The dynamics of cognition and action: Mental processes inferred from speed-accuracy decomposition.

Abstract

Measurements of reaction time have played a major role in developing theories about the menial processes that underlie sensation, perception, memory, cognition, and action. The interpretation of reaction time data requires strong assumptions about how subjects trade accuracy for speed of performance and about whether there is a discrete or continuous transmission of information from one component process to the next. Conventional reaction time and speed-accuracy trade-off procedures are not, by themselves, sufficiently powerful to test these assumptions. However, the deficiency can be remedied in part through a new speed-accuracy decomposition technique. To apply the technique, one uses a hybrid mixture of (a) conventional reaction time trials in which subjects must process a given test stimulus with high accuracy and (b) peremptory response-signal trials in which subjects must make prompted guesses before stimulus processing has been finished. Data from this "titrated reaction time procedure" are then analyzed in terms of a parallel sophisticated-guessing model, under which normal mental processes and guessing processes are assumed to race against each other in producing overt responses. With the model, one may estimate the amount of partial information that subjects have accumulated about a test stimulus at each intermediate moment during a reaction time trial. Such estimates provide deeper insights into the rate at which partial information is accumulated over time and into discrete versus continuous modes of information processing. An application of speed-accuracy decomposition to studies of word recognition illustrates the potential power of the technique. People do not think or act instantaneously. The time required to take action depends systematically on mental and physical processes that precede an overt response. Thus, throughout many areas of psychology, conclusions about the nature of mind and body have been based on measurements of human reaction time.' Past uses of reaction time data extend from studies of elementary sensory mechanisms (e.g., Green & Luce, 1973) to studies of perception (e.g.

Effects of pallidal deep brain stimulation and levodopa treatment on reaction-time performance in Parkinson’s disease

“…If DBS affected the arousal level of patients, then it should primarily influence the slower (right tail) part of the RT distribution, because slow RTs are most severely affected by arousal fluctuations, as has been shown by different authors (Sanders 1983;Miller 1988). On the other hand, if DBS affected one processing stage, i.e., the motor stage, then it should have affected all parts of the RT distributions, i.e., the fast, middle, and slow RTs (Ratcliff 1979;Meyer et al 1988;Schubert 1999). Figure 2 shows that DBS led to a leftward shift of all RTs in all three tasks.…”

“…According to the arousal assumption, DBS should have affected the slowest RTs, because these are most severely affected by an unstable arousal level (Sanders 1983;Miller 1988). In contrast, when affecting a particular processing stage, DBS should influence the RTs in all trials irrespective of their speed (Ratcliff 1979;Meyer et al 1988). The results of our RT distribution analysis primarily indicated an influence on one particular processing stage, i.e., the motor stage, although an additional influence of DBS on the general arousal level cannot be ruled out.…”

A dynamic approach to recognition memory.

“…By varying the lag between stimulus onset and the response signal, it is possible to estimate a speed-accuracy trade-off (SAT) curve which measures how the probability of responding "old" to a test item changes as additional time is allowed to process the stimulus. Although it is possible to make more complex assumptions about the cognitive processes that underlie SAT curves, for instance that they reflect a mixture of ongoing and already-finished processes (Meyer, Irwin, Osman, & Kounios, 1988;Ratcliff, 1988), we adopt a much simpler assumption. As just described, we assume that features of the test item begin to be sampled to join the context features already present in the probe of memory, starting at time t 0 after stimulus onset.…”

A Diffusion Model Account of the Lexical Decision Task.

“…The model can fit all aspects of the data-the probabilities of and the RTs for both correct and error responses and the shapes of RT distributions (and their hazard functions). It accurately accounts for the patterns of data that result from manipulations of speed versus accuracy and from manipulations that curtail decision processes, such as deadline, response signal, and speed-accuracy decomposition procedures (e.g., Meyer, Irwin, Osman, & Kounios, 1988). Diffusion models have also been applied in the domains of simple RT (Smith, 1995) and decision making (Busemeyer & Townsend, 1992Diederich, 1997;Roe, Busemeyer, & Townsend, 2001), and diffusion models are close cousins of random walk models (Laming, 1968;Link, 1975;Link, 1992;Link & Heath, 1975;Smith, 1990;M.…”