Mode Choice Analysis with Imprecise Location Information Toshiyuki - PowerPoint PPT Presentation

Mode Choice Analysis with Imprecise Location Information Toshiyuki Yamamoto & Ryosuke Komori Nagoya University

Background  Public transit such as LRT and Flexible bus is regarded one of the alternatives for EST projects, which improves the access and egress conditions to public transit.  Centroid of TAZ is used as the origin and destination in conventional mode choice models Location information of trip is not precise enough IATBR2006 2006/08/17 2

Objective Effective use of large-scale person trip data to investigate the effects of access condition to the station and bus stop by  Combine information to get precise location information  Develop a model to overcome the impreciseness IATBR2006 2006/08/17 3

Chukyo Metro. Person trip data 1971 1971 8. 8.3 6.4 6. 31.3 31. 12. 12.9 41.2 41. Trai ain 1981 1981 9.9 9. 3. 3.1 39.2 39. 17. 17.9 29. 29.9 Bus Bu Car ar Mot oto. o. 1991 1991 10.5 10. 2.1 2. 49.4 49. 16.9 16. 21.1 21. Wal alk 2001 2001 10 10 1.4 1. 56.3 56. 14. 14.5 17.8 17. 0% 0% 20% 20% 40% 40% 60% 60% 80% 80% 100% 100% IATBR2006 2006/08/17 4

Zone system  True access = 730 m Ground truth Station A Station B Destination IATBR2006 2006/08/17 5

Zone system  True access = 730 m  Access to zonal centroid = 490 m Zone system Station A Station B Destination IATBR2006 2006/08/17 6

Zone system  True access = 730 m  Access to zonal centroid = 490 m  If larger zone system is used = 330 m Larger zone system Station A Station B Destination IATBR2006 2006/08/17 7

Choice structure Nested logit model with  Line haul mode in upper level  Access and egress modes for train in lower level Line haul Train Bus Car Access Walk Walk Bus Bus Car Car & egress Walk Bus Walk Bus Walk Bus IATBR2006 2006/08/17 8

Our approach: egress  Trip to governmental office, hospital and school can be identified by information on destination type  Multiple governmental offices are not located together, and usually, one zone contain at most one governmental office  The same thing applies hospital and school  Note: trips to small clinics might be included as noise  Precise location and access to station and bus stop is calculated by using GIS data IATBR2006 2006/08/17 9

Destinations 20 zones with Gifu City largest number of trip destinations for each type of destination Toyota City are used Nagoya City Yokkaichi City Destination IATBR2006 2006/08/17 10

Descriptive analysis Relationship between egress and auto share 100% 80% 自 Auto share 動 60% 車 分 担 40% 官公庁 率 Gov. （％） 病院 Hospital 20% School 学校 0% 0 1 2 3 4 イグレス距離（km） Egress (km) IATBR2006 2006/08/17 11

Gov. office: 1981 to 2001 100% 90% 80% 70% Auto share 自動車分担率 60% 1981 50% 1991 2001 40% 30% 20% 10% 0% 0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60 距離（ km ） Distance (km) IATBR2006 2006/08/17 12

Hospital: 1981 to 2001 100% 90% 80% 70% Auto share 自動車分担率 60% 1981 50% 1991 2001 40% 30% 20% 10% 0% 0 1 2 3 4 5 6 距離（ km ） Distance (km) IATBR2006 2006/08/17 13

School: 1981 to 2001 100% 90% 80% 70% Auto share 自動車分担率 60% 1981 50% 1991 2001 40% 30% 20% 10% 0% 0 1 2 3 4 5 6 Distance (km) 距離 (km) This university operated school bus from station IATBR2006 2006/08/17 14

Location choice in motorization Hospitals which located after 1970 Rank Year Length (km) 1 st /16 Hospital A 1984 4.9 4 th /16 Hospital B 1974 2.9 5 th /16 Hospital C 1972 2.8 Buildings which moved after 1960 Length （ km ） Year Before After Municipality Hall D 1966 1.1 1.4 City Hall E 1976 0.2 1.0 Hospital F 1974 0.4 0.8 Hospital G 1978 0.6 0.6 IATBR2006 2006/08/17 15

Our approach: access  About home-based trip, it is impossible to identify the house, origin of the trip  Census data provide distribution of residents of specific age/sex within survey zone  Access length is treated as probabilistic variable in estimating the mode choice model IATBR2006 2006/08/17 16

Calculation of choice probability Access ＝２４０ｍ Probability of living this city block e.g. ）２０ yrs. Male → ３ . ３％ ４０ yrs. Male → ７ . ７％ Smaller city block IATBR2006 2006/08/17 17

Calculation of choice probability  Q(j): Probability of living in city block j  Calculated by census and treated as known  P(i| j): Probability of choosing mode i given that he lives in city block j  Precise access information is used as explanatory variable  P(i): Marginal probability of choosing mode i is ∑ calculated by P(i| j)Q(j) j  Unknown parameters to be estimated are only utility function, which is indifferent among Q(j) IATBR2006 2006/08/17 18

Effects of preciseness of egress info. （ zone system is used for access ） Coefficient estimate and t-stat. in parenthesis GIS based Zone system Upper level Bus egress -2.0 (-3.4) -1.7 (-4.2) Lower level Sta. egress -2.8 (-18.9) -1.8 (-17.8) Log-likelihood -2944 -3063 GIS based egress has  Better log-likelihood  Larger coefficient estimates in absolute value IATBR2006 2006/08/17 19

Effects of preciseness of access （ GIS base is used for egress ） Coefficient estimate and t-stat. in parenthesis Proposed Zone Larger zone model system Upper level Bus access -9.2 (-5.6) -2.5 (-8.8) -2.5 (-9.3) Lower level Bus access -1.4 (-4.0) -1.2 (-4.6) -1.1 (-4.7) Sta. access -1.4 (-11.9) -0.8 (-11.8) -0.7 (-11.0) Log-likelihood -2900 -2944 -2944 Proposed model has  Better log-likelihood  Larger coefficient estimates in absolute value IATBR2006 2006/08/17 20

Comparison between access & egress Access: proposed method Egress ： GIS based Coef. t-stat. Upper level Bus access -9.2 (-5.6) Bus egress -1.8 (-3.2) Lower level Sta. access -1.4 (-11.8) Sta. egress -2.9 (-18.7) Use of bus: access to bus stop from home is dominant  Terminal mode for train: egress has larger effect than access  IATBR2006 2006/08/17 21

2 ρ Conclusion  Efficient use of conventional person trip data is proposed, and confirmed by empirical analysis  Rail ridership can be increased by locating closer to station, but move from 3 km to 2 km doesn’t mean anything  Egress from station is more important than access to station IATBR2006 2006/08/17 22