” Annual Review of Medicine, 53:
113-31.
29 Figure 1.
Distribution of Pmk Reported Inattentive/Hyperactive Scores for Children Aged 4 to 9
.2 .15
Density
.1
.05 0 0
5
10
Inattentive/Hyperactive Behaviour Score
Boy
15 Girl
Figure 2. Illustration of Difference in Difference Estimation Approach. Inattentive/Hyperactive
Scores for April to June Births versus October to December Births in Quebec and Ontario.
4.8
4.6
4.4
4.2
Quebec
4
Ontario
3.8
3.6
3.4
Young in QC but not ON
Young in ON but not QC
30 Figure 3. Discontinuity in Inattention/Hyperactivity at School Start Date. Boys and Girls Aged
4 to 9.
0
1
2
Mean Value
3
4
5
6
PMK Assessed Hyperactivity Score
-10
0
15-Day Blocks of Age Relative to the Cut-Off Date
10
Note: Each circle represents the average hyperactivity score by 15-day blocks of age for all children aged between
4-11 years from 1994-2008 in the NLSCY. Relative age of zero (i.e. the middle of the x-axis) is the school entry age
cut-off date for a province at a point in time when the child was 4/5 years old. The curve is predicted from a linear
regression fitted on un-weighted individual observations that additionally controls for school grade, cycle dummies,
province dummies and a third-order polynomial of the forcing variable (i.e days from cut-off).
31 Figure 4. Discontinuity in Inattention/Hyperactivity at School Start Date. Boys Compared to
Girls at Ages 4 to 9.
6
PMK Assessed Hyperactivity Score - Girls
0
0
1
1
2
2
Mean Value
3
4
Mean Value
3
4
5
5
6
PMK Assessed Hyperactivity Score - Boys
-10
0
15-Day Blocks of Age Relative to the Cut-Off Date
10
-10
0
15-Day Blocks of Age Relative to the Cut-Off Date
10
Note: Each circle represents the average hyperactivity score by 15-day blocks of age for all children aged between
4-11 years from 1994-2008 in the NLSCY. Relative age of zero (i.e. the middle of the x-axis) is the school entry age
cut-off date for a province at a point in time when the child was 4/5 years old. The curve is predicted from a linear
regression fitted on un-weighted individual observations that additionally controls for school grade, cycle dummies,
province dummies and a third-order polynomial of the forcing variable (i.e days from cut-off).
Figure 5. Distributions of Inattention/Hyperactive Scores at ages 2/3. Boys compared to Girls.
0
.05
Density
.1
.15
.2
Distribution of Pre-Kindergarten Hyperactivity Score
0
5
10
15
Hyperactivity Score
Boy
Girl
32 Figure 6. Distribution of Pre-school Boys Compared to Pre-school Girls Across Hyperactivity
Score ‘Quartiles’
