1. Transportation Demand Analysis
Mode Choice
Mode Choice

2. Outline Introduction Binary Diversion Analysis Abstract Mode Models
Behavioral Models
Transportation Demand Analysis - Mode Choice

3. Transportation Demand Analysis
Mode Choice
Introduction

4. Third Step in Urban Transportation Modeling System
Socio-economic Forecasts
(Population, Employment, …)
Trip Generation
Trip Distribution
Transportation Demand Analysis - Mode Choice
Mode Choice
Trip Assignment

5. Situation Trip-end models Trip interchange models Tim=Tif(m)
Restricted to the characteristics of origin (socio-economic variables)
Trip interchange models
Tijm=Tijf(m)
Restricted to the characteristics of origin-destination (socio-economic and LOS variables)
Transportation Demand Analysis - Mode Choice

6. Situation Four-step model: Given Tij  Tijm= f(Tij, …m)
Direct approach:
Tijm = f(Ti,Aj,…m)=f(g(popi),h(popj),m(ttijm)) =Ψ(popi, popj, ttijm)
More general:
Tijm = f( …i, …j, …ij, …ijm)
Example:
Tijm = f(popi, popj, Number of callsij, ttijm)
Transportation Demand Analysis - Mode Choice

7. Hierarchy of mode and route choice
Simultaneous approach
Useful in many application (e.g., policy making)
Desirable, if the scope of the problem at hand is manageable
Sequential approach
It is generally agreed to be the mode followed by the route choice
Preferable for large scale networks with excessive number of alternatives
Transportation Demand Analysis - Mode Choice

8. Scale of mode and route choice problem
Microscopic (individual based) due to policy making
Mode choice examples
Exclusive lanes for high occupancy vehicles
Increased parking taxes for private automobiles …
Route choice example
Toll collection
Metering on freeway ramps
Transportation Demand Analysis - Mode Choice

9. Binary Diversion Analysis
Transportation Demand Analysis
Mode Choice
Binary Diversion Analysis

10. Early developments Aggregate diversion models Binary choice case
Splitting traffic flows into modes based on some simple formulation of their relative attributes
Binary choice case
Traffic splits into two modes
Automobile (private)
Transit (public)
Transportation Demand Analysis - Mode Choice

11. Concept
Breaking the traffic down based on the attributes of private and public transportation
It has been used for modal split both for
trip-end (before trip distribution analysis)
trip interchanges (after trip distribution analysis)
Transportation Demand Analysis - Mode Choice

12. Trip-end Models Placed before trip distribution step
The accounted variables:
Socio economic variables
Accessibility ratio to each mode
The model is independent of system variables
The model is almost obsoleted due to increasing car usage after the 2nd world war
Transportation Demand Analysis - Mode Choice

13. Example (Binary Diversion Curves)
Accm= ∑Djfij,
Dj: trip attraction of zone j
fijm: distance between i and j for mode m
Transportation Demand Analysis - Mode Choice
Note: Curves may stratified on some socioeconomic characteristics (e.g., Cars per household, household size).

14. Examples of Binary Diversion Curves (Trip-end)
Wisconsin Model (1963):
Advanced trip-end model including trip type (purpose), individual characteristics (number of cars) and System characteristics limited to zone i (accessibility ratio in the form of private to public)
Chicago Model (CATS):
Including Land use, Gender, Occupation, Driver, Age, Car ownership
Transportation Demand Analysis - Mode Choice

15. Example of Binary Regression Model (Trip-end)
Pittsburg model (PATS):
Trip type (purpose), car ownership, residential density, distance from CBD
Ln(number of edu. trips by public/1000people)= Ln (residential density)
Ln(number of other trips by public/1000people with no car)= (residential density) Ln (residential density)2
Ln(number of other trips by public/1000people with 1 car)= (residential density) Ln (residential density)2
Ln(number of other trips by public/1000people with 2+ car)= (residential density) Ln (residential density)2
Generally, regression is avoided because of 0,1 boundaries of the probability
Transportation Demand Analysis - Mode Choice

16. Trip Interchange Models
Placed after trip distribution step
The attributes are specific to origin-destination pairs
More than a single measure of accessibility:
travel time ratios
travel cost ratios …
Transportation Demand Analysis - Mode Choice

17. Examples of Binary Diversion Curves (Trip-interchange)
Bay Area Rapid transit (BART)
Trip type (purpose), peak and off-peak period, CBD and non-CBD, Travel time ratio
Washington, Philadelphia and Toronto study:
Cost ratio, Travel time ratio, out of vehicle time ratio, Income level
Transportation Demand Analysis - Mode Choice

18. Washington, Philadelphia and Toronto Study
Transportation Demand Analysis - Mode Choice

19. Washington, Philadelphia and Toronto Study
CR: Ratio of transit to auto cost
EC: Annual median income per worker
L: Ratio of transit to auto service
Transportation Demand Analysis - Mode Choice
Range for (L) service time ratio:
L4: 0.0 to 1.5
L3: 1.5 to 3.5
L2: 3.5 to 5.5
L1: 5.5 and over

