Mixed Logit Model And Its Application In Analysis Of The Health Care Choice In Rural China

Discrete choice analysis is used to understand how individuals make choices under choice set of econometric theory, the individual choice of explanatory variables can be divided into two categories affected by the choice, the other is unique to the individual, the former may be the price, time, location, etc, the latter category is often used as age, gender, income, and so variable. Simple two choice produced in the sixties and seventies of last century, because of its problem-solving effectiveness, and gradually developed into a large family of models. The first choice is to promote a number of binomial Logit model, the choice probability have analytical solutions, and it is globally convex, so its parameter estimation is also very convenient. But there are some restrictions of Logit model because of IIA assumption. Although the IIA assumption is mathematically convenient, and in actual use can be achieved, but it still limits the modeling of random utility models of discrete choice. For this reason, people want to do some expansion to Logit model to enable it avoid the ⅡA assumption. One solution is Mixed Logit models. Mixed Logit model assumes that the parameter is a random parameter to obey a certain distribution, which contains its selection probability integral, so that the numerator and denominator can not cancel each other out, whish avoiding the IIA assumption. Meanwhile Mixed Logit model can approximate any random utility models. Therefore, due to the above two reasons Mixed Logit model became the focus of research Difficulties in Mixed Logit model estimating is how to simulate it’s selection probability, it is necessary to use the numerical method to calculate the likelihood function. There are five methods of computing selection probabilities in history. Because of commercial software in general and did not specify the default condition estimation methods, users often do not know the meaning behind it. For this reason we implement the first three methods do some numerical simulations. They are Monte Carlo, Quasi Monte Carlo, Gauss Quadrature. Focused on comparing the accuracy of their estimates and time consumption, and concluded that the consolidated accuracy and time consuming, Quasi Monte Carlo is the best estimation method.In our empirical study, We chose to use the CHNS data, site selection for rural residents for medical treatment were analyzed. Empirical results show that the initial fee, the maximum amount of compensation, and the final cost for the farmer decided to go to the doctor what medical institutions price factor does not play a dominant role.