Balancing cognitive demands: control adjustments in the stop-signal paradigm.

Abstract

Cognitive control enables flexible interaction with a dynamic environment. In 2 experiments, the authors investigated control adjustments in the stop-signal paradigm, a procedure that requires balancing speed (going) and caution (stopping) in a dual-task environment. Focusing on the slowing of go reaction times after stop signals, the authors tested 5 competing hypotheses for post-stop-signal adjustments: goal priority, error detection, conflict monitoring, surprise, and memory. Reaction times increased after both successful and failed inhibition, consistent with the goal priority hypothesis and inconsistent with the error detection and conflict hypotheses. Post-stop-signal slowing was greater if the go task stimulus repeated on consecutive trials, suggesting a contribution of memory. We also found evidence for slowing based on more than the immediately preceding stop signal. Post-stop-signal slowing was greater when stop signals occurred more frequently (Experiment 1), inconsistent with the surprise hypothesis, and when inhibition failed more frequently (Experiment 2). This suggests that more global manipulations encompassing many trials affect post-stop-signal adjustments.

Full text

Page 1

Balancing Cognitive Demands:
Control Adjustments in the Stop-Signal Paradigm
Patrick G. Bissett and Gordon D. Logan
Vanderbilt University
Cognitive control enables flexible interaction with a dynamic environment. In 2 experiments, the authors
investigated control adjustments in the stop-signal paradigm, a procedure that requires balancing speed
(going) and caution (stopping) in a dual-task environment. Focusing on the slowing of go reaction times
after stop signals, the authors tested 5 competing hypotheses for post-stop-signal adjustments: goal
priority, error detection, conflict monitoring, surprise, and memory. Reaction times increased after both
successful and failed inhibition, consistent with the goal priority hypothesis and inconsistent with the
error detection and conflict hypotheses. Post-stop-signal slowing was greater if the go task stimulus
repeated on consecutive trials, suggesting a contribution of memory. We also found evidence for slowing
based on more than the immediately preceding stop signal. Post-stop-signal slowing was greater when
stop signals occurred more frequently (Experiment 1), inconsistent with the surprise hypothesis, and
when inhibition failed more frequently (Experiment 2). This suggests that more global manipulations
encompassing many trials affect post-stop-signal adjustments.
Keywords: post-stop-signal slowing, proactive control, cognitive control, inhibition, stop-signal paradigm
Cognitive control, the ability to adapt mental processes to the
demands of the task environment, is of central importance to
goal-directed behavior (Logan, 1985; Miyake et al., 2000). Cog-
nitive control often involves resolving competing demands. Many
instances of control involve achieving a balance between going
and stopping, speed and caution. For example, driving involves
balancing when to accelerate and when to brake. In two experi-
ments, we investigated the adjustments subjects made to balance
speed and caution in the stop-signal paradigm (Logan & Cowan,
1984), which directly pits going against stopping. Previous re-
search has shown that reaction time (RT) increases after both
successful and failed inhibition (Rieger & Gauggel, 1999; Ver-
bruggen, Logan, Liefooghe, & Vandierendonck, 2008). The first
motivation for this research was to evaluate five competing hy-
potheses for these post-stop-signal adjustments: goal priority, error
detection, response conflict, surprise, and memory. There is also
evidence that subjects make adjustments proactively, increasing
RT in anticipation of the need to stop (Logan, 1981; Logan &
Burkell, 1986; Verbruggen & Logan, 2009b). The second motiva-
tion for this research was to investigate whether proactive in-
creases in RT can be explained by the accumulation of post-stop-
signal slowing.
Stop-Signal Paradigm
The stop-signal paradigm is a common procedure for investi-
gating response inhibition (Logan, 1994; Logan & Cowan, 1984;
Verbruggen & Logan, 2008c). In the stop-signal paradigm, sub-
jects typically perform a choice RT task (the “go” task), and they
are asked to withhold their response when a stop signal occurs on
a random subset of trials. The delay between the presentation of
the go stimulus and the stop signal (stop-signal delay, or SSD) is
varied to manipulate the probability of inhibition. When SSD is
short, subjects usually inhibit their responses. As SSD increases,
the probability of inhibition decreases. These findings have been
explained with the “horse race” model (Logan & Cowan, 1984),
which assumes that a go process, initiated by go stimulus onset,
races against a stop process, initiated by stop-signal onset. If the go
process finishes before the stop process, subjects fail to inhibit,
producing a signal-respond trial. If the stop process finishes before
the go process, subjects succeed at inhibiting, producing a signal-
inhibit trial. Stop-signal delay biases the race between stop and go
processes: Short SSDs bias the race in favor of stopping and
thereby increase the probability of inhibiting; long SSDs bias the
race in favor of going and thereby increase the probability of
responding.
Post-Stop-Signal Slowing
Subjects can also bias the race between stopping and going by
strategically speeding or slowing go RT. Speeding go RT de-
creases the probability of inhibition, whereas slowing go RT
increases it. These adjustments can be triggered by local events in
the experiment, such as the presentation of a stop signal or a cue
indicating that stop signals are likely, or by global factors, such as
the percentage of trials on which stop signals are presented. One
category of adjustments in the stop-signal paradigm is post-stop-
This article was published Online First December 20, 2010.
Patrick G. Bissett and Gordon D. Logan, Department of Psychology,
Vanderbilt University.
This research was supported by Grant FA9550-07-1-0192 from the Air
Force Office of Scientific Research, Grant R01-MH073878-01 from the
National Institute of Mental Health, and Grant BCS 0957074 from the
National Science Foundation.
Correspondence concerning this article should be addressed to Patrick G.
Bissett or Gordon D. Logan, Department of Psychology, Vanderbilt Uni-
versity, Nashville, TN 37240. E-mail: Patrick.g.bissett@vanderbilt.edu or
Gordon.logan@vanderbilt.edu
Journal of Experimental Psychology:
Learning, Memory, and Cognition
2011, Vol. 37, No. 2, 392–404
© 2010 American Psychological Association
0278-7393/10/$12.00DOI: 10.1037/a0021800
392

Page 2

signal slowing: RT often increases on trials following a stop signal
(Rieger & Gauggel, 1999; Verbruggen et al., 2008). Some research
has found that slowing occurs after both successful and failed
inhibition (Rieger & Gauggel, 1999), some research has found
greater slowing after successful inhibition (Emeric et al., 2007;
Verbruggen et al., 2008), and some research has found greater
slowing after failed inhibition (Schachar et al., 2004; Verbruggen
et al., 2008). A major goal of the present research is to determine
the conditions under which post-stop-signal slowing is triggered.
Different hypotheses predict that different events will trigger post-
stop-signal slowing, and our goal was to distinguish between the
hypotheses.
Proactive Adjustment or Accumulation of Post-Stop-
Signal Adjustments?
A second category of adjustments in the stop-signal paradigm is
proactive slowing of go RT in anticipation of stop signals. Ver-
bruggen and Logan (2009b) found strong evidence for proactive
adjustment by presenting cues that indicated whether stop signals
would be relevant for the next few trials. They found slowing on
the first trial following the cue, before any stop signals were
presented. Several investigators have found that go RTs are longer
when stop signals occur more frequently (Emeric et al., 2007;
Logan, 1981; Logan & Burkell, 1986), which they interpreted as
proactive slowing in anticipation of stop signals. An alternative
hypothesis is that the slowing results from accumulated post-stop-
signal slowing. The more frequently stop signals occur, the more
instances there are of post-stop-signal slowing, and these instances
could aggregate to produce longer go RTs. A second goal of the
present research was to distinguish between these hypotheses by
evaluating how quickly go RT returns to baseline after a stop-
signal trial.
Hypotheses for Post-Stop-Signal Adjustments
Our goal of determining the conditions under which post-stop-
signal slowing is triggered is important because it allows us to
distinguish between five hypotheses that have been proposed to
account for post-stop-signal adjustments. Each hypothesis predicts
that post-stop-signal slowing is triggered by a particular kind of
event on a stop-signal trial: the mere occurrence of a stop signal,
failed inhibition, successful inhibition, the occurrence of a rare
event, or the occurrence of a stimulus associated with inhibition.
The five hypotheses link these events and the adjustments in
response to them to broader goals of the executive control system,
such as balancing competing demands and adapting to dynami-
cally changing environments.
The goal priority hypothesis (Leotti & Wager, 2010; Liddle et al.,
2009) suggests that subjects shift their priority to the stop task after
stop signals. Performance on the stop task requires caution, and
performance on the go task requires speed. The occurrence of a
stop signal indicates the need for caution, so subjects increase RT.
The goal priority hypothesis predicts post-stop-signal slowing after
all stop-signal trials (both signal-respond and signal-inhibit trials;
see Figure 1a).
The error detection hypothesis (Laming, 1968; Rabbitt,
1966) suggests that subjects adjust their behavior in response to
errors. When applied to the stop-signal paradigm (Schachar et
al., 2004; Verbruggen et al., 2008), the error detection hypoth-
esis claims that subjects slow go RT when they fail to inhibit,
in order to reduce the probability of stopping errors in the
future. The error is the event that signals adjustment, so the
error detection hypothesis predicts slowing after failed inhibi-
tion (signal-respond trials) but not after successful inhibition
(signal-inhibit trials; see Figure 1b).
The response conflict hypothesis (Botvinick, Braver, Barch,
Carter, & Cohen, 2001) suggests that the recruitment of control
processes occurs as a result of conflict, or coactivation of compet-
ing responses. This hypothesis has been applied to a variety of
cognitive tasks (Botvinick et al., 2001). Schall and colleagues have
applied it to the stop-signal paradigm (Ito, Stuphorn, Brown, &
Schall 2003; Stuphorn & Schall, 2006; Stuphorn, Taylor, & Schall,
2000), arguing that the greatest coactivation of competing re-
sponses occurs on signal-inhibit trials, so slowing should be
greater after signal-inhibit trials than after signal-respond trials
(see Figure 1c). We evaluate this specific prediction and consider
alternative interpretations of the conflict hypothesis in the General
Discussion.
The surprise hypothesis predicts slowing after rare events,
which elicit an orienting response that competes with normal
processing (Notebaert et al., 2009; Nu ´n ˜ez Castellar, Ku ¨hn, Fias, &
Notebaert, 2010). When applied to the stop-signal paradigm, the
surprise hypothesis predicts slowing after all stop signal trials
(signal-respond and signal-inhibit) because stop signals occur on a
minority of trials. In this respect, its predictions are similar to the
goal priority hypothesis. However, this hypothesis also predicts
greater slowing when stop signals are less likely, so it can be
distinguished from the goal priority hypothesis by manipulating
the percentage of stop signals, as we do in Experiment 1 (see
Figure 1d).
The memory hypothesis proposes that stimuli that occur on
stop trials become associated with stopping, and the association
is stronger when inhibition is successful (signal-inhibit trials;
Verbruggen & Logan, 2008a, 2008b; Verbruggen et al., 2008).
If the stimuli repeat, they retrieve associations with stopping,
and that slows RT. If the stimuli do not repeat, they do not
retrieve associations with stopping to slow RT. Thus, the mem-
ory hypothesis predicts greater post-stop-signal slowing when
stimuli repeat after stop trials than when stimuli do not repeat,
and this difference should be larger after successful inhibition
than after unsuccessful inhibition. In principle, the responses
that were supposed to occur on stop trials could become asso-
ciated with stopping as well, so the memory hypothesis is
consistent with slower RT following stop trials when the re-
sponse repeats. In practice, associations between stimuli and
stopping are stronger than associations between responses and
stopping; post-stop-signal slowing is greater for stimulus rep-
etitions than for response repetitions (Verbruggen et al., 2008).
In the present experiments, we distinguish the memory hypoth-
esis from the other four hypotheses by separating stimulus
repetitions, response repetitions, and response alternations. The
memory hypothesis predicts greater slowing with stimulus rep-
etitions, and possibly after response repetitions (see Figure 1e).
The other hypotheses do not make any predictions about the
effects of stimulus repetition. The other hypotheses can be
distinguished from each other by examining trials in which
stimuli do not repeat.
393
CONTROL ADJUSTMENTS IN THE STOP-SIGNAL PARADIGM