40
35.8
35
31
30
24.7
25
26
22.5
21
20
19.6
Boys
15
15
Girls
10
5
0
Quartile 1
Quartile 2
Quartile 3
Quartile 4
33 Figure 7. Distributions of Birth Month, by Province
.4
Nova Scotia
0
0
.1
.1
Density
.2
Density
.2
.3
.3
.4
Newfoundland and Labrador
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Month of Birth
Sep
Oct
Nov
Jan
Dec
Feb
Mar
Apr
May
Jun
Jul
Aug
Month of Birth
Sep
Oct
Nov
Dec
Quebec
.3
.4
New Brunswick .3
.2
Density
.2
Density
0
Jan
Feb
Mar
Apr
May
Jun
Jul
Month of Birth
Aug
Sep
Oct
Nov
0
.1
.1
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Month of Birth
Sep
Oct
Nov
Dec
British Columbia
0
0
.1
.1
Density
Density
.2
.2
.3
.3
.4
Ontario
Jan
Feb
Mar
Apr
May
Jun Jul Aug
Month of Birth
Sep
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Month of Birth
Sep
Oct
Nov
Dec
34 Figure 8.
60
Parent:"Based on Your Knowledge of His/Her Schoolwork, including report cards, how well is your child doing overall at school?" %
40
Poorly
Average
Well
Very Well
34
29 30
30
27
20
20
7
10
5.4
6
4
2
4
0
0
Boys
Girls
Boys
Child: "How often do you do homework when it is assigned? %
33
30
Never
22
20
Rarely
17
Some of
the time
Most of
the time
15
10
7
5
4
2 1
1
Girls
Child: "How often can you understand hard questions?" %
41
39 38
45
40
27
25
5
8
49
50
35
Parent: "Has your child ever repeated a grade at school?" %
7.6
35
31
Rarely
30
Sometimes
22
25
20
14
15
10
Very Often
8
6
Often
5
0
0
Boys
Girls
Boys
Child: "How important is it to you to get good grades? " %
80
Girls
Child: "How Far Do You Hope to Go in School? %
60
66.7
70
50
36.6
40
48
50
58.4
60
30.6
30
Not
important
at all
Not very
important
39
40
0.6
4.4
0.3
2.5
0
Boys
Girls
College
28
Univ
22
20
Very
important
10
20
10
HS
29
30
Somewhat
important
Jr. High
19
2+
Degrees
10
6
1
1
0
Boys
Girls
35 Table 1. Cut-Off Dates for Entry into Kindergarten in 8 Provinces and School Boards/CMAs of Alberta and
Saskatchewan (1994 to 2008)
September 1 September
October November December December 31
January Last Day March 1
30
31
1
31
of
30
February
PEI
Alberta - Alberta
British
(1994Medicine Edmonton
Columbia
2002)
Hat
Ontario
Saskatchewan
(7 school
boards/CMAs)
Manitoba
(1997-2000)
New
Brunswick
Newfoundland
and Labrador
Alberta
– Lethbridge
– Red Deer
–
Lloydminster
– Grande
Prairie
– Wetaskiwin
Note: 1. 7 school boards/CMAs in Saskatchewan include Regina –Regina District School Division, Yorkton –
Yorkton School Division, Moose Jaw – Moose Jaw School Division, Swift Current – Swift Current School
Division, Saskatoon – Saskatoon District School Division, North Battleford – Battleford School Division, Prince
Albert (cannot be linked with any school division, judging from the name so is not used in the analysis), and Estevan
– Estevan School Division.
Alberta –
Calgary
Quebec
Nova Scotia
PEI (2006)
PEI
(2005
and
2008)
PEI
(2004)
Manitoba
(19941996)
36 Table 2. Difference in Difference Estimates of the Effect of Being ‘Young in Class’ on Inattentive
Behaviour at 4/9 for Children in Public Schools.
Boys+Girls
Boys
Girls
(1)
(2)
(1)
(2)
(1)
(2)
Mean Score
4.099
4.615
3.562
(St Deviation)
(3.030)
(3.157)
(2.792)
Young in Class
0.431***
0.403***
0.404***
0.432***
0.462***
0.380***
(0.062)
(0.030)
(0.099)
(0.084)
(0.040)
(0 .046)
Boy
1.072***
1.065***
--(0.039)
(0.030)
Log Equivalent
-0.219***
-0.219***
-0.218***
Household Income
(0.052)
(0.052)
(0.063)
Pmk University
-0.671***
-0.615***
-0.737***
(0.071)
(0.089)
(0.104)
Pmk College
-0.311***
-0.226
-0.408***
Diploma
(0.083)
(0.197)
(0.122)
Pmk High School
-0.242**
-0.239
-0.254**
(0.103)
(0.213)
(0.096)
Step Family
0.711***
0.788**
0.633***
(0.105)
(0.213)
(0.104)
Lone Parent
0.488***
0.581***
0.387***
Family
(0.085)
(0.101)
(0.096)
Pmk Immigrant
-0.312***
-0.410***
-0.212*
(0.063)
(0.081)
(0.097)
Child Age in
0.059**
0.029
0.041
0.016
0.080***
0.043*
Months
(0.022)
(0.019)
(0.026)
(0.017)
(0.023)
(0.022)
Child Age in
-0.0003**
-0.0002
-0.0002
-0.0001
-0.0004***
-0.0003**
Months Squared
(0.0001)
(0.0001)
(0.0001)
(0.0001)
(0.0001)
(0.0001)
Number of
Observations
30970?
34500
15656
17454
15314
17046
These models also include but do not report NLSCY cycle. Robust standard errors clustered at the provincial level
are reported in parentheses. *** indicates statistically significant at 1 percent; ** indicates statistically significant at
5 percent; * indicates statistically significant at 10 percent. Note that the 'young' coefficient is robust to alternative
specifications (linear and cubic) for age in months.
