恒常法における反応カテゴリー数の影響

抄録

Simulation of experiments using methods of constant stimuli demonstrated that from a statistical perspective the three- or four-category (3AFC or 4AFC) methods are preferable to the two-category (2AFC) method. Point estimates of the slope and position parameters, sigma and mu, respectively, of psychometric functions of cumulative normal distributions were calculated using Bayesian analysis of the data generated by simulation experiments using methods of constant stimuli. The root mean square errors (RMSEs) of estimates of sigma for the two-category method were approximately 1.4 times those for the three- or four-category methods. The just noticeable difference is given by a constant time of sigma. RMSEs of estimates of mu corresponding to the point of subjective equality were approximately of the same magnitude in the two- and four-categories methods, but those for the three-category method was about three fourths of those for the two- or four-category methods.