A baseline is not a formality. It is the measurement contract for the whole program. Here is how to design one that will still be usable at endline.
Start from the endline, not the baseline
The most common baseline failure is designing it as a standalone data collection exercise. A baseline exists to make a future comparison possible, so the design question is always: what will we need to be able to say at endline, and what must we measure now to say it credibly?
That means indicator definitions, recall periods, units of analysis and sampling frames should be fixed at baseline with the endline already in view. Changing a denominator two years later quietly destroys comparability.
Get the sampling frame right before anything else
In Kenya, sampling frames vary widely in quality between counties, sub-locations and program registries. Program beneficiary lists are often incomplete, duplicated or out of date, while national frames may not align with program catchment areas.
Invest early in reconciling the frame. Decide whether the study is representative of the program population or of a wider geography, document the decision, and state the resulting limits on generalisation in the report.
Design instruments for the field, not the desk
Long questionnaires reduce data quality. Every additional module increases interview fatigue, response error and cost. A disciplined baseline measures the indicators in the results framework plus a limited set of explanatory variables, not everything that might be interesting.
- Translate and back-translate into the languages actually spoken in the sampled areas
- Pilot in conditions similar to the real field, not in the office
- Build skip logic, constraints and validation rules into the digital tool
- Pre-code enumerator instructions for sensitive or ambiguous questions
Plan quality assurance as part of the design
Back-checks, spot checks, GPS verification and daily data review should be specified in the inception report with targets, for example a defined percentage of interviews back-checked per enumerator per week. Quality assurance that is improvised mid-fieldwork rarely holds.
Finally, write the analysis plan before data collection starts. If you cannot specify how an indicator will be calculated, you are not ready to measure it.