You are on page 1of 55

The Effect of Teacher’s Aides on Students with Special Needs

Simon Calmar Andersen Louise Beuchert Helena Skyt Nielsen


Aarhus University The Danish Center for Aarhus University
sca@ps.au.dk Social Science hnielsen@econ.au.dk
Research (VIVE)
lobe@vive.dk

February 25, 2023

Abstract:

This study investigates whether students with special education needs (SEN) in regular classrooms

benefit from increasing the educational support available. We exploit two randomized controlled

trials that added teacher’s aides to grade 6 classrooms combined with rich register data informative

about special education needs and school assignment. The teacher’s aides comprised three types:

A co-teacher with a teaching degree, a teaching assistant without a teaching degree, and a coach.

We find that the treatments resulted in 6–7 percentage points higher inclusion of SEN students in

regular classrooms and that SEN students gained academically. While the academic gains

evaporated over time, the treated students were able to stay in regular classrooms throughout

compulsory education and largely follow the same progression as their peers in the control group

when they transited to upper secondary education. Especially the co-teacher intervention was

effective in achieving these ends. A cost-benefit analysis suggests that teacher’s aides constitute a

cost-effective educational support for SEN students.

Keywords: segregated classrooms; special class; special school; randomized controlled trial.

JEL codes: I21; J14; J24.

We acknowledge financial support from TrygFonden and helpful comments from Beatrice
Schindler Rangvid and seminar participants at VIVE and Aarhus University. Andersen:
Department of Political Science, Aarhus University. Beuchert: The Danish Center for Social
Science Research (VIVE). Nielsen: Department of Economics and Business Economics,
Trygfonden’s Centre for Child Research and CIRRAU, Aarhus University. The authors
declare they have no relevant material financial interests that relate to the research described
in the paper.

Electronic copy available at: https://ssrn.com/abstract=4182253


1 Introduction

As the number of students enrolled in special education and the amount of money spent

on these services has grown (Schwartz et al., 2021), it becomes a pressing concern to

examine the most effective ways of providing education to this group of children. Some

research estimates positive causal effects of segregated special education (Hanushek et al.,

2002; Hurwitz et al., 2020) or special education support in regular classrooms (Ballis &

Heath, 2021; Schwartz et al., 2021) as opposed to treatment as usual in the regular school

system. However, school authorities spend two to four times as much money per student in

special education as per student in regular classrooms (Setren, 2021). Therefore, an

alternative to special education may be to include special education needs (SEN) students in

the regular classrooms while investing some of the saved resources in higher school quality in

terms of more teacher support. Besides potentially being a cost-effective strategy, the added

resources to regular classrooms may have positive spillovers on the peers. 1 Importantly, it

will also comply with the commitment to inclusive education programs in the Salamanca

Declaration 2 where SEN students are supposed to have access to regular schools and to the

extent possible be educated alongside their peers.

A recent study from Massachusetts indicated some positive effects of co-teaching on

students with disabilities in language arts in primary grades, but the effects were negative in

secondary grades, and there were no (or small positive) effects in math (Jones & Winters,

forthcoming). However, many questions relating to the effect of teacher’s aides on SEN

students remain to be answered. One type of question relates to which type of teacher’s aide

is most effective for included SEN students. Do SEN students learn more from untrained

teaching assistants that spend more time in the classroom (due to lower wages and less

preparation time) than from trained teachers that may provide higher quality teaching? Or is it

even more effective to have teachers with special qualifications coach the ordinary classroom

Electronic copy available at: https://ssrn.com/abstract=4182253


teachers a few hours every week? Another group of questions relate to the long-term impacts.

What happens to the SEN students if the teacher’s aides are no longer supported? Do the

students revert to special education classrooms, or are they able to stay in the classroom? And

if they stay, will that damage their completion of secondary education, or do they stay on par

with non-included SEN students? Answers to all these questions have direct implications for

the costs and benefits of using teacher’s aides to include SEN students.

To study these questions, we exploit two parallel randomized controlled trials

implemented in Denmark. The trials provided three different types of teacher’s aides to

classrooms: One treatment used teaching assistants without teaching degrees who spend 14.5

lessons per week in the classroom during nine months in grade 6. The second treatment used

co-teachers with teaching degrees who spend 10.5 lessons per week in the classroom. The

costs of these two interventions were the same (because collective agreements required that

educated teachers had more preparation time and higher salary). The third treatment used

coaches with expertise in special education needs (or similar qualifications) who were

allocated 2.5 hours per week per class to work with the teachers in the classrooms as well as

outside the classrooms. The third treatment cost about 14% of each of the other two

treatments (= 21,700/156,000 DKK per class). We examine (1) how these investments

attracted students from special education services, (2) how they affected the SEN students

during the intervention, (3) the downstream effects on the SEN students up to seven years

after the intervention ended, and (4) the short- and long-term costs and benefits of the

teacher’s aides.

The average treatment effects on regular students of the first two treatment arms have

been reported elsewhere (Andersen et al., 2020). However, due to apparent endogenous

sorting of SEN students into treated schools and across classrooms within treated schools, we

could not single out effects for SEN students as part of the original study. In the present

Electronic copy available at: https://ssrn.com/abstract=4182253


study, we have gathered administrative data on the allocation of SEN students in the years

prior to the intervention when treatment assignment could not possibly have affected student

allocation. In this way we can handle the issue of selection of SEN students into treatment.

Furthermore, we have not previously reported effects of the coaching treatment, which was a

parallel trial.

Our results first document that assignment of schools to these three forms of teacher’s

aide indeed increased the inclusion of SEN students in these classrooms. At schools assigned

a co-teacher or teaching assistant, the inclusion of SEN students increased significantly by 6–

7 percentage points compared to the inclusion at control schools, while the increased

inclusion was less and not significantly higher at schools assigned a coach compared to

controls schools.

Second, we find that SEN students benefitted from two of the treatments—the co-teacher

and the teaching assistant—during the period of the intervention. They increased their

reading test scores by 0.158 and 0.184 of a standard deviation, respectively, which is at par

with the effects of regular students reported earlier (Andersen et al., 2020). Third, after the

end of the intervention, the immediate effects evaporated. We found no effects in grade 8, no

effects on school leaving exams at the end of grade 9, and no effects on their choice of high

school versus other post-compulsory education. Also, we found no effects of the low-cost

coaching intervention.

Taken together these results indicate that assigning teacher’s aides to regular classrooms

attracts more SEN students, and as long as they are exposed to additional resources, SEN

students improve their reading skills just like regular students do. As reported elsewhere, the

two interventions at the same time improved test scores of the other students in the classroom

(Andersen et al., 2020), so we find no negative effects of the inclusion. When the two

interventions ended, the SEN students stayed in the regular classrooms with their new peers.

Electronic copy available at: https://ssrn.com/abstract=4182253


Their educational performance and attainment thereafter resemble that of SEN students in the

control group. The costs of a teaching assistant or a co-teacher for one classroom in our

intervention is similar to the additional costs of assigning 1.1 student to a special class rather

than a regular classroom for one school year. In comparison, we estimate that the teacher’s

aide treatments caused an additional 1.5 students to become included in general education per

class. So, the teacher’s aide policy was a cost-effective way of enhancing the inclusion of

special needs students. However, the low-cost coaching intervention was not enough to make

a difference.

Our study is closely related to the literature on the consequences of assignment of SEN

students to segregated classrooms as opposed to regular classrooms. Hanushek et al. (2002)

and Hurwitz et al. (2020) study the effects of special education programs on academic

achievement by exploiting that children move in and out of targeted programs. Both find that

the average special education program significantly increases academic achievement of

students with special needs entering the program. Also Schwartz et al. (2021) study the

intent-to-treat effects of special education classification and find positive effects on academic

achievement for students who are at some point assigned to special education after school

entry. In contrast to Hanushek et al. (2002) and Hurwitz et al. (2020), Schwartz et al. study

four types of special education ranging from integrated to more segregated settings: i)

support services such as counseling, physical therapy, or speech/language therapy, ii)

supplemental instruction, iii) integrated teaching support, and iv) self-contained services

where SEN students are educated in special classes. Similarly, Ballis and Heath (2021) study

the effects of reducing access to special education in Texas in 2005, which constrained the

special education caseload to 8.5%. They exploit district-specific reform exposure, and

results suggest that the marginal student benefits from special education services.

Electronic copy available at: https://ssrn.com/abstract=4182253


Our study is consistent with this line of research showing positive effects of the additional

resources invested in special education services. Even more relevant to our results, Setren

(2021) uses admission lotteries to charter schools to examine the effects of a high-performing

general education program (including high intensity tutoring, data-driven instruction, and

increased instructional time) on SEN students. She finds that the inclusion of SEN students in

regular—but high-quality—schools increases the likelihood of them graduating from two-

year colleges.

Our study also speaks to the literature on effects of school resources on academic

outcomes. These studies exploit plausibly exogenous variation to make inference on effects

of school funding (Hægeland et al., 2012; Holmlund et al., 2010; Hyman, 2017; Jackson et

al., 2016; Lafortune et al., 2018; Rothstein & Schanzenbach, 2022), staffing (Fredriksson et

al., 2013; Hemelt et al., 2021) or teacher quality (Chetty et al., 2014; Jackson, 2018). They

find favorable effects of school resources that are larger for students from low-income

families. Our findings are in line with these results and further suggest that availability of

abundant resources attracts SEN students, who—in particular—gain from extra school

resources.

In the following section, we present the institutional context and policy content. Next, we

present the experimental design of the trial and the data used. Afterwards, we present the

results—both in terms of inclusion, short-term effects at the end of the intervention period,

and long-term effects after the intervention ended. In the final section, we conclude the paper.

2 Inclusion of Special Needs Students in Danish Public Schools

At age four, the vast majority of Danish children are enrolled in some form of subsidized

public daycare. For the children in our study, education was compulsory from the calendar

year in which they turned seven 3 until completing grade 9.

Electronic copy available at: https://ssrn.com/abstract=4182253


At the end of grade 9 (around age 16), students sit compulsory exit exams, which must be

passed with satisfactory results in order to enroll in upper secondary education. 4 After grade

9, students may enroll directly in upper secondary education (if qualified), or they may attend

grade 10 (53% of the cohort did that in 2020) 5 either at the same public schools that also offer

grades 0 to 9 or at independent schools (i.e., private schools or boarding schools) and then

transit to upper secondary school afterwards (i.e., high school or vocational training).

Compulsory education, as well as most post-compulsory education, is free of charge at public

schools, whereas independent schools charge tuition fees. In grade 10, public as well as

independent schools have many degrees of freedom to focus on developing particular skills in

students and to teach according to their views and values. However, they are obligated to

provide a minimum mandatory coursework, a large independent written assignment, and

prepare students for transitioning to upper secondary schools (cf. Law of Independent and

Private Schools).

Public daycare, schools, and after-school programs are the responsibility of the

municipalities, who are also responsible for providing additional support for SEN children.

