When awareness gets in the way: Reactivation aversion effects resolve the generality/specificity paradox in sensorimotor interference tasks.

Interference tasks combining different distractor types usually find that between-trial adaptations (congruency sequence effects [CSEs]) do not interact with each other, suggesting that sensorimotor control is domain-specific. However, within each trial, different distractor types often do interact, suggesting that control is domain-general. The present study presents a solution to this apparent paradox. In 3 experiments, testing 130 participants in total, we (a) confirm the simultaneous presence of between-trial domain-specific (noninteracting) CSEs and within-trial "domain-general" interactions in a fully factorial hybrid prime-Simon design free of repetition or contingency confounds; (b) demonstrate that the within-trial interaction occurs with supraliminal, but not with subliminal primes; and (c) show that it is disproportionately enlarged in older adults. Our findings suggest that whereas interference (priming and Simon) effects and CSEs reflect direct sensorimotor control, the within-trial interaction does not reflect sensorimotor control but "confusion" at higher-level processing stages (reactivation aversion effect [RAE]). (PsycINFO Database Record (c) 2020 APA, all rights reserved).

These seemingly contradictory results present a problem for any theory of inhibitory cognitive control, and consequently for developmental and clinical research into inhibitory deficits. If there is a general inhibitory control system that can affect distractor-triggered activity from different domains, as indicated by the within-trial interaction, then why are CSEs typically domain-specific? Conversely, if distractor activation and inhibition is mediated by distractor-domain-specific, independent processes (as suggested by the domainspecificity of CSEs), then how do different distractors interact within a trial?

Reactivation Aversion Effect: Beyond Domain-Specific Sensorimotor Processes?
A potential solution to this conundrum was proposed by SRM11. Using a hybrid prime × Simon interference task (i.e., targets preceded by task-irrelevant primes and presented at task-irrelevant locations) with additional neutral primes and neutral target locations, SRM11 (a) replicated the seemingly paradoxical pattern of a within-trial interaction with betweentrial domain specificity, and (b) established that the former was driven by a selective response slowing on trials with a goal-corresponding prime and a goal-noncorresponding target location. (For convenience, we will refer to a prime's goal-correspondence as its 'compatibility' [compatible or incompatible], and to a target-location's goal-correspondence as its 'congruency' [congruent or incongruent] -hence, selective slowing was observed on compatible-incongruent trials.) These trials are uniquely characterized by a sequence of alternating activations in which a just-discarded response has to be reactivated (e.g., a primetriggered right-hand activation, discarded in favor of a location-triggered left-hand activation, then reactivated to execute a right-hand target response 5 ; see Figure 1, bottom left). Having to reactivate a just-discarded response typically results in behavioral costs, as evidenced by numerous negative aftereffects like negative compatibility effects (NCEs) with subliminal (e.g., Eimer & Schlaghecken, 1998) or supraliminal (e.g., Machado, Wyatt, Devine, & Knight, 2007) primes, inhibition of return (IOR; Taylor & Klein, 2000), negative priming (NP; Frings, Schneider, & Fox, 2015), and N-2 repetition costs (Mayr & Keele, 2000). The most parsimonious explanation of the specificity/interaction paradox thus seems to be that responses on prime-compatible location-incongruent trials similarly suffer from reactivation costs (termed by SRM11 a 'reactivation aversion effect', RAE).
Potentially, the RAE contains the solution to the specificity/interaction paradox because, rather than reflecting interacting sensorimotor processes, it might be due to processing difficulties or confusion at a higher-level monitoring or decision stage ("go right… no, go left… hold on, go right after all..."). In support of the 'high-level confusion' account, SRM11 observed RAEs with clearly visible (non-masked) primes, but not with masked, subjectively invisible primes. However, this evidence is not conclusive. At the relatively long (150-ms) prime-target interstimulus interval (ISI) employed in that study, motor processes with masked and non-masked primes are not equivalent: whereas nonmasked prime × Simon trials elicit a sequence of three motor responses (prime-activated, location-activated, target identity-activated), masked prime × Simon trials elicit a series of four, as an additional inhibition process reverses the initial prime activation (Eimer & Schlaghecken, 1998), thus potentially obscuring any RAEs. Because of its potential relevance to the cognitive control literature, the present study therefore aimed to obtain more conclusive evidence about the nature of the RAE, in particular, whether it originates at the level of sensorimotor interference, and thus poses a conceptual problem for theories based on domain-specificity of sensorimotor processes, or whether it originates at subsequent, higherlevel monitoring and decision stages, thereby resolving the apparent specificity/interaction paradox.

