Phasic Alertness and Residual Switch Costs in Task Switching

Transcription

1 Journal of Experimental Psychology: Human Perception and Performance 2017, Vol. 43, No. 2, American Psychological Association /17/$ Phasic ness and Residual Switch osts in Task Switching Darryl W. Schneider Purdue University Residual switch costs are deficits in task-switching performance that occur despite considerable time to prepare for a task switch. In the present study, the author investigated whether increased phasic alertness modulates residual switch costs. In 2 experiments involving the task-cuing procedure, subjects performed numerical categorization tasks on target digits, with and without an alerting stimulus presented shortly before the target (alert and no-alert trials, respectively). Switch costs were obtained that decreased with a longer cue target interval, indicating subjects engaged in preparation, but large residual switch costs remained. ing effects were obtained in the form of faster overall performance on alert than on no-alert trials, indicating the alerting stimuli increased phasic alertness. ritically, residual switch costs were similar on alert and no-alert trials in both experiments, unaffected by manipulations of alert type, alert availability, and alert target interval. Implications of the results for understanding the relationship between phasic alertness and cognitive control in task switching are discussed. Public Significance Statement People are slower and more error-prone at switching than repeating tasks, even when they are provided with time to prepare for a task switch. In the present study, the author investigated whether the cost of task switching is influenced by temporary changes in alertness. In 2 experiments, college students completed numerical categorization tasks faster when alerting stimuli were presented than when they were not, indicating that simple alerts can improve general task performance. However, large costs of switching tasks still occurred and were unaffected by the meaning, availability, and timing of the alerts. These findings suggest that preparation and control associated with switching tasks are not influenced by a person s state of alertness, clarifying the role of attention in task switching. Keywords: task switching, residual switch cost, preparation, phasic alertness One of the reasons why the topic of task switching has attracted interest in cognitive psychology is that it provides insight regarding preparatory aspects of cognitive control (for reviews, see Kiesel et al., 2010; Vandierendonck, Liefooghe, & Verbruggen, 2010). For example, consider the task-cuing procedure, in which a cue indicates which of two tasks to perform on a target on each trial (for an overview, see Meiran, 2014). By presenting the cue before the target and manipulating the cue target interval (), one varies the time available to prepare for the cued task. A common finding is that switch costs worse performance for task switches than for task repetitions decrease as the gets longer (e.g., Meiran, 1996; Monsell & Mizon, 2006; Schneider & Logan, 2011), indicating that preparatory processing occurred during the. However, there seems to be a limit to this preparation because switch costs sometimes persist despite s of 1,000 ms or longer (e.g., Longman, Lavric, Munteanu, & Monsell, 2014; Meiran, This article was published Online First November 10, I thank Allison Blake, Megan Boston, Jamie olton, Natalie Harpenau, Bailey Masterson, and hasity Ricker for assistance with data collection. orrespondence concerning this article should be addressed to Darryl W. Schneider, Department of Psychological Sciences, Purdue University, 703 Third Street, West Lafayette, IN dws@purdue.edu 1996; Meiran, horev, & Sapir, 2000; Schneider, 2016). These residual switch costs are the focus of the present study. Efforts to reduce or eliminate residual switch costs have been based primarily on De Jong s (2000) failure-to-engage theory. The central premise of the theory is that subjects occasionally fail to engage in preparation, even when considerable time is provided for it. When these failures happen, control processing involved in switching tasks must occur after target onset, resulting in residual switch costs. In support of his theory, De Jong showed that response time (RT) distributions for task-switch trials at long preparation intervals can be modeled as mixtures of prepared and unprepared states (for revised mixture models, see Brown, Lehmann, & Poboka, 2006; Poboka, Karayanidis, & Heathcote, 2014). If residual switch costs arise from failures to engage in preparation, then they might be attenuated by inducing subjects to prepare more often. Nieuwenhuis and Monsell (2002) found that incentivizing preparation via instructions and payoffs had only a modest effect on the estimated probability of preparation and still produced large residual switch costs. Lien, Ruthruff, Remington, and Johnston (2005) found that encouraging preparation by imposing strict time deadlines still resulted in residual switch costs, although not for all stimulus response pairs (but see Lindsen & De Jong, 2010). Verbruggen, Liefooghe, Vandierendonck, and Demanet (2007) found that briefly presenting the cue yielded smaller 317

2 318 SHNEIDER (and nonsignificant) residual switch costs relative to having the cue available for the full and after target onset, but the robustness of that finding is questionable (see Schneider, 2016). On the basis of these studies, it is unclear what can be done to effectively reduce residual switch costs. One possibility that has received limited investigation concerns increasing phasic alertness, a component of attention that reflects rapid but temporary changes in sensitivity to external stimulation (Posner, 1978, 2008; Posner & Boies, 1971). Phasic alertness is usually increased by presenting an alerting stimulus shortly before the target on which a task is to be performed. ing effects are indicated by shorter RTs on alert trials than on no-alert trials. For example, when subjects classify the direction of a target arrow appearing above or below fixation in the Attention Network Test (ANT; Fan, Mcandliss, Sommer, Raz, & Posner, 2002), RTs are typically shorter when an alerting stimulus (e.g., a pair of asterisks) is presented 500 ms before the target than when no alerting stimulus is presented. Phasic alertness has been investigated with multiple tasks hypothesized to involve cognitive control, such as the flanker (Fan et al., 2002; Redick & Engle, 2006), Simon (Böckler, Alpay, & Stürmer, 2011; Fischer, Plessow, & Kiesel, 2010), and Stroop (Weinbach & Henik, 2012) tasks. ing effects on overall RTs are commonly observed, but a counterintuitive finding is that interference effects attributed to cognitive control are often larger on alert trials than on no-alert trials (for overviews, see MacLeod et al., 2010; Nieuwenhuis & De Kleijn, 2013). The interpretation of