Learning from the Past - Do Historical Data Help to Improve Progress Indicators in Web Surveys?

Video

DOI

As a typical part of the interface of web surveys, progress indicators show the degree of completion for participants. These progress indicators influence the dropout and response behavior as various studies suggest. For this reason, the indicator should be chosen carefully. However, calculating the progress in adaptive surveys with many branches is often difficult. Recently related work has provided algorithms for such surveys based on different prediction strategies and has identified the Root Mean Squared Error as a valuable measure to compare different strategies. However, all previously mentioned strategies have shown poor predictions in some cases. In this paper, we present a new strategy which learns from historical data. A simulation study with 10k randomly generated surveys shows its benefits and its limits. As an example of application, we confirm our results of the simulation by comparing different prediction strategies for two large real-world surveys.