Trends in Cognitive Sciences
Feature ReviewDiffusion Decision Model: Current Issues and History
Section snippets
Modeling Simple Decision Making
Decision making is intimately involved in all of our everyday activities. Many decisions are made rapidly and at a low level cognitively, for example, deciding whether to drive left or right round a car in front. Others, such as deciding which candidate to vote for or which car to buy, are made at a higher level with prolonged deliberation. The diffusion models we discuss are of the former type. In the real world, they involve a rapid matching of a perceptual representation to stored knowledge
The Two-Choice Diffusion Model
Figure 2 shows simulated paths that represent the accumulation of evidence on individual trials (Figure 2A) and it shows the effects of changes in drift rate (Figure 2B) and boundary settings (Figure 2C) on RT distributions. Because there is a minimum on RTs but no maximum, the model automatically produces right-skewed distributions that have the same shape as those found in most simple two-choice tasks. That the model predicts RT distributions that are the same as those found experimentally is
Across-Trial Variability in Model Components
A problem with early random walk models, which were discrete time precursors of diffusion models, was that they predicted identical RT distributions for correct and incorrect responses (when the starting point is equidistant from the boundaries), which is never observed empirically. Several different approaches to this problem have been investigated, including: dynamically changing decision boundaries; nonlinear evidence accumulation processes; and non-normal random walk increments. The most
Response Signal and Go/No-Go Tasks: Implicit Boundaries
In the response signal task, a stimulus is presented and then after some amount of time, a signal is given. Participants are asked to choose between two alternatives just as in the usual two-choice procedure except that they are asked to respond as quickly as possible after the signal (in, say, 200–300 ms). The stimulus-to-signal time varies from trial to trial 47, 48, 49, 50, which means that processing can be assumed to be the same for all signal times up to each specific signal time [2]. A
Conflict Tasks
Another class of paradigms that appear to require dynamic changes in diffusion model parameters over time are conflict paradigms. In a reinforcement learning paradigm, subjects had to choose one of a pair of letters and feedback indicated which one was ‘correct’ on that trial. Feedback was probabilistic with one letter of the pair being reinforced more often than the other. Later in the session, conflict conditions were created by pairing the letters from different pairs that had low
Optimality
Considerations of optimality have played a significant role in the theoretical and experimental analysis of human and animal decision making. Theories of optimality prescribe how the available evidence should be used to produce a best decision, in some specified sense; experimental studies of optimality have investigated whether actual behavior approximates the theoretical ideal. In simple decision making, two different senses of optimality have been promoted (Box 6). One of these is based on
Collapsing Bounds
The link with optimality theory, on the one hand, and neural studies of decision making, on the other, has led to models in which decision bounds collapse over time. In the collapsing bound model, less evidence is required to trigger a decision as time passes, that is, the boundaries collapse from initially wide spacing toward the center (Figure 4B). Another assumption with much the same effect is that fixed boundaries are maintained, but an ‘urgency signal’ is added to the accumulated evidence
Expanded Judgment Tasks
Most recent applications of diffusion models have been to experimental tasks in which a single stimulus is presented and the noise in the evidence accumulation process arises from moment-by-moment variability in the cognitive representation of the stimulus. However, some of the earliest applications of random walk models 8, 22 were to expanded judgment tasks in which a noisy sequence of stimulus elements has to be integrated to make a decision. Studying such tasks was motivated by the Wald SPRT
Brief Stimulus Presentation
There is a growing interest in whether there is nonstationarity in processing that reflects changing stimulus information over time. This is often studied with stimuli flashed for very brief times (e.g., 50 ms) and is also partly motivated by the thought that stimulus quality can change during the time course of an expanded judgment task. In a highly relevant example from about 15 years ago, the question was whether drift rate tracks the stimulus or not. In other words, whether drift rate begins
Concluding Remarks
Current research in modeling decision processes has used diffusion models extensively. They are being applied in clinical and educational domains, economic decision making (Box 7), and the neuroscience of decision making. Here we have examined current issues that have a history in psychology and we have discussed the earlier research and how it complements new research. In some cases, the earlier research provides an answer to new research questions. Although we have separated the issues, many
Acknowledgments
We would like to thank Jerome Busemeyer, Josh Gold, and Tim Hanks for comments on the paper. Preparation of this article was supported by National Institute on Aging grant R01-AG041176, Department of Education/Institute of Educational Sciences grant R305A120189, and Australian Research Council grants DP140102970 and FT120100244.
Glossary
- Accumulator
- an assumed structure in an evidence accumulation model that has the purpose of gathering evidence in favor of one response.
- Across-trial variability
- the assumption that drift rates vary from decision to decision, motivated by the idea that, even if physical stimulus conditions are identical, the internal representation of the decision-relevant information is not.
- Attractor model
- a network (graph-based) model of interconnected nodes with a dynamic updating process. The updating process
References (210)
The relative judgement theory of two choice response time
J. Math. Psychol.
(1975)A diffusion model analysis of the effects of aging on recognition memory
J. Mem. Lang.
(2004)Modeling response signal and response time data
Cogn. Psychol.
(2006)Evaluating the unequal-variability and dual-process explanations of zROC slopes with response time data and the diffusion model
Cogn. Psychol.
(2012)Optimal strategies for seeking information: models for statistics, choice reaction times, and human information processing
J. Math. Psychol.
(1965)Prior probability and feature predictability interactively bias perceptual decisions
Neuropsychologia
(2014)Stochastic models for choice reaction time with applications to experimental results
J. Math. Psychol.
(1965)Subjective probability in choice-reaction experiments
J. Math. Psychol.
(1969)- et al.
The simplest complete model of choice response time: linear ballistic accumulation
Cogn. Psychol.
(2008) - et al.
Decision making as a window on cognition
Neuron
(2013)