this finding continues to be debated, with proposed explanations based on inhibition of cognitive control (allejas, Lupiáñez, & Tudela, 2004), activation of established stimulus response associations (Fischer, Plessow, & Kiesel, 2012), and early onset of evidence accumulation (Nieuwenhuis & De Kleijn, 2013). Drawing an analogy from the alerting literature to task switching, one might predict larger residual switch costs on alert than on no-alert trials, especially if residual switch costs reflect interference, as suggested by some authors (e.g., Allport, Styles, & Hsieh, 1994; Meiran, 2000; for discussion, see Vandierendonck et al., 2010). Alternatively, if residual switch costs reflect failures to engage in preparation (De Jong, 2000), then an alerting stimulus might trigger preparatory processing, leading to the prediction of smaller residual switch costs on alert than on no-alert trials. Thus, there are competing predictions regarding the relationship between phasic alertness and residual switch costs. ing effects on task-switching performance have been investigated in few studies. Rogers and Monsell (1995, Experiment 5) examined performance in the alternating-runs procedure, in which task switching was predictable because targets were presented in successive spatial locations associated with tasks. On no-alert trials, the target was presented after a 1,000-ms intertrial interval from the previous response. On alert trials, an alerting stimulus (a rectangle at the location of the next target) was presented 500 ms before the target and remained visible after target onset. Rogers and Monsell found significant alerting benefits and costs for overall RTs and error rates, respectively, but nonsignificant differences in switch costs between alert and no-alert trials. They concluded that increased phasic alertness induced a speed accuracy tradeoff that did not affect task switching. Meiran et al. (2000, Experiment 4) examined performance in the task-cuing procedure, in which subjects classified the up down or left right position of a target appearing in a 2 2 grid based on arrow cues. On no-alert trials, the target was presented after one of four s ranging from 166 to 1,182 ms. On alert trials, an alerting stimulus (a highlighted grid) was presented before the cue at one of four alert cue intervals (AIs) ranging from 132 to 1,082 ms, then the target was presented after a of 166 ms. Meiran et al. found that switch costs decreased with on no-alert trials, indicating subjects engaged in preparation. They also found a significant alerting benefit for overall RTs when the three longest AIs were compared with the shortest AI, but switch costs did not vary significantly across AIs. They concluded that the control processing involved in preparing for a task switch (yielding reductions in switch costs with ) is not a byproduct of phasic alertness. Meiran and horev (2005) examined performance in the taskcuing procedure using the same spatial tasks as did Meiran et al. (2000), except all trials involved alerting stimuli. In their Experiments 1 3, the alerting stimulus (a double-lined grid or an outer frame surrounding the entire stimulus display) was presented before the target at one of four alert target intervals (s) ranging from 66 to 1,032 ms. The two s used in each experiment were always relatively long (1,800 and 2,800 ms in Experiments 1 and 3; 1,800 and 10,000 ms in Experiment 2). In their Experiment 4, the alerting stimulus (a double-lined grid) was presented before the cue at one of four AIs ranging from 116 to 916 ms, then the target was presented after a of either 116 or 816 ms. Meiran and horev found significant alerting benefits for overall RTs when the two or three longest s or AIs were compared with the shortest intervals. They also found that residual switch costs (verified to be residual in Experiments 1 3 because of nonsignificant differences in switch costs across the long s) were significantly smaller at the two longest s or AIs compared with the shortest intervals in all four experiments. They concluded that increased phasic alertness reduces residual switch costs by strengthening task-goal representations. These previous studies do not paint a clear picture of the relationship between phasic alertness and task switching. On the one hand, all the studies consistently showed alerting benefits on overall RTs, indicating subjects processed the alerting stimuli and attained a heightened state of alertness. On the other hand, two studies showed no alerting effects on switch costs (Meiran et al., 2000; Rogers & Monsell, 1995), one study showed smaller residual switch costs on alert trials associated with more optimal alerting intervals (Meiran & horev, 2005), and no studies showed larger residual switch costs on alert than on no-alert trials the pattern seen for interference effects in the alerting literature (MacLeod et al., 2010; Nieuwenhuis & De Kleijn, 2013). However, previous studies were limited in ways that constrain the conclusions that can be drawn from them. One limitation is that using a single intertrial interval (Rogers & Monsell, 1995, Experiment 5) or only long s (Meiran & horev, 2005, Experiments 1 3) does not allow one to determine whether subjects actually engaged in preparation. Recall that preparation is typically inferred from a reduction in switch cost with, which can be established only by using short and long s. Another limitation is that using a single intertrial interval (Rogers & Monsell, 1995, Experiment 5), a single short on alert trials (Meiran et al., 2000, Experiment 4), or a single long (Meiran & horev, 2005, Experiment 4) does not allow one to determine whether the observed switch

3 RESIDUAL SWITH OSTS 319 costs are actually residual (i.e., that they would not be reduced at a longer interval). Stronger claims about switch costs being residual can be made by finding similar switch costs at multiple long s (as in Meiran & horev, 2005, Experiments 1 3). A final limitation is that presenting alerting stimuli on all trials (Meiran & horev, 2005) does not allow one to determine whether the observed residual switch costs differ from those obtained on no-alert trials under typical task-cuing conditions. The purpose of the present study was to investigate the relationship between phasic alertness and residual switch costs in task-switching performance while avoiding the limitations of previous studies. In two experiments involving the task-cuing procedure, subjects performed odd even and small large tasks on target digits, cued by the letters and U. I chose these tasks and cues because they produced large and highly reliable residual switch costs on RTs in a previous study (Schneider, 2016, Experiment 4). On no-alert trials, the cue preceded the target by a of 200, 1,500, or 2,000 ms, which allowed me to determine whether preparation occurred (as indicated by a reduction in switch cost from the 200-ms to the longer s) and whether residual