Abstract
The predictability of Nested Logit (NL) models was examined at the aggregate level as well as the disaggregate sample level, using the before-and-after data on a new railway station, The NL models tested have a three-level structure, containing line-haul mode, station, and access mode choices.
We showed (1) the temporal stability of the choice structure, the independent variables and the parameters is very high, and (2) the prediction error on the usage of the new railway station is within 20 percent, using simplified aggregation procedures. So, we concluded that the NL models have high applicability for the prediction of the station's usage.