School authorities may offer support in terms of extra teacher resources or aides in regular

classrooms (Andersen et al., 2020), or they may use segregated settings in terms of special

classes or special schools (Egelund & Rangvid, 2014). Kindergarten class, public school, and

after-school programs are typically located in the same geographical location, and in many

cases special classes for children with special needs are integrated at these locations, whereas

special schools are located at other physical locations.

In 2011, 6% of Danish schoolchildren were educated in segregated settings in terms of

either special classes or special schools (henceforth “segregated education”). In 2015, the

number had decreased slightly to 5%, and financial incentives had been installed to bring the

number further down to enhance inclusive settings. 6 Segregated settings are targeted children

Electronic copy available at: https://ssrn.com/abstract=4182253


with particular needs (e.g., ADHD, ASD) and characterized by 1–2 professionals in a

classroom of 5–7 students (Mortensen et al., 2020).

Teaching and learning content in segregated settings is vastly different from that in

regular settings. According to Egelund and Rangvid (2014), who gathered information from a

survey of special school principals, the principals focus on stimulating the students’ personal

and social development rather than teaching them the course content. Furthermore, Egelund

and Rangvid (2014) report that 70% of students are taught below their age-appropriate grade-

level, and only one third of children are taught the full course package stipulated in the Law

of Public Schools. Most often the focus is on language arts, math, and English (Mortensen et

al., 2020). Hence, being taught in a segregated setting shifts focus from course content to

socio-emotional education.

Each municipality has an authority responsible for pedagogical-psychological counseling,

who assesses and refers children to segregated classrooms after referral from local schools or

public daycare institutions in close collaboration with parents and, if children have mental

health diagnoses, specialist physicians. 7 If the assessment concludes that the SEN student is

suited for a regular classroom, the pedagogical-psychological counseling assists the local

school in how to include the student. Referral to segregated classrooms is driven by three

main reasons (mentioned in order of importance): (i) the local public school cannot

accommodate the special needs of the student at the relevant grade level, (ii) the classmates

are negatively affected by noise or conflicts in rare cases leading to school refusal, or (iii) the

teacher cannot cope with the mental and physical stress from including the student in a

regular class. The type of segregated classroom the student is referred to depends on the

potential mental health diagnosis (Mortensen et al., 2020). A study by Junge (2022) using

school-cohort fixed effects documents that teachers’ referrals of students to special education

is influenced by the academic performance of the classmates.

Electronic copy available at: https://ssrn.com/abstract=4182253


In 2012, the Danish government implemented a reform of special needs education that

had an explicit goal of reducing the share of students in segregated education to 4% by 2015.

In practice, this was accomplished by redefining the characterization of SEN students and by

introducing economic incentives for schools to accommodate special needs students in

regular classrooms. Rangvid (2019) shows that the reform worked as intended and caused a

movement of some students already enrolled in segregated classrooms into regular

classrooms in particular during 2012 and 2013. Rangvid (2021) shows that this movement

back into regular classrooms is important for results on the grade 9 exit exam and successful

transitioning from lower to upper secondary school.

3 Experimental Methods

3.1 Interventions

Along with the reform of special needs education, the Danish government launched a

randomized controlled trial that should test the effect of three types of teacher’s aide. The

experiment should inform municipalities about the effects of these types of investments in

regular classrooms on regular and SEN students.

The intervention consisted of assigning a teacher’s aide to the grade 6 classroom in all of

the classes at a given school. The intervention was in place for roughly 85% of the school

year, from October 1, 2012 to June 20, 2013. The intervention and control groups were

announced on August 15, 2012, leaving the schools 1.5 months to search for and employ the

respective teacher’s aides.

Three types of teacher’s aide were part of the experiment:

(1) A co-teacher with a teaching degree (minimum 10.5 lessons per week per class)

(2) A teaching assistant without a teaching degree (minimum 14.5 lessons per week per

class)

Electronic copy available at: https://ssrn.com/abstract=4182253


(3) A coach with a teaching degree with expertise in special education or similar

qualifications (minimum 2.5 hours per week per class)

The first two interventions cost about 156,000 DKK per class. The third intervention cost

about 21,700 DKK per class or 14% of the other two interventions. By comparing the three

interventions, we are able to assess the relative costs and benefits of having more educated

teacher’s aides versus more time in the classroom with less educated aides.

As regards the first two interventions, those teacher’s aides were present in most language

arts and math lessons as well as a number of other lessons (Andersen et al., 2020). Out of

approximately 28 lessons per week, the co-teacher’s presence corresponds to 38%, and the

teaching assistant’s presence corresponds to 52%. When asked about their primary duties,

about 80% of the main teachers responded that the main duty of the aide was improving

academic achievement or wellbeing (avoiding conflict and the like), while 12% responded

that they supported students with special education needs and only 7–8% responded practical

support. While the co-teachers spent some time teaching students in various flexible ways

(inside/outside the classroom or smaller groups), the teaching assistants spent more time

handling conflicts and disruptive students. The overall picture, based on the main duties and

how they spent their time, supports that both types of teacher’s aide may have played an

important role in facilitating inclusion of more SEN students in regular classes with teacher’s

aides. As regards the coach intervention, they spent very little time in class and were to a

higher extent used for professional development of the class’ regular teacher, e.g.,

instructions on classroom management or how to teach students with reading disabilities.

Electronic copy available at: https://ssrn.com/abstract=4182253


3.2 Experimental Design and Participants

The experiment was run as two parallel, stratified, cluster-randomized trials with two

treatment arms (co-teacher and teaching assistant interventions) and one control arm in trial

1, and one treatment arm (coach intervention) and one control arm in trial 2 (see figure 1).

Figure 1: Two parallel trials with randomization at school level

Note: This figure illustrates the experimental design with two parallel trials.

The Ministry of Education invited all 98 municipalities in Denmark to sign up their

schools for the trial. A total of 68 municipalities applied for participation. Of these, 12

municipalities were selected for participation in trial 1, and 6 municipalities were selected for

trial 2. Municipalities were selected in order to create variation in municipality size, school

size, and geography while excluding municipalities engaged in school consolidations. This

selection process was chosen to create balance (with few municipalities, random selection

might have created more imbalance) and handle logistics of having coaches working at

multiple schools. It means that effects of the two trials may not be perfectly comparable.

However, since they were run within the same country at the same time with the same

outcome measures and without any self-selection into the trials, we assess that effect sizes are

at least as comparable as meta-analyses comparing intervention effects across multiple trials.

The number of enrolled schools were based on the available funding for the intervention,

which was financed by the ministry.

10

Electronic copy available at: https://ssrn.com/abstract=4182253


Within participating municipalities and to further increase statistical power, schools that

signed up were stratified (based on their predicted average achievement level for the relevant

grade 6 cohort) and randomly allocated to either treatment or control group within the strata.

Table A1 presents summary statistics by participating schools. The table illustrates that the

selection of municipalities and random allocation of schools into treatment and control

groups succeeded in creating balanced groups on a long list of observable school

characteristics.

3.3 Sample Selection

Figure 2: Illustration of time line, reallocation of students, and final estimation sample
(pooling trial 1 and 2)

Note: This figure pools sample size from trial 1 and 2. Table A2 in the appendix reports the corresponding
numbers by trial and treatment group.

In this paper, we are interested in the effect of the intervention on SEN students, i.e.,

students classified with special educational needs by school authorities. Figure 2 illustrates

how the final estimation sample of SEN students is constructed.

11

Electronic copy available at: https://ssrn.com/abstract=4182253


In total, 183 schools participated in the trials (105 schools in trial 1 and 78 schools in trial

2, see figure 1). Figure 2 shows that at the time of randomization, the treatment groups

included a total of 109 schools, and the control groups included 74 schools.

If a school was assigned to treatment, all regular grade 6 classrooms received the

treatment. This means that there was no within-school selection concerning which regular

classrooms received the teacher’s aides. Some schools also had special classes for students at

grade 6 (see figure 2), which would not receive the additional teaching resource/intervention.

Importantly, figure 2 shows that within participating schools some students shifted from

special classrooms to regular classrooms after the announcement of the randomization result

(i.e., between August 15 and October 1 when the intervention began). 8 The figure also shows

that some students entered the participating schools from outside. Such transitions happened

in both treatment and control group and are partly due to normal processes when students are

assessed ready for regular instruction and/or switch school (e.g., because their family moves

to another school district). However, in the following analyses, we test whether the

interventions increased the influx of SEN students to the treated classrooms. The intervention

was not and could not be blinded to participating schools and families, so the investment in

teacher’s aides in the regular classrooms may have caused more SEN students to be included

in the regular classrooms.

At the time when the intervention period began, the treatment groups included 4923

students in regular classrooms of which 547 students (11%) had special needs and an

additional 57 SEN students were moved from a segregated to a regular classroom resulting in

604 SEN students in total. In the control schools, the regular classes included 3297 students

of which 396 (12%) had special needs and an additional 14 SEN students were moved from

segregated to regular classrooms resulting in 410 SEN students. In total, 1014 students

constitute the initial sample of this study: 604 students with special needs in treated schools

12

Electronic copy available at: https://ssrn.com/abstract=4182253


and 410 with special needs at the control schools. In the initial sample, 660 were associated

with trial 1, and 354 were associated with trial 2.

During the intervention period, an additional 22 SEN students enrolled in the treated

schools, while 11 SEN students enrolled in the control schools. If we include students in

segregated classrooms as well, we have a total of 1284 students constituting the final

estimation sample: 772 SEN students in treated schools and 512 SEN students in control

schools (see figure 2). In the final estimation sample, 829 were associated with trial 1, and

455 were associated with trial 2.

In the analyses, we consider all SEN students in grade 6 in both regular and segregated

classes in treatment and control group schools in order to estimate the total effect of adding

additional teacher resources in regular classrooms for all students with special needs at the

school. Our final estimation sample thus includes SEN students in regular classrooms, SEN

students reallocated from segregated to regular classrooms, SEN students staying in

segregated classrooms, and incoming SEN students. This is to ensure that we compare SEN

students included due to the treatment with their best available counterfactual, i.e., SEN

students at control schools who stayed in or arrived at the segregated classrooms. Leaving out

the latter group would potentially underestimate the effect of the intervention because the

sample of students with special needs at the treated schools would include students with

worse disabilities than the sample at control schools. We find that treated schools included

more SEN students from other schools compared to control schools, however, we cannot

identify the corresponding additional students in the control group. Yet, analyses based only

on SEN students that were enrolled at participating schools before the beginning of the

intervention, generate results very similar to the results presented.

13

Electronic copy available at: https://ssrn.com/abstract=4182253


3.4 Data and Outcomes

To evaluate the effect of the interventions on the short- and long-term outcomes of the

students, we combine several data sources. Data collected as part of the randomized

controlled trial includes survey data on teacher’s aides used to characterize their background

and tasks in the classroom. By using personal identifiers (similar to social security numbers),

we are able to merge the trial data with administrative register data on both the assignment of

students (in special or regular classrooms) in the years prior to the initiation of the trial, and

on student data after the beginning of intervention including data on the students’ reading and

math skills and their continuation into post-compulsory education up to seven years after the

end of the intervention. Furthermore, we augment the data with socio-demographic

background characteristics of children and their parents from administrative registers. Below,

we explain how we define background variables, SEN status, and outcome variables.

