Experimental Design Details
The intervention’s treatment arms are as follows:
1) [Awareness treatment] General statements about attendance, lateness and assignment completion (e.g. “attending school is important”) – T1
2) [Awareness + information treatment] Child-level attendance, lateness and assignment completion – T2
Comparing T2 to T1 and T1 to control allows separating the effects of information and awareness.
There are two main concerns about how this design may potentially underplay the effects of information. The first is that parents may already have (to a reasonable extent) information about their child, such that the key piece of information missing is how to place their child relatively to his or her classmates. In fact, other studies often focus on relative behavior: e.g. Rogers and Feller (2016) inform parents about how their children’s attendance fares relatively to his/her classroom modal attendance.
To deal with this concern, we pursue two strategies. First, we survey parents at baseline about their best guess for their child’s attendance, lateness and assignment completion, so as to investigate heterogeneity of treatment effects by baseline accuracy (Annex 2). Second, for a subsample of schools, we add an alternative awareness + information treatment that conveys parents both with pupil- and classroom-level information, to test whether that treatment has additional effects.
3) [Awareness + relative information treatment] Child- and classroom-level attendance, lateness and assignment completion – T3
The second concern is contamination, or peer effects. While there is a concern that assigning different treatments within the same classroom may lead to contamination, we are less worried about it in this setting parents typically have no recurring interactions at this age – most of them no longer take their children to school, and parent-teacher meetings are rather infrequent in Brazilian public schools. However, peer effects may lead us to underestimate treatment effects.
To deal with this concern, our design varies the exposure to the different treatments across different subsamples of schools, allowing us to estimate spillovers. Randomization will be performed in two steps. First, schools will be randomly assigned to 4 different subsamples (A-D), determining the treatment arms each school will have access. Then, students will be randomized within class to each treatment arm:
A. Pure control – 25 schools
B. T1 + control – 25 schools
C. T1 + T2 + control – 100 schools
D. T1 + T2 + T3 + control – 50 schools
Subsample C allows separating the effects of information and awareness; subsamples A and B allow estimating spillover effects. Subsample D is meant to address the concern about relative vs. absolute child-level information.
In order to collect cellphones and information from parents in the control group, and also to control for the proportion of parents registered in the program, we will offer the control and the treatment group access to send school events through the platform.