SERIES

IZA DP No. 10065

PAPER

The Impact of Immigrant Peers on Native Students’

Academic Achievement in Countries Where
Parents of Immigrants Are Relatively Skilled
DISCUSSION

Kelvin KC Seah

July 2016

Forschungsinstitut
zur Zukunft der Arbeit
Institute for the Study
of Labor
 The Impact of Immigrant Peers on
Native Students’ Academic Achievement in
Countries Where Parents of Immigrants
Are Relatively Skilled

Kelvin KC Seah
National University of Singapore
and IZA

Discussion Paper No. 10065

July 2016

IZA

P.O. Box 7240

53072 Bonn
Germany

Phone: +49-228-3894-0
Fax: +49-228-3894-180
E-mail: [email protected]

Any opinions expressed here are those of the author(s) and not those of IZA. Research published in
this series may include views on policy, but the institute itself takes no institutional policy positions.
The IZA research network is committed to the IZA Guiding Principles of Research Integrity.

The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center
and a place of communication between science, politics and business. IZA is an independent nonprofit
organization supported by Deutsche Post Foundation. The center is associated with the University of
Bonn and offers a stimulating research environment through its international network, workshops and
conferences, data service, project support, research visits and doctoral program. IZA engages in (i)
original and internationally competitive research in all fields of labor economics, (ii) development of
policy concepts, and (iii) dissemination of research results and concepts to the interested public.

IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion.
Citation of such a paper should account for its provisional character. A revised version may be
available directly from the author.
IZA Discussion Paper No. 10065
July 2016

ABSTRACT

The Impact of Immigrant Peers on Native Students’

Academic Achievement in Countries Where Parents of
Immigrants Are Relatively Skilled

This study examines how exposure to immigrant students affects the academic achievement
of native students in the three largest immigrant-receiving countries – United States,
Australia, and Canada. Using a large cross-country dataset, variation in the share of
immigrant children between different grade levels within schools is exploited to identify the
impact of immigrant peers. I find that exposure to immigrant children has dissimilar effects on
native students’ achievements across the three countries. While exposure has a positive
impact on Australian natives, it has a negative impact on Canadian natives. Exposure has no
effect on U.S. natives. More importantly, I find that institutional factors, such as the way in
which countries organise their educational systems, have a crucial bearing on how immigrant
students affect their peers.

JEL Classification: I21, J15

Keywords: academic achievement, immigrant children, peer effects, within-school estimation

Corresponding author:

Kelvin KC Seah
Department of Economics
National University of Singapore
AS2 #06-02, 1 Arts Link
Singapore 117570
E-mail: [email protected]
1 INTRODUCTION

Over the past three decades, developed countries have witnessed a

dramatic increase in the number of incoming international migrants. A product
of this movement has been the rising enrolment of immigrant students and
concomitant changes in the ethnic and nativity composition of students in
schools and classrooms of the receiving countries (Betts and Fairlie, 2003;
Jensen and Rasmussen, 2011; OECD, 2012; Brunello and Rocco, 2013).

One question which is of relevance for the receiving country’s education

policy is whether the presence of immigrant students has an effect on the
academic performance of peers in the same learning environment. Despite the
importance of such knowledge for policy-making, research on this issue is
scant. The small number of existing studies examining the effects that
immigrant students have on the educational outcomes of peers has, until now,
been largely based on evidence from countries where the skill composition of
immigrants relative to natives is low (Szulkin and Jonsson, 2007; Gould et al.,
2009; Neymotin, 2009; Jensen and Rasmussen, 2011; Hardoy and Schøne,
2013; Ohinata and van Ours, 2013; Schneeweis, 2013; Hermansen and
Birkelund, 2015). These studies generally find that exposure to immigrant
students has either negative or no effects on peers’ academic achievement1.
Apart from a study by Friesen and Krauth (2011), no attempt has been made to
examine the effects of such children in countries where migrants are, on
average, more skilled than natives (e.g. in Canada, Australia, or New

1
Exceptions are Neymotin (2009), Ohinata and van Ours (2013), and
Hermansen and Birkelund (2015), who find some patchy evidence that
exposure to immigrant children have positive impacts on native students’
educational outcomes.

2
Zealand2)3. Yet, one would expect immigrant peer effects to be quite different,
possibly even positive, given that the school-going immigrant population in
these countries tend to be from more privileged socioeconomic backgrounds.

Results from a study by Schnepf (2006) reveal that the socioeconomic

backgrounds of immigrant children in countries that typically attract high-
skilled migrants tend to be similar to or even surpass those of native children.
Immigrant children in such countries also perform either no differently or only
marginally poorer, academically, compared to their native counterparts. By
contrast, immigrant children in countries that generally receive less-skilled
migrants tend to be markedly less advantaged socioeconomically than native
children. They also tend to perform considerably worse in school.

Because the immigrant children who are at the focus of existing studies
typically come from less-privileged backgrounds, peer socioeconomic
composition and school immigrant concentration are likely to be negatively
correlated in these studies (Jensen and Rasmussen, 2011). Since most studies
in the immigrant peer effects literature ignore modelling explicitly the effects

2
Unlike in most other high immigration countries, the average educational
level of immigrants in Canada, Australia, and New Zealand stand out as being
much higher than that of natives. See, for example, Antecol et al. (2003).
3
Though Friesen and Krauth (2011) provides some indication on the possible
effects of immigrant peers in a country where the migrant population tends to
be relatively skilled, the results are not generalisable. There are 2 reasons for
this. Firstly, the study focuses only on the effects of 2 specific groups of
immigrant students – those that speak Chinese and those that speak Punjabi at
home. This hardly covers the universe of immigrant students in their country
of study (Canada). Secondly, the evidence is based on student experience in
British Columbia and may not be representative of student experience in the
other Canadian provinces.

3
which are due to the socioeconomic backgrounds of immigrant schooling
peers, the estimated effects documented in these studies will also include any
effects which are due to the unfavourable socioeconomic characteristics of
migrant peers. Since the estimates capture a “total treatment effect”, the
findings may not be generalisable to those countries where immigrant children
are, on average, from more-privileged socioeconomic backgrounds.

Given the above considerations, the objective of this study is to fill the
gap in the literature by examining whether the negative immigrant peer effects
documented in much of the literature still persists when immigrant students
have relatively higher-skilled parents than native students.

Specifically, 4 research questions are addressed in this study:

(1) How do first-generation immigrant peers affect the Mathematics and

Science achievements of native students if parents of immigrants are, on
average, more skilled than those of natives?

(2) How does student achievement vary with the share of immigrant peers?

(3) If immigrant peers indeed have spill-over effects on the academic

achievements of natives, what are the mechanisms behind these effects?

(4) Do institutional factors, such as the way in which countries organise their
educational systems, have an influence on the peer effects exerted by
immigrants?

While there have been a small number of previous studies which have
tried to investigate whether academic achievement and the share of immigrant
peers varies non-linearly4, none have attempted to entirely relax the linearity

4
These include Gould et al. (2009), Schneeweis (2013), Szulkin and Jonsson
(2007), and Hardoy and Schøne (2013). Gould et al. (2009) and Schneeweis
(2013) allow for a quadratic relationship between the grade share of

4
assumption by using non-parametric regression methods to recover the
underlying functional relationship between these variables. The present study
will be the first to model any possible non-linearity this way. Existing studies
also rarely investigate how and why immigrant peer effects arise. With the
exception of Hardoy and Schøne (2013) and Ohinata and van Ours (2013), no
other study I know of has attempted to investigate the mechanisms behind
these effects. The present research is among the few which attempts to study
whether the peer effects are possibly generated by differences in the language
ability of immigrants and educational attainment of their parents. The final
research question adds to the literature most appreciably. To my knowledge,
no study has yet examined the relationship between immigrant peer effects
and educational institutions. Yet, the fact that educational policies and
institutions have been found to affect the relative academic achievement of
immigrants (Schneeweis, 2011; Cobb-Clark et al., 2012) implies that it may be
possible for institutional factors to influence the way in which immigrant
students affect others.

In this study, we are interested in examining the peer effects only from
first-generation immigrant students. Given that differences in attributes (e.g.
differences in language ability and/or host-country-specific human capital) are
likely to be sharper between native students and first-generation immigrant

immigrants and student outcomes. On the other hand, Szulkin and Jonsson
(2007) and Hardoy and Schøne (2013) first create dummy variables for the
shares of immigrant peers by grouping these into arbitrarily-sized bins (e.g. 0-
5%, 5-10% immigrants and so on) before regressing student outcomes on
these dummies to allow for non-linearities. Both approaches have their
disadvantages. The former approach restricts the possible functional
relationships analysed to only quadratic and linear forms while the conclusions
yielded by the latter approach are sensitive to the base category specified and
to the size of the bins used to group the migrant shares.

5
students than between native students and later generations of immigrant
students, these students constitute a particularly interesting group to study.
Hence, throughout the rest of this paper, immigrants refer to first-generation
migrants – that is, all students not born in the country where the test was
conducted5. Note also that, throughout the paper, the level of skills possessed
by parents is proxied by their educational attainment.

The study proceeds by providing an international comparative study of

the peer effects generated by migrants in 3 major immigrant-receiving
countries – Australia, Canada, and the United States. Because parents of
immigrants in the U.S. are, on average, lower-skilled than parents of natives
whilst parents of immigrants in Australia and Canada are, on average, higher-
skilled than parents of natives, a comparison of the immigrant peer effects
across these 3 countries will provide an indication on whether exposure to
immigrant peers is indeed less adverse when the parents of migrants are
relatively high-skilled.

