JaLCDOI 10.18926/fest/19719
フルテキストURL 004_057_075.pdf
著者 大竹 正徳| 田中 豊| 冨田 誠|
抄録 An analytic approach that provides explicit estimates of risk on cataract and epilation data is evaluated by reasonableness of conceivable relative risk models regarding a simple, odds, logistic or Gompertz regression method, assuming a binomial distribution. In these analyses, we apply relative risk models with two thresholds between epilators and nonepilators from a highly characteristic lesion of which radiation cataract does not occur around 2 gray for a single acute exposure. The risk models are fitted to the data assuming 10 as a constant relative biological effectiveness of neutron. The likelihood of observing the entire data set in these models fitted is evaluated by an individual binary-response array. Estimation of a threshold with or without severe epilation and the 100(1-α )% confidence limits are derived from the maximum likelihood approach. The relative risk model with two thresholds can be expressed as a formula with structure of Backgroun × RR, where RR includes threshold models with or without epilation. The radiosensitivity of ionizing radiation to cataracts has been examined for the relationship between epilators and nonepilators.
キーワード radiation cataract threshold model profile approach risk model radiosensitivity
出版物タイトル 岡山大学環境理工学部研究報告
発行日 1999-02-26
4巻
1号
開始ページ 57
終了ページ 75
ISSN 1341-9099
言語 English
論文のバージョン publisher
NAID 120002309188
JaLCDOI 10.18926/fest/19714
フルテキストURL 004_037_055.pdf
著者 崔 承培| 田中 豊|
抄録 One of major problems in spatial analysis is to estimate the value z(s(0)) at an unknown location s(0) using the information about observations z(s(α)), α = 1,…,n. In this article, we will perform a numerical study about some methods for this problem. That is, we examine both the tranditional statistical method which does not take into account spatial correlation and the spatial statistical method which takes into account spatial correlation by applying them to a set of non-stationary spatial data. We compare the predictive powers of these methods. More precisely, we choose Universal Kriging(UK) and Median-Polish Kriging(MPK) as spatial statistical methods, and locally weighted regression or LOESS as a traditional method. As the major criterion for comparison, we use the so-called PRESS statistic, and also draw the prediction surface plot and the prediction standard error surface plot as minor criteria. A real numerical example of non-stantionary spatial data is analyzed for the comparison among the above three methods.
キーワード Stationary Variogram UK MPK LOESS
出版物タイトル 岡山大学環境理工学部研究報告
発行日 1999-02-26
4巻
1号
開始ページ 37
終了ページ 55
ISSN 1341-9099
言語 English
論文のバージョン publisher
NAID 120002309080
JaLCDOI 10.18926/fest/19712
タイトル(別表記) Assessment of Local Influence in Statistical Modelling with Equality Constraints
フルテキストURL 004_019_035.pdf
著者 張 方紅| 森 裕一| 田中 豊|
抄録(別表記) Influence analysis in the sense of Cook (1986)'s local influence is derived for statistical modelling with equality constraints. It is a supplement to Tanaka and Zhang (1998) and its formulation is an extension of that of Kwan and Fung (1998) so that it can deal with the influence on an arbitrary subset of parameters. As a special case the influence in principal component analysis is discussed and a numerical example is shown. The numerical results are compared with those by Zhang et al. (1998), in which local influence is evaluated by using the perturbation theory of eigenvalue problems, and the relationship is discussed with the influence functions derived by Tanaka and Watadani (1992).
キーワード Cook's local influence equality constraints principal component analysis subsets of parameters
出版物タイトル 岡山大学環境理工学部研究報告
発行日 1999-02-26
4巻
1号
開始ページ 19
終了ページ 35
ISSN 1341-9099
言語 Japanese
論文のバージョン publisher
NAID 120002309028
JaLCDOI 10.18926/fest/11611
フルテキストURL 001_065_076.pdf
著者 金 鉉彬| 田中 豊|
抄録 Partial least squares linear discriminant function (PLSD) as well as ordinary linear discriminant function (LDF) are used in pattern recognition analysis of writer identification based on are patterns extracted from the writings written with Hangul letters by 20 Koreans. Also a simulation study is performed using the Monte Carlo method to compare the performances of PLSD and LDF. PLSD showed remarkably better performance than LDF in the Monte Calro study and slightly better performance in the analysis of the real pattern recognition data.
