Lecture 6 - Mode & Destination Choice Models
CIVE 461/861: Urban Transportation Planning
Outline
- Utility function variables
- Direct & cross-elasticity
- Forecasting & aggregation
- Limitations
- Extension to other choices
Trip-End Models
- Trip end mode split (choice) models apply before trip distribution & split trip ends estimated in trip generation among ends (destinations)
- Basically mode-specific trip generation models (e.g., regression or cross-classification)
- Trip-end mode split is a function of socio-economic variables such as income & auto ownership
- Cannot include modal level-of-service attributes (travel time, cost, etc.) since do not yet have O-D flows
- Works better for short-term & when there is no modal competition & traffic congestion – small urban & rural areas where transit is a social service
Trip End Model Example
Trip Interchange Model
- Trip interchange mode split (choice) models apply after trip distribution & split trip interchanges estimated in trip distribution
- Since O-D flows are known, can compute travel times, costs, etc. for competing modes
- Trip interchange models generally applied in medium to large urban areas where alternative modes are competitive & traffic congestion is a factor
- Can assess broad range of policies:
- Improved transit service (e.g., headways, coverage, travel time)
- Road pricing, gasoline taxes, etc.
- Transit fare policy
- Parking supply & cost
- HOV policies
Explanatory Variables in Mode Choice Models
Alternative Specific Variables
- Travel time
- In-vehicle
- Out-of-vehicle
- Walk (access & egress)
- Wait (initial & transfer)
- Out-of-pocket travel costs
- Transit fare or “in-vehicle” auto cost
- Parking cost
- Other factors? Reliability, safety, comfort, convenience, etc.
Explanatory Variables in Mode Choice Models
Tripmaker Specific Variables
- Income
- Vehicle availability
- No. of vehicles in household
- Driver’s license? (0/1)
- Age
- Gender
- Occupation
- Household composition
Modeled Alternatives: The Choice Set
The number & definition of modeled modes depends on
- Problem application
- Available data
- Network modeling capabilities
- At a minimum, requires some representation of competition between auto & transit
Auto Mode Representation
- Common to distinguish between drivers & passengers:
- Auto driver mode: trip-maker drives a vehicle from origin to destination
- Auto passenger mode: trip-maker is a passenger in car from origin to destination
- Alternatively, we can distinguish between drive alone & shared trips
- Drive alone mode: trip-maker is sole occupant of vehicle from origin to destination (single occupancy vehicle (SOV) trip)
- Shared-ride mode: trip-maker is one of several vehicle occupants from origin to destination
- Car driver for shared ride
- Car passenger for shared ride with household member
- Car pooling: driver or passenger in vehicle with non-household members
Transit Mode Representation
- Possible ways to categorize transit include
- Local (e.g., StarTran or Metro Transit) vs. regional (e.g., OMALiNK or N-E Ride)
- Surface shared right-of-way (bus, streetcar) vs. dedicated ROW (subway, LRT, busway)
- Bus vs. rail
- Regular service vs. express vs. other premium service
Mixed Modes
- Combined auto-transit modes exist, in which auto is used to access transit system
- Park & Ride: trip-maker drives to a transit station & parks their vehicle at station
- Kiss & Ride: trip-maker is driven as a passenger to a transit station & is dropped off there
- Auto access greatly expands catchment area of transit services
Other Modes
There has been an explosion of modes in the last decade
- Transportation network companies (TNCs): Uber, Lyft
- Single ride
- Pooled rides & van pools
- Regular taxis
- Motorcycles
- Jitneys
- Ferries, water taxis
- Non-motorized modes
- Walking
- Bicycling
- E-bikes
- E-Scooters
Decision Structure
- Mode choice can be represented as decision tree
- Each node of tree represents an alternative & relationships among choices indicated via hierarchical tree structure (often called nesting structure)
![]()
Alternative Decision Structures
Example: Work Trip Mode Choice Model
A regional morning peak-period work trip mode choice model is a three-mode logit model as follows
d = auto-drive
p = auto passenger
t = transit
Example: Work Trip Mode Choice Model
