Selective Stopping in Task Switching

Selective Stopping in Task Switching The Role of Response Selection and Response Execution Frederick Verbruggen, Baptist Liefooghe, and André Vandierendonck Department of Experimental Psychology, Ghent University, Belgium Abstract. Recently, several studies stressed the role of response selection in cued task switching. The present study tried to investigate directly the hypothesis that no switch cost can be found when there was no response selection. In two experiments, we combined a cued task switching paradigm with the selective stopping paradigm. Results of the experiments demonstrated that a switch cost was found when participants selected a response, even without response execution. Alternatively, when the response was inhibited without the need of response selection, no switch cost was found. These results provide direct evidence for the distinct role of response selection in cued task switching and suggest that response execution is not a necessary factor to obtain a switch cost. Keywords: response inhibition, response selection, response execution, task switching, selective stop signal task Introduction It is a common finding that switching between two different tasks is associated with a cost in reaction times and accuracy (see Monsell, 2003, for a review). Different proposals have been made to explain this switch cost and it seems that at least two different kinds of processes contribute to the switch cost. First, each task is assumed to be associated with internal constraints (i.e., the task set), enabling a correct performance of the task. Switching would take more time compared to repetition because it involves the additional active reconfiguration process of changing the task set (e.g., Rogers & Monsell, 1995). Second, Allport suggested that at least part of the switch cost is due to carry-over effects of the previous trial (Allport, Styles, & Hsieh, 1994). Later on, Wylie (Wylie & Allport, 2000; Wylie, Javitt, Fox, 2004) hypothesized that the retrieval of previous stimulus-response associations causes between-task interference on the current trial due to the response requirements of the task. This between-task interference delays the responding on the current trial and explains why there is still a switch cost present even when participants have sufficient time to prepare the task before the stimuli are presented. Experimental Psychology 2006; Vol. 53(1):48Ð57 DOI: 10.1027/1618-3169.53.1.48 Recently, also much interest has arisen in the role of response selection in the establishment of the switch cost. The first study that directly addressed the role of response selection in task switching was the paper of Schuch and Koch (2003). These authors suggested that in task switching there are overlapping stimulus-response rules of the different tasks. However, at the stage of response selection the irrelevant stimulus-response rules are inhibited. This inhibition of the task set is still observable on the next trial, causing a delay when participants have to switch to the task that was previously inhibited. Schuch and Koch (2003) demonstrated the importance of response selection by integrating a go/no-go task with the cued task switching paradigm and the backward inhibition paradigm. They found that both the switch cost and the backward inhibition effect were absent after a no-go trial and suggested that this resulted from the absence of a response selection in a no-go trial. No response selection means also that there is no application of the relevant task set. For that reason, there was also no inhibition of the irrelevant task set. In other words, there is no residual inhibition after a no-go trial. Later on, Verbruggen, Liefooghe, Szmalec and Vandierendonck (2005a) replicated this finding with the simple stop signal task (see Logan, ” 2006 Hogrefe & Huber Publishers F. Verbruggen et al.: Selective Stopping in Task Switching 1994, for a review). Unlike a go/no-go task, the stop signal is always presented after the stimulus presentation. Basically, these authors found the same results as Schuch and Koch (2003). When participants could correctly inhibit their response on the previous trial (signal-inhibit trial), no switch cost was found. However, when participants responded in spite of a stop signal (signal-respond trial), there was still a switch cost. Verbruggen et al. (2005a) also suggested that these results are in favor of an account that stresses the role of response selection in task switching. However, contrary to the explanation of Schuch and Koch (2003), Kleinsorge and Gajewski (2004) hypothesized that participants were less willing to engage in advance task set reconfiguration when occasionally no-go trials were presented. Furthermore, they suggested that this lack of advanced preparation resulted in the disappearance of the switch cost. Although it is not exactly clear what can be expected after a signal-respond trial based on the motivational account of Kleinsorge and Gajewski (2004), one could assume that there would be no difference between trials that followed a signal-respond trial compared to trials that followed a signal-inhibit trial. In both types of trials, there was a stop signal presented and Kleinsorge and Gajewski (2004) argued that the motivational aspect was context (or block) based. Thus, the important difference is that in signal-respond trials, the inhibition failed and participants responded whereas on signal-inhibit trials there was no response. Therefore, Verbruggen et al. (2005a) suggested that their results obtained with the stop signal task were best explained by the hypothesis that response selection, or task set application, is an important and mediating factor in task switching. In the present study, we wanted to further investigate the hypothesis that response selection and task set application is indeed a mediating factor in task switching, by means of two different selective stop