キーワード Writer identification Arc patterns Linear discriminant function Partial least squares
出版物タイトル 岡山大学環境理工学部研究報告
発行日 1996-03
1巻
1号
開始ページ 65
終了ページ 76
ISSN 1341-9099
言語 English
論文のバージョン publisher
NAID 120002314147
JaLCDOI 10.18926/fest/11564
フルテキストURL 005_047_056.pdf
著者 大竹 正徳| 大坪 龍介| 冨田 誠| 田中 豊|
抄録 Main content of this paper is to classify IQ individuals into two categories of normal and abnormal groups. It is too difficult to divide IQ individuals into two groups of normal IQ group and abnormal group because of sparse number of cases with mental retardation. Therefore, we examined a normality of 1673 IQ individuals, but a significant difference was noted for the IQ data. The lowest three mentally retarded cases of less than or equal to 59 IQ score were excluded, the IQ data then fitted to a normal distribution well. The critical value which minimizes the probability of classification is obtained on the basis on an approximate technique with regard to normality. An approximate probability of misclassification for individuals at random from mixture of two normal populations is 25.5%.
キーワード Compound distribution normality Prenatal exposure IQ misclassification
出版物タイトル 岡山大学環境理工学部研究報告
発行日 2000-02-29
5巻
1号
開始ページ 47
終了ページ 56
ISSN 1341-9099
言語 English
論文のバージョン publisher
NAID 120002313674
JaLCDOI 10.18926/fest/11563
フルテキストURL 003_047_062.pdf
著者 大竹 正德| 田中 豊| Lee Sunghee|
抄録 Exposure to ionizing radiation has been testified to have harmful effects on the developing human brain, particularly in the highly vulnerable period of 8-15 weeks after ovulation as an incereased frequency of severe mental retardation. However, the distribution of cases of severe mental retardation suggests a threshold in the low dose region. Data are composed of a binary (1,0) for mentally retarded or normal individuals. The statistical approaches with a threshold are conceivable for a simple, odds, logistic and Gompertz regression models. Estimation of threshold and 100(1-α)% confidence limits are derived from the maximum likelihood technique based on a profile approach.
キーワード mental retardation threshold model prenatal exposure profile approach atomic bombings
出版物タイトル 岡山大学環境理工学部研究報告
発行日 1998-01-14
3巻
1号
開始ページ 47
終了ページ 62
ISSN 1341-9099
言語 English
論文のバージョン publisher
NAID 120002313935
JaLCDOI 10.18926/fest/11562
フルテキストURL 003_037_046.pdf
著者 Kim Hyum-Jeong| 垂水 共之| 田中 豊|
抄録 The present paper deals with multivariate analyses applied to the maximum likelihood estimate(s) for (the mean vector and) the covariance matrix based on incomplete data, and derives influence functions for the mean vector, the covariance matrix and some statistics in multivariate analyses. Influential directions in the sense of Cook's local influence are also derived. A numerical example is given to show the usefulness of the proposed method.
出版物タイトル 岡山大学環境理工学部研究報告
発行日 1998-01-14
3巻
1号
開始ページ 37
終了ページ 46
ISSN 1341-9099
言語 English
論文のバージョン publisher
NAID 120002313576
JaLCDOI 10.18926/fest/11561
フルテキストURL 005_035_046.pdf
著者 崔 承培| 田中 豊|
抄録 Spatial data is analyzed in three stages of 1) estimating the variograms, 2) fitting a model for the estimated variograms and 3) predicting the value at unknown location based on the information at known locations (kriging). Recently, it has become a subject of interest to detect influential observations in these stages. Choi and Tanaka(1999) have derived influence functions in the above three stages and have proposed sensitivity analysis procedure. So far influence functions have only been derived for variograms by Gunst and Hartfield(1996). The present article makes a comparison of the performances between those influence functions for variograms derived by Choi and Tanaka(1999) and by Gunst and Hartfield(1996). A real numerical example is given to discuss the validity or usefulness of those influence functions.