Background variables. We use pre-treatment data on both school, student, and parent

characteristics to assess the balance of the experimental groups at baseline and to increase

precision in the statistical estimates. Appendix tables A1–A3 show balance across

experimental conditions at baseline.

Special education needs (SEN). We define SEN students based on the administrative

register on special education in public schools. The school administration reports to Statistics

Denmark yearly (in April) if they have assessed that students have special education needs

(e.g., due to behavioral or social disabilities), the primary cause (classified according to ICD-

10 H-classification), and the type of special education received and number of hours per

week. We define a student as having special education needs if the student was registered

with a cause in grades 4, 5, or 6. By including SEN assessments from the two school years

prior to the intervention, we include students that may have been reclassified in the

intervention year due to either the 2012-reform (that redefined and increased the requirements

14

Electronic copy available at: https://ssrn.com/abstract=4182253


to be classified as a SEN student) or the treatment assignment. In total, 71% (72%) of the

SEN students in our sample had a SEN assessment already in grade 4 and 71% (88%) in

grade 5 for trial 1 (2), respectively, see table 1. On average, 30% (24%) had learning

disabilities, 36% (31%) had reading disabilities (e.g. dyslexia), and 44% (48%) had other

disabilities (e.g. ADHD or autism) or causes of special needs in trial 1 (2).

Table 1: Students with SEN, sample means


Trial 1 Trial 2
Mean Mean
Boys 0.62 0.62
Psychiatric diagnosis 0.17 0.17
Timing and cause of SEN status
SEN assessment from grade 4 0.71 0.72
SEN assessment from grade 5 0.71 0.88
SEN assessment from grade 6 0.21 0.23
SEN cause: Learning disabilities 0.30 0.24
SEN cause: Reading disabilities 0.36 0.31
SEN cause: Other 0.44 0.48
Segregated education
Special class in grade 4 0.18 0.18
Special class in grade 5 0.17 0.20
Special class in grade 6 0.21 0.23
Special school in grades 4–6 0.01 0.02
Inflow compared to initial sample
Additional students 0.20 0.22
Observations 829 455
Note: Final estimation sample. Table summarizes sample characteristics as regards gender, psychiatric
diagnosis, and timing and cause of SEN assessment and segregated education. SEN causes are not mutually
exclusive and include all causes registered in grades 4–6. Segregated education entails special class or
special school, and the left out category is regular class possibly with some special support services.

Segregated education. To classify whether a student attends a special classroom or a

regular classroom, we again use the administrative register on special education in public

schools. This register indicates what type of classroom the student is enrolled in: a regular

class (possibly with special needs assistance), a special class, or a special school. Table 1

shows that the dominant fraction of SEN students are included in regular classes whereas

around 20% attend special class and only 1–2% attend a special school. For our empirical

15

Electronic copy available at: https://ssrn.com/abstract=4182253


analysis, we create an indicator variable for attending segregated education, which is equal to

one if a student attends a special class or a special school, and zero otherwise. We register

status of all students in our sample in grades 4 and 5 (prior to the intervention), in grade 6

(towards the end of the intervention), and in grades 7–10 (after the intervention). Segregated

education as of grades 4 and 5 is used to study inclusion due to the intervention; the

segregated education status as of grades 7–10 is used to study whether inclusion effects

persists.

Reading skills. We measure reading skills using national, standardized, computerized,

adaptive tests. The tests are adaptive in the sense that after a run-in period of five items, the

system selects the next item from an item bank based on the estimated skill level of the

students (calculated based on responses to previous items) and the estimated difficulty level

of the items. The students are tested simultaneously within three reading skill domains:

language comprehension, decoding, and text comprehension. Following (Beuchert &

Nandrup, 2018) and (Andersen et al., 2020), the final test score within each domain is

standardized (mean of 0 and standard deviation of 1) across the full population of public-

school students taking the test. To compute a composite total reading test score, we calculate

the average across the three standardized domain scores and re-standardize the total score.

The national reading tests were mandatory for public school students in Denmark in grades 2,

4, 6, and 8 at the time of the intervention. In our final estimation sample, about 80% of SEN

students took the reading test in grade 6.

Math skills. Math skills are measured using the same type of national test. They cover

three domains: numbers and algebra, geometry, and statistics and probability. We standardize

and summarize the math test scores in the same way as the reading test scores. Math tests

were mandatory for grades 3 and 6 at the time of the intervention. In our final estimation

sample, 76% took the math test in grade 6.

16

Electronic copy available at: https://ssrn.com/abstract=4182253


Grade 9 GPA. Oral examination in language arts, English, and science, and written

examination in language arts and math are mandatory. Students are graded on a 7-point scale

(-3, 0, 2, 4, 7, 10, 12), where 2 is the minimum grade required to pass a test. In order to pass

the grade 9 exam, the student needs a GPA of at least 2 in language arts and written math.

Language arts GPA is a weighted average of the results in oral exams (weight 1/2), reading

(weight 1/8), grammar (weight 1/8), and writing (weight 1/4). Math written GPA is a

weighted average of written tests in math with access to aids (weight 1/2) and without access

to aids (weight 1/2). Each test score is standardized with mean zero and a standard deviation

of one before calculating the total GPA. Finally, the GPA is standardized with mean 0 and

standard deviation of 1. The standardization is performed on the whole population of grade 9

exams in spring 2016 (including sickness exams). In our final estimation sample, 72% and

74% of SEN-students obtained a GPA in language arts and math, respectively.

On track versus falling behind. Student enrolment and progress through the Danish

education system is registered in a central register maintained by Statistics Denmark. The

register includes all types of public schools, independent schools (e.g., boarding schools and

private schools), special schools, and post-compulsory education institutions. The register

covers all grade levels. It is updated yearly on September 5 with information on student ID,

institution, education and grade level, and date of enrollment and end date (end due to drop

out or completion). Using this register, we define the following outcomes, which indicate

whether the students are on track or fall behind: enrollment in grade 9 in school year 2015/16

(including school type), enrollment in grade 10 in 2016/17 (including school type), and

enrollment in upper secondary education in school years 2016/17– 2019/20. Students on track

with peers are expected to enroll in grade 9 in August 2015 and in the optional grade 10 or

upper secondary education in August 2016 and later.

17

Electronic copy available at: https://ssrn.com/abstract=4182253


Highest completed education in year 2020. Finally, we measure the students’ highest

completed education at year end 2020. In year 2020, students turn 20 years old if they started

school according to legislated school starting age, and they turn 21 if they postponed school

start by one year or repeated a grade before the intervention in grade 6. At this point in time,

they are expected to have completed at least compulsory education and—if on track—high

school or vocational training as well (even if they have spent a year in the optional grade 10).

3.5 Statistical Models

To evaluate the effect of the interventions on SEN students, we use the following statistical

models depending on the trial evaluated:

Trial 1: 𝑌𝑌1,𝑖𝑖𝑖𝑖𝑖𝑖 = 𝛼𝛼 + 𝛽𝛽1 𝑇𝑇𝑇𝑇𝑠𝑠 + 𝛽𝛽2 𝑇𝑇𝑇𝑇𝑠𝑠 + 𝜕𝜕1 𝑋𝑋0,𝑖𝑖𝑖𝑖𝑖𝑖 + 𝛾𝛾𝑟𝑟 + 𝜀𝜀𝑖𝑖𝑖𝑖𝑖𝑖 (1)

Trial 2: 𝑌𝑌1,𝑖𝑖𝑖𝑖𝑖𝑖 = 𝛼𝛼 + 𝛽𝛽3 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑠𝑠 + 𝜕𝜕1 𝑋𝑋0,𝑖𝑖𝑖𝑖𝑖𝑖 + 𝛾𝛾𝑟𝑟 + 𝜀𝜀𝑖𝑖𝑖𝑖𝑖𝑖 (2)

where Y1 is the outcome variable for individual i, in school s, in strata r. The indicator

variable TC is equal to one if the school, s, is randomly assigned to the co-teacher

intervention, while TA is equal to one if the school, s, is randomly assigned to the

intervention with a teaching assistant. X0 includes student characteristics measured prior to

the intervention. In our main model specification, X includes student i’s pre-test scores in

reading (from grade 4) and math (from grade 3) to control for student ability. 𝛾𝛾𝑟𝑟 is a

randomization-strata fixed effect, and ε is the error term. We evaluate trial 1 and 2 separately.

Thus, for Trial 1, we use equation (1), and for Trial 2, we use equation (2) where the only

change is the inclusion of an indicator variable COACH equal to one if the school, s, is

randomly assigned to the coach intervention.

4 Results

In the results section, we examine (1) how these interventions attracted students from

segregated education, (2) how they affected the SEN students during the intervention, (3) the

18

Electronic copy available at: https://ssrn.com/abstract=4182253


downstream effects on the SEN students up to seven years after the intervention ended, and

(4) the financial costs and benefits of the treatments.

4.1 Effects on Inclusion of SEN Students

Table 2 reports the effect of the announcement of treatment assignment on enrollment of

students that were in segregated education prior to the intervention. 9 Model 1 in table 2

shows that treatment schools had more students enrolled in segregated education in grades 4–

6 prior to the intervention compared to the control schools. An additional 1 percentage point

in the co-teacher intervention, 0.8 percentage points in the teaching assistant intervention, and

0.6 percentage points (not statistically significant) in the coach intervention enrolled. The

same pattern is seen when limiting the sample to SEN students, but the point estimates are

scaled up to 7–8% instead of 0.6–1% (see table 2, model 2). This corresponds to 1.5 students

per classroom (or 3.7 students per school).

Table 2: Effects of the interventions on inclusion of SEN students in segregated


education prior to intervention (initial sample)
(1) (2)
Sample All students SEN students
Trial 1
Co-teacher w/ degree 0.010** 0.071**
(0.005) (0.029)
Teaching assistant w/o degree 0.008** 0.081***
(0.004) (0.029)

Observations 5,213 660


Adjusted R-squared 0.0455 0.214
Test: CO=TA (p-value) 0.639 0.811

Trial 2
Coach 0.006 0.077*
(0.005) (0.043)

Observations 3,022 354


Adjusted R-squared 0.063 0.287
Note: Initial sample including all students in the experiment. Each regression model includes students’ reading
and math pre-tests, indicator for missing pre-test, and randomization strata fixed effects. Robust standard errors
in parentheses are clustered at intervention school. *** p<0.01, ** p<0.05, * p<0.1.

