Impact of Preschool on School Preparedness: Evidence from a Nationwide RCT in Finland

Intervention:
The two year preschool intervention advances the starting age of mandatory preschool by one year from age 6 to age 5. In Finland, compulsory schooling starts at age 6 in preschools which are often attached to day care centers. At age 7 elementary school begins. A special bill was passed by the parliament to mandate preschool for the 5 year old children in the treatment group. The treatment group attends preschool for two years instead of one. The treatment group follows a curriculum made by educational specialists specifically for the two-year preschool. The curriculum is centrally directed by the Ministry of Education but locally adapted my municipal boards of education. There are potentially large variations across municipalities in the exact content of the curriculum. In general, relative to day care, preschool curricula are more structured including sessions with directed learning of school skills.

Counterfactual:
In the absence of treatment, at age 5, most of the treatment group would attend day care and some would stay home.

2021-08-11

2024-06-30

- Arithmetic and reading skills.
- Psycho-social skills and self esteem.

We will examine the impact of the intervention on school preparedness as measured by standardized tests on arithmetic and reading skills, psycho-social skills and self-esteem. The first follow-up will be conducted roughly 14 months (during the second year of preschool) and the second follow-up 26 months (during first grade of elementary school) after the start of the intervention. The survey instrument for the first follow-up will be finalized during the summer of 2022 and for the second follow-up during the summer of 2023. Each four primary outcomes are composed of multiple questionnaire items. We will update our trial registration with details of the measures once the survey instruments are finalized.

Individual test items or sub components of the primary outcomes composites.

We will update our trial registration with details of the measures once the survey instruments are finalized.

The experimental program targets all eligible day care centers in Finland. Eligibility was determined by the ability to offer both services: day care and preschool. The number of eligible day care centers in each municipality was estimated by a survey. Based on the survey, municipalities with fewer than 5 eligible daycare centers formed their own experimental sample (small RCT). Based on this sample split, we run two separate designs in parallel.

Small RCT (randomization at the municipal level)
- 83 municipalities with estimated more than one but less than five eligible day care centers.
- Randomization is done at the level of the municipality.
- 41 municipalities are randomized into treatment group.
- In the treated municipalities all eligible day care centers start offering two-year preschool.

Large RCT (randomization at the day care center level)
- 63 out of 92 large municipalities are first randomized to participate in the experiment using stratified sampling.
- The strata were defined by geographical areas and the number of eligible daycare centers.
- All children born in 2016-2017 are assigned to a day care center based on current customership, or distance.
- Within each of the 63 municipalities in the sample, 40% of eligible day care centers were randomly assigned to the treatment group
- Students that are assigned to a treatment center are offered two-year preschool.

For both designs (small and large municipalities), re-randomization was applied to guarantee covariate balance. We simulated one million potential assignment vectors and chose from them 2000 with the best covariate balance according to the Mahalanobis distance criterion. The final treatment assignment vector was drawn from among those 2000 potential assignment vectors. The covariates used for the criterion were:

Small RCT (municipal level covariate averages/shares)
- Geographical/administrative area
- Foreign born
- Earnings
- Secondary education
- College education

Large RCT (day care center level covariate averages/shares)
- Herfindahl index for language concentration
- Earnings
- Single parents
- Primary education
- Secondary education

In addition, in order to stay within the budget, we balanced on the number of children and cost per child to minimize variation in the expected cost of the experiment.

The re-randomization algorithm was implemented using an R script with a random number seed. The seed was generated by requesting the current computer clock time by email from our Ministry of Education partners.

Small RCT: Randomization is done at the level of the municipality.
Large RCT: Randomization is done at the level of the day care center.

Yes

Small RCT: 83 clusters (municipalities).
Large RCT: 752 clusters (day care centers).

For the cohort born in 2016 Small RCT: 3700 children. Large RCT: 14700 children. The cohort born in 2017 is expected to have 0.95 times as many children. In total, the above figures are multiplied by 1.95 to arrive at an estimated number of observations. Total children in across two cohorts: Small RCT: 7200 children. Large RCT: 28600 children.

Small RCT: 41 small municipalities in treatment, 42 small municipalities control.
Large RCT: 299 day care centers in treatment, 453 day care centers in control.

Large RCT: We are powered to detect a minimum of 0.1 standard deviation treatment effect for a continuous test variable, for example, mathematics. To determine the MDE, we assumed 80% power, test size of 0.05, and intra-cluster correlation of 0.2. Small RCT: We are powered to detect a minimum of 0.25 standard deviation treatment effect for a continuous test variable, for example, mathematics. To determine the MDE, we assumed 80% power, test size of 0.05, and intra-cluster correlation of 0.2. We expect to improve precision by including baseline measures as controls.

Preschool RCT pre-analysis plan (Izadi, Sarvimaki, Harjunen)