A household roster lists the people in a household and asks the same questions about each of them. Survey tools implement it as a repeat group, also called a roster or loop. Rosters carry much of a household survey's analytical weight, from household size to school attendance, and a roster error spreads into every indicator built on it.
Start with a membership rule
Decide who counts as a household member before writing questions. The definition and its reference period come from the study protocol: for example, people who usually live and eat together, counted over a stated period. If results will be compared with a census or another survey, use their definition or document the difference. Put the rule in the question text and the interviewer manual, and train on borderline cases: students away at school, seasonal migrants, domestic workers, newborns and recent deaths.
Number members and keep identities stable
- Give each member a line number when they are listed, and keep it for the whole interview. If a person was listed by mistake or twice, mark the line as removed, with a reason, instead of deleting it and renumbering the others.
- Never reassign a line number to a different person within an interview. Later questions, such as “Who is this child's mother?”, refer to these numbers.
- Record each member's relationship to the head of household and, where needed, the line numbers of parents and spouses, so that family structure can be rebuilt in analysis.
- Within one round, the household ID together with the line number identifies each member's row. Document any correction to either.
- A line number is a position in one round's list, not a permanent identifier. In a panel survey, line numbers may legitimately change between rounds as members join, leave or are listed again. To link people across rounds, give each person a stable person ID that is carried from round to round, and document every change to it, such as a correction or a merged duplicate. Never silently give an existing ID, or a line number, to a different person.
Make counts agree
If the questionnaire asks “How many people live here?” and then lists them, the two numbers must match. Either derive the count from the roster, or compare the stated count with the listed members and ask the interviewer to resolve differences while still in the household. The same applies to totals across members, such as the number of children under five or the sum of individual incomes: calculate them from the roster instead of asking for them again.
Ask only what applies
Many roster questions apply to some members only: schooling for children, employment for adults. Filter these questions by age or relationship inside the roster, and record nonresponse explicitly, so that “not asked” can be told apart from “not answered” in the data.
Plan the shape of the data
Roster data is naturally long: one row per member, linked to the household by its ID. Wide files with one column per member and question are hard to analyse and break down for large households. Before fieldwork, check that every member row links to exactly one household and that member identifiers are unique.
An example
A household reports six members, but the roster lists five: a grandmother who lives with the family half of the year was left out. A count check alerts the interviewer, who applies the study's residence rule with the respondent. If the grandmother meets the rule, she is added with the next line number; if not, the stated household size is corrected. Either way, the decision is made in the household, not guessed later.
Before fieldwork
- Pilot the membership rule with real households and refine the borderline cases.
- Test the roster with a one-person household, a large household, and a person who is listed by mistake and then removed.
- Open a test export and check that member IDs are unique, every member links to a household and the counts are consistent.
- In a panel, test that person IDs from the previous round carry forward correctly, including for members who have left and for new members.