Mode Effects

Interviewer Presence

Mode differences can be thought of in terms of dimensions on which modes differ. One of these dimensions is the degree of interviewer involvement. A purely self-administered mode (e.g. a paper-and-pencil survey that is mailed to respondents) removes this component completely.

The characteristics of the interviewer and respondent can impact responses to some kinds of questions in studies in which an interviewer is involved in the data collection. For example, it has been found that the match between the race of the interviewer and the race of the respondent can influence responses to racial attitude questions. More moderate answers have been found to be reported by African American respondents to white interviewers than to African American interviewers. Similar results have been found for the gender of the interviewer. The match of the gender of the interviewer and respondent may be important, as more “feminist” answers are reported to female interviewers.

Interviewer variance is another type of interviewer-related mode effect that arises when responses of respondents interviewed by the same interviewer tend to be correlated with each other. This phenomenon is found in both forms of interviewer-administered surveys (face-to-face surveys and telephone surveys), and it contributes to the variance component of statistical error. Interviewer variance, noted by pint, is a specific application of the intraclass correlation. If pint = 1.0, then responses within an interviewer's set of respondents are completely correlated. If it is zero, they are completely uncorrelated. The study of interviewer variance requires an interpenetrated design in which interviewers are randomly assigned to respondents so that natural intraclass correlation (e.g. those due to neighborhood or region) can be separated from intraclass correlation caused by the interviewer. For this reason, few valid studies have been done. Nonetheless, ρint values tend to be higher for attitude questions than for factual questions. They are also found in open-ended questions, when interviewers need to probe in order to get further responses. The absolute values of ρint tend to be fairly small, though they have been found to be larger in face-to-face interviews than in phone interviews. However, the absolute value is of less concern than the impact on the error of a statistic. The impact of ρint on a given statistic is determined by the size of the interviewer's workload. The impact of interviewer variance on a statistic is , where is the average interviewer workload. This form of mode-related variance is not found in self-administered data collection modes.

...