19

Electronic copy available at: https://ssrn.com/abstract=4182253


4.2 Effects on SEN Students During the Interventions

In this section, we examine the effect on SEN students during the interventions. The final

analysis sample of SEN students includes all students in segregated classrooms at the school

to correct for the sample selection (see figure 2). The sample is generally disadvantaged (see

table A3): Students scored on average between 0.7–1.0 standard deviation (SD) below the

population means in math and reading in grades 3 and 4, only half of the students live with

both parents, among whom three fourths have no college degree, and 13% are diagnosed with

a psychiatric diagnosis such as ADHD. Table A3 tests sample balance across treatment arms

after extension of the sample to include all SEN students. The sample is well balanced

especially for trial 1. In total, 2/48 and 4/48 characteristics are significantly different across

treatment status at the 5%-level and the 10%-level for trial 1, and 2/28 and 3/28 for trial 2.

Table 3: Effects of interventions on academic outcomes, during intervention


(1) (2)
Reading Math
grade 6 grade 6
Trial 1
Co-teacher w/ degree 0.158** 0.196*
(0.074) (0.105)
Teaching assistant w/o degree 0.184** 0.054
(0.074) (0.106)

Observations 699 679


Adjusted R-squared 0.404 0.288
Test: CO=TA (p-value) 0.768 0.215
Control group mean -0.850 -0.635
Trial 2
Coach -0.102 0.024
(0.099) (0.120)

Observations 359 363


Adjusted R-squared 0.382 0.257
Control group mean -0.867 -0.650
Note: Final estimation sample. Each regression model includes reading and math pre-tests, indicator for missing
pre-test, and randomization strata fixed effect. Robust standard errors in parentheses are clustered on
intervention school. *** p<0.01, ** p<0.05, * p<0.1.

20

Electronic copy available at: https://ssrn.com/abstract=4182253


Table 3 reports the effects on the students’ academic achievements measured at the

end of the intervention period (grade 6). The table shows large and significant effects on

reading of the co-teacher (0.158 SD) and the teaching assistant intervention (0.184 SD), but

not of the coach intervention. Furthermore, the effect of the co-teacher intervention on math

is of the same magnitude but only borderline significant, while the other two treatments show

no statistically significant effects on math. In tables A4 and A5, columns 1–4, we gradually

add control variables and show that the effects are robust after controlling for reading pre-

test. In columns 5-6, we add interaction effects with ‘additional students’, i.e. SEN students

included after the intervention. The results suggest that effects are driven by SEN students

already in the class before the intervention. However, the point estimates on the interaction

effects suggest that also ‘additional students’ gain from the co-teacher and the teaching

assistant intervention but the sample is too small to make firm conclusions about this. In table

A6, we split by student gender, which reveals no systematic pattern.

The effects of the treatments on test scores are, of course, conditional on test

participation, which to some extent is reduced by treatment. Table A4 (upper panel) shows

that there are no significant differences in test participation across control and treatments in

trial 1. On the other hand, table A5 shows that the coach intervention induced significantly

fewer SEN students to participate in the test compared to the control group. This may indicate

that the coach assisted the main teacher in seeking exemption for taking the tests, or that they

reclassified students to make the intervention look better, or that although more students were

included in the regular class room, the low-intensity coach intervention could not support test

participation of the additional students. 10 The insignificant results on test scores for the coach

intervention should be interpreted in the light of higher non-participation. If exempted

students have systematically lower academic progression than average, this means that the

effects may be overstated. 11

21

Electronic copy available at: https://ssrn.com/abstract=4182253


4.3 Effects on SEN Students After the End of the Interventions

4.3.1 Effects on Inclusion

Table 4 reports results of the treatment on attending segregated classrooms in grades 7

through 10. The table shows no significant effects, which suggests that once included,

students are not referred back to segregated education when the extra teacher resources are

removed, and SEN students who were already in the regular classroom before the

intervention are also able to stay in the regular classroom afterwards to the same extent as in

control schools. We observe a tendency for students exposed to the coach intervention to be

more likely to be in segregated classrooms in grades 9 and 10, though.

Table 4: Effects of the interventions on attending special education, after intervention


Post intervention
Segregated Segregated Segregated Segregated
education education education education
grade 7 grade 8 grade 9 grade 10
Trial 1
Co-teacher w/ degree 0.007 0.010 -0.007 -0.014
(0.046) (0.044) (0.036) (0.031)
Teaching assistant w/o degree 0.056 0.064 0.050 0.029
(0.043) (0.042) (0.037) (0.032)

Observations 829 829 829 829


Adjusted R-squared 0.298 0.257 0.186 0.123
Test: CO=TA (p-value) 0.392 0.322 0.218 0.272
Control group mean 0.178 0.185 0.185 0.113
Trial 2
Coach 0.030 0.015 0.055 0.061*
(0.050) (0.048) (0.048) (0.034)

Observations 455 455 455 455


Adjusted R-squared 0.421 0.369 0.293 0.166
Control group mean 0.218 0.232 0.218 0.123
Note: Final estimation sample. Each regression model includes students’ reading and math pre-tests, indicator
for missing pre-test, and randomization strata fixed effect. Robust standard errors in parentheses are clustered at
intervention school. Segregated education is defined as special class or special school. *** p<0.01, ** p<0.05, *
p<0.1.

22

Electronic copy available at: https://ssrn.com/abstract=4182253


4.3.2 Effects on Academic Outcomes

Table 5 reveals that effects on academic achievement found during the intervention have

evaporated by grades 8 and 9, although the point estimate of the effect of the co-teacher on

reading in grade 8 is large (and imprecise). The point estimates for the effects of the coach

intervention on academic outcomes are negative, and some are even significant or borderline

significant.

4.3.3 Transition from lower secondary to upper secondary education

Evaluating the effects on educational attainment after the mandatory grade 9 is complicated

because there are many routes through the educational system in Denmark via optional grade

10 (in public or independent schools) to completion of upper secondary school at either

academic high schools (preparing for college) or vocational schools (preparing for skilled

jobs). In the following analyses, we first examine whether treatments make the SEN students

fall behind their peers in grades 7–9 in the first three years after the interventions. Then, we

examine if they are enrolled in (or have completed) upper secondary education within the

next three years. We summarize these analyses in a test of their highest completed level of

education seven years after the end of the interventions.

Figure 3 evaluates the effect of treatments on being on track vs. falling behind the first

three years after the intervention. Being on track means being enrolled in grade 7 in 2013/14,

grade 8 in 2014/15, and grade 9 in 2015/16 (the numbers behind the figure are reported in

appendix table A8). Students falling behind (i.e., not enrolled in grades 7–9 in the

corresponding years 2013/14–2015/16) are either repeating a grade or have dropped out of

the ordinary education system temporarily or permanently.

23

Electronic copy available at: https://ssrn.com/abstract=4182253


Table 5: Effects of interventions on academic outcomes, after intervention
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
Language arts, sub-tests
Reading Science Enroll in Exam Passed Math Language arts Oral Reading Grammar Writing
Grade 8 Grade 8 Grade 9 Participation Language arts (GPA) (GPA)
(0/1) (0/1) and math (0/1)
Trial 1
Co-teacher w/ degree 0.161 0.112 -0.034 -0.004 -0.032 -0.020 0.010 0.040 0.035 0.081 -0.081
(0.098) (0.084) (0.020) (0.038) (0.039) (0.082) (0.078) (0.086) (0.082) (0.074) (0.079)
Teaching assistant w/o degree 0.051 0.037 0.000 -0.064* -0.038 0.041 -0.042 -0.050 0.075 0.033 -0.083
(0.109) (0.080) (0.017) (0.037) (0.041) (0.084) (0.072) (0.079) (0.093) (0.072) (0.074)

Observations 622 600 829 829 829 643 643 643 643 643 643
Adjusted R-squared 0.297 0.254 0.082 0.186 0.161 0.283 0.224 0.095 0.227 0.269 0.175
Test: CO=TA (p-value) 0.352 0.464 0.079 0.162 0.904 0.529 0.522 0.295 0.709 0.547 0.989
Control group mean -0.857 -0.572 0.952 0.743 0.640 -0.734 -0.742 -0.532 -0.655 -0.879 -0.566

Trial 2
Coach -0.058 -0.156 0.012 -0.057 -0.055 -0.036 -0.247** -0.299*** -0.124 -0.159 -0.114
(0.100) (0.104) (0.023) (0.041) (0.061) (0.102) (0.099) (0.109) (0.086) (0.105) (0.091)

Observations 318 316 455 455 455 347 347 347 347 347 347
Adjusted R-squared 0.378 0.280 0.063 0.276 0.248 0.222 0.236 0.162 0.266 0.328 0.142
Control group mean -0.933 -0.559 0.941 0.732 0.595 -0.799 -0.790 -0.544 -0.662 -0.983 -0.621
Note: Final estimation sample. Regression model includes reading and math pre-test, indicator for missing pre-test, and randomization strata fixed effect. Robust standard
errors in parentheses are clustered on intervention school. 9th grade exit exams in spring 2016. Exam participation includes participation in at least mandatory language arts
and math exams. Language arts GPA is a weighted average of results in the oral, reading, grammar, and writing exam. Math GPA is a weighted average of written tests in
math with and without aids. Number of observations for trial 1 (2) are reported as 643 (347) to ensure anonymity even though <5 are missing for some outcomes. *** p<0.01,
** p<0.05, * p<0.1

24

Electronic copy available at: https://ssrn.com/abstract=4182253


Figure 3: Effect of interventions on enrollment, grades 7–9

Figure 3 shows that the treatments led students to fall slightly behind track immediately

after the intervention ended, and according to table A8, the effects are about 2–6 percentage

points and statistically significant for several school years. 12

In contrast, figure 4 shows that by the time students are supposed to be enrolled in upper

secondary education (in the school year 2017/18), the treated students in trial 1 have caught

up. The treatment effects are virtually zero and not statistically significant.

In the school year 2016/17, less than one third of the cohort is enrolled in upper

secondary education. Instead, students may either attend the optional grade 10 at a public

school or an independent school, a segregated school program, or be out of the school system.

Figure 4 shows that significantly fewer students exposed to the teaching assistant are enrolled

in upper secondary right after compulsory schooling compared to the two other student

groups. However, table A10 shows that students exposed to the teaching assistant

25

Electronic copy available at: https://ssrn.com/abstract=4182253


intervention are 9 percentage points more likely to be enrolled in the optional grade 10 in

2016/17 compared to the control group, which means they are still in school but delay

enrollment in upper secondary education. 13

Students treated by the teaching assistant and coach intervention are just as likely to be

enrolled in upper secondary education two years after compulsory schooling. Yet, figure 4

also illustrates that the students from the teaching assistant group are more likely to drop out

of upper secondary education in the following years compared to the control group. This may

be because although the students are induced to enroll in an upper-secondary education they

no longer receive the necessary support that they have become used to and drop out to a

larger extent that students from the control schools.

Figure 4: Effects of interventions on enrollment or completion of upper secondary


education

Trial 1 Trial 2
.9

.9
.8

.8
Proportion of students
.7

.7
Proportion of students
.6

.6
.5

.5
.4

.4
.3

.3
.2

.2

2016/17 2017/18 2018/19 2019/20


2016/17 2017/18 2018/19 2019/20

Control
Co-teacher Control

Teaching Assistant Coach

Note: Enrollment means the student is enrolled in upper secondary education in the given school year.
Completion of upper secondary education is obtained in December in a given school year. See table A11.

