Personal mindset and information processing in education: Experimental evidence from Cambodia

In our intervention, students will go through an “interest exploration tool” on a tablet. At the end of the tool, students will be revealed their strongest personality types, as well as shown a list of 18 jobs (with description of typical tasks and required educational degree), ordered by their strongest type. After the students completed the tool, they participate in an information session on education paths, which provides detailed information on high schools and vocational training options as well as scholarship possibilities. This information session will be conducted in small groups.

2020-02-17

2020-04-30

Primary outcomes are (1a) the accuracy of reported educational costs and (1b) retention of information on high school scholarship opportunities provided in the information session.

For (1a) students are asked at baseline and endline to estimate (i) the costs of transportation per month when going to high school (by motorbike), (ii) the cost of extra classes (i.e. informal tuition) per month, and (iii) the annual expenses for school material when going to high school. Additionally, students are asked to estimate the (iv) travel time (by motorbike) to high school.
The primary outcome for (1b) will be the accuracy of students' responses at endline when asked about non-governmental organizations that provide scholarships for the nearest high school.

Secondary outcomes measure (2a) interest in jobs that are outside the typical reference window and (2b) whether students are more likely to see themselves as being able to go to high-school.

(2a) will be measured by the the number (and type) of jobs a student clicks on in the app when presented with a list of 18 jobs, and asked “Which of these jobs do you think could be interesting to you?” (elicited as part of the IET and the Placebo app). (2b) will be measured by (i) the weight that a student assigns to constraints that could keep her from pursuing higher education (measured at midline), and (ii) the interpretation of a graph that depicts average educational expenditures per grade in Cambodia (measured at midline).

We target students from grade 9 in northeast Cambodia. We construct a sample of lower secondary schools, and invite all students of one (randomly) selected class to participate in a workshop carried out at their school. All students that are willing to participate are randomly allocated into one of three arms: a) treatment group (40%), b) placebo group (40%), c) control group (20%).

We construct a sample of lower secondary schools from which we randomly select 50% for our intervention. The school sample consists of 49 pre-selected schools that are cooperating with Child’s Dream, one of our partner NGOs. These schools are distributed across four provinces: Banteay Meanchey, Battambang, Oddar Meanchey, and Siem Reap, provinces that experience particularly high drop-out rates. We exclude schools with less than 30 students in grade 9 (which drops 10 schools). To increase power, we additionally sample 23 lower secondary schools from other districts in the same provinces that are similar in characteristics to the partner schools of Child’s Dream. Of these schools, we randomly select 2 schools to serve as back-up, the remaining 21 schools are included in the main sample. We stratify schools by size (more or less than 100 students in grade 9), by Child’s Dream partnership, and by district, and randomly select half of the schools to participate in our intervention. We then randomly select one class of grade 9 to participate in the intervention.
This sample is a non-random subset of the population of grade 9 students in rural Cambodia. The schools that are in the sample are slightly smaller than the average in rural Cambodia. On average there are about 1.95 grade 9 classes (88 grade 9 students) per school in rural Cambodia, while there are about 1.63 (74) in our sample.

The assignment into treatment arms is done randomly in a two-stage process. First, we randomly assign one class of each school to receive our intervention. Randomization will be done by computer. Students within selected classes in treatment schools will then randomly be allocated to one of three treatment arms before the intervention begins (one the same day). Randomization will be done by having students blindly draw numbered badges from a bag.

Student

No

1350 students

1350 students

a) 540 students in treatment, b) 540 students in placebo, c) 270 students in control

We calculate the smallest real effect sizes that we can detect at a 5% significance level with 80% power. We perform the calculation based on 540 students in treatment and 270 students in control (i.e. assuming the most conservative approach). When calculating the minimum detectable effect size (MDE), we take into account that we have baseline information on student age, gender, test scores, attendance and class size. We expect that these baseline characteristics can explain between 0% and 50% of the variation in the outcome variable. The MDEs corresponding to the main treatment (i.e. comparing students in the treatments group to students in the control group) range between 0.15 of a standard deviation (SD) and 0.21 of a SD. MDEs are slightly smaller, when comparing students in treatment and placebo, and range between 0.12 (0.5 explained variance), and 0.17 (0.0 explained variance). These are the detectable effect sizes for outcomes with mean zero and a standard deviation of one (such as standardized cost estimates).