switch costs were obtained (as indicated by similar switch costs at the 1,500- and 2,000-ms s) under typical task-cuing conditions. On alert trials, there was a fixed of 2,000 ms, during which time an alerting stimulus was presented. I chose to present the alerting stimulus after (instead of before) the cue to provide an opportunity to trigger preparatory processing that might not have occurred following cue onset. The type of alerting stimulus and the timing of its presentation were manipulated, as described later for each experiment. I compared performance on no-alert and alert trials in two respects to assess the influence of alerting stimuli. First, I examined whether overall performance (primarily with respect to RTs) was better on alert than on no-alert trials, which served as a manipulation check that the alerting stimuli actually increased phasic alertness. Second, I compared switch costs at the 2,000-ms for alert and no-alert trials, which allowed me to determine whether increased phasic alertness modulated residual switch costs. Experiment 1 The type of alerting stimulus presented on alert trials was manipulated between subjects in Experiment 1. An example of each alert type is shown in Figure 1, which illustrates the sequence of events on alert trials. For subjects in the cue alert group, the alerting stimulus was a more salient version of the cue letter. For subjects in the target-location alert group, the alerting stimulus was a more salient version of the fixation cross that indicated the spatial location of the forthcoming target digit, analogous to the alerting stimulus in Rogers and Monsell (1995, Experiment 5). For subjects in the outer-frame alert group, the alerting stimulus was an outlined rectangle that surrounded the main stimulus display, analogous to the alerting stimulus in Meiran and horev (2005, Experiment 3). I chose these alert types to investigate whether their informational value would modulate any alerting effects on task-switching performance. The cue alert is the most informative because it signals the relevant task. If there were a failure to engage in preparation on the basis of the originally presented cue, then the cue alert might trigger preparation, thereby reducing residual Blank (500 ms) switch costs. The target-location alert is less informative because it signals where the target will be but neither the target identity nor the task to be performed. However, attracting attention to the forthcoming target location could facilitate target encoding, thereby shortening RT and possibly also affecting target processing associated with residual switch costs. The outer-frame alert is mostly uninformative because it provides no information related to either the identities or the locations of the cue and the target. However, on the basis of Meiran and horev s (2005) findings, it might be expected to reduce residual switch costs on alert trials relative to no-alert trials. A single of 500 ms was used in Experiment 1, which meant that all three alert types reliably predicted the time of target onset. I chose that particular value because it was associated with optimal or near-optimal performance in previous work involving tasks performed on single-character stimuli (Posner, 1978; Posner & Boies, 1971) and because it is the typical value used in the ANT (Fan et al., 2002). It was the value used by Rogers and Monsell (1995), and it falls in the middle of the range of values examined by Meiran and horev (2005). The potential effects of manipulating are examined in Experiment 2. Method Fixa on (500 ms) ue ue Time Target Loca on Subjects. A total of 120 students (40 per alert-type group) from Purdue University participated for course credit. Data from three additional subjects were excluded because their mean error rates exceeded a preset inclusion criterion of 20%. All subjects in the present study reported having normal or corrected-to-normal visual acuity. Apparatus. The experiment was conducted on desktop computers that displayed stimuli on monitors and registered responses with millisecond-accurate timing from hronos response devices (Psychology Software Tools, Inc.). Except for the alerting stimuli described later, the stimuli were displayed in white 18-point Arial font on a black background at a viewing distance of approximately 50 cm. Responses were made by using the left and right index (one of) ue Target Interval (; ms) Outer Frame 8 Target Target Interval (; ms) Figure 1. Schematic of events on alert trials in Experiment 1. The category response mapping reminders at the bottom of each display have been omitted.

4 320 SHNEIDER fingers to press the leftmost and rightmost buttons, respectively, on the response device. The response buttons (each 1.3 cm 2.5 cm) were 10.3 cm apart and separated by three unused buttons. White light emitting diodes located approximately 5 cm behind the response buttons were illuminated continuously during the experiment to indicate the relevant buttons. Tasks. Two numerical categorization tasks were performed on the target digits 1 9, excluding 5. The odd even task involved pressing a response button to categorize a target as odd or even, whereas the small large task involved pressing a response button to categorize a target as smaller or larger than 5. The tasks were cued by the letters and U, with cue task mappings counterbalanced across subjects. One of the categories associated with each task was mapped to each response button (e.g., odd and small categories mapped to the left button; even and large categories mapped to the right button), with category response mappings counterbalanced across subjects. Reminders of the category response mappings were displayed throughout each trial at the bottom of the screen. Procedure. Subjects were seated at computers in individual testing rooms after providing informed consent for a study protocol approved by the Purdue University Institutional Review Board. Instructions were presented onscreen and read aloud by the experimenter. Subjects completed eight example trials (with accuracy feedback and, if necessary, experimenter guidance) during the instructions before beginning the experiment proper, which consisted of six blocks of 64 trials per block (without accuracy feedback or experimenter guidance). Each block included 48 noalert trials and 16 alert trials presented in random order. For no-alert trials, every possible combination of, cue, and target occurred once per block. For alert trials, which involved a fixed, every possible combination of cue and target occurred once per block. Key aspects of the trial procedure are illustrated in Figure 1. Each trial started with a blank display for 500 ms, followed by a fixation display consisting of two vertically arranged crosses (the distance between the centers of the crosses was 1.1 cm) presented centrally for 500 ms. The top fixation cross was then replaced by a cue. On no-alert trials, the cue display appeared