The impact of educational television on Children's Developmental Outcomes

2023-06-24

2024-03-27

- Literacy Outcomes: Reading Fluency Score, Comprehension Score
- Gender Attitudes: Gender stereotype knowledge and flexibility score, Gender roles, In- and Out-group attitudes
- Socio-emotional lerning: Confidence and Curiosity Score

Literacy Outcomes: employs the Early Grade Reading Assessment Tool (USAID, 2016).
- Reading Fluency Score aggregates scores for letter identification, non-word reading, and oral fluency.
- Comprehension Score aggregates scores for oral and listening comprehension.

Gender Attitudes:
- Gender stereotype knowledge and flexibility score: Following a mapping of which stereotypes are considered “male” or “female” in the literature, responses are coded with “1” if they follow the stereotype, and “0” otherwise. These items are then aggregated into one score denoting the number of gender stereotypes that the child subscribes to. A higher score means that the child has strong gender stereotype knowledge and low stereotype flexibility.
- Gender roles: An aggregate score measuring belief in traditional gender roles. The score aggregates individual questions about different behaviors and traits typically associated with either “male” or “female.” Each child is asked both how likely they are to perform each behavior and how often they actually perform the behavior. Based on each child’s gender and their responses, a final score is constructed denoting the strength of their gender role beliefs. A higher score represents stronger beliefs in traditional gender roles.
- In- and Out-group attitudes: An aggregate score measuring attitudes toward oth- ers with the same and different genders. The score aggregates individual questions about activities with both the in- and out-group. Each child is asked both how willing they are to perform each behavior and how often they actually perform the behavior. Based on each child’s gender and their responses, a final score is con- structed denoting the strength of in-group and out-group attitudes. A higher score represents strong preference for in-groups while a lower score represents flexibility between in- and out-groups.

Socio-Emotional Learning:
- Confidence and Curiosity: We use measures of confidence and curiosity as a proxy for self-efficacy using the RTI confidence and curiosity scale, which has been vali- dated in both our East African context and our age group (Jukes et al., 2021). The score is the simple sum of all items, where relevant items are reverse coded so that a higher final score reflects higher confidence and curiosity. Our main outcome here, defined as the “SEL Score” pools the scores for Confidence and Curiosity.

Evaluate sub-categories of the different main outcomes. We will also explore mechanisms (see pre-analysis plan attached)

We are conducting a school-level cluster randomized controlled trial. Our sampling strategy consisted of 3 stages: (1) identify eligible counties and sub-counties, obtain a random draw; (2) identify eligible schools within randomly selected sub-counties and obtain a random draw; (3) obtained final sample of children within eligible schools.

Schools were randomly assigned to a pure control or treatment condition where they received encouragement to watch the show. For this randomization we employ a “multivariate stratified quadruple matching” technique. Traditionally, pair-wise matching was believed to be the most ideal approach to maximize balance and power in field experiments following Bruhn and McKenzie (2009). However, more recent work suggests a more conservative approach of using quadruplets or larger groups (McKenzie, 2022). We follow this approach by employing quadruplets given our clustered design and moderate number of units within each cluster.

To adopt this approach, we find groups of four clusters, based on a list of baseline characteristics, and randomly assign two into the treatment group, and two into the control group. Quadruplets were matched based on clusters with the lowest Mahalanobis distance cal- culated using these characteristics.

Selecting counties and sub-counties:
We use data on public primary school enrollment from the Kenya Basic Education Statistical Booklet 2019 \citep{kenya2019basic} and data from the Kenyan Census to obtain sub-county school-level estimates of the number of eligible children. We define an eligible child as one who is enrolled in grades 1-3 and has a TV at home.

We then keep sub-counties where we estimated there were at least 40 eligible children enrolled in a public primary school. We further dropped sub-counties that were the only eligible unit within their county to reduce enumerator travel costs. We also excluded Nairobi because as the capital city there are many schools in close geographic proximity and therefore spillovers would be more likely.

We made the final selection of sub-counties by first specifying all possible combinations of 6 counties from the pool of eligible counties identified above, such that each county is in a different region.\footnote{There are 8 regions in Kenya.} Each possible combination yields a list of at least 3,000 government primary schools.\footnote{We targeted combinations of counties yielding at least 3,000 schools because we were hoping to find 500 schools for the study that were at least 4km away from each other. However, schools tend to be quite close together, so we used a conservative factor of 6 when defining the size.} Finally, we randomly selected one of the combinations of 6 counties yielding a list of over 3,000 government primary schools within their eligible sub-counties.

Selecting schools within sub-counties

We followed a procedure like the one used to select sub-counties to select schools. We first calculated the expected number of eligible boys and girls (i.e., in grades 1-3 and, potentially, with a TV at home). We then defined an eligible school as one that we had estimated at least 12 eligible girls and 12 eligible boys, so that there were at least 24 estimated eligible children. We use a threshold of 12 to ensure there was a large enough sample of children with access to TV.

From this group of eligible schools, we randomly sampled schools, such that each additional randomly sampled school was at least 4 kilometers away from any other schools already included in our sample.


Sampling children within these schools

Once we had our randomly selected schools, we then obtained samples of children within these schools. Field officers were sent to the locations where our sampled schools were located. With the help of the village chief, the village elders, and the teachers, we were able to obtain household contact information.

We then conducted a phone screening and recruitment exercise to ensure that we had a good number of eligible households that met our study criteria to be included in the study and these were interviewed during baseline. The criteria included children's age and TV access. Due to a fairly low number of households without access to a functioning TV, we included all eligible households in our sample. We also included children in the same household in our sample.

The school-level strata characteristics used to form clusters include:
1. Proportion of female students in the school
2. Mean asset score
3. Mean grade within school ranging from 1 to 3
4. Caregiver education level
5. Mean reading fluency score at baseline
6. Cluster size or number of sampled students
7. Self-reported measure of how much TV children watch on average

Randomization done in office by computer

School level

Yes

346 schools

4,373 children

Treatment: 173 schools (2,198 children)
Control: 173 schools (2,175 children)