perfectjt 发表于 2008-6-13 20:29
各位大虾,哪里能找到一些非集计的一些资料,比如logit模型的推导与应用实例等,小弟急用。谢谢
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’