The empirical analysis is based on data from the Third International

Mathematics and Science Study of 1995. This dataset provides, for all 3
countries, information on the nativity statuses and Math and Science abilities
of students from 2 adjacent grades within sampled schools. This allows me to
exploit plausibly exogenous variation in the share of immigrants across

5
I do not further restrict first-generation migrants to those who have at least 1
foreign-born parent because foreign-born children born abroad of 2 native-
born parents may also have missed out on certain educational experiences
specific to the country during their time of absence. Hence, they are also of
interest to this study. In any case, the results do not differ substantively when
the definition of first-generation migrant students is restricted to those who
were born-abroad and who had at least 1 foreign-born parent.

6
adjacent grades within schools to identify the causal impact of immigrant
peers on natives’ academic achievement.

The identification strategy employed is an improvement over a number

of those previously used in the literature. Studies in the literature have used
within-school between-class variation in the share of immigrant pupils
(Ohinata and van Ours, 2013) and instrumental variable (IV) strategies (Jensen
and Rasmussen, 2011) to address the potential endogeneity in the variable
measuring the degree of exposure to immigrant pupils. However, these
approaches face various methodological issues. The former approach requires
a strong assumption that students are randomly assigned across classes within
schools while the latter approach requires the difficult task of finding a
variable which is correlated with the share of immigrant peers but which
otherwise is unrelated to student achievement. Jensen and Rasmussen (2011)
suggest using immigrant concentration in a larger geographical area as an
instrument for immigrant concentration in the school. However, the suggested
variable is unlikely to satisfy the requirements needed for an IV since previous
studies have shown that it is possible for the characteristics of the larger
community to have a direct influence on students’ educational outcomes (see,
for instance, Ainsworth (2002)). In comparison, the identification strategy
used in this paper requires only a relatively weak assumption that variations in
the share of immigrants between grades in schools are random.

There are a number of reasons why the presence of immigrant children

can have an effect on the academic achievements of native peers. Firstly,
coming from a different country and culture, immigrant children may possess
different language skills, knowledge, aspirations, and attitudes towards
education. Through socialisation within the same learning environment,
transmission of skills and educational aspirations between migrants and
natives can result (Szulkin and Jonsson, 2007). Since immigrant students may
possess more positive attitudes towards learning and/or perform academically

7
better than native students, this channel allows for immigrants to have a
positive impact on peers’ learning experiences.

Secondly, immigrant students generally possess a poorer command of

the host society’s native language. As a result, teachers may slow the pace of
instruction more than they otherwise would in order to accommodate them
(Hunt, 2012). It is also possible that teachers may alter their pedagogical
methods, using less language-intensive methods to deliver their lessons (Betts
and Fairlie, 2003). These could either enhance or hinder the learning
experiences of those studying with immigrants, though the latter seems more
likely. Immigrant students may also divert teaching resources away from other
students if teachers allocate a greater amount of class time providing
individual assistance to them.

Taken together, it seems unclear whether the presence of immigrant

students will have a net beneficial or adverse effect on peers’ academic
achievement. The overall effect of exposure to immigrant peers is therefore an
empirical question.

The results from this study indicate that exposure to immigrant students
does have an effect on the academic achievement of peers. In particular,
results from both non-parametric regressions and school fixed effects
estimations suggest that while exposure to immigrant peers has a positive
impact on the Math achievement of Australian natives, it has a negative
impact on the Math achievement of Canadian natives. There is no evidence
that exposure to immigrant peers has an effect on the academic achievement of
U.S. natives. Interestingly, the result for Canada suggests that even if
immigrant students have more educated parents than natives, this does not
guarantee that immigrants will have non-adverse impacts on the academic
achievements of peers. Additional tests to uncover the mechanisms behind the
respective peer effects reveal that the peer effects of migrants are more
adverse when they are non-native speakers of the test language and when they

8
have less-educated parents. Hence, within countries, an improvement in the
quality of immigrant children (as measured by language proficiency and
parental education) will likely mitigate any negative effects or enhance any
positive effects which these students may have on the educational outcomes of
natives.

Because it is surprising that immigrants in Australia and Canada exert

such different impacts (given the similarity in immigrant selection criteria of
both countries), I additionally conduct an analysis to examine whether
differences in the peer effects of immigrants may be explained by differences
in the way educational systems are organised across countries. I find that the
peer effects of immigrants are more positive when schools have greater
autonomy over the setting of curriculum, when the share of immigrant
students who speak the test language frequently at home is higher, and when
immigrant children arrive in the host country at younger ages. This finding –
that “institutional context matters” – is a novel one in the immigrant peer
effects literature.

The findings from this study will be useful for education policy-makers
because it enables them to know whether country immigration policies (e.g.
policies aimed at increasing the educational attainments of incoming migrants)
and educational institutions have an influence on the way migrant students
affect their peers.

2 DATA

The data used in this paper is from the Third International Mathematics
and Science Study (TIMSS 1995)6. TIMSS 1995 is an international study

6
Available online from: http://www.iea.nl/data.html. This study was renamed
the “Trends in International Mathematics and Science Study” after 1995.

9
which was conducted across more than 40 countries in 1995. The aim of the
study is to assess the Mathematics and Science achievements of students in 5
grade levels (3rd, 4th, 7th, and 8th grades and the grade constituting the final year
of secondary education) of the participating countries.

Because background characteristics are deemed important in explaining

academic performance, TIMSS 1995 also fielded questionnaires to students in
order to collect contextual information. Among other things, the student
questionnaire sought information regarding each student’s demographic
characteristics (e.g. sex, age, migration status, age at arrival to the country if
the child was born abroad) and family background (e.g. educational attainment
of each parent, number of books at home, number of people living in the
home, whether the test language is frequently spoken at home, whether the
child lives with both parents).

Although data on student achievement are available for 5 grade levels, I

focus my study on examining student achievement in only the 7th and 8th
grades. While the 1995 wave is not the most up-to-date version of TIMSS, it
does provide a very unique advantage. In particular, it is the only
internationally comparable dataset which provides achievement and migration
information on students from 2 different grades of the same school. This
makes it well-suited for the identification strategy employed in this paper (this
will be discussed in Section 3). No other datasets which I am aware of – not
even the subsequent waves of TIMSS7 – fulfil the strict data requirements
needed to implement the empirical strategy proposed.

7
The other waves of TIMSS (i.e.1999, 2003, 2007, and 2011 waves) assess
the academic achievements of students belonging to only 2 grade levels (4th
and 8th grade). Students in the 4th and 8th grades are typically not drawn from
the same school, making within-school type comparisons infeasible.

10
 TIMSS 1995 employed a two-stage sampling design. In the first stage,
schools were sampled with a probability proportional to school size. In the
second stage, intact classes of students were randomly selected from within
the sampled school. Generally, within each school, one classroom would be
randomly selected from the 7th grade and another would be randomly selected
from the 8th grade. Because each respondent provides information on his/her
nativity status (whether born abroad or born in the country of test), one is able
to identify whether a particular observation is an immigrant student. This
information allows one to construct the key variable of interest – a variable
measuring the share of immigrant students in the 7th and the 8th grade of each
school8. In all subsequent analyses, I define immigrant students as all those
born abroad.

8
The share of immigrant children in each grade (both 7th and 8th) of a school is
calculated as follows: First, each student is weighted by the product of the
class weighting factor (inverse of the probability of selection of the classroom
within a school), the student weighting factor (inverse of the probability of
selection of a student within a classroom), and the student weighting
adjustment (adjustment to account for non-participating students in the
selected classroom). Next, for each grade in a particular school, the proportion
of immigrant students to all students is computed, applying the
abovementioned weight. All results are similar if I had instead restricted my
sample to only those schools with one class per grade. In this case, the share of
immigrant children in the 7th grade class would be akin to the share of
immigrant children in the 7th grade of the school. Likewise, the share of
immigrant children in the 8th grade class would be akin to the share of
immigrant children in the 8th grade of the school. The disadvantage of the
latter analysis, though, is that very few schools (and hence, student
observations) would remain in the sample, leading to large standard errors in
the impact estimates.

11
2.1 Sample

The dataset consists of individual-level student observations. A total of

12,852, 16,581, and 10,973 student observations from 161, 380, and, 183
schools are available respectively for Australia, Canada, and the U.S.. The
number of foreign-born observations in each of the 3 countries is sizeable. Of
the 12,578 student observations which had non-missing responses on the
country of birth in the Australian dataset, 1,401 observations are foreign-born
while the rest (11,177) are native-born. Of the 16,232 student observations
which had non-missing country of birth information in the Canadian dataset,
1,326 are foreign-born while the rest (14,906) are native-born. Similarly, of
the 10,774 student observations in the U.S. dataset which had non-missing
responses on the country of birth, 938 are foreign-born while the rest (9,836)
are native-born. These figures are consistent with the fact that these 3
countries have been successful in attracting considerable numbers of foreign
students.

Since I am mainly interested in investigating whether the exposure to

immigrant peers has an effect on the academic achievements of native
students, I restrict my analytic sample to those students who reported their
country of birth in the background questionnaires. All students who omitted
providing a response as to whether or not they were born in the country of test
were excluded. Finally, I link each student observation to the share of foreign-
born children in his/her grade in school.

2.2 Measuring Student Achievement

Student achievement is measured using the subject plausible values

(plausible value 1) in TIMSS 1995. To minimise the burden on participating
students, each student was only required to complete 1 of 8 possible test
booklets. Plausible values are estimates of the achievement scores that

12
students would have obtained if they had completed an assessment consisting
of all the items in the 8 test booklets. They are derived using multiple
imputation methods. Throughout this paper, test scores are normalised within
each country and subject, to a mean of 0 and a standard deviation of 1 so that
the values presented can be interpreted as fractions of a standard deviation.

