fMRI evidence for a dual process account of the speed-accuracy tradeoff in decision-making

Abstract

Background: The speed and accuracy of decision-making have a well-known trading relationship: hasty decisions are more prone to errors while careful, accurate judgments take more time. Despite the pervasiveness of this speed-accuracy trade-off (SAT) in decision-making, its neural basis is still unknown.

Methodology/principal findings: Using functional magnetic resonance imaging (fMRI) we show that emphasizing the speed of a perceptual decision at the expense of its accuracy lowers the amount of evidence-related activity in lateral prefrontal cortex. Moreover, this speed-accuracy difference in lateral prefrontal cortex activity correlates with the speed-accuracy difference in the decision criterion metric of signal detection theory. We also show that the same instructions increase baseline activity in a dorso-medial cortical area involved in the internal generation of actions.

Conclusions/significance: These findings suggest that the SAT is neurally implemented by modulating not only the amount of externally-derived sensory evidence used to make a decision, but also the internal urge to make a response. We propose that these processes combine to control the temporal dynamics of the speed-accuracy trade-off in decision-making.

Figure 1. An illustration of a simple accumulator model of the speed-accuracy tradeoff (SAT).

According to this type of model, task-relevant evidence (e.g. motion coherence) accumulates over time from a starting point until a threshold for decision is reached (horizontal grey bars). The vertical stippled lines mark the time, along the x-axis, when a response would be made. (a) In a flexible threshold account of the SAT, emphasizing the speed (SPD) of responding lowers the threshold for decision-making relative to emphasizing accuracy (ACC) of responding, thus reducing the amount of accumulated evidence (and time) prior to the response. (b) In a flexible baseline account of the SAT, emphasizing the speed of responding would increase the baseline level of activity towards a decision threshold, thereby reducing the amount of evidence, and hence the time, that is required to reach that threshold. These two models are not necessarily mutually exclusive. (c) If the accumulation functions are aligned to the time that the decision is made, and normalized to onset of the signal, both the flexible threshold and flexible baseline accounts predict less accumulated signal-based evidence at the time of the decision when response speed is stressed.

Figure 2. (a) Trial sequence.

A cue (SPD or ACC) instructed participants to heed the speed or accuracy of their decisions at the onset of a block of seven trials. A red-framed display containing randomly moving dots appeared 3 s later. Trial onset was signaled by the frame changing from red to green. Once a response was made, either within the 19.5 s trial duration or at the end of the trial, the frame returned to red until the next trial. (b) Trial types. ′Coherence′ trials consisted of a 1%/s rise in motion coherence that began either 0 s, 4.5 s, or 9 s following trial onset. ′Baseline′ trials contained no motion coherence (0%) throughout the trial. A third trial type included a 2%/s rise in coherence that began at trial onset.

Figure 3. Sensitivity to detect coherent motion (d') versus response time (s) as a function of Speed and Accuracy instructions.

Figure 4. fMRI results for MT+.

(a) SPM of right MT+ (white arrow) in localizer experiment. (b) Speed and accuracy time courses for the coherence trials, time-locked and normalized to the onset of stimulus coherence. (c) Speed and accuracy time courses for the coherence trials, time-locked to the onset of the response (R) and normalized to the onset of stimulus coherence. (d) Speed and accuracy time courses for the baseline trials, time-locked and normalized to the onset of the trial.

Figure 5. Activation differences between accuracy and speed conditions at response time (i.e., the average of the pre-response and response volumes) for sensory, motor, and premotor ROIS (see Table 1 for complete ROI list).

(a) ′ Coherence ′ and (b) ′ Baseline ′ conditions. Error bars reflect the SEM of the difference. *: p<0.05. Although the right ventral inferior parietal lobe (vIPL) and postcentral gyrus ROIs in (a) and the right anterior insular ROI in (b) showed significant activation differences (accuracy vs. speed), these brain regions did not demonstrate significant activity increase above baseline in both accuracy and speed conditions (see Tables S1 and S2).

Figure 6. fMRI results for right primary motor cortex (M1).

(a) Isolation of M1 (left) with the left-right manual response contrast of the event-related experiment. White arrow indicates Right M1. (b) Speed and accuracy time courses for the coherence trials, time-locked to the onset of the response (R) and normalized to the onset of stimulus coherence. (c) Speed and accuracy time courses for the baseline trials, time-locked and normalized to the onset of the trial.

Figure 7. fMRI results for pLPFC (left column) and the average of left and right pre-SMA (right column): (a) Location of right pLPFC (white arrow) and pre-SMA (white arrow) in the localizer task.

The anterior and posterior horizontal bars in the pre-SMA figure indicate the coronal planes of the anterior and posterior commissures. (b) Speed and accuracy time courses for the coherence trials, time-locked to the onset of the response (R) and normalized to the onset of stimulus coherence. (c) Speed and accuracy time courses for the baseline trials, time-locked and normalized to the onset of the trial.

Figure 8. fMRI results for right pLPFC (left column) and average pre-SMA (right column).

(a) Speed and accuracy time courses for the coherence+baseline trials, time-locked to the onset of the response, and normalized to the % signal change in the baseline trials. (b) Linear regression plots between the accuracy-speed differences in activity in the coherence and baseline trials. (c) Linear regression plots between the accuracy-speed difference in the decision criterion (c) and the accuracy-speed difference in activity for the coherence trials.