Negative binomial regression and Poisson regression results in the change of water to reduce fluoride in the comparative analysis.doc

 PAGE \* MERGEFORMAT 6 Negative binomial regression and Poisson regression results in the change of water to reduce fluoride in the comparative analysis Author: Chu Huizhu Gao Yan-Hui ZOU Yu-hua LI Bo-ling [Abstract] Objective: To investigate the existence of discrete data over discrete phenomenon of regression analysis. Methods: To compare the negative binomial regression and Poisson regression in the changes of water to reduce fluoride Evaluation of data analysis and goodness of fit. Results: Poisson regression underestimated the variance of parameter estimation, the negative binomial regression goodness of fit higher. Conclusion: The negative binomial regression can be used to analyze the existence of discrete data over discrete phenomenon. [Keywords:] Negative binomial regression; Poisson regression; over-dispersion; endemic dental fluorosis Poisson regression used in research in one or more independent variables on the intensity of the impact of the incident, when the outcome variable model required to obey Poisson distribution, ie events are independent, and has the overall mean and variance equal to the overall characteristics. However, in medical research, many incidents of non-independent, such as infectious diseases, genetic diseases, endemic diseases and so on. This information is characterized by the observed variation (variance) are often greater than the Poisson distribution of mutations, namely, the phenomenon of over-dispersion (over-dispersion). In general, such information may be based on negative binomial distribution, using negative binomial regression method to analyze the various factors on the strength of the impact of the incident. In this study, Guangdong Province, Chaoyang City, change the water to reduce fluoride data, using negative binomial regression of water improvement measures to reduce the assessment of dental fluorosis prevalence rate of primary school results, compared with Poisson regression. 1 Materials and methods 1.