One numerator, four reports

Four partners screened in the same twelve communes and reported to the same nutrition cluster. All four used the same register. All four reported “global acute malnutrition by MUAC” and all four got a different number.

The cluster coordinator has asked you to work out who is wrong. The answer is nobody, and that is the finding.

The file

muac-screening-artibonite-2024.v1.csv — 4,218 screening records from twelve communes in Artibonite. Synthetic, and carrying the missing measurements, sentinel codes and duplicate registrations a real campaign register carries.

The four figures

Partner A — 8.58%. “Cases over children screened.”

Partner B — 8.73%. “Cases over children with a measurement.”

Partner C — 8.9%. “We report the referral rate; a referral means a case was found.”

Partner D — 9.1%. “Cases over children screened, excluding the communes where our teams did not work.”

The task

Part one. Reproduce all four numbers from the register. Each is a small number of lines, and getting all four to match is the point — if one is off, your reading of that partner’s sentence is wrong, not their arithmetic.

Part two. For each, write down in one line what it actually measures and what its denominator is. Then rank them from most to least defensible, and say what would have to be true for each to be the right choice.

Part three. Write the indicator reference sheet the cluster should have issued before the round, using the thirteen fields from the course. It must be specific enough that all four partners would have produced the same figure.

Where each of the four goes wrong, or does not

You are not looking for arithmetic errors. You are looking for the sentence in each partner’s definition that a reference sheet would have settled. Some prompts:

• One partner’s denominator includes children whose nutritional status was never established.
• One partner’s numerator is a decision rather than a measurement, and the register contains records where the two disagree.
• One partner has restricted the geography without restricting the claim.
• One partner is arguably correct and still needs to say more than it did.

The questions to answer in prose

Three sentences each.

1. Partner C reports the referral rate as a prevalence. Say what a referral rate measures that a prevalence does not, name a specific feature of this register that makes the two diverge, and say whether C’s figure is useless or merely mislabelled.

2. Partner D’s number is the highest. Explain what its exclusion does to comparability with the other three, and what D should have written next to its figure to make it usable by the cluster.

3. The four figures span half a percentage point. Give one decision this spread would not affect and one it would, and use the second to argue for the reference sheet.

What to hand in

A Python script that prints all four figures with their denominators and n, a ranking table with your one-line assessment of each, and the reference sheet as a JSON file in indicators/. Your script should compute the sheet’s own definition at the end and show that it lands where you expect.

How to know you are done

All four figures reproduce to two decimal places. Your reference sheet fills every one of the thirteen fields, and the decision_informed field names something more specific than “reporting”. Somebody who has not read this exercise could compute your figure from your sheet alone.