RMResearch Metrics AfricaMethodology & trustStart a project

HOW THE WORK IS CONTROLLED

Trace the evidence.
Keep the limits visible.

RMA treats the process as part of the deliverable. A serious project should show what came in, what changed, why it changed, which method was used, what remains uncertain and who reviewed the result.

THE CONTROLLED LIFECYCLE

Audit. Clean. Analyse. Validate. Report.

01

Triage

Define the decision, source, deadline, sensitivity and constraints. Return Fit, Needs clarification or Out of scope.

02

Scope

Document inputs, work, deliverables, timeline, price logic, exclusions and responsibilities before material work begins.

03

Protect

Preserve the raw source and create timestamped working, clean and final versions rather than overwriting evidence.

04

Audit

Check completeness, duplicates, coding, ranges, logic, dates, outliers and available field-quality signals.

05

Analyse

Map every objective to variables, descriptive output, method, assumptions and a fallback before interpretation.

06

Validate

Require human review of method, diagnostics, interpretation, limitations and final delivery state.

07

Report

Deliver an executive layer, technical annex, reproducibility assets and an Evidence Passport appropriate to scope.

Data Integrity Score

A summary for prioritisation—not a certificate.

The 0–100 score uses only applicable dimensions. A missing GPS field is not silently treated as perfect when GPS was never collected. Each sub-score remains connected to its underlying finding and the detailed issue register always takes precedence.

Completeness

Missingness and required-field completion

Consistency

Cross-field and logical agreement

Duplicate risk

Repeated rows or identifiers requiring review

Range validity

Values outside declared boundaries

Skip logic

Unexpected pathways or answers

GPS integrity

Missing or implausible location signals where lawfully collected

Timing

Unusually short, long or concentrated submissions

Enumerator patterns

Concentration, repetition or straight-lining signals

Statistical QA

A model output is not automatically a conclusion.

ASSUMPTION GATE

Three explicit decisions

  • Appropriate
  • Proceed with limitations
  • Alternative recommended
ANALYST VALIDATION

Accountable human review

Method choice, diagnostics, exclusions, interpretation and material limitations are reviewed before a high-impact conclusion is released.

SECOND OPINION

Independent challenge

A distinct reviewer can examine suitability, assumptions, interpretation, unsupported claims and missing limitations.

Evidence Passport

A one-page chain of evidence for the delivered project.

The authenticated passport records dataset identity and version, checks completed, issue status, analysis-plan and assumption state, cleaning/reproducibility availability, report version, human QA ownership and a verification ID—without publishing confidential data.

Reproducibility & privacy

Enough evidence to review the work, no more personal data than necessary.

Versioned evidence

Depending on scope: raw reference, clean dataset, cleaning log, codebook, analysis plan, syntax/code, tables, charts and reports.

Named ownership

Prepared, reviewed and approved roles are shown honestly. RMA does not invent a team, credential or independent reviewer.

Minimum data

Public tools use anonymised input and explain their limits. Identifiable participant data belongs only in an agreed secure workflow with lawful purpose and bounded access.

Signals, not accusations

GPS, timing, duplicate and enumerator patterns are review signals. They are not automatic proof of fraud, intent or misconduct.

PRACTICAL TEMPLATES

Use the same questions before you buy a service.

Print the free checklists and planning templates, then bring the completed record into a project intake.

Open five research templatesUse the three public toolsRead privacy & security