37 Table 3. Robustness Checks for DID Estimates of Being Young in Class on Inattentive Behaviour at 4/9.
Excluding Children on Ritalin
Excluding Children in Top Decile of Boy +
Girl Inattentive Score
Boys + Girls
Boys
Girls
Boys + Girls
Boys
Girls
Young
0.373***
0.319***
0.431***
0.202**
0.128*
0.272***
(0.053)
(0.079)
(0.039)
(0.067)
(0.068)
(0.077)
Boy
0.964***
0.673***
(0.031)
(0.062)
Number of
30402
15219
15183
28462
13952
14510
Observations
Including Children Not in Public School
System
Young
Boy
Attended Junior
K at Age 4-5
Attended Formal
Daycare at 4/5
Number of
Observations
Boys + Girls
0.425***
(0.053)
1.066***
(0.037)
--
Boys
0.387***
(0.114)
Girls
0.472***
(0.032)
--
--
--
--
--
32366
16377
16029
Restricting to Longitudinal Data to Control for
Attendance at Structured Daycare or Junior
Kindergarten at Age 3/4
Boys+Girls
Boys
Girls
0.364***
0.337***
0.400***
(0.065)
(0.093)
1.086***
(0.067)
-0.035
0.021
-0.078
(0.065)
(0.066)
(0.105)
0.341***
0.318*
0.372**
(0.098)
(0.167)
(0.116)
28152
14218
13934
These models also include but do not report child age and age squared family income, parental education, family
structure, immigrant status and NLSCY cycle. Robust standard errors clustered at the provincial level are reported in
parentheses. *** indicates statistically significant at 1 percent; ** indicates statistically significant at 5 percent; *
indicates statistically significant at 10 percent.
38 Table 4. Regression Discontinuity Estimates of the Effect of Being ‘Young in Class’ on Inattentive
Behaviour of Boys+Girls at Age 4/9.
All
Excluding
Excluding
Including
Controlling for
Boys+Girls Children on
Children in
Children Not
Attendance at
Ritalin
Top Decile of
in Public
Daycare or Junior
Score
School
Kindergarten at
Age 3/4
2-month window
Mean Score
4.105
3.991
3.479
4.080
4.133
(St Deviation)
(3.063)
(2.954)
(2.379)
(3.058)
(3.040)
Young in Class
0.481**
0.431*
0.216
0.502*
0.150
(0.237)
(0.222)
(0.192
(0.237)
(0.231)
Boy
1.007***
0.879***
0.620***
0.961***
0.914***
(0.108)
(0.106)
(0.090)
(0.110)
(0.113)
+ Covariates
x
x
x
x
x
Number of
10156
9966
9286
10626
9125
Observations
3-month window
Mean Score
4.085
3.975
3.466
4.068
4.109
(St Deviation)
(3.067)
(2.951)
(2.369)
(3.054)
(3.044)
Young in Class
0.540***
0.437**
0.231
0.573***
0.292
(0.198)
(0.188)
(0.154)
(0.198)
(0.195)
Boy
0.962***
0.847***
0.605***
0.932***
0.947***
(0.088)
(0.089)
(0.075)
(0.088)
(0.092)
+ Covariates
x
x
x
x
x
Number of
15464
15174
14175
16189
13885
Observations
These models also include f(bdi-ci), in a linear specification, grade, in a cubic specification, child gender, family
income, parental education, family structure, immigrant status as well as province and NLSCY cycle. Robust
standard errors clustering at distance from the cut-off are reported in parentheses. *** indicates statistically
significant at 1 percent; * indicates statistically significant at 10 percent.
39 Table 5. Difference in Difference Estimates of the Effect of Being ‘Young in Class’ on Inattentive
Behaviour at 4/9. Controlling for Hyperactive/Inattentive Behaviour at 2/3.
(2)
(3)
(4)
(5)
Mean Score
4.115
4.115
4.115
4.115
(St Deviation)
(3.036)
(3.036)
(3.036)
(3.036)
Young in Class
0.364***
0.315***
0.319***
0.264*
(0.084)
(0.077)
(0.079)
(0.111)
Boy
1.090***
0.906***
0.938***
0.936***
(0.083)
(0.073)
(0.076)
(0.074)
Inattentive Score at 2/3
-0.400***
--(0.004)
Quartile of Inattentive Score at 2/3
Second
--0.923***
0.990***
(0.072)
(0.169)
Third
--1.624***
1.520***
(0.077)
(0.075)
Top
-2.850***
2.526***
(0.059)
(0.105)
Quartile X Young Interactions
Second X Young
----0.132
(0.172)
Third X Young
---0.041
(0.121)
Top X Young
---0.448***
(0.145)
+ Covariates
x
x
x
x
Number of Observations
25842
25842
25842
25842
These models also include but do not report child age and age squared family income, parental education, family
structure, immigrant status and NLSCY cycle. Robust standard errors clustered at the provincial level are reported in
parentheses. *** indicates statistically significant at 1 percent; ** indicates statistically significant at 5 percent; *
indicates statistically significant at 10 percent.
40 Table 6. Estimates of Implications of Parent-Reported Inattentive Behaviour at 6/7 for Self-Assessed
Inattentive Behaviour at Age 13/14. Boys+Girls. Using 'Young in Class' as IV.
