|
Abstract:
|
Computer Assisted Recorded Interviewing (CARI) has long been used by field management to monitor interviewer performance and to assess questionnaire items. Prior research found that providing feedback to interviewers based on CARI was effective at improving interviewer performance. Conventionally, a coder needs to first listen to the audio recording of the interactions between the interviewer and the respondent, and then evaluate and code features of the question-and-answer sequence using a pre-specified coding scheme. Such coding tends to be labor intensive and time consuming. In practice, often a small proportion of completed interviews or a selected group of questionnaire items can be evaluated in a timely manner. In this study, we will present a pipeline we developed at Westat that heavily draws on the use of machine learning. We will present how to use the pipeline to detect potential interviewer falsification and identify interviewers with undesirable behaviors. We will evaluate the performance of the pipeline using both mock interviews and field interviews. We will also discuss the time and cost implications of using the pipeline as compared to conventional human coding.
|
