Validation rules stop impossible answers at the moment they can still be fixed: while the respondent is present. Used carelessly, they do the opposite. A rule that cannot be satisfied truthfully teaches interviewers to enter whatever the questionnaire accepts, and the result looks like real data.
Three kinds of checks
| Check | What it does | Use it for |
|---|---|---|
| Required answer | The interview cannot continue until the question is answered | Questions every eligible respondent can answer truthfully, such as the main outcome, or the consent decision, where declining is a valid answer |
| Hard limit | Rejects a value outside a range or format | Values that are impossible under the study's definitions: an age of 250, a 13th month |
| Soft check (warning) | Asks the interviewer to confirm an unusual value | Rare but possible values, such as a very large household or a very high price |
Required does not mean answerable
A required question works only if every respondent who sees it can give a true answer. Where “Don't know” or “Refused” are real possibilities, include them as answer options. Otherwise a required question forces a guess.
- Require a recorded consent decision, never consent itself. Declining must always be a valid answer, and the questionnaire should then end or continue exactly as the study protocol says.
- Require eligibility answers and the key variables your analysis depends on.
- Offer explicit nonresponse options where respondents may not know or may decline.
- Make sure a required question can always be reached: required settings and skip conditions must not contradict each other.
Set limits from the study's definitions and evidence
Hard limits should exclude only what is impossible under the study's own definitions and eligibility criteria. If only women aged 15 to 49 are eligible for a module, the age check belongs in the eligibility screen, not in a later limit. Plausible but rare values belong in a soft check, so that a genuinely large household is recorded with the interviewer's confirmation instead of being rejected. Base the values on defensible evidence such as census data, earlier survey rounds or pilot results, and record their source.
Identifiers are text
Phone numbers, ID card numbers and postal codes identify things; they are not quantities. Store them as text with a format check, such as the expected number of digits, so that leading zeros are kept and nobody averages them by accident.
Messages interviewers can act on
Write each message as an instruction that says what is wrong and what to do: “This household size is unusual in this area. Confirm the number with the respondent.” A message such as “Invalid value” leaves the interviewer guessing.
An example
Suppose a survey asks how many people usually live in the household. A hard limit rejects only values that are impossible under the household definition, such as zero members. A soft check asks the interviewer to confirm sizes that census data and the pilot show to be rare in the study area. Such a check might flag a few very large households that turn out to be real extended families; they are kept, and the confirmation is recorded. The values come from the study's own definitions and evidence; there is no universal threshold.
Limitations
Validation catches implausible values, not wrong ones: a plausible but incorrect answer passes every rule. High-frequency checks, back-checks and supervision are still needed.