STRUCTURAL COMPARATIVE CALIBRATION USING THE EM ALGORITHM

被引:9
作者
BOLFARINE, H [1 ]
ROJAS, MG [1 ]
机构
[1] UNIV SAO PAULO,IME,DEPT ESTAT,BR-14152990 SAO PAULO,BRAZIL
关键词
D O I
10.1080/757584622
中图分类号
O21 [概率论与数理统计]; C8 [统计学];
学科分类号
020208 ; 070103 ; 0714 ;
摘要
The problem of comparing the linear calibration equations of several measuring methods, each designed to measure the same characteristic on a common group of individuals, is discussed. We consider the factor analysis version of the model and propose to estimate the model parameters using the EM algorithm. The equations that define the 'M' step are simple to implement and computationally inexpensive, requiring no additional maximization procedures. The derivation of the complete data log-likelihood function makes it possible to obtain the expected and observed information matrices for any number p(greater than or equal to 3) of instruments in closed form, upon which large sample inference on the parameters can be based. Re-analysis of two actual data sets is presented.
引用
收藏
页码:277 / 292
页数:16
相关论文
共 10 条