The Challenge of Using Data about Household-level Characteristics obtained from Multiple Informants: Experience in Rural Alaska
METHODS: During an investigation into a skin infection epidemic in rural Alaska, we conducted a Knowledge-Attitudes-Behavior survey of adults in three communities. Surveys were administered in-person from a standard questionnaire. A subset of the questionnaire contained questions about the household (HH) unit in which the participant resided. Since different members of one HH were surveyed, we were able to compare the responses to these questions among multiple respondents from the same HH. Differences were assessed by means of kappa statistics (κ) and intra-class correlation (ICC). All reported κ-values have P<0.001.
RESULTS: The study included 126 HH with more than one respondent. Among the dichotomized variables the greatest agreement was for questions about the presence of piped water to the HH (κ=0.94) and use of self-hauled natural water (κ=0.82). Substantial agreement was also seen in whether the HH contained 6 or more persons (κ=0.72), or one or more children under 6 years of age(κ=0.68) , and use of community laundry facilities by the HH (κ=0.75). Less agreement was seen for presence of someone in HH who was diagnosed with a furuncle in the last 6 last months (κ=0.44) and use of self-hauled treated water (κ=0.46). Intra-class correlation of additional variables, showed generally similar results in frequency of laundry (ICC=0.44, P<0.001) and water re-use for laundry (ICC=0.60, P<0.001), and lesser correlation for steambath cleaning frequency (ICC=0.10, P=0.135).
CONCLUSIONS: The use of in-person interviews to collect household data can result in varying responses dependent on who is being questioned. Use of alternative data sources or targeting of ideal informants may be considered.