Post-secondary (PS) student travel represents a significant travel market in many urban regions. Despite this, PS student travel behaviour has generally not been extensively studied and is typically poorly modelled in operational travel demand models. This paper presents an investigation into issues and alternatives for modelling PS school location choice. The Greater Toronto-Hamilton Area (GTHA) is the case study region for this investigation. Two custom surveys of PS student travel are used to explore alternative school location model approaches and specifications: the 2015 and 2019 StudentMoveTO (SMTO) surveys, which were conducted by a consortium of universities in the City of Toronto. Two model formulations are investigated and compared: random utility models and random forest models. Both methods are explored in detail using the 2015 dataset, including the exploration of a wide range of model specifications. Based on the 2015 analysis, the random utility approach was then also applied to the 2019 dataset, and results for the two modelled years are compared. Key findings of this analysis include: (1) classic “gravity” variables of distance from student home to school location and school size (campus enrollment) are important explanatory variables; (2) modal accessibility measures do not appear to out-perform a simple distance metric; (3) whether the student “lives at home” with their family or lives independently significantly impacts the prediction of school location choice; (4) university and college students display different school location choice behaviour; and (5) while machine learning methods such as random forest models can reproduce observed behaviour well, they may be problematic for application to this problem for a variety of technical reasons. The paper concludes with a brief discussion of possible directions for future work.
