東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

Regression analysis for large binary...

Carey, Vincent James.

FindBook

Google Book

Amazon

博客來

Regression analysis for large binary clusters.

紀錄類型:	書目-語言資料,印刷品 : Monograph/item
正題名/作者:	Regression analysis for large binary clusters./
作者:	Carey, Vincent James.
面頁冊數:	186 p.
附註:	Adviser: Scott L. Zeger.
Contained By:	Dissertation Abstracts International54-01B.
標題:	Biology, Biostatistics. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=9313342

Regression analysis for large binary clusters.
Carey, Vincent James.

Regression analysis for large binary clusters. - 186 p.

Adviser: Scott L. Zeger.

Thesis (Ph.D.)--The Johns Hopkins University, 1993.

Marginal models for multivariate binary data permit separate modelling of (1) the relationship of each response with explanatory variables and (2) the pairwise odds ratios for pairs of responses. When the former is the scientific focus, the generalized estimating equation method (GEE) (Liang and Zeger, 1986) is easy to implement and gives reasonably efficient estimates of regression coefficients. However, estimates of the association among the binary outcomes can be highly inefficient. When the association model is also a focus, simultaneous modelling of each response and all their pairwise products using quadratic estimating equations (Prentice, 1988) gives efficient estimates of association parameters as well. However, this becomes computationally infeasible as the dimension of the binary vector (cluster size) gets large. This dissertation studies an alternate approach, called "alternating logistic regressions" (ALR) for simultaneously modelling the regression and association in problems with large clusters. The ALR algorithm iterates between (a) logistic regression of the response on explanatory variables using GEE to estimate regression coefficients and (b) logistic regression of every response on each of the other responses from the same cluster using an appropriate offset. Phase (a) yields estimates of population-averaged effects of covariates, and phase (b) yields estimates of the clustering parameters, which correspond to a linear regression analysis of pairwise log odds ratios. For clusters of size n, alternating logistic regression involves evaluation and inversion of matrices of order Subjects--Topical Terms:

1018416
Biology, Biostatistics.

Regression analysis for large binary clusters.
LDR:02945nam 2200265 a 45 001 949783
005 20110525
008 110525s1993 ||||||||||||||||| ||eng d
035 $a (UMI)AAI9313342
035 $a AAI9313342
040 $a UMI $c UMI
100 1 $a Carey, Vincent James. $3 1273178
245 1 0 $a Regression analysis for large binary clusters.
300 $a 186 p.
500 $a Adviser: Scott L. Zeger.
500 $a Source: Dissertation Abstracts International, Volume: 54-01, Section: B, page: 0312.
502 $a Thesis (Ph.D.)--The Johns Hopkins University, 1993.
520 $a Marginal models for multivariate binary data permit separate modelling of (1) the relationship of each response with explanatory variables and (2) the pairwise odds ratios for pairs of responses. When the former is the scientific focus, the generalized estimating equation method (GEE) (Liang and Zeger, 1986) is easy to implement and gives reasonably efficient estimates of regression coefficients. However, estimates of the association among the binary outcomes can be highly inefficient. When the association model is also a focus, simultaneous modelling of each response and all their pairwise products using quadratic estimating equations (Prentice, 1988) gives efficient estimates of association parameters as well. However, this becomes computationally infeasible as the dimension of the binary vector (cluster size) gets large. This dissertation studies an alternate approach, called "alternating logistic regressions" (ALR) for simultaneously modelling the regression and association in problems with large clusters. The ALR algorithm iterates between (a) logistic regression of the response on explanatory variables using GEE to estimate regression coefficients and (b) logistic regression of every response on each of the other responses from the same cluster using an appropriate offset. Phase (a) yields estimates of population-averaged effects of covariates, and phase (b) yields estimates of the clustering parameters, which correspond to a linear regression analysis of pairwise log odds ratios. For clusters of size n, alternating logistic regression involves evaluation and inversion of matrices of order $n \sp2$ rather than $n \sp4$ as is needed for extended estimating equations. Asymptotic calculations and finite-sample simulations are used to show that alternating logistic regression estimates are reasonably efficient over a range of problems. The new method is illustrated with an analysis of data on diarrhea prevalence in rural Malawi, in which clusters range in size to n = 329, and in a study of neuropsychological tests on patients with epileptic seizures, in which clusters are size n = 18.
590 $a School code: 0098.
650 4 $a Biology, Biostatistics. $3 1018416
650 4 $a Statistics. $3 517247
690 $a 0308
690 $a 0463
710 2 $a The Johns Hopkins University. $3 1017431
773 0 $t Dissertation Abstracts International $g 54-01B.
790 $a 0098
790 1 0 $a Zeger, Scott L., $e advisor
791 $a Ph.D.
792 $a 1993
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=9313342