Define the population and unit of analysis precisely
A household survey needs a clear definition of who is eligible, what constitutes a household and which person should answer each module. These details become especially important when programmes work with pastoralist households, displaced populations, female-headed households, youth, farmers or participants in multiple intervention components.
Eligibility rules should be written into the field manual and, where possible, reflected in the digital tool. Ambiguous respondent selection can introduce systematic bias even when the sample-size calculation is correct.
Build a sampling approach that can be implemented
A statistically sound sampling design must still work with the available frame and field conditions. Clients should establish whether reliable household lists exist, whether cluster sampling is necessary, how replacement or non-response will be handled, and which levels of geographic representation are required.
Sample allocation should also reflect the analysis plan. If the study needs comparisons by region, sex, intervention type or other subgroup, the design needs enough observations in those groups. These requirements should be resolved before field deployment rather than discovered during analysis.
Plan language and localisation early
Ethiopia’s multilingual context makes translation a technical component of survey design. Direct word-for-word translation is rarely sufficient. Concepts, response categories, units, time periods and sensitive terms should be reviewed for meaning and usability in the target language.
A good workflow may include translation, independent review, back-translation where useful, harmonisation and testing during enumerator training. Digital forms should also be checked for layout, special characters and interviewer navigation in every language that will be used.
Use training and piloting to test the full protocol
Training should cover the questionnaire, consent, eligibility, interviewing behaviour, sampling procedures, use of devices, referral or safeguarding procedures where relevant, and what to do when respondents are unavailable. Role plays and supervised practice help identify variation in how enumerators interpret the protocol.
The pilot should test more than question wording. It should also test interview duration, navigation, travel assumptions, respondent availability, supervisor workflow, data syncing and issue escalation. Findings should feed into a final field-readiness decision.
Design field logistics around geography and season
Travel assumptions that look reasonable on a map can be unrealistic in practice. Road condition, rainfall, dispersed settlements, market days, agricultural seasons, pastoral mobility, security and local administrative schedules can affect how many interviews a team can complete safely each day.
Realistic routing and team structure are therefore part of methodological quality. Overloaded daily targets encourage rushed interviews, substitution and weak supervision. A good field plan balances productivity with quality and duty of care.
Monitor the survey while it is happening
Supervision should combine direct field observation with data review. Daily dashboards or review tables can highlight missing data, implausible values, unusual distributions, short interviews, repeated coordinates or enumerator-specific patterns that require follow-up.
The fieldwork plan should specify who reviews these indicators, how quickly questions are returned to supervisors and what evidence is required before a suspicious record is accepted. This turns quality assurance into an operational process rather than an end-of-study cleaning exercise.
Clarify response rates and substitution rules in advance
Clients should agree before fieldwork how non-contact, refusal, temporary absence and ineligibility will be recorded and managed. Pressure to reach a target sample can otherwise encourage informal replacement, which changes the probability of selection and can bias results.
A fieldwork tracker should show the status of sampled households and the reason for every non-completion. Where revisit attempts are required, the protocol should state how many are expected and under what circumstances a replacement can be approved.
Plan the analysis before finalising the questionnaire
The questionnaire should be designed with the intended indicators, tables and comparisons in mind. This reduces the common problem of collecting large amounts of information that are difficult to analyse while omitting variables required for disaggregation or indicator calculation.
An analysis plan can specify core outcomes, denominators, subgroup comparisons, weighting requirements and derived variables. When this is done early, form programming, validation rules and data handover can be aligned with the final reporting needs.