Interviewer Effects in a Telephone Survey: A Word to the Wise

Abstract

Interviewer effects, that is, systematic differences in the data collected by different interviewers, are a relatively common hazard of social science research (Bradburn, 1983; Kintz, Delprato, Mettel, Persons & Shappe, 1965; Selltiz, Wrightsman & Cook, 1976). Surprisingly, some nursing research texts pay little attention to this problem (e.g. Polit & Hungler, 1983; Wilson, 1985). Even when nursing researchers are alerted to this potential source of invalidity, when confronted with the practical difficulties of gathering data from large numbers of subjects, the possible impact of interviewer differences is often ignored (e.g. Hash, Donlea & Walljasper, 1985). What follows is the description of a study in which the interviewer variable was not considered until the analysis phase of the research, and in which interviewer differences proved to have a pervasive influence on study outcomes. It is hoped that, by presenting this result, other researchers will be reminded of the potential biases that can occur even when experienced and well-trained interviewers are involved in data collection. The present research was conducted during the planning phases for a Quebec-wide health survey. In designing this survey the question of the stability of psychological symptoms assessed with the major survey index of mental health, the Psychiatric Symptom Inventory (PSI; Ilfeld, 1976), was of particular concern. Although itself relatively new, the 29-item PSI is a brief version of the well-known Hopkins Symptom Distress Checklist (Derogatis, Lipman, Covi, Rickles & Uhlenhuth, 1970) and was designed to assess psychoneurotic symptoms in community surveys. Initial validity studies conducted on a large community sample found that PSI symptom levels were significantly related to "having sought out professional help for emotional problems, having recently used psychoactive drugs, and interviewers' ratings of respondent's degree of tension" (Ilfeld, 1976; p.1215). In spite of this promising validity data, prior to the present study test-retest reliability had