26

Electronic copy available at: https://ssrn.com/abstract=4182253


Table 6 reports effects on having obtained compulsory education by year end 2020

(model 1) and highest obtained education by year end 2020 (models 2–5). According to

model 1, the treatments do not affect the probability of completing compulsory education

(grade 9). The point estimates are small and statistically insignificant. This again implies that

the students who fell behind gradually caught up.

According to models 2–5, the students exposed to the co-teacher treatment have caught

up completely in terms of highest attained education at the end of year 2020, which is in line

with the favorable effects on academic achievement in grades 6–9 seen in tables 3 and 4. On

the contrary, the students exposed to the teaching assistant or the coach intervention, are 4–8

percentage points less likely to have completed high school compared to the control group in

2020. For the coach intervention, this is in line with the unfavorable effects on academic

achievement in grade 9 seen in table 5.

Overall, the results in this section show that even though the co-teacher intervention

attracted more SEN students, and even though the positive effects on academic outcomes

faded out, they still completed upper secondary education to the same extent as SEN students

in the control group. In the teaching assistant group, SEN students completed high school to a

lower extent (about 8 percentage points) than the control group. The effect of the coach on

highest completed education was statistically insignificant but of negative sign as for the

teaching assistant.

27

Electronic copy available at: https://ssrn.com/abstract=4182253


Table 6: Effects of interventions on highest completed education by year end 2020
(1) (2) (3) (4) (5)
Highest completed education
Completed Upper Upper Upper secondary:
Compulsory
compulsory secondary: secondary: Vocational
education
education Any High school training
Trial 1
Co-teacher w/ degree -0.001 0.003 -0.004 -0.005 0.001
(0.009) (0.040) (0.040) (0.040) (0.016)
Teaching assistant w/o degree 0.007 0.068* -0.061 -0.083** 0.022
(0.010) (0.039) (0.038) (0.034) (0.016)

Observations 829 829 829 829 829


Adjusted R-squared 0.028 0.103 0.113 0.110 0.028
Test: CO=TA (p-value) 0.472 0.100 0.154 0.033 0.233
Control group mean 0.986 0.582 0.404 0.356 0.048
Trial 2
Coach -0.001 0.044 -0.045 -0.039 -0.006
0.003 (0.055) (0.055) (0.052) (0.025)

Observations 455 455 455 455 455


Adjusted R-squared 0.010 0.084 0.097 0.130 0.067
Control group mean 0.995 0.677 0.318 0.255 0.064
Note: Final estimation sample. Each regression model includes students’ reading and math pre-tests, indicator
for missing pre-test, and randomization strata fixed effect. Robust standard errors in parentheses are clustered at
intervention school. *** p<0.01, ** p<0.05, * p<0.1.

4.4 Cost-benefit analysis

In table A12, we report cost measures to be used for a rough cost-benefit analysis. The

costs of a teaching assistant or a co-teacher for one classroom in our intervention is similar to

the additional costs of assigning 1.1 student to a special class rather than a regular classroom

for one school year. In comparison, we estimate that the teacher’s aide treatments caused an

additional 1.5 students to become included in regular education per class (see section 4.1).

So, the teacher’s aide policy has an immediate favorable effect on the costs of educating SEN

students.

The immediate effect compares the costs of a teacher’s aide for 75% of a school year with

the saved costs of including a SEN student in regular class for one school year. When the

included SEN student stays in the regular classrooms until the end of compulsory education

28

Electronic copy available at: https://ssrn.com/abstract=4182253


(i.e., four years)—as our results show they do—the cost-benefit ratio is obviously even more

favorable.

Figure 5 illustrates the costs of the intervention compared to benefits in terms of saved

costs from including additional SEN students. The figure shows (i) the accumulated saved

costs from including one student in a regular class who would otherwise have been in a

special class (dark grey) as well as (ii) the accumulated saved costs from including one

student who would otherwise have received special education support in a regular class (light

grey). While there is break-even during the first year (grade 6) if the intervention makes it

possible to include one student from a special class with no further special education support,

it takes another two years to recover the saved costs if the intervention “only” makes it

possible to get rid of special education support in regular class for one student. The

illustration does not take any other costs or benefits into account that may relate to

intervention effects or external effects on peers.

Figure 5: Cost-benefit illustration

Note: Accumulated (special class) is the accumulated saved cost assuming the student no longer has to attend a
special class after inclusion in the regular class. Accumulated (special support) assumes that the student no
longer has to receive special educational support in the regular class. The calculations only include the direct
costs and benefits (saved costs on other special education services); benefits in terms of saved costs on other
public services or in terms of higher academic skills are not included.

29

Electronic copy available at: https://ssrn.com/abstract=4182253


5 Conclusion

Existing research tends to show that inclusion of SEN students in regular classes has negative

impacts on both the SEN students themselves and their peers. However, these studies

typically compare regular classes without any additional teacher support to segregated

education, which is much more expensive due to a higher teacher-student ratio.

Our results document that assignment of a co-teacher, a teaching assistant, or a coach

increased the inclusion of SEN students in regular classrooms. The teacher’s aides

interventions increased the proportion of SEN students by around 7 percentage points

corresponding to 1.5 students per classroom (or 3.7 students per school). As long as the co-

teachers and the teaching assistants were present in the classroom, the SEN students (those

included after announcement of the intervention and those already included in the

classrooms) gained academically in terms of better reading skills and—in the case of co-

teachers—better math skills. The point estimates are in the same ballpark as the effects of the

treatments on regular students (Andersen et al., 2020). The much cheaper coach intervention

did not improve the academic skills.

After the end of the interventions, the academic edge of the treated students faded out.

However, they stayed included in the regular classrooms, and despite a little delay in the

progression through the educational system, by the age of 20, the co-teacher treated SEN

students had completed upper secondary education to the same extent as the SEN students in

the control group (where fewer SEN students were included in the first place). In the other

two interventions, about 6–7 percentage points fewer students had completed upper

secondary education by the age of 20. This corresponds to the 7 percentage points students

who were included in these classrooms in the first place as an effect of the intervention.

In sum, the results of these trials show that investing in regular classrooms using teacher’s

aides is a cost-effective way of including SEN students in the daily lives of peers without

30

Electronic copy available at: https://ssrn.com/abstract=4182253


special needs. Especially the co-teacher intervention had immediate positive effects on the

academic skills, and the students benefitting from this intervention stayed in the regular

school system throughout upper secondary education. By comparing the co-teacher

intervention to the much cheaper coach intervention, we can conclude that while the coach

intervention was enough to attract students from segregated education into the regular

classrooms, it was not strong enough to have a positive impact on their academic skills.

There are limitations to the study presented here. Especially, we can only speculate about

what the effects would have been if the co-teacher intervention had been maintained

throughout grades 7–9. The school system saved the additional costs of special class

education during these years, and if these resources had been invested in the SEN students,

they might have maintained the benefits of the intervention.

Another limitation relates to the selection of students from segregated classrooms into the

treated classrooms. We cannot distinguish statistically between effects on SEN students who

were already in inclusion before the trial and those who moved to inclusion as a result of the

treatment, because the latter group is too small. If the positive results of the present study

inspire school leaders to test the effect of similar interventions, it would be very relevant to

test more systematically what groups of SEN students may benefit the most from being

included in regular classrooms together with a co-teacher support.

With these caveats in mind, the results presented here points to a direction for future

research and policy development aiming at fulfilling the promises of the Salamanca

Declaration where SEN students are supposed to have access to regular schools and to the

extent possible be educated alongside their peers.

31

Electronic copy available at: https://ssrn.com/abstract=4182253


References

Andersen, S. C., Beuchert, L. V., Nielsen, H. S., & Thomsen, Mette K. (2020). The Effect of

Teacher’s Aides in the Classroom: Evidence from a Randomized Trial. Journal of the

European Economic Association, 18(1), 469–505. https://doi.org/10.1093/jeea/jvy048

Balestra, S., Eugster, B., & Liebert, H. (2022). Peers with special needs: Effects and Policies.

Review of Economics and Statistics, 104(3), 602–618.

Ballis, B., & Heath, K. (2021). The Long-Run Impacts of Special Education. American

Economic Journal: Economic Policy, 13(4), 72–111.

https://doi.org/10.1257/pol.20190603

Beuchert, L. V., & Nandrup, A. B. (2018). The Danish national tests at a glance.

Nationaløkonomisk Tidsskrift, 2018(1).

Chetty, R., Friedman, J. N., & Rockoff, J. E. (2014). Measuring the Impacts of Teachers II:

Teacher Value-Added and Student Outcomes in Adulthood. American Economic

Review, 104(9), 2633–2679. https://doi.org/10.1257/aer.104.9.2633

Egelund, N., & Rangvid, B. S. (2014). Determinants of quality of special schools: A survey of

special schools and pedagogical psychological counselling (In Danish: Faktorer af

betydning for kvalitet i specialskoler: En survey-kortlægning blandt specialskoler og

PPR). Danish School of Education, Aarhus University.

Fletcher, J. (2010). Spillover effects of inclusion of classmates with emotional problems on

test scores in early elementary school. Journal of Policy Analysis and Management,

29(1), 69–83. https://doi.org/10.1002/pam.20479

Fredriksson, P., Öckert, B., & Oosterbeek, H. (2013). Long-term effects of class size.

Quarterly Journal of Economics, 128(1), 249–285.

32

Electronic copy available at: https://ssrn.com/abstract=4182253


Hægeland, T., Raaum, O., & Salvanes, K. G. (2012). Pennies from heaven? Using exogenous

tax variation to identify effects of school resources on pupil achievement. Economics

of Education Review, 12, 601–614.

Hanushek, E. A., Kain, J. F., & Rivkin, S. G. (2002). Inferring Program Effects for Special

Populations: Does Special Education Raise Achievement for Students with

Disabilities? The Review of Economics and Statistics, 84(4), 584–599.

https://doi.org/10.1162/003465302760556431

Hemelt, S. W., Ladd, H. F., & Clifton, C. R. (2021). Do Teacher Assistants Improve Student

Outcomes? Evidence From School Funding Cutbacks in North Carolina. Educational

Evaluation and Policy Analysis, 43(2), 280–304.

Holmlund, H., McNally, S., & Viarengo, M. (2010). Does money matter for schools?

Economics of Education Review, 29, 1154–1164.

Hurwitz, S., Perry, B., Cohen, E. D., & Skiba, R. (2020). Special Education and

Individualized Academic Growth: A Longitudinal Assessment of Outcomes for

Students With Disabilities. American Educational Research Journal, 57(2), 576–611.

Hyman, J. (2017). Does Money Matter in the Long Run? Effects of School Spending on

Educational Attainment. American Economic Journal: Economic Policy, 9(4), 256–

280.

Jackson, K. (2018). What Do Test Scores Miss? The Importance of Teacher Effects on Non–

Test Score Outcomes. Journal of Political Economy, 126(5), 2072–2107.

https://doi.org/10.1086/699018

Jackson, K., Persico, C., & Johnson, R. (2016). The Effects of School Spending on

Educational & Economic Outcomes: Evidence from School Finance Reforms.