2.3 Parental Education of Immigrant and Native Children

<Insert Table 1 here>

To ascertain if the parents of immigrants in Australia and Canada indeed

have higher average educational attainments than those of natives, I compute
the share of students having fathers who have not completed secondary
education and the share of students having fathers who have at least some
university education, separately for the group of immigrant students and native
students, by country. These are found in Table 1. As expected, the share of
those having fathers with at least some university education is notably larger
for immigrant students than for native students in Australia and in Canada,
indicating that immigrants in these countries have relatively higher educated
parents. As for the United States, although the share of those having fathers
with at least some university education is similar for immigrants and natives, a
larger proportion of immigrant students have fathers that have not completed
secondary education. Therefore, on average, immigrant students in the U.S.
have relatively lower educated parents than native students. These findings
match those previously found from census data in Antecol et al. (2003).

2.4 Preliminary Analysis

Most studies in the immigrant peer effects literature assume the

existence of a linear relationship between student test scores and the degree of

13
exposure to immigrant peers. However, there is no reason a priori to believe
why a linear relationship should hold. In this section, I relax the linearity
assumption entirely and, instead, let the data characterise the functional
relationship underlying the variables.

<Insert Figure 1 here>

Figure 1 presents results from non-parametric regressions (locally

weighted smoothed scatterplots), showing the relationship between native
students’ Math test scores and the share of immigrant peers in each of the 3
countries9. Panel A of the Figure depicts the relationship between the average
Math test scores achieved by native students in a school-grade and the share of
immigrant students in the school-grade. For Australia, average Math
achievement increases with the share of immigrant grade peers throughout
most of the distribution. For Canada, average Math achievement rises initially
until the share of immigrant students in the school-grade reaches a non-zero
value. Thereafter, it falls as the share of immigrant grade peers increase. For
the U.S., average Math achievement falls with the share of immigrant grade
peers until the value of this share reaches roughly 0.38. Thereafter, it remains
constant as the share of immigrant grade peers increase.

Because the above relationships could be driven by selective enrolment

into schools (for example, native students with lower abilities could have a
tendency to sort into schools with higher shares of immigrants), I next present
results from non-parametric regressions showing the relationships between
differences in the share of immigrant students across adjacent grades of a
school and differences in grade average native Math test scores. These are
provided in Panel B of Figure 1. Because variation in the shares of immigrants

9
In Figures 1 and 2, grade effects are partialed out semi-parametrically prior to
estimation. The bandwidth is set to 0.4 for all regressions. I chose the smallest
bandwidth that provided me with a relatively smooth curve.

14
between adjacent grades of each school are used to identify the peer effects of
immigrants, the relationships depicted in this panel account for the non-
random sorting of students between schools.

The figures in Panel B of Figure 1 indicate that when school differences

are taken into account, average Math achievement of native Australian
students still rise with the share of immigrant students in the school-grade.
Similarly, for Canada, average native Math achievement still falls with the
share of immigrant peers10. However, for the U.S., the negative correlation
between average native Math achievement and the share of immigrant peers
disappears and there now appears to be no relationship between them. This
suggests that in the U.S., academically weaker students tend to attend schools
with larger shares of immigrants. This comparative exercise highlights the
importance of using across-grade variation in immigrant shares within schools
to identify the causal effects of immigrant peers. The within transformation
plots provide preliminary evidence that while exposure to immigrant peers
hurts the Math achievement of Canadian students, it actually benefits the Math
achievement of Australian students. Math achievements of American students
do not appear to be affected by the presence of immigrant peers.

The results from non-parametric regressions showing the relationships

between native students’ Science test scores and the share of immigrant peers

10
Throughout most of the distribution, the relationship between the grade
share of immigrants and the grade average Math test scores is negative. The
part of the curve to the left of -0.2 and to the right of 0.2 (of the horizontal
axis) can, for practical purposes, be ignored since the difference in the shares
of immigrants across the 2 grades in the vast majority of schools in the sample
does not exceed 0.2 in absolute value. As such, the regression function is
likely to be poorly estimated at both tails of the distribution where the absolute
difference in the share of immigrants exceeds 0.2.

15
are displayed in Figure 2. The estimated relationships are very similar to those
found in Figure 1. Accordingly, results for Math achievements apply as well to
Science.

<Insert Figure 2 here>

Because the relationships between school-grade average test scores and

school-grade immigrant shares appear to be reasonably well approximated by
a linear function and because the aggregate results presented in this section
could mask compositional differences in the student population across both
grades, I next turn towards estimating linear models with individual-level data.

3 EMPIRICAL STRATEGY

If students were randomly assigned to schools and classes, identifying

the causal effect of exposure to immigrant peers would be straightforward.
Suppose we were interested to know how the share of immigrant children in a
school or class affects the academic performance of peers, we could obtain an
estimate of this effect by performing an OLS regression of test scores on the
share of immigrant children studying in the school or class. The coefficient on
the variable measuring the share of immigrants would then provide an
unbiased estimate of the causal effect of exposure to immigrant children.

In practice, however, estimating the causal effect of immigrant students

is more challenging. This is because students are neither randomly assigned
across schools nor across classrooms within schools. Often, parents choose the
neighbourhoods to settle in as well as the schools to send their children to.
School administrators may also place students into classes in accordance with
their demonstrated academic potential or prior academic achievements.

The potential non-random sorting of immigrant and native students

across schools and across classrooms within schools implies that the

16
unobserved determinants of student achievement may be correlated with the
share of immigrant students in schools and classes. As such, OLS estimators
of the achievement effect of immigrant peers are likely to be biased.

To address the potential endogeneity arising from the non-random

sorting of students across schools as well as across classrooms within schools,
this paper employs a combination of 2 strategies. The first involves
aggregating our measure of immigrant concentration to the school-grade level.
This eliminates the bias due to the non-random sorting of students, by
immigrant status, across classes within schools (in any particular grade). The
second involves identifying the effect of having immigrant peers in a school-
grade by relating variations in the academic performance of students over 2
adjacent grades – grade 7 and grade 8 – of each school to variations in the
share of immigrant students over the 2 grades of the school.

To elucidate, consider the educational production function:

𝑦𝑖𝑔𝑠 = 𝑿′𝑖𝑔𝑠 𝛽 + 𝛿𝐼𝑔𝑠 + (𝛼𝑔 + 𝛼𝑠 + 𝜀𝑖𝑔𝑠 ) (1)

Where 𝑦𝑖𝑔𝑠 denotes the academic performance of student 𝑖 from grade 𝑔 of

school 𝑠. 𝑿𝑖𝑔𝑠 denotes a vector of individual and family background
characteristics for the student. 𝐼𝑔𝑠 denotes the share of immigrant children in
grade 𝑔 of school 𝑠. The error, (𝛼𝑔 + 𝛼𝑠 + 𝜀𝑖𝑔𝑠 ), is a composite given by 3
terms: 𝛼𝑔 and 𝛼𝑠 respectively reflect unobserved grade and school
characteristics which are correlated with both 𝐼𝑔𝑠 and 𝑦𝑖𝑔𝑠 while 𝜀𝑖𝑔𝑠 denotes a
mean zero random error term. Of interest here is the parameter, 𝛿, which
captures the effect of the share of immigrant peers in a school’s grade on the
academic performance of students in that school’s grade.

In order to eliminate any bias due to the non-random sorting of students

across schools, the unobserved school heterogeneity, 𝛼𝑠 , has to be removed

17
prior to estimation. To do this, equation (1) is first averaged over all students
in each school, 𝑠, to yield:

𝑦̅𝑠 = ̅̅̅
𝑿′𝑠 𝛽 + 𝛿𝐼̅𝑠 + 𝛼𝑠 + 𝜀̅𝑠 (2)

Subtracting (2) from (1) then yields:

𝑦𝑖𝑔𝑠 − 𝑦̅𝑠 = (𝑿′𝑖𝑔𝑠 − ̅̅̅

𝑿′𝑠 )𝛽 + 𝛿(𝐼𝑔𝑠 − 𝐼̅𝑠 ) + 𝛼𝑔 + (𝜀𝑖𝑔𝑠 − 𝜀̅𝑠 ) (3)

In the transformed equation (3), our parameter of interest remains

unchanged: 𝛿. In addition to the interpretation given before, 𝛿 can now also be
interpreted as measuring how variations in student test scores around the mean
test score for each school are related to variations in immigrant shares around
the mean immigrant share for each school. Clearly, for this identification
strategy to be implemented, data on student test scores and immigration shares
from at least 2 grades of each school are required. The dataset that I use fulfils
these unique requirements since it provides me with information on the test
scores and nativity statuses of students from the 7th and the 8th grades of each
sampled school. The assumption used to identify the effect of immigrant peers
is that any variation in the share of immigrants over the 2 grades is due
entirely to random demographic differences between age cohorts in schools.
This identification strategy is closely related to the one employed by
Wößmann and West (2006). In particular, that study also uses variation in
educational inputs between grade-levels of a school in order to identify the
effects of particular educational inputs on student achievement. Our
approaches towards addressing within-school sorting differ, however. While
Wößmann and West use an IV strategy to address the potential endogeneity in
their main variable of interest (class size) arising from the non-random
placement of students across classes within schools, the present study uses a
strategy of aggregating the main variable of interest to a level such that the
effect estimates are not biased by such sorting patterns. (i.e. immigrant share is
measured at the school-grade level).

18
 The empirical strategy used will allow one to capture causal effects as
long as parents do not react to a high concentration of immigrant students in a
particular grade by moving their children to another school. The estimates
would be biased otherwise. The direction of the bias will depend on the types
of native students who move. If the natives who move are the higher-
achieving ones, then our estimates will be subject to a negative bias (i.e. we
will overestimate the negative effects, or alternatively, underestimate the
positive effects of immigrant students). On the other hand, if the natives who
move are the lower-achieving ones, then our estimates will be positively
biased.

