Estimating the Potential Modal Split of Any Future Mode Using Revealed Preference Data
Algorithm 1
Estimate future modal split using revealed preference data.
Initialize
(1)
Import full OViN dataset
(2)
Perform latent class analysis to define a “minimum performance benchmark”
(3)
Define clusters based on personal and trip attributes using k-means and elbow function in the full dataset
(4)
Retrieve train (80%) and test (20%) dataset
(5)
Define general utility function
Estimate current modal split (with and without alternative-specific constant)
(6)
Estimate parameters of the utility function of a discrete choice model with 5 modes per cluster using the train dataset
(7)
Calculate the modal split of 5 modes per cluster in the test dataset
(8)
Compare calculated modal split with recorded modal split in the full test dataset
Estimate future modal split
(9)
Define the attributes of future mode, incl. variations of 20% for sensitivity analysis (SA)
(10)
Calculate the similarity of all modes and the future mode to estimate the scaling parameter in a nest (only for nested logit), see equation (2)
(11)
Calculate modal split ranges (SA) of 6 modes per cluster in the test dataset using results of the modal split of step 6 (without alternative-specific constant)