Quarterly Journal of Economics, 131(1), 157–218.

33

Electronic copy available at: https://ssrn.com/abstract=4182253


Jones, N., & Winters, M. A. (forthcoming). Are Two Teachers Better Than One? The Effect

of Co-Teaching on Students With and Without Disabilities. Journal of Human

Resources.

Junge, S. Y. (2022). Administrative Groupings and Equality in Public Service Provision.

Journal of Public Administration Research and Theory, 32(2), 252–268.

https://doi.org/10.1093/jopart/muab031

Justice, L. M., Logan, J. A. R., Lin, T.-J., & Kaderavek, J. N. (2014). Peer Effects in Early

Childhood Education: Testing the Assumptions of Special-Education Inclusion.

Psychological Science, 25(9), 1722–1729. https://doi.org/10.1177/0956797614538978

Lafortune, J., Rothstein, J., & Schanzenbach, D. W. (2018). School Finance Reform and the

Distribution of Student Achievement. American Economic Journal: Applied

Economics, 10(2), 1–26.

Lynggaard, M., & Rangvid, B. S. (2014). Specialskolelevers resultater ved skolegangens

afslutning og 5 år senere. Danish School of Education, Aarhus University.

Ministry of Finance. (2010). Special education support in public schools—Ways to better

organization and management. [In Danish: Specialundervisning i folkeskolen – veje

til en bedre organisering og styring.]. Local Government Denmark, Ministry of

Education and Ministry of Finance.

Mortensen, N. P., Andreasen, A. G., & Tegtmejer, T. (2020). Educational results and

educational patterns for children and youth with disabilities. (In

Danish:Uddannelsesresultater og -mønstre for børn og unge med

funktionsnedsættelser). The Danish Centre for Social Science Research (VIVE).

Rangvid, B. S. (2019). Returning special education students to regular classrooms:

Externalities on peers’ reading scores. Economics of Education Review, 68, 13–22.

https://doi.org/10.1016/j.econedurev.2018.11.002

34

Electronic copy available at: https://ssrn.com/abstract=4182253


Rangvid, B. S. (2021). Special educational needs placement in lower secondary education:

The impact of segregated vs. Mainstream placement on post-16 outcomes. Education

Economics, 30(4), 399–425. https://doi.org/10.1080/09645292.2021.1995850

Rothstein, J., & Schanzenbach, D. W. (2022). Does Money Still Matter? Attainment and

Earnings Effects of Post-1990 School Finance Reforms. Journal of Labor Economics,

40(S1). https://doi.org/10.1086/717934

Schwartz, A. E., Hopkins, B. G., & Stiefel, L. (2021). The Effects of Special Education on

the Academic Performance of Students with Learning Disabilities. Journal of Policy

Analysis and Management, 40(2), 480–520. https://doi.org/10.1002/pam.22282

Setren, E. (2021). Targeted vs. General Education Investments Evidence from Special

Education and English Language Learners in Boston Charter Schools. Journal of

Human Resources, 56(4), 1073–1112. https://doi.org/10.3368/jhr.56.4.0219-10040R2

35

Electronic copy available at: https://ssrn.com/abstract=4182253


Appendix (for online publication)

Table A1: Baseline data, school level (initial sample)


Trial 1 Trial 2
Control Co-teacher Teaching assistant Control Coach
Mean coef. s.e. p-val coef. s.e. p-val Mean coef. s.e. p-val
No. Students per school 51.11 0.71 3.82 0.85 -5.11 4.00 0.20 39.05 -0.62 4.22 0.88
No. Classes per school 2.46 -0.06 0.19 0.76 -0.20 0.19 0.30 1.90 0.00 0.17 1.00
No. SEN students per school 6.94 -0.34 1.07 0.75 -1.60 1.04 0.13 4.31 0.46 0.99 0.64
No. Students per class 21.30 0.29 0.98 0.77 -1.34 0.93 0.15 20.14 -1.39 1.13 0.22
No. SEN students per class 2.93 -0.24 0.40 0.56 -0.49 0.40 0.23 2.38 -0.36 0.39 0.36
Student characteristics:
Reading score grade 4 -0.03 0.04 0.06 0.50 0.03 0.07 0.62 -0.01 0.01 0.06 0.82
Math score grade 3 -0.01 0.01 0.06 0.86 0.06 0.06 0.34 0.02 0.06 0.09 0.50
Boys 0.51 0.00 0.02 0.98 0.02 0.02 0.48 0.49 0.02 0.03 0.56
Living with both parents 0.62 0.03 0.02 0.11 0.04 0.02 0.04 0.67 -0.02 0.03 0.48
Non-western origin 0.13 0.01 0.03 0.68 0.00 0.03 0.97 0.10 -0.01 0.01 0.49
ADHD and similar diagnoses 0.04 0.00 0.01 0.67 0.01 0.01 0.33 0.04 0.01 0.01 0.53
Special education needs 0.14 0.00 0.02 0.98 -0.01 0.02 0.81 0.12 -0.02 0.02 0.28
Segregated class in grades 4–6 0.01 0.01 0.01 0.16 0.01 0.01 0.44 0.00 0.00 0.00 0.46
Segregated school in grades 4–6 0.00 0.00 0.00 0.36 0.00 0.00 0.15 0.00 0.00 0.00 0.36
Parental characteristics:
Mothers with F-diagnoses 0.12 -0.01 0.01 0.62 0.00 0.01 0.84 0.11 0.02 0.01 0.20
Fathers with F-diagnoses 0.09 0.02 0.01 0.08 0.02 0.01 0.22 0.09 0.00 0.01 0.89
Mothers employed 0.77 -0.01 0.02 0.83 -0.01 0.03 0.71 0.80 -0.05 0.03 0.13

36

Electronic copy available at: https://ssrn.com/abstract=4182253


Fathers employed 0.80 -0.01 0.02 0.80 0.00 0.02 0.89 0.82 -0.02 0.03 0.40
Both parents low educated 0.12 0.01 0.02 0.76 -0.02 0.02 0.25 0.10 0.00 0.01 0.79
Both parents no college degrees 0.62 -0.01 0.03 0.81 -0.03 0.03 0.35 0.68 -0.05 0.03 0.15
At least one parent holds a college degree 0.38 0.01 0.03 0.81 0.03 0.03 0.35 0.32 0.05 0.03 0.15
Note: Initial sample. Table reports the results from regressing each baseline covariate on the set of treatment indicators including randomization strata fixed effect and
clustering at the school level. Number of observations: 105 in trial 1 and 78 in trial 2. School level. *** p<0.01; ** p<0.05; *p<0.1.

37

Electronic copy available at: https://ssrn.com/abstract=4182253


Table A2: Sample selection and sample size by trial and treatment group
Sample size
Total Trial 1 and 2
Trial 1 Trial 2
Date Experimental timeline and sample selection (Figure 2)
Co- Teacher
Control Total Coach Control Total Treatment Control Total
teacher assistant
03.08.2012 Randomization at school level
Number of schools 35 35 35 105 39 39 78 109 74 183
Number of students in regular classrooms 1793 1593 1782 5168 1480 1516 2996 4866 3298 8164
- students w/o SEN 1583 1423 1547 4553 1313 1355 2668 4319 2902 7221
- students w SEN 210 170 235 615 167 161 328 547 396 943
Number of students in special classrooms 68 79 51 198 58 54 112 205 105 310
13.08.2012 School year begins
15.08.2012 Randomization result announced
Inflow of students w/SEN from special classroom to regular
classroom 21 17 7 45 19 7 26 57 14 71
31.08.2012 Deadline reporting students
Number of students in regular classrooms (INITIAL SAMPLE) 1814 1610 1789 5213 1499 1523 3022 4923 3312 8235
- students w/o SEN 1583 1423 1547 4553 1313 1355 2668 4319 2902 7221
- student w/SEN 231 187 242 660 186 168 354 604 410 1014
Number of students in special classrooms 47 62 44 153 39 47 86 148 91 239
Total number of students w/SEN 278 249 286 813 225 215 440 752 501 1253
October 2012
01.10.2012 Intervention begins
Oct - June Number of inflow students from other schools
- to regular classrooms, students w/o SEN 25 17 35 77 46 19 65 88 54 142
- to regular classrooms, students w/SEN 7 5 6 18 10 5 15 20 11 31
June 2013
28.06.2013 Intervention ends
FINAL ESTIMATION SAMPLE 1893 1692 1874 5459 1594 1594 3188 5179 3468 8647
- students w/o SEN 1608 1440 1582 4630 1359 1374 2733 4407 2956 7363
- students w/SEN 285 252 292 829 235 220 455 772 512 1284

38

Electronic copy available at: https://ssrn.com/abstract=4182253


Note: Initial sample size includes students with missing ID (28 students in trial 1.33 students in trial 2); we assume they have no SEN. Final estimation sample includes
incoming students during the school year (minus students outflowing) as well as students in special education classrooms. SEN information obtained from register data from
grades 4–6. SEN: Special education needs.

39

Electronic copy available at: https://ssrn.com/abstract=4182253


Table A3: Students’ baseline data and balance test (final estimation sample)
Trial 1 Trial 2

