School improvement through customized support: Evaluation of a randomized targeted intervention program

The intervention consists of targeted school improvement support provided by the Swedish National Agency for Education (SNAE; Skolverket) to compulsory schools with persistently low student outcomes. The support program is part of a government mandate that SNAE support school development and its design has been developed over time by the SNAE and used before. The research team has not been involved in the design of the program.
The program spans three to four years and includes two main phases. During the first year (the needs assessment phase), a local development team at each participating school, together with representatives from the school district and SNAE, conducts a structured analysis of the school’s strengths, weaknesses, and development needs. Based on this analysis, a customized action plan is formulated.
During the subsequent two to three years (the implementation phase), the action plan is implemented and continuously adjusted. Implementation is supported through financial resources and guidance from SNAE. Interventions typically include professional development activities, coaching, and organizational development efforts aimed at improving instructional quality, strengthening school leadership, and enhancing the school’s capacity for improvement.

2025-06-30

2029-06-30

Student academic achievement measured using:
• National standardized test scores
• Compulsory school grades
• Educational attainment outcomes (upper secondary school eligibility and program choice)

The primary outcomes are student academic achievement measured using Swedish administrative register data.
National standardized test outcomes include:
1. Average standardized test scores across mathematics and Swedish (and, where available, English).
2. Subject-specific standardized test scores in mathematics and Swedish.
Test scores are available for grades 3, 6, and 9.
Grade outcomes include:
• Average standardized grade point average (GPA) in grades 6 and 9.
• Average standardized grades in core subjects (mathematics, Swedish, and English) in grades 6 and 9.
All outcomes will be standardized by year and grade using the distribution among all students. For the main analysis, outcomes will be stacked across grades to improve statistical power, while grade-specific effects will also be reported.
The primary estimand is the effect of the intervention. Outcomes will be measured from randomization and onwards. We will estimate the dynamic effects of treatment during the analysis phase, the implementation and post-intervention phase.
Administrative data sources are used to link student outcomes in school registers to information on parents (and siblings) and contain information on education, labor market outcomes, place of residence, migration status, countyr of origin etc.
Pre-specified heterogeneity analyses will examine differential effects by:
• Student gender,
• Family socioeconomic status (based on e.g., parental education and income),
• Migration background,
• Baseline achievement (defined using pre-intervention test scores where available).
These subgroup analyses will be interpreted as exploratory complements to the main estimates. Data and
Given the presence of multiple primary outcome measures, results will be presented separately by outcome domain. The main conclusions will be based on the overall pattern of effects across related outcomes rather than statistical significance of any single estimate. Where relevant, adjusted p-values or standardized indices aggregating related outcomes may be reported as robustness checks.

We also intend to study the effects of the interventions on pathways through which the intervention might affect student outcomes relating to teaching resources, classroom organization and climate. These include: student-to-teacher ratios, teacher qualifications, , instructional practices, targeted student support, classroom organization and climate, student attendance, and teacher turnover.

The intervention is expected to affect student outcomes through improvements in schools’ organizational capacity, instructional quality, and ability to identify and support low-performing students. These changes are hypothesized to operate through enhanced teacher collaboration and professional learning, improved classroom practices, strengthened leadership, and more effective allocation of student support resources.
We will also assess whether the intervention affects student mobility, test participation, grade retention, and observable student composition. This is important for the interpretation of any effects on student outcomes. Balance between treatment and control groups will be evaluated both at baseline and over time.
Secondary outcomes are intended to shed light on potential mechanisms through which the intervention affects student achievement.
Using administrative teacher and student registers, we will assess effects on:
• Teacher-student ratios
• Teacher qualifications and experience
• Teacher mobility and turnover
• Classroom organization (e.g. class size, grade-for-age, composition)
• Student attendance and access to special educational support
Survey data from the Swedish Schools Inspectorate will be used to construct measures of instructional practices and classroom and school climate from the perspectives of students and teachers.
In addition, a teacher survey developed for this study will measure collegial collaboration, professional learning, perceived teaching capacity, and quality-enhancing school development practices.
These analyses are exploratory and aim to improve understanding of the mechanisms underlying any observed impacts on student outcomes.

In collaboration with the Swedish National Agency for Education (SNAE), IFAU has developed a selection and randomization procedure to allocate targeted support to schools.
Eligible school districts (municipalities) were identified using a pre-determined needs index incorporating prior academic outcomes, student composition, and teacher qualifications. Priority was given to districts with a high share of students attending low-performing schools, i.e. schools in the lower half of the national distribution of student outcomes. Priority was also given to districts with multiple schools.
School districts were informed that at most half of their eligible schools would receive support. Prior to randomization, districts were allowed to exclude schools deemed unsuitable due to planned reorganizations or mergers.
Within participating school districts, low performing schools were ranked according to the needs index. Schools were stratified by grade configuration. Pair-wise randomization within strata determined whether a school was offered the intervention or assigned to the control group.
The randomization aimed to assign approximately 100 schools to treatment and control. In practice, 23 school districts participated, resulting in 86 treated schools and 86 control schools.

Randomization motivates the following statistical analysis plan
The main analysis will estimate intention-to-treat (ITT) effects using linear regression models of the form:
Y_ist=α+β⋅Treatment_s+γ^' X_s+δ_k+ε_ist,

where Y_ist the outcome of student i in school s at time t, Treatment_s is an indicator for assignment to the intervention, X_s is a vector of pre-determined school-level characteristics, and δ_k are fixed effects for randomization pair. Standard errors will be clustered at the school level.
The primary specification will not condition on any post-treatment variables such as student characteristics at time t, since changes in composition of students may be endogenous to treatment. Potential compositional changes will be assessed explicitly by examining effects of treatment on student mobility, test participation, grade retention, and observable student characteristics. Balance between treatment and control groups will be evaluated both at baseline and over time. Sensitivity of the primary specification to student characteristics will be assessed.

Estimates will be based on assignment to treatment, regardless of intervention take-up. However, to the extent schools dropped out from treatment before the intervention started, we will also consider IV-estimates where assignment to treatment is used as an instrument for actual participation at the school level.

Administrative data on students, teachers, and parents will be obtained from Statistics Sweden’s education registers. Survey data on school climate, student perceptions of teaching practices and teachers view of collegial collaboration and school climate will be obtained from the Swedish Schools Inspectorate.
During the early stages of the intervention, interviews with SNAE process support agents (approximately 3–6 individuals) will be conducted to document the content and intended mechanisms of the support packages. This information will be used to design a teacher survey administered to teachers in treated and control schools in collaboration with SNAE.
The use of these administrative data for research purposes, which in a legal sense, are not sensitive, formally does not require ethical approval. Ethical approval will, however, be sought from the Swedish Ethical Review Authority. The application is planned for spring 2026.

Not available

Randomization was conducted using a computer algorithm. Re-randomization was used to achieve balance between treatment and control schools on pre-determined student characteristics and baseline performance measures.

Randomization took place at the school unit level within school heads and strata of grade configurations.

Yes

Treatment it clustered at the school level. In total 172 schools were pairwise randomly assigned.

The average school size among participating schools is some 294 students. This implies an expected treated student population of roughly 27,000 students.

In total, 172 schools were pairwise randomly assigned: 86 schools to treatment, 86 schools to be controls.

Based on prior evidence from Swedish schools, between-school variation in student outcomes conditional on student characteristics is expected to lie between 0.025 and 0.05 standard deviations. Given this variation, a sample of 86 treatment and 86 control schools provides sufficient statistical power to detect average treatment effects in the range of 0.10–0.15 standard deviations.