4 RESULTS

<Insert Table 2 here>

Summary statistics of all variables included in the estimations are

presented for the samples of native students in Table 2. There are marked
differences in the share of immigrant children which native students are
exposed to in the 3 countries. On average, natives in the Australian sample
have the highest share of immigrant peers in their grade level in school whilst
natives in the Canadian sample have the lowest. The educational attainment of
native parents is also quite different across the 3 countries. Canadian natives
have the most educated parents, with 35.6% and 33.4% respectively having
fathers and mothers who completed university. U.S. natives come next, with
corresponding values of 27.2% and 23.7%. Australian natives have the lowest
proportion of university educated parents, with only 24.9% and 20.9%
respectively having fathers and mothers who completed university.

<Insert Table 3 here>

19
 Table 3 presents results from a variety of regressions showing the
estimated effect of the share of immigrant grade peers on the Math and
Science achievements of native Australian, Canadian, and U.S. students11. For
each subject, 3 columns of results are displayed. The first column shows
estimates from the most basic specification: an OLS specification that controls
only for grade fixed effects. The second column shows the estimates when
school fixed effects are included in the regressions. By including school fixed
effects, the coefficient on the immigrant share variable is identified on the
basis of variation in the exposure to immigrant peers across adjacent grades
within schools. Hence, the results in the second column are free from any bias
due to the non-random sorting of students across schools. Finally, the third
column shows estimates from the most comprehensive specification. This
specification controls for student12 and family background13 characteristics

11
The full set of coefficient estimates for the regressions in Table 3 can be
found in Appendix Tables A1 and A2.
12
Included in the set of student characteristics are dummy variables for the
student’s sex (male or female) and whether the student speaks the test
language frequently at home (this dummy is set to 1 if the student reports
always or almost always speaking the test language at home). It also includes a
continuous variable indicating the student’s age (measured in months).
13
Included in the set of family background characteristics are dummy
variables for the highest level of education attained by the student’s mother
and father, the number of books in the student’s home, and whether the
student lives with both parents. It also includes a continuous variable
indicating the total number of people living in the student’s home.
Observations may fall into 1 of 6 parental education categories: Father
(Mother) had at most primary education or less; Father (Mother) had some
secondary education; Father (Mother) completed secondary education; Father
(Mother) had some / completed vocational education; Father (Mother) had

20
(which may be correlated with both the share of immigrant grade peers and
subject test scores) in addition to grade and school fixed effects.

In all analyses, I present standard errors which are clustered at the school
level to correct for arbitrary correlations in individual error terms within
schools14.

For Australia, the coefficient estimate in column (1) of Table 3 indicates

that the Math achievement of native students is positively associated with the
share of immigrant peers in the grade. Specifically, a 10 percentage point
increase in the grade share of immigrants is associated with a 0.091 standard
deviation increase in the Math test score (this relationship is statistically
significant at the 5% level). The estimated effect increases slightly in
magnitude when school fixed effects are incorporated in the regression,
indicating that academically weaker Australian students tend to attend schools
with larger concentrations of immigrants. The estimate does not change much
when student and family background characteristics are additionally controlled
for. The estimate in column (3) suggests that a 10 percentage point rise in the
grade share of immigrants increases native Math test scores by about 0.094
standard deviations (significant at the 5% level). Is this effect sizeable? Recall
from table 2, that the mean and standard deviation of the share of immigrant
students in a grade for Australia are 10.0% and 8.9% respectively. So the
values imply that increasing the share of immigrants in a grade by 1 standard

some university education; Father (Mother) completed university education.

Observations may fall into 1 of 5 categories with regards to the number of
books owned: 0-10; 11-25; 26-100; 101-200; or more than 200 books at home.
14
Throughout the paper, all regressions are un-weighted. A comparison
reveals that weighting has very little effect on either the point estimates or
standard errors.

21
deviation leads to a 0.084 standard deviation improvement in test scores15. So
the estimated effects, while non-zero, are arguably small.

For Canada, the coefficient estimate in column (1) of Table 3 indicates

that the share of immigrant students in the grade is negatively related to
natives’ Math test scores (this relationship is statistically significant at the 5%
level). A 10 percentage point increase in the grade share of immigrants is
associated with a 0.044 standard deviation decrease in natives’ Math test
score. When across-school sorting is accounted for through inclusion of school
fixed effects, the coefficient on the immigrant share variable increases slightly
in magnitude. This indicates that, in Canada, academically more-able students
tend to sort towards schools with greater concentrations of immigrants. The
estimate further increases slightly in magnitude when individual and family
background attributes are controlled for. The estimate in column (3) suggests
that the share of immigrant grade peers has a negative and statistically
significant (at the 10% level) impact on the Math test scores of natives. A 10
percentage point increase in the share of immigrant grade peers is estimated to
lead to a 0.056 standard deviation decline in the Math test score of natives.

For the U.S., the coefficient estimate in column (1) indicates that native
Math achievement is negatively related to the share of immigrant peers in the
grade. However, the estimate in column (2) reveals that this association is

15
An alternative approach is to consider that the school-grade at the 50th
percentile of the immigrant share distribution had 7.4% immigrants while that
at the 75th percentile had 14.6% immigrants. This immigrant share difference
is 7.2 percentage points or 0.81 (=7.2/8.9) standard deviations. The estimate
implies that an increase in the share of immigrants which is sufficient to move
a school-grade from the 50th to the 75th percentile of the immigrant share
distribution would only result in a 0.068 (=0.81×0.084) standard deviation
improvement in Math test score.

22
actually an artefact of the non-random sorting behaviour of students across
schools. Once school differences are accounted for, the statistically significant
relationship between the variables disappears and the coefficient falls to
essentially zero. This result is unaltered when controls for individual and
family background characteristics are added (column (3)).

The above analysis suggests that while exposure to immigrant peers has
a positive impact on the Math achievement of Australian students, it has a
negative impact on the Math achievement of Canadian students. There is no
evidence that exposure to immigrant peers has an effect on the Math
achievement of U.S. students. These results are consistent with those from
non-parametric regressions.

The estimated peer effects of immigrants on natives’ Science test scores

are displayed in columns (4) to (6) of Table 3. As can be seen, for all 3
countries, the patterns described for Math achievement apply similarly to
Science. Exposure to immigrant peers is shown to have a positive effect on the
Science achievement of Australian students and a negative effect on the
Science achievement of Canadian students. Here again, exposure to immigrant
peers has no impact on the Science achievement of U.S. students. However,
the size of the estimated peer effects are much smaller for Science
achievement. In fact, none of the estimates from the most comprehensive
specification (column (6)) are statistically significant. Hence, at least for the
countries considered, immigrant peers appear to have a larger influence on the
Math achievements compared to the Science achievements of natives16.

16
None of the conclusions described for Table 3 changes even when log-linear
or log-log specifications are used.

23
4.1 Mechanisms Explaining Immigrant Peer Effects

In this section, I investigate the possible pathways leading to the

observed immigrant peer effects. Could the language deficiencies of Canadian
immigrants or the skill deficiencies of their parents be driving the negative
achievement effects which these immigrants seem to impose? Similarly,
would the positive achievement impacts of Australian immigrants be lower
had their parents been lower skilled or had they possessed lower proficiency in
the English language? To study whether the quality of immigrant students and
their parents (as measured by language ability and parental educational
attainment) affects the peer effects that they generate, I split immigrants into
dichotomous groups based on skill / adaptation as measured by (1) parental
education and (2) native language, before comparing how exposure to
immigrants differing on these characteristics affect student achievement. In all
cases, I control for individual and family background characteristics as well as
school and grade fixed effects.

4.1.1 Educational Attainment of Immigrant Parents

<Insert Table 4 here>

I begin by investigating whether the achievement effects from exposure

to immigrants with more-educated parents differ from those of exposure to
immigrants with less-educated parents. To do this, I create 2 variables: (1) the
share of immigrant grade peers with at least one parent who received some
education beyond the secondary level (relatively more-educated parents) and
(2) the share of immigrant grade peers with neither parent having any
education beyond the secondary level (relatively less-educated parents). I then
re-estimate equation (3), incorporating these as explanatory variables instead
of the single variable measuring the grade share of immigrants. If parental
education contributes positively to the peer effects generated by immigrants,

24
then the impact from immigrant peers with less-educated parents should be
more negative (or less positive) than the impact from immigrant peers with
more-educated parents. This is indeed what I find. Panel A of Table 4 shows
that for all 3 countries, the share of immigrant peers with less-educated parents
has a larger negative impact on natives’ Math and Science test scores than the
share of immigrant peers with more-educated parents (though effects are only
precisely estimated for Australia). Noteworthy are the findings for Australia.
For Math, exposure to immigrant peers with more-educated parents has a
statistically significant positive impact on the achievement of Australian
natives while for Science, exposure to immigrant peers with less-educated
parents has a statistically significant negative impact on their achievement.

4.1.2 Language Abilities of Immigrant Students

Next, I investigate whether spill-over effects may be due to the language

deficiencies suffered by immigrant students. I create 2 variables for this
purpose: (1) the grade share of immigrant peers that are non-native speakers of
the test language and (2) the grade share of immigrant peers that are native
speakers of the test language17. Equation (3) is then re-estimated with these
variables instead of the single explanatory variable measuring the grade share
of immigrants. The idea is that if language ability contributes positively to
peer effects, then any impact from those immigrant peers who are non-native
speakers should be more adverse than the impact from those who are native
language speakers. This is indeed what I find for Math achievement in
Canada. Panel B of Table 4 shows that while the estimated effect (-0.806) of
the share of non-native language speaking immigrant peers on natives’ Math

17
A student is classified as a native speaker of the test language if he/she
reports always or almost always speaking the language of test at home. He/she
is classified as a non-native speaker of the test language otherwise.

25
test scores is sizeable and statistically significantly negative at the 5% level,
the estimated effect of the share of native language speaking immigrant peers
is small and non-significant. This provides some support for the idea that the
overall negative peer effects documented for Canada are partly driven by the
language deficiencies of immigrant students. Little can be said for
achievements in the U.S. (and for Science achievement in Canada) because the
coefficients on the variables of interest do not reach statistical significance in
any of these estimations.