Control Co-teacher Teaching assistant Control Coach


Mean Coef. s.e. p-val. Coef. s.e. p-val. Mean Coef. s.e. p-val.
Pre-tests:
Reading grade 4 -0.81 -0.09 0.11 0.45 -0.03 0.13 0.85 -0.96 -0.08 0.10 0.44
Math grade 3 -0.69 -0.05 0.09 0.55 0.03 0.11 0.78 -0.72 0.11 0.14 0.43
Took reading test 0.82 0.01 0.04 0.79 -0.02 0.05 0.66 0.79 -0.05 0.05 0.33
Took math test 0.76 0.00 0.05 0.97 -0.02 0.06 0.79 0.82 -0.11 0.06 0.07
Student characteristics:
Boy 0.60 -0.01 0.03 0.69 0.07 0.03 0.00 0.62 0.04 0.05 0.46
ADHD and similar diagnoses 0.17 -0.02 0.03 0.58 0.04 0.04 0.24 0.16 0.01 0.04 0.75
Special needs: Learning disabilities 0.27 -0.02 0.05 0.67 0.14 0.06 0.03 0.25 0.09 0.05 0.11
Special needs: Reading disabilities 0.37 0.01 0.06 0.93 -0.13 0.07 0.06 0.35 -0.08 0.07 0.23
Special needs: Other disabilities 0.46 0.00 0.06 0.94 -0.04 0.07 0.55 0.45 0.00 0.05 0.97
Non-Western origin 0.15 0.01 0.03 0.75 0.01 0.03 0.85 0.11 0.01 0.03 0.76
Low birthweight 0.09 0.00 0.02 0.93 -0.01 0.02 0.80 0.11 -0.04 0.02 0.10
Living with both parents 0.53 0.05 0.04 0.26 -0.01 0.04 0.88 0.58 -0.08 0.05 0.12
Mother’s characteristics:
ADHD or similar diagnosis 0.14 0.04 0.03 0.26 0.03 0.03 0.26 0.15 0.01 0.04 0.77
Age at childbirth 23.19 0.25 1.58 0.88 -3.58 1.92 0.06 21.98 -1.69 1.80 0.35
Single 0.30 0.01 0.03 0.69 0.00 0.03 0.98 0.28 0.10 0.05 0.04
No college education 0.76 -0.02 0.03 0.39 0.02 0.03 0.54 0.78 0.01 0.04 0.77
Employed 0.69 -0.01 0.04 0.78 0.01 0.04 0.87 0.68 0.00 0.04 0.93
Unemployed 0.08 -0.01 0.02 0.58 -0.02 0.02 0.46 0.07 0.02 0.03 0.59
Inactive in labor market 0.22 0.03 0.03 0.42 -0.02 0.04 0.63 0.23 -0.02 0.02 0.39
Income 209,688 -3,771 11,263 0.74 9,510 11,470 0.41 191,225 -2,645 18,292 0.89
Father’s characteristics:
ADHD or similar diagnosis 0.13 0.04 0.03 0.11 0.01 0.02 0.74 0.15 0.00 0.04 0.93
Age at childbirth 24.20 0.10 1.60 0.95 -2.29 1.81 0.21 23.74 -3.61 2.17 0.10
Single 0.29 -0.01 0.03 0.72 -0.02 0.03 0.49 0.30 0.04 0.05 0.43
No college education 0.78 0.01 0.03 0.61 0.04 0.03 0.29 0.86 -0.13 0.03 0.00
Employed 0.73 0.02 0.04 0.63 0.04 0.04 0.26 0.76 -0.04 0.04 0.37
Unemployed 0.05 0.00 0.02 0.79 0.00 0.02 0.89 0.06 -0.05 0.02 0.02
Inactive in labor market 0.15 0.01 0.03 0.84 -0.01 0.04 0.88 0.13 0.04 0.03 0.29
Income (DKK) 270,202 13,138 18,390 0.48 21,235 17,643 0.23 259,741 3,450 20,316 0.87
No. students 292 285 252 220 235

40

Electronic copy available at: https://ssrn.com/abstract=4182253


Notes: Final estimation sample. Table reports the results from regressing each baseline covariate on the set of treatment indicators including randomization strata fixed effect
and clustering at the school level. Trial 1 and 2 are tested separately. For each trial, the first column reports the control group’s sample means, while the remaining columns
report the coefficients and standard errors from a regression of the covariate on the treatment indicators as well as p-values from z-tests of the treatment indicators being zero.
School level is defined by the student’s initial school allocation (August 2012). Baseline data is obtained from register data collected before the beginning of the intervention.
Pre-test scores are standardized in the population within each year, grade, and subject.

41

Electronic copy available at: https://ssrn.com/abstract=4182253


Table A4: Analysis of attrition and robustness of effects of interventions on academic outcomes, during intervention. Trial 1.
Reading (grade 6) Math (grade 6)
(1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6)
Outcome: 0/1 has
completed test
Co-teacher w/ degree -0.050 -0.053 -0.053 -0.041 -0.023 -0.014 -0.049 -0.049 -0.051 -0.034 -0.023 -0.014
(0.050) (0.037) (0.037) (0.034) (0.034) (0.031) (0.046) (0.042) (0.035) (0.031) (0.034) (0.031)
Teaching assistant w/o -0.042 -0.031 -0.031 0.006 0.008 0.021 -0.044 -0.041 -0.033 0.008 0.008 0.021
degree (0.052) (0.037) (0.037) (0.033) (0.033) (0.031) (0.051) (0.044) (0.037) (0.030) (0.033) (0.031)
Additional students -0.281*** -0.195 -0.281*** -0.195
(0.104) (0.120) (0.104) (0.120)
Co-teacher X Additional -0.123 -0.163 -0.123 -0.163
students (0.134) (0.131) (0.134) (0.131)
Teaching assistant X -0.055 -0.067 -0.055 -0.067
Additional students (0.144) (0.141) (0.144) (0.141)

Observations 829 829 829 829 829 829 829 829 829 829 829 829
Adjusted R-squared 0.079 0.337 0.336 0.397 0.424 0.431 0.063 0.152 0.293 0.360 0.424 0.431

Outcome: Test score


Co-teacher w/ degree 0.113 0.164** 0.158** 0.165** 0.133* 0.135* 0.178 0.183 0.196* 0.206* 0.133* 0.135*
(0.096) (0.073) (0.074) (0.073) (0.076) (0.074) (0.121) (0.115) (0.105) (0.105) (0.076) (0.074)
Teaching assistant w/o 0.174* 0.198*** 0.184** 0.200** 0.205** 0.207** 0.064 0.038 0.054 0.104 0.205** 0.207**
degree (0.097) (0.075) (0.074) (0.078) (0.078) (0.081) (0.121) (0.112) (0.106) (0.105) (0.078) (0.081)
Additional students -0.382 -0.361 -0.382 -0.361
(0.260) (0.234) (0.260) (0.234)
Co-teacher X Additional
students 0.343 0.326 0.343 0.326
(0.407) (0.349) (0.407) (0.349)
Teaching assistant X 0.022 -0.000 0.022 -0.000
Additional students (0.339) (0.325) (0.339) (0.325)

Observations 699 699 699 699 699 699 679 679 679 679 679 679

42

Electronic copy available at: https://ssrn.com/abstract=4182253


Adjusted R-squared 0.045 0.385 0.404 0.425 0.409 0.427 0.054 0.235 0.288 0.302 0.409 0.427

Included controls:
Randomization strata FE x x x x x x x x x x x x
Pre-test in reading x x x x x X X x x
Pre-test in math x x x X X x X x X
Students controls x X x x
Note: Final estimation sample. In trial 1, 82% and 76% of the control group have a national test in reading and math grade 6, respectively. For each outcome, we run two
regressions: First, the probability of having the outcome (i.e., having completed the national test in that subject) on the treatment indicators and a set of control variables, and
second, the student’s test score on the treatment indicators and the same set of control variables. Each regression model includes randomization strata fixed effect and the set
of controls noted in the bottom panel. Student characteristics include the following: Enrolled in special class in grades 4 and/or 5, learning disabilities, reading disabilities,
other special needs or disabilities, low birthweight, non-western origin, and mother and father's education level (college or no college degree). Missing variables are set to
zero, and an indicator for missing is included in the set of controls. Robust standard errors in parentheses are clustered on intervention school. *** p<0.01, ** p<0.05, *
p<0.1.

43

Electronic copy available at: https://ssrn.com/abstract=4182253


Table A5: Analysis of attrition and robustness of effects of interventions on academic outcomes, during intervention. Trial 2.
Reading (grade 6) Math (grade 6)
(1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6)
Outcome: 0/1 has
completed test
Coach -0.160** -0.138*** -0.137*** -0.106** -0.143*** -0.124*** -0.160*** -0.133*** -0.136*** -0.098** -0.143*** -0.124***
(0.062) (0.048) (0.049) (0.043) (0.045) (0.041) (0.058) (0.049) (0.045) (0.040) (0.045) (0.041)
Additional students -0.401*** -0.314*** -0.401*** -0.314***
(0.098) (0.111) (0.098) (0.111)
Coach X Additional 0.054 0.066 0.054 0.066
students (0.144) (0.133) (0.144) (0.133)

Observations 455 455 455 455 455 455 455 455 455 455 455 455
Adjusted R-squared 0.209 0.357 0.359 0.426 0.432 0.455 0.231 0.296 0.389 0.456 0.432 0.455

Outcome: Test score


Coach -0.104 -0.072 -0.102 -0.069 -0.102 -0.078 0.043 0.018 0.024 0.052 -0.102 -0.078
(0.141) (0.106) (0.099) (0.088) (0.092) (0.090) (0.142) (0.125) (0.120) (0.103) (0.092) (0.090)
Additional students -0.617** -0.418 -0.617** -0.418
(0.237) (0.280) (0.237) (0.280)
Coach X Additional -0.014 -0.046 -0.014 -0.046
students (0.360) (0.375) (0.360) (0.375)

Observations 359 359 359 359 359 359 363 363 363 363 363 363
Adjusted R-squared 0.129 0.354 0.382 0.400 0.399 0.404 0.112 0.224 0.257 0.286 0.399 0.404
Included controls:
Randomization strata FE x x x x x x x x x x x x
Pre-test in reading x x x x x x x x x
Pre-test in math x x x x x x x x x
Students controls x x x x
Note: Final estimation sample. In trial 2, 79% and 82% of the control group have a national test in reading and math grade 6, respectively. For each outcome, we run two
regressions: First, the probability of having the outcome (i.e., having completed the national test in that subject) on the treatment indicators and a set of control variables, and
second, the student’s test score on the treatment indicators and the same set of control variables. Each regression model includes randomization strata fixed effect and the set

44

Electronic copy available at: https://ssrn.com/abstract=4182253


of controls noted in the bottom panel. Student characteristics include the following: Enrolled in segregated class in grades 4 and/or 5, learning disabilities, reading disabilities,
other special needs or disabilities, low birthweight, non-western origin, and mother and father's education level (college or no college degree). Missing variables are set to
zero and an indicator for missing is included in the set of controls. Robust standard errors in parentheses are clustered on intervention school. *** p<0.01, ** p<0.05, * p<0.1.

45

Electronic copy available at: https://ssrn.com/abstract=4182253


Table A6: Effects of interventions on academic outcomes by gender, during intervention
Reading score Math score
grade 6 grade 6
(1) (2) (3) (4)
Boys Girls Boys Girls
Trial 1
Co-teacher w/ degree 0.195** 0.116 0.091 0.266**
(0.086) (0.111) (0.117) (0.126)
Teaching assistant w/o degree 0.160 0.202* 0.033 0.013
(0.110) (0.103) (0.111) (0.129)

Observations 430 269 421 258


Adjusted R-squared 0.384 0.423 0.264 0.414
Co-teacher=TA (p-value) 0.760 0.511 0.642 0.083
Control group mean -0.863 -0.832 -0.627 -0.646
Trial 2
Coach -0.108 -0.096 -0.018 0.086
(0.146) (0.160) (0.161) (0.270)

Observations 218 141 225 138


Adjusted R-squared 0.323 0.418 0.216 0.276
Control group mean -0.984 -0.698 -0.662 -0.634
Note: Final estimation sample. Each regression model includes reading and math pre-tests, indicator for missing
pre-test, and randomization strata fixed effect. Robust standard errors in parentheses are clustered on
intervention school. *** p<0.01, ** p<0.05, * p<0.1.