signal tasks. Verbruggen et al. (2005a) used a simple stop task in which all responses had to be inhibited when an auditory stop signal occurred. However, a selective stop task requires that the stop is cognitively controlled. A selective stop task can be based on a perceptual discrimination by using different tones (e.g., only stop when you hear a high tone; e.g., Bedard et al., 2003), or it can be based on a motor discrimination. Logan, Kantowitz and Riegler (1986), cited by Logan (1994), used this motor version of the selective stop task. On presentation of the stop signal, participants were required to withhold their response with the right hand but to ignore the signal when the response was to be made by the left hand. Logan et al. (1986) suggested that in this version of the stop task, motor inhibition should be focused on a single response instead of cancelling all responses as in the ” 2006 Hogrefe & Huber Publishers 49 simple stop task. Unlike the perceptual variant of the selective stop signal task, the motor variant of the selective stop task implies a response selection in the primary task before the response inhibition since only half of the responses (e.g., only left-handed responses) should be inhibited. For an investigation of the mediating role of response selection, the present study used cued task switching combined with selective stopping based on the response selection of the primary task in Experiment 1 and with selective stopping based on a perceptual discrimination in Experiment 2. This procedure has two important advantages in comparison with previous studies. First, Schuch and Koch (2003) found a general increase of choice reaction times (CRTs) after a no-go trial and suggested that this increase was due to a switch from a no-go trial to a go trial. Only when this switch was made (i.e., deciding whether they had to respond or inhibit), participants would proceed processing the stimulus and the appropriate response. But this implied, as they pointed out (Schuch & Koch, 2003, p. 96), that the go/no-go switch and the task switch should have additive effects in order to preserve their hypothesis that the absence of response selection in a no-go trial caused the disappearance of the switch cost on the next trial. By using the different forms of selective stopping, we avoided this problem since after different forms of stopping the same switch had to be made. Second, in both the studies of Schuch and Koch (2003) and Verbruggen et al. (2005a) response selection and response execution were confounded in the sense that the absence of response selection was always associated with an absence of response execution. Schuch and Koch (2003) tackled this problem indirectly in their Experiment 3 and 4 by demonstrating that response execution without response selection did not cause a switch cost on the next trial. However, one could argue that their manipulation (‘tap both response buttons’) influenced also other processes besides response selection. In Experiment 1 of the present study, there was a direct test of the suggestion that response selection without response execution is a sufficient factor in the establishment of the switch cost. If the response selection hypothesis is correct, one would expect a switch cost after a correctly inhibited trial in the selective stop signal task based on response selection. In order to be sure that participants could not base their decision about the validity of the stop signal on stimulus features, we used eight different digits in two different tasks: a parity task and a magnitude task. We predicted that in Experiment 1 a switch cost should be present after a correctly inhibited trial (i.e., a signal-inhibit trial) when response selection in the primary task is a mediating factor in the establishment of the switch cost on the next trial. Experimental Psychology 2006; Vol. 53(1):48Ð57 50 F. Verbruggen et al.: Selective Stopping in Task Switching Experiment 1 Method Subjects Twenty first-year psychology students (18 females and 2 males) at Ghent University (Belgium) participated for course requirements and credits. All participants had normal or corrected-to-normal vision, were righthanded, and all were naive as to the purpose of the experiment. Materials The experiment was run on a Pentium 4 PC running Tscope (Stevens, Lammertyn, Verbruggen, & Vandierendonck, in press) and the stimuli were presented on a 17-inch monitor. We used the digits 1Ð9 (0.6 ¥ 0.3 cm), excluding 5. The white digits always appeared in the centre of the screen on a black background (see Figure 1). The task cues were presented on the left and the right of the digit. The letters “On” (for “oneven,” meaning odd) and “Ev” (for “even,” meaning even) indicated the parity task; the letters “Kl” (for “kleiner,” meaning smaller) and “Gr” (for “groter,” meaning larger) indicated the magnitude task. The position of the cues always corresponded to the relevant response mapping. For example, “On” was always presented on the left of the digit whereas “Ev” was always presented on the right of the digit. Responses were collected via a response box connected to the parallel port of the PC. Occasionally (one third of the trials), an auditory stop signal (750 Hz, 50 dB, 75 ms) was presented through closed headphones (Sennheiser HD 265-1) shortly after the stimulus onset in the primary task. The validity of the stop signal was presented at the center of the top and bottom of the screen. For example, when participants had to stop their responses with the left hand and ignore the stop signal in case of right-handed responses, we presented “LEFT = STOP” in Dutch (“LINKS = STOPPEN”) at the top and bottom of the screen (see Figure 1). This information remained on the screen during the whole experiment. Task and Procedure There were two different tasks and the same two response buttons were used for both tasks. In the parity task, odd was mapped on the index finger of