The use of observation techniques has been promoted for the study of hygiene practices; however, questions still remain about the validity and repeatability of such techniques. In this article we compare data on hygiene behaviours obtained from questionnaires with data obtained using a structured observation approach and examine the repeatability of structured observations of behaviours and spot observations of environmental conditions. Poor agreement between questionnaire responses and observations was found for child defecation and stool disposal practices (kappa statistic: 0.25 and 0.28, respectively). There was evidence of over-reporting of "good" behaviours (P < 0.0001). Repeated observations of child defecation and stool disposal behaviours showed better agreement (kappa statistic: 0.76 and 0.62, respectively) based on small sample sizes. These findings suggest that our questionnaire data are less valid than data obtained by direct observation. However, different approaches to questioning may be less prone to over-reporting of "good" behaviours than our approach. Further research into the validity of different forms of question is warranted. Behaviours and conditions related to hygiene vary. Observations may be useful in determining the frequency of different behaviours/conditions in the community. However, individual practices may be too variable to assign individuals to exposed and non-exposed groups for the purpose of identifying links with health outcomes. Further studies on the variability of behaviours and the repeatability of observations are therefore needed.
PIP: The measurement of health behavior has included point-in-time structured questionnaires and structured observation. The correlation between these 2 approaches for assessing hygiene practices has not been found to be high. In this study, questionnaire responses, direct observations of behavior, repeated observations of hygiene behavior, and repeated observations of environmental conditions related to hygiene are compared. The unweighted kappa statistic was used to assess the degree of agreement between questionnaire responses and direct observation, and between repeated observations. McNemar's test and chi-square tests were used for examining general associations and trends. The case control study of childhood diarrhea was conducted in Bobo-Dioulasso, Burkina Faso under between January 15, 1990 and March 31, 1991, among a sample of children under 36 months old who had been admitted to Sanou Souro Hospital. Each child upon discharge was visited at home and matched to a control in the same neighborhood. A precoded data collection form was used during the observation to record behaviors. There were 10 households that were observed on 6 separate occasions out of the 57 (10%) households with repeat observations. Home interviews were conducted for 2775 children. 548 households were involved in direct observation. The findings revealed that agreement between questionnaire responses and observations on child defecation and stool disposal practices were very poor when chance agreement was taken into account. "Good practices" were overreported. Repeated observations were in greater agreement. The results were consistent with the hypothesis that questionnaire responses are less valid. One caveat was the way the question was posed; different results might have been achieved had the question been: "What happened the last time the child defecated?" Another difficulty is the assumption that behaviors are habitual as a justification for an initial observation. The suggestion is that hygiene behaviors are variable and the degree of consistency may vary substantially; an example is given to show the consequences of this variability. Measurement may be useful to gauge a general level of incidence, but not behavioral change or assignment into risk or nonrisk groups. Further research is warranted.