1 Introduction to the Guide

The public release of the AMPLab microdata has created new opportunities for researchers, policymakers, and practitioners to explore primary learning outcomes in participating African countries. The datasets can be downloaded from the UNESCO Institute for Statistics (UIS), making it possible to conduct independent analyses of student achievement, educational inequalities, school contexts, and the factors associated with learning.

AMPLab (Assessment of Minimum Proficiency Levels) measures attainment of the SDG 4.1.1 Minimum Proficiency Levels in reading and mathematics: MPLa at the end of lower primary (AMPL-a) and MPLb at the end of primary (AMPL-b). The public files include student assessment data and contextual questionnaires for students and schools. Participating countries may administer AMPL-a only, AMPL-b only, or the combined AMPL-ab design.

This guide is a practical companion for new users of the AMPLab data, prepared by AFLEARN at DataFirst, University of Cape Town. It explains the key concepts needed to analyse the data correctly, especially plausible values and replicate weights, and provides step-by-step examples in both R and Stata for common tasks such as calculating means, estimating proficiency levels, testing group differences, and running regression models.

Working with AMPLab data differs from working with many conventional survey datasets because AMPLab uses two statistical features designed to produce valid population estimates:

Replicate weights, which account for the complex sampling design and are used to calculate correct standard errors. AMPLab uses paired jackknife replicate weights (fwgt and rwgt1–rwgt135).
Plausible values, which account for uncertainty in the measurement of student proficiency. AMPLab provides five plausible values for each of mathematics and reading.

Why does this matter?

Ignoring these features can lead to incorrect estimates of uncertainty and potentially misleading conclusions.
In practice, ignoring replicate weights typically leads to standard errors that are too small, which inflates test statistics and produces confidence intervals that are too narrow.
Ignoring plausible values introduces bias in distributional estimates and underestimates the uncertainty around all proficiency-related statistics.
Packages such as Rrepest (R) and repest (Stata) make it relatively straightforward to analyse AMPLab data while correctly accounting for both plausible values and replicate weights.

The information in this guide has been compiled and synthesised from UIS and ACER AMPLab technical materials, including international reporting, codebooks, and documentation on Minimum Proficiency Level cut scores. The aim is to consolidate key technical information into a single, accessible reference for data exploration, management, and analysis.