Alternative Accounts
First, however, it needs to be demonstrated that this paradox -within-trial interactions in the presence of domain-specific, but not domain-general, CSEs -is a genuine phenomenon, not an artefact resulting from experimental confounds. The issue of such confounds is hotly debated in the CSE literature. On the one hand, CSEs are often assumed to reflect cognitive control in the sense of strategic top-down adjustment of attention toward task-relevant (target) and away from task-irrelevant (distractor) information (e.g., Duthoo et al., 2014aDuthoo et al., , 2014bGratton et al., 1992;Verguts & Notebaert, 2008). However, various accounts have challenged this interpretation, attributing CSEs to bottom-up associative processes instead.
According to the repetition priming account (e.g., Mayr, Awh, & Laurey, 2003) and the feature integration account (e.g., Hommel, Proctor, & Vu, 2004), CSEs merely reflect response time (RT) differences between trials that are identical repetitions of the previous trial and trials that are not. Partly in response to this challenge, various studies have increased the number of S-R pairings in order to be able to implement trial-type repetitions without feature repetitions. However, because each target can only have one goal-corresponding feature, but multiple goal-noncorresponding ones, this produces a problem with balancing correspondence conditions and stimulus features. Contingency learning accounts of CSEs (e.g., Mordkoff, 2012) point out that (a) participants can learn that distractors are informative with regard to the required response and thus respond more quickly on goal-corresponding trials, and that (b) this effect is cumulative, that is, responses to corresponding/highly contingent trials are even faster when the previous trial was also corresponding/highly contingent, resulting in larger interference effects following corresponding/highly contingent trials than following noncorresponding/non-contingent trials. Consequently, in studies with more than two S-R mappings, CSEs might really be 'contingency sequence effects' -again only mimicking cognitive control of sensorimotor processes.

The Present Study
If, as these alternative accounts suggest, CSEs are not 'real' -that is, if they result from experimental confounds rather than reflecting cognitive control processes -then the apparent conflict between domain-specific between-trial and domain-general within-trial effects could similarly be regarded as a quirk driven by specific experimental conditions rather than a fundamental paradox in cognitive control research. Therefore, the initial step of the present study (Experiment 1) was to demonstrate that within-trial interactions in a hybrid sensorimotor interference task do indeed occur together with domain-specific CSEs in a paradigm that is free of any confounds. This can be achieved by splitting an interference task with four stimulus-response assignments (e.g., left-, right-, up-, and down-pointing arrows, requiring corresponding left, right, up, and down responses) into two separate, nonoverlapping tasks (e.g., left/right vs. up/down). Tasks alternate across trials (thus avoiding immediate repetitions), and within each, there are as many corresponding as noncorresponding stimulus combinations (thus preventing contingency learning). Previously, the same logic has been applied to study CSEs for a single distractor domain (e.g., Mayr et al., 2003;see Egner, 2008see Egner, , 2017. To the best of our knowledge, however, the present study is the first to use this design in a hybrid interference task with two independent and fully factorially combined distractor types. The results will therefore not only be of interest to the study of the RAE, but will also provide a new type of evidence regarding the domain-specificity versus domain-generality of CSEs.
Using the same hybrid prime-Simon task -though only in the two-alternative version -Experiments 2 and 3 then investigated whether the RAE is best understood in terms of inhibitory interactions affecting low-level sensorimotor processes, or as reflecting 'confusion' at higher-level action-or decision-monitoring stages. To this purpose, Experiment 2 examined whether stimuli of which participants are not consciously aware would still give rise to RAEs: if they do, then the RAE would need to be explained in terms of low-level sensorimotor processes, but if they do not, this would provide evidence in favor of a more high-level account. Finally, Experiment 3 investigated whether RAEs in older adults are smaller or larger than those in young adults: again, the former would be consistent with a sensorimotor inhibition interpretation of RAEs, the latter with a high-level account.

