A note on logit choices in strategy transit assignment

Since it was first developed [see Spiess and Florian, Transp Res 23: 83-102 (1989)], the strategy-based transit assignment has been extensively used and its properties are well understood now. The computation of an optimal strategy is relatively fast and is comparable to the computation of a shortest path tree for one destination. However, since it is the solution of a linear program, it produces extremal solutions. As a consequence, the sensitivity analysis of strategy flows is not smooth. This work parallels the contribution of Nguyen et al. (Transp Sci 32: 54-64 1998) who developed a logit choice of strategies following a basic idea due to Dial (Transp Res 5: 88-111, 1971), in order to consider a larger variety of strategies by allowing walk choices at nodes of the transit network. Nevertheless, since the network representation used is different from the one used by Nguyen et al. (Transp Sci 32: 54-64, 1998), the development is different. This modified logit strategy transit assignment algorithm was shown to produce more realistic results in dense transit networks where relatively short walks are required for access to attractive alternative transit paths. It also models better access from centroids representing large zones.