Paradata and How Interviewers Perceive Interview Quality – an Explorative Study Based on the Social Survey Austria 2021

Despite advances in computer-assisted web interviewing (CAWI), face-to-face (F2F) interviews remain widely regarded as the gold standard for collecting high-quality survey data. This is primarily due to the interviewer's ability to control the interview environment, monitor social surroundings, and ensure that questions are correctly understood by respondents. In particular, the interviewer’s capacity to interact directly with respondents allows for clarification of misunderstandings and supports more accurate responses.

While interviewer effects can influence data quality researchers employing F2F methods often implement strategies to mitigate such biases (e.g. interviewer training, standardization of interview protocols) as well as plausibility checks (e.g. such as cross-referencing education levels with respondents' ages or comparing income with living situations) to assure data quality.

However, (1) how do interviewers assess whether an interview was successful or of high quality? (2) To what extent do these assessments relate to situational issues or the broader social context of the interview? (3) How do those assessments relate to traditional measurement quality indicators?

We discuss these questions, using data from Austrian Social Survey 2021, which included 352 face-to-face interviews, with the aim to contribute to the growing body of survey data quality assessments, published in the aftermath of the COVID-19 pandemic.

As our the data was collected during the covid-19-pandemic ample paradata is available to research the social situation (interview-constellation, presence of third parties, interruptions etc.), mandates tied to covid-19 (mask wearing, distance between parties etc.) as well as typical interview-related information (needs for assistance or clarification, delays in responses etc.). Furthermore, more traditional measure of interview quality like e.g. straight lining answers or using only extreme categories are used as control variables.