Abstract

Traditionally, diary studies are used to capture time use behaviour, such as traveling, that is otherwise prone to recall errors, telescoping and other measurement errors when asked retrospectively. However, the traditional diary modes, web and paper, pose a high burden for respondents, resulting in nonresponse and measurement errors such as rounding errors, and underreporting of short and walking trips. Given these limitations and the increased use of smartphones, growing attention is given to smartphone-based travel surveys. Smartphone travel apps can make use of geolocation sensors present on the smartphone to construct a travel diary.
Currently, the common approach in travel surveys is to survey one to three days in order to minimize response burden, despite recommendations from researchers to collect data for a longer period (in order to capture within-person variability). In smartphone travel surveys the response burden is different. The respondent has to install an app on a device where geolocation data can be collected passively. Therefore, increasing the number of diary days contributes less to response burden compared to traditional diary surveys. This reduction in burden opens up the possibilities to prolong the study period.
Statistics Netherlands developed a state-of-the-art smartphone app to measure mobility behavior in the Netherlands. The smartphone app automatically compiles a travel diary, consisting of a series of movements and stops, based on passively collected geolocation data. In 2022 and 2023, the app was tested in a large-scale field test (initial sample n=3,200). This test aimed to analyze and evaluate design choices such as the data collection strategy (response rate and participation duration), the app sensor tracking implementation (measurement frequency and battery management), the respondent interaction (active-passive trade-offs), the data quality and validation (quality requirements), and the AI-ML predictions (travel mode/motive prediction).
The field test included several randomizations to study these design choices, which we will report on in this paper. In particular, we address an influential choice: the duration of the travel diary. For this choice we look at data quantity and data quality as a function of time in-study. Our conclusions and recommendations have relevance across a wide range of research areas concerning studies on (time-use) behaviour, studies considering similar design choices, and studies using (smartphone) sensors.