混合型態的相關資料之強軔概似推論;Robust Likelihood Inference for Correlated Mixed Type Data

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/82287

Title:	混合型態的相關資料之強軔概似推論;Robust Likelihood Inference for Correlated Mixed Type Data
Authors:	鄒宗山
Contributors:	國立中央大學統計研究所
Keywords:	相關混合資料;強軔概似函數;潛在變數模型;位置模型;Correlated mixed data;Robust likelihood;Latent variable model;Location model
Date:	2020-01-13
Issue Date:	2020-01-13 14:36:48 (UTC+8)
Publisher:	科技部
Abstract:	本研究計劃想探知, 如使用了在分析單維度資料時皆具有可被強軔化的分配, 如gamma; Poisson; normal; negative binomial與binomial, 做為latent variables與location model方法中分配假設時，對這兩種模型的強軔性的影響為何?是賦予了latent variables與location model方法擁有原來這些分配單維度時可被強軔化的性質?還是這些具單維度分配之強軔性被latent variables與location model模型破壞掉? 我們針對的資料是多維混合型資料(如:相關的(個數，連續)型資料)，相關的(個數，名目(nominal)型資料)，相關的(名目，連續)型資料)。我們亦將試著以個人所長期研究的強軔概似函數法來分別與latent variables與location model模型比較它們在(1) 一致性 (consistency) (2) 有效性 (efficiency) 的表現。 ;Amongst various methods for analyzing mixed correlated data, latent variables, location model, and Copula are frequently adopted. However, once model assumptions are incorrect, Copula not only faces the problems of giving inconsistent maximum likelihood parameter estimates, the estimation of the association parameter in Copula can be very much unstable. This makes application of Copula in real data analysis quite questionable. The performance of latent variables method and location model in terms of robustness under model misspecification are rarely studied. That is, how sensitive is the validity of inference derived from latent variables method and location model, when, in fact, data distributions do not conform to the model assumptions? There are several univariate distributions, including gamma, Poisson, normal, negative binomial and binomials that can be robustified under model misspecification. The objectives of this research project includes 1. When any of the above univariate distributions that can be robustified are adopted as part of the latent variables method/location model, will the property of robustness be reserved?2. If that is the case, can the latent variables/location model be robustified to deliver legitimate likelihood inference? How do the adjusted robust likelihood derived from the latent variables/location model compare to the robust likelihood derived from the independence working model approach that my researches in recently years are most devoted to?3. If that is not the case, what contributes to the problem and any possible remedies?We will focus our research on data of mixing types, such as correlated (count, continuous), (nominal, continuous), (count, nominal) data. These data types are certainly the scenarios that latent variables method, location model and Copula are most popular.
Relation:	財團法人國家實驗研究院科技政策研究與資訊中心
Appears in Collections:	[Graduate Institute of Statistics] Research Project

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	293	View/Open

社群 sharing

Loading...