\(V_m\) = utility for mode m
\(COST_m\) = out-of-pocket travel cost ($) for mode m
\(IVTT_m\) = in-vehicle travel time (min.) for mode m
\(OVTT_m\) = out-of-vehicle travel time (min.) for mode m
NVEH = avg. no. of vehicles per household in home zone
TWY = 1 if employment zone is located within the catchment area of a transitway station outside the CBD = 0 otherwise
REGION = 1 if home zone is located in a specific area = 0 otherwise
Example: Work Trip Mode Choice Model
Systematic utility functions
\[𝑉_𝑑=−0.55−0.57×𝐶𝑂𝑆𝑇_𝑑−0.20×𝐼𝑉𝑇𝑇_𝑑+0.75×𝑁𝑉𝐸𝐻\]
\[𝑉_𝑝=−2.28−0.57×𝐶𝑂𝑆𝑇_𝑝−0.20×𝐼𝑉𝑇𝑇_𝑝−0.30×𝑂𝑉𝑇𝑇_𝑡
+0.45×𝑁𝑉𝐸𝐻\]
\[𝑉_𝑡=−0.57×𝐶𝑂𝑆𝑇_𝑡−0.20×𝐼𝑉𝑇𝑇_𝑡−0.30×𝑂𝑉𝑇𝑇_𝑡+1.07×𝑇𝑊𝑌
−0.98×𝑅𝐸𝐺𝐼𝑂𝑁\]
Example: Work Trip Mode Choice Model
Important Notes on Mode Choice Example
- Cost, IVTT, & OVTT are generic variables
- Enter all utility with same parameter values (no OVTT for driver)
- NVEH is an alternative specific variable
- Enters two utility functions with different parameter values
- There are alternative specific constants for drive & passenger modes but not transit
- If there are M alternatives, we can statically identify M-1 alternative specific constants (ASC)
- ASC account for unobserved but non-random variation – ASC will match market share for multinomial logit model
Important Notes on Mode Choice Example
- Socio-economic variables do not vary between alternatives
- If included in model, must enter as alternative specific variables
- If included in model as generic variables, would add same utility to all alternatives & have no impact on choice probability
- Can interact socio-economic variables with level of service (LOS) variables
- E.g., travel cost / household income
Calculating Value of Travel Time (VTT) from Utility Function
- We can calculate an implied VTT from model coefficients
- From our previous example
\[−0.57×𝐶𝑂𝑆𝑇($)=−0.20×𝐼𝑉𝑇𝑇(𝑚𝑖𝑛)\]
- How many $ would one pay to reduce their travel time by one hour?
\[𝑉𝑇𝑇=𝛽_{𝑡𝑖𝑚𝑒}/𝛽_{𝑐𝑜𝑠𝑡} =(−0.20)/(−0.57)=$21/ℎ𝑜𝑢𝑟\]
- Similarly, VTT for OVTT = $31.6/hour
- Question: How do you think VTT for IVTT compares for auto vs. transit modes?
Direct & Cross Elasticity
- Direct elasticity: direct effect of changing a variable value related to a good on demand for the same good
- E.g., elasticity of transit demand wrt transit fare, transit travel time, or transit headway
- Cross elasticity: effect of changing a variable value related to a good on demand for a different good
- E.g., elasticity of transit demand wrt auto travel time
Logit - Direct Elasticity
- Direct elasticity of the probability of an individual \(n\) choosing alternative \(i\) wrt a change in an attribute/independent variable \(X_{ik}\) with coefficient \(\beta_k\) (ignoring \(n\) subscripts for simplicity)
\[e_{direct} = \frac{\partial P_i}{\partial X_{ik}}\frac{X_{ik}}{P_i} = (1-P_i)X_{ik}\beta_k\]
Logit - Cross Elasticity
- Cross elasticity of the probability of an individual \(n\) choosing alternative \(i\) wrt a change in an attribute/independent variable \(X_{jk}\) with coefficient \(\beta_k\) for a different alternative
\[e_{cross}=-P_j X_{jk} \beta_k\]
- Above is same regardless of alternative \(i\). I.e., all modes have the same cross-elasticity wrt an attribute \(k\) of mode alternative \(j\)
- Above results from independence of irrelevant alternatives (IIA) property of basis logit model
Forecasting Mode Choice (Model Application)
- Mode choice models generate choice probability predictions for individuals
- For planning purposes, we are interested in total number of people in a zone or study area likely to choose each mode
- Must aggregate individual choices
- Total enumeration/microsimulation
- Sample enumeration
- Naïve aggregation
- Classify with naïve aggregation
Total Enumeration
- Simplest method but often not practical
- Enumerate all individuals in study area
- Calculate mode choice probabilities
- Sum probabilities over all individuals
- Problem: extremely expensive to survey all individuals & difficult to simulate them
- Microsimulation models can be used to simulate individuals
Sample Enumeration
- Second best to total enumeration
- Requires representative sample of population
- Calculate mode choice probabilities for sample
- Sample prediction of mode shares can be substituted for population prediction
- Problem: How to synthesize a sample for a future year?
