Calculated fields save interviewers from mental arithmetic and give analysts ready-made variables: a household's total income, a child's age in months, a score built from several answers. They also fail quietly. A calculation that is empty or wrong in a few interviews can go unnoticed until analysis, when those interviews can no longer be repeated.

The research problem

Most calculation errors have a few causes: inputs that are not answered, numbers stored as text, divisions by zero, rounding at the wrong moment, and calculations evaluated before their inputs are available. Each is easier to prevent at design time than to repair afterwards.

Unanswered inputs

A question can be unanswered because the respondent skipped it, because skip logic hid it, or because the interviewer has not reached it yet. A sum that includes an unanswered amount has no meaningful value. Tools differ in what they do with it: some leave the result empty, others treat the blank as zero, which silently understates the total.

  • Decide for each calculation what a missing input means: the result is unknown, or the input genuinely counts as zero.
  • If zero is a real answer, ask for it. Offer 0 as a valid response instead of relying on a blank.
  • Handle divisions explicitly: a share or rate with a zero denominator should be guarded, not left to produce an error or an empty value.

Text that looks like a number

An amount typed into a text field is text, even when it reads “25”. Depending on the tool, adding it to a number fails, joins the characters together or returns nothing. Collect amounts, counts and ages in numeric questions, and keep text questions for identifiers such as phone numbers, where leading zeros must be kept.

Round once, at the end

Rounding intermediate results makes totals drift. If a price per unit is rounded before it is multiplied by the quantity, the household total can be off by more than the rounding suggests. Calculate with full precision and round only what is displayed, or leave rounding to the analysis.

When calculations are evaluated

Tools differ in when they evaluate a calculation. Some follow the order of the questionnaire, some recalculate whenever an input changes, and some evaluate again when the interview is finalized. A result shown on an early screen can therefore be blank or provisional even when the final saved value is right, or the other way round. Learn how your tool handles dependencies and recalculation, place calculations after their inputs where order matters, and test the value at every point where it is used, such as in a question text or a skip condition.

An example

A survey computes monthly income per adult as household income divided by the number of adults. In one interview, the respondent does not give income from casual work; in another, no adults are recorded because the roster was skipped. Without explicit handling, the first case may be recorded as if casual income were zero, and the second divides by zero. Asking “Did anyone in the household earn income from casual work?” before the amount, and checking that at least one adult is listed before dividing, turns both into results an analyst can interpret.

Before fieldwork

  1. Test each calculation with complete answers, with a skipped input and with boundary values such as zero.
  2. Show key results to the interviewer where a wrong value would be noticed, such as a total the respondent can confirm.
  3. Document every calculated variable: its formula, its unit and how missing inputs are handled.