Somebody has already defined this

The question to ask before writing a definition

Almost every indicator this sector reports has already been defined by somebody with more time, more consultation and more scrutiny than you will get. Finding that definition takes twenty minutes and it changes what you write.

The reason is not deference. It is that a number nobody can compare to anything answers half the question. A GAM prevalence that cannot be read against the IPC phase thresholds, or a coverage figure computed differently from the ministry’s, has thrown away most of its value in exchange for local convenience.

Where the definitions live

Source	Covers	Note
WHO	Growth standards, immunisation, disease surveillance case definitions	The growth standards are the authority for anthropometry, full stop
UNICEF	Child protection, WASH, education, nutrition indicator definitions	Usually the most operationally detailed
Sphere	Minimum standards and their indicators across humanitarian sectors	Gives thresholds, not just definitions
The SDG framework	231 global indicators with metadata sheets	The metadata sheet is a reference sheet; read it as a model
IPC	Acute food insecurity and acute malnutrition classification	Defines the evidence, not just the indicator
SMART	Nutrition survey methodology and plausibility criteria	Defines how the data is collected, which is half the definition
JMP	Water, sanitation and hygiene service ladders	The ladders are indicator definitions in disguise
The cluster	Whatever the response is reporting on this year	Least stable, most immediately binding

Two of those deserve emphasis for this audience. The JMP service ladders are definitions, not categories — “basic” water service is an improved source within a 30-minute round trip, and a programme classifying on source alone is not reporting the JMP indicator whatever it calls it. And the SDG metadata sheets are the best free model of a reference sheet you will find; read two of them before writing your own.

Adopt, adapt, or invent

A rule that resolves most cases:

Adopt when a standard definition exists and you can compute it. This is the default and it should cover most of your LogFrame. Adopting means adopting the whole thing — numerator, denominator, age bands, recall period — not the name.

Adapt when the standard definition exists but you genuinely cannot compute it, usually because a data element is missing. Adaptation is legitimate and it must be documented, because an adapted indicator carrying the standard name is the worst of both worlds.

Invent only when the thing you need to measure is specific to your programme and nobody else measures it. Rarer than people think. Before inventing, check whether the thing you want is really a disaggregation of a standard indicator, which it usually is.

Documenting a departure

The reference sheet gets two more fields when you depart from a standard.

Standard          WHO/UNICEF, penta3 coverage (WUENIC methodology)
Departure         Denominator uses MoH facility catchment estimates rather than
                  UN WPP surviving infants. Numerator restricted to facilities
                  reporting in the month, where the standard uses all facilities
                  with non-reporters imputed from the previous year.
Effect            Our figure is a lower bound relative to the standard. On 2024
                  data the difference is largest in August, where reporting
                  completeness was 29%.
Comparable to     Other districts using the same MoH denominators. Not directly
                  comparable to WUENIC national estimates.

Four lines, and the last one is the one a reader needs most. Say what your number can be compared to. Most misuse of an indicator is comparison against something incomparable, and it is almost always done in good faith by somebody who had no way to know.

Map to the standard where you can

Where you have person-level data, you can often compute both your definition and the standard one and publish the pair.

DEFINITIONS = {
    "local_gam_muac": lambda d: (d["muac_mm"] < 125) | (d["oedema"] == True),
    "who_gam_muac_only": lambda d: d["muac_mm"] < 125,
}

for name, rule in DEFINITIONS.items():
    eligible = d = muac[muac["muac_mm"].notna()]
    print(f"{name:22} {rule(d).sum() / len(eligible):.2%}  (n={len(eligible)})")

muac |>
  filter(!is.na(muac_mm)) |>
  summarise(
    local_gam_muac    = mean(muac_mm < 125 | oedema),
    who_gam_muac_only = mean(muac_mm < 125),
    n = n()
  )

Publishing both costs two lines of code and settles a question that would otherwise take an email thread. It also makes the effect of your departure a measured quantity rather than an assertion — which is the same move the cleaning log makes for every cleaning rule.

Comparability is the whole point

Three places where a locally-invented definition costs you concretely:

• Against yourself, over time. A definition changed between rounds breaks the series, and the break is invisible unless versioned.
• Against your neighbours. Cluster reporting aggregates partner figures. A partner using a different denominator makes the cluster total meaningless, and nobody finds out.
• Against the threshold that triggers action. IPC phases, Sphere minimum standards and the 15% GAM emergency threshold are all defined against specific indicator definitions. A GAM computed differently cannot be read against the threshold, and reading it anyway is how a response gets scaled on a number that does not mean what the threshold means.

That last one is not hypothetical. MUAC-based and weight-for-height-based GAM give different prevalences on the same children, and the IPC acute malnutrition thresholds are defined for specific measures. Name the measure in the indicator name — gam_muac_percent and gam_whz_percent are different indicators that both get called “GAM”.

SDG tiers, and what they tell you

The SDG framework classifies its indicators into tiers, and the classification is useful well beyond the SDGs.

• Tier 1 — definition agreed, data regularly produced by most countries.
• Tier 2 — definition agreed, data not regularly produced.
• Tier 3 — no agreed definition yet. (Formally retired as a tier, but the situation it described is everywhere.)

If your indicator is tier 3 in substance — no agreed definition — you are inventing, and you should expect to spend time defending it in every review. That is sometimes the right call. It is never the cheap one, and knowing which tier you are in tells you how much documentation the indicator will need for the rest of its life.

When the standard is wrong for you

Occasionally it genuinely is, and adopting it uncritically is its own failure.

• The age band does not match your programme. A standard defined for under-fives in a programme serving 6 to 23 months needs the standard computed on the standard’s band and yours computed on yours, both published.
• The recall period does not fit the context. A seven-day food consumption recall in a context with a fortnightly distribution cycle will oscillate.
• The denominator is unavailable, and the standard assumes a census that does not exist for your area.

In each case, adapt and document. What you must not do is keep the standard’s name, change its guts, and let a reader assume comparability that is not there.

Adopting a standard definition is not a loss of analytical independence. It is what makes your number legible to everyone who has to act on it.

What comes next

You have a definition, a denominator and a standard to compare it against. The last unit is about the two numbers placed beside it — the baseline it started from and the target it is supposed to reach — and what each of them does to the behaviour of the people being measured.