for a of 200, 1,500, or 2,000 ms, then the bottom fixation cross was replaced by a target. On alert trials, the cue display appeared for 1,500 ms, then one of three types of alerting stimuli (see Figure 1) appeared for an of 500 ms, resulting in a of 2,000 ms. For the cue alert group, the alerting stimulus was a more salient version of the cue (displayed in green 28-point Arial bold font) that replaced the original cue. For the target-location alert group, the alerting stimulus was a more salient version of the bottom fixation cross (displayed in green 28-point Arial bold font) that replaced the original cross. For the outer-frame alert group, the alerting stimulus was an outlined rectangle (displayed in green, measuring 1.4 cm 2.1 cm) that surrounded the cue and the bottom fixation cross. After the elapsed, the alerting stimulus disappeared (for the cue alert group, the original cue reappeared in its place) and the bottom fixation cross was replaced by a target. On both no-alert and alert trials, the final display consisted of the cue and the target, which remained onscreen until the subject responded or a response deadline of 2,500 ms elapsed, and then the next trial commenced. Subjects were made aware of the response deadline and instructed to respond quickly and accurately. They were informed that the cues would be selected randomly (which meant that tasks would occur unpredictably) but that the could be used to prepare for the cued task. They were also informed about the nature (but not the purpose) of the alerting stimulus that would sometimes appear during the. Results The first block, the first trial of each subsequent block, and any trials on which subjects failed to respond before the deadline (.8% of trials) were excluded from analysis. Error trials were excluded from the RT analyses. Mean RTs and error rates are presented in Figure 2, collapsed over alert type because it had no significant effects on performance (see Table 1 for the means for all conditions). Statistical results from analyses of variance (ANOVAs) and contrasts are summarized in Table 2. Baseline analysis. This analysis focused on no-alert trials to determine whether preparation effects occurred and whether switch costs at the long s could be considered residual. Mean RTs and error rates for no-alert trials were submitted to 2 (transition: task switch or task repetition) 3 (: 200, 1,500, or 2,000 ms) repeated-measures ANOVAs. 1 As shown in Figure 2, RTs were longer and error rates were higher for task switches (946 ms and 5.9%) than for task repetitions (818 ms and 3.6%), yielding significant main effects of transition. Performance improved from the 200-ms (923 ms and 5.4%) to the 1,500-ms (864 ms and 4.5%), yielding significant main effects of. Switch costs for RTs and error rates decreased by 99 ms and 1.5% from the 200-ms to the 1,500-ms, yielding significant interactions between transition and. Planned contrasts for the interactions revealed that switch costs did not differ significantly between the 1,500-ms (94 ms and 1.9%) and the 2,000-ms (95 ms and 1.8%). Bayesfactor analyses using scaled JZS Bayes factors with r 1 (Rouder, Speckman, Sun, Morey, & Iverson, 2009) for the same contrasts indicated that no difference in switch costs was 13.8 and 13.7 times more likely than was a difference for RTs and error rates, respectively. ing analysis. This analysis focused on no-alert and alert trials at the longest to determine whether alerting effects occurred. Mean RTs and error rates were submitted to 3 (alert type: cue, target location, or outer frame) 2 (transition: task switch or task repetition) 2 (alert status: no alert or alert) mixed-measures ANOVAs. RTs were shorter on alert trials (810 ms) than on no-alert trials (858 ms), yielding a significant main effect of alert status. As shown in Figure 2, switch costs were similar on no-alert and alert trials for both RTs (95 and 86 ms, respectively) and error rates (1.8% and 1.9%, respectively), yielding significant main effects of transition but nonsignificant interactions involving alert status. Bayes-factor analyses indicated that no difference in switch costs between no-alert and alert trials was 9.0 and 13.7 times more likely than was a difference for RTs and error rates, respectively. No other effects were significant in the ANOVAs. 1 The analysis was repeated with alert type (cue, target location, or outer frame) included as a between-subjects factor to assess contextual effects on no-alert trials. There were no significant effects involving alert type.

5 RESIDUAL SWITH OSTS 321 Discussion A Mean Response Time (ms) Switch ost (ms) 1,400 1,300 1,200 1,100 1, Task Switch (No ) Task Repe on (No ) Task Switch () Task Repe on () ,500 2,000 2,000 No ,500 2,000 2,000 There were several important findings in Experiment 1. The baseline analysis revealed switch costs that decreased with on no-alert trials, indicating subjects often engaged in preparation. That analysis also revealed that switch costs did not differ significantly between the two long s (and Bayes factors favored no differences), indicating they can be considered residual switch costs. The alerting analysis revealed that the large residual switch costs at the longest did not differ significantly between noalert and alert trials (and Bayes factors favored no differences), indicating the alerting stimuli did not influence task-switching performance in particular. However, that analysis also revealed an alerting benefit on overall RTs, indicating subjects attended to the alerting stimuli and experienced a general improvement in performance. Finally, there were no significant effects of alert type, indicating the informational value of the alerting stimulus affected B Mean Error Rate (%) D Switch ost (%) Task Switch (No ) Task Repe on (No ) Task Switch () Task Repe on () ,500 2,000 2,000 No ,500 2,000 2,000 Figure 2. Experiment 1 data. Mean response times (Panel A) and error rates (Panel B) as a function of transition, alert status, and cue target interval (; ms). Switch costs for response times (Panel ) and error rates (Panel D) as a function of alert status and. The alert target interval (; ms) is included on the x-axes. neither task switching in particular nor performance more generally. Experiment 2 Even though alerting effects on overall RTs were observed in Experiment 1, there are two potential concerns regarding how alerting was implemented. The first concern is that because the of 500 ms falls in the middle of the four s (66, 350, 632, and 1,032 ms) examined by Meiran and horev (2005, Experiments 1 3), who found that residual switch costs at their two longest s were significantly smaller than those at their shortest, it might not have been long enough to increase phasic alertness to the extent that it would modulate residual switch costs. It is possible that a longer might be needed to reliably detect a difference in residual switch costs between alert and no-alert trials.