Interestingly, for Australia, we see that the positive peer effects of

immigrants come mainly from those immigrants that are non-native speakers.
The share of non-native speaking immigrant grade peers is shown to have a
sizeable positive impact on the Math achievement of Australian natives (the
estimated coefficient of 1.925 is statistically significant at the 1% level). In
comparison, the estimated achievement effect of the share of native language
speaking immigrant grade peers is much smaller and statistically non-
significant. While this seems to suggest that the language deficiencies of
immigrants enhance the learning experiences of peers in Australia, we must be
careful in making such causal statements. This is because immigrants who are
non-native speakers may differ systematically from those who are native
speakers. And these differences, which are not accounted for in this analysis,
may be creating the illusion that linguistic deficiency makes for better peers.
For example, an analysis using TIMSS 1995 reveals that non-native speaking
immigrant children in Australia actually outperform native-speaking
immigrants and native-born children in Math. While the average Math test
scores achieved by native speaking immigrant and native-born children are
0.091 and -0.006 respectively, it is 0.104 for non-native speaking immigrants.
Since non-native speaking immigrants seem to perform academically better
than their native speaking counterparts, it is not surprising that increased
exposure to them has a more positive effect on natives’ test scores. However,
it is not language deficiency per se that is responsible for the more positive

26
peer effect. Rather, it is the un-modelled difference in attributes possessed by
the two groups of students (i.e. native speaking and non-native speaking
immigrant students) which contributes spuriously to this result. Hence,
attaching a causal interpretation to the above result would be erroneous.

Overall, the results in sub-sections 4.1.1 – 4.1.2 are in agreement with

those from Hardoy and Schøne (2013) and suggest that peer quality is a
potential channel through which immigrants affect the achievements of peers.

5 DISCUSSION

It seems surprising that the peer effects of immigrants should be so

different in Australia and Canada despite the fact that both countries share
such similar immigration policies18 and histories and levels of economic
development. In both countries, immigrant students have parents that are, on
average, more highly skilled than those of natives. It is therefore puzzling why
only immigrant students in Australia should have positive impacts on the
achievements of peers (while immigrant students in Canada not only fail to
have positive impacts, but even have adverse impacts on peers’ achievement).
Furthermore, given that the parents of immigrants are relatively less-skilled
(than those of natives) in the U.S. than in Canada, one would expect
immigrant peer effects to be more adverse in the U.S. than in Canada. Yet, the
evidence suggests that U.S. immigrants actually have less adverse impacts on
the achievements of natives than Canadian immigrants. How can these
counterintuitive findings be explained?

18
Both countries have a skills-based admissions policy where admission is
based on economic criteria such as educational attainment, occupational
demand, language ability, and age.

27
 One potential explanation lies in the fact that how immigrant students
perform relative to natives (and therefore how immigrant students affect peers’
behaviours) is not only influenced by their socioeconomic characteristics but
also by institutional factors such as the way in which educational systems are
organised (Schneeweis, 2011; Cobb-Clark et al., 2012). This means that even
if immigrant students possess better socioeconomic characteristics than native
students, any positive influence that this may have on peer effects may still be
undone if countries adopt education policies or organise their educational
systems in ways that hinder the educational progress of migrants. Because
educational systems are inherently differently organised across countries, the
impacts from exposure to immigrant students can be different in different
countries even if the relative socioeconomic characteristics of immigrant
students in these countries are identical.

In fact, I do find some evidence that the way in which educational

systems are organised across the 3 countries are quite different. For instance,
responses from the TIMSS 1995 school administrator questionnaire reveal that
the degree to which schools have the autonomy to make curricular decisions
independently of external (i.e. district, regional, or national) entities is lowest
for Canada and highest for Australia19. While 94% and 86% of the student
population respectively in Australia and the U.S. had subject teachers who
possessed either “some” or “a lot” of influence in determining curriculum,
only 59% of Canadian students had teachers with such influence20. Similarly,
while 81% and 78% of the student population respectively in Australia and the
U.S. had principals who had either “some” or “a lot” of influence in
determining curriculum, only 65% of Canadian students had principals with

19
Data from TIMSS 1995, accessed from: https://nces.ed.gov/timss/idetimss/.
20
School principals were posed with the following question: “How much
influence do teachers of a subject have in determining curriculum?” They
could choose among 4 options: “none”, “a little”, “some”, or “a lot”.

28
such influence21. The extent to which schools are given the autonomy over
curriculum setting is important, because as pointed out by the OECD (2012),
“schools with more autonomy over curricular decisions may be better able to
cater to the particular needs of immigrant students”. They are therefore better
able to present immigrants with more educational opportunities. Because of
the lower autonomy that Canadian schools possess over curricular decisions,
schools there may not be as able as schools in either Australia or the U.S. to
cater to the specific needs of immigrant students. Hence, the peer effects
exerted by immigrants may be worse in Canada than in the other two
countries.

Two other potential explanations lie in age at arrival differences of

immigrant children and differences in the extent to which immigrant students
are exposed to the test language at home in these countries. Computations
using TIMSS 1995 indicate that the average age at arrival of 7th and 8th grade
immigrant children is lowest for Australia (5.596 years) and highest for
Canada (6.556 years)22. They also indicate that the share of immigrant
children who speak the test language frequently at home is highest for
Australia (64.3% of immigrant children in Australia report “always” or
“almost always” speaking the test language at home) and lowest for Canada
(only 53.5% of immigrant children in Canada report “always” or “almost
always” speaking the test language at home)23. Since age at arrival and the
extent to which students are exposed to the assessment language at home are

21
School principals were asked: “How much influence does the Principal/Head
of School have in determining curriculum?” Again, they could choose from 4
options: “none”, “a little”, “some”, or “a lot”.
22
In the U.S, the average age at arrival of 7th and 8th grade immigrant children
is 6.276 years.
23
In the U.S., 56.5% of 7th and 8th grade immigrant children report that they
either “always” or “almost always” speak the test language at home.

29
strong predictors of immigrant students’ academic performance (see Cobb-
Clark et al. (2012) and OECD (2012)), the older ages at which immigrant
children arrive in Canada (compared to Australia and the U.S.) and the lower
tendency for Canadian immigrant students to speak the test language at home
(compared to Australian and American immigrant students) may, to some
extent, account for the more adverse effects that these children have on peers’
educational achievements.

5.1 Immigrant Peer Effects and the Role of Educational Institutions

Is it possible to empirically validate the reasoning above? In other

words, is there any evidence to suggest that differences in the peer effects of
immigrants are related to differences in the way educational systems are
organised across countries? To assess whether and how institutional
arrangements affect the peer effects exerted by immigrants, I follow the
approach by Schütz et al. (2007) and exploit institutional variation at the
country level. Specifically, I pool observations from the 3 countries and
estimate the following educational production function:

𝑦𝑖𝑔𝑠𝑐 = 𝑿′𝑖𝑔𝑠𝑐 𝛽 + 𝛿𝐼𝑔𝑠𝑐 + 𝛾(𝐼𝑔𝑠𝑐 × 𝑅𝑐 ) + 𝛼𝑐 + 𝛼𝑠 + 𝛼𝑔 + 𝜀𝑖𝑔𝑠𝑐 (4)24

Where 𝑦𝑖𝑔𝑠𝑐 denotes the academic performance of native student 𝑖 from grade
𝑔 of school 𝑠 in country 𝑐. 𝑿𝑖𝑔𝑠𝑐 denotes a vector of individual and family
background characteristics for the student. 𝐼𝑔𝑠𝑐 denotes the share of immigrant
children in grade 𝑔 of school 𝑠 in country 𝑐. 𝑅𝑐 is a measure of the institution
of interest for country 𝑐 and so 𝐼𝑔𝑠𝑐 × 𝑅𝑐 represents an interaction between the

24
Institutional variables which vary at the country level cannot be included as
main explanatory variables in specification (4) because they would be
perfectly collinear with the country fixed effects.

30
grade share of immigrants and the institutional characteristic. 𝛼𝑐 , 𝛼𝑠 , and 𝛼𝑔
capture unobserved country, school, and grade heterogeneity respectively and
are accounted for in the regressions through a set of country, school, and grade
dummy variables. The coefficient of interest here is 𝛾 for it captures how the
effect of the share of immigrant peers on student achievement changes as the
institutional feature of interest varies across countries. If the estimate of this
coefficient is negative, then this indicates that the influence of the share of
immigrants on student achievement becomes smaller (i.e. becomes more
negative) as the indicator measuring the institutional feature increases. The
converse is true if the estimated coefficient is positive25.

I concentrate the analysis on 4 indicators measuring national education

systems. These measure features relating to schools’ autonomy over
curriculum and characteristics of the immigrant student population and include
specifically: (1) the share of students with subject teachers who possess
“some” or “a lot” of influence in determining curriculum, (2) the share of
students with principals who possess “some” or “a lot” of influence in
determining curriculum, (3) the average age at arrival of (7th and 8th grade)
immigrant children, and (4) the share of immigrant children who speak the test
language frequently at home. As with Schütz et al. (2007), all institutional
indicators are measured as averages at the country level. This aggregation is
done so as to avoid problems of within-country endogeneity.

<Insert Table 5 here>

25
More precisely, the marginal effect of the grade share of immigrant peers on
𝜕𝑦
student achievement is given by = 𝛿 + 𝛾𝑅 (subscripts dropped for
𝜕𝐼

convenience). Therefore, 𝛾 captures how the peer effects of immigrants

change as 𝑅 is varied across countries.

