Interviewer voice characteristics and data quality

As an aural mode, interviewer voices play an important part in telephone surveys. Telephone interviewers are typically instructed to read questions with a proper phrasing and inflection and to read questions at a speech rate of 2 words per second (wps). However, there is no study that examines whether these interviewer voices affect data quality. In this dissertation, I examine how interviewer voice characteristics are associated with data quality in socially desirable, undesirable, and complex questions. Data for this study come from the Work and Leisure Today Survey (NSF SES-1132015). I examined the first turn that interviewers read a survey question (n=4,689). Pitch, intonation, speech rate, and disfluencies are both objectively measured by the Praat program and subjectively evaluated by coders. In addition, coders evaluated five interviewer personality traits (expertise, trustworthiness, reliability, confidence, and easiness to understand) from interviewer voices. I examined four sets of data quality indicators including problematic respondent behaviors, item nonresponse, the directional hypothesis of “more/less is better,” and rounding. Analyses showed both objective and subjective voice characteristics affect data quality; however, the effects are inconsistent across data quality indicators. Interviewers obtain better data quality when they read questions with moderate intonation and disfluencies. The voice characteristic with the largest effect on data quality is speech rate. Interviewers obtain better data quality when they read neutral questions with 2 wps, but read socially undesirable questions more quickly. Results suggest that interviewers should be trained to read questions with moderate intonation and disfluencies. In addition, to maximize data quality, interviewers should read neutral questions with the recommended speech rate of 2 wps, but read socially undesirable questions more quickly. I also found that listeners can perceive interviewers’ personality traits (credibility and easiness to understand) from interviewers’ voices, and these personality traits tend to affect data quality. Credibility affects data quality in sensitive questions while easiness to understand affects data quality in complex questions. In addition, I found credibility mediates the effect of speech rate on respondents interrupting questions with answers. Moreover, easiness to understand mediates the effects of intonation and speech rate on item nonresponse rates