Quasi-likelihood Method For Spatially Binary Data

Spatial data is a hot spot in current statistical studies. It appears in more and more areas, such as biological species potential distribution studies, biological neural imaging studies and epidemiological studies. Spatially binary data is a special kind of spatial data. Specificity of variable values makes parameter estimation more di?-cult, especially considering the case of spatial autocorrelation in the model. This paper mainly focuses on parameter estimation in logistic regression model of spatially binary data with spatial autocorrelation and corresponding method which is based on composite likelihood method. Composite likelihood method is used in parameter estimation and statistical inference by reducing complexity and increasing computability. The composite likelihood method for spatially binary data in this paper estimates parameters by maximize quasi-likelihood function constructed by weighted paired likelihood function. In order to verify estimation results of the composite likelihood method,this paper compares it with the ordinary maximum likelihood estimation by simulation. The simulation results show that the composite likelihood method is better than the ordinary maximum likelihood estimation. Then a real data analysis about invasion species Solanum rostratum Dunal is conducted based on the composite likelihood method, which can provide suggestions and guidance for monitoring and management of Solanum rostratum Dunal. A discussion about further study of the composite likelihood method is given at the end of this paper.