31
 Equation (4) is estimated separately for each institutional indicator of
interest. Table 5 reports the results from these estimations. For brevity, only
estimates of the coefficient on the interaction term between the grade share of
immigrant peers and each institutional indicator of interest (i.e. estimates of
the parameter 𝛾 in equation (4)) are reported. These tell us directly how the
effect of the share of immigrant peers on native students’ test scores changes
as the institutional indicator of interest is increased. Statistically significant
negative coefficients indicate that the peer effects of immigrants become
smaller (more negative) as the institutional indicator is increased. Statistically
significant positive coefficients indicate the opposite.

The results in Table 5 support the hypothesis that differences in

educational institutional arrangements are, at least, partly responsible for the
differential impacts imposed by immigrant students in the 3 countries. In
particular, the results for Math achievement indicate that the peer effects of
immigrants are more positive when the shares of teachers and principals in the
country who possess at least some influence in the setting of curriculum are
higher and when the share of immigrant students who speak the test language
frequently at home is higher. In addition, the peer effects of immigrants are
more negative in countries where immigrant children arrive at older ages. The
results for Math achievement apply largely to Science, though for the latter,
none of the institutional measures have a significant influence on the peer
effects exerted by immigrants.

The results from this exercise provide an explanation for why

immigrants in Australia and Canada have such different effects on peers
despite the adoption of similar immigration policies by both countries. The
fact that schools have less autonomy over curricular decisions and that
immigrant children tend to speak the test language less frequently and arrive at
older ages in Canada than in Australia possibly accounts for why immigrants
in Canada have more adverse achievement impacts. Perhaps more importantly,

32
they provide a possible explanation for the apparently conflicting findings
within the literature. As noted, while most studies have found immigrants to
have either adverse or no effects on natives’ achievements, a small number do
actually find beneficial effects. This analysis shows that the way immigrant
students affect their peers is ultimately influenced as well by the educational
institutions under which they operate. This is a novel finding in the literature
on immigrant peer effects.

6 CONCLUSION

Using an internationally comparable dataset which provides

achievement and migration information on students from 2 different grades of
each school, I estimate the effects that immigrant peers have on the
Mathematics and Science achievements of native students in 3 countries –
Australia, Canada, and the United States. To derive causal estimates of the
impacts of immigrant students on peers’ academic achievement, I relate
variations in the test scores of students over 2 adjacent grades of each school
to variations in the share of immigrant students over the 2 grades of the
school.

The results from this study indicate that immigrant students do have an
effect on native peers’ academic achievement. However, the effects of
immigrants are not the same across the 3 countries. While immigrant students
in Australia affect the academic achievements of natives positively, immigrant
students in Canada affect the academic achievements of natives adversely.
However, in both cases, it is only the impacts on natives’ Math achievements
which are statistically significant. The impacts on natives’ Science
achievements, though qualitatively similar, do not reach statistical
significance. There is no evidence that immigrants in the U.S. affect the
academic achievements of natives.

33
 The results from Canada demonstrate that even if immigrant children
have better-educated parents than natives, this does not guarantee that they
will have non-adverse effects on native peers’ academic outcomes. How
immigrants affect peers’ academic achievement depends also, and perhaps
more importantly, on the way a country’s educational system is organised.
However, within countries, improvements in immigrant quality (as measured
by language ability and parental education) may lessen the negative effects or
increase the positive effects that immigrants have on the educational
achievements of peers.

A unique implication of the findings is that the peer effects of

immigrants found in any one country cannot be simply extrapolated to another
even if these countries share similar immigration policies. Though immigrants
in two countries may share similar socioeconomic positions relative to natives,
institutional factors, such as the way in which the countries organise their
educational systems, could have an influence on how immigrant students
affect the behaviour of others. It would therefore be a mistake if researchers
attempt to generalise the peer effects found in one country to others on the
basis of similarities in the relative socioeconomic positions of migrants. A
reliable assessment of immigrant peer effects would need to be country and
context specific and would have to consider both the socioeconomic
characteristics of immigrants as well as the institutional arrangements under
which these students operate.

34
 ACKNOWLEDGMENTS

I am indebted to Steven Stillman, Tarja Viitanen, Murat Genç, and

Marta Tienda for their invaluable help, feedback, and suggestions. I also thank
Alicia Adsera, Frank D. Bean, Henry Farber, Kirabo Jackson, Alexandre Mas
and participants of the Population Association of America 2015 Annual
Meeting, the Princeton University Spring ’14 Labor Lunch Seminar, the
Rimini Conference in Economics and Finance 2014, and the Singapore
Economic Review Conference 2015 for insightful feedback. All errors are
mine.

35
 REFERENCES

Antecol, Heather, Cobb-Clark, Deborah A., and Trejo, Stephen J. (2003)

Immigration Policy and the Skills of Immigrants to Australia, Canada, and the
United States. Journal of Human Resources, 38(1), 192-218.

Betts, Julian R. and Fairlie, Robert W. (2003) Does Immigration Induce

‘Native Flight’ from Public Schools into Private Schools?. Journal of Public
Economics, 87(5-6), 987-1012.

Brunello, Giorgio and Rocco, Lorenzo. (2013) The Effect of Immigration on

the School Performance of Natives: Cross Country Evidence using PISA Test
Scores. Economics of Education Review, 32(C), 234-246.

Cobb-Clark, Deborah A., Sinning, Mathias, and Stillman, Steven. (2012)

Migrant Youths’ Educational Achievement: The Role of Institutions. Annals
of the American Academy of Political and Social Science, 643(1), 18-45.

Friesen, Jane and Krauth, Brian. (2011) Ethnic Enclaves in the Classroom.
Labour Economics, 18(5), 656-663.

Gould, Eric D., Lavy, Victor, and Paserman, Daniele M. (2009) Does
Immigration Affect the Long-Term Educational Outcomes of Natives? Quasi-
Experimental Evidence. The Economic Journal, 119(540), 1243-1269.

Hardoy, Inés and Schøne, Pal. (2013) Does the Clustering of Immigrant Peers
Affect the School Performance of Natives?. Journal of Human Capital, 7(1),
1-25.

Hermansen, Are Skeie and Birkelund, Gunn Elisabeth. (2015) The Impact of
Immigrant Classmates on Educational Outcomes. Social Forces, 94(2), 615-
646.

Hunt, Jennifer. (2012) The Impact of Immigration on the Educational

Attainment of Natives. NBER Working Paper No. 18047, May.

36
International Association for the Evaluation of Educational Achievement.
(n.d) Third International Mathematics and Science Study [Online] Available
from: http://www.iea.nl/timss_1995.html [Accessed 17 Nov 2012].

International Association for the Evaluation of Educational Achievement.

(n.d) International Data Explorer: Third International Mathematics and
Science Study [Online] Available from: https://nces.ed.gov/timss/idetimss/
[Accessed 5 February 2013].

Jensen, Peter and Rasmussen, Astrid Wurtz. (2011) The Effect of Immigrant
Concentration in Schools on Native and Immigrant Children’s Reading and
Math Skills. Economics of Education Review, 30(6), 1503-1515.

Neymotin, Florence. (2009) Immigration and its Effect on the College-Going

Outcomes of Natives. Economics of Education Review, 28(5), 538-550.

OECD. (2012) Untapped Skills: Realising the Potential of Immigrant

Students. OECD Publishing. [Online] Available from:
http://dx.doi.org/10.1787/9789264172470-en [Accessed 20 November 2013].

OECD. (n.d) International Data Explorer: Programme for International

Student Assessment [Online] Available from:
https://nces.ed.gov/surveys/pisa/idepisa/ [Accessed 21 November 2013].

Ohinata, Asako and van Ours, Jan C. (2013) How Immigrant Children Affect
the Academic Achievement of Native Dutch Children. The Economic Journal,
123(570), F308-F331.

Schneeweis, Nicole. (2011) Educational Institutions and the Integration of

Migrants. Journal of Population Economics, 24(4), 1281-1308.

Schneeweis, Nicole. (2013) Immigrant Concentration in Schools:

Consequences for Native and Migrant Students. IZA Discussion Paper No.
7230, February.

37
Schnepf, Sylke V. (2006) How Different are Immigrants? A Cross-Country
and Cross-Survey Analysis of Educational Achievement. In: Parsons, Craig A.
and Smeeding, Timothy M. (eds.) Immigration and the Transformation of
Europe. Cambridge, United Kingdom, Cambridge University Press. pp. 200-
234.

Schütz, Gabriela, West, Martin, and Wößmann, Ludger. (2007) School

Accountability, Autonomy, Choice, and the Equity of Student Achievement:
International Evidence from PISA 2003. OECD Directorate for Education
Working Paper No. 14, December.

Szulkin, Ryszard and Jonsson, Jan O. (2007) Ethnic Segregation and

Educational Outcomes in Swedish Comprehensive Schools. SULCIS Working
Paper No. 2007:2.

Wößmann, Ludger and West, Martin. (2006) Class-Size Effects in School

Systems around the World: Evidence from Between-Grade Variation in
TIMSS. European Economic Review, 50(3), 695-736.

