Baselines set targets, evaluations decide whether programmes continue, and verification triggers payments. All of them rest on field data. Before you accept a survey's numbers, ask these six questions.
1. Does the sample represent the people the decision is about?
Ask how the list of people or places to sample from (the sampling frame) was built, and who it leaves out. Check that the sample is large enough for the comparisons the report makes, especially for sub-groups such as women, young people or refugees. Clustered samples, such as villages, need larger sizes than simple random ones.
2. Was the questionnaire tested and translated properly?
A question that works in English can mean something different in Luganda, Acholi or Ateso. Good surveys translate tools, translate them back to check meaning, and pilot them with real respondents before fieldwork starts. Ask to see the pilot notes and what changed afterwards.
3. Were the enumerators trained and tested?
Training should cover every question, consent, the tablet or phone, safety and safeguarding, with practice interviews and a test at the end. Only those who pass should go to the field. Ask for the training agenda and the list of who passed.
4. Were the data checked while fieldwork was still running?
Digital surveys allow daily checks on interview length, GPS location, time stamps, skipped questions and unusual answers by enumerator. Problems found on day two can be fixed. Problems found after fieldwork usually cannot.
5. Were some interviews done again?
Back-checks mean a separate team re-visits a random share of respondents, often around one in ten, and asks a few key questions again. Big differences point to errors or invented interviews. Ask what share was back-checked and what was found.
6. Can every number be traced from raw data to report?
Cleaning and analysis should be done with written code or documented steps, so anyone can repeat them from the raw data. Consent should be recorded and personal data protected, as Uganda's Data Protection and Privacy Act, 2019, requires.
A simple test: ask for the raw data, the cleaning and analysis files, and the field quality reports. A firm that collected good data can share them quickly.
What this means for buyers of research
Write these checks into terms of reference and contracts. Budget for back-checks and piloting; they cost little compared with a decision taken on bad data. And ask bidders to describe their quality controls, not just their sample sizes.