Experiment 1: Four Response Alternatives
Experiment 1 aimed to (a) replicate SRM11's central finding of a within-trial interaction of prime compatibility and target-location congruency driven by a selective slowing on compatible incongruent trials (RAE) in the absence of domain-general (but presence of domain-specific) CSEs in (b) a paradigm in which CSEs can unequivocally be interpreted as reflecting sensorimotor adaptations to the goal-correspondence or goalnoncorrespondence of the preceding trial's distractor. To achieve this, we combined the leftright prime × Simon task of the original study with a conceptually identical up-down prime × Simon task such that stimulus and response directions (horizontal vs. vertical) alternated trialto-trial. As noted above, such a design not only prevents stimulus and response repetitions, but also avoids any contingencies between stimuli and goal-correspondence conditions. Furthermore, because targets are directional arrows that require a spatially corresponding response, working memory load does not increase from the two-to the four-alternative choice version of the paradigm (i.e., there is still only one thing to remember: "respond to the direction of the target arrow").

Method
Participants. The original SRM11 study had a sample size of 20 young participants.
Following Simonsohn's (2015) advice for investigating null effects in a replication, we aimed for a sample size 2.5 times as large as the original, namely, 50 participants. To achieve this number, we tested 73 volunteers. Data from two participants were lost due to a recording error, and a further 21 participants were excluded due to insufficient numbers of valid trials in at least one of the cells (see below). According to self-report, the remaining 50 participants were 18-22 years old (M = 18.5; SD = 0.9), three were male and 47 were female, six were left-handed and 44 were right-handed, and all had normal or corrected-to-normal vision. All were first-year Psychology students at the University of Warwick who volunteered to take part for course credit. 6 Stimuli. Primes were directional (up, down, left, or right) double arrows or single plus signs, and targets were directional double arrows (see Figures 1 and 2). All stimuli were presented in black on a white background and subtended a visual angle of 0.9° × 0.9°. Primes always appeared at the center of the screen. Targets appeared 6° to the left or right of the center or 4° above or below the center (in line with the steeper decline of contrast sensitivity along the vertical than the horizontal axis [e.g., Rijsdijk, Kroon, & van der Wildt, 1980], and with pilot testing indicating that these distances produced similar RTs for horizontal and vertical targets).
Procedure. Testing took place in individual sessions. Participants were seated in a dimly lit, sound-attenuated cubicle approximately 100 cm in front of a CRT monitor. Each trial began with the presentation of a prime for 33 ms in the center of the screen, followed 150 ms later by a target presented for 100 ms in one of the four possible target locations. The inter-trial interval (ITI) was 1317 ms. The experiment consisted of alternating horizontal 6 All experiments reported in this paper were approved by the University of Warwick's Humanities and Social Sciences Research Ethics Committee. Before signing up, and again upon arrival at the lab, participants were warned that the experiment contained rapidly flashing high-contrast stimuli and were explicitly told not to take part if they were photosensitive. Participants were then given a demonstration of the procedure before providing written informed consent. Across all experiments reported here, no participant initially reported to be photosensitive; after seeing the demonstration, one participant stated that the flashing stimuli felt uncomfortable and consequently aborted the experiment. For all participants taking part for course credit, alternative means to gain credits were available. trials and vertical trials (see Figure 2). On horizontal trials, all stimulus and response dimensions were horizontally aligned: primes were either left-or right-pointing arrows or a plus sign, target arrows appeared on the left or right of fixation and pointed either to the left or to the right, and participants had to give a corresponding left or right response. On vertical trials, all stimulus and response dimensions were vertically aligned: primes were either up-or down-pointing arrows or a plus sign, target arrows appeared above or below fixation and pointed either up or down, and participants had to give a corresponding up or down response.
