|
Activity Number:
|
291
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, July 31, 2007 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistics and the Environment
|
| Abstract - #309782 |
|
Title:
|
A Correlation Analysis of Bivariate Normal Data Subject to Partial Left-Censoring
|
|
Author(s):
|
Wendy Leith*+ and Jong Kim and Tom Fielden
|
|
Companies:
|
Portland State University and Portland State University and Portland State University
|
|
Address:
|
Department of Mathematics and Statistics, Portland, OR, 97207,
|
|
Keywords:
|
EM algorithm ; left-censored data ; bivariate normal ; bivariate lognormal ; correlation
|
|
Abstract:
|
Researchers working with bivariate normal data subject to partial left-censoring have not had access to software which could treat the data appropriately. In an effort to treat the data appropriately and to fully utilize the information the censored observations provide we use the EM Algorithm to obtain the MLEs of the parameters. After arriving at a final estimate we use Monte Carlo Imputation to estimate the standard errors. Finally, we extend our work to cover partially left-censored bivariate lognormal data, which is commonly studied in environmental sciences. Simulations show that the EM algorithm is an excellent choice for estimating the parameters of the bivariate normal distribution when both variables are subject to partial left censoring, and that the method is robust. The estimates perform very well regardless of sample size, strength of correlation or amount of censoring.
|
