Nested Logit Model课程.pptVIP

Nested Logit Model by Asif Khan Contents Independence of Irrelevant Alternatives Independence of Irrelevant Alternatives IIA property convenient for estimation but fails on consumer behavior. Unrealistic assumption: why? b/coz: people will travel by white bus, if grey bus is not available without switching to taxi, which may be expansive. More realistic situation may be: White bus = 40% Taxi = 60% IIA biggest drawback of MNLM model In NLM we assume that unoberved utility has GEV distribution: exp(-∑kk=1 (∑e-εnj/λk) λk) Generalization of univerate distribution in logit model. The unoberved utility is correlated within nests. εnj uncorrelated across nests Parameter λk is a measure of degree of independence in εnj in nest k. Higher λk means less correlation higher independence vice versa. McFadden (1978) used 1-λk as indication for correlation If λk =1 means complete independence or no correlation If λk =1 nested logit model reduce to standard logit model Observed Utility (Unj) is: Unj = UT + UC|T UT = utility from travel mode e.g., auto or transit UC|T = utility from travel choice e.g., car, bus etc. Random utility = Marginal utility + Conditional utility Probabilities in nested logit is a product of two simple logits. Pi = Prob (nest containing i) x Prob (i, given nest containing i) e.g., Pi = Prob (auto) x Prob (car, given auto) Pi = Pn * Pi|n Greene, William H. 2003. Econometric Analysis. 5th ed. Prentice Hall, USA. Jeffrey, Wooldridge M. 2001. Econometric Analysis of Cross Section and Panel Data. The MIT, USA. The Nested Logit Regression Model. /~statmath/stat/all/cdvm/cdvm8.html Kenneth, Train. 2003. Discrete Choice Methods with Simulation. Cambridge University Press, USA. Discrete Dependent Variable Models. /onlinepubs/nchrp/cd-22/v2chapter5.html Maddala, G. S. 1983. Limited-Dependent and Qualitative Variables in Econometrics. Cambridge University Press USA. McFadden