Decision Delay

delayindexThis post is a continuation of the theme of when decisions are made and how we delay or wait or decide not to decide.  This post is based on a 2014 paper by Teichert, Ferrera, and Grinband, “Humans Optimize Decision-Making by Delaying Decision Onset,” in PLoS ONE. Again, this paper is beyond my understanding at least as to the details. It has some excellent figures and graphics that are pretty, but I do not think that I really understand them. These are my shortcomings. What interests me about this is the contrast with my previous post Deciding not to Decide.   This paper examines decision onset and nondecision time while “Deciding not to Decide” suggested an explicit decision to inhibit the decision. I find making an inhibitory decision a more satisfying explanation than delaying decision onset, although they could be the same thing or the situations may be so different that there is no real comparison.

Teichert et al suggest that humans have the remarkable ability to make fast and accurate decisions in a seemingly endless number of different tasks. However, in some situations it is unnecessary or undesirable to initiate a decision. Current research still does not make clear whether decision onset is under cognitive control, or whether decisions are initiated automatically by the presence of an appropriate sensory stimulus. Teichert and colleagues tested whether humans will delay the onset of a decision process to increase response accuracy in a situation when it is beneficial to ignore the initial pulse of sensory evidence that can be dominated by salient rather than task-relevant information. Most decisions are based on a subset of stimuli that appears in close temporal and/or spatial proximity to other, task-irrelevant information. Often, the irrelevant stimuli interfere with the processing of the relevant stimuli especially when the distractors are physically salient and/or are associated with behavioral responses that are similar to those of the current task. Thus, according to Teichert, accurate decision-making critically depends on top-down attentional selection that enhances task-relevant and/or suppresses irrelevant information. The engagement of selective attention takes time, and it has been suggested that errors occur in part because the decision process (evidence accumulation) is initiated before selective attention has isolated task-relevant sensory information from salient but irrelevant distractors. The authors say that they are the first to suggest that subjects might increase accuracy by delaying the onset of the decision process. This possibility is not precluded by the generally accepted finding that subjects trade speed for accuracy by increasing the amount of evidence that will trigger a response (the threshold mechanism), as both mechanisms could be used in parallel.

The current study examined the role of decision onset for optimal decision-making when selective feature-based attention is needed to reduce interference from salient but irrelevant
distractors. The study results suggest that it takes ,120 to 150 ms before the less salient but task-relevant target stimulus exclusively determines the input to the decision process. It also shows that human subjects indeed use the onset mechanism to trade speed for accuracy: they automatically delay decision onset until selective attention can begin to isolate the relevant target when they are required to stress response accuracy over speed. Teichert et al state that on average, the observed 50 ms delay of decision onset alone would have accounted for an estimated 75% of the observed improvement in response accuracy.

The model splits this value into two independent parameters: decision onset and nondecision
time. The authors indicate that they were able to distinguish whether subjects wait before initiating the decision process (delayed decision onset), or whether they wait to execute the response once they have made up their mind (longer non-decision time). The current study provides support for the idea that subjects can strategically adjust decision onset to trade speed for accuracy. Manipulating decision onset provides an independent mechanism to improve accuracy by finding the optimal balance between decision onset and response threshold.