Page 3

These hypotheses are logical alternatives, but they are not nec-
essarily mutually exclusive. Some combinations of hypotheses are
logically possible and may make similar predictions in some
circumstances. Our experiments were designed to allow us to
discriminate between hypotheses and between some combinations
of hypotheses.
The Present Experiments
Our first goal was to test the five hypotheses for post-stop-signal
slowing. Both Experiment 1 and Experiment 2 were designed to
address this first goal. Our second goal was to test hypotheses
about proactive adjustments in the stop-signal paradigm. In Ex-
periment 1, we manipulated the percentage of stop-signal trials,
presenting stop signals on 40% versus 20% of all trials. This
allowed us to determine whether slowing when stop signals are
more frequent is due to anticipatory proactive adjustments or the
accumulation of post-stop-signal adjustments. In Experiment 2, we
manipulated the probability of successful inhibition, holding
the percentage of stop signals constant, to determine whether
stop-signal frequency or probability of success affects post-stop-
signal slowing.
Figure 1.
tions are trial sequences in which Trial N and Trial N ? 1 have different go stimuli; stimulus repetitions are
instances in which Trial N and Trial N ? 1 have the same go stimulus; response repetitions are instances in which
Trial N and Trial N ? 1 have the same correct response but a different go stimulus; and response alternations
are instances in which the stimulus and the response change from Trial N to Trial N ? 1.
The predictions made by the five hypotheses for post-stop-signal adjustments. Nonstimulus repeti-
394
BISSETT AND LOGAN

Page 4

Experiment 1
In Experiment 1, subjects made a manual response to one of
four shapes in the go task (circle, square, diamond, triangle) and
attempted to stop when they heard a stop signal. Subjects
completed two sessions: one session in which 40% of trials
included a stop signal and another session in which 20% of
trials included a stop signal. Mapping four go stimuli onto two
responses allowed us to distinguish stimulus repetition trials, in
which the stimulus and response repeated (e.g., circle3circle),
response repetition trials, in which the stimulus changed but the
response repeated (e.g., square3circle if square and circle were
mapped onto the same response), and response alternation trials,
in which both the stimulus and the response changed (e.g.,
diamond3circle if diamond and circle were mapped onto different
responses). We evaluated the memory hypothesis by comparing
stimulus repetition trials and response repetition trials with re-
sponse alternation trials. We evaluated the goal priority, error
detection, conflict, and surprise hypotheses on response repetition
and response alternation trials. In both the 40% and 20% condi-
tions, the goal priority hypothesis predicts slowing after both
signal-respond and signal-inhibit trials (Figure 1a), the error de-
tection hypothesis predicts slowing only after signal-respond trials
(Figure 1b), the conflict hypothesis predicts slowing only after
signal-inhibit trials (Figure 1c), and the surprise hypothesis pre-
dicts slowing after both signal-inhibit and signal-respond trials.
The surprise hypothesis predicts greater slowing in the 20% con-
dition than in the 40% condition (Figure 1d). The memory hypoth-
esis predicts greater slowing after stimulus repetitions, especially
after signal-inhibit trials, in both the 40% and 20% conditions, and
could accommodate greater slowing after response repetitions than
response alternations (Figure 1e).
Manipulating the percentage of stop-signal trials allowed us to
determine why mean go RT increases with more frequent stop
signals (Emeric et al., 2007; Logan, 1981; Logan & Burkell, 1986).
The increase in RT may reflect proactive adjustment in anticipa-
tion of frequent stop signals, or it may reflect the accumulation of
post-stop-signal slowing. To evaluate these hypotheses, we as-
sessed how quickly RT returned to baseline after a stop signal. If
RT returns to baseline quickly, we can rule out accumulated
post-stop-signal slowing and support proactive slowing. If RT
returns to baseline slowly, then accumulated post-stop-signal slow-
ing is a viable explanation of slower go RT with more frequent
stop signals.
Method
Subjects.
Nashville area and were compensated $24 for two separate 1-hr
sessions on consecutive days. All subjects had normal or
corrected-to-normal vision, and all were naive as to the purpose of
the experiment. Subjects did not participate in Experiment 2 or any
similar studies during the previous two months. We replaced two
subjects whose probabilities of successful stopping fell outside the
95% confidence interval of .5 probability of stopping to a stop
signal.
Apparatus and stimuli.
The experiment was run on a Pen-
tium Dual-Core PC running E-Prime 1 (Psychology Software
Tools; www.pstnet.com). The stimuli were presented on a 19-in.
Twenty-four subjects were recruited from the
(48.3-cm) cathode ray tube monitor. The go task was to respond to
a single black shape on a white background presented in the center
of the screen. The shape was chosen from a set of four shapes:
triangle, circle, square, or diamond. The height of the triangle was
4 cm, the diameter of the circle was 4 cm, each side of the square
was 4 cm, and the diamond was 4 cm from point to point. Subjects
responded by pressing the “1” or “0” key on the top row of a
QWERTY keyboard with the left or right index finger, respec-
tively. The stop signal was an auditory tone (70 dB, 100 ms, 500
Hz) presented through closed headphones (Sennheiser eH 150).
Procedure.
Each trial began with a 500-ms fixation display,
followed by a single black shape chosen randomly from a set of
four shapes. Subjects were instructed to respond as quickly and
accurately as possible to the shape according to a specific response
mapping. Two of the shapes were mapped onto the keyboard
number “1,” and two of the shapes were mapped onto the keyboard
number “0,” and the response mapping was counterbalanced
across subjects. The shape stimulus remained on the screen for 850
ms. In order to discourage a strategy of waiting for stop signals to
occur, the experimenter instructed subjects to respond while the
shape was displayed on the screen. The shape was followed by a
1,000-ms intertrial interval (ITI) during which the screen was left
blank.
Between sessions, we manipulated the percentage of trials that
included a stop signal. In one session, stop signals occurred on
40% of trials; in the other session, stop signals occurred on 20% of
trials. The order of stop-signal percentages was counterbalanced
across subjects. The stop signal indicated to subjects that they
should withhold their response for that trial. The stop signal was
initially presented 250 ms after the presentation of the shape (i.e.,
SSD ? 250 ms). The SSD was continually adjusted according to
a tracking procedure, increasing by 50 ms after each signal-inhibit
trial and decreasing by 50 ms after each signal-respond trial. This
tracking procedure yields a probability of .5 of responding to a stop
signal (Levitt, 1971). If the probability of responding is .5, the race
between going and stopping is tied, and the stop-signal RT (SSRT)
can be estimated by subtracting mean SSD from mean go RT
(Logan, Schachar, & Tannock, 1997).
The experiment began with written and verbal instructions. Sub-
jects were instructed to respond quickly and accurately to the shape
and to do their best to withhold their response when stop signals
occurred. Subjects were instructed not to wait for the stop signal.
After the instructions, subjects were given 24 trials of
experimenter-supervised practice. At the completion of practice,
subjects were given feedback on speed and accuracy. If subjects
did not respond with accuracy above 75% and RT below 1,000 ms,
the practice was repeated until these thresholds were met. After
practice, subjects completed the main task, which included five
blocks of 240 trials. At the end of each block, subjects were given
feedback on RT and accuracy, as well as the percentage of trials on
which they responded while the shape remained on the screen.
Results and Discussion
Mean RTs and accuracies from no-stop-signal trials were cal-
culated for each cell of a 2 (pre or post: Trial N ? 1 or Trial N ?
1) ? 2 (percentage of stop signals: 40% or 20%) ? 3 (Trial N: go,
signal-respond, or signal-inhibit) ? 3 (repetition: response alter-
nation, response repetition, or stimulus repetition) experimental
395
CONTROL ADJUSTMENTS IN THE STOP-SIGNAL PARADIGM