38
 Figure 1: Non-Parametric Estimates of the Effect of School-Grade Immigrant Concentration on Natives’ Math Test Scores

Panel A: Relationship between Native Students' Math Test Scores and Share of Immigrant Children at the School-Grade Level

Australia Canada United States

2

2
Average Math Test Score of School-Grade (Natives)

Average Math Test Score of School-Grade (Natives)

1
1

0
0

-1
-1

-1

-2
-2

-2

-3
0 .2 .4 .6 .8 0 .2 .4 .6 .8 0 .2 .4 .6
Share of Immigrant Children in School-Grade Share of Immigrant Children in School-Grade Share of Immigrant Children in School-Grade
bandwidth = .4 bandwidth = .4 bandwidth = .4

Panel B: Relationship between Differences in the Share of Immigrant Children between 2 Grades of a School and Differences in Native Students' Math Test Scores

Australia Canada United States

Difference in Average Math Test Scores between Grades (Natives)

Difference in Average Math Test Scores between Grades (Natives)

2

2
2

1
1

0
0

-1
-1

-2
-1

-.2 -.1 0 .1 .2 .3 -.6 -.4 -.2 0 .2 .4 -.4 -.2 0 .2 .4

Difference in Share of Immigrant Children between Grades Difference in Share of Immigrant Children between Grades Difference in Share of Immigrant Children between Grades
bandwidth = .4 bandwidth = .4 bandwidth = .4

Notes: Data from the Third International Mathematics and Science Study of 1995. Sample weights used in all computations. Grade effects are partialed out semi-
39
parametrically prior to estimation. The bandwidth is set to 0.4 for all regressions.
 Figure 2: Non-Parametric Estimates of the Effect of School-Grade Immigrant Concentration on Natives’ Science Test Scores

Panel A: Relationship between Native Students' Science Test Scores and Share of Immigrant Children at the School-Grade Level

Australia Canada United States

2
Average Science Test Score of School-Grade (Natives)

Average Science Test Score of School-Grade (Natives)

1

1
0

0
-1

-1

-1
-2

-2

-2
-3

-3

-3
0 .2 .4 .6 .8 0 .2 .4 .6 .8 0 .2 .4 .6
Share of Immigrant Children in School-Grade Share of Immigrant Children in School-Grade Share of Immigrant Children in School-Grade
bandwidth = .4 bandwidth = .4 bandwidth = .4

Panel B: Relationship between Differences in the Share of Immigrant Children between 2 Grades of a School and Differences in Native Students' Science Test Scores

Australia Canada United States

Difference in Average Science Test Scores between Grades (Natives)

Difference in Average Science Test Scores between Grades (Natives)

2

2
2

1
1
1

0
0
0

-1
-1

-2
-1

-.2 -.1 0 .1 .2 .3 -.6 -.4 -.2 0 .2 .4 -.4 -.2 0 .2 .4

Difference in Share of Immigrant Children between Grades Difference in Share of Immigrant Children between Grades Difference in Share of Immigrant Children between Grades
bandwidth = .4 bandwidth = .4 bandwidth = .4

Notes: Data from the Third International Mathematics and Science Study of 1995. Sample weights used in all computations. Grade effects are partialed out semi-
parametrically prior to estimation. The bandwidth is set to 0.4 for all regressions. 40
 Table 1: Parental Education of Foreign-Born and Native-Born Children, by Country

Australia Canada United States

Foreign Native Foreign Native Foreign Native
Father Did Not Complete Secondary Education 0.225* 0.360 0.180* 0.248 0.176* 0.131
Father Had Some / Completed University
0.473* 0.299 0.557* 0.447 0.520 0.506
Education

Sample Size 1,401 11,177 1,326 14,906 938 9,836

Notes: Data from the Third International Mathematics and Science Study of 1995. Sample weights used in all computations so
that mean values are nationally representative. Values are proportions. * indicates that the mean value is significantly different
at the 5% level between native and immigrant students.

41
 Table 2: Summary Statistics of Variables Used in Estimations (for the Sample of Native Students)

Australia Canada U.S.

Variable Mean S.D. Mean S.D. Mean S.D.
Standardised Math Test Score -0.003 0.999 0.026 0.986 0.032 0.981
Standardised Science Test Score 0.015 0.991 0.043 0.979 0.052 0.978
Share of Immigrant Students in Grade 0.100 0.089 0.066 0.097 0.078 0.084
Share of Immigrant Students in Grade with More Educated Parents 0.066 0.069 0.048 0.079 0.054 0.066
Share of Immigrant Students in Grade with Less Educated Parents 0.034 0.044 0.017 0.044 0.019 0.036
Share of Immigrant Students in Grade who are Non-Native Speakers 0.032 0.055 0.030 0.065 0.037 0.057
Share of Immigrant Students in Grade who are Native Speakers 0.074 0.060 0.036 0.052 0.041 0.048
8th Grade (0/1) 0.558 0.497 0.503 0.500 0.646 0.478
Female (0/1) 0.529 0.499 0.495 0.500 0.510 0.500
Age 13.889 0.648 13.540 0.701 13.865 0.704
Speak Test Language at Home (0/1) 0.956 0.205 0.857 0.350 0.906 0.292
Living with Both Parents (0/1) 0.749 0.434 0.733 0.442 0.624 0.484
Household Size 4.660 1.272 4.508 1.208 4.746 1.741
Mother’s Highest Edu – Primary Education or Less (0/1) 0.020 0.139 0.065 0.246 0.030 0.170
Mother’s Highest Edu – Some Secondary Education (0/1) 0.385 0.487 0.141 0.348 0.110 0.313
Mother’s Highest Edu – Completed Secondary Education (0/1) 0.233 0.423 0.239 0.426 0.260 0.439
Mother’s Highest Edu – Some/Completed Vocational Edu (0/1) 0.091 0.288 0.104 0.305 0.092 0.289
Mother’s Highest Edu – Some University Education (0/1) 0.061 0.240 0.118 0.323 0.271 0.444
Mother Completed University Education (0/1) 0.209 0.407 0.334 0.471 0.237 0.425
Father’s Highest Edu – Primary Education or Less (0/1) 0.030 0.170 0.079 0.269 0.032 0.176
Father’s Highest Edu – Some Secondary Education (0/1) 0.325 0.468 0.164 0.370 0.118 0.323
Father’s Highest Edu – Completed Secondary Education (0/1) 0.173 0.378 0.192 0.394 0.246 0.430
Father’s Highest Edu – Some/Completed Vocational Edu (0/1) 0.174 0.379 0.113 0.316 0.121 0.326
Father’s Highest Edu – Some University Education (0/1) 0.050 0.219 0.097 0.296 0.211 0.408
Father Completed University Education (0/1) 0.249 0.432 0.356 0.479 0.272 0.445
Has 0-10 Books at Home (0/1) 0.020 0.141 0.036 0.185 0.086 0.281
Has 11-25 Books at Home (0/1) 0.056 0.230 0.100 0.300 0.131 0.338
Has 26-100 Books at Home (0/1) 0.231 0.421 0.282 0.450 0.284 0.451
Has 101-200 Books at Home (0/1) 0.260 0.439 0.244 0.430 0.203 0.403
Has more than 200 Books at Home (0/1) 0.433 0.496 0.338 0.473 0.296 0.456

Number of Students 11,177 14,906 9,836

Notes: Unweighted means and standard deviations of all variables included in the estimations, for the sample of native students.

42
 Table 3: Estimated Effect of the Share of Immigrant Peers on Native Students' Test Scores

Math Science
OLS OLS (w/School FE) OLS OLS (w/School FE)
(1) (2) (3) (4) (5) (6)
Controls for Controls for
Student Student
No Controls No Controls No Controls No Controls
& Family & Family
Characteristics Characteristics
Australia
Share of Immigrant Students in Grade 0.910** 1.095*** 0.935** 0.440 0.563 0.263
(0.383) (0.390) (0.375) (0.382) (0.404) (0.397)
Observations 11,177 11,177 7,488 11,177 11,177 7,488
R-Squared 0.025 0.323 0.385 0.026 0.280 0.371

Canada
Share of Immigrant Students in Grade -0.435** -0.494* -0.557* -0.226 -0.411 -0.205
(0.206) (0.284) (0.289) (0.221) (0.253) (0.264)
Observations 14,906 14,906 9,882 14,906 14,906 9,882
R-Squared 0.041 0.270 0.346 0.026 0.285 0.375

U.S.
Share of Immigrant Students in Grade -1.669*** -0.006 -0.035 -2.612*** -0.280 -0.089
(0.510) (0.530) (0.466) (0.497) (0.413) (0.369)
Observations 9,836 9,836 7,855 9,836 9,836 7,855
R-Squared 0.037 0.404 0.459 0.065 0.412 0.473

Notes: Grade fixed effects are included in all regressions. The student and family background variables control for students’ sex, age, whether or not they speak the test
language frequently at home, household size, mothers’ and fathers’ highest level of education, number of books at home, and whether or not they live with both parents.
Standard errors in parentheses are cluster-robust standard errors that allow for correlation in individual error terms within schools. *** p-value<0.01, ** p-value<0.05, * p-
value<0.1.

43
 Table 4: Estimated Effects of the Shares of Immigrant Peers in the High Skill Group and in the Low Skill Group on Native Students' Test Scores

Math Science
Skill/Adaptation Measured Share of Immigrant Peers in High Share of Immigrant Peers in Low Share of Immigrant Peers in High Share of Immigrant Peers in Low
by: Skill/Adapted Group Skill/Adapted Group Skill/Adapted Group Skill/Adapted Group
Panel A
Australia 1.225*** -0.662 0.564 -1.296**
(0.401) (0.558) (0.385) (0.639)
Canada Parental Education -0.242 -1.117 0.024 -0.847
(0.302) (0.723) (0.282) (0.560)
U.S. 0.211 -0.948 0.166 -0.656
(0.614) (0.970) (0.426) (0.757)
Panel B
Australia 0.214 1.925*** -0.103 0.784
(0.422) (0.724) (0.397) (0.644)
Canada Language Use -0.157 -0.806** -0.243 -0.128
(0.424) (0.351) (0.413) (0.316)
U.S. -0.170 0.072 -0.109 -0.070
(0.683) (0.738) (0.577) (0.602)

Notes: This table presents estimates from various regressions, showing the impact of immigrant peers in the high skill/adapted group and the impact of immigrant peers in
the low skill/adapted group. Skill/adaptation is measured by: (1) parental education and (2) language use. Immigrant students are defined as low skilled/adapted when skill
is measured by parental education and if neither of the student’s parents have an education above the secondary level. When adaptation is measured by language use,
immigrant students are low skilled/adapted if they are non-native speakers of the test language. In each case, the number of observations from Australia, Canada, and the
U.S. is 7,488, 9,882, and 7,855 respectively. All regressions control for the grade, sex, age, and household size of students, the number of books the student has at home,
whether the student frequently speaks the test language at home, whether the student lives with both parents, and the highest education attained by both parents. All
regressions also include school fixed effects. Standard errors in parentheses are cluster-robust standard errors that allow for correlation in individual error terms within
schools. *** p-value<0.01, ** p-value<0.05, * p-value<0.1.