OLS
OLS
IV
Mean Score
3.919
3.919
3.919
(St Deviation)
(2.892)
(2.892)
(2.892)
Inattentive Behaviour at 6/7
-0.189***
0.898**
(0.027)
(0.323)
Boy
0.358***
0.174***
-0.512*
(0.030)
(0.030)
(0.254)
These models also include but do not report child age and age squared family income, parental education, family
structure, immigrant status and NLSCY cycle. Robust standard errors clustered at the provincial level are reported in
parentheses. *** indicates statistically significant at 1 percent; ** indicates statistically significant at 5 percent;*
indicates statistically significant at 10 percent.
Table 7. Estimates of Implications of Inattentive Behavour at 6/7 for Outcomes at Age 14/15. Boys+Girls.
Using 'Young in Class' as IV.
Parent-Reported Educational Success
Ever Repeated a Grade at School
OLS
OLS
IV
OLS
OLS
IV
Mean Score
4.065
4.065
4.065
0.065
0.065
0.065
(St Deviation)
(0.943)
(0.943)
(0.943)
(0 .247)
(0 .247)
(0 .247)
Inattentive
--0.073***
-0.268*
-0.014
0.079*
Behaviour at 6/7
(0.004)
(0.099)
(0.007)
(0.040)
Boy
-0.329***
-0.255***
-0.068
0.028**
0.011
-0.051
(0.026)
(0.025)
(0.124)
(0.009)
(0.006)
(0 .044)
Self-Reported ‘I Do Homework when
Self-Reported ‘I Can Understand Hard
Assigned’
Questions’
OLS
OLS
IV
OLS
OLS
IV
Mean Score
5.152
5.152
5.152
2.687
2.687
2.687
(St Deviation)
(1.051)
(1.051)
(1.051)
(0.850)
(0.850)
(0.850)
Inattentive
--0.037***
-0.401***
-0.024**
-0.292**
Behaviour at 6/7
(0.008)
(0.092)
(0.008)
(0.115)
Boy
-0.256***
-0.217***
0.202*
0.188**
0.219***
0.474***
(0.055)
(0.058)
(0.108)
(0.060)
(0.060)
(0.111)
Self-Reported Importance of Getting
Self-Reported Educational Aspirations
Good Grades
OLS
OLS
IV
OLS
OLS
IV
Mean Score
2.584
2.584
2.584
3.838
3.838
3.838
(St Deviation)
(0.580)
(0.580)
(0.580)
(0.905)
(0.905)
(0.905)
Inattentive
-0.005
-0.136**
--0.031***
-0.254**
Behaviour at 6/7
(0.007)
(0.044)
(0.005)
(0.095)
Boy
-0.112***
-0.107***
0.025
-0.279***
-0.232***
0.002
(0.014)
(0.009)
(0.023)
(0.017)
(0.019)
(0.096)
These models also include but do not report child age and age squared family income, parental education, family
structure, immigrant status and NLSCY cycle. Robust standard errors clustered at the provincial level are reported in
parentheses. *** indicates statistically significant at 1 percent; ** indicates statistically significant at 5 percent; *
indicates statistically significant at 10 percent.
41 Appendix Table 1. Example of Provincial Variation in Construction of ‘Young’ Variable for Difference
in Difference Estimates.
September 1 September
October 31
December
End
March 1
30
31
February
Calgary
Quebec and PEI
BC, SK,
Medicine
Edmonton
Nova Scotia
MB, ON,
Hat
NB, NL,
Other AB
January
Young
Young
February
Young
Young
March
Young
April
Young
Young
May
Young
Young
Young
June
Young
Young
Young
July
Young
Young
Young
Young
August
Young
Young
Young
Young
September
Young
Young
Young
Young
Young
October
Young
Young
Young
Young
November
Young
Young
Young
December
Young
Young
Young
Note: This example table using the most recent school cut-off dates. As noted in Table 1, some changes in school
cut-off dates have taken place during our study period.
Appendix Table 2. Rates of Non-Compliance with School Entry Regulations. Children in Grades K
through 4.
Boys + Girls
Boys
Girls
Newfoundland %
2.6
2.9
2.3
PEI %
0.0
0.0
0.0
Nova Scotia %
2.0
2.3
1.7
New Brunswick %
2.7
2.8
2.6
Quebec %
1.5
1.5
1.4
Ontario %
2.6
2.4
2.7
Manitoba %
2.3
1.5
3.0
Saskatchewan %
0.0
0.0
0.0
Alberta %
0.0
0.0
0.0
BC %
2.4
3.0
1.6
42 
Download