01 / 06 — The brief
The client needed primary field evidence — not a repackaging of existing data.
An institution needed primary data from a target population that existing published sources did not adequately cover. The evidence gap was specific: existing data either used different sampling frames, asked different questions, or was too dated to support the decision the organisation needed to make. Primary collection was necessary.
MMK's task was to design a system that would produce reliable, usable evidence: from instrument design through field deployment logistics and data structure.
Primary collection was justified by a specific evidence gap — not chosen by default.
02 / 06 — Instrument design
Every question in the instrument was linked to a research objective.
The questionnaire was built using a question-objective mapping process: each question was required to connect to at least one stated research objective, and each research objective needed to be covered by at least one question. This prevented the instrument from accumulating questions that were interesting but not decision-relevant.
The final instrument was piloted with a small group of representative respondents before field deployment. Piloting identified three questions with interpretation ambiguity and two with response-option design problems — all fixed before data collection began.
Question-objective mapping + pilot testing before deployment — not after.
03 / 06 — Sampling design
Sampling rules were clear enough for any enumerator to apply without improvisation.
The sampling protocol specified: target respondent criteria, selection process within each sampling location, replacement rules when target respondents were unavailable, and documentation requirements for refusals. The protocol was written in plain language and tested against hypothetical field scenarios to verify it produced consistent decisions without supervisor input.
Sampling documentation was structured so that the composition of the final sample could be fully reconstructed from field records — a requirement for any evidence that would face external scrutiny.
Sampling protocol: field-testable, fully documentable, no improvisation needed.
04 / 06 — Enumerator preparation
Training covered interpretation, not just form completion.
Enumerator training was structured around two objectives: ensuring consistent question interpretation across the field team, and ensuring that enumerators knew what to do when situations arose that the protocol had not explicitly anticipated. The training included read-aloud practice of every question, mock interviews with structured feedback, and a session on common problematic scenarios from comparable field operations.
Training emphasis: consistent interpretation and protocol for unexpected situations — not just tool navigation.
05 / 06 — Quality control
QC was built into the collection process, not applied at the end.
A daily supervisor review covered: completion rate against target, time-per-interview distribution (flagging unusually fast completions), skip logic errors, and a spot-check of five to eight completed instruments per enumerator. Back-checks were conducted on eight percent of completed interviews — a random sample using a shortened re-contact questionnaire that asked the same questions in a different order to detect fabricated responses.
Daily review + eight-percent back-check rate — QC built into the process, not bolted on at the end.
06 / 06 — Data structure and deliverable
The data package was structured for analysis, not just archiving.
The final data package included: the clean dataset with variable documentation, a data dictionary explaining each variable's source question and coding logic, a field operations log recording sample composition and QC actions, and a brief cleaning report documenting every decision made in preparation of the analysis dataset — missing value treatment, outlier handling, and derived variable construction.
The package was handed over in a format that allowed a different analyst to work with the data without a handover call.
Deliverables: clean dataset, data dictionary, field log, and cleaning report — structured for independent analysis use.