Page 5

design (we use N when the trial of interest is either a go or a stop
signal, S when the trial is a stop trial, and G when the trial is a go
signal). Repeated measures analyses of variance (ANOVAs) were
conducted on the mean go RTs and accuracies with this design.
Accuracy results did not contradict RTs, and accuracies were high,
so we focused on RTs. Summary tables for these ANOVAs appear
in Table 1. Mean go RTs and accuracies across subjects appear in
Table 2. We included go RTs from correct trials that were shorter
than 1,850 ms (shape presentation time plus ITI). To increase the
number of observations, we included S ? 1 trials that preceded
multiple stop signals and S ? 1 trials that followed multiple stop
signals if the multiple stop signals all had the same outcome
(signal-respond or signal-inhibit). To ensure that this inclusion did
not affect our conclusions, we compared the ANOVA in Table 1
with a matching ANOVA including trials with only one interven-
ing stop signal between Trials S ? 1 and S ? 1. The means in the
two ANOVA designs were very similar (r ? .95, p ? .01), and the
ANOVAs led to the same conclusions. Our interpretations of
the data are based primarily on planned comparisons instead of
ANOVA factors. We conducted the ANOVAs to provide error
terms for the planned comparisons.
In addition to traditional null hypothesis significance testing, we
computed Bayes factors (Bs) for our contrasts to compare the
likelihood of the data under the null hypothesis of no difference
with the likelihood of the data under the alternative hypothesis of
some difference. This allowed us to quantify the extent to which
the data support the null hypothesis, giving the odds in favor of the
null hypothesis (Rouder, Speckman, Sun, Morey, & Iverson,
2009). The maximum possible odds ratio in support of the null
hypothesis with 24 subjects is approximately 6.38 (i.e., if the
difference is 0), but there is no maximum odds ratio for the
alternative hypothesis, because, in principle, the difference is lim-
itless, but if the odds ratio is greater than 10, we present it as ?10.
In general, odds ratios greater than 3 are considered as substantial
support for the favored hypothesis (Rouder et al., 2009). We
present Bayes factors as single numbers. Those that occur with
significant effects (accompanied by p values) should be interpreted
as the odds in favor of the alternative hypothesis; those that occur
with nonsignificant effects (not accompanied by p values) should
be interpreted as the odds in favor of the null hypothesis.
The percentage of stop signals had no effect on the probability
of responding given a stop signal (M ? .50 for 20% and 40% stop
signals), F(1, 23) ? 1, B ? 5.86, indicating that the tracking
algorithm was successful. Consistent with previous research
(Logan, 1981; Logan & Burkell, 1986), there was no significant
effect of percentage of stop signals on SSRT (Ms ? 243 and 224
Table 1
Experiment 1 Analysis of Variance Summary Table for Go Reaction Time and Go Accuracy
Measure
df
Reaction time Accuracy
MSEF MSEF
Pre or post stop signal
Percentage of stop signals
Trial N
Repetition
Pre or Post ? Trial N
Pre or Post ? Percentage of Stop Signals
Pre or Post ? Repetition
Percentage of Stop Signals ? Trial N
Percentage of Stop Signals ? Repetition
Trial N ? Repetition
Pre or Post ? Percentage of Stop Signals ? Trial N
Pre or Post ? Trial N ? Repetition
Pre or Post ? Percentage of Stop Signals ? Repetition
Percentage of Stop Signals ? Trial N ? Repetition
Pre or Post ? Percentage of Stop Signals ? Trial N ? Repetition
1, 23
1, 23
2, 46
2, 46
2, 46
1, 23
2, 46
2, 46
2, 46
4, 92
2, 46
4, 92
2, 46
4, 92
4, 92
968 3.93
10.37??
21.29??
8.81??
12.94??
0.43
10.19??
4.68?
0.79
6.60??
7.02??
6.29??
0.03
0.04
1.28
.001
.006
.004
.005
.004
.001
.005
.003
.001
.003
.002
.002
.002
.002
.003
0.01
0.92
0.49
19.20??
2.50
1.53
15.04??
1.37
0.65
1.96
0.10
0.41
1.88
1.51
0.88
62,409
1,236
1,823
1,511
366
2,010
612
714
1,221
611
375
416
1,508
330
?p ? .05.
??p ? .01.
Table 2
Experiment 1 Mean Go Reaction Time and Go Accuracy Before
(Pre) and After (Post) a Stop Signal
Measure
Go
Signal
respond
Signal
inhibit
Pre PostPrePost PrePost
40% stop signals
Response alternations
RT
ACC
Response repetitions
RT
ACC
Stimulus repetitions
RT
ACC
516
93
504
94
503
94
522
93
511
92
530
94
511
92
513
90
518
93
539
88
516
93
544
89
515
94
463
96
503
94
511
95
531
92
543
95
20% stop signals
Response alternations
Reaction time
Accuracy
Response repetitions
Reaction time
Accuracy
Stimulus repetitions
Reaction time
Accuracy
458
94
461
94
455
93
468
94
466
93
470
93
458
93
468
90
454
92
477
87
468
93
480
92
459
93
430
97
458
94
449
97
463
93
466
98
396
BISSETT AND LOGAN

