Datasets
Every dataset carries a data dictionary, provenance, documented quality problems, and worked examples in both Python and R. All are synthetic or fully de-identified.
1,100 admissions across six sites, carrying both MUAC and weight-for-height at admission, so the disagreement between the two admission criteria is measurable rather than assertable, and the treatment outcome every CMAM report is judged on.
996 household interviews drawn from a 470-area frame by stratified PPS sampling, shipped with the frame rather than with the weights, so selection probabilities have to be reconstructed before any estimate is made.
Synthetic mid-upper arm circumference screening for 4,218 children under five across twelve communes, carrying the missing values, unit errors and duplicate registrations a real campaign register carries.
A 30-cluster SMART survey of 930 children aged 6-59 months, shipped as raw weight and height so z-scores must be computed, and carrying the specific defects a SMART plausibility report is designed to catch.
975 cases recorded one at a time across sixteen weeks and three districts, shipped with the population they came from, so attack rates and case fatality have denominators and age standardisation is possible.
Monthly DHIS2-style aggregate reporting of six antigens across 38 health facilities, built so that genuinely low coverage and simply-did-not-report look identical until you separate them.
975 cases recorded one at a time across sixteen weeks and three districts, shipped with the population they came from, so attack rates and case fatality have denominators and age standardisation is possible.
996 household interviews drawn from a 470-area frame by stratified PPS sampling, shipped with the frame rather than with the weights, so selection probabilities have to be reconstructed before any estimate is made.
2,629 synthetic monitoring visits to 242 water points across three districts, monthly through 2024 — the repeat-visit structure a functionality rate needs and a household survey cannot provide.
2,403 synthetic household interviews across three districts covering water source, collection time, quantity, water quality testing, sanitation and handwashing — enough to place every household on the JMP service ladders.
996 household interviews drawn from a 470-area frame by stratified PPS sampling, shipped with the frame rather than with the weights, so selection probabilities have to be reconstructed before any estimate is made.
2,273 synthetic monthly price observations from twelve markets, covering seven commodities and a casual labour wage over twenty-four months — the series a terms-of-trade analysis needs and a single survey round cannot supply.
2,112 synthetic household interviews carrying the raw components of the Food Consumption Score, Household Hunger Scale and reduced Coping Strategy Index, so the composite indicators have to be computed rather than read off.
2,471 enrolment records across two school years, the school-age population that is their denominator, and 2,798 learning assessment results in two rounds — the three files gross and net enrolment, cohort survival and learning outcomes each need.
Two joinable synthetic files — a roster of 1,200 students across 24 schools and 70,245 daily attendance records — built so dropout is only visible if you look for consecutive absences rather than a term average.