Information criterions based on quasi-likelihood with application to over-dispersion problems

2016 Fiscal Year Final Research Report

• Project/Area Number: 25330034
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Statistical science
• Research Institution: Chiba University
• Principal Investigator: Wang Jinfang 千葉大学, 大学院理学研究科, 教授 (10270414)
• Project Period (FY): 2013-04-01 – 2017-03-31
• Keywords: 一般化線形モデル / conditional independence / cain / quasi-likelihood / causal inference / Bayesian inference / coq/SSReflect / cell regression

Outline of Final Research Achievements

(1)We proposed some new model selection criterions for semi-parametric regression models. These models only use the assumptions on mean and variance functions instead of the full parametric assumptions as used in traditional generalized linear model. (2)We formalized the theory of cain (2010) using the interactive theorem-prover Coq/SSReflect. As a consequence, we formalized the theory on conditional independence based on cain. (3)We proposed a new Bayesian prediction method by combining some independent source data with the target detailed data. For the summary source data in the form of tables, we proposed cell regression methods based on integrated predictive probabilities.

Free Research Field

数理統計学