the left hand and even was mapped on the index finger of the right hand. Smaller than five was mapped on the left Experimental Psychology 2006; Vol. 53(1):48Ð57 finger and larger than five was mapped on the right finger. The validity of the stop signal was dependent on the response hand. One half of the participants had to ignore the stop signal when the response was with the right hand and had to inhibit left handed responses. This mapping was reversed for the other half of the participants. Each trial started with the presentation of the task cue. After 300 ms, the digit appeared in the middle of the screen and required a response within 2,000 ms in case of no-signal trials or invalid signal trials. Both the cue and the stimulus remained on the screen until the response was given, after which the trial ended. When a valid stop signal was presented, the trial ended after 1,500 ms unless participants had responded. The intertrial interval was 1,250 ms. Participants received oral instructions. The experiment consisted of one practice phase and one experimental phase. First, there was one practice block of 20 trials without stop signals. In a second practice block of 48 trials, stop signals could occur. During the practice phase, participants received immediate feedback. The word “FOUT” (meaning wrong) appeared in the centre of the screen for 500 ms when participants made an error. When participants incorrectly suppressed a response on an invalid stop signal, the word “REAGEER” (meaning react) was presented. Finally, when the inhibition failed, the word “STOP” appeared. The experimental phase consisted of eight blocks of 96 trials. On a random selection of one third of the trials, a stop signal was presented. Half of the stop signals was valid, half of the stop signals was invalid. This resulted in 64 valid and 64 invalid stop signals for repetition trials and 64 valid and 64 invalid stop trials for the switch trials. During the experiment, participants received feedback at the end of each block only: The number of errors made during the block, the mean reaction times (CRT), the amount of false alarms (i.e., no response when an invalid stop signal was presented) and the mean probability of stopping were presented. The stop signal delay was initially set at 250 ms and continuously adjusted according to separately staircase tracking procedures for repetition and switch trials to obtain a probability of stopping of .50. In order to avoid “waiting” strategies, participants were informed about the tracking procedure and about the fact that the probability of stopping will approximate 50 %, irrespective of whether they were postponing their response or not. Each time a participant responded to the stimulus in the presence of a valid stop signal, the stop signal delay decreased with 50 ms. When inhibition succeeded after a valid stop signal, the stop signal delay increased with 50 ms. After an invalid stop signal, the stop signal delay was not ad” 2006 Hogrefe & Huber Publishers F. Verbruggen et al.: Selective Stopping in Task Switching 51 cost on trial n, F(1, 19) = 67.28, p , .001. The signal properties of trial n-1 also affected the CRTs on the trial n, F(3, 17) = 76.40, p , .001. Both main effects interacted significantly, F(3, 17) = 39.64, p , .001. Secondly, two-tailed t-tests were performed as a function of the signal properties of trial n-1. After all types of trials, we found a switch cost. There was a switch cost after a no-signal trial, t(19) = -3.85, p , .005, or when an invalid stop signal was presented on trial n1, t(19) = -11.74, p , .001. We found also a switch cost after both a signal-respond trial, t(19) = -5.17, p , .001, and after a signal-inhibit trial, t(19) = -7.05, p , .001. A similar pattern was observed for the error data. A 2 (Trial n: repetition vs. switch) ¥ 4 (Trial n-1: nosignal, invalid signal, signal-respond, signal-inhibit) repeated measures MANOVA revealed a main effect of trial n, F(1, 19) = 24.94, p , .001, and trial n-1, F(3, 17) = 168.6, p , .001. The interaction was also significant, F(3, 17) = 4.48, p , .05. A switch cost was observed after a no-signal trial, t(19) = -3.50, p , .005, and after a signal-inhibit trial, t(19) = -4.30, p , .001. However, after an invalid stop signal and after a signal-respond trial, the switch cost disappeared, t(19) = 1.12, p = .23 and t(19) = -1.11, p = .28, respectively. Figure 1. Example of the display of the screen during the experiment. justed. Based on the assumptions of the horse-race model, SSRT can be calculated by simply subtracting mean SSD from mean CRT (Logan, 1994). Results CRT data were subjected to a within-participant trimming procedure. Mean CRTs of correct trials were calculated after removal of outlying CRTs (3 standard deviations above the mean). This resulted in a data reduction of 0.8 %. Since there were few false alarms (1.6 %; i.e., no response when an invalid stop signal was presented), these data were not further analyzed. All reported F values are approximations to Wilks’ lambda. Signal Data Although the stop signal inhibition was complex, the staircase tracking procedure still produced relatively good results (probability of responding given a stop signal was .47). Thus, SSRTs could be reliably estimated. As can be seen in Table 2, response inhibition in the selective stop task was influenced by task switching, indicated by longer SSRTs for switch trials than for repetition trials, t(19) = -6.17, p , .001. When a stop signal was presented but participants responded (i.e., valid signal-respond trials), a switch cost was observed, t(19) = -4.96, p , .001. Also, when an invalid stop signal was presented and participants correctly ignored the signal (i.e., invalid stop trials), there was also a switch cost, t(19) = -3.81, p , .005. Discussion No-Signal Data No-signal data are presented in Table 1. CRTs were analyzed by means of a 2 (trial n: repetition vs. switch) by 4 (trial n-1: