http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Ahn Byoung Jin(安秉珍) 건국대학교 경제경영연구소 1985 商經硏究 Vol.10 No.1
Collinearity among independent variables in multiple linear regression can have sever effects on the precision of prediction equation. To overcome this problem, many alternate type of biased estimation have been proposed. In this we have compared the ordinary least squares, principal component and restricted estimators of regression coefficient using mean-squared error sum criterion. we have shown that the principal component estimator is identical to the least squared estimator with a special restriction and using this fact we have proposed a specific method which determines the number of latent vectors to be deleted in principal component regression. Since the proposed method in this paper contains the least squares estimator which may be poor under collinearity, the estimation problem of proposed criterion deserved further consideration.
로지스틱 모형에서 변수선택에 영향을 미치는 관측값에 대한 연구
안병진(Byoung Jin Ahn) 건국대학교 경제경영연구소 1994 상경연구 Vol.19 No.1
As is the case in linear regression, model fitting via logistic regression is also sensitive to influential cases. The values of statistic used for variable selection criteria can be reduced remarkably by excluding only a few influential cases. Furthermore, different subsets of explanatory variables change the influence patterns for the same response variable. Leger and Altman(1993) introduce a statistic, namely unconditional Cook’s distance, to assess the influence of each case on the variable selection procedure. We adopt same idea to logistic model and obtain an unconditional likelihood distance for the detection of influential cases. And some variable selection criteria for logistic model are also discussed.
안병진(Byoung Jin Ahn),오상은(Sang Eun Oh) 건국대학교 경제경영연구소 1996 상경연구 Vol.21 No.1
It is well known fact that the least squares estimators in multiple linear regression model have several serious problems when the collinearity among the independent variables exists. When ridge regression is used to reduce the effect of collinearity, not all the cases in a data set play an equal role in determining estimater and the influence of each case changes as a function of shrinkage parameter. For these reasons, it seems necessary to monitor the changes in shrinkage parameter as individual cases are in turn removed from the set of data. In this paper, a deletion formula of shrinkage parameter is proposed and modified versions of DFFITS and Cook’s D for detection of influential points are also disscussed for ridge regression.
안병진(Byoung Jin Ahn),장인석(Jang In Seog) 건국대학교 경제경영연구소 1997 상경연구 Vol.22 No.2
A new method, called combined type estimator, is proposed for estimation in linear model. It combines ridge estimator and shrinkage estimator with subset selection. If the regression equations generated by a procedure do not change drastically with small changes in the data, the procedure is called stable. Subset selection is unstable, ridge and shrinkage are very stable, and combined type estimator is intermediate. A method for choosing shrinkage parameters is also discussed and this method is based on a prediction-oriented criterion, which is a generalized version of Mallows Cp statistic.
안병진(Byoung Jin Ahn) 건국대학교 경제경영연구소 1998 상경연구 Vol.23 No.2
Two problems which often plague researchers using regression techniques are multicollinearity and nonnormal error distributions. A number of authors have proposed robust regression procedure that are robust to nonnormal error distributions and have suggested biased estimation methods for multicollinearity problem. Although we usually think of there two problems separately, in a significant number of practical situations nonnormality and multicollinearity occur simultaneously. Since robust regression estimates are frequently unstable when the design matrix is ill-conditioned, it would be desiable to have a technique for dealing directly with both problems. Askin and Montgomery(1980) discuss augmented robust estimators as a way of combining biased and robust regression techniques. Based on this idea, this paper suggests the robust versions of Nonnegative Garrot and Lasso. It seems necessary that Monte Carlo study be done to compare the performance of the various type of robust biased estimators.
Ahn Byoung Jin(安秉珍) 건국대학교 경제경영연구소 1985 상경연구 Vol.10 No.1
Collinearity among independent variables in multiple linear regression can have sever effects on the precision of prediction equation. To overcome this problem, many alternate type of biased estimation have been proposed. In this we have compared the ordinary least squares, principal component and restricted estimators of regression coefficient using mean-squared error sum criterion. we have shown that the principal component estimator is identical to the least squared estimator with a special restriction and using this fact we have proposed a specific method which determines the number of latent vectors to be deleted in principal component regression. Since the proposed method in this paper contains the least squares estimator which may be poor under collinearity, the estimation problem of proposed criterion deserved further consideration.
A Ridge-type Estimator For Generalized Linear Models
Byoung Jin Ahn(안병진) 한국통계학회 1994 응용통계연구 Vol.7 No.1
일반화 선형모형에서도 회귀모형에서와 마찬가지로 다공선성이 존재할 경우 여러가지 문제가 발생한다. 이를 극복하기 위한 한가지 방법으로 능형형태의 추정량과 그의 축소 모수를 결정하는 방법에 대하여 다루었다. It is known that collinearity among the explanatory variables in generalized linear models inflates the variance of maximum likelihood estimators. A ridge-type estimator is presented using penalized likelihood. A method for choosing a shrinkage parameter is discussed and this method is based on a prediction-oriented criterion, which is Mallows`s C_L statistic in a linear regression setting.