När går bussen? : En studie kring metoder för kvalitetsbedömning av SL:s bussavgångsprognoser

Sammanfattning: As a result of a growing population, the city of Stockholm is facing many challenges. Getting more people to travel by public transportation is a key factor in coping with this increased urbanization. In the strive for increased ridership, it is the Stockholm Public Transport Administration’s (SL) job to make sure that the services provided are of high quality. One of these services is the real time bus departure predictions provided to the travellers through digital signs or by web and mobile applications. Due to a lack of proper tools, SL has unfortunately not yet been able to establish a systematic assessment of the quality of these bus prediction. The goal of this study was to help SL find such tools and solutions for assessing the quality of bus predictions. More specifically, the purpose of the study was to investigate the concept of prediction quality and identify suitable statistical tools for measuring quality. In order to do this a comprehensive literature study has been conducted. The findings of the literature study were then tested in practice in order to answer how such quality measurements should be made in the context of SL’s ITinfrastructure. This was answered by carrying out a pilot study in which the prediction quality was assessed on data from one week for a specific bus line. From the initial literature study, it was concluded that there are many dimensions that potentially affect the traveller’s perception of bus prediction quality. However, it was also concluded that a quality assessment plausibly should start with an evaluation of the precision. In order to assess the precision, several types of descriptive measures and analytical perspectives were proposed. As of how these findings should be made in the context of SL’s IT-systems, a method for creating observations from the available prediction data was presented. It was also concluded that in order to mirror the travellers experience, the prediction data should be collected “late” in the process of bus prediction generation.