Labs
Longer than an exercise, and against a platform dataset the course did not teach on. Some run in a hosted notebook; some expect your own machine, because setting it up is part of the work.
A management model on 2,625 water point visits says NGO management, spring sources, point age and one district all matter. The register holds 242 points visited twelve times each. Correct for that and four of those findings stop being findings, and the coefficients never move.
You are handed a working analysis in a single notebook with hard-coded paths, no environment file and a chart exported by hand. Turn it into something a colleague can clone and run — and prove it by running it somewhere it has never been.
You are handed the numeracy assessment and asked for the effect of the school feeding programme. Write the one-page plan first, run all four designs, compute what the study could have detected, and deliver the answer the commission did not want.
Build a dashboard and the report around it for a vaccination programme, starting from a twenty-tile request. The deliverable is not the page — it is the page, the definition panel behind it, and the paragraph refusing seventeen of the tiles.
Build a small figures pipeline of your own, produce two charts for a WASH report from the committed survey, and prove the arrangement works by correcting the dataset and watching the chart and its caption move together.
Build one indicator from each of the three populations, on four files and six denominators, and then write the paragraph explaining why the three cannot be drawn as a funnel.
Build all four food security indicators from raw components, put them on the same households, and find they disagree by a factor of seven. Then assemble the evidence table that reports the disagreement instead of resolving it.
Run a five-dimension assessment on a food security survey, quantify what every defect costs the headline, and discover that the largest single driver of the published figure is a threshold choice nobody wrote down.
Rank three districts on water point functionality, then work out that you cannot. Build the three rates, bound each one against the visits that were never made, and write the sentence that declines to rank them.
A district table shows six districts, one of them with zero open defecation and a significant difference against its neighbour. There are three districts, the difference is a spelling variant, and the interval on the zero reaches 16%.
A daily attendance register is an aggregate reporting system that has not been described as one. Rebuild it as data elements, org units, periods and completeness, then answer what was counted.
A 30-cluster SMART survey with no weights and no frame. Work out what it can and cannot support, compute GAM with a design-adjusted interval, and decide whether it can be read against the 15% emergency threshold.
A household survey measured one child under five per household. Estimating acute malnutrition from it needs a weight nobody applied, a design nobody declared, and an honest statement about which threshold it can be read against.
Turn a treatment register into a cohort, count person-days instead of children, and watch three sites that share a defaulter proportion separate once you account for how long each one held its children.
Produce the disaggregation a partner asked for, measure its disclosure risk, apply a suppression rule, and then write the reply that declines it and offers three things instead. The deliverable is the reply.
Write reference sheets for the JMP water, sanitation and hygiene service ladders, compute all three from one household survey, and discover that the three indicators cannot share a denominator.
A hands-on lab classifying a real-shaped MUAC dataset against WHO thresholds, then checking how the GAM rate moves when you handle missing values differently.
One area recorded disability as Yes and No instead of true and false, splitting the equity finding into four categories and leaving a cell of three cases at 0%. Clean the caseload, defend the denominator, and write a log you could attach to a protection report.
Join 930 SMART survey children to the WHO 2006 reference table on a three-part key where one part is a rounded float and another is derived from a third column. Four children fail to match, and those four are the finding.
A district's penta3 coverage appears to collapse in August. Build the script that decides whether that is a programme failure or a reporting failure, and make it survive next quarter's export without editing.
A WASH survey where one enumerator team wrote a district name four ways, splitting the worst-performing district into fragments. Normalise it with forcats, put every household on the JMP ladders, and wrap the whole thing in a function that runs on the next round.