In this thesis, we propose tests for detecting over-dispersion and excess of zeros in count data. Current tests for these two purposes mostly assume there is no measurement error in covariates. However, this kind of measurement error is inevitable. Therefore, we build new tests for over-dispersion and extra zero counts under the framework of classical measurement error model. We carry out simulation studies to evaluate the empirical power and level of the proposed tests and compare them with the existing tests. We also apply these tests to a real data concerning how health-related quality-of-life influences the number of clinical consultations for colorectal neoplasm patients and check which model should be chosen among Poisson model, negative binomial model and zero-inflated Poisson...[ Read more ]

In this thesis, we propose tests for detecting over-dispersion and excess of zeros in count data. Current tests for these two purposes mostly assume there is no measurement error in covariates. However, this kind of measurement error is inevitable. Therefore, we build new tests for over-dispersion and extra zero counts under the framework of classical measurement error model. We carry out simulation studies to evaluate the empirical power and level of the proposed tests and compare them with the existing tests. We also apply these tests to a real data concerning how health-related quality-of-life influences the number of clinical consultations for colorectal neoplasm patients and check which model should be chosen among Poisson model, negative binomial model and zero-inflated Poisson model.