Download presentation
Presentation is loading. Please wait.
Published byAlessandro Sinclair Modified over 9 years ago
1
1 ESTIMATION OF DRIVERS ROUTE CHOICE USING MULTI-PERIOD MULTINOMIAL CHOICE MODELS Stephen Clark and Dr Richard Batley Institute for Transport Studies University of Leeds, U.K.
2
2 Introduction Where ‘panel’ data is available on choices made by individuals, it is reasonable to assume that previous experiences somehow condition these choices
3
3 Data set 1 Number plate matching exercise conducted in the City of York, U.K. 100% survey 08:00 to 09:00 27, 28 June; 7, 8, 9, 11, 13, 27 September; 18 October, 2000
4
4 Repetition and contiguity Of the vehicles observed on 2 ‘adjacent’ survey days, over 50% used the same route on both days As the period between 2 survey days increased, this percentage dropped –e.g. for 2 survey days 14 working days apart, the percentage was 35%-40%
5
5 Repetition and O-D pairs For a particular O-D movement on 4 ‘consecutive’ days: –3 vehicles travelled O-D on all 4 days –8 vehicles travelled O-D on any 3 ‘consecutive’ days –28 vehicles travelled O-D on any 2 ‘consecutive’ days Of these 39 repeat vehicles, only on 3 occasions out of the 53 possibilities did they follow a different route on a ‘consecutive’ day
6
6 Habit Suggests route choice data contains a high degree of habitual information If habitual behaviour is explicitly modelled, then its strength can be estimated Failure to account for repetition and route experience may undermine the validity of any models
7
7 Random effects probit
8
8 Modelling problems LIMDEP failed to estimate a parameter in the range 1 GAUSS code applying Chamberlain’s conditional maximum likelihood estimation detected lack of variability in explanatory variables unable to estimate model
9
9 Multinomial multi-period probit Autoregressive structure, correlations between alternatives and time periods, unobserved heterogeneity across individuals, differential variances across alternatives
10
10 Geweke’s GAUSS code Multinomial multi-period probit AR(1) errors for each choice Individual-specific random effects for each choice Two estimation methods 1.Method of Simulated Moments 2.Simulated Maximum Likelihood
11
11 Modelling problems Both methods failed to estimate a model Presence of singular matrix Again, suspected artefact of lack of variability in explanatory variables
12
12 Data set 2 Stated preference study of route and departure time choice in City of York, U.K. –How do people respond to an increase in travel time and/or travel time variability? –2-stage study, involving customisation –5 cards
13
13 Stage 1
14
14 Derived time variables ‘get off earlier’ time (G) journey time (Q) journey time variability (S) late time (L)
15
15 Stage 2
16
16 Field study Members of staff at the York Health Services Trust Prize draw incentive 165 usable first stage questionnaires 56 usable second stage questionnaires 34% response rate for second stage Ranked data ‘exploded’ into binary choice data 840 binary choice observations
17
17 MNL
18
18 Random parameters logit
19
19 Random parameters logit
20
20 RPL 1
21
21 RPL 2
22
22 Summary Where ‘panel’ data is available on choices made by individuals, it is reasonable to assume that previous experiences somehow condition these choices Data set 1: RP route choice data –Random effects probit –Multi-period multinomial probit Data set 2: SP route and departure time choice –Pseudo-panel –Random parameters logit –Evidence of repeated observations effects
Similar presentations
© 2024 SlidePlayer.com Inc.
All rights reserved.