no-signal, invalid signal, signal-respond, signal-inhibit) repeated measures MANOVA. First, for CRTs there was a general switch ” 2006 Hogrefe & Huber Publishers In Experiment 1, we hypothesized that if response selection is indeed the mediating factor in task switching, as suggested by Schuch and Koch (2003), a switch cost should be present, after a signal-inhibit trial since participants had to make a response selection before they knew the validity of the stop signal. Experimental Psychology 2006; Vol. 53(1):48Ð57 52 F. Verbruggen et al.: Selective Stopping in Task Switching Table 1. Choice reaction times (CRT) and error percentages (E%) in Experiment 1 (SDs in parentheses) of repetition trials and switch trials as a function of the signal properties of trial n-1. The switch cost was computed by subtraction the means of repetition trials of the means of the switch trials (* p , .005; ** p , .001). Trial n-1 No-signal Repetition trial Switch trial Switch cost Invalid signal Signal-inhibit CRT E% CRT E% CRT E% CRT E% 717 (119) 754 (144) 37* 2 (1.8) 4 (2.6) 1.8* 760 (31) 836 (50) 76** 4 (0.8) 4 (1.5) -0.5 812 (36) 888 (81) 76** 4 (1.4) 5 (1.4) 0.7 852 (52) 868 (52) 16** 2 (1.0) 3 (0.5) 0.9** Table 2. Stop signal reaction times (SSRT), signalrespond RTs (V-SRT) and RTs of invalid stop trials (IV-SRT) in Experiment 1 (SDs in parentheses; * p , .005; ** p , .001). Repetition trial Switch trial Switch cost Signal-respond SSRT V-SRT IV-SRT 254 (107) 694 (131) 656 (118) 303 (108) 49** 761 (158) 67** 725 (155) 69* The results of Experiment 1 provided direct evidence for this hypothesis since the switch cost was present after all types of trials, even after a correctly inhibited response. Although this cost was smaller, it was still significant. This finding clearly indicates that response selection is a mediating, sufficient factor in task switching, without the execution of the response. A second important the finding is that response inhibition in the selective stop task and task switching do interact. This is different from the findings with the simple stop task. Verbruggen et al. (2005a) found that the SSRTs of switch trials were comparable to the SSRTs of repetition trials. However, in Experiment 1, they did differ significantly. We will get round to this finding in the general discussion. Experiment 2 In Experiment 1, a switch cost was observed when the inhibition on the previous trial succeeded. In other words, there was no response execution on the previous trial. We argued that this finding was due to the fact that participants had to make a response selection before they knew whether they had to stop or not. However, another possibility would be that the different findings in Experiment 1 and the study of Verbruggen et al. (2005a) are due to the more complex Experimental Psychology 2006; Vol. 53(1):48Ð57 form of selective stopping Ð in comparison with simple stopping Ð in Experiment 1, unrelated to the response selection. Therefore, in Experiment 2, cued task switching was combined with a selective stopping based on a perceptual discrimination by using different tones (e.g., Bedard et al., 2003). This form of selective stopping does not require a response selection in the primary task. If the findings of Experiment 1 were indeed due to the response selection in the selective stop task, no switch cost should be present after a signal-inhibit trial in case of selective stopping based on a perceptual discrimination. Method Subjects Nineteen first-year psychology students (17 females and 2 males) at Ghent University (Belgium) participated for course requirements and credits. None of the participants participated in Experiment 1. All participants had normal or corrected-to-normal vision, were right-handed, and all were naive as to the purpose of the experiment. Materials, Task and Procedure The only difference in comparison with Experiment 1 is related to the stop signals. The validity of the stop signal was no longer dependent on the response hand. Instead, we used a perceptual variant of the selective stop signal task; i.e., the pitch of a tone determined whether participants had to stop or not. One half of the participants had to ignore a low tone (250Hz) and suppress their response when a high tone (750Hz) occurred. This mapping was reversed for the other half of the participants. Information about the validity remained on the screen during the experiment. For example, when a high pitched stop signal was valid, ” 2006 Hogrefe & Huber Publishers F. Verbruggen et al.: Selective Stopping in Task Switching 53 Table 3. Choice reaction times (CRT) and error percentages (E%) in Experiment 2 (SDs in parentheses) of repetition trials and switch trials as a function of the signal properties of trial n-1. The switch cost was computed by subtraction the means of repetition trials of the means of the switch trials († p , .05, * p , .005; ** p , .001). Trial n-1 No-signal Repetition trial Switch trial Switch cost Invalid signal Signal-inhibit CRT E% CRT E% CRT E% CRT E% 616 (90) 668 (110) 52* 4 (4.2) 5 (5.5) 1.7† 656 (21) 714 (48) 58** 4 (1.1) 4 (0.5) -0.2 719 (26) 763 (47) 44** 4 (2.5) 3 (1.9) -0.6 702 (19) 699 (19) -3 3 (0.7) 2 (1.7) -0.2 Table 4. Stop signal reaction times (SSRT), signalrespond RTs (V-SRT) and RTs of invalid stop trials (IV-SRT) in Experiment 4 (SDs in parentheses; * p , .05; ** p , .005). Repetition trial Switch trial Switch cost Signal-respond SSRT V-SRT IV-SRT 277 (108) 575 (71) 758 (119) 306 (116) 29* 632 (90) 57** 814 (107) 56* “HIGH = STOP” was presented in Dutch (“HOOG = STOPPEN”) at the top and bottom of the screen. There were no other changes in comparison with Experiment 1. Results We used the same trimming procedure as in Experiment 1. This resulted in a data-loss of 1.6 %. The percentage of false alarms was again very low (1.3 %) and was not further analyzed. All reported F values are approximations to Wilks’ lambda. No-Signal Data Like in Experiment 