1999 年 13 巻 p. 73-86
This paper reports on the effect on test reliability of converting selected response dichotomous items into polychotomous ones. A 34-item test was administered to 90 Japanese university students who were instructed, in addition to choosing one correct option for each item, to choose one of three confidence level ratings-3 (definitely sure), 2 (not very sure), or 1 (wild guess). 6-point-scale polychotomous data were produced by combining response correctness data and confidence level ratings in such a way that a correct response with higher confidence was scored more highly than one with lower confidence. Item analyses revealed that incorporation of confidence level ratings generally improved the overall item reliability, replicating the results of an earlier study (Shizuka 1999).