6 322 SHNEIDER Table 1 Mean Response Times (in ms) and Error Rates (as Percentages) for Experiment 1 Transition and (ms) status Response time Error rate type: ue Task switch 200 No alert 990 (30) 8.3 (.9) 1,500 No alert 873 (30) 5.4 (.8) 2,000 No alert 886 (31) 5.3 (.9) 831 (31) 4.8 (.8) Task repetition 200 No alert 805 (26) 4.2 (.7) 1,500 No alert 804 (25) 3.8 (.7) 2,000 No alert 798 (26) 4.1 (.6) 747 (27) 2.6 (.6) type: Target location Task switch 200 No alert 1,021 (24) 6.6 (.9) 1,500 No alert 925 (29) 5.1 (.7) 2,000 No alert 909 (30) 5.4 (.7) 861 (32) 5.6 (.9) Task repetition 200 No alert 822 (18) 3.0 (.5) 1,500 No alert 820 (23) 2.9 (.6) 2,000 No alert 817 (27) 3.0 (.6) 771 (24) 2.7 (.4) type: Outer frame Task switch 200 No alert 1,048 (34) 6.2 (1.0) 1,500 No alert 937 (34) 5.8 (.9) 2,000 No alert 923 (34) 5.2 (.8) 865 (35) 4.5 (.8) Task repetition 200 No alert 852 (29) 3.9 (.7) 1,500 No alert 828 (26) 4.0 (.7) 2,000 No alert 817 (27) 3.3 (.6) 783 (30) 3.9 (.7) Note. Standard errors of the means appear in parentheses. cue target interval. A second potential concern is that the alerting stimulus was transient in Experiment 1: After the elapsed, the alerting stimulus was replaced by the original cue (cue alert) or the target (target-location alert), or it simply disappeared (outer-frame alert). Although transient alerting stimuli are typical in studies using the ANT (Fan et al., 2002), persistent alerting stimuli that remain present after target onset have tended to be used in previous studies of phasic alertness in task switching (Meiran & horev, 2005; Rogers & Monsell, 1995). It is possible that the offset of a transient alerting stimulus has a disruptive effect on performance that makes it difficult to detect a difference in residual switch costs between alert and no-alert trials. I introduced two manipulations in Experiment 2 to address these concerns. First, was manipulated within subjects at three levels (100, 500, and 900 ms) to determine whether alerting effects on overall RTs and possibly also residual switch costs vary across s. I chose those s because the shortest one is close to the least optimal (66 ms) and the longest one is in between the two most optimal s (632 and 1,032 ms) used by Meiran and horev (2005); the middle was included to maintain continuity with Experiment 1. Second, alert availability was manipulated between subjects. For subjects in the transient alert group, the alerting stimulus which was always an outer frame disappeared at target onset, as in Experiment 1. For subjects in the persistent alert group, the alerting stimulus remained present after target onset, such that the outer frame surrounded the cue and the target until either a response was made or the response deadline elapsed. I chose the outer frame as the alerting stimulus because it most closely resembles the alerting stimuli used in Meiran and horev s study, in which the manipulation of was found to affect residual switch costs. Thus, Experiment 2 conceptually replicates some of the key methodological aspects of their study. Method Subjects. A total of 80 students (40 per alert-availability group) from Purdue University participated for course credit. None of the subjects had participated in Experiment 1. Data from six additional subjects were excluded because their mean error rates exceeded 20% (four subjects) or they failed to respond before the deadline on at least 10% of trials (two subjects). Apparatus and tasks. These aspects of the experiment were identical to those of Experiment 1. Procedure. The procedure was identical to that of Experiment 1, except for the following changes. The experiment consisted of 14 blocks of 48 trials per block. Each successive pair of blocks included 48 no-alert trials and 48 alert trials presented in random order. The no-alert trials were identical to those of Experiment 1, including the manipulation of (200, 1,500, or 2,000 ms). The alert trials involved the outer-frame alerting stimulus from Experiment 1 and two new manipulations. First, whereas the remained fixed at 2,000 ms for all alert trials, the was 100, 500, or 900 ms (occurring equally often with every possible combination of cue and target). Second, alert availability was manipulated between subjects: For subjects in the transient alert group, the Table 2 Analyses of Variance and ontrasts for Experiment 1 Analysis and effect df Response time Error rate F MSE 2 p F MSE 2 p Baseline T 1, , , , T 2, , ontrast a 1, , ing Y 2, , T 1, , A 1, , Y T 2, , Y A 2, , T A 1, , Y T A 2, , Note. T transition; cue target interval; Y alert type; A alert status. a Planned contrast for the interaction, comparing switch costs at the two longest cue target intervals. p.05.

