Kodo Keiryogaku (The Japanese Journal of Behaviormetrics)
Online ISSN : 1880-4705
Print ISSN : 0385-5481
ISSN-L : 0385-5481
Volume 41, Issue 1
Displaying 1-6 of 6 articles from this issue
Remembrance
Articles
  • Takahiro Tsuchiya, Nagatomo Nakamura
    Article type: Articles
    2014 Volume 41 Issue 1 Pages 3-16
    Published: 2014
    Released on J-STAGE: March 10, 2015
    JOURNAL FREE ACCESS
    This paper discusses a linear regression model for the aggregate means which result from some partitions of the independent variable. These are called stratified data. The data at hand indicate the aggregate means of independent variable and the number of data in intervals, as well as the means and the variances (or standard deviations) of the dependent variable calculated by strata or intervals. The regression coefficients estimated by this data are approximately equal to the regression coefficients obtained from the data prior to aggregation. However, the coefficient of determination is clearly different. The differences of these regression coefficients are verified, and a method for estimating the original coefficient of determination is proposed by using the aggregate means and the variance of each stratum. The efficiency of the proposed method is illustrated through some numerical experiments.
    Download PDF (992K)
  • —— For the Purpose of Building Item Bank for a Common Achievement Test for Nursing Colleges——
    Haruhiko Mitsunaga, Haruo Yanai, Hiroaki Nishikawa, Kei'ichiro Saiki, ...
    Article type: Articles
    2014 Volume 41 Issue 1 Pages 17-34
    Published: 2014
    Released on J-STAGE: March 10, 2015
    JOURNAL FREE ACCESS
    Background: As Japanese nursing colleges increasingly require common criteria for assessing practical nursing ability prior to entering clinical hospital practice, it is important to construct a test item bank that can facilitate the evaluation of multiple domains. However, ordinal IRT models, such as the 2 parameter logistic model (2PL), operate under the assumption of unidimensionality, preventing application to comprehensive testing of multiple domains. Method: We conducted a computer-based test with items from 20 domains, classified into three areas: (1) basic medicine, (2) basic nursing, and (3) clinical nursing. About 780 students answered items, which were applied to common-item design and calibrated item parameters using two strategies; the first strategy assumed one-factor model for each area, the second strategy assumed unidimensionality by domain. Conclusion: For constructing an item bank, estimating item parameters by domain results in larger test information and more appropriate parameter estimates than estimating parameters by area.
    Download PDF (1115K)
  • Mingzhe Jin
    Article type: Articles
    2014 Volume 41 Issue 1 Pages 35-46
    Published: 2014
    Released on J-STAGE: March 10, 2015
    JOURNAL FREE ACCESS
    Text classification results often vary depending on the detailed factors in data analysis, including feature data, classification method, and parameter sets adopted in the analysis. The author of an anonymous text can be generally identified by extracting a set of distinctive features of the text, and then using the features to find the most likely author. Numerous efforts have been made to develop the feature extraction technique with more robustness and the classification algorithm, but an important issue is how to select the features datasets and classification method. To address this issue, we propose an integrated classification algorithm that extracts multiple feature datasets from differing viewpoints and aspects of a text and applies multiple strong classifiers to the datasets. Our proposed method achieved 100% accuracy in identifying the authors of literary works and student essays, and identified the author of all but 1 out of 60 diaries which were written by 6 different people.Our proposed method achieved equivalent or better accuracy than the case when any a strong classifier applied to individual feature dataset. Furthermore, the accuracy in identifying the authors of student essays increased by roughly two percentage points.
    Download PDF (890K)
Note
  • ——Experimental survey research for structural analyses on Japanese public opinion——
    Insook Choi
    Article type: Note
    2014 Volume 41 Issue 1 Pages 47-62
    Published: 2014
    Released on J-STAGE: March 10, 2015
    JOURNAL FREE ACCESS
    This study has conducted an experimental survey on Japanese Nuclear Power Generation by using a dynamic method that makes it possible to exchange political opinion in order to accurately measure public opinion and gain more exact results. The process of the survey is following: First, interviewees are asked if they are in favor of or against nuclear power. Second, they are provided with “full arguments” and “empty arguments” as counter arguments. After this, they are asked their opinion again. To this end, we have had anticipation as to how their first opinion, their interest in politics and nuclear power, their knowledge of politics and their information perception of nuclear power, are influenced when confronted with a counter argument. The results of our analysis have showed that the interviewees with higher interest, higher knowledge of politics and nuclear power, and higher information perception find it harder to change their opinion. But, there is no significant effect on the interviewees of each side after presenting the “full argument” and the “empty argument”. This means that Japanese public opinion on nuclear power is pretty consistent.
    Download PDF (792K)
Miscellanea
feedback
Top