東華大學圖書館 |

Language: English

Help

回圖書館首頁

手機版館藏查詢

Back

Switch To: Labeled | MARC Mode | ISBD

Evaluation of Bayesian diagnostic me...

Yan, Guofen.

Linked to FindBook

Google Book

Amazon

博客來

Evaluation of Bayesian diagnostic methods for hierarchical data.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Evaluation of Bayesian diagnostic methods for hierarchical data./
Author:	Yan, Guofen.
Description:	160 p.
Notes:	Source: Dissertation Abstracts International, Volume: 64-05, Section: B, page: 2259.
Contained By:	Dissertation Abstracts International64-05B.
Subject:	Statistics. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3092040

Evaluation of Bayesian diagnostic methods for hierarchical data.
Yan, Guofen.

Evaluation of Bayesian diagnostic methods for hierarchical data. - 160 p.

Source: Dissertation Abstracts International, Volume: 64-05, Section: B, page: 2259.

Thesis (Ph.D.)--Case Western Reserve University, 2003.

In the initial stages of data analysis where a single model is tentatively proposed and the nature of the departure of this model is not well specified, techniques are needed to assess whether this model provides an adequate fit of the data. Over the past twenty years a number of Bayesian model diagnostic techniques have been developed, but there has been little evaluation of their properties. This study is an investigation of several Bayesian methods that use posterior predictive assessment to detect incorrect specification of the variance (covariance) structure of a fitted model. We focus on the common situation where the data coine from a model with a two-stage hierarchical structure while the fitted model doesn't have this structure. We only consider methods that permit the use of noninformative prior distributions, and that are potentially useful for detecting misspecification of the variance structure.Subjects--Topical Terms:

517247
Statistics.

Evaluation of Bayesian diagnostic methods for hierarchical data.
LDR:03173nmm 2200301 4500 001 1862401
005 20041202093443.5
008 130614s2003 eng d
035 $a (UnM)AAI3092040
035 $a AAI3092040
040 $a UnM $c UnM
100 1 $a Yan, Guofen. $3 1949957
245 1 0 $a Evaluation of Bayesian diagnostic methods for hierarchical data.
300 $a 160 p.
500 $a Source: Dissertation Abstracts International, Volume: 64-05, Section: B, page: 2259.
500 $a Adviser: Joseph Sedransk.
502 $a Thesis (Ph.D.)--Case Western Reserve University, 2003.
520 $a In the initial stages of data analysis where a single model is tentatively proposed and the nature of the departure of this model is not well specified, techniques are needed to assess whether this model provides an adequate fit of the data. Over the past twenty years a number of Bayesian model diagnostic techniques have been developed, but there has been little evaluation of their properties. This study is an investigation of several Bayesian methods that use posterior predictive assessment to detect incorrect specification of the variance (covariance) structure of a fitted model. We focus on the common situation where the data coine from a model with a two-stage hierarchical structure while the fitted model doesn't have this structure. We only consider methods that permit the use of noninformative prior distributions, and that are potentially useful for detecting misspecification of the variance structure.
520 $a We have considered these methods: distribution of the Bayesian standardized residuals, distribution of the individual posterior predictive p values, the posterior predictive p value and the partial posterior predictive p value using a checking function. These methods were first applied to simulated data sets created under different scenarios. Then theoretical (and additional simulation) studies were conducted to investigate the properties of the most promising methods.
520 $a Our study suggests that a wrongly fitted model can be easily identified if it simply misspecifies the variance structure of a linear model. It is difficult to distinguish between two models where the fitted model doesn't have any hierarchical structure, while the true data model has a hierarchical stage. The method of the distribution of the individual posterior predictive p values looks the most promising. The posterior predictive p value with an effective checking function is a useful method. Also, the usual F statistic is an effective checking function even for a small sample size. The use of a checking function separately for subpopulation characteristics may be an important method to disclose specific aspects of the model that are deficient. It is also clear that one should not use the sufficient statistics of the fitted model as a checking function.
590 $a School code: 0042.
650 4 $a Statistics. $3 517247
650 4 $a Biology, Biostatistics. $3 1018416
650 4 $a Health Sciences, Public Health. $3 1017659
690 $a 0463
690 $a 0308
690 $a 0573
710 2 0 $a Case Western Reserve University. $3 1017714
773 0 $t Dissertation Abstracts International $g 64-05B.
790 1 0 $a Sedransk, Joseph, $e advisor
790 $a 0042
791 $a Ph.D.
792 $a 2003
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3092040