1, CRTs were analyzed by means of a 2 (trial n: repetition vs. switch) by 4 (trial n-1: no-signal, invalid signal, signal-respond, signal-inhibit) repeated measures MANOVA. Results are presented in Table 3. There was a switch cost on trial n, F(1, 18) = 42.41, p , .001, and an effect of trial n-1, F(3, 16) = 129.33, p , .001. Both main effects interacted significantly, F(3, 16) = 13.33, p , .001. There was a switch cost when the previous trial was a no-signal trial, t(18) = -3.54, p , .005. We also found a switch cost when an invalid stop signal was presented on trial n-1, t(18) = -7.16, p , .001, or when ” 2006 Hogrefe & Huber Publishers participants responded when a valid stop signal was presented, t(18) = -6.22, p , .001. However, there was no switch cost after a signal-inhibit trial, t(18) = 1.40, p = .18. For the error data, the 2 (trial n: repetition vs. switch) by 4 (trial n-1: no-signal, invalid signal, signal-respond, signal-inhibit) repeated measures MANOVA, revealed a main effect of trial n-1, F(3, 16) = 61.33, p , .001. There was no main effect of task switching, F , 1. The interaction tended to be marginally significant, F(3, 16) = 3.03, p = .06. We found only a switch cost after a no-signal trial, t(19) = -2.61, p , .05. There was no switch cost after an invalid stop signal, t(18) = .77, p = .25, after a signalrespond trial, t(18) = 1.40, p = .18, or after a valid stop signal, t(18) = .50, p = .62. Signal Data Signal data are presented in Table 4. Again, the staircase tracking procedure produced good results (probability of responding given a stop signal was .51). Response inhibition in the selective stop task at a perceptual level was also influenced by task switching, t(19) = -2.73, p , .05. When a valid stop signal occurred but participants responded, a switch cost was observed, t(19) = -3.93, p , .001. This was also the case when an invalid stop signal was presented, t(19) = -2.80, p , .05. Discussion The results of Experiment 2 are straightforward. First, we replicated the interaction of Experiment 1 between response inhibition and task switching, indicating that there are indeed common mechanisms or shared resources in both types of tasks. Second, we predicted no switch cost after a signal-inhibit trial because no response selection had to be made in the primary task. This hypothesis was confirmed. These results are in Experimental Psychology 2006; Vol. 53(1):48Ð57 54 F. Verbruggen et al.: Selective Stopping in Task Switching Table 5. Choice reaction times (CRT) in both experiments (SDs in parentheses) of repetition trials and switch trials for both response repetitions and response alternations. The switch cost was computed by subtraction the means of repetition trials of the means of the switch trials († p , .05, * p , .005; ** p , .001). Experiment 1 Task repetition Task alternation Switch cost Response repetition Response alternation Response repetition Response alternation 674 (106) 746 (135) 72** 772 (148) 793 (154) 21† 617 (101) 682 (112) 65* 647 (88) 679 (101) 32* line with the findings of Verbruggen et al. (2005a) and suggest that the finding of Experiment 1 that a switch cost was present after a signal-inhibit trial was not simply due to the fact that a selective stop task was used. However, there remains another mediating factor that can have contributed to the present results.1 After all, there are no response repetitions after a signalinhibit trial and it is a common finding in the literature about task switching that the switch cost is smaller for a response alternation (see, e.g., Rogers & Monsell, 1995; Meiran, 2000). Thus, if the effects of task switching in Experiment 2 are only due to response repetition trials, no switch cost is expected after a signal-inhibit trial, regardless of whether participants had to make a response selection or not. Therefore, in order to exclude this possibility, we performed post hoc analyses for both experiments and looked what the influence of response alternations was in our study. Experiment 1 and 2: Response Repetitions vs. Response Alternations For no-signal trials in both experiments, we analyzed the effect of response repetitions vs. response alternations on the switch cost by means of a 2 (Response: repetition vs. alternation) ¥ 2 (Task: repetition vs. alternation) repeated measures ANOVA. In Experiment 1, there were main effects of response alternation, F(1, 19) = 45.01, p , .001, and task alternation, F(1, 19) = 38.38, p , .001. As can be seen in Table 5, both main effects interacted, F(1, 19) = 16.49, p , .001. Post hoc two-tailed t tests revealed that the switch cost was significant for both response repetitions, t(19) = -6.55, p , .001, and response alternations, t(19) = -2.48, p , .05. Thus, although the switch cost was significantly smaller for a response alternation, the switch cost was still significant. Interestingly, the switch cost for a response alternation, was statistically not different from the switch cost found after a signal-inhibit trial (21 ms vs. 16 ms), F(1, 19) , 1. 1 Experiment 2 In Experiment 2, similar results were found. There was a marginally significant main effect of the response alternation, F(1, 18) = 3.63, p = .07, and a significant effect of task alternation, F(1, 19) = 15.17, p , .001. The interaction between both main effects was again significant, F(1, 19) = 8.54, p , .01. Twotailed t-tests revealed that the switch cost was significant for both response repetitions, t(18) = -3.81, p , .01, and response alternations, t(18) = -3.54, p , .01. In sum, the fact that there are no response repetitions after a signal-inhibit trial cannot explain why there is no switch cost observed after this type of trial in Experiment 2. On the other hand, it can explain why the switch cost is smaller after a signal-inhibit trial compared to the trial that followed a no-signal trial in Experiment 1. When we looked only at trials that followed a no-signal trial but where the response