The Construction And Analysis Of Spatial Weight Matrix Based On AMOEBA Improvements And WASG Methods

In the field of spatial econometrics and other fields,in order to describe and understand the potential structure of different research units or regions,a convenient way is to show through the definition of spatial weight matrix.In the empirical,we can not put an end to the subjective purpose of the scholar.And the choice of the weight matrix will make the final estimation result of the model deviate or even be(big)wrong if it is unreasonable.Therefore,its choice becomes the key problems of spatial analysis.In view of this,this paper first introduces the definition,classification,common type construction of spatial weight matrix,the spatial autocorrelation statistics and spatial econometrics model.Then,We use the local Moran'I instead of the local Gi*statistic on the basis of AMOEBA(A Multidirectiornal Optimum Ecotope—Based Algorithm)and combined with the "economic distance" weight to build the weight matrix.At the same time,we use Monte Carlo simulation,using Maximum Likelihood to estimation model and comparative studies under different weight ma-trices.Then this paper presents a spatial weights matrix automatic selection gen-erating method(Recorded as WASG method)based on the local Moran'I,and through Monte Corlo simulation for comparative analysis under different weight matrices.Finally,we also made a spatial autocorrelation analysis of the total ener-gy consumption of 15 cities in Xinjiang in 2014.The main conclusions and innovations are:(1)In the simulation,the estimation result corresponding to the weight matrix constructed by the improved AMOEBA is better,and the corresponding estimation result under the large sample is better than that of the other weight matrix.(2)Using the WASG method by us,the results of the spatial weight matrix obtained are more close to the actual value(true value)than those of the other types in the numerical simulation experiment(large and small samples).(3)In the empirical,we have that the spatial autocorrelation and heterogeneity coexist in the total energy consumption of 15 states in Xinjiang in 2014,and the weight matrix constructed in this paper can reflect the spatial autocorrelation of spatial units.(4)Through lonte Carlo simulation and demonstration,we found that AIC(Akaike Information Criterion),the global Moran'I and so on can provide the basis for the selection or comparison of weight matrix or model,and play an im-portant role in the follow-up analysis and explanation of the phenomena of various correlations with the weight matrix.