On prime-compatible trials, prime and target arrows pointed in the same direction, on prime-incompatible trials, they pointed in opposite directions, and on prime-neutral trials, the prime was a plus sign. On location-congruent trials, the location and the pointing direction of the target arrow matched (e.g., a down-pointing arrow appearing below fixation), whereas on location-incongruent trials, they were in opposition (e.g., a down-pointing arrow appearing above fixation). Within each orientation (horizontal, vertical), each trial type (3 prime compatibilities, 2 location congruencies, 2 target identities) were equiprobable and completely randomized. Horizontal and vertical trials strictly alternated, so that no primes (except neutral ones), locations, targets, or responses could repeat from one trial to the next (see Figure 2). Order of trial types was randomized to ensure that all types of transitions (e.g., horizontal-compatible-congruent-followed-by-vertical-compatible-congruent, verticalcompatible-congruent-followed-by-horizontal-incompatible-incongruent, etc.) were approximately equiprobable. Excluding the first trial of each block, each transition type appeared on average 15.9 times (lowest average: 14.7, highest average: 16.9).
Participants were instructed to ignore the prime and the target location, and to respond to the direction of the target arrow as quickly and accurately as possible by making a corresponding key-press with their right index or middle finger (whichever they preferred) on the number key pad of a standard QWERTY keyboard. The center key (5) was the designated 'home' key, and responses were made by moving the finger from the home key to the key that corresponded to the arrow direction (i.e., to the 8 [above the center] for an up-ward pointing arrow, to the 2 [below] for a down-ward pointing one, and correspondingly to the 4 [left] or to the 6 [right] for left-or right-pointing target arrows). After pressing the response key, they were to return immediately to the home key in preparation for the next response (key-presses on the home key were neither required nor recorded).
In order to familiarize participants with task requirements, two practice blocks were administered. In the first (16 trials), stimulus presentation was slowed to half of the normal speed; in the second (32 trials), stimulus presentation was the same as in the experiment. If participants produced more than 25% errors in the second practice block, were exceedingly slow (average RTs > 1000 ms), or otherwise indicated that they struggled with the task, the practice block was repeated. During practice, the experimenter remained in the cubicle to ensure that participants understood and complied with task instructions.
Following practice, the experiment consisted of eight experimental blocks, each containing 144 trials. Participants started the next block when they felt ready to do so. They were encouraged to leave the cubicle for a break after the fourth block, during which they were offered tea/coffee and biscuits. Including instruction, practice, breaks, and debriefing, the experiment took approximately 50 min.
Data analyses. Data were sorted according to orientation (horizontal, vertical), prime compatibility (compatible, neutral, incompatible), location congruency (congruent, incongruent), previous trial's prime compatibility, and previous trial's location congruency, resulting in 72 trial types. For RT analyses, only correct-response trials following a correct response were taken into account (93.9% of all correct responses). Participants who had fewer than five valid trials in any condition were excluded from analysis; no further data trimming procedures were employed. For error rate analyses, only errors following a correct response were taken into account (89.9% of all errors). Mean RTs were analyzed using a 2 (orientation) × 3 (prime compatibility) × 2 (location congruency) × 3 (previous prime compatibility) × 2 (previous location congruency) within-subject ANOVA. For the sake of brevity, we only report within-trial error rate analyses here, based on a 2 (orientation) × 3 (prime compatibility) × 2 (location congruency) within-subject ANOVA. Greenhouse-Geisser corrections were applied where appropriate. For a full analysis of sequential (as well as response-latency based) error rate effects, the reader is referred to Supplemental Materials.
For follow-up analyses, priming effects were calculated as prime benefits (neutral minus compatible) and prime costs (incompatible minus neutral), separately for each orientation (2), location congruency (2), and -for RTs only -previous trial type (6), resulting in 24 RT benefits and 24 RT costs, and in 4 error-rate benefits and 4 error-rate costs.