44
 Table 5: Estimated Impacts of Various Institutional Arrangements on the Peer Effects Exerted by Immigrants

Math Science
Institutional Indicator 1
Share of Students with Teachers who Possess Influence in Curriculum Setting 3.712*** 1.488
(1.315) (1.267)
Full Set of Controls? Yes Yes
Observations 25,225 25,225
R-Squared 0.387 0.399
Institutional Indicator 2
Share of Students with Principals who Possess Influence in Curriculum Setting 7.829*** 3.098
(2.841) (2.711)
Full Set of Controls? Yes Yes
Observations 25,225 25,225
R-Squared 0.387 0.399
Institutional Indicator 3
Average Age at Arrival of Immigrant Students -1.582*** -0.721
(0.508) (0.526)
Full Set of Controls? Yes Yes
Observations 25,225 25,225
R-Squared 0.387 0.399
Institutional Indicator 4
Share of Immigrant Students who Speak the Test Language Frequently at Home 14.012*** 6.403
(4.513) (4.670)
Full Set of Controls? Yes Yes
Observations 25,225 25,225
R-Squared 0.387 0.399
Notes: The coefficients in this table represent estimates of the parameter in equation (4). Each is derived from a separate regression. Negative coefficients imply that the peer effects of
immigrants become smaller (more negative) as the institutional indicator is increased. Positive coefficients imply that the peer effects of immigrants become larger (more positive) as the
institutional indicator is increased. All regressions control for the grade, sex, age, and household size of students, the number of books the student has at home, whether the student
frequently speaks the test language at home, whether the student lives with both parents, and the highest education attained by both parents. All regressions also include country and school
fixed effects. Standard errors in parentheses are cluster-robust standard errors that allow for correlation in individual error terms within schools. *** p-value<0.01, ** p-value<0.05, * p-
value<0.1.

45
 Appendix Table A1: Full Set of Coefficient Estimates for Table 3 Regressions – Math Achievement

Australia Canada United States

OLS OLS (w/School FE) OLS OLS (w/School FE) OLS OLS (w/School FE)
Regressor (1) (2) (3) (4) (5) (6) (7) (8) (9)
Controls for Controls for Controls for
No Controls No Controls Student No Controls No Controls Student No Controls No Controls Student
and Family Charac and Family Charac and Family Charac
Share of Immigrant Students in Grade 0.910** 1.095*** 0.935** -0.435** -0.494* -0.557** -1.669*** -0.006 -0.035
(0.383) (0.390) (0.375) (0.206) (0.284) (0.289) (0.510) (0.530) (0.466)
8th Grade 0.261*** 0.320*** 0.451*** 0.389*** 0.374*** 0.583*** 0.272*** 0.299*** 0.523***
(0.034) (0.033) (0.042) (0.023) (0.023) (0.032) (0.050) (0.045) (0.044)
Female – – -0.077*** – – -0.061*** – – -0.156***
(0.022) (0.018) (0.018)
Age – – -0.130*** – – -0.230*** – – -0.249***
(0.026) (0.022) (0.022)
Speak Test Language at Home – – 0.291*** – – 0.121*** – – 0.123***
(0.057) (0.037) (0.043)
Living with Both Parents – – 0.096*** – – 0.137*** – – 0.094***
(0.024) (0.021) (0.019)
Household Size – – -0.025** – – -0.028*** – – -0.029***
(0.010) (0.007) (0.006)
Mother had Some Secondary Education – – 0.231*** – – 0.126** – – 0.123*
(0.078) (0.049) (0.072)
Mother Completed Secondary Education – – 0.169** – – 0.183*** – – 0.123*
(0.083) (0.052) (0.073)
Mother had Vocational Education – – 0.286*** – – 0.247*** – – 0.085
(0.086) (0.053) (0.075)
Mother had Some University Education – – 0.262*** – – 0.268*** – – 0.120
(0.086) (0.053) (0.074)

46
Mother Completed University Education – – 0.381*** – – 0.275*** – – 0.156**
(0.086) (0.049) (0.072)
Father had Some Secondary Education – – 0.156*** – – 0.179*** – – -0.071
(0.059) (0.044) (0.055)
Father Completed Secondary Education – – 0.159** – – 0.234*** – – -0.030
(0.061) (0.045) (0.052)
Father had Vocational Education – – 0.290*** – – 0.330*** – – 0.001
(0.058) (0.048) (0.056)
Father had Some University Education – – 0.205*** – – 0.293*** – – 0.042
(0.069) (0.046) (0.057)
Father Completed University Education – – 0.377*** – – 0.408*** – – 0.091
(0.062) (0.044) (0.058)
Has 11-25 Books at Home – – 0.223** – – 0.067 – – 0.082*
(0.089) (0.055) (0.042)
Has 26-100 Books at Home – – 0.497*** – – 0.288*** – – 0.319***
(0.081) (0.054) (0.040)
Has 101-200 Books at Home – – 0.644*** – – 0.397*** – – 0.446***
(0.081) (0.054) (0.041)
Has more than 200 Books at Home – – 0.762*** – – 0.417*** – – 0.506***
(0.080) (0.054) (0.043)
Observations 11,177 11,177 7,488 14,906 14,906 9,882 9,836 9,836 7,855
R-Squared 0.025 0.323 0.385 0.041 0.270 0.346 0.037 0.404 0.459

Notes: Grade fixed effects are included in all regressions. Standard errors in parentheses are cluster-robust standard errors that allow for correlation in individual error terms within schools. ***
denotes a coefficient significant at the 1% level, ** denotes a coefficient significant at the 5% level, * denotes a coefficient significant at the 10% level.

47
 Appendix Table A2: Full Set of Coefficient Estimates for Table 3 Regressions – Science Achievement

Australia Canada United States

OLS OLS (w/School FE) OLS OLS (w/School FE) OLS OLS (w/School FE)
Regressor (1) (2) (3) (4) (5) (6) (7) (8) (9)
Controls for Controls for Controls for
No Controls No Controls Student No Controls No Controls Student No Controls No Controls Student
and Family Charac and Family Charac and Family Charac
Share of Immigrant Students in Grade 0.440 0.563 0.263 -0.226 -0.411 -0.205 -2.612*** -0.280 -0.089
(0.382) (0.404) (0.397) (0.221) (0.253) (0.264) (0.497) (0.413) (0.369)
8th Grade 0.309*** 0.376*** 0.434*** 0.311*** 0.305*** 0.433*** 0.258*** 0.281*** 0.453***
(0.030) (0.029) (0.038) (0.022) (0.021) (0.028) (0.043) (0.037) (0.035)
Female – – -0.194*** – – -0.219*** – – -0.238***
(0.024) (0.017) (0.017)
Age – – -0.063** – – -0.144*** – – -0.188***
(0.026) (0.020) (0.017)
Speak Test Language at Home – – 0.422*** – – 0.235*** – – 0.268***
(0.057) (0.034) (0.040)
Living with Both Parents – – 0.084*** – – 0.110*** – – 0.055***
(0.024) (0.021) (0.019)
Household Size – – -0.045*** – – -0.064*** – – -0.042***
(0.009) (0.007) (0.006)
Mother had Some Secondary Education – – 0.276*** – – 0.164*** – – -0.063
(0.083) (0.047) (0.071)
Mother Completed Secondary Education – – 0.255*** – – 0.274*** – – -0.010
(0.083) (0.047) (0.069)
Mother had Vocational Education – – 0.333*** – – 0.299*** – – -0.037
(0.088) (0.050) (0.072)
Mother had Some University Education – – 0.291*** – – 0.313*** – – 0.041
(0.090) (0.049) (0.069)

48
Mother Completed University Education – – 0.444*** – – 0.337*** – – 0.035
(0.086) (0.045) (0.070)
Father had Some Secondary Education – – 0.157*** – – 0.081* – – 0.043
(0.060) (0.042) (0.055)
Father Completed Secondary Education – – 0.143** – – 0.113*** – – 0.084
(0.062) (0.043) (0.051)
Father had Vocational Education – – 0.318*** – – 0.283*** – – 0.116**
(0.057) (0.045) (0.053)
Father had Some University Education – – 0.225*** – – 0.154*** – – 0.144**
(0.073) (0.045) (0.056)
Father Completed University Education – – 0.372*** – – 0.295*** – – 0.194***
(0.062) (0.043) (0.057)
Has 11-25 Books at Home – – 0.141 – – 0.048 – – 0.115**
(0.093) (0.051) (0.045)
Has 26-100 Books at Home – – 0.398*** – – 0.279*** – – 0.343***
(0.087) (0.050) (0.042)
Has 101-200 Books at Home – – 0.566*** – – 0.460*** – – 0.477***
(0.084) (0.052) (0.045)
Has more than 200 Books at Home – – 0.723*** – – 0.534*** – – 0.553***
(0.087) (0.051) (0.046)
Observations 11,177 11,177 7,488 14,906 14,906 9,882 9,836 9,836 7,855
R-Squared 0.026 0.280 0.371 0.026 0.285 0.375 0.065 0.412 0.473

Notes: Grade fixed effects are included in all regressions. Standard errors in parentheses are cluster-robust standard errors that allow for correlation in individual error terms within schools. ***
denotes a coefficient significant at the 1% level, ** denotes a coefficient significant at the 5% level, * denotes a coefficient significant at the 10% level.

49