纵向观测计数数据的对数线性模型

Generalized log linear Models for Longitudinal Count Data

目的纵向观测数据是按时间顺序对个体的某一变量进行多次观测获得的资料。本文利用广义线性模型对纵向计数数据进行了分析，充分考虑重复观测间的相关性。方法采用Ｚｅｇｅｒ和Ｌｉａｎｇ提出的广义估计方程，在拟合对数广义线性模型的同时，引入偏离参数，讨论三种协方差矩阵的结构。结果同时获得回归参数、相关参数、偏离参数的估计，完成了较为实用的运行程序，并进行了实例讨论。结论医学研究和临床试验中经常接触到纵向观测数据。

Objective The defining characteristic of logitudinal data is that individuals are measured repeatedly through time.This thesis used generalized linear models to analyze longitudinal count data.The correlation between repeated measures was considered.Methods Generalized estimating equations(GEE)proposed by Zeger and Liang were used.While fitted generalized loglinear models,we introduced dispersed parameters for three covariance structures.Results Regression,correlation and dispersed parameters were estimated simultaneously.A program was finished and an example was illustrated.Conclusion Longitudinal data often occur in medical researches and clinical trials.It is necessary to use some special methods to cope with this kind of data.