Correspondingly, Simon effects were calculated as incongruent minus congruent, separately for each orientation (2), prime compatibility (3), and -for RTs only -previous trial type (6), resulting in 36 RT Simon effects and 6 error-rate Simon effects. RAEs were calculated as Simon effects on prime-compatible trials minus Simon effects on prime-neutral trials, separately for each orientation and -for RTs only -previous trial type, producing 12 RT RAE and 2 error-rate RAE values. Repeated-measures ANOVAs and t-tests were used to analyze these effects (for 95% confidence intervals for effects in all three experiments, see Supplemental Materials).

Results
Participants produced an average of 14.2 valid correct responses and an average of 0.7 valid errors per condition. Figure 3) to 5.31). That is, depending on the exact test parameters and comparisons, priming and Simon effects are estimated to be approximately 2 to 5 times more likely to be unaffected by the previous trial's opposite distractor type than to be affected by it.

Discussion
Experiment 1 produced two main results. First, it successfully replicated the RAE pattern reported in SRM11: within a given trial, prime compatibility and target location congruency interacted, and whereas for error rates, this interaction was primarily driven by exaggerated error rates on prime-incompatible location-incongruent trials, for RTs, it was primarily -for horizontal trials even exclusively -driven by a slowing of responses to primecompatible location-incongruent targets. Second, the experiment confirmed the existence of domain-specific and absence of domain-general CSEs in a paradigm free of any repetition or contingency confounds. Consequently, confound models cannot account for the present results, nor can models based on the notion of abstract, general-purpose conflict adaptation (e.g., Duthoo et al., 2014aDuthoo et al., , 2014bGratton et al., 1992;Verguts & Notebaert, 2008), at least not without making substantial additional assumptions. Instead, this pattern strongly suggests that CSEs do indeed reflect trial-by-trial adaptations to a distractor's goalcorrespondence or goal-noncorrespondence (in contrast to confound accounts), and that these adaptations occur through up-and down-regulations within distractor-specific sensorimotor pathways rather than reflecting a top-down controlled enhancement of target-related processing (in contrast to conflict adaptation accounts).
At first glance, the claim that distractor-related activation and inhibition processes occur within separate sensorimotor pathways seems inconsistent with the finding of within-trial distractor interactions, which might be taken as evidence for domain-general inhibitory control processes (e.g., Boy et al., 2010;Feldman & Freitas, 2019;Frühholz et al., 2011;Rey-Mermet & Gade, 2016) or cross-talk between sensorimotor channels (e.g., Stoffels & van der Molen, 1988). However, the RAE pattern of these interactions -that is, the selective slowing of responses on prime-compatible location-incongruent trials -suggests a different interpretation. As outlined in the Introduction, these trials are uniquely characterized by the need to reactivate a just-discarded response activation (e.g., right prime-induced activation → left location-induced activation → right target-related activation; see Figure 1). Given the number and variety of tasks producing negative aftereffects of just-discarded activity, such reactivation seems inherently difficult and behaviorally costly. However, the reasons for these costs are not necessarily clear, and might indeed differ between different tasks or task features (see, e.g., Frings et al., 2015;Kiesel et al., 2010;and Taylor & Klein, 2000, for discussions of various mechanisms of NP, N-2 repetition costs, the NCE, and IOR, respectively).
For the RAE, four -not necessarily mutually exclusive -explanations seem intuitively plausible. The first one is trivial: given the disproportionately enlarged number of errors on incompatible-incongruent trials, might the RT-RAE merely be the result of a speed-accuracy trade-off? This seems unlikely. First, as noted above, the prime × Simon interaction on error rates is likely to be due to floor effects for compatible and neutral primes. Second, it should be emphasized that the overall data pattern does not indicate a speed-accuracy trade-off, as higher error rates coincide with slower responses and lower error rates with faster ones. In particular, third, the critical prime-compatible location-incongruent condition, assumed to be selectively slowed, shows error rates that are numerically (though not statistically) higher than those in the prime-neutral location-incongruent condition, that is, the opposite of a speed-accuracy trade-off pattern. Finally, it should be noted that the vast majority of errors occur within the fastest 20% of all responses, yet the RAE emerges only at longer latencies and is noticeably absent within the first RT quintile (see Supplemental Materials for details).