Page 6

ms for the 20% and 40% conditions, respectively), F(1, 23) ?
3.07, MSE ? 1,395, p ? .10, but the Bayes factor does not suggest
substantial support for the null hypothesis (B ? 1.58).
We conducted planned comparisons to evaluate the effect of stop
signals on RT, calculating double difference scores and testing
their significance. First, we subtracted RT for go trials preceding
stop-signal trials (Trial S ? 1) from RT for go trials following
stop-signal trials (Trial S ? 1) to remove the effect of slow
fluctuations in RT over the course of the experiment (Gilden,
Thornton, & Mallon, 1995; Wagenmakers, Farrell, & Ratcliff,
2004). Nelson, Boucher, Logan, Palmeri, and Schall (2010)
showed that slow fluctuations may contaminate estimates of
post-stop-signal slowing. These researchers found that trials pre-
ceding signal-inhibit trials tended to be slower than trials preced-
ing signal-respond trials. Ignoring these preceding trials led to an
overestimate of post-stop-signal slowing after signal-inhibit trials
and an underestimate after signal-respond trials. In order to deter-
mine whether our estimates would be affected by slow RT fluc-
tuations, we conducted a 2 (percentage of stop signals: 40% or
20%) ? 3 (Trial N: go, signal-respond, or signal-inhibit) ? 3
(repetition: response alternation, response repetition, or stimulus
repetition) repeated measure ANOVA on Trial N ? 1 RT. This
ANOVA revealed an effect of Trial N on N ? 1 RT, F(2, 46) ?
6.89, MSE ? 728, p ? .01, B ? 2.90. A planned comparison
showed that RT on trials preceding a signal-inhibit trial (M ? 491
ms) was longer than RT on trials preceding a signal-respond trial
(M ? 481 ms), F(1, 46) ? 10.03, MSE ? 728, p ? .01, B ? 8.81.
This showed that we needed to subtract each condition’s N ? 1 RT
from our estimates of post-stop-signal slowing in order to remove
the effects of Trial N ? 1. Then we calculated the corresponding
difference for go trials preceding (G ? 1) and following (G ? 1)
go trials to control for the effect of repetition on the intervening
trial (G in this case; S in the stop-signal case). Finally, we sub-
tracted the (G ? 1) ? (G ? 1) difference from the (S ? 1) ? (S ?
1) difference to isolate the effect of the stop signal. The double
difference scores measuring post-stop-signal slowing are presented
in Figure 2. Post-stop-signal slowing would be evident as a posi-
tive double difference score, which is what we found (M ? 24 ms),
F(1, 46) ? 27.42, MSE ? 1,511, p ? .01, B ? 10.
To evaluate goal priority, error detection, conflict, and surprise
explanations of post-stop-signal slowing, we focused on response
repetition and response alternation trials, as these hypotheses do
not make predictions about the effects of stimulus repetition. We
found significant slowing (M ? 17 ms) collapsed across response
repetitions and response alternations, F(1, 92) ? 18.84, MSE ?
375, p ? .01, B ?10.
Post-stop-signal slowing was similar for signal-respond trials
(M ? 19 ms) and signal-inhibit trials (M ? 15 ms), F(1, 92) ? 1,
MSE ? 375, B ? 4.48. This result supports the goal priority
hypothesis, the surprise hypothesis, or a combination of the error
detection and conflict hypotheses. The Bayes factor analysis sug-
gests that it provides substantial evidence against error detection
and conflict monitoring alone.
The surprise hypothesis predicts greater slowing in the 20%
stop-signals condition than in the 40% stop-signals condition
(Notebaert et al., 2009; Nu ´n ˜ez Castellar et al., 2010). The results
revealed the opposite effect. There was greater slowing in the 40%
condition (M ? 29 ms) than in the 20% condition (M ? 6 ms), F(1,
92) ? 19.39, MSE ? 330, p ? .01, B ? 10, disconfirming the
surprise hypothesis.
To evaluate the conflict hypothesis further, we compared slow-
ing after signal-inhibit trials with long and short SSDs. Conflict
occurs when there is strong concurrent activation of response
processes (Botvinick et al., 2001). Stop process activation must be
strong on signal-inhibit trials, because the stop process wins the
race. Go process activation should be higher when SSD is long
because the go process has more time to accumulate activation
than when SSD is short. Thus, conflict should be greater for
long-SSD signal-inhibit trials than for short-SSD signal-inhibit
trials (Ito et al., 2003; Stuphorn & Schall, 2006; Stuphorn et al.,
2000). To test this, we compared signal-inhibit trials with SSDs
above and below the median SSD and found no difference in
post-stop-signal slowing (14 ms and 12 ms for long and short
SSDs, respectively), F(1, 92) ? 1, MSE ? 375, B ? 5.93. This
constitutes strong evidence against the conflict hypothesis, and
consequently against the combination of the conflict and error
detection hypotheses, as an explanation of equal slowing after
signal-inhibit and signal-respond trials. Post-stop-signal slowing is
best explained by the goal priority hypothesis, as stop signals
Figure 2.
ble difference score [(S ? 1) ? (S ? 1)] ? [(G ? 1) ? (G ? 1)] for 40%
(Figure 2a) and 20% (Figure 2b) stop signal sessions. Stimulus repetitions
(Stim Rep) are instances in which Trial N and Trial N ? 1 have the same
go stimulus; response repetitions (Resp Rep) are instances in which Trial
N and Trial N ? 1 have the same correct response but a different go
stimulus; and response alternations (Resp Alt) are instances in which the
stimulus and the response change from Trial N to Trial N ? 1.
Experiment 1 post-stop-signal adjustment calculated as a dou-
397
CONTROL ADJUSTMENTS IN THE STOP-SIGNAL PARADIGM

Page 7

appear to shift priority to caution and the stop task independently
of success at stopping.
The memory hypothesis predicts substantial slowing when the
go stimulus repeats, especially after signal-inhibit trials (Verbrug-
gen & Logan, 2008a, 2008b; Verbruggen et al., 2008) and could
accommodate greater slowing after response repetitions than after
response alternations. We found greater slowing after stimulus
repetitions (M ? 44 ms) than after the combination of response
repetitions and response alternations (M ? 17 ms), F(1, 92) ?
47.76, MSE ? 375, p ? .01, B ?10. There is weak evidence that
slowing was greater after stimulus repetition when subjects suc-
cessfully inhibited (M ? 48 ms) than when they failed to inhibit
(M ? 40 ms) on Trial N, F(1, 92) ? 4.06, MSE ? 375, p ? .05,
B ? .96, consistent with previous research (Verbruggen et al.,
2008). There was greater slowing when stimuli repeated in the
40% condition (M ? 62 ms) than in the 20% condition (M ? 27
ms), F(1, 92) ? 45.41, MSE ? 330, p ? .01, B ?10. Slowing was
the same for response repetitions (M ? 15 ms) and response
alternations (M ? 18 ms), F(1, 92) ? 1.0, MSE ? 375, B ? 4.25,
suggesting that subjects do not associate responses with stopping.
These results suggest that subjects associate stimuli with stopping,
which is consistent with previous evidence for the memory hy-
pothesis (Verbruggen & Logan, 2008a, 2008b; Verbruggen et al.,
2008).
Mean RT was longer in the 40% condition (M ? 514 ms) than
in the 20% condition (M ? 459 ms), F(1, 23) ? 10.37 MSE ?
62,409, p ? .01, B ? 9.86, which is consistent with proactive
slowing (Logan, 1981; Logan & Burkell, 1986; Verbruggen &
Logan, 2009b) or the accumulated effects of post-stop-signal slow-
ing. To distinguish between these interpretations, we compared the
rate at which RT returned to baseline after a stop signal. We
compared RT on the second trial after a stop signal (S ? 2) with
RT on the trial before (S ? 1) and found no residual slowing in
Trial S ? 2. Trial S ? 2 RT was slightly faster than S ? 1 in the
40% condition (Ms ? 504 ms and 510 ms for S ? 2 and S ? 1,
respectively), F(1, 23) ? 6.88, MSE ? 64.41, p ? .05, B ? 2.89,
and there was no difference in the 20% condition (Ms ? 456 ms
and 459 ms for S ? 2 and S ? 1, respectively), F(1, 23) ? 1.73,
MSE ? 46.97, B ? 2.84. However, Trial S ? 2 is always preceded
by a go Trial S ? 1, which may result in speeding on Trial S ? 2.
So we analyzed sequences in which Trial S ? 2 was also a go trial,
compared Trial S ? 2 RT with Trial S ? 1 RT, and found no
differences in the 40% condition (S ? 2: 499 ms, S ? 1: 498 ms),
F(1, 23) ? 1, MSE ? 135.19, B ? 5.33, or the 20% condition
(N ? 2: 457 ms, N ? 1: 456 ms), F(1, 23) ? 1, MSE ? 54.03,
B ? 4.58. As a further test of the hypothesis that longer RTs in the
40% condition were not due to accumulated post-stop-signal slow-
ing, we calculated mean RT excluding S ? 1 trials and found that
RT in the 40% condition (M ? 501 ms) remained longer than in
the 20% condition (M ? 456 ms), F(1, 23) ? 8.40, MSE ?
2,834.26, p ? .01, B ? 5.02. This suggests that accumulated
post-stop-signal slowing cannot account for the increase in RT
with stop-signal frequency and that proactive adjustments play a
key role in the effect.
Conclusions
The results for post-stop-signal support the goal priority hypoth-
esis (Leotti & Wager, 2010; Liddle et al., 2009), and provide
evidence against the error detection (Laming, 1968; Rabbitt,
1966), conflict monitoring (Ito et al., 2003; Stuphorn & Schall,
2006; Stuphorn et al., 2000), and surprise (Notebaert et al., 2009;
Nu ´n ˜ez Castellar et al., 2010) hypotheses. The results for stimulus
repetition trials support the memory hypothesis (Verbruggen &
Logan, 2008a, 2008b; Verbruggen et al., 2008), but there was no
support for responses becoming associated with stopping.
The finding that post-stop-signal slowing in the 40% condition
was larger than that in the 20% conditions suggests that global,
experiment-wide manipulations can influence the slowing after a
stop signal. The hypotheses for post-stop-signal adjustments gen-
erally consider only the immediately preceding trial, suggesting a
local explanation for slowing. Our results suggest that post-stop-
signal slowing is affected by global factors.
Mean RT was longer in the 40% stop signal condition than in
the 20% condition. The results suggest that the accumulation of
post-stop-signal slowing cannot account for this effect and that
proactive adjustments occur in anticipation of stop signals (Logan,
1981; Logan & Burkell, 1986; Verbruggen & Logan, 2009b).
Experiment 2
In Experiment 1, we found more post-stop-signal slowing when
stop signals were frequent than when they were infrequent. Our
results suggest that global manipulations encompassing many tri-
als affected post-stop-signal adjustments. It is not clear whether
this effect was due to the prevalence of stop-signal trials or to the
prevalence of failed (or successful) inhibition. To distinguish be-
tween these interpretations, in Experiment 2 we manipulated the
probability of successful inhibition while holding the percentage of
stop signals constant. We used two different tracking procedures
for setting SSD that were designed to produce successful inhibition
on 30% and 70% of the trials. In a third condition, we intermixed
the two tracking procedures to produce an overall success rate of
50%. If post-stop-signal slowing depends on the probability of a
stop signal, there should be no differences among the three con-
ditions. If post-stop-signal slowing depends on the probability of
successful (or failed) inhibition, then we should see large differ-
ences between the three conditions.
Method
Subjects.
Nashville area and were compensated $36 for three separate 1-hr
sessions on three consecutive days. All subjects had normal or
corrected-to-normal vision, and all were naive as to the purpose of
the experiment. Subjects did not participate in Experiment 1 or any
similar studies during the previous two months. We replaced seven
subjects whose probabilities of successful stopping fell outside the
95% confidence interval of the expected probability of stopping in
any of the three conditions.
Apparatus and stimuli.
The apparatus and stimuli were the
same as in Experiment 1.
Design and procedure.
The basic trial structure and instruc-
tions for Experiment 2 were identical to those of Experiment 1
with the following exceptions: The probability of a stop signal was
held constant at .33. We manipulated the tracking algorithm to
produce different probabilities of successful inhibition. There were
three different conditions involving different tracking procedures
Twenty-four subjects were recruited from the
398
BISSETT AND LOGAN

