CVEN9405 Chap.6 Random Utility and Mode Choice
Random Utility and Mode Choice
Mode choice models represent an individual's selection among available travel alternatives. The course reading separates utility into an observed systematic component and an unobserved random component, allowing analysts to predict probabilities rather than claim certainty about one traveller. Systematic utility can combine time, cost, access, transfers, reliability, traveller characteristics and interactions.
Multinomial logit converts relative utilities into probabilities through exponentials and a shared denominator. This chapter also examines availability, reference alternatives, coefficient directions and the independence of irrelevant alternatives property, especially where close substitutes share unobserved features and simple logit predicts an implausible substitution pattern.
What this chapter covers
- 01
Choice sets and explicit availability rules
- 02
Random utility as systematic utility plus unobserved error
- 03
Alternative attributes, traveller characteristics and interactions
- 04
Alternative-specific constants and reference alternatives
- 05
Multinomial logit probabilities and relative utility
- 06
IIA, close substitutes and the case for nested structures
Worked example · free
Calculate choice probabilities after a fare change
- +1Base denominator = 1.221 + 0.741 = 1.962.
- +1Base transit share = 0.741/1.962 = 37.77%.
- +1New transit utility is -0.3 + 0.4 = 0.1; denominator = 1.221 + 1.105 = 2.326.
- +1New transit share = 1.105/2.326 = 47.51%, a gain of about 9.74 percentage points.
Key terms
- Choice Set
- The alternatives available to a decision maker for the choice being modelled.
- Random Utility
- A framework that represents utility as an observed systematic part plus an unobserved random part.
- Systematic Utility
- The modelled component of utility constructed from observed attributes, characteristics and their interactions.
- Reference Alternative
- The alternative against which constants or coefficients are normalised so relative utility can be identified.
- IIA Property
- The multinomial logit restriction that pairwise probability odds are independent of the attributes of other alternatives.
Random Utility and Mode Choice FAQ
Why does a mode-choice model predict probabilities instead of certain choices?
Analysts cannot observe every perception, preference or circumstance affecting an individual decision. Random utility treats that missing information as an unobserved component, so people with the same observed attributes are represented through population choice probabilities rather than deterministic certainty.
Why must unavailable modes be removed from the logit denominator?
An unavailable alternative cannot receive probability. Leaving it in the denominator assigns some share to an impossible choice and reduces every feasible alternative's share. Availability can depend on licence, vehicle access, operating hours, mobility constraints or whether the service reaches the trip.
When is the IIA property a problem?
IIA becomes questionable when alternatives share unobserved qualities and are close substitutes, such as two nearly identical transit services. Simple logit can draw too much share from unrelated modes. Better observed attributes or a behaviourally justified nested structure may be required.
Why are utility levels not interpreted directly?
Only utility differences determine logit probabilities, so adding the same constant to every alternative changes no prediction. One alternative-specific constant is therefore normalised for identification. Interpret coefficients through contrasts, probability changes or willingness-to-pay measures, while keeping the scale and reference alternative explicit.
How does a change in one alternative affect the other mode shares?
Improving one alternative raises its systematic utility and probability while reducing the probabilities assigned to available competitors because all shares must sum to one. In multinomial logit, the pattern follows the IIA structure. Check the full probability vector after a policy change rather than reporting only the improved mode.
What distinguishes availability from a very unattractive mode?
An unavailable mode is excluded from the choice set and receives exactly zero probability. A feasible but unattractive mode remains in the denominator and can receive a small positive share. Confusing these cases distorts substitution, especially for travellers without a licence, vehicle access or a usable service at the required time.
Exam move
Write the available choice set before calculating utility. Build a table with one row per alternative and separate columns for time, cost, access, service and personal interactions. Multiply each attribute by its coefficient, add the alternative-specific term and check the sign of every contribution. Then exponentiate utilities, sum the denominator and confirm probabilities total one.
Practise adding the same constant to all utilities to verify that shares do not change, which reinforces that only differences matter. Use one policy change at a time and explain the resulting share movement through the altered utility term. For IIA, sketch a red-bus and blue-bus example and ask which alternatives should substitute most strongly.
Before choosing a nested model, check missing observed attributes and availability rules. End each question with a behavioural sentence, not only percentages: identify which intervention changed which component of utility and whether the predicted source of new riders is plausible for the planning objective.