7 RESIDUAL SWITH OSTS 323 outer-frame alerting stimulus disappeared at target onset, as in Experiment 1, whereas for subjects in the persistent alert group, the alerting stimulus remained present after target onset, until the end of the trial. Results The first block, the first trial of each subsequent block, and any trials on which subjects failed to respond before the deadline (.7% of trials) were excluded from analysis. Error trials were excluded from the RT analyses. Mean RTs and error rates are presented in Figure 3, collapsed over alert availability because it had only one significant effect on performance (described later; see Table 3 for the means for all conditions). Statistical results from ANOVAs and contrasts are summarized in Table 4. Baseline analysis. This analysis focused on no-alert trials to determine whether preparation effects occurred and whether switch costs at the long s could be considered residual. Mean RTs and error rates for no-alert trials were submitted to 2 (transition: task switch or task repetition) 3 (: 200, 1,500, or 2,000 ms) repeated-measures ANOVAs. 2 As shown in Figure 3, RTs were longer and error rates were higher for task switches (913 ms and 5.9%) than for task repetitions (797 ms and 3.4%), yielding significant main effects of transition. Performance improved from the 200-ms (909 ms and 5.4%) to the 1,500-ms (839 ms and 4.8%), yielding significant main effects of. Switch costs for RTs and error rates decreased by 100 ms and 2.0% from the 200-ms to the 1,500-ms, yielding significant interactions between transition and. Planned contrasts for the interactions revealed that switch costs did not differ significantly between the 1,500-ms (87 ms and 2.0%) and the 2,000-ms (71 ms and 1.6%). Bayes-factor analyses for the same contrasts indicated that no difference in switch costs was 3.9 and 8.6 times more likely than was a difference for RTs and error rates, respectively. ing analysis. This analysis focused on no-alert and alert trials at the longest to determine whether alerting effects occurred. Mean RTs and error rates were submitted to 2 (alert availability: transient or persistent) 2 (transition: task switch or task repetition) 2 (alert status: no alert or alert) mixed-measures ANOVAs, averaging over for the alert trials. RTs were shorter on alert trials (773 ms) than on no-alert trials (817 ms), yielding a significant main effect of alert status. This alerting effect was larger when the alerting stimulus was persistent (53 ms) than when it was transient (34 ms), yielding a significant interaction between alert availability and alert status. As shown in Figure 3, switch costs were similar on no-alert and alert trials for both RTs (71 and 75 ms, respectively) and error rates (1.6% and 1.8%, respectively), yielding significant main effects of transition but nonsignificant interactions involving alert status. Bayes-factor analyses indicated that no difference in switch costs between no-alert and alert trials was 10.2 and 10.6 times more likely than was a difference for RTs and error rates, respectively. No other effects were significant in the ANOVAs. ing interval analysis. This analysis focused on alert trials to determine whether alerting effects differed across s. Mean RTs and error rates for alert trials were submitted to 2 (transition: task switch or task repetition) 3 (: 100, 500, or 900 ms) repeated-measures ANOVAs. RTs became progressively shorter (782 to 774 to 764 ms) from the shortest to the longest, yielding a significant main effect of. As shown in Figure 3, switch costs were similar from the shortest to the longest for both RTs (80, 74, and 70 ms) and error rates (1.9%, 2.0%, and 1.6%), yielding significant main effects of transition but nonsignificant interactions involving. Bayes-factor analyses indicated that no difference in switch costs between the shortest and the longest s was 6.4 and 9.9 times more likely than was a difference for RTs and error rates, respectively. No other effects were significant in the ANOVAs. Discussion There were several important findings in Experiment 2 that replicated those of Experiment 1. The baseline analysis revealed switch costs that decreased with on no-alert trials, indicating subjects often engaged in preparation. That analysis also revealed that switch costs did not differ significantly between the two long s (and Bayes factors favored no differences), indicating they can be considered residual switch costs. The alerting analysis revealed that the large residual switch costs at the longest did not differ significantly between no-alert and alert trials averaged across s (and Bayes factors favored no differences), indicating the alerting stimuli did not influence task-switching performance in particular. However, that analysis also revealed an alerting benefit on overall RTs (the effect was larger for persistent than for transient alerting stimuli), indicating subjects attended to the alerting stimuli and experienced a general improvement in performance, especially when the alerting stimulus remained present after target onset. The alerting interval analysis revealed that the residual switch costs on alert trials did not differ significantly across s, despite progressively shorter RTs as the varied from 100 to 900 ms, indicating that a more optimal alerting interval was ineffective for reliably reducing residual switch costs (cf. Meiran & horev, 2005). General Discussion My goal in the present study was to achieve insight regarding the relationship between phasic alertness and residual switch costs in task-switching performance. I conducted two taskcuing experiments in which subjects did odd even and small large tasks on target digits, with and without an alerting stimulus presented shortly before the target (alert and no-alert trials, respectively). Both experiments involved short and long s on no-alert trials to assess preparation effects and residual switch costs, whereas a long was used on alert trials to assess potential alerting effects on overall performance and residual switch costs. In Experiment 1, I manipulated alert type (cue, target location, or outer frame) to determine whether the informational value of alerting stimuli would modulate alerting effects. In Experiment 2, I manipulated the and alert availability (transient or persistent) to determine whether more optimal conditions existed for detecting alerting effects on residual switch costs. The experiments produced similar patterns of results (see Figures 2 and 3). Switch costs were obtained that decreased with 2 The analysis was repeated with alert availability (transient or persistent) included as a between-subjects factor to assess contextual effects on no-alert trials. There were no significant effects involving alert availability.

8 324 SHNEIDER A Mean Response Time (ms) Switch ost (ms) 1,400 1,300 1,200 1,100 1, Task Switch (No ) Task Repe on (No ) Task Switch () Task Repe on () ,500 2,000 2,000 2,000 2,000 No ,500 2,000 2,000 2,000 2,000 on no-alert trials, indicating subjects engaged in preparatory processing that benefitted task-switching performance. Switch costs were similar at the two long s on no-alert trials, indicating those s were long enough for subjects to finish whatever preparation they could do, and the switch costs could be considered residual. ing effects were obtained in the form of shorter RTs on alert than on no-alert trials, which served as a manipulation check that the alerting stimuli actually increased phasic alertness. ritically, residual switch costs were similar on alert and no-alert trials in both experiments for RTs and error rates, regardless of alert type, alert availability, and. This finding represents evidence that increased phasic alertness does not modulate residual switch costs. The results of the present study are consistent with some previous findings in the literature. The reduction in switch costs from the shortest to the long s replicates extant evidence of B Mean Error Rate (%) D Switch ost (%) Task Switch (No ) Task Repe on (No ) Task Switch () Task Repe on () ,500 2,000 2,000 2,000 2,000 No ,500 2,000 2,000 2,000 2,000 Figure 3. Experiment 2 data. Mean response times (Panel A) and error rates (Panel B) as a function of transition, alert status, cue target interval (; ms), and alert target interval (; ms). Switch costs for response times (Panel ) and error rates (Panel D) as a function of alert status, cue target interval (), and alert target interval (). preparation in task switching (e.g., Meiran, 1996; Monsell & Mizon, 2006; Schneider & Logan, 2011). The observation of large and reliable residual switch costs at the long s fits with other findings of performance decrements in task switching despite ample preparation time (e.g., Longman et al., 2014; Meiran, 1996; Meiran et al., 2000; Rogers & Monsell, 1995), in addition to replicating recent results from my laboratory using the same tasks and cues (Schneider, 2016, Experiment 4). The observation of alerting effects on overall RTs is consistent with what has been found in both the alerting literature (e.g., Fan et al., 2002; Nieuwenhuis & De Kleijn, 2013) and the task-switching literature (Meiran & horev, 2005; Rogers & Monsell, 1995). Given my goal of determining whether phasic alertness modulates residual switch costs, it is reassuring that my experimental methods were suitable for producing robust alerting effects on overall RTs and large residual switch costs in the first place.