Page 8

run in different sessions, and the order of tracking conditions was
counterbalanced across subjects. In the high-SSD condition, SSD
increased by 50 ms each time subjects successfully inhibited their
response (signal-inhibit) but decreased by 50 ms when subjects
failed to inhibit (signal-respond) on two consecutive stop-signal
trials. This tracking procedure should yield a .707 probability of
responding to the stop signal (Levitt, 1971). In the low-SSD
condition, SSD increased by 50 ms when subjects successfully
inhibited their response on two consecutive stop-signal trials but
decreased by 50 ms each time subjects failed to inhibit. This
tracking procedure should yield a .293 probability of responding to
a stop signal (Levitt, 1971). In the mixed-SSD condition, the high-
and low-SSD tracking procedures were interleaved within the
session. The mixed condition should yield an average probability
of responding to the stop signal of 0.5. Because the different
tracking procedures yield probabilities of responding that are dif-
ferent from 0.5, we used the integration method to calculate SSRT
(Logan & Cowan, 1984).
Results and Discussion
Mean go RTs and accuracies from no-stop-signal trials were
calculated for each cell of a 2 (pre or post: Trial N ? 1 or Trial
N ? 1) ? 3 (tracking condition: high, mixed, or low) ? 3 (Trial N:
go, signal-respond, or signal-inhibit) ? 3 (Repetition: response
alternation, response repetition, or stimulus repetition) experimen-
tal design. Summary tables for repeated measures ANOVAs on go
RTs and accuracies based on this design are presented in Table 3.
Accuracy results did not contradict RTs, and accuracies were high,
so we focused on RTs. Mean go RTs and accuracies across
subjects appear in Table 4, and post-stop-signal slowing double
difference scores are depicted in Figure 3. The inclusion criteria
were the same as in Experiment 1. Our interpretations of the data
are based primarily on planned comparisons instead of ANOVA
factors. We conducted the ANOVAs to provide error terms for the
planned comparisons. We also included Bayes factors to quantify
the odds in favor of the null or alternative hypothesis, as in
Experiment 1.
The main ANOVA presented in Table 3 collapses across the
high and low tracking procedures in the mixed condition. This was
necessary because we could not separate go Trial N in high- and
low-tracking procedures. To test the effect of tracking procedure
on post-stop-signal effects in the mixed condition, we conducted a
separate 2 (pre or post: Trial S ? 1 or Trial S ? 1) ? 2 (tracking
procedure: high or low) ? 2 (Trial S: signal-respond or signal-
inhibit) ? 3 (repetition: stimulus repetition, response repetition, or
response alternation) repeated measures ANOVA on go RTs. We
found an interaction of Pre or Post ? Tracking, F(1, 23) ? 6.41,
MSE ? 221.86, p ? .05, B ? 2.43, but no main effect of tracking
or other interactions of tracking with post-stop-signal effects (all
ps ? .15). This significant effect is driven by larger post-stop-
signal slowing in the mixed-low condition (M ? 20 ms) than in the
mixed-high condition (M ? 13 ms). This result does not distin-
guish post-stop-signal slowing hypotheses, and we do not have a
ready explanation, so we do not discuss it further. Complete
mixed-condition RT and go accuracy results are presented in
Table 4.
As expected, tracking condition had strong effects on the prob-
ability of responding given a stop signal, F(2, 46) ? 8,477.64,
MSE ? .0001, p ? .01. The observed probabilities (Ms ? .70, .50,
and .31 for high-, mixed-, and low-SSD conditions, respectively)
matched the expected probabilities closely, indicating that the
tracking procedure worked as we intended (Levitt, 1971). Tracking
condition also affected SSRT (Ms ? 190 ms, 205 ms, and 236 ms
for high-, mixed-, and low-SSD conditions, respectively), F(2,
46) ? 9.97, MSE ? 1,330, p ? .01. This is expected from the race
model: The stop process should beat the go process more often, the
shorter the SSD; therefore, the effective SSRT will include more of
the upper tail of the SSRT distribution, the shorter the SSD (Logan
& Burkell, 1986).
In order to determine whether there were slow RT fluctuations
in these data, we conducted a 3 (tracking condition: high, mixed,
Table 3
Experiment 2 Analysis of Variance Summary Table for Go Reaction Time and Go Accuracy
Measure
df
Reaction time Accuracy
MSEF MSEF
Pre or post stop signal
Tracking condition
Trial N
Repetition
Pre or Post ? Trial N
Pre or Post ? Tracking Condition
Pre or Post ? Repetition
Tracking Condition ? Trial N
Tracking Condition ? Repetition
Trial N ? Repetition
Pre or Post ? Tracking Condition ? Trial N
Pre or Post ? Trial N ? Repetition
Pre or Post ? Tracking Condition ? Repetition
Tracking Condition ? Trial N ? Repetition
Pre or Post ? Tracking Condition ? Trial N ? Repetition
1, 23
2, 46
2, 46
2, 46
2, 46
2, 46
2, 46
4, 92
4, 92
4, 92
4, 92
4, 92
4, 92
8, 184
8, 184
49421.76??
2.21
47.31??
10.89??
57.25??
1.59
18.12??
6.72??
0.55
15.25??
1.88
14.28??
0.77
0.88
0.88
.001
.008
.002
.004
.002
.001
.004
.003
.002
.001
.002
.002
.002
.002
.002
5.91?
0.52
4.41?
15.10??
0.07
0.01
26.37??
0.55
1.17
3.56?
0.56
5.69??
1.69
1.51
1.01
49,746
1,703
1,064
789
382
1,348
2,871
575
492
1,058
363
565
697
646
?p ? .05.
??p ? .01.
399
CONTROL ADJUSTMENTS IN THE STOP-SIGNAL PARADIGM

Page 9

or low) ? 3 (Trial N: go, signal-respond, or signal-inhibit) ? 3
(repetition: response alternation, response repetition, or stimulus
repetition) repeated measure ANOVA on Trial N ? 1 RT. This
ANOVA revealed a large effect of trial N on N ? 1 RT, F(2, 46) ?
41.34, MSE ? 508, p ? .01. A planned comparison showed that
RT on trials preceding a signal-inhibit trial (M ? 499 ms) was
longer than trials preceding a signal-respond trial (M ? 482 ms),
F(1, 46) ? 62.23, MSE ? 508, p ? .01, B ? 10. As in Experiment
1, this justifies the use of double-difference scores as the measure
of post-stop-signal slowing.
Averaged over conditions, there was significant post stop-signal
slowing (31 ms), F(1, 46) ? 135.45, MSE ? 789, p ? .01, B ? 10.
To determine whether the magnitude of post-stop-signal slowing
depends on the probability of successful stopping, we compared
double difference scores in the high-SSD condition (p[inhibit] ?
.30; slowing ? 40 ms) and the low-SSD condition (p[inhibit] ?
.69; slowing ? 24 ms) and found a significant difference overall,
F(1, 92) ? 8.43, MSE ? 1,058, p ? .01, B ? 5.08. The difference
was significant for stimulus repetitions (64 ms vs. 47 ms, respec-
tively), F(1, 184) ? 5.31, MSE ? 646, p ? .05, B ? 1.58, and
response repetitions (49 ms vs. 11 ms, respectively), F(1, 184) ?
28.03, MSE ? 646, p ? .01, B ? 10, but not for response
Figure 3.
ble difference score [(S ? 1) ? (S ? 1)] ? [(G ? 1) ? (G ? 1)] for (a)
the high-stop-signal-delay (SSD) tracking condition, (b) the mixed-SSD
tracking condition, and (c) the low-SSD (Figure 3c) tracking condition.
Stimulus repetitions (Stim Rep) are instances in which Trial N and Trial
N ? 1 have the same go stimulus; response repetitions (Resp Rep) are
instances in which Trial N and Trial N ? 1 have the same correct response
but a different go stimulus; and response alternations (Resp Alt) are
instances in which the stimulus and the response change from Trial N to
Trial N ? 1.
Experiment 2 post-stop-signal adjustment calculated as a dou-
Table 4
Experiment 2 Mean Go Reaction Time and Go Accuracy Before
and After a Stop Signal
Condition/measure
Go
Signal
respond
Signal
inhibit
Pre Post PrePost PrePost
High-SSD tracking
Response alternations
Reaction time
Accuracy
Response repetitions
Reaction time
Accuracy
Stimulus repetitions
Reaction time
Accuracy
Mixed-high-SSD tracking
Response alternations
Reaction time
Accuracy
Response repetitions
Reaction time
Accuracy
Stimulus repetitions
Reaction time
Accuracy
Mixed-low-SSD tracking
Response alternations
Reaction time
Accuracy
Response repetitions
Reaction time
Accuracy
Stimulus repetitions
Reaction time
Accuracy
Low-SSD tracking
Response alternations
Reaction time
Accuracy
Response repetitions
Reaction time
Accuracy
Stimulus repetitions
Reaction time
Accuracy
507
95
498
94
498
95
512
95
523
96
537
97
504
94
494
91
495
94
534
90
523
95
563
94
504
95
455
97
501
95
501
97
525
94
554
95
488
95
483
95
479
96
484
94
488
95
500
95
487
96
484
93
473
95
490
90
495
94
524
95
488
95
446
98
478
93
479
98
483
95
505
94
488
95
483
95
472
96
495
95
488
95
501
97
487
96
484
93
478
96
519
91
486
96
516
96
488
95
446
98
483
94
481
96
493
97
511
96
487
95
482
95
467
95
485
95
485
95
497
95
480
95
486
91
475
95
496
90
484
96
497
92
484
94
440
97
481
93
480
98
490
95
500
97
Note.
SSD ? stop-signal delay.
400
BISSETT AND LOGAN