Naive Aggregation
- Treat individual choice model as an aggregate model
- Use zonal averages for variables as inputs to compute “average” zonal probability
- Problem: Choice probabilities are non-linear functions of variables so aggregation bias
- Avoid this method!
Classification with Naive Aggregation
- Classify population into relatively homogenous groups
- E.g., by vehicle ownership & transit availability
- Determine variable averages for each classification group
- Use group averages to compute average mode shares
- Sum mode shares
- Most common method used in practice
Classification with Naïve Aggregation Example
A work trip mode split logit model includes the no. of hh. vehicles (NVEH) and whether a worker has a driver’s license (DLIC) as variables. The aggregation/forecasting procedure used is to:
- Divide workers into “n” NVEH & DLIC categories
- Estimate percentage of workers in each category by O-D pair
- Compute mode choice probabilities for each category
- Compute weighted average mode splits for each O-D pair
Aggregation Bias
- Consider a zone in which people are identical except for their income, and that the probability of a person using transit depends only on the person’s income.
Aggregation Bias
- Bias becomes worse for more heterogenous populations
Forecasting Example
\[𝑉_𝑎=\beta_1+\beta_2 𝐼𝑉𝑇𝑇_a+\beta_4 𝑂𝑉𝑇𝑇_𝑎+\beta_5 ((𝑂𝑃𝑇𝐶_a)/𝐼𝑁𝐶)+\beta_6 𝐴𝑂\]
\[𝑉_𝑡=\beta_3 𝐼𝑉𝑇𝑇_𝑡+\beta_4 𝑂𝑉𝑇𝑇_𝑡+\beta_5 ((𝑂𝑃𝑇𝐶_𝑡)/𝐼𝑁𝐶)\]
- IVTT = in-vehicle travel time
- OVTT = out-of-vehicle travel time
- OPTC = out-of-pocket travel cost
- INC = household income
- AO = auto ownership level
![]()
Forecasting Example - Full Enumeration & Naive Aggregation
In this example, naive aggregation results in 3% overprediction of auto
![]()
![]()
Forecasting Example - Full Enumeration & Classification with Naive Aggregation
In this example, classification with naive aggregation results in <1% overprediction of auto
![]()
![]()
Destination Choice
- Like mode choice, we can consider the trip distribution process using disaggregate choice models
- Destination choice models attempt to represent the decision-process by individuals when deciding where to make a trip
![]()
Destination Choice
- Unfortunately, destination choice is more complex than mode choice due to choice set explosion
- We can address challenge by making use of logit properties to draw a random sample of alternatives
![]()
Joint Mode-Destination Choice
Often mode-destination choice a joint process: choosing to walk will affect destination choice & choosing a close destination will affect mode choice
![]()
Other Choices
- Home location choice (multinomial logit) – similar to destination choice
- Vehicle ownership choice? 1, 2, 3, 4, etc. vehicles (ordered logit)
- Departure time choice? 8AM, 9AM, 10AM, etc. (ordered logit)
- Party size choice? 1, 2, 3, 4, 5, etc. persons (ordered logit)
- Choice set formation
- What destinations do I consider? Do I consider only destinations within my neighborhood? Do I consider only destinations within a certain distance or travel time?