2024-08-10T14:31:15Z
https://nagoya.repo.nii.ac.jp/oai
oai:nagoya.repo.nii.ac.jp:00002145
2023-11-09T02:16:33Z
323:350:373:391
ON THE BASIS FOR THE PSYCHOPHYSICAL SCALING
<原著>心理物理的尺度構成の基礎について
村上, 隆
5601
MURAKAMI, Takashi
5602
1980-12-15
This study is an attempt to establish a logically sound and practically useful scaling model which gives the numerical descriptions of unidimensional perceptual attributes, such as brightness, loudness or heaviness. Traditonally, two major classes of procedures are distinguished in psychophysical scaling methods : direct methods such as magnitude estimation or category ratings, and indirect methods such as Fechnerian scaling or paired comparison method processed according to Thurstone's law of comparative judgment. Both classes have some shortcomings respectively. Direct methods are liable to be distorted by many bias sources inherent in the number assigning behavior of subjects. Indirect methods presuppose the untestable assumptions to determine the scale values, such as discriminal process. We sought more valid scaling model which is free from the defects of traditional methods. The standpoint of this study, which is a compromise between direct and indirect methods, is that a human subject can indicate not only which of two sensory impressons appears greater (brighter, louder or behavior), but also which of two sensory intervals greater. More concretely, it is assumed that a subject can respond which of pairs of stimuli is more dissimilar with respect to the attribute being scaled when a pair of pairs of stimuli (s, t) and (u, w) is presented, but is not assumed that he can evaluate the size of dissimilarities numerically. In our framework, therefore, the most recommendable scaling method is based on the ordering of sensory intervals. From the theoretical viewpoint, a psychological attribute should be measured fundamentally at first. "Fundamental" means not to involve the previous numerical representations, and a psychophysical scale as a fundamental measurement should be established through the qualitative, not quantitative, operations. The method based on the ordering of sensation intervals fulfills the requirement. Practically, ordinal judgment of two intervals is relatively easy and is expected to be free from contextual effects of judgmental bias. Moreover, it can be applied to series of stimuli spacing supraliminally and requires no untestable assumptions. Since orderings or intervals define a system of linear inequalities, we can easily solve the scale. values through ORDMET (McClleland & Coombs) or linear programming algorithm. Using the linear programming, it is demonstrated that these procedures based on merely ordinal information produce the approximate interval scales when relatively large number of stimuli are used. This method of direct comparison of intervals have two weakness. One is that the number of trials is liable to be too large, the other is that the judgmental error causes system of linear inequalities to be inconsistent in most empirical situations. To settle these problems, an interactive method using the high speed computer is proposed.
国立情報学研究所で電子化したコンテンツを使用している。
departmental bulletin paper
名古屋大学教育学部
1980-12-15
名古屋大學教育學部紀要. 教育心理学科
27
1
15
http://hdl.handle.net/2237/3568
03874796
jpn