A logistic regression model of variant preference in Japanese kanji: an integration of mere exposure effect and generalized matching law

The word hinoki or ‘cypress’ can be transcribed in two variant forms, 檜 (the so-called “traditional” variant) and 桧 (the “simplified” variant), in Japanese kanji. Such variant forms are called kanji variants. The present paper reviews a series of studies on Japanese kanji recognition (Yokoyama, 2006a, 2006b, 2006c), and proposes a model which accounts for performance in a preference judgment task based on kanji frequency data. Yokoyama (2006a) administers preference judgment task in which the participants were presented with 263 pairs of traditional and simplified variants and asked to choose the more preferable variant of each pair. The analyses indicate a positive contribution of frequency to variant preferences, supporting the so-called “mere exposure effect” theory of Zajonc (1968). This finding leads to a logistic regression model that describes preference behavior in kanji recognition, based on Fechner’s law. Yokoyama (2006b) shows that the model is comparable to the so-called “the generalized matching law” of Baum (1974) and to “the ideal free distribution theory” of Fagen (1987). Yokoyama (2006c) further examines the predictive validity of the model with empirical data obtained from a preference judgment task, administered in the Tokyo and Kyoto areas. Logistic regression analyses are performed with the ratio of preference for the given variants and the logit of the character frequencies, yielding significant correlations between the predicted probabilities and the observed responses (r = .804 for Asahi newspaper data). The present paper synthesizes these studies and proposes a logistic regression model that efficiently describes preference behavior in Japanese kanji recognition, integrating the theoretical perspectives of mere exposure effect and the generalized matching law.