20. Washington, Philadelphia and Toronto Study
CR: Ratio of transit to auto cost
EC: Annual median income per worker
L: Ratio of transit to auto service
Transportation Demand Analysis - Mode Choice
TTR: Ratio of transit to auto time

21. Binary Diversion Analysis Limitations
The method is too aggregate
It is not applicable for policy-making
(Policies affect different groups of urban travelers differently)
Do not accounted for captive users
Captive transit users
Captives car users
Limited to binary choice situations
Nowadays, multiple choices are available in most transportation systems
Transportation Demand Analysis - Mode Choice

22. Transportation Demand Analysis
Mode Choice
Abstract Mode Models

23. Assumptions
Mode choice is based on the modes characteristics (not on their names)
A mode characteristics are sufficient to find its transportation share
A new mode usage can be determined by its characteristics
General form (Gravity structure):
T=αHβCγNθ
H: travel time, C:Cost, N:Number of modes
Transportation Demand Analysis - Mode Choice

24. Abstract Mode Example 1 Mode 3 Mode 2 Mode 1 3 2 1 Travel Time (hr)
Ratio to best 5
Travel cost (100$)
1.5
2.5
Tk= Hb-60Cb-60Hrk-50Crk-50N
Tk: Number of trips by mode k
Hb: Best travel time Hrk: Travel time ratio of mode k to the best
Cb: Best cost Crk: Travel cost ratio of mode k to the best
N: Number of modes
Tmode1= *1-60*2-60*1-50*2.5-50*3
Tmode1=445, Tmode2=435, Tmode3=  TTotal=1280
Transportation Demand Analysis - Mode Choice

25. Abstract Mode Example 1 (Introduction of mode 4)
1.5 3 2 1
Travel Time (hr)
Ratio to best 5
Travel cost (100$)
Tk= Hb-60Cb-60Hrk-50Crk-50N
Tk: Number of trips by mode k
Hb: Best travel time Hrk: Travel time ratio of mode k to the best
Cb: Best cost Crk: Travel cost ratio of mode k to the best
N: Number of modes
Tmode1= *1-60*1-60*1-50*5-50*4
Tmode1=330, Tmode2=370, Tmode3= Tmode4=500
 TTotal=1560
Transportation Demand Analysis - Mode Choice

26. Abstract Mode Example 1 (Introduction of mode 5)
0.5
1.5 3 2 1
Travel Time (hr) 6 4
Ratio to best 5
Travel cost (100$)
Introduction of Mode 5:
Tmode1=385, Tmode2=315, Tmode3= Tmode5=520
 TTotal=1440
Introduction of Mode 4 and Mode 5:
Tmode1=270, Tmode2=250, Tmode3= Tmode4=410, Tmode5=480
 TTotal=1590
Transportation Demand Analysis - Mode Choice

27. Abstract Mode Example 2
Tijk= α0(Pi)α1(Pj)α2(Yi)α3(Yj)α4(Mi)α5(Mj)α6 (Nij)α7 f1(H)f2(C)f3(D)
f1(H)= (Hijb)β0 (Hijrk)β1
f2(C)= (Cijb)γ0 (Cijrk)γ1
f3(D)= (Dijb)δ0 (Dijrk)δ1
Hb: Best travel time Hrk: Travel time ratio of mode k to the best
Cb: Best cost Crk: Travel cost ratio of mode k to the best
Db: Best convenience Drk: Convenience ratio of mode k to the best
N: Number of modes between i and j
Y: Income M: Land-use index (Labor source)
P:population
Transportation Demand Analysis - Mode Choice

28. Abstract Mode Model Limitations
The total number of trips depends on the number of modes (This may not be the case in work trips)
Sometimes, it is difficult to find a corridor with flow similar to the studied model
Ratios are more important than real values
Transportation Demand Analysis - Mode Choice

29. Transportation Demand Analysis
Mode Choice
Behavioral Models

30. Variables Category Application Socioeconomic demand variables
Level of service or supply variables
Application
Significant
Can reflect the policy analysis or planning
Transportation Demand Analysis - Mode Choice

31. Socio-economic Variables
Income
Age and role in household
Car ownership
Household size
Residential location
Profession …
Transportation Demand Analysis - Mode Choice

32. Supply Variables In-vehicle travel time.
Access, waiting, and transfer times.
Travel cost.
Qualitative and attitudinal variables …
Transportation Demand Analysis - Mode Choice

33. Two Postulations in Choice Modeling
Unlabeled Choice models (Mode-abstract)
Consumers decide on the basis of characteristics of goods and services rather than the goods themselves (Lancaster, 1966)
E.g., Two distinct modes which have the same level of service attributes would be treated as one
Labeled choice (Mode-specific)
Choices are influenced by the level of service attributes, but these influences will vary from one alternative to another
Transportation Demand Analysis - Mode Choice

