The project FACES (Feasibility, Acceptance, and Data Quality of New Multimodal Surveys) is investigating how virtual reality (VR) and artificial intelligence (AI) can be used to make traditional face-to-face surveys more innovative. The aim is to use new technologies to better address challenges in interviewer-based surveys, such as rising survey costs and decreasing response rates.
Avatars, i.e. virtual representations of people, and AI systems open up new possibilities for conducting survey-based studies. With FACES, a research team from the Leibniz Institute for Educational Trajectories (LIfBi) and Goethe University Frankfurt aims to develop an avatar-based survey system, test it in practice and investigate its applicability.
“The integration of VR and AI in survey research can help to develop solutions for current problems in face-to-face surveys. For example, we hope to improve response rates and the quality of the data as well as reduce the costs of conducting surveys in the long term,” says Prof. Dr. Corinna Kleinert, project manager at FACES.
New approaches to data quality
A new system for interviewer-based online surveys is to be developed with FACES. It will integrate different avatars as well as AI technologies for the automatic processing of speech and behavioral data.
“We aim to find out how different avatar designs are accepted by survey participants in certain interview situations and how this affects the acceptance and data quality of a survey,” explains Prof. Dr. Christian Aßmann, who is also project manager at FACES. “The question is particularly interesting: can virtual interviewers reduce typical biases in surveys - for example, by increasing the willingness to participate in the interview or by answering the avatar more honestly?”
Investigations and tests in practice
The first step is to develop an open source system for avatar-based and video-based interviews. In experiments, different characteristics of the avatars (e.g. photorealism), the interview situation (e.g. room design) and the immersiveness of the interview are tested in order to find combinations that are well suited for interviews. Both interviews in virtual environments and video-based formats are used.
These combinations of characteristics are then tested in interviews with participants from Starting Cohorts 3 and 5 of the National Educational Panel Study (NEPS). Both cohorts started in 2010 with students in the 5th grade and first-year university students. Since then, both groups have been accompanied on their educational journeys in the NEPS. The experimental surveys are investigating whether interviews with avatars have advantages compared to video interviews with real people, for example with regard to acceptance, feasibility or non-response bias (bias in surveys due to systematic non-response). The aim is also to find out which characteristics of the situation and the avatars help to reduce interviewer effects. These findings will ultimately be incorporated into the AI-based training of avatar-based chatbots.
FACES is funded by the German Research Foundation (DFG) as part of the priority infrastructure program SPP 2034 “New Data Spaces” and is being carried out jointly by LIfBi and Goethe University Frankfurt.
Project website
Website "New Data Spaces"