The three remaining, theoretically more interesting, explanations for the RAE are motor inhibition, visual attention, and high-level response scheduling confusion. According to the inhibition hypothesis, the initial, prime-triggered motor activation will be inhibited by the opposite-hand location-triggered motor activation, and as a result will take longer to become sufficiently reactivated for response execution. This hypothesis has the advantage of being simple and intuitively plausible. However, because it assumes an interaction between distractor processes, it cannot easily be reconciled with the CSE results. More importantly, the hypothesis predicts not only reduced prime benefits on location-incongruent trials (i.e., the RAE), but also reduced prime costs on location-congruent trials. On these trials, primeand location-triggered activations are in opposition when primes are incompatible (e.g., left prime → right location → right target). If opposing location activation ('right') inhibits the initial prime-triggered activity ('left'), then this activity should have a correspondingly reduced impact on overt performance ('right') than on trials where prime-and locationtriggered activity are in accordance (i.e., costs caused by an incompatible prime should be smaller on location-congruent than on location-incongruent trials). Clearly, this was not the case in the present experiment, nor in the SRM11 study.
The visual attention hypothesis faces a similar issue. This hypothesis is based on the possibility that directional arrow primes might automatically shift the focus of visual attention to the indicated location, and that relative to an 'unfocussed' condition, target identification will be facilitated for targets appearing inside this focus, and will be delayed for targets appearing on the opposite side (see Supplemental Materials for a schematic illustration). Consequently, Simon effects will be enlarged on compatible relative to neutral trials (i.e., the RAE), because attention-related costs and benefits coincide with location-related costs and benefits. By the same token, however, Simon effects should show a corresponding reduction on incompatible relative to neutral trials, because on these trials, attention-related and location-related costs and benefits are in opposition. Again, this was not the case. Although vertical Simon effects were smaller on incompatible than on neutral trials, this reduction was small relative to the Simon-effect increase on compatible trials (16 vs. 33 ms), and for horizontal Simon effects, no reduction at all occurred. (Moreover, recall that for error rates, the opposite pattern -enlarged rather than reduced Simon effects on primeincompatible trials -was observed.) In sum, it seems that neither direct interactions of distractor-related inhibitory processes nor prime-induced attentional modulation of target identification can account for the RAE, because both predict mirror-symmetrical effects rather than an effect on only one specific trial type (i.e., not only reduced prime benefits on location-incongruent trials, but correspondingly reduced prime costs on congruent ones; not only enlarged Simon effects on prime-compatible trials, but correspondingly reduced Simon effects on incompatible ones).
We therefore propose that the unique feature of this trial type -the back-and-forth alternation of motor responses associated with prime, target location, and target identity -is the cause of the RAE. Specifically, we propose that this back and forth produces confusion at higher-level (meta-cognitive) processing stages, which results in delayed response execution. This account is based on the assumption that the initial prime-triggered activation registers as incorrect when the target location indicates the opposite response, which then in turn registers as incorrect when the target identity does not correspond with it. Overt response errors are known to be aversive and to produce a reflexive withdrawal response (Hajcak & Foti, 2008;Hochman, Milman, & Tal, 2017). In the case of compatible-incongruent trials, there is no overt error, but all available response options have already proven themselves to be problematic in the face of new evidence. We believe that it is possible that this produces an 'aversion' to return to these responses, similar to the aversion produced by overt errors, which manifests as a delay in initiating response activation. In the remainder of this paper, we will explore this possibility in more detail.

Experiment 2: Conscious Awareness
Initial evidence that the RAE might reflect a higher-level process was obtained in SRM11, as only participants who performed the hybrid prime × Simon task with non-masked (visible) primes produced the RAE, whereas the pattern was absent in participants who performed the task with masked (subliminal) primes. However, although this pattern suggests that conscious awareness of the prime is a necessary precondition for the RAE to occur, this evidence is not conclusive. In those experiments, prime-target ISI was 150 ms (as in Experiment 1 above), and at this interval, masked primes do not produce normal priming effects, but reversed priming effects (NCE, with performance benefits for incompatible and costs for compatible trials relative to neutral trials). The NCE reflects an inhibition of the initially primed response and corresponding disinhibition of the non-primed response (Eimer & Schlaghecken, 1998;Schlaghecken, Bowman, & Eimer, 2006). Consequently, instead of a sequence of three motor activations (prime → location → target identity), as in the nonmasked prime × Simon task, trials in the masked-prime version elicit a sequence of four (prime → reversal → location → target identity). It is not a priori clear where in this sequence a reactivation-aversion-like effect might occur, if at all: in principle, it might affect the third step (location-triggered activation), the fourth (target-related activation), or both. As a result, it seems conceivable that the RAE was not truly absent in this version of the masked prime × Simon task, but was merely obscured by the intervening reversal phase.
