Choice Probabilities
Role of Choice Probabilitiesβ
- Essential in modeling discrete choices.
- Cannot predict choices definitively from representative utility alone because of the unobserved component.
- Choice probabilities crucial in discrete choice models.
Calculationβ
- Probability of choosing alternative i: Pniβ.
- Considers the cumulative distribution of unobserved utility components.
- The probability that decision maker n chooses alternative i is:
Pniββ=Pr(Uniβ>Unjβ,βjξ =i)=Pr(Vniβ+Ο΅niβ>Vnjβ+Ο΅njβ,βjξ =i)=Pr(Ο΅njββΟ΅niβ<VniββVnjβ,βjξ =i)=β«Ο΅β1(Ο΅njββΟ΅niβ<VniββVnjββjξ =i)f(Ο΅nβ)dΟ΅nββ
Note: Above integral is multi-dimensional integral and f(Ο΅nβ) is joint density function of Ο΅n1β,Ο΅n2β,...,Ο΅njβ.
We have to make assumptions about this joint density function.
Example from statisticsβ
Let,
fX,Yβ(x,y)then,Β P(X>Y)β={6eβ(2x+3y),0,βx,yβ₯0otherwiseβ=β«0βββ«yββ6eβ(2x+3y)dxdy=β«0ββ3eβ3ydy=53β.β
Choice Probabilities Exampleβ
Scenario:
- A person decides whether to take a car (c) or a bus (b) to work.
- Observable attributes: time (T) and cost (M) for each mode of transport.
Representative Utility Specification:
- Car: Vncβ=Ξ²0cβ+Ξ²1βTncβ+Ξ²2βMncβ.
- Bus: Vnbβ=Ξ²0bβ+Ξ²1βTnbβ+Ξ²2βMnbβ.
Assuming Known Coefficients:
- If Ξ² coefficients are known, Vncβ and Vnbβ can be calculated.
- However, unobserved factors (Ο΅ncβ and Ο΅nbβ) also affect the decision.
Choice Probability Calculation:
- For driving:
- Pncβ=Pr(Ο΅nbββΟ΅ncβ<VncββVnbβ).
- Expanded: Pncβ=Pr(Ο΅nbββΟ΅ncβ<(Ξ²0cββΞ²0bβ)+Ξ²1β(TncββTnbβ)+Ξ²2β(MncββMnbβ)).
Using Choice Probabilities To Estimate Parametersβ
Objective:
- Estimating structural parameters of the model that describe the decision makerβs preferences and behavior.
Role of Choice Probabilities:
- Choice probabilities assist in fitting the model to observed data.
- A good fit is indicated when the choice probability for the selected alternative is close to 1, and for all other alternatives, it is close to 0.
Fitting the Model:
- Finding structural parameters that align the model's choice probabilities with observed choices.
Methodology:
- The approach depends on the assumptions made about the distribution of the unobserved utility component (Ο΅).