Automatic vigilance for negative words is categorical and general.

With other factors controlled, negative words elicit slower lexical decisions and naming than positive words (Estes & Adelman, 2008). Moreover, this marked difference in responding to negative words and to positive words (i.e., between-category discontinuity) was accompanied by relatively uniform responding among negative words (i.e., within-category equivalence), thus suggesting a categorical model of automatic vigilance. Larsen, Mercer, Balota, and Strube (this issue) corroborated our observation that valence predicts lexical decision and word naming latencies. However, on the basis of an interaction between linear arousal and linear valence, they claim that automatic vigilance does not occur among arousing stimuli and they purport to reject the categorical model. Here we show that (a) this interaction is logically irrelevant to whether automatic vigilance is categorical; (b) the linear interaction is statistically consistent with the categorical model; (c) the interaction is not observed within the categorical model; and (d) despite having 5 fewer parameters, the categorical model predicts word recognition times as well as the interaction model. Thus, automatic vigilance is categorical and generalizes across levels of arousal.