同问~~~~

同问问。。。

Random utility model Setup
•Let n be labeled as a consumer and j be labeled as a choice
•Let Unj  be utility of a consumer’s choice j
•Decision rule:

       choose alternative i if and only if  Uni  > Unj      ∀ j <> i
•Utility is decomposed as :   Unj  =       Vnj               +    εnj

                                                      = measureable utility + error
•Pni = Prob(Uni > Unj )  = Prob(εnj − εni < Vni − Vnj )

    =∫ I (εnj − εni < Vni − Vnj) f (εn) dεn                    ∀ j <> i      (1)

  I = indicator function

Note:   Different error distribution assumptions will results in different   likelihood functions, and hence different estimation models

  If an error distribution is normal(logistic) with two choices, it is   equivalent to a probit/logit model.

  When there are many choices, the computation of a probit is very   ‘expensive’