Analysis of Binary Data via Spatial-Temporal Autologistic Regression Models

摘要: Spatial-temporal autologistic models are useful for binary data that measured repeatedly over time on a spatial lattice. They can account effects of potential covariates and spatial-temporal statistical dependence among the data. However, traditional parametrization model presents difficulties in interpreting parameters across varying levels dependence, where its non-negative autocovariates could bias realizations toward 1. In order to achieve interpretable parameters, centered regression has been developed. Two efficient inference approaches, expectation-maximization pseudo-likelihood approach (EMPL) Monte Carlo likelihood (MCEML), have proposed. Also, Bayesian is considered studied. Moreover, performance efficiency these three approaches various sizes sampling lattices numbers points through both simulation study real example addition, We consider imputation missing values models. Most existing methods not admissible impute values, because they disrupt inherent structure lead serious during or computing issue. methods, iteration-KNN maximum entropy imputation, proposed, them relatively simple yield reasonable results. summary, main contributions this dissertation development with parameterization, proposal EMPL, MCEML, obtain estimations parameters. presented data, which generate reliable imputed time.