Page 10

alternations (23 ms vs. 19 ms), F(1, 184) ? 1, MSE ? 646, B ?
5.65. The probability of a stop signal was held constant at .33, so
these differences suggest that post-stop-signal adjustments are
sensitive to the overall probability of failed inhibition.
The response repetition and response alternation conditions al-
lowed additional tests of the goal priority, error detection, and
conflict hypotheses. As in Experiment 1, we collapsed across
response repetitions and response alternations in evaluating these
hypotheses. Post-stop-signal slowing was similar for signal-
respond trials and signal-inhibit trials (Ms ? 25 ms and 23 ms,
respectively), F(1, 92) ? 1.0, MSE ? 363, B ? 4.55, supporting
the goal priority hypothesis or a combination of the error detection
and conflict hypotheses. To evaluate the conflict hypothesis, we
compared slowing after signal-inhibit trials in the two tracking
procedures in the mixed session. The mixed-high SSD procedure
yielded longer SSDs than did the mixed-low SSD procedure
(Ms ? 324 ms and 211 ms for mixed-high and mixed-low SSD
procedures, respectively), F(1, 23) ? 227.26, MSE ? 674, p ?
.01, B ? 10, which means that there should be greater conflict and
therefore greater slowing in the mixed-high SSD procedure. Ex-
cluding stimulus repetitions, there was no difference in signal-
inhibit slowing between mixed-high and mixed-low SSD condi-
tions (Ms ? 32 ms and 33 ms for mixed-high and mixed-low SSD,
respectively), F(1, 23) ? 1, MSE ? 819, B ? 6.33, providing
strong evidence against the conflict hypothesis, and consequently
against the combination of the conflict and error detection hypoth-
eses, as an explanation of equivalent slowing after signal-inhibit
and signal-respond trials. Replicating Experiment 1, the post-stop-
signal slowing data appear best explained by the goal priority
hypothesis, as stop signals appear to shift priority to the stop task
independently of stopping success.
The memory hypothesis predicts greater slowing when the go
stimulus repeats, especially after signal-inhibit trials, and could
accommodate greater slowing when the response repeats. There
was greater slowing after stimulus repetitions (M ? 54 ms) than
after nonstimulus repetitions (M ? 26 ms), F(1, 92) ? 81.84,
MSE ? 363, p ? .01, B ? 10. Within stimulus repetitions, there
was greater slowing after signal-inhibit trials (M ? 65 ms) than
after signal-respond trials (M ? 44 ms), F(1, 92) ? 41.45, MSE ?
363, p ? .01, B ?10. There was also greater slowing after
response repetitions (M ? 31 ms) than after response alternations
(M ? 20 ms), F(1, 92) ? 10.55, MSE ? 363, p ? .01, B ? 10.
These results suggest that both stimuli and responses become
associated with stopping. However, response repetition slowing
did not differ between signal-inhibit trials (M ? 31 ms) and
signal-respond trials (M ? 30 ms), F(1, 92) ? 1 MSE ? 363, B ?
5.77, suggesting that different processes may underlie response
repetition and stimulus repetition effects. Response repetition
slowing may reflect an increase in response threshold, but we did
not see the corresponding increase in accuracy that usually comes
with an increase in threshold (Ratcliff & Smith, 2004).
Conclusions
Experiment 2 tested whether post-stop-signal slowing depends
on the probability of a stop signal or the probability of failed
inhibition. We held the probability of a stop signal constant and
found greater slowing, the higher the probability of failed inhibi-
tion. In future research, we plan to manipulate the probability of a
stop signal and the probability of failed inhibition independently.
Replicating Experiment 1, the results for post-stop-signal slow-
ing support the goal priority hypothesis (Leotti & Wager, 2010;
Liddle et al., 2009) and provide evidence against the error detec-
tion (Laming, 1968; Rabbitt, 1966) and conflict monitoring (Ito et
al., 2003; Stuphorn & Schall, 2006; Stuphorn et al., 2000) hypoth-
eses. The results for stimulus repetition trials support the memory
hypothesis (Verbruggen & Logan, 2008a, 2008b; Verbruggen et
al., 2008). In contrast to Experiment 1, we found support for the
memory hypothesis working at the level of responses.
General Discussion
Consistent with previous experiments, we found evidence for
control adjustments that adapt to the competing demands of stop-
ping and going in the stop-signal paradigm. Subjects slowed go RT
after stop trials (Experiments 1 and 2). Their slowing was greater
when stop signals were more frequent (Experiment 1) and when
failed inhibition was more frequent (Experiment 2). These results
allowed us to discriminate among five hypotheses that explain
post-stop-signal slowing: We found equal slowing after signal-
respond and signal-inhibit trials, which is consistent with the goal
priority hypothesis (Leotti & Wager, 2010; Liddle et al., 2009) and
inconsistent with the error detection (Laming, 1968; Rabbitt, 1966)
and conflict monitoring hypotheses (Ito et al., 2003; Stuphorn &
Schall, 2006; Stuphorn et al., 2000). We found greater slowing
with higher percentages of stop signals, which is inconsistent with
the surprise hypothesis (Notebaert et al., 2009; Nu ´n ˜ez Castellar et
al., 2010). We found greater slowing for stimulus repetitions in
both experiments, which is consistent with the memory hypothesis
(Verbruggen & Logan, 2008a, 2008b; Verbruggen et al., 2008),
but we found mixed evidence for the memory hypothesis working
at the level of responses. In addition, we found evidence against
the hypothesis that longer RTs with higher proportions of stop
signals (Experiment 1) result from accumulated post-stop-signal
slowing. RT returned to pre-stop-signal baseline by the second trial
after a stop signal. Thus, longer RT with more stop signals appears
to result from proactive adjustments (Verbruggen & Logan,
2009b).
Much of the research on post-stop-signal adjustments focuses on
local events from the preceding trial as signals that recruit control
processes, but we showed that this is insufficient. We showed that
control adjustments in the stop-signal paradigm depend on several
factors, including the immediately preceding trial and global fac-
tors that extend beyond it. We showed that the magnitude of
post-stop-signal adjustment depends on the probability of stop
signals and the probability of failed inhibition. Thus, post-stop-
signal adjustments are like proactive adjustments in that they
depend on aggregates of several trials. However, we showed that
post-stop-signal adjustments are different from proactive adjust-
ments, demonstrating that proactive slowing does not result from
aggregated post-stop-signal slowing. The mechanisms by which
aggregates of several trials influence both reactive and proactive
control adjustments are an important topic for future research.
Error Detection in the Stop-Signal Paradigm
The error detection hypothesis suggests that the recruitment of
control processes occurs when an error is detected (Rabbitt, 1966).
401
CONTROL ADJUSTMENTS IN THE STOP-SIGNAL PARADIGM

