Methodology and Research Standards

Research begins with the question and the decision the finding is intended to support—not with the instrument. Concepts, population, time, and geography are reviewed before selecting design, data sources, sample, instrument, and analysis plan.

1. Define the problem, purpose, and intended users.
2. Formulate questions, objectives, concepts, and indicators.
3. Select design, data sources, and sample.
4. Develop, translate, review, and test instruments.
5. Prepare fieldwork, training, quality, and ethics plans.
6. Collect data, monitor implementation, and document deviations.
7. Clean, code, weight, and analyze data.
8. Review findings, tables, narrative, and limitations.
9. Prepare deliverables and methodological disclosure.
10. Publish or deliver and manage corrections and data.

• Define population, frame, and unit of selection.
• Determine size and allocation according to precision, comparison needs, and resources.
• Document exclusions, nonresponse, and replacement.
• Apply and describe weighting where needed.
• Do not call a sample representative unless the design supports it.
• Use caution with small cells and multiple comparisons.

• Clear, neutral language appropriate to participants.
• One concept per item where possible.
• Balanced and reasonably exhaustive response options.
• Ordering that reduces leading and context effects.
• Logical, technical, and cognitive testing.
• Version, translation, and change documentation.

Quality controls differ by collection mode and may include training, testing, observation, verification calls, duration and logic checks, duplicate and location checks, authorized audio review, and open-response review. Automated indicators are not used alone for punitive decisions without human review.

• Analysis plan tied to research questions.
• Clear indicators, rules, and missing-value treatment.
• Attention to sample design and uncertainty.
• Distinction among description, association, and causation.
• Alternative specifications and sensitivity analysis where needed.
• Presentation of meaningful differences without selection.
• Interpretation connected to context and limitations.

• Problem, purpose, and intended users.
• Primary and secondary questions, concepts, and indicators.
• Population, unit of analysis, and geographic and temporal scope.
• Data sources, sampling design, and collection mode.
• Analysis plan, deliverables, and methodological disclosure.
• Quality, ethics, data-protection, and safety plan.
• Roles, approvals, schedule, and scope-change rules.

A sampling frame is assessed for coverage, duplication, currency, and accessibility. Where a complete probability frame is unavailable, the alternative and resulting limitations are explained. Depending on the question, studies may use stratified, clustered, systematic, multistage, purposive, theoretical, or typical-case sampling.

• Define contact outcomes, ineligibility, refusal, and partial completion.
• Record attempts, timing, and outcomes consistently.
• Do not automatically replace a selected unit without an approved rule.
• Examine respondent and nonrespondent differences where feasible.
• Use weights or sensitivity analysis where appropriate.
• Report response or completion indicators understandably.

Translation is reviewed for conceptual equivalence and comprehensibility rather than literal matching alone. The process may include initial translation, bilingual review, back-translation where useful, cognitive testing, and documentation of terms without a direct equivalent.

• Variable dictionary and consistent naming convention.
• Untouched source copy and dated working versions.
• Cleaning, coding, transformation, and indicator-derivation log.
• Independent code or table review where feasible.
• Separation of identifiers from analytical data.
• Backups, permissions, and deletion according to plan.

Analysis should be reproducible through code or a clear processing record, with fixed data versions and rules. Where confidentiality prevents data or code release, a methodological description or intermediate tables may support reasonable review without exposing restricted information.

• Identify whether integration is concurrent or sequential.
• Explain whether one component explains, builds on, or validates another.
• Do not erase conflict among sources without examination.
• Connect each conclusion to the source or component supporting it.
• Present convergence, divergence, and evidence gaps.

Margins of error or confidence intervals are reported where appropriate, together with design effects, weighting, and small-cell limitations. Sensitivity analysis may test missing data, definitions, rules, or assumptions. Excessive numerical precision should not conceal genuine uncertainty.

After delivery or publication, problems, deviations, lessons, and required updates are reviewed. A close-out record or learning meeting may distinguish individual error from a system issue and recommend an implementable change to templates, training, or systems.