Teacher Professional Development in Rural Nepal

Last registered on September 21, 2026

Pre-Trial

Trial Information

General Information

Title
Teacher Professional Development in Rural Nepal
RCT ID
AEARCTR-0019304
Initial registration date
September 15, 2026

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
September 21, 2026, 9:56 AM EDT

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Locations

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Primary Investigator

Affiliation
University of Washington

Other Primary Investigator(s)

PI Affiliation
University of Washington

Additional Trial Information

Status
In development
Start date
2027-04-26
End date
2028-06-26
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
Abstract
Procuring education hardware school by school tends to raise learning where teacher capacity already exists and can depress it where it does not, widening rather than narrowing gaps between better- and worse-resourced schools. This trial tests whether teacher capacity delivered directly to teachers can improve instruction and student foundational learning in rural Nepal. Using a school-level cluster-randomized waitlist design, we evaluate a teacher-facing AI instructional tool, co-designed with teachers, and measure teacher uptake, student learning, and engagement.
External Link(s)

Registration Citation

Citation
Griffith, Alan and Tsering Sherpa. 2026. "Teacher Professional Development in Rural Nepal." AEA RCT Registry. September 21. https://doi.org/10.1257/rct.19304-1.0
Sponsors & Partners

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information
Experimental Details

Interventions

Intervention(s)
AI tool delivered directly to teachers.
Intervention Start Date
2027-05-31
Intervention End Date
2028-05-29

Primary Outcomes

Primary Outcomes (end points)
Grades 2-4 oral English proficiency, measured with a short, standardized oral-language instrument independent of DeepGyan.
Primary Outcomes (explanation)
The instrument is an EGRA-style set of oral sub-tasks: letter/word recognition, a fixed timed connected-text passage (oral reading fluency, scored in words correct per minute), and a small number of oral-comprehension / expressive-language prompts. This will be administered identically in both arms at baseline and endline. Scores are standardized to mean zero, unit variance within the control-group endline distribution. A single pre-registered primary keeps the significance bar low and preserves power; all other measures are secondary or exploratory.

Secondary Outcomes

Secondary Outcomes (end points)
Teacher uptake/usage of the tool; student engagement; teacher self-efficacy.
Secondary Outcomes (explanation)
Uptake and engagement measured through tool usage logs and survey items (specified once finalized).

Experimental Design

Experimental Design
A school-level cluster-randomized trial with a waitlist (staggered rollout) design, stratified by lab access and school level.
Experimental Design Details
Not available
Randomization Method
Randomization is done using a pre-registered R script with a fixed seed committed to the registry in the office by a computer.
Randomization Unit
School (cluster).
Was the treatment clustered?
Yes

Experiment Characteristics

Sample size: planned number of clusters
80 schools.
Sample size: planned number of observations
~2,000
Sample size (or number of clusters) by treatment arms
Blocks are lab(2) × geography(2) × phone-ownership(2) = 8. Within each non-empty block, schools are assigned to treatment or control with probability 0.5 under equal allocation; odd-membership blocks use a pre-recorded coin flip for the residual school. School enrollment is not a block; it is retained as an ANCOVA covariate.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
At the full frame J ≈ 80 schools (20 with a computer lab, 60 without), with a median grades 2-4 cohort of m ≈ 13 pupils per school and a mean cohort size of approximately 25 pupils per school (≈2,000 pupils assessed total; the gap between median and mean reflects larger cohorts concentrated at the 12 secondary schools relative to the 68 basic schools) and a baseline oral-English covariate, the main-effect minimum detectable effect (MDE) is ≈ 0.24 SD at the headline ICC = 0.25 (≈ 0.27 SD at a conservative ICC ≈ 0.39), under the efficient inverse-variance-weighted (precision-weighted / GLS) estimator, which is the pre-registered primary specification. A naive unweighted OLS with cluster-robust standard errors is materially weaker (≈ 0.30-0.32 SD), because per-school cohort sizes vary widely; this is why the primary specification is precision-weighted rather than unweighted. The covariate R² is ≈ 0.51 alone, ≈ 0.66 combined with stratification, and ≈ 0.62 after a cross-administrator reliability discount. The headline 0.24 SD MDE uses R² ≈ 0.62 (the reliability-discounted figure); R² ≈ 0.66 is reported for comparison as the pre-discount value. These design parameters (covariate R², the ICCs, and the reliability discount lambda) were re-estimated on the grades 2-4 band the trial actually assesses, which lowered the combined R² from 0.85 (the prior, broader-band estimate) to 0.66, moving the headline MDE from ≈ 0.19 to ≈ 0.24 SD. They will be re-confirmed, not re-derived, once baseline data are collected; no outcome data inform these parameters prior to baseline. The pre-specified 0.20 SD target is not cleared, and we do not claim it is: 0.20 SD is retained as the policy-relevant threshold the design is built around, and ≈ 0.24 SD is the honest minimum detectable effect on the conservative grades 2-4 proxy. The binding exposure constraint is consent-gated adoption rather than pupil attrition: the intent-to-treat MDE stays near 0.24-0.27 SD across plausible school-level loss (0-20%), but the Bloom-adjusted treatment-on-the-treated effect among adopters must be at least 0.36, 0.46, or 0.63 SD at adoption rates of 70%, 55%, or 40% respectively. Adequate power therefore rests on adoption being reasonably high, a bar the baseline pilot will establish. Within-stratum and interaction tests are powered only to sign a direction, not to size an effect: within-stratum MDE ≈ 0.29 SD in the non-lab stratum (J = 60) to ≈ 0.41 SD in the lab stratum (J = 20); the lab × phone interaction MDE is ≈ 0.47 SD.
IRB

Institutional Review Boards (IRBs)

IRB Name
IRB Approval Date
IRB Approval Number