The MOBIS dataset: a large GPS dataset of mobility behaviour in Switzerland

Abstract

This article presents the MOBIS dataset and underlying survey methods used in its collection. The MOBIS study was a nation-wide randomised controlled trial (RCT) of transport pricing in Switzerland, utilising a combination of postal recruitment, online surveys, and GPS tracking. 21,571 persons completed the first online survey, and 3680 persons completed 8 weeks of GPS tracking. Many continued tracking for over a year after the study was completed. In the field experiment, participants participated through the use of a GPS tracking app, Catch-my-Day, which logged their daily travel on different transport modes and imputed the trip segments and modes. The experiment lasted 8 weeks, bookended by two online surveys. After the first 4-week control phase, participants were split into two different treatment groups and a continued control group. An analysis of the survey participation shows that the technology is capable of supporting such an experiment on both Android and iOS, the two main mobile platforms. Significant differences in the engagement and attrition were observed between iOS and Android participants over the 8-week period. Finally, the attrition rate did not vary between treatment groups. This paper also reports on the wealth of data that are being made available for further research, which includes over 3 million trip stages and activities, labelled with transport mode and purpose respectively.

Mobility in pandemic times: Exploring changes and long-term effects of COVID-19 on urban mobility behavior

“…Third, it is worth mentioning that the modes in the dataset result from estimation by means of an algorithm. In the study by a Swiss research team, the algorithm was able to detect major transportation modes with very high accuracy (see ( Molloy et al, 2022 ), Table 7), independent of the operating system. Trips traveled by bus were detected with relatively low accuracy of about 66%.…”

“…Trips traveled by bus were detected with relatively low accuracy of about 66%. One important reason is that either car trips were wrongfully labeled as bus trips or bus trips were mistaken for car trips (see ( Molloy et al, 2022 ), Table 8). However, for the setup in this study, these misclassifications are of little consequence.…”

“…On this platform, stages and travel modes are assigned from the GPS tracks, which implies a high spatial accuracy of the data collection. Furthermore, accuracy was found to be above 80% and partly above 90% for the mode detection algorithm ( Molloy et al, 2022 ). Achieving additional accuracy; app users can validate whether a trip really took place and whether the mode was correctly classified.…”

“…Achieving additional accuracy; app users can validate whether a trip really took place and whether the mode was correctly classified. If necessary, both pieces of information can be corrected and the corrected data and the algorithmic detection are stored in the analytics platform ( Molloy et al, 2022 ). In this work; a trip represents the direct movement of an individual from point A to point B using a single mode of travel.…”

Spatial risk for a superspreading environment: Insights from six urban facilities in six global cities across four continents

“…Travel data in Chicago, Hong Kong and São Paulo were extracted from the most updated large-scale travel diary surveys conducted by local authorities. The Zurich dataset was compiled from a research project by ETH Zurich that has applied the methods of both survey questionnaires and GPS tracking on mobile phones to record human travel behavior ( 45 ). Datasets of Chicago, São Paulo and Zurich used in this study were trimmed from a larger survey database.…”