Travel Demand Model - OKI

OKI/MVRPC Travel Demand Model – Version 6.06. Non-Home-Based Model Estimation6.1 Sample SizeThe NHB estimation dataset includes 6,385 records, of which 5,188 were available for estimation.As in the case of the HBW and HBO models, records that could not be used for estimation includeintrazonal trips, records where the chosen mode was transit but no transit path was available andrecords with a highway in-vehicle time longer than 60 minutes. Table 6.1 shows the number ofobservations by chosen mode. A common rule of thumb suggests that at least 300 records permode are required to obtain robust estimation results. For this reason, local and express busmodes were collapsed into a single transit mode, and park & ride and kiss & ride transit accessmodes were collapsed into a single transit access mode. Table 6.1 also shows the modal shares,i.e., the weighted number of observations. Please see Section 3.3 for an explanation of theweight factors. All estimation results were derived using estimation weight factors, ascorresponds given that the sample is choice-based.Table 6-1 Non-Home-Based Sample SizeMode ChosenNo. ObsModalShareDrive Alone 2807 60.61%Shared Ride 2 1090 23.54%Shared Ride 3+ 679 14.66%TransitWalk 422 0.82%Drive 190 0.37%Total 5188 100.00%Modal shares are weighted proportions.6.2 Estimation ResultsThe NHB models were the last set of models estimated, and thus much had been learned aboutthe dataset from the HBW and HBO model estimation efforts. For this reason comparatively fewruns were required to arrive at a good NHB model for the OKI/MVRPC region. All NHB estimationruns are included in Appendix 3.Table 6.2 shows the recommended NHB model. NHB trips are almost as sensitive to in-vehicletime as HBW trips, as indicated by the similar in-vehicle time coefficients obtained for NHB andHBW. The NHB model however exhibits a lower value of time compared to the HBW model, asexpected. In fact, NHB values of time equal to one-half of the HBW value of time are common.In the recommended model, the relationship of centroid walk to in-vehicle time is kept constantat 2.5. This is necessary because in models with an unconstrained centroid walk coefficient theout-of-vehicle to in-vehicle time ratio is unacceptably high, while the in-vehicle time coefficient ispositive and insignificant (see for example Run #1).CBD indicator variables were used to capture differences in the attractiveness of the CBD bymode, after controlling for time and cost. The estimated CBD coefficients indicate that, all elseequal, transit trips are more likely than auto trips to have a CBD destination. This result makessense given that the transit network is primarily radial and CBD oriented, that is, it providesbetter service to suburb-to-CBD trips than to suburb-to-suburb trips.Mode Choice - Non-Home-Based Model Estimation 27