9 RESIDUAL SWITH OSTS 325 Table 3 Mean Response Times (in ms) and Error Rates (as Percentages) for Experiment 2 Transition and (ms) status (ms) Response time Error rate availability: Transient Task switch 200 No alert (31) 7.2 (.9) 1,500 No alert (31) 5.2 (.8) 2,000 No alert (31) 3.9 (.6) (30) 4.1 (.5) (32) 5.1 (.6) (31) 4.2 (.6) Task repetition 200 No alert (24) 2.5 (.4) 1,500 No alert (25) 3.4 (.5) 2,000 No alert (25) 2.5 (.5) (25) 2.4 (.5) (26) 2.0 (.3) (26) 2.7 (.5) availability: Persistent Task switch 200 No alert 0 1,030 (23) 7.5 (1.0) 1,500 No alert (24) 6.4 (.8) 2,000 No alert (25) 5.4 (.8) (27) 5.0 (.7) (26) 4.0 (.6) (23) 5.0 (.8) Task repetition 200 No alert (20) 4.2 (.7) 1,500 No alert (22) 4.1 (.6) 2,000 No alert (22) 3.6 (.6) (19) 2.9 (.5) (19) 3.1 (.6) (19) 3.4 (.5) Note. Standard errors of the means appear in parentheses. cue target interval; alert target interval. In contrast, the present results are inconsistent with the competing predictions outlined in the Introduction section. Recall that if residual switch costs reflect interference, then on the basis of findings with various interference effects in the alerting literature (e.g., MacLeod et al., 2010; Nieuwenhuis & De Kleijn, 2013), one would expect larger residual switch costs on alert than on no-alert trials. However, if residual switch costs reflect failures to engage in preparation (De Jong, 2000), then an alerting stimulus might trigger preparatory processing, yielding smaller residual switch costs on alert than on no-alert trials. The present finding of no differences in residual switch costs between alert and no-alert trials (supported by Bayes factors favoring no differences) in two experiments does not support either prediction. A logical interpretation of this finding is that phasic alertness does not affect task switching. One possible explanation for this outcome can be derived from the early onset hypothesis proposed by Nieuwenhuis and De Kleijn (2013) to explain why alerting stimuli speed responding but increase interference effects in the ANT and related tasks. The hypothesis is based on the assumptions that increased phasic alertness shortens target encoding time (enabling response selection to start sooner) and that cognitive control is implemented during response selection. The first assumption is supported by behavioral data showing that alerting stimuli can increase temporal orienting to facilitate target encoding (orrea, Lupiáñez, & Tudela, 2005) and electrophysiological data showing that alerting stimuli have neural effects linked to target onset (Jepma, Wagenmakers, Band, & Nieuwenhuis, 2009). The second assumption is supported by modeling work showing that interference effects in the flanker task were captured best by a model with gradual adjustment of attentional control during response selection (White, Ratcliff, & Starns, 2011). Nieuwenhuis and De Kleijn reported simulation results showing the early onset hypothesis could account for the relationship between phasic alertness and interference effects. A variant of the early onset hypothesis can explain the present findings if one assumes that control processing in task switching has to finish before response selection starts. onsider how control processing might have occurred on alert trials in Experiment 1, in which the and were always 2,000 and 500 ms, respectively. The data from no-alert trials indicate preparation was finished by 1,500 ms, given the similar switch costs obtained at the 1,500-ms and 2,000-ms s. This implies that the control processing involved in preparation had already finished by the time the alerting stimulus was presented on alert trials, providing no opportunity for changes in phasic alertness to influence such processing. Increased phasic alertness still could have facilitated target encoding, though, resulting in a benefit to overall RTs on alert trials. If subjects failed to engage in preparation during the and control processing started after target onset (De Jong, 2000) or if the target triggered additional control processing (Rogers & Monsell, 1995), then faster target encoding would allow control processing to start sooner on alert than on no-alert trials. However, if control processing is responsible for residual switch costs and has to finish before response selection starts, then in- Table 4 Analyses of Variance and ontrasts for Experiment 2 Analysis and effect df Response time Error rate 2 2 F MSE p F MSE p Baseline T 1, , , , T 2, , ontrast a 1, , ing V 1, , T 1, , A 1, , V T 1, , V A 1, , T A 1, , V T A 1, , ing interval T 1, , , , T 2, , Note. T transition; cue target interval; V alert availability; A alert status; alert target interval. a Planned contrast for the interaction, comparing switch costs at the two longest cue target intervals. p.05.

