Detecting inherent bias in lexical decision experiments with the LD1NN algorithm

A basic assumption of the lexical decision task is that a correct response to a word requires access to a corresponding mental representation of that word. However, systematic patterns of similarities and differences between words and nonwords can lead to an inherent bias for a particular response to a given stimulus. In this paper we introduce LD1NN, a simple algorithm based on one-nearest-neighbor classification that predicts the probability of a word response for each stimulus in an experiment by looking at the word/nonword probabilities of the most similar previously presented stimuli. Then, we apply LD1NN to the task of detecting differences between a set of words and different sets of matched nonwords. Finally, we show that the LD1NN word response probabilities are predictive of response times in three large lexical decision studies and that predicted biases for and against word responses corresponds with respectively faster and slower responses to words in the three studies.