行動計量学
Online ISSN : 1880-4705
Print ISSN : 0385-5481
ISSN-L : 0385-5481
34 巻 , 1 号
選択された号の論文の9件中1~9を表示しています
原著
  • 衛藤 俊寿, 下川 敏雄, 後藤 昌司
    2007 年 34 巻 1 号 p. 1-20
    発行日: 2007年
    公開日: 2007/05/30
    ジャーナル フリー
    In survival analysis, it is one of important subjects to explore influencing factors on prognosis of relevant diseases. The useful tool for exploration of the factors is tree structured method. In this paper, we proposed a multi-split tree structured method. Then, we can present the complicated interactions of those influencing factors on prognosis using graphical display. Then, we evaluated performance of the multi-split tree structured method by using certain literature examples and conducting simulation. As a result, the multi-split tree had the small number of terminal nodes rather than ordinary binary-split tree. When true tree structure had many (rather than two) branches, total mean squared error of estimated hazard of the multi-split tree structured method was less than one of the binary-split tree structured method. That is, the performance of the former was superior to one of the latter. Moreover, we can use the multi-split tree structured method in order to diagnose the result of the binary-split tree structured method.
  • 宮崎 慧, 田村 玄, 繁桝 算男
    2007 年 34 巻 1 号 p. 21-31
    発行日: 2007年
    公開日: 2007/05/30
    ジャーナル フリー
    The purpose of this paper is to propose an extension model of the existing Latent Class Analysis (LCA). This model assumes multi-level latent classes so that the interpretation of the latent classes should be made clearer. The inference about the parameters involved in the model was made numerically using Gibbs Sampler. This inference procedure was applied to real data, and yielded the meaningful results.
  • 星野 崇宏
    2007 年 34 巻 1 号 p. 33-48
    発行日: 2007年
    公開日: 2007/05/30
    ジャーナル フリー
    We showed that the propensity score weighted M-estimation method proposed by Hoshino (2005) can be applied in order to adjust for the bias in sampling surveys for marketing research. Further, using real data sets, we investigated the capabilities and the limitations of the propensity score adjustment method to the adjustment of the biased internet surveys and demonstrated the validity of Hoshino's method. By using the covariates selected the method proposed by Hoshino and Maeda (2006) in two similar surveys, we found that the effectiveness of the adjustment method using the propensity scores was reproducible.
  • 竹谷 誠, 安木 伸明, 富田 聡
    2007 年 34 巻 1 号 p. 49-66
    発行日: 2007年
    公開日: 2007/05/30
    ジャーナル フリー
    The traditional paper test is able to measure mainly individual student's the understanding level of individual bits of knowledge. However, it is difficult to measure the internal connectedness among bits of knowledge. The authors had been presented a new testing method, called a Logical Flow Testing (LFT) for formative evaluation. The teacher may draw a graph, called a Logical Flow Graph (LFG) according to the contents to be taught, and then teach the subject based on the LFG. After lecturing, the teacher may give the LFT with elements which are equal to those of the teacher's LFG. The teacher compares individual students' LFGs with the teacher's LFG, and evaluates the structural understanding levels of individual students. This paper discusses similarity index based on qualitative degree of sequencing. As a result of linear translation ψ of the similarity index, the authors show that the index ψ is an expansion of traditional rank order correlation coefficients, such as the γ coefficient. Lastly, this paper shows that similarity index is effective for LFGs with partially ordered data structure from both theoretical and practical view points.
