How do transformative early learning gains shape learning and life in adolescence? Evidence from West Africa

The interventions being studied are described in greater detail in Eble et al. (2021) and Fazzio et al. (2021). The basic model motivating both interventions is to combine, or “bundle” multiple interventions, known to work in isolation, with the goal of having a synergistic effect whose overall impact on learning is far greater than the sum of the anticipated benefits from each intervention in isolation.

In The Gambia, the intervention was delivered via 12 hours per week of after-school lessons, combining several important components:
● Previously untrained “para teachers”;
● Scripted lessons;
● Extensive monitoring of teachers (also known as “coaching”) with the goal of increasing teacher skill and raising student learning;
● Extensive community sensitization; and
● A high level of per-child resources spent to ensure fidelity of the intervention.

In Guinea Bissau, the intervention was delivered by running schools in lieu of the government. Unlike in The Gambia, schools were run by individuals who had previously trained and worked as teachers, but that intervention also used the other components of the bundle in The Gambia. Both studies took place in small, remote rural areas of the country in question.

2013-10-01

2017-07-01

Hypotheses and outcome variables
Our research is designed to test the broad hypothesis that the learning gains from the two studies persisted into adolescence and have yielded salutary effects on outcomes related to education, such as belief formation, life satisfaction, aspirations, and early fertility-related decisions. The main outcomes of interest can be divided into families of outcomes, as in Kling, Liebman, and Katz (2007). In our study, we have the following four families: educational progression; learning; beliefs and aspirations; and family formation and gender norms.

The primary outcome for each family of outcomes is as follows:

● For educational progression, the primary outcome will be the highest grade attended. Secondary outcomes will include whether the trial participant is still in school, whether they were in school in the previous year, the following outcomes collected conditional on the participant being in school this year or the previous: {annual educational expenditure on the trial participant’s education, location of school relative to trial participant’s village of enumeration, type of school}, variables collected conditional on the trial participant not being in school this year: {reason left school, intention to go back to school, main daytime activity, whether they are employed, type of job, work hours, and work income}; time use, and the highest grade attended of the participant’s next older and younger siblings.
● For learning, the primary outcome will be a composite score on a test comprising assessment of early grade reading and math skills and higher-grade learning. Secondary outcomes will include summary scores on of these two categories of skills plus on a third category of other skills, commonly known as “non-cognitive” or “life” skills; scores on subtasks within each category; and spillover of learning to the nearest older and younger sibling of the participant, as measured by ASER-style tests.
● For beliefs and aspirations, the primary outcome will be an index of the participant’s future-oriented beliefs aspirations, created by binarizing their responses to the relevant beliefs and aspirations questions in the survey, harmonizing them so that positive values connote higher aspirations, and summing them. Secondary outcomes will include an analysis of the first principal component of this same index of outcomes as a y variable, individual responses on these participant aspirations questions, an index of caregiver aspirations for the participant, and individual responses on caregiver aspirations questions.
● For family formation and gender norms, the primary outcome will be an index of the participant’s responses to relevant questions on family formation and gender norms, created by binarizing the responses, harmonizing them so that positive values connote higher aspirations, and summing them. Secondary outcomes will include an analysis of the first principal component of this same index of outcomes as a y variable, as well as individual level responses to each question.

In all cases, outcomes will be measured at the level of the study participant.

Described in response to the primary outcomes prompt

Our analysis will follow the research design of the original studies. We will test our hypotheses with linear regression following the research design in the original two studies, regressing the outcome variable on treatment status and the original stratification variables and nothing else, clustering standard errors at the cluster (Gambia) or tabanca (Guinea Bissau) level.

We will estimate the main parameter of interest, the coefficient on treatment status, separate by study country, with one set of regressions and estimates for Guinea Bissau and a separate set of regressions and estimates for The Gambia. To account for potential differentials in attrition across trial arms in both contexts, we will compute and report Lee Bounds.

Not available

Randomization done in office by a computer

Tabancas (Guinea Bissau); Clusters of adjacent villages (Gambia)

Yes

Guinea Bissau: 49 tabancas (villages)
Gambia: 111 clusters of adjacent villages

At least 4,641 observations (described in greater detail below)

For our primary outcome, the unit will be learning at the individual trial participant level. We will travel to all clusters (55 control and 56 treatment in The Gambia; 33 control and 16 treatment in Guinea Bissau), and based on a pilot done in summer of 2023, we estimate that we will be able to reach at least 70 percent of participants in each study (original number of participants: 2,458 control and 2,060 intervention in The Gambia; 1,463 control and 649 intervention in Guinea Bissau).

Under these assumptions, and using an intra-cluster correlation coefficient of 0.1647 in The Gambia and 0.1493 in Guinea Bissau (calculated using the control group test scores at endline), we estimate that our MDE will be 0.2344 in The Gambia, and 0.3605 in Guinea Bissau. Given that the original treatment effects were 3.2 and 5.3 SD in the original studies, this ensures that we are powered to detect any difference that is as large or larger than a tenth of the original effect size.

Pre-analysis plan