Internal redundancy in simple response time tasks

点击翻译按钮获取中文摘要

Many models of simple decision making are based on two major mechanisms, a decision threshold and a rate of accumulation for the perceived information. The decision thresholds are influenced by contextual information such as instructions, response frequencies, etc. whereas rates are influenced by the physical properties of the stimuli such as contrast, signal-to-noise ratio, etc. Highly-cited models include Ratcliff's Diffusion model and Brown and Heathcote's Linear Ballistic Accumulator model. In recent years, these models have incorporated an additional component: variability in the decision threshold across trials. This component was added to account for fast errors, i.e., erroneous responses that are faster than expected. The fact that a new component of a model improves fit is not a sufficient justification for its addition. Variability in threshold should be made observable so that its existence and impact on performance can be assessed directly. In this research project, I first propose a new type of plot, the within-condition latency-accuracy curve. This curve will allow us to observe variability in threshold and its subsequent effect on the speed of correct and erroneous responses. Second, I examine whether variability in thresholds has a role in learning. One possibility is that the thresholds are fluctuating so that the system can find optimal threshold values. Third, I examine the impact of threshold variability on response time distributions. If a decision threshold is fluctuating between trials, the response time distribution is the superposition of many distributions, each with slightly different offsets, resulting in a left tail. Finally, as the Diffusion and the Linear Ballistic Accumulator models cannot accommodate fast errors, high accuracy and modest skew in response time distributions, a new model of simple decision making based on internal redudancy is proposed. As part of the project, experiments with humans explore simple decision making tasks. The lexical decision task is a commonly-used task in which participants must decide as quickly as possible if a stimulus composed of letters is a word or a non-word. I also explore two tasks where the stimulus attributes are manipulated formally. In one task, the attributes must be weighed in conjunction (all the attributes matter for a decision, as in the same-different task, Bamber, 1969); in the other, the attributes must be weighed disjunctively (only one attribute is sufficient for a decision, as in the redundant-attribute target detection task, Miller, 1982). Fast errors have often been treated in the literature as fast guesses caused by an extrinsic mechanism. Here, I assume that fast errors are on a continuum of response speed and threshold variability. This unifying view should help understand the causes and the role of fast errors, better specify models of decision making and have implications for training programmes.