Fig 1.
The zones are depicted in different colours with their centroids in blue. The entry/exit points to the network are depicted with yellow points. There is an average of 5.82 routes connecting each origin and destination. The geodata used to render the plot is from ©OpenStreetMap contributors, licensed under the Open Data Commons Open Database License (ODbL). The zones are from Contours…Iris® licensed under the Open Licence v1.0 from Etalab. The maps were rendered using The R Project for Statistical Computing which is distributed as Free Software under the terms of the Free Software Foundation’s GNU General Public License.
Fig 2.
The vector is composed of the attributes of the OD pair and the three routes connecting the origin and destination, with length(Route_1) ≤ length(Route_2) ≤ length(Route_3).
Table 1.
Hyperparameters of the prior distribution h.
Fig 3.
Determining the number of clusters k.
The sum of squared errors for k-means clustering of the 624,490 OD-routes with k = 1, …, 30. After k = 9 the decrease in the mean WCSS is marginal.
Fig 4.
Selected OD-routes for the route choice experiments.
The geodata used to render the plot is from ©OpenStreetMap contributors, licensed under the Open Data Commons Open Database License (ODbL). The maps were rendered using The R Project for Statistical Computing which is distributed as Free Software under the terms of the Free Software Foundation’s GNU General Public License.
Table 2.
Cluster analyis results.
Fig 5.
Cluster centroids as representative OD-routes.
The distribution of the attributes of the selected OD pairs are similar to that of the whole network. The p-values of the Kolmogorov-Smirnov, presented in red in the top of each panel, indicate the lack of statistical evidence (with a confidence level of 0.90) to reject the hypothesis that the two distributions are the same.
Fig 6.
Route choice distribution in the nine cluster centroids.
The choices of the informed participants are different from those of the not informed participants.
Table 3.
MXL models estimation results.
Fig 7.
The MPEs of model M* are smaller than the MPEs of models Mr in the majority of the cases (blue dots). Furthermore, in the cases where the MPEs of models M* are bigger (red dots), the differences are small (close to identity line).
Fig 8.
Mean predictive errors by information group.
The models estimated with the cluster centroids are clearly better in predicting the choices for the not informed participants.
Fig 9.
Distributions of the predictive errors the 21 validation OD-routes.
The level and the variability of the errors amongst the different OD-routes imply that the choices in some OD-routes are difficult to predict, regardless of the training set used to estimate the models.
Fig 10.
Distributions of the prediction errors of the models M* on the 21 validation OD-routes.
There are no significant differences between the error distributions.