34. γi: a mode-specific constant term (different values for A and B)
Mode-Abstract and Mode-Specific Comparison
Unlabeled choice
Choice function parameters have no relation to specific modes
Labeled choice
Choice function parameters would also vary depending on the alternatives
They may have dummies or mode specific constants
α,β: parameters
i: A or B
𝑉 𝑖 =𝛼 𝑇 𝑖 +𝛽 𝐶 𝑖
Transportation Demand Analysis - Mode Choice
𝑉 𝑖 =𝛼 𝑇 𝑖 + 𝛽 𝑖 𝐶 𝑖
𝑉 𝑖 =𝛼 𝑇 𝑖 + 𝛽 𝑖 𝐶 𝑖 + 𝛾 𝑖
γi: a mode-specific constant term (different values for A and B)

35. Mode-Abstract and Mode-Specific Comparison
Build on the framework of modern microeconomic theory
Requiring a lower number of model parameters
Mode-specific
More consistent with choice behavior
Transportation Demand Analysis - Mode Choice

36. Examples of Urban Mode Choice Models
Stochastic binary model
Earlier applications of stochastic choice models
Due to the absence of efficient, computer-based estimation techniques for multinomial models
Trinomial model
Multimodal model
Transportation Demand Analysis - Mode Choice

37. A Stochastic Binary Model (De Donnea, 1970)
Choice between auto and bus transit in Rotterdam, the Netherlands.
Calibrated both a logit and probit mode-specific model
Insignificant difference in the results
Transportation Demand Analysis - Mode Choice

38. A Stochastic Binary Model (De Donnea, 1970)
𝑉 1 = 𝛼 1 + 𝛼 2 𝑌 𝑡 1 + 𝛼 3 𝐻
𝑉 2 = 𝛼 2 𝑌 𝑡 2
V(1): Auto choice function
V(2): Transit choice function
Y: individual's income
t: travel time
H: a dummy variable (H = 1, if the individual is head of a household)
α1,α2,α3 : model parameters
Transportation Demand Analysis - Mode Choice

39. A Stochastic Binary Model (De Donnea, 1970)
L= V(1) - V(2) = a1+ a2 Y (t1– t2 ) + a3 H
Logit model:
Probit model:
Transportation Demand Analysis - Mode Choice

40. A Stochastic Binary Model (De Donnea, 1970)
Insignificant difference in the results
Logit Model
Probit Model a1
0.0226
0.0117 a2 a3
1.0800
0.6600
Transportation Demand Analysis - Mode Choice

41. A Trinomial Model (Ganek and Saulino, 1976)
A model calibrated for the Pittsburgh, Pennsylvania
Three modes
Automobile
Transit
Carpool
a mode between drive alone and public transportation
Transportation Demand Analysis - Mode Choice

42. Variables (Ganek and Saulino, 1976)
In-vehicle time
Access time and egress times
Total cost
Relative comfort and convenience
Flexibility
Mode reliability
Car availability
Location of work place
Household income
Sex
Mode constants
Transportation Demand Analysis - Mode Choice

43. A Multinomial Model (Train, 1976)
A multinomial logit choice mode for San Francisco Bay Area households
Four alternatives
Auto alone
Bus with walking access
Bus with automobile access
Carpool
Transportation Demand Analysis - Mode Choice

44. Variables (Train, 1976)
Cost divided by post tax wage (cents per cent per minute)
On-vehicle time in minutes
Walk time in minutes
Transfer time in minutes
Number of transfers
Bus headways in minutes (below or above 8 minutes)
Family income classes
Length of residence in the community
Number of persons in the household who can drive
Mode dummy variables
Transportation Demand Analysis - Mode Choice

45. A Multinomial Model (Ben Akiva and Richards, 1976)
A multinomial choice mode for some Dutch cities
Six alternatives
Walking
Bicycle
Moped
bus
Train
automobile
Transportation Demand Analysis - Mode Choice

46. Variables (Ben Akiva and Richards, 1976)
In-vehicle time
Out-of-vehicle time
Out-of-pocket cost
Household income
Car availability
Occupation
Mode-specific dummy variables
Transportation Demand Analysis - Mode Choice

47. Notes Urban mode choice models can be used to evaluate
Effects of minor and gradual changes
In the area where they have been calibrated
Mode-specific models are rather successful in predicting behavior than mode-abstract models
According to the literature:
Most of the variations of mode choice behavior:
Mode-specific dummy variables
Some socioeconomic variables
(particularly income or car ownership)
Transportation Demand Analysis - Mode Choice

48. New developments
Mode specific models containing a number of alternatives
Developing models for new urban types
Relaxing the assumption of independency in IID and developing the nested structure
Relaxing the assumption of independency and identically in IID and developing the mixed structure
Relaxing the assumption of homogeneity of population and developing the latent class structure
Transportation Demand Analysis - Mode Choice

49. Transportation Demand Analysis - Mode Choice
Finish