alternated, there was no longer a difference in switch cost. This can also be seen as extra evidence for the fact that response selection, or task application, and not response execution is the crucial factor in task switching. General Discussion In the present study, we further investigated the role of response selection in task switching. Schuch and Koch (2003) demonstrated that after a no-go trial without response selection, no switch cost was observed. They interpreted these findings as evidence for the hypothesis that response selection on the previous trial is necessary to observe a switch cost on the current trial. A similar data pattern was observed by Verbruggen et al. (2005a) who used a simple stop task. By using two different selective stop signal tasks in the present study, the hypotheses that the response selection is the mediating factor, was further investigated. We used selective stopping requiring a response selection in the primary task in Experiment 1 and se- We would like to thank Ulrich Mayr for this suggestion. Experimental Psychology 2006; Vol. 53(1):48Ð57 ” 2006 Hogrefe & Huber Publishers F. Verbruggen et al.: Selective Stopping in Task Switching lective stopping requiring a perceptual discrimination in the selective stop task in Experiment 2. In the first type of the task, participants had to select the (correct) response in the primary task before they knew the validity of the stop signal. In perceptual version of the stop task, the validity of the stop signal was determined by the pitch of the tone, and this could be done independent of the primary task. Results of both experiments are straightforward. In both experiments, we found a switch cost after a nosignal trial and after an invalid stop signal, which is of course not surprising. But the most important comparison between both experiments was what happened after a valid stop signal. Both experiments had in common that participants slowed down their responses when a stop signal was presented on the previous trial, irrespective of the validity of the stop signal. This post-signal adaptation is a common finding in the literature about the stop signal task (see Logan, 1994) and suggests that strategic factors come into play. However, besides this common post-signal adaptation, there was an important difference between both experiments regarding what happened after a signal-inhibit trial. In Experiment 1, there was a switch cost both after a signal-respond and after a signal-inhibit trial. In Experiment 2, this was not the case: Only after signal-respond trials, the switch cost was present. Therefore, this experiment also dissociates neatly between the effect of stop signal presentation (i.e., go vs. no-go trials in terms of the no-go paradigm) and the effect of stopping itself: successful inhibition and not signal presentation cause the disappearance of the switch cost. This difference between the two forms of selective stopping has some important implications. First, in both experiments, after a signal-inhibit trial a switch occurred from a signal trial to a no-signal trial. Thus the problem of additivity of Schuch and Koch (2003) is absent since we dissociated between the two forms of selective stopping. Second, the presence of a switch cost after a signal-inhibit trial in Experiment 1 is indeed in line with the hypothesis that response selection is a mediating factor in task switching. This was previously suggested by Schuch and Koch (2003) and Verbruggen et al. (2005a). Participants had to apply the task set and make a response selection before they knew whether they had to stop or not and we argue that this caused the switch cost after a signal-inhibit trial in Experiment 1. Additionally, Experiment 1 demonstrated beyond doubt that response execution was not necessary to obtain a switch cost. At first sight, the switch cost after a signal-inhibit trial was smaller than after a nosignal trial, 37 ms vs. 16 ms; F(1, 19) = 4.77, p , .05. However, post hoc analyses revealed that this differ” 2006 Hogrefe & Huber Publishers 55 ence is probably due to the difference between response repetitions and response alternations. The switch cost was significantly higher for a response repetition, which is a common finding in the literature about task switching (see, e.g., Rogers & Monsell, 1995; Meiran, 2000) and after a signal-inhibit trial, there are no response repetitions, simply because no response was executed. Therefore, in Experiment 1 we compared the switch cost found after a signal-inhibit trial with the switch cost found for response alternations. Interestingly, there was no longer a difference in the switch cost. This finding can be interpreted as extra evidence that response selection is an important mediating factor in task switching and that response execution is clearly not necessary to observe a switch cost on the next trial. We mentioned already in the introduction that there is also an alternative account for the findings of Schuch and Koch (2003). Kleinsorge and Gajewski (2004) suggested that in a go/no-go paradigm, participants are less willing to prepare the task in advance because they know that a no-go trial may occur. This motivational account, proposed for the go/no-go paradigm, can also easily explain the data of Experiment 1. Participants knew in this experiment that whether they had to stop or not, the stimuli had to be processed and a response selection had to be made. Thus, preparation in advance would be beneficial. This could indeed explain why we found a switch cost in Experiment 1. However, due to the differences in the go/nogo paradigm and the selective stop paradigm that was used in Experiment 2, it is more complicated and therefore, probably harder to explain the data pattern of this experiment in terms of the motivational account of Kleinsorge and Gajewski. In a go/no-go paradigm, the effect of motivation is dependent on the context (Kleinsorge & Gajewski, 