| Generalized linear models play an important role in the analysis of longitudinal data,but such models have limited flexibility in adapting to skewed data.In addition,existing estimation methods often rely heavily on the specification of random effects,so the corresponding distribution is sensitive to the selection of random effects.We incorporate sequence-dependent and undistributed random effects into the Tweedie generalized linear model to accommodate the range of skewness and the structure of covariance under longitudinal data.For Tweeide random effects model,this paper adopts the traditional estimation method of the traditional frequency school,and the main research contents can be summarized as follows:1.On the basis of Tweedie random effects model,the regression parameters and the covariance of discrete parameters are respectively estimated.Next,we performed statistical diagnosis on the model,carried out local impact analysis on the model,and identified potential strong impact points in the data points.Four disturbance modes of case weight disturbance,response variable disturbance,covariable disturbance and discrete parameter disturbance are established respectively.Then the effectiveness of our perturbation method is proved by simulation and case analysis.2.On the basis of the original Tweedie random effects model,we added the nonparametric part to study the parameter estimation of the semi-parametric Tweedie random effects model.Finally,the data were incorporated into the semi-parametric Tweedie random effects model by simulation and case analysis to prove the feasibility of our method. |
