Adapting to a Response Deadline in Categorization

Abstract

Listen

Translate article

Adapting to a Response Deadline in Categorization A. J. Wills (a.j.wills@ex.ac.uk) School of Psychology, University of Exeter, Perry Road, Exeter. EX4 4QG. England Abstract The effect of a response deadline on categorical decisions was investigated. Time available for response was manipulated in the test phase, along with stimulus difficulty. Effects of these manipulations were observed in response accuracy, and in the mean, standard deviation and skew of the reaction times. The effects observed demonstrate that participants responded to the deadline in an adaptive manner - reducing their reaction time to long-latency decisions whilst leaving short latency decisions relatively unaffected. A simple connectionist model of categorical decisions (Wills & McLaren, 1997) is shown to account for this behavior. Introduction Categorization is a basic and essential cognitive function. Our ability to engage it has been well studied, and a number of different theories of the underlying processes have been proposed (e.g. Ashby & Gott, 1988; Gluck, 1991; Nosofsky, 1986; Nosofsky, Palmeri & McKinley, 1994). At first, attempts to quantitatively fit models of categorization to empirical data concentrated on categorization accuracy. However, in recent years, models which have the potential to predict reaction time distributions in categorization have been developed and evaluated (e.g. Ashby, 2000; Maddox, Ashby & Gottlob, 1998; Lamberts, 2000; Nosofsky & Palmeri, 1997; Wills & McLaren, 1997). This paper focuses on the effects of imposing a response deadline on a) participants' response accuracy and b) the nature of their reaction time distributions. It has been known for some time that categorical decisions made under time pressure may be different to those made without time pressure (eg. Smith & Kemler Nelson, 1984). More recently, this avenue of research has been developed by investigation of the effects of time pressure with more complex stimuli (e.g. Lamberts, 1995; Palmeri & Blalock, 2000) coupled with formal modeling of the results found (e.g. Lamberts, 1995). It is worth considering Lambert's (1995) study in a little more detail as it provides one motivation for the current work. At one level, the results found are intuitive. In these experiments, Lamberts employed a simple deadline procedure. Participants first learned, in the absence of time pressure, to categorize artificial stimuli (schematic faces) into two categories. Following this training, participants had to categorize test stimuli before a given deadline (e.g. 1600ms from stimulus onset). Failure to respond in time resulted in an error tone, followed by the presentation of the next stimulus. Participants were informed about the time available for response, which changed at regular intervals. In one experiment, the deadlines employed were 600ms, 1100ms, 1600ms and no deadline. Participants were less accurate at shorter deadlines. Interestingly, the effect was stimulus specific, with some stimuli being considerably more affected by time pressure than others. Lamberts proposed a particular formal model of this effect (the Extended Generalized Context Model or EGCM, Lamberts, 1995) and showed that it provided a good fit to the accuracy data. Time pressure and reaction time Lamberts' experiments reveal another result. In his experiments, categorization in the absence of a response deadline takes approximately 1500ms (Lamberts, 1995, experiment 2). As the stringency of the deadline increases, so the mean reaction times decrease, with categorization under a 600ms deadline taking about 450ms. In other words, categorical decisions appear to take considerably less time when there is time pressure than when there is not. This is, of course, intuitively obvious. The interest, from the perspective of the current paper, is that there seem to be at least three distinct reasons why it might happen. When considering the following, it is important to remember that the descriptions relate to observed reaction time distributions - they are not statements about underlying process: Non-selective adaptation: The participant reacts to the imposition of the deadline in a manner that decreases all reaction times in the distribution by a fixed amount. As a consequence, mean of the distribution will drop, but the standard deviation and skew will be unaffected. Selective, linear adaptation: The participant reacts to the imposition of the deadline in a manner which decreases all reaction times in the distribution by a fixed factor (i.e. RT deadline = f × RT no deadline ). As a result, the mean and standard deviation of the distribution will drop, but the skew will be unaffected. Selective, non-linear adaptation: The participant reacts to the imposition of the deadline in a way that cannot be characterized as non-selective, or selective, non-linear, by the definitions above. Changes in the mean, standard deviation, and skew of the distribution may all be observed.