2004). Alternatively, it was suggested by several authors that strategic adjustments in the stop signal task are based on the properties of the previous trail (e.g., Logan, 1994). For example, after signal-respond trials, participants tend to make strategic adjustments, and are more cautious to respond compared to no-signal trials. The CRT data of Experiment 1 and 2 demonstrated that participants were indeed more cautious and that they tended to slow down their responses after signal-respond trials, compared to no-signal trials. These findings are in favour of some kind of trial-based strategic adjustment. But even though adjustments were made after signalrespond trials, there was still a switch cost in both experiments. Thus, this seems to suggest that these motivational/strategic differences induced to the failure of response inhibition in the stop task, cannot fully explain the present data pattern and previous results of Verbruggen et al. (2005a). On the other hand, given the above mentioned differences between the go/noExperimental Psychology 2006; Vol. 53(1):48Ð57 56 F. Verbruggen et al.: Selective Stopping in Task Switching go paradigm and the stop signal paradigm, one has to be careful in generalizing the results of the present study. Also, as pointed out by T. Kleinsorge (personal communication), these differences between paradigms make it rather difficult to formulate, based on the Kleinsorge and Gajewski account, specific predictions about the motivational consequences of the selective stop task. All in all, the results of the present paper do not necessarily contradict the results of Kleinsorge and Gajewski (2004), but seem to suggest that there are at least differences in motivational effects of the go/no-go task and the stop signal task. Another inevitable question is what actually becomes inhibited when a stop signal is presented. Mostly, it is assumed that stop signal inhibition is targeted on the inhibition of the response execution (see, e.g., Band & Van Boxtel, 1999, for a neuro-anatomical model). Although the present study does not allow any strong conclusions, one could also hypothesize that under certain conditions, not only the response execution becomes inhibited, but probably also the whole task set. In Experiment 1, the task sets have to be activated and applied. This allows a response selection and based on the result of this response selection, a response is selectively inhibited or executed. In other words, the task set may not be inhibited because the task set is needed to perform correctly the response inhibition. This is in line with Logan et al. (1986), who suggested also that in a simple stop task, all responses become inhibited whereas in the selective stop task used in Experiment 1, inhibition is focused on a single response. This picture could change in Experiment 2. Here, response selection and task set application in the primary task are no longer needed when a stop signal is presented. Under these conditions, one could hypothesize that not only the response execution becomes inhibited, but also the whole task set. If the task set is also inhibited when a stop signal is presented, one expects no longer a difference after a signal-inhibit trial between task repetitions and task alternations. This is because for both types of trials, the task set was inhibited on the previous trial, and as a consequence, task repetition would no longer be beneficial. Note that this explanation does not contradict the hypothesis that response selection is necessary in task switching. After all, we argue that response selection is an important mediating factor, but not a causing factor. The results of the present study can also be related to the proposal of Wylie and Allport (2000), who suggested that part of the switch cost is due to interference caused by the retrieval of stimulus-response associations. The possibility that stimulus-response associations are differentially influenced by simple and selective stopping is also supported by another Experimental Psychology 2006; Vol. 53(1):48Ð57 paradigm. Verbruggen, Liefooghe and Vandierendonck (2005b) found that the negative priming effect (i.e., slower reactions when the target was previously ignored) disappeared after a signal-inhibit trial in the simple stop task, but not in a selective stop task similar to the one used in Experiment 1 of the present study. Recently, Rothermund, Wentura and De Houwer (2005) suggested that negative priming is also due to the retrieval of stimulus-response associations. Given the similarity of both designs, we therefore suggest that in both studies stimulus-response associations are established after a valid stop signal in the selective stop task at response level, even without the actual response execution. Besides the fact that we demonstrated that response selection without response execution is sufficient for the establishment of a switch cost, there was still another important finding in both experiments. Both forms of selective stopping interacted with task switching. Verbruggen et al. (2005a) did not find such an interaction with simple stopping. However, it is not surprising to find such an interaction. First, Logan et al., (1986) already demonstrated that the selective stop task was more susceptible for manipulations in task difficulty, probably due to the higher cognitive demands Ð indicated by larger SSRTs in selective stopping compared with simple stopping. These authors demonstrated that inhibiting one out of four responses was more difficult than inhibiting one out of two responses in the selective stop task. Second, neuro-imaging data demonstrated that there is at least a neuro-anatomical overlap between response inhibition and task switching. More precisely, Aron, Robbins, and Poldrack (2004) suggested on the basis of a meta-analytical study that the right-inferior cortex is strongly activated in both the stop signal task and the task switching paradigm, suggesting