The Effects of Study Duration on Nonresponse and Measurement Quality in a Smartphone App-Based Travel Diary

Abstract

The use of smartphones for passive measurement can help increase response rates as such measurement may reduce the burden for participants by simplifying respondent tasks. Additionally, smartphone-based passive data collection can improve measurement quality by mitigating underreporting and reducing recall errors. In diary studies with reporting periods that last  multiple days or weeks making these studies very burdensome, response rates are usually low and measurement errors can be substantial. In such burdensome studies smartphone sensor data can be particularly beneficial. We report the results of a randomized experiment in an app-assisted travel diary study in a probability-based sample in the Netherlands. The app uses passively collected geolocation measurements to compile a diary consisting of trips and stops which respondents can edit and enrich. We invited half of the respondents to participate in the travel app for one-day, and the other half for seven-days (overall N = 2544). We also offered a one-day web diary as an alternative and varied the moment of offering the web diary.   More people registered in the app diary (12%) than in the web diary (7%). Unexpectedly, the initial app registration is significantly higher in the seven-day sample than in the one-day sample (11% vs. 14%). Study completion is significantly higher for the one-day participants (29% vs. 43%), making the absolute app completion rate the same (7%) for both groups. Using the population registry data, we analyzed whether socio-demographic variables related to travel behavior can predict participation in the app or web diary. We find differences in sample composition between the app and the web diary participants and between the one-day and the seven-day participants. Measurement quality in the app – measured with two dimensions – the amount of passive data collected and the active engagement of participants with their data – differs between the groups.