46

Electronic copy available at: https://ssrn.com/abstract=4182253


Table A7: Analysis of imputation of missing outcomes
Reading score Math score
Grade 6 Grade 6
(1) (2) (3) (4)
Missing outcome is imputed with Control group’s Control group’s
-1 SD -1 SD
value: mean score mean score
Trial 1
Co-teacher w/ degree 0.150** 0.142** 0.196** 0.177*
(0.068) (0.067) (0.095) (0.090)
Teaching assistant w/o degree 0.186*** 0.181*** 0.088 0.076
(0.067) (0.066) (0.094) (0.092)

Observations 829 829 829 829


Adjusted R-squared 0.365 0.378 0.256 0.278
Co-teacher=TA (p-value) 0.658 0.620 0.258 0.275

Trial 2
Coach -0.051 -0.069 0.074 0.027
(0.104) (0.102) (0.107) (0.102)

Observations 455 455 455 455


Adjusted R-squared 0.319 0.331 0.220 0.257
Note: Final estimation sample with imputed missing outcomes. Each regression model includes reading and
math pre-tests, indicator for missing pre-test, and randomization strata fixed effect. Robust standard errors in
parentheses are clustered on intervention school. In trial 1, the control group’s mean score is -0.85 SD in reading
and -0.63 SD in math. In trial 2, the control group’s mean score is -0.87 SD in reading and -0.65 SD in math.
*** p<0.01, ** p<0.05, * p<0.1.

47

Electronic copy available at: https://ssrn.com/abstract=4182253


Table A8: Effects of interventions on enrollment, grades 7–9
(1) (2) (3)
Enrolled in grade 7 Enrolled in grade 8 Enrolled in grade 9
in 2013/14 in 2014/15 in 2015/16
Trial 1
Co-teacher w/ degree -0.028** -0.058*** -0.034
(0.013) (0.017) (0.020)
Teaching assistant w/o degree -0.020 -0.013 0.000
(0.016) (0.015) (0.017)

Observations 829 829 829


Adjusted R-squared 0.082 0.112 0.082
Co-teacher=TA (p-value) 0.586 0.009 0.079
Control group mean 0.983 0.983 0.952
Trial 2
Coach -0.053** -0.056** 0.012
(0.025) (0.026) (0.023)

Observations 455 455 455


Adjusted R-squared 0.118 0.063 0.063
Control group mean 0.986 0.959 0.941
Note: Final estimation sample. Each regression model includes students’ reading and math pre-test, indicator for
missing pre-test, and randomization strata fixed effect. Robust standard errors in parentheses are clustered on
intervention school. Cols. 1–3 form the basis of figure 3. *** p<0.01, ** p<0.05, * p<0.1.

48

Electronic copy available at: https://ssrn.com/abstract=4182253


Table A9: Effects of interventions on enrollment in grade 9 in year 2015/16, type of
school
School type
Enrolled in Public Private Boarding Special
Grade 9 school school school school
Trial 1
Co-teacher w/ degree -0.034 -0.055 -0.004 0.039* -0.002
(0.020) (0.034) (0.016) (0.021) (0.017)
Teaching assistant w/o degree 0.000 -0.035 0.016 -0.012 0.040**
(0.017) (0.036) (0.014) (0.021) (0.019)

Observations 829 829 829 829 829


Adjusted R-squared 0.082 0.050 -0.003 0.022 0.054
Co-teacher=TA (p-value) 0.079 0.610 0.191 0.029 0.041
Control group mean 0.952 0.719 0.041 0.106 0.086
Trial 2
Coach 0.012 0.088 -0.065** -0.031 0.001
(0.023) (0.055) (0.027) (0.023) (0.039)
Observations
Adjusted R-squared 455 455 455 455 455
Control group mean 0.063 0.020 0.005 0.060 0.009
Note: Final estimation sample. Each regression model includes students’ reading and math pre-tests, indicator
for missing pre-test, and randomization strata fixed effect. Robust standard errors in parentheses are clustered on
intervention school. *** p<0.01, ** p<0.05, * p<0.1.

49

Electronic copy available at: https://ssrn.com/abstract=4182253


Table A10: Effects of interventions on enrollment in grade 10 in year 2016/17, type of
school
School type
Enrolled in Public Private Boarding Special
Grade 10 school school school school
Trial 1
Co-teacher w/ degree -0.054 -0.012 0.024* -0.021 -0.044**
(0.034) (0.037) (0.013) (0.025) (0.022)
Teaching assistant w/o degree 0.093** 0.037 0.049*** 0.007 0.006
(0.041) (0.042) (0.017) (0.032) (0.022)

Observations 829 829 829 829 829


Adjusted R-squared 0.082 0.052 0.044 0.081 0.074
Co-teacher=TA (p-value) 0.001 0.265 0.148 0.368 0.042
Control group mean 0.575 0.229 0.034 0.185 0.127
Trial 2
Coach 0.033 0.029 -0.037* 0.046 -0.005
(0.036) (0.049) (0.022) (0.044) (0.029)

Observations 455 455 455 455 455


Adjusted R-squared 0.001 0.040 0.026 0.067 0.039
Control group mean 0.600 0.223 0.032 0.186 0.159
Note: Final estimation sample. Each regression model includes students’ reading and math pre-tests, indicator
for missing pre-test, and randomization strata fixed effect. Robust standard errors in parentheses are clustered on
intervention school. *** p<0.01, ** p<0.05, * p<0.1.

50

Electronic copy available at: https://ssrn.com/abstract=4182253


Table A11: Effect of interventions on enrollment or completion of upper secondary
education by years
(1) (2) (3) (4)
2016/17 2017/18 2018/19 2019/20
Trial 1
Co-teacher w/ degree 0.002 -0.036 0.017 -0.015
(0.037) (0.040) (0.036) (0.036)
Teaching assistant w/o degree -0.131*** -0.039 -0.034 -0.083**
(0.042) (0.040) (0.034) (0.033)

Observations 829 829 829 829


Adjusted R-squared 0.097 0.144 0.121 0.21
Co-teacher=TA (p-value) 0.001 0.951 0.237 0.084
Control group mean 0.301 0.712 0.702 0.685
14
Trial 2
Coach -0.006 -0.011 -0.048 -0.080*
(0.033) (0.039) (0.039) (0.046)

Observations 455 455 455 455


Adjusted R-squared 0.069 0.212 0.225 0.161
Control group mean 0.218 0.641 0.645 0.627
Note: Final estimation sample. Each regression model includes students’ reading and math pre-test, indicator for
missing pre-test, and randomization strata fixed effect. Upper secondary education includes high school and
vocational training. Completion is measured by December a given school year, e.g., for school year 2019/20;
completion is by December 2019. Robust standard errors in parentheses are clustered on intervention school.
*** p<0.01, ** p<0.05, * p<0.1.

51

Electronic copy available at: https://ssrn.com/abstract=4182253


Table A12: Cost calculations
Cost of the intervention DKK for the
DKK per DKK per US$ per
intervention period
school year student/year student/year
(Oct.–June)
Prices as of year 2012 2012 2012
Co-teacher w/ degree 156,000 208,000 9,905 1,565
Teaching assistant w/o degree 156,000 208,000 9,905 1,565
Coach 21,739 28,985 1,380 218

Cost of students in Danish public schools DKK per US$ per


student/year student/year
Prices as of year 2012
Exchange rate 6.33
Public school student in regular class 53,892 8,514
Public school student in:
- Regular class with special education
90,425 14,285
support
- Special class 198,720 31,393
- Special school 304,754 48,144

Cost-benefit calculations
Saved costs of moving one student from a
144,828 22,879
special class to a regular classroom

Ratio (costs of a co-teacher or teaching 1.1


assistant for one classroom relative to the
saved costs of moving one student from a
special class to a regular classroom)
Source: (Ministry of Finance, 2010). Special education support in regular classrooms includes SEN students
who receive support for at least 12 lessons per week. All cost measures are extrapolated from the school year
2008/09 based on price indices from Statistics Denmark.

1
Previous studies examining peer effects of including SEN students without providing

additional teacher support in the classroom indicate that learning of peers may be attenuated

(Balestra et al., 2022; Fletcher, 2010; Justice et al., 2014; Rangvid, 2019). However, this

result may no longer hold if additional teacher support is provided.


2
https://unesdoc.unesco.org/ark:/48223/pf0000098427, signed by 92 countries.

52

Electronic copy available at: https://ssrn.com/abstract=4182253


3
Kindergarten class was not compulsory for students in our study. It became compulsory for

the cohort starting school in 2009. Before 2009, average enrolment in the optional

kindergarten class was 83% (2005 figures; UNI-C 2012).


4
Since 2006, school principals have been responsible for ensuring that students sit the

compulsory exit exams. Students with special needs are offered circumstances

accommodating their needs (extra time, assistive technology, digital rather than handwritten

exam etc.), and only in rare cases students are formally exempted from sitting the exam.

Government auditors have recently expressed concern that this regulation is not enforced

(www.ft.dk/statsrevisorerne).
5
See www.statbank.dk/UDDAKT20.
6
See Lynggaard and Rangvid (2014), who also show that SEN students in special schools

and special classes are quite similar along many observable dimensions, including

background characteristics, test attendance, and test scores.


7
Note, however, that students may or may not be considered “disabled“, and the existence of

“disability” is not formally linked to the services that the student is entitled to. In this respect

the institutional setting may differ from other countries, e.g. the US, where many of the cited

studies concern “students with disabilities”, who have certain guaranteed rights to appropriate

education.
8
It was necessary to announce the randomization result prior to the intervention start to allow

schools time to hire the additional teaching staff meeting the intervention requirements.
9
This showed up as a slight unbalance and was reported as such by (Andersen et al., 2020).
10
In columns 5–6 in tables A4 and A5 we show that attrition in the reading test (grade 6) is to

some extent driven by additional students, but additional students were few, and no

interaction terms are statistically significant, and effects on test scores are no different across

groups.

53

Electronic copy available at: https://ssrn.com/abstract=4182253


11
In table A7, we impute missing outcomes to test the robustness for the results on reading

and math. In column 1, we assume the missing observations would have obtained at least the

same value as the control group’s mean test outcome, and in column 2, we assume they

would have scored 1 SD below the population mean. For trial 1, the effects are of the same

magnitude and significance in both reading and math (as expected), while for trial 2, the

effects remain insignificant after imputing the missing outcomes.


12
Table A9 distinguishes between enrollments in different school types in grade 9. For trial

1, the point estimate on the co-teacher for enrollment in grade 9 on time is negative (-3.4)

although insignificant. It appears that students have moved from public to boarding school to

a larger degree than at the control schools. It also appears that students have moved from

public to segregated school after the teaching assistant intervention to a larger degree than at

the control schools. For trial 2, treated students have a 1.2 percentage points higher

probability of enrolling in grade 9 on time, concealing two counteracting effects: a significant

decrease in enrollment in private schools, which is more than offset by an (insignificant)

increase in enrollment in public schools.


13
Table A10 shows that effects of treatment on specific school types in grade 10 vary

somewhat. The co-teacher and teaching assistant treatment tend to increase the probability of

attending grade 10 at a private school whereas the coach intervention tends to increase the

probability of attending grade 10 at a public school or at a boarding school (not significant,

though).

54

Electronic copy available at: https://ssrn.com/abstract=4182253

You might also like