A modification of Nakatani's model (Nakatani, 1972,
Journal of Mathematical Psychology, 9, 104-127) is proposed, in which stimulus
Si and response
Ri are represented as a common point in a multidimensional space. When
Si is presented, the distances between
Si and all the responses are perturbed by an identical error process. If the distance between
Si and
Rj is below the response threshold
tj,
Rj is acceptable. The observer chooses a response from among the acceptable responses according to a choice probability distribution. The advantages over the original are a great reduction in computing time, incorporation of the INDSCAL model, and use of the maximum likelihood estimation procedure, but the fit of the proposed model is slightly worse than Nakatani's original model. An application to a letter recognition data revealed that relative importance of the differences in shape of the lower part of letters (C, E, F, G, P, and R) to that of the right or left part of them increased as display times were increased.
View full abstract