Develpooing data analytic techniques for the entrance examinations utilizing mgroup regression analysis and decision theoretic apporach.
Project/Area Number  05680189 
Research Category 
GrantinAid for Scientific Research (C).

Research Field 
Educational technology

Research Institution  The National Center for University Entrance Examinations 
Principal Investigator 
MAEKAWA Shinichi The National Center for University Entrance Examinations. Research Division, Associate Professor, 研究開発部, 助教授 (70190288)

CoInvestigator(Kenkyūbuntansha) 
山田 文康 大学入試センター, 研究開発部, 教授 (40158217)
繁桝 算男 東京工業大学, 工学部, 教授 (90091701)
SHIGEMASU Kazuo The Tokyo Institute of Technology Department of Engeneering Professor
YAMADA Fumiyasu The National Center for University Entrance Examinations.Research Division, Asso

Project Fiscal Year 
1993 – 1994

Project Status 
Completed(Fiscal Year 1994)

Budget Amount *help 
¥2,000,000 (Direct Cost : ¥2,000,000)
Fiscal Year 1994 : ¥400,000 (Direct Cost : ¥400,000)
Fiscal Year 1993 : ¥1,600,000 (Direct Cost : ¥1,600,000)

Keywords  mgroup regression / neural network regression / utility scaling / 階層的回帰分析 / m群回帰分析 / 効用の測定 / 経験的ベイズ推定法 
Research Abstract 
Mgroup regression analysis technique so far developed has two major problems. The first one is that the model always assumes the exchangeability of all the regression coefficients included in the model. In this research, we developped a general method which can handle a mixture of 1) common regression coefficients across groups, 2) exchangeable regression coefficients across groups, and 3) nonexchangeable regression coefficients. A computer program based on this model was developped. The second problem is that the usual ingroup regression analysis is restricted to the linear model. In this research, we investigated the possibility of extending the original idea to nonlinear models including spline model and neural network model. As the first step, an efficient algorithm for neural netrowk regression was developped. We found that developping the mgroup version of the newral network regression models seems promissing. As for measuring utility functions, a computer program was developped which uses the maximum likelihood estimation technique based on the assumption that the indifference probabilities have beta distribution.

Report
(4results)
Research Output
(3results)