10 326 SHNEIDER creased phasic alertness would affect the onset but not the duration of response selection, thereby failing to modulate residual switch costs. This explanation suggests that the effects of phasic alertness depend on the nature and the timing of control processing in task performance. Weinbach and Henik (2012) provided evidence in support of this point, showing that the typical modulation of interference effects by phasic alertness occurs only for tasks that involve spatial attentional control. This is consistent with neuropsychological work showing that alerting stimuli improve spatial attention in patients with unilateral neglect (Robertson, Mattingley, Rorden, & Driver, 1998). It might also help to explain why Meiran and horev (2005) observed modulations of residual switch costs across s when subjects switched between spatial tasks. An important goal for future research will be to delineate the specific circumstances under which phasic alertness affects performance in task-switching situations and in other contexts that involve cognitive control. References Allport, A., Styles, E. A., & Hsieh, S. (1994). Shifting intentional set: Exploring the dynamic control of tasks. In. Umiltà & M. Moscovitch (Eds.), Attention and Performance XV: onscious and nonconscious information processing (pp ). ambridge, MA: MIT Press. Böckler, A., Alpay, G., & Stürmer, B. (2011). Accessory stimuli affect the emergence of conflict, not conflict control: A Simon-task ERP study. Experimental Psychology, 58, Brown, S., Lehmann,., & Poboka, D. (2006). A critical test of the failure-to-engage theory of task switching. Psychonomic Bulletin & Review, 13, allejas, A., Lupiáñez, J., & Tudela, P. (2004). The three attentional networks: On their independence and interactions. Brain and ognition, 54, orrea, A., Lupiáñez, J., & Tudela, P. (2005). Attentional preparation based on temporal expectancy modulates processing at the perceptual level. Psychonomic Bulletin & Review, 12, De Jong, R. (2000). An intention-activation account of residual switch costs. In S. Monsell & J. Driver (Eds.), Attention and Performance XVIII: ontrol of cognitive processes (pp ). ambridge, MA: MIT Press. Fan, J., Mcandliss, B. D., Sommer, T., Raz, A., & Posner, M. I. (2002). Testing the efficiency and independence of attentional networks. Journal of ognitive Neuroscience, 14, Fischer, R., Plessow, F., & Kiesel, A. (2010). Auditory warning signals affect mechanisms of response selection: Evidence from a Simon task. Experimental Psychology, 57, Fischer, R., Plessow, F., & Kiesel, A. (2012). The effects of alerting signals in action control: Activation of S-R associations or inhibition of executive control processes? Psychological Research, 76, Jepma, M., Wagenmakers, E.-J., Band, G. P. H., & Nieuwenhuis, S. (2009). The effects of accessory stimuli on information processing: Evidence from electrophysiology and a diffusion model analysis. Journal of ognitive Neuroscience, 21, Kiesel, A., Steinhauser, M., Wendt, M., Falkenstein, M., Jost, K., Philipp, A. M., & Koch, I. (2010). ontrol and interference in task switching A review. Psychological Bulletin, 136, Lien, M.-., Ruthruff, E., Remington, R. W., & Johnston, J.. (2005). On the limits of advance preparation for a task switch: Do people prepare all the task some of the time or some of the task all the time? Journal of Experimental Psychology: Human Perception and Performance, 31, Lindsen, J. P., & De Jong, R. (2010). Distinguishing between the partialmapping preparation hypothesis and the failure-to-engage hypothesis of residual switch costs. Journal of Experimental Psychology: Human Perception and Performance, 36, Longman,. S., Lavric, A., Munteanu,., & Monsell, S. (2014). Attentional inertia and delayed orienting of spatial attention in task-switching. Journal of Experimental Psychology: Human Perception and Performance, 40, MacLeod, J. W., Lawrence, M. A., Mconnell, M. M., Eskes, G. A., Klein, R. M., & Shore, D. I. (2010). Appraising the ANT: Psychometric and theoretical considerations of the Attention Network Test. Neuropsychology, 24, Meiran, N. (1996). Reconfiguration of processing mode prior to task performance. Journal of Experimental Psychology: Learning, Memory, and ognition, 22, Meiran, N. (2000). Modeling cognitive control in task-switching. Psychological Research, 63, Meiran, N. (2014). The task-cuing paradigm: A user s guide. In J. A. Grange & G. Houghton (Eds.), Task switching and cognitive control (pp ). New York, NY: Oxford University Press. Meiran, N., & horev, Z. (2005). Phasic alertness and the residual taskswitching cost. Experimental Psychology, 52, Meiran, N., horev, Z., & Sapir, A. (2000). omponent processes in task switching. ognitive Psychology, 41, Monsell, S., & Mizon, G. A. (2006). an the task-cuing paradigm measure an endogenous task-set reconfiguration process? Journal of Experimental Psychology: Human Perception and Performance, 32, Nieuwenhuis, S., & de Kleijn, R. (2013). The impact of alertness on cognitive control. Journal of Experimental Psychology: Human Perception and Performance, 39, Nieuwenhuis, S., & Monsell, S. (2002). Residual costs in task switching: Testing the failure-to-engage hypothesis. Psychonomic Bulletin & Review, 9, Poboka, D., Karayanidis, F., & Heathcote, A. (2014). Extending the failure-to-engage theory of task switch costs. ognitive Psychology, 72, Posner, M. I. (1978). hronometric explorations of mind. Hillsdale, NJ: Erlbaum. Posner, M. I. (2008). Measuring alertness. Annals of the New York Academy of Sciences, 1129, Posner, M. I., & Boies, S. J. (1971). omponents of attention. Psychological Review, 78, Redick, T. S., & Engle, R. W. (2006). Working memory capacity and attention network test performance. Applied ognitive Psychology, 20, Robertson, I. H., Mattingley, J. B., Rorden,., & Driver, J. (1998, September 10). Phasic alerting of neglect patients overcomes their spatial deficit in visual awareness. Nature, 395, Rogers, R. D., & Monsell, S. (1995). osts of a predictable switch between simple cognitive tasks. Journal of Experimental Psychology: General, 124, Rouder, J. N., Speckman, P. L., Sun, D., Morey, R. D., & Iverson, G. (2009). Bayesian t tests for accepting and rejecting the null hypothesis. Psychonomic Bulletin & Review, 16, Schneider, D. W. (2016). Investigating a method for reducing residual switch costs in cued task switching. Memory & ognition, 44, Schneider, D. W., & Logan, G. D. (2011). Task-switching performance with 1:1 and 2:1 cue-task mappings: Not so different after all. Journal of Experimental Psychology: Learning, Memory, and ognition, 37, Vandierendonck, A., Liefooghe, B., & Verbruggen, F. (2010). Task switching: Interplay of reconfiguration and interference control. Psychological Bulletin, 136,