キーワード Stationary spatal data Influence function Sample variogram Median-polish residual
出版物タイトル 岡山大学環境理工学部研究報告
発行日 2000-02-29
5巻
1号
開始ページ 35
終了ページ 46
ISSN 1341-9099
言語 English
論文のバージョン publisher
NAID 120002313332
JaLCDOI 10.18926/fest/11526
フルテキストURL 006_025_034.pdf
著者 山西 芳裕| 田中 豊|
抄録 In functional principal component analysis (PCA), we treat the data that consist of functions not of vectors (Ramsay and Silverman, 1997). It is an attractive methodology, because we often meet the cases where we wish to apply PCA to such data. But, to make this method widely useful, it is desirable to study advantages and disadvantages in actual applications. As alternatives to functional PCA, we may consider multivariate PCA applied to 1) original observation data, 2) sampled functional data with appropriate intervals, and 3) coefficients of basis function expansion. Theoretical and numerical comparison is made among ordinary functional PCA, penalized functional PCA and the above three multivariate PCA.
キーワード Functional data Multivariate data Principal component analysis Eigenvalue Eigenvecotor
出版物タイトル 岡山大学環境理工学部研究報告
発行日 2001-02-28
6巻
1号
開始ページ 25
終了ページ 34
ISSN 1341-9099
言語 English
論文のバージョン publisher
NAID 120002313939
JaLCDOI 10.18926/fest/11521
フルテキストURL 007_031_038.pdf
著者 張 方紅| 田中 豊|
抄録 Relationship has been discussed by Tanaka and Zhang(1999) between the sensitivity analyses based on influence functions and on Cook's local influence, and it has been shown that equivalent results are obtained under general conditions by these analyses in statistical modeling without/with equality constraints. However, a condition implicitly assumed in the proof in Tanaka and Zhang(1999) in the case with equality constraints does not necessarily hold. The present paper gives a complete proof without assuming the condition. Also a formula for the normal curvature is derived for the convenience of practical computation.
キーワード Local influence influence function equality constraints
出版物タイトル 岡山大学環境理工学部研究報告
発行日 2002-03-22
7巻
1号
開始ページ 31
終了ページ 38
ISSN 1341-9099
言語 English
論文のバージョン publisher
NAID 120002313336
JaLCDOI 10.18926/fest/11519
フルテキストURL 007_023_029.pdf
著者 成 祉旻| 梁 祐誠| 田中 豊|
抄録 Analysis of covariance selection models is a useful multivariate method to analyze the covariance structure of a multivariate normal distribution. It is used to reveal cause-and-effect relationships. In the present paper we review the theory and study numerically how the stepwise procedure of covariance selection works in actual data analysis. Then we try to develop a method of influence analysis in covariance selection, and show a numerical example to illustrate the usefulness of the method of influence analysis.
キーワード Covariance Selection Local Influence Influence Function Stepwise Procedure
出版物タイトル 岡山大学環境理工学部研究報告
発行日 2002-03-22
7巻
1号
開始ページ 23
終了ページ 29
ISSN 1341-9099
言語 English
論文のバージョン publisher
NAID 120002314018
JaLCDOI 10.18926/fest/11453
フルテキストURL 009_037_044.pdf
著者 Sung Jimin| 田中 豊|
抄録 There have been proposed so far many methods of statistical diagnostics in Cox regression for checking the goodness of the estimated model or checking the adequacy of the data. The former type contains the checking of the overall goodness of fit, the validity of the assumption of proportional hazards and the proper functional forms of the effects of covariates. While the latter type contains the checking whether there exist singly and/or jointly influential observations in the data set. In the present paper we study numerically the performances of various methods of diagnostics including our method of influence analysis for multiple-case diagnostics (Sung and Tanaka, 2003) by analyzing a real data set of lung cancer patients.
キーワード Cox regression Influence function Local influence Influential Subsets Cox-Snell residuals Martingale residual Deviance residual
出版物タイトル 岡山大学環境理工学部研究報告
発行日 2004-02-27
9巻
1号
開始ページ 37
終了ページ 44
ISSN 1341-9099
言語 English
論文のバージョン publisher
NAID 120002313595