Page 11

This hypothesis was developed in the context of choice errors in
RT tasks, where posterror slowing is commonly observed (Lam-
ing, 1968; Rabbitt, 1966). Recent studies have applied it to the
stop-signal paradigm (Schachar et al., 2004; Verbruggen et al.,
2008), where it predicts greater slowing after failed inhibition
(signal-respond trials) than after successful inhibition (signal-
inhibit trials). The data from both experiments contradicted this
prediction, suggesting that the error detection hypothesis may not
apply to the stop-signal paradigm. One possibility is that stop
errors are different in important ways from choice errors. They are
typically more frequent: Many stop-signal experiments aim for a
50% success rate (e.g., Logan et al., 1997), whereas most RT
experiments aim for success rates (accuracy) above 90%. Subjects
may be less concerned with correcting errors when they are more
frequent (Gehring, Goss, Coles, Meyer, & Donchin, 1993). Also,
stop errors are errors of timing rather than errors of choice.
Subjects try to stop but do not stop quickly enough. Choice errors
involve confusion among stimuli or responses and may be re-
sponded to differently than errors of timing. Another possibility is
that the error detection hypothesis is not a valid explanation of
posterror slowing in choice RT tasks. Other explanations are
available, including surprise (Notebaert et al., 2009; Nu ´n ˜ez Cas-
tellar et al., 2010) and goal priority (Leotti & Wager, 2010; Liddle
et al., 2009). Additional research is necessary to bridge these two
literatures.
Conflict Monitoring in the Stop-Signal Paradigm
The conflict-monitoring hypothesis (Botvinick et al., 2001) sug-
gests that the recruitment of control processes occurs as a result of
coactivation of competing responses. Evidence from countermand-
ing eye movements in monkeys suggests that coactivation of
movement (go) and fixation (stop) neurons occurs only on signal-
inhibit trials (Ito et al., 2003; Stuphorn & Schall, 2006; Stuphorn
et al., 2000), so post-stop-signal slowing should be greater on
signal-inhibit than on signal-respond trials. Our results showed no
greater slowing for signal-inhibit trials, contradicting this version
of the conflict monitoring hypothesis (Ito et al., 2003; Stuphorn &
Schall, 2006; Stuphorn et al., 2000). Possibly, monkey results do
not generalize to humans or eye movement results do not gener-
alize to keypress responses. Another possibility is that coactivation
of fixation and movement neurons does not index response conflict
in the stop-signal paradigm. Indeed, monkey data show coactiva-
tion of fixation and movement neurons at the beginning and end of
every saccade (Hanes, Patterson & Schall, 1998), so coactivation
of fixation and movement neurons would not be a valid signal for
recruiting control processes.
Another possibility is that conflict occurs on every stop signal
trial because the go stimulus and the stop signal both occur and
require responses that are incompatible. Indeed, the race model
(Logan & Cowan, 1984) assumes that the stop process and the go
process are engaged on every stop-signal trial. One or the other
may win the race, but both runners are running. Executive pro-
cesses may detect response conflict whenever two runners com-
pete and produce post-stop-signal slowing regardless of the out-
come of the race. The issue is whether response conflict depends
on the entire race, as we are suggesting here, or on the competition
at the end of the race, as Schall and colleagues (Ito et al., 2003;
Stuphorn & Schall, 2006; Stuphorn et al., 2000) suggested. Further
research will be required to distinguish these possibilities.
Surprise in the Stop-Signal Paradigm
Nu ´n ˜ez Castellar et al. (2010) and Notebaert et al. (2009) showed
that subjects slowed their responses after low-probability events,
including errors in easy discriminations and correct responses in
difficult discriminations, suggesting that surprising events recruit
control processes. Our data contradicted the surprise hypothesis,
showing greater slowing after frequent than after infrequent events
(stop signals). It is possible that the slowing after correct trials
observed by Nu ´n ˜ez Castellar et al. (2010) and Notebaert et al.
(2009) was a response to the feedback presented after each trial in
those studies, which may have recruited a control process that
raised the threshold for a response in the subsequent trials, increas-
ing both RT and accuracy. We did not present trial-by-trial feed-
back after go or stop trials, so feedback processing cannot explain
our results.
Goal Priority in the Stop-Signal Paradigm
Our results were most consistent with the goal priority hypoth-
esis (Leotti & Wager, 2010; Liddle et al., 2009). We found equal
slowing after signal-inhibit and signal-respond trials, suggesting
that stop signals recruited control processes that adjusted the
balance between stopping and going independently of stop suc-
cess. These results encourage further exploration of the role of
motivation and optimization of performance in the stop-signal task
(see also Wong-Lin, Eckhoff, Holmes & Cohen, 2010).
The goal priority hypothesis is not as constrained as the error
detection (Laming, 1968; Rabbitt, 1966) and conflict monitoring
(Botvinick et al., 2001) hypotheses. Error detection and conflict
monitoring attribute post-stop-signal slowing to a specific event
(errors and response conflict, respectively) that occurs on the
immediately preceding trial. By contrast, goal priority can depend
on factors that extend beyond the immediately preceding trial, such
as subjects’ perception of the percentage of stop signals (Logan,
1981), explicit cues that indicate the importance of the stop signal
(Verbruggen & Logan, 2009b), and the magnitude of reward given
for stopping and going (Liddle et al., 2009). Thus, goal priority is
a more flexible hypothesis, and that flexibility may underlie its
ability to account for a broader range of phenomena. Future
research should exert stronger constraints on the goal priority
hypothesis.
Strategic Adjustment or Automatic Priming?
We suggest that subjects strategically shift their priority toward
caution after a stop signal, but an alternative account is that a stop
signal on Trial S automatically primes the stop process on Trial
S ? 1, slowing go RT (Verbruggen & Logan, 2008a, 2008b;
Verbruggen et al., 2008). Our current design cannot distinguish
strategic adjustments from automatic priming after stop signals,
and further research will be necessary to distinguish these expla-
nations (McNamara, 2005). One possibility would be to implement
a contingency in the sequence of trials that eliminates stop trial
repetitions. If post-stop-signal slowing results from the stop Trial
S automatically priming the stop goal on Trial S ? 1, then
402
BISSETT AND LOGAN

Page 12

post-stop-signal slowing should not be affected by this contin-
gency. If post-stop-signal slowing results from a strategic shift in
task priority after a stop trial, then subjects may incorporate this
contingency into their strategy and eliminate or reverse post-stop-
signal slowing. If post-stop-signal slowing results from automatic
priming, the contingency should not affect it.
Both Experiments 1 and 2 support the memory hypothesis,
which involves priming from the go stimulus. The go stimulus on
a stop Trial S becomes associated with stopping; when it repeats
on Trial S ? 1, it retrieves the association, which primes the stop
task, increasing S ? 1 reaction time (Verbruggen & Logan, 2008a,
2008b; Verbruggen et al., 2008). It is possible that similar priming
effects could result from associations between stopping and re-
sponses. Indeed, we find greater slowing after response repetitions
than response alternations in Experiment 2. However, the response
repetition effects may also be explained in terms of strategic
adjustments of response criteria, so further research would be
required to demonstrate response priming. However, Verbruggen
and Logan (2009a) investigated priming in the stop-signal para-
digm, presenting the words GO and STOP as primes for the go
process and the stop process, and found effects smaller than 10 ms.
The effects in the present experiments were larger than that, so it
seems unlikely that our effects were entirely due to automatic
priming. Moreover, proactive adjustment of mean RT appears to
be strategic, as subjects become more cautious in advance of stop
trials (see also Verbruggen & Logan, 2009b). Therefore, we have
evidence for both strategic (proactive slowing) and automatic
(stimulus priming) adjustments, and additional research is neces-
sary to investigate priming as an explanation for post-stop-signal
slowing after trials that are not stimulus repetitions.
What Do Post-Stop-Signal and Proactive Control
Processes Adjust?
A theory of cognitive control must specify what is controlled,
and that requires a theory of the subordinate processes that are
subject to cognitive control (Logan & Gordon, 2001). For choice
RT tasks, the subordinate processes that respond to stimuli and
generate responses are well described by stochastic accumulator
models (for a review, see Ratcliff & Smith, 2004). In these models,
a choice response is generated when evidence accumulates to a
threshold. Evidence is stochastic, so the accumulation process
produces distributions of finishing times that correspond well to
RT distributions. Stochastic accumulator models include three
basic parameters that could be adjusted in response to control
signals, such as errors, response conflict, or the occurrence of stop
signals: the onset of accumulation, the rate of accumulation, and
the threshold of accumulation (Ratcliff & Smith, 2004). Delaying
the onset, reducing the rate, or raising the threshold could all
produce longer RTs. Fortunately, changes in these parameters have
different consequences for RT and accuracy that can be used to
discriminate among them: Delaying the onset increases RT with-
out affecting accuracy; reducing the rate increases RT and reduces
accuracy; and increasing the threshold increases RT and accuracy.
Indeed, Rabbitt (1966) and Laming (1968) interpreted posterror
slowing as an adjustment of response threshold that increased
accuracy (i.e., a change in the speed–accuracy trade-off).
It is difficult to interpret the post-stop-signal slowing in our
experiments in terms of changes in onset, rate, or threshold be-
cause there were complex interactions involving pre versus post
and trial type (stimulus repetition, response repetition, and re-
sponse alternation). An experiment with more trials would be
required to provide enough data to fit a stochastic accumulator
model to RT distributions for correct and error responses in each
condition (Ratcliff & Smith, 2004). This is an important goal for
future research.
The proactive slowing observed in Experiment 1 is easier to
interpret because percentage of stop signals had no main effect on
accuracy and did not produce any significant interactions. Thus,
the proactive slowing we observed is most likely due to a delay in
the onset of accumulation. A reduction in accumulation rate is
implausible because control adjustments should increase cognitive
control and therefore increase accumulation rate (Botvinick et al.,
2001). An increase in threshold would produce an increase in
accuracy, which we did not observe. Verbruggen and Logan
(2009b) fit the diffusion model to RT and found that proactive
adjustments delayed onset but also increased threshold (also see
Logan, 1981). Further investigation of the mechanisms that medi-
ate control is an important goal for future research.
Conclusions
We have shown that control adjustments in the stop-signal
paradigm are sensitive to the immediately preceding trial and to
aggregates of trials beyond the immediately preceding one. The
results allow us to rule out three of five hypotheses that account for
slowing after stop-signal trials, leaving goal priority and memory
as viable explanations. The evidence that global factors influence
post-stop-signal slowing sets the stage for new research that ad-
dresses the nature of the signals to the control processes, how the
signals are processed, and how the results of executive processing
adjust subordinate processes to balance competing demands.
References
Botvinick, M. M., Braver, T. S., Barch, D. M., Carter, C. S., & Cohen, J. D.
(2001). Conflict monitoring and cognitive control. Psychological Re-
view, 108, 624–652. doi:10.1037/0033-295X.108.3.624
Emeric, E. E., Brown, J. W., Boucher, L., Carpenter, R. H. S., Hanes, D. P.,
Harris, R., . . . Schall, J. D. (2007). Influence of history on countermand-
ing saccade performance in humans and macaque monkeys. Vision
Research, 47, 35–49. doi:10.1016/j.visres.2006.08.032
Gehring, W. J., Goss, B., Coles, M. G. H., Meyer, D. E., & Donchin, E.
(1993). A neural system for error detection and compensation. Psycho-
logical Science, 4, 385–390. doi:10.1111/j.1467-9280.1993.tb00586.x
Gilden, D. L., Thornton, T., & Mallon, M. W. (1995, March 24). 1/f noise
in human cognition. Science, 267, 1837–1839. doi:10.1126/science
.7892611
Hanes, D. P., Patterson, W. F., & Schall, J. D. (1998). The role of frontal
eye field in countermanding saccades: Visual, movement and fixation
activity. Journal of Neurophysiology, 79, 817–834.
Ito, S., Stuphorn, V., Brown, J. W., & Schall, J. D. (2003). Performance
monitoring by the anterior cingulate cortex during saccade countermand-
ing. Science, 302, 120–122. doi:10.1126/science.1087847
Laming, D. (1968). Information theory of choice-reaction times. London,
England: Academic Press.
Leotti, L. A., & Wager, T. D. (2010). Motivational influences on response
inhibition measures. Journal of Experimental Psychology: Human Per-
ception and Performance, 36, 430–447. doi:10.1037/a0016802
Levitt, H. (1971). Transformed up–down methods in psychoacoustics.
403
CONTROL ADJUSTMENTS IN THE STOP-SIGNAL PARADIGM