This problem can be avoided by reducing the prime-target ISI. If targets follow the primes immediately, then there is no time for the reversal phase to develop, and normal priming effects occur (Eimer, 1999;Schlaghecken & Eimer, 1997. The present Experiment 2 therefore used a masked prime × Simon task with a prime-target ISI of 0 ms to investigate the role of prime awareness for the RAE. In particular, we aimed to directly compare performance with subliminal (not consciously available) versus supraliminal primes.
Primes were therefore presented either for 16 ms or for 33 ms prior to mask onset (based on pilot data indicating that with the present set-up, as described below, prime identification performance was near chance with the former and clearly above chance with the latter duration). Like other subliminal perception tasks, masked-prime tasks are known to be affected by exposure effects (i.e., experiencing supraliminal primes can change participants' threshold of awareness; e.g., Lamy, Carmel, & Peremen, 2017;Schlaghecken, Blagrove, & Maylor, 2008). To prevent such effects in the present study, prime duration was varied between rather than within participants.
With a 0-ms interval between prime offset and target onset, trials in a masked prime × Simon task elicit the same sequence of three response activations as do trials in the nonmasked version of the task (i.e., prime-triggered activation → location-triggered activation → identity-based activation); the reversal phase developing during longer prime-target ISIs and resulting in the NCE is absent in this design. Consequently, if SRM11 failed to obtain an RAE with masked primes merely because it was obscured by the reversal phase, then in the present experiment, RAEs should occur with both 33-ms (supraliminal) and 16-ms (subliminal) primes. Conversely, if the RAE is a high-level phenomenon that requires conscious awareness of the stimuli involved, then only the former but not the latter duration should produce RAEs. 8

Method
Participants. Forty-five volunteers from an opportunity sample participated without payment. Three participants were excluded from further analysis due to high overall error rates (> 15% valid errors 9 ), and one participant was excluded because of failure to follow task instructions, leaving 20 participants in the 16-ms group and 21 in the 33-ms group. According to self-report, participants were 18-32 years old (M = 20.8, SD = 2.3), eight were male, all but five were right-handed, and all had normal or corrected-to-normal vision.
Stimuli. Primes and targets were left-or right-pointing arrows (< and >), and a plus sign (+) served as an additional neutral prime stimulus, all subtending a visual angle of approximately 2.0° × 1.5° at a viewing distance of approximately 60 cm. Masks were constructed on the basis of a virtual 9 × 9 matrix, randomly filled with overlapping horizontal, vertical and oblique lines of different lengths, resulting in a roughly rectangular array of approximately 5° × 3°. A new random mask was created on each trial to avoid perceptual learning of the mask and correspondingly increased prime identification (Schlaghecken et al., 2008).
Procedure. The experiment consisted of a hybrid prime × Simon task followed by a prime identification task. In the former (see Figure 5, left panel), trials began with a centrally presented prime, followed immediately by a centrally presented mask and a target appearing 11° to the left or right of the center. Prime duration was 16 ms for one group of participants and 33 ms for the other; masks and targets were presented for 100 ms. The ITI was 1460 ms.
In the identification task (see Figure 5, right panel), each trial began with a left-or right-pointing arrow presented randomly and equiprobably for 16, 33, 50, 66, 83, 100, or 116 ms at the center of the screen and immediately followed by a 100-ms mask, but no target. 9 As in Experiment 1, only errors following a correct response were regarded as valid.
Participants were instructed to indicate