  • 石井 秀宗, 椎名 久美子, 前田 忠彦, 柳井 晴夫
    2007 年 34 巻 1 号 p. 67-77
    発行日: 2007年
    公開日: 2007/05/30
    ジャーナル フリー
    Differences between attributes of those who discuss decline of ability in pupils and students sometimes cause confusion in the arguments. In this paper, we try to reveal what attribute of faculty members effects on their views for decline of ability in college students. As the result, we find that the difference of faculty and the difference of foundation, that is, national university, public college and private college, have relatively strong effects on the views. We also find that decline of ability can be explained in terms of the following two points of views. One is the basic ability of reasoning thinking, numerical thinking and competence in foreign languages. The other is the learning motivation. Roughly speaking, elder faculty members or the members in private colleges are afraid of decline of the basic ability. And, as the difference among faculties, decline of the basic ability is suspected in the faculty of sociology, gymnastics and domestic science; decline of the learning ability is in science; both are in information science, economics, technology and pharmacy.
  • 田崎 勝也
    2007 年 34 巻 1 号 p. 79-89
    発行日: 2007年
    公開日: 2007/05/30
    ジャーナル フリー
    The purpose of study was to test validity of self-construal scales through an examination of differential item functioning (DIF) using multigroup mean and covariance structure (MG-MACS) analysis. The data analyzed were composed of 368 Japanese and 152 European Americans. A series of MG-MACS analyses was conducted on 6 items, which have displayed two-factor model of self-construal in the previous studies. As a result, no items demonstrated nonuniform DIF, though uniform DIF was detected on two items. Since the number of DIF items was small and partial factorial invariance was established, factor mean comparisons were conducted. Consistent with the theories of self-construal, Japanese participants were significantly higher on interdependent self-construal and were significantly lower on independent self-construal than the American counterparts. Based on the results of DIF analyses and factor mean comparisons, cross-cultural validity of self-construal scales was discussed.
研究ノート
  • 岩崎 学, 廉 民善
    2007 年 34 巻 1 号 p. 91-100
    発行日: 2007年
    公開日: 2007/05/30
    ジャーナル フリー
    An analysis method is considered that concerns count data with zero truncation, i.e. frequencies of occurrence of an event greater than or equal to 1 are only reported and the frequency of zero occurrence is not available. Relationship between the counts of occurrences of two related events is particularly interested. Such relationship is modeled by a bivariate gamma-Poisson distribution in this paper. To estimate the unknown parameters involved in the model, a simple method based on sample means is proposed. The estimation procedure is applied to some real datasets and interpretations of the results obtained are given. An additional point emphasized in this paper is that the frequency of zero count is not always trusted even the number is reported.
資料
  • 豊田 秀樹, 川端 一光, 中村 健太郎, 片平 秀貴
    2007 年 34 巻 1 号 p. 101-110
    発行日: 2007年
    公開日: 2007/05/30
    ジャーナル フリー
    This study proposes the adoption of a neural network as an alternative to logistic regression analysis, which is conventionally used to estimate the propensity score (Rosenbaum & Rubin, 1983). Moreover, covariates that are frequently obscured are presented.
    Considering the response pattern to a mail survey by random sampling as a criterion, we examined how is the response pattern to a Web survey by purposive selection rectified using the propensity score. The propensity score was estimated using the subjects' demographic variables as covariates.
    The results of adopting a neural network were compared with those of the logistic regression analysis. As a result, the accuracy of bias reduction by the threelayer neural networks was found to be greater than that by the logistic regression analysis.
    In addition, detailed contents of the covariates were presented, and a decision tree was produced to examine the influence of covariates on allocation of the subjects to survey forms.
  • 辻谷 将明, 外山 信夫
    2007 年 34 巻 1 号 p. 111-131
    発行日: 2007年
    公開日: 2007/05/30
    ジャーナル フリー
    This article presents an introduction to generalized additive models using R for data of mutually exclusive groups and a set of predictor variables. Illustrated herein are a number of resampling methods, that is cross-validation when selecting the optimum smoothing parameter, and bootstrapping applications that implement the bootstrap-based information when using the deviance in order to summarize the measure of goodness-of-fit on generalized additive models. The cross-validation is also adapted for influential analysis in order to verify the appropriateness of the model and to detect observations that do not agree with the rest of the data.
feedback
Top