| Co-Investigator(Kenkyū-buntansha) |
KANDA Takashi Hiroshima Inst. of Technology, Fac.of Enviromentrics, Prof., 環境学部, 教授 (40098679)
OHOTAKI Megu Hiroshima Univ., Research Inst.for Radiat.Biol. and Med., Prof., 原爆放射能医学研究所, 教授 (20110463)
SHOHOJI Takao Hiroshima Univ., Fac.of Integrated Arts and Sciences, Prof., 総合科学部, 教授 (00033910)
FUJISAWA Hironori Tokyo Institute of Technology, Graduate School of Information Science and Engineering, Research Assistant, 大学院・情報理工学研究科, 助手 (00301177)
WAKAKI Hirofumi Hiroshima Univ., Graduate School of Science, Associate Professor, 大学院・理学研究科, 助教授 (90210856)
|
|---|
| Research Abstract |
The purpose of this project is to development statistical methods and softwares for growth or repeated measures data. As one of the main results we introduced the growth curve model with explanatory variables which are hierarchically related, and gave some results on its statistical analysis. More precisely, we pointed an importantce of the model, and proposed estimation methods based on the maximum likelihood, tests for mean parameters, simultaneous confidence intervals for severale growth curves, tests for covarinace structures, model selection criteria, etc. (Ann.Inst.Statist.Math., Comm.Statist., Multivariate Anal.). For unblanced data, assuming random effect covariance structure, we derived an extention of the previous statistical inference procedures (submitted). Further, based on these results, we completed software 「"GROWT3" Ver 1.0 : Program correspond to Windows for statistical analysis on growth curve models with linear structures of parameters.」 (Megu Ohotaki, Akari Satoh, Kennichi Satoh and Yasunori Fujikoshi). Related to this software, we also completed software for estimating a mixture distribution.
In this project we dealt with traking analysis of height and weight of human. We proposed growth curve fitting based on a non-linear model and obtained some new findings for relationships among biological parameters (Acta Med.Auxol., Environmentrics).
As basic works related to this project, we examined modifications of model selection criteria and their basic properties. Especially, we derived a modified AIC criterion in multivariate linear model (Hiroshima Math.J.), and obtained some basic properties on cross- validation criterion in growth curve model. As another result, we proposed an estimation method for misclassification errors based on binary repeated measures (Biometrics).
|