Page 13

Journal of the Acoustical Society of America, 49, 467–477. doi:10.1121/
1.1912375
Liddle, E. B., Scerif, G., Hollis, C. P., Batty, M. J., Groom, M. J., Liotti,
M., & Liddle, P. F. (2009). Looking before you leap: A theory of
motivated control of action. Cognition, 112, 141–158. doi:10.1016/
j.cognition.2009.03.006
Logan, G. D. (1981). Attention, automaticity, and the ability to stop a
speeded choice response. In J. Long & A. D. Baddeley (Eds.), Attention
and performance IX (pp. 205–222). Hillsdale, NJ: Erlbaum.
Logan, G. D. (1985). Executive control of thought and action. Acta
Psychologica, 60, 193–210. doi:10.1016/0001-6918(85)90055-1
Logan, G. D. (1994). On the ability to inhibit thought and action: A users’
guide to the stop signal paradigm. In D. Dagenbach & T. H. Carr (Eds.),
Inhibitory processes in attention, memory, and language (pp. 189–239).
San Diego, CA: Academic Press.
Logan, G. D., & Burkell, J. (1986). Dependence and independence in
responding to double stimulation: A comparison of stop, change, and
dual-task paradigms. Journal of Experimental Psychology: Human Per-
ception and Performance, 12, 549–563. doi:10.1037/0096-1523
.12.4.549
Logan, G. D., & Cowan, W. B. (1984). On the ability to inhibit thought and
action. A theory of an act of control. Psychological Review, 91, 295–
327. doi:10.1037/0033-295X.91.3.295
Logan, G. D., & Gordon, R. D. (2001). Executive control of visual
attention in dual-task situations. Psychological Review, 108, 393–434.
doi:10.1037/0033-295X.108.2.393
Logan, G. D., Schachar, R. J., & Tannock, R. (1997). Impulsivity and
inhibitory control. Psychological Science, 8, 60–64. doi:10.1111/j.1467-
9280.1997.tb00545.x
McNamara, T. P. (2005). Semantic priming: Perspectives from memory
and word recognition. New York, NY: Psychology Press. doi:10.4324/
9780203338001
Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., Howerter, A., &
Wager, T. (2000). The unity and diversity of executive functions and their
contributions to complex “frontal lobe” tasks: A latent variable analysis.
Cognitive Psychology, 41, 49–100. doi:10.1006/cogp.1999.0734
Nelson, M. J., Boucher, L., Logan, G. D., Palmeri, T. J., & Schall, J. D.
(2010). Nonindependent and nonstationary response times in stopping
and stepping saccade tasks. Attention, Perception, & Psychophysics, 72,
1913–1929. doi:10.3758/APP.72.7.1913
Notebaert, W., Houtman, F., Van Opstal, F., Gevers, W., Fias, W., &
Verguts, T. (2009). Post-error slowing: An orienting account. Cognition,
111, 275–279. doi:10.1016/j.cognition.2009.02.002
Nu ´n ˜ez-Castellar, E., Ku ¨hn, S., Fias, W., & Notebaert, W. (2010). Outcome
expectancy and not accuracy determines posterror slowing: ERP sup-
port. Cognitive, Affective, & Behavioral Neuroscience, 10, 270–278.
doi:10.3758/CABN.10.2.270
Rabbitt, P. M. A. (1966). Errors and error correction in choice-response
tasks. Journal of Experimental Psychology, 71, 264–272. doi:10.1037/
h0022853
Ratcliff, R., & Smith, P. L. (2004). A comparison of sequential sampling
models for two-choice reaction time. Psychological Review, 111, 333–
367. doi:10.1037/0033-295X.111.2.333
Rieger, M., & Gauggel, S. (1999). Inhibitory after-effects in the stop signal
paradigm. British Journal of Psychology, 90, 509–518. doi:10.1348/
000712699161585
Rouder, J. N., Speckman, P. L., Sun, D., Morey, R. D., & Iverson, G.
(2009). Bayesian t test for accepting and rejecting the null hypothesis.
Psychonomic Bulletin & Review, 16, 225–237. doi:10.3758/PBR
.16.2.225
Schachar, R. J., Chen, S., Logan, G. D., Ornstein, T. J., Crosbie, J., Ickowicz,
A., & Pakulak, A. (2004). Evidence for an error monitoring deficit in
attention deficit hyperactivity disorder. Journal of Abnormal Child Psychol-
ogy, 32, 285–293. doi:10.1023/B:JACP.0000026142.11217.f2
Stuphorn, V., & Schall, J. D. (2006). Executive control of countermanding
saccades by the supplementary eye field. Nature Neuroscience, 9, 925–
931. doi:10.1038/nn1714
Stuphorn, V., Taylor, T. L., & Schall, J. D. (2000, December 14). Perfor-
mance monitoring by the supplementary eye field. Nature, 408, 857–
860. doi:10.1038/35048576
Verbruggen, F., & Logan, G. D. (2008a). Automatic and controlled re-
sponse inhibition: Associative learning in the go/no-go and stop-signal
paradigms. Journal of Experimental Psychology: General, 137, 649–
672. doi:10.1037/a0013170
Verbruggen, F., & Logan, G. D. (2008b). Long-term aftereffects of re-
sponse inhibition: Memory retrieval task goals and cognitive control.
Journal of Experimental Psychology: Human Perception and Perfor-
mance, 34, 1229–1235. doi:10.1037/0096-1523.34.5.1229
Verbruggen, F., & Logan, G. D. (2008c). Response inhibition in the
stop-signal paradigm. Trends in Cognitive Sciences, 12, 418–424. doi:
10.1016/j.tics.2008.07.005
Verbruggen, F., & Logan, G. D. (2009a). Automaticity of cognitive con-
trol: Goal priming in response-inhibition paradigms. Journal of Exper-
imental Psychology: Learning, Memory and Cognition, 35, 1381–1388.
doi:10.1037/a0016645
Verbruggen, F., & Logan, G. D. (2009b). Proactive adjustments of re-
sponse strategies in the stop-signal paradigm. Journal of Experimental
Psychology: Human Perception and Performance, 35, 835–854. doi:
10.1037/a0012726
Verbruggen, F., Logan, G. D., Liefooghe, B., & Vandierendonck, A.
(2008). Short-term aftereffects of response inhibition: Repetition prim-
ing or between-trial control adjustment? Journal of Experimental Psy-
chology: Human Perception and Performance, 34, 413–426. doi:
10.1037/0096-1523.34.2.413
Wagenmakers, E.-J., Farrell, S., & Ratcliff, R. (2004). Estimation and
interpretation of 1/f?noise in human cognition. Psychonomic Bulletin &
Review, 11, 579–615.
Wong-Lin, K., Eckhoff, P., Holmes, P., & Cohen, J. D. (2010). Optimal
performance in a countermanding saccade task. Brain Research, 1318,
178–187. doi:10.1016/j.brainres.2009.12.018
Received April 8, 2010
Revision received September 13, 2010
Accepted September 30, 2010 ?
404
BISSETT AND LOGAN