Assess Secondary Data for a New M&E Question

Evidence and applicability: use only supplied evidence for historical facts, values, quotations, completed checks and approvals. Cite source labels and row, page or section references. Distinguish observations, source claims, calculations, assumptions and proposals. Keep missing evidence unknown or not evidenced, not proof an activity never occurred. Preserve conflicts and their provenance. Do not invent data, dates, permissions, authority, agreement requirements or sign-off. Apply only supplied applicable obligations. A recommendation is conditional on unresolved checks and the responsible owner's decisions; it does not authorize collection, reuse or publication. Descriptive change alone does not establish program impact.

Use only material supplied in this conversation. Ask me for the new question and proposed claims, original collection purpose and codebook, safe source extracts or substantive summaries, version and processing history, population/coverage information, and documented terms governing the intended reuse. Do not request personal data when metadata or safe aggregates suffice. If source contents or use terms are absent, keep substantive findings or permission status unresolved rather than assuming them from access.

Assess whether the supplied secondary source is fit for each proposed new M&E question and claim. Separate analytic suitability from permission to use the source.

1. Define the new question, intended decision and audience, proposed construct/indicator or estimand, target population, unit, geography, time window and level of inference. Request these if absent. Inventory the actual extract/version, original collection purpose and process, custodian if documented, codebook/instruments, collection/reference dates, extraction date, coverage rules and supplied provenance/processing records. A platform name, source reputation, access credential or dashboard label does not establish the underlying data or its suitability. If substantive material is absent, provide only targeted requests and a conditional assessment plan.
2. Build a claim-to-source crosswalk. For each proposed claim identify the actual fields/items, original definition and unit, operational coding, denominators, time periods and inclusion rules needed. Explain direct fit, defensible proxy with limitations, incompatibility or unknown fit. For example, service records may describe recorded service users without measuring prevalence among all residents. Do not equate records, events, households and distinct people, or infer an unobserved construct from a convenient available field. Make any proposed derived measure explicit, with its formula, assumptions and needed evidence.
3. Examine original selection and reporting processes against the new target. Identify who/what enters and is missing, duplicates or repeat events, incentives to report, access/registration exclusions, changes in eligibility or capture, missing reports and nonresponse where relevant. Separate source-covered units from the intended population. Use a compatible denominator only when its coverage, unit and time match. Do not assume administrative records are a census, a large dataset is representative, a missing record means no event, or a service-use rate is a population outcome. State how aggregation or different geographic boundaries limit individual or subgroup inference.
4. Trace extraction, cleaning, recoding, deduplication, linkage and aggregation already documented, separating them from proposed transformations. Preserve raw provenance and versions, identify transformations needing source variables or verification, and retain conflicting definitions. Assess definition and collection changes before proposing a time comparison or pooled estimate. Calculate only supported quantities from authorized supplied inputs with reproducible rules and references; do not invent old values, linkage matches or weights. No record linkage or individual-level recovery is authorized by this assessment. Data quality checks alone cannot establish construct validity or causal identification.
5. In a separate permitted-use column, summarize only the supplied consent/purpose terms, data-sharing agreement, license, custodian restrictions and relevant institutional requirements for this specific new purpose, environment, linkage, retention and intended disclosure. Mark each applicable item evidenced, conflicting or unknown with its source, and identify clarification needed from the responsible owner without inventing an authority. Technical access, public availability, pseudonymization and analytic fitness do not by themselves establish permission or safe disclosure. Do not infer a legal basis or certify compliance. If permission for sensitive use is unclear, assess from safe metadata/aggregates and leave that use pending clarification; do not request restricted records or treat this prompt as permission.
6. Deliver a question/claim-specific reuse matrix with analytic suitability, coverage and comparability limits, permitted-use evidence status, required transformations/checks, unresolved items and conditional recommendation. Distinguish supported narrower use, use pending evidence/conditions, incompatible use and not assessable from supplied evidence; an analytic recommendation never overrides an unresolved permission issue. Explain which claims are unsupported and whether a different question, source or prospective collection would be needed. Include a prioritized verification plan and a reporting/provenance note that acknowledges the original collection purpose, version, exclusions and limits. Do not turn observed associations into causal effects or call an advisory assessment approval to reuse or publish.