Stop vs. Go: Re-Examining the Race Horse Model of Inhibition

Although "executive function" may seem like an elusive topic for study, in cognitive neuroscience it is largely approached simply as the ability to control one's own behavior in accord with some goal, despite interference from previous experiences. Central to many accounts of executive function is the ability to "cancel" or "inhibit" actions that are not currently appropriate - this is the same capacity that is said by many to deteriorate with age, with alcohol, with the onset of ADHD, and to be under-developed in young children. Despite the utility of "inhibition" as an explanatory construct, there are reasons to suspect this construct is faulty.

To reiterate this Monday's analysis, "inhibition" does not seem to be a unitary construct, even when the same paradigm is administered via different sensory modalities. For example, the ability to "cancel" eye movements seems mechanistically different than the ability to "cancel" hand movements.

Worse yet, Ozyurt, Colonious, and Arndt's 2003 Perception & Psychophysics article demonstrates that the assumptions underlying the influential "Stop Signal" test of inhibition seem problematic, at least in oculomotor tasks. Specifically, this task involves the measurement of how long it takes to cancel a planned movement.

Of course, it's difficult to measure the reaction time of cancelled movements. To get around this problem, researchers find the delay after which a stop signal can be only 50% effective (i.e., subjects will only correctly refrain from responding 50% of the time). This delay is subtracted from the average reaction time on trials without such "stop" signals. Theoretically, this gives the amount of time for processing the stop signal and cancelling the planned movement; but this rests on the assumption that there is a "go" process that is independent of the "stopping" process (known as the race model).

Critically, this calculation also assumes that the "stop" process is faster than the "go" process. Unfortunately, this is not always the case. The authors review previous work demonstrating that the calculated stopping time was larger than the "go" time at extremely short delays between the "go" and "stop" signals.

Just as critically, this calculation predicts that trials in which a response is mistakenly given (despite a stop signal) should be faster than the average reaction time on "Go" trials. (Ozyurt et al go on to show that this assumption is also violated in their current data.)

Why is it so important for the calculation of stop signal reaction time (SSRT) that there are independent "stop" and "go" processes? First, and most clearly, if there are not independent processes it makes no sense to talk about the efficiency of an inhibitory process per se - planned movements might not be cancelled so much as not fully planned in the first place. Secondly, assuming there is a distinct "stopping" processes, if it somehow affects an ongoing "go" process in a way other than cancelling it - for example, by slowing down "go" processing or movement planning - then the actual reaction time on stopping trials would have been much longer, which leads to an overestimate of the efficiency of a stopping process.

To test these assumptions of the race model, Ozyurt et al. had subjects complete a simple visual reaction time task - they just had to look towards a visual stimulus, presented hundreds of times either to the left or to the right. This allowed for a calculation of their average "go" time.

In a second task, the subjects were told to continue as before, except that an auditory "stop" signal would appear unpredictably throughout the rest of the trials. When that signal was given, they should refrain from looking at the stimulus. Because many subjects will strategically slow their "go" responses in this paradigm to increase the probability they will be able to successfully cancel a movement, the researchers provided occasional feedback to ensure that subjects were responding at the same speed they had on the previous visual task.

For each subject, the stop signal reaction time increased with decreasing delays between the "go" and "stop" signal. This demonstrates that processing of the stop signal may slow the processing of the go signal, or vice versa, in a way similar to the psychological refractory period (in which the second of two rapidly presented stimuli is processed more slowly).

It remains to be seen whether the assumptions of independence between stop and go processes are similarly violated in manual tasks, given that previous research shows oculomotor and manual inhibition tasks differ in certain ways.

In conclusion, Ozyurt et al. show that assumptions underlying the "race model" do not hold at short delays between go and stop signals (perhaps due to a psychological refractory period). Secondly, the data presented by Ozyurt et al contradicts predictions motivated by the race horse model: failed stop reaction time can be larger than the go reaction time.