期刊
JOURNAL OF STATISTICAL PLANNING AND INFERENCE
卷 113, 期 2, 页码 527-539出版社
ELSEVIER SCIENCE BV
DOI: 10.1016/S0378-3758(01)00320-2
关键词
delta method; errors in variables; exponential family; estimating equations; measurement error; sandwich estimator
Measurement error occurs when one or more regression model covariates are measured with error, and is a common problem in occupational health and other fields. If the effects of measurement error are not corrected, estimates of the regression coefficients and their variances are biased. We consider the situation when an external validation study provides information about the measurement error process. Two related approaches for correcting measurement error, both called regression calibration, have been proposed. Although the estimators for the corrected regression coefficients can be shown to be the same under the two approaches, the asymptotic variances are derived using two different methods and have not been compared. In this paper, we show that these two methods give algebraically identical estimators for the corrected regression coefficients and their asymptotic variances, in the general case of multiple covariates measured with error, for a generalized linear model of exponential family form. (C) 2002 Elsevier Science B.V. All rights reserved.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据