Response Effects in a Network Survey

A reverse record check study was conducted to measure the response effects in a network survey designed to locate a rare population subgroup. In this field experiment, diagnosed cancer patients and certain specified relatives were interviewed using a survey instrument based on the National Health Interview Survey questionnaire and a supplemental set of questions designed to identify persons with cancer. The measurement errors associated with obtaining reports on cancer diagnosis and the date of diagnosis are presented. In general, high rates of patient identification were found in both patient and relative households; in only 6% of the patient-relative pairs was the patient not reported. Females and Whites were more likely to be reported than males and non-Whites. A significant amount of variation and error in reporting date of diagnosis was found for both the patient and the relative households. Contrary to expectations, most households were likely to backward rather than to forward telescope the date of diagnosis. Suggestions on how to deal with error due to memory factors are proposed.