Measurement errors stemming from nonrespondents present at in-person interviews

Data are frequently collected from in-person interviews in epidemiologic studies. Despite the advantages of this mode of data collection, the presence of a third party during the interview can contribute to measurement error, especially if third-party presence is related to case status.

Using data obtained from a case-control study of colon cancer, we evaluated the frequency of third-party presence during in-person interviews, and how having someone else present during the interview influences reporting of exposure data.

Interviews were conducted in the presence of a third party for 28% of cases and 22% of controls who lived in a household of two or more individuals. Men with a third party present reported significantly lower age-adjusted mean levels of alcohol consumption (P < 0.01). Associations, as indicated by odds ratios, between colon cancer and alcohol intake were not statistically different among those with a third party present and those without a third party present. Although not statistically significant, energy intake was more strongly associated with colon cancer among those without a third party present during the interview.

These results emphasize the need to review questions to be asked and decide whether privacy should be emphasized before data collection begins. If privacy is required, interviewers need to be given better skills to ensure privacy during interview.