the right inferior cortex might play an important role in inhibition processes in different types of tasks. Based on the findings of the present study and the study of Aron et al. (2004), we can hypothesize that the same inhibitory processes work in the selective stop task and in the cued task switching paradigm. However, it might also be the case that it is not necessarily the inhibition in task switching that interacts with the response inhibition in the selective stop signal task. First, several authors suggested that the inhibition of task switching is a more lateral kind of inhibition (e.g., Schuch & Koch, 2003). Mayr and Keele (2000, p. 22) also suggested that the inhibitory process involved in task-switching may be “relatively impenetrable for higher-level control” and preferred the notion of lateral inhibition above the concept of a more active form of self inhibition. Second, in a recent paper, Derrfuss, Brass and von Cramon (2004) found ” 2006 Hogrefe & Huber Publishers F. Verbruggen et al.: Selective Stopping in Task Switching evidence for cognitive control in the posterior frontal cortex. This region was commonly activated in task switching, an interference task (these authors used the Stroop task), and an n-back working memory task. Based on these results, they suggested that the common activation is due to the amount of cognitive control in those different tasks. Given the fact that selective stopping requires a more cognitive controlled stop, it seems plausible to assume that the interaction between task switching and response inhibition in the selective stop task is not necessarily due to common inhibitory mechanisms. Instead, the interaction may be due to the higher cognitively control in both paradigms. In sum, the present study demonstrated that response selection, even without response execution, is indeed an important factor in task switching. In other words, only when the task set is applied, a switch cost is observed on the next trial. This finding is in accordance with the accounts of Schuch and Koch (2003) and Wylie and Allport (2000). Moreover, although it is still unclear where the overlap is precisely situated, the present data pattern also suggests that task switching and response inhibition in the selective stop task seem to rely on common structures or mechanisms. Acknowledgement Frederick Verbruggen is a fellow (grant no. 011D06102) of the Special Research Fund at Ghent University. This research was also supported by grant no. 10251101 to the third author. We would like to thank Thomas Kleinsorge and Ulrich Mayr for their helpful comments on a previous version this manuscript. References Allport, D. A., Styles, E. A., & Hsieh, S. (1994). Shifting intentional set: Exploring the dynamic control of tasks. In: C. Umilta & M. Moscovitch (Eds.), Attention and performance XV (pp. 421Ð452). Cambridge, MA: MIT Press. Aron, A. R., Robbins, T. W., & Poldrack, R. A. (2004). Inhibition and the right inferior frontal cortex. Trends in Cognitive Sciences, 8, 170Ð177. Band, G. P. H., & van Boxtel, G. J. M (1999). Inhibitory motor control in stop paradigms: Review and reinterpretation of neural mechanisms. Acta Psychologica, 101, 179Ð211. Derrfuss, J., Brass, M., & von Cramon, D. Y. (2004). Cognitive control in the posterior frontolateral cortex: Evidence from common activations in task coordination, interference control, and working memory. Neuroimage, 23, 604Ð606. ” 2006 Hogrefe & Huber Publishers 57 Kleinsorge, T., & Gajewski, P. D. (2004). Preparation for a forthcomming task is sufficient to produce subsequent shift costs. Psychonomic Bulletin and Review, 11, 302Ð 306. Logan, G. D. (1994). On the ability to inhibit thought and action: A user’s guide to the stop signal paradigm. In D. Dagenbach & T. H. Carr (Eds.), Inhibitory processes in attention, memory and language (pp. 189Ð239). San Diego: Academic Press. Mayr, U., & Keele, S. (2000). Changing internal constraints on action: The role of backward inhibition. Journal of Experimental Psychology: General, 129, 4Ð26. Meiran, N. (2000). Modeling cognitive control in taskswitching. Psychological Research, 63, 234Ð249. Monsell, S. (2003). Task switching. Trends in Cognitive Sciences, 7, 134Ð140. Rogers, R. D. & Monsell, S. (1995). Costs of a predictable switch between simple cognitive tasks. Journal of Experimental Psychology: General, 124, 207Ð231. Rothermund, K., Wentura, D., & De Houwer, J. (2005). Retrieval of incidental stimulus-response associations as a source of negative priming. Journal of Experimental Psychology: Learning, Cognition and Memory, 31, 482Ð 495. Schuch, S., & Koch, I. (2003). The role of response selection for inhibition of task sets in task shifting. Journal of Experimental Psychology: Human Perception and Performance, 29, 92Ð105. Stevens, M., Lammertyn, J., Verbruggen, F., & Vandierendonck, A. (in press). Tscope: A C library for programming cognitive experiments on the MS Windows platform. Behavior Research Methods. Verbruggen, F., Liefooghe, B., Szmalec, A., & Vandierendonck, A. (2005a). Inhibiting responses when switching: Does it matter? Experimental Psychology, 52, 125Ð130. Verbruggen, F., Liefooghe, B., & Vandierendonck, A. (2005b). On the difference between response inhibition and negative priming: evidence from simple and selective stopping. Psychological Research, 69, 262Ð271. Wylie, G., Allport, A. (2000). Task switching and the measurement of “switch costs”. Psychological Research, 63, 212Ð233. Wylie, G. R., Javitt, D. C., & Foxe, J. J. (2004). The role of response requirements in task switching: Dissolving the residue. Neuroreport, 15, 1079Ð1087. Received January 20, 2005 Revision received May 12, 2005 Accepted May 18, 2005 Frederick Verbruggen Department of Experimental Psychology Ghent University Henri Dunantlaan 2 B-9000 Ghent (Belgium) Tel: +32 9 264 6441 Fax: +32 9 264 6496 E-mail: Frederick.Verbruggen@Ugent.be Experimental Psychology 2006; Vol. 53(1):48Ð57