1664It has been common in reaction time (RT) experiments to use the ANOVA to analyze means across covariate conditions. For example, the Stroop task (Stroop, 1935) employs the nominal covariates of a color word and the text color in which it is written, and mean RTs of covariate categories are the objects analyzed (see MacLeod, 1991, for a review). However, when dealing with continuous covariates, such as age, IQ, stimulus intensity, symbolic distance, word frequency, and so on, the ANOVA approach-with its categorized means-tends to ignore the potentially rich relations between covariates and RTs. To remedy this situation, researchers have started to develop alternative, process-based methodologies for RT analysis. For example, Rouder and colleagues (Rouder, Lu, Speckman, Sun, & Jiang, 2005;Rouder, Sun, Speckman, Lu, & Zhou, 2003) have provided a general Bayesian framework to model RT (distributions) as a function of continuous covariates.In the present article, we challenge the nominal-covariate approach for continuous covariates in a within-subjects design from a different perspective. We provide a principled method for integrating continuous covariates into RT modeling. The idea originated from the adaptive methods commonly used in psychophysical research for detection and discrimination. We introduce the notion of "sensitivity" and describe the algorithms to estimate it. In principle, this method can be applied to any continuous covariate. For instance, to study the relationship between word frequency and naming times in the so-called "go/no-go" experimental paradigm, in which subjects are instructed to press a button if the presented string is an English word but to refrain from pressing if the presented string is a nonword (see, e.g., Gomez, Ratcliff, & Perea, 2007), one may use the method to estimate the word frequency needed to elicit a certain RT at a given percentile. 1 In the following section, we will briefly describe a "sensitivity function" in the context of a simple RT experiment and a commonly studied continuous covariate: the signal intensity/contrast. We will also motivate our development from a small-sample perspective, arguing that our method is an appropriate RT design for estimating the sensitivity function efficiently. Simple RTIn a standard simple RT experiment, a trial starts with a warning signal, followed by a foreperiod, at the end of which the signal is presented. The participant is required to react to the presence of the signal as quickly as possible Luce, 1986).Historically, Piéron's law has been one of the most widely cited functional forms specifying the relation between the RT and the signal intensity i (Luce, 1986): ( ) from which one sees that the parameters R, , and K are now functions of the probability . 3 Influenced by the seminal work of Falmagne (1985) on the near miss to Weber's law in psychophysics for detec- Applications of nonparametric adaptive methods for simple reaction time experiments YUNG-FONG HSU AND YEN-HO CHEN National Taiwan University, Taipei, TaiwanA...