Performance indicator reference sheets (PIRS)

What a PIRS fixes

An indicator’s published value rests on dozens of small decisions: who counts, from which records, over what period, with which borderline cases in or out. None of those decisions appear in the indicator’s name. “Number of farmers adopting improved practices” reads identically whether farmer means landholder or household member, whether adopting means once or sustained across a season, and whether the count is new adopters or a running total — six defensible indicators wearing one name (see Indicator types and levels for the reporting-period arithmetic alone).

The performance indicator reference sheet is where those decisions are made once, written down, and governed. Its absence produces three failures with long lag times, which is why they surface as crises rather than bugs:

• Silent definition drift. Field offices resolve ambiguities differently, or one office quietly changes its reading between years. Every value remains individually plausible; the aggregate and the trend become fiction. This is precisely the reliability failure that data quality assessments are built to detect [2] — detection that a PIRS makes unnecessary.
• Turnover loss. The reporting officer who knew what the indicator “really meant” leaves. Their successor reconstructs a meaning from the name — a different meaning. The time series now has a seam no one can see.
• Apples-to-oranges aggregation. Values computed from different sources or rules are summed into a portfolio figure. The arithmetic works; the number means nothing.

The PIRS is the standards-layer artefact; the practice of defending indicator data when a donor’s auditors arrive — source documents, filing discipline, the audit trail — is the subject of the Monival practice companion, and the formal assessment instruments that test data against the sheet are covered under data quality assessment.

The canonical fields

The field set below follows the structure USAID recommends for reference sheets on its own indicators [1] — the most widely copied template in donor-funded practice — and matches the system-design decisions Kusek and Rist place at the centre of a results-based M&E system [3].

Field by field, what good looks like:

1. Indicator name — exactly as it appears in the results matrix, so the sheet and the matrix cannot diverge. The name states what is measured, never the target (the grammar rule from Types and levels).
2. Precise definition — the load-bearing field; treated in its own section below.
3. Unit of measure — including, for monetary indicators, currency and magnitude (“KES, absolute” vs “KES thousands”): the classic silent-aggregation killer.
4. Disaggregation — the dimensions (sex, age band, disability, geography and any programme-specific cuts) with their exact categories. Deciding this at design time is what makes disaggregated reporting possible at all; the design reasoning sits with baselines, targets and disaggregation.
5. Data source — the named artefact (the register, the form, the dataset), not a genre. “Facility records” is not a source; “the ANC first-visit register, MOH 405” is.
6. Collection method / construction — how the value is computed from the source, including numerator and denominator rules for percentages and the treatment of duplicates and repeat participants.
7. Reporting frequency — with the aggregation rule across periods (cumulative, snapshot, milestone) stated explicitly.
8. Responsible party — a role, kept current, so the sheet survives reorganisations.
9. Known data limitations — recall bias, incomplete coverage, self-report, denominators estimated from projections. Stated plainly, with any mitigation.
10. Change log — every revision, dated, with what changed and from which reporting period it takes effect.

The lightweight version. A six-person NGO does not need USAID’s full apparatus, but there is a floor below which the sheet stops working: definition, unit, disaggregation, source, frequency, owner. Those six fields are the minimum that lets a new officer produce the same number the old officer produced. Everything else is proportionate elaboration.

Writing definitions that survive a handover

The definition field fails when it paraphrases the name. It succeeds when it decides the borderline cases in advance — which means it is written as rules, not prose description. A useful drafting test, borrowed from field-measurement practice [4]: hand the definition and three deliberately awkward cases to two colleagues who did not write it, and check they classify all three identically.

A worked pair:

Weak: “Number of youth trained in entrepreneurship skills.”

Strong: “Number of distinct individuals aged 18–35 (at first session) who attended at least 6 of the 8 sessions of the entrepreneurship curriculum in the reporting quarter. Each individual is counted once per lifetime of the project, in the quarter they first reach 6 sessions; repeat attendance in later cohorts is not recounted. Attendance evidence: signed session registers. Excludes: facilitators, project staff, walk-ins not enrolled in a cohort.”

The strong version has made five decisions — age boundary and its reference date, completion threshold, deduplication rule, evidence artefact, exclusions — that the weak version leaves to whoever reports next quarter. Note what it has not done: no target, no direction of change, no aspiration. Definitions measure; targets sit elsewhere in the results architecture.

The sheet as a governed artefact

A PIRS that can be edited casually is a PIRS that will be edited casually. Two governance rules give the document its value:

• Versioning with effect dates. Definition changes are sometimes right — a threshold proves wrong, a source is replaced. The requirement is not immutability but a logged change: what changed, why, approved by whom, effective from which period. The log is what lets an analyst — or an evaluator — interpret a discontinuity in the series as a definition seam rather than a programme effect.
• One authoritative copy. The moment sheets circulate as email attachments, the organisation has several PIRS per indicator and does not know it. A single stored, access-controlled location — and ideally a system that enforces the sheet’s unit, categories and frequency at data entry rather than trusting reporters to consult it — closes the gap between the documented definition and the collected data.

The payoff for this discipline is collected downstream. When a routine data quality assessment traces reported values back to source [2], the PIRS is the standard the assessment tests against; where no sheet exists, the assessment has nothing to verify against, and every office’s practice is unfalsifiably “correct”. Documented limitations, similarly, convert what would be audit findings into disclosed design facts — the difference between “the programme knew and said so” and “the evaluators discovered”.