A Fundamental Study On Private Car Ownership And Usage

With the rapid development of motorization,China now is challenged by traffic congestion,traffic accident,traffic emission,and traffic energy consumption problems which are caused by the usage of private cars.Investigating private car ownership and usage is not only useful in predicting future car ownership and energy consumption,but also helpful for government departments and auto industry to make policies related to cars.To achieve this goal,this paper employed Bayesian MCMC(Markov Chain Monte Carlo)approach to study household private car ownership and usage behavior.The main contents are listed as follows:1)Firstly,this paper examined the determinants of private car ownership of 32 provincial capital cities from 2001 to 2011.Moreover,differences between large scale cities and small scale cities as well as eastern cities and middle_western cities were investigated.In order to capture the individual effects(heterogeneity),the fixed and random effects models based on double logarithm form were adopted and compared.The estimated results showed that economy condition,income per capital,highway mileage per capital,and average road pavement could significantly promote the growth of private car ownership.Additionally,large scale cities and small scale cities exhibited distinct travel demand patterns.Specifically,travel demand in large scale cities increased slowly and was almost saturated.By comparison,travel demand in small scale cities rose rapidly.Meanwhile,demand for intercity travel was greater in large scale cities and eastern cities than their corresponding parts.Finally,this paper studied the the inhibiting effect of metro in preventing the growth of private car ownership.The estimation results suggested that although metro system could slow the growth of private car ownership,the influence was small and not statistically significant.2)Secondly,this paper studied household private car ownership considering the interdependence of households.The approach employed was spatial autoregressive Binary Probit model based on bayesian MCMC.The interdependence was captured by weight matrix which was defined by the similarity of household structure.The estimation results suggested that households were indeed interrelated and model performance was greatly improved compared with the traditional Binary Probit model.Moreover,children and high consumption level were significantly and positively associated with ownership level.The influence of apartment was large.Households which already had their own apartments tended to buy private cars.3)Thirdly,a sample selection model based on bayesian MCMC estimation was employed to jointly study household private car ownership and usage.This model was proposed by Hasselt in 2011,but seldom used in the empirical analysis.It assumes that two error terms from ownership equation and usage equation follow the joint normal distribution,which is a traditional assumption in the sample selection model.In order to prove effectiveness of the model,a simulation example was conducted.Simulation results suggested that it had enough power in revealing parameters' true values.Then it was used in the empirical study.It was found that ownership decision and usage decision were indeed correlated.If the interdependence was ignored,parameters estimated would be biased.In terms of influence of variables,estimation results suggested that existence of children and apartment,high consumption level stimulated households to purchase private cars.Moreover,cars with high displacement were found to be used more.And maintenance fee were negatively correlated with the private car usage.4)Finally,this paper put forward a sample selection model based on mixture of normals and bayesian MCMC.This model can relax the joint normal distribution in the traditional sample selection model.In order to justify the model,a simulation example was proposed and analyzed.Simulation results suggested that the new model could reveal true values of parameters and outperformed the traditional one.However,the model may experience overfitting problem when the number of components is large.To avoid this problem,different tests were proposed.In the empirical analysis,the model with 3 components was more suitable considering goodness of fit,estimates stability,as well as model complexity.Comparing estimates with those from the traditional normal model,the study found that estimates from the normal model were biased in terms of significance level and magnitude.Finally,both of simulation example and empirical analysis suggested that new model with 2 or 3 components could capture most traits of the data.Therefore,the new model is very efficient.