@ARTICLE{16515842_1985,
author = {Chesher, Andrew},
keywords = {errors, mathematical model, statistical methods, test, математическая модель, статистический метод},
title = {Score Tests for Zero Covariances in Recursive Linear Models for
Grouped or Censored Data },
journal = {Journal of Econometrics},
year = {1985},
month = {},
volume = {28},
number = {3},
pages = {291-305},
url = {http://ecsocman.hse.ru/text/16515842/},
publisher = {},
language = {ru},
abstract = {Efficient estimation of normal linear simultaneous equations systems
is frequently easier when error covariances are zero. The score test
is examined for the hypothesis that an error covariance is zero in a
2-equation recursive normal linear simultaneous equations system
where endogenous variates may be completely observed, censored, or
grouped. The model contains the seemingly unrelated regression
equation model and its analogues for grouped and censored data as
special cases. The score test for the hypothesis explores the sample
covariance of suitably defined residuals and is closely related to
the Information Matrix test calculated for the restricted model in
which the error covariance is zero. These results are obtained by
viewing a non-zero error covariance as emerging from correlated
random variation in intercept parameters that can be detected through
the use of Chesher's (1984) test for neglected heterogeneity },
annote = {Efficient estimation of normal linear simultaneous equations systems
is frequently easier when error covariances are zero. The score test
is examined for the hypothesis that an error covariance is zero in a
2-equation recursive normal linear simultaneous equations system
where endogenous variates may be completely observed, censored, or
grouped. The model contains the seemingly unrelated regression
equation model and its analogues for grouped and censored data as
special cases. The score test for the hypothesis explores the sample
covariance of suitably defined residuals and is closely related to
the Information Matrix test calculated for the restricted model in
which the error covariance is zero. These results are obtained by
viewing a non-zero error covariance as emerging from correlated
random variation in intercept parameters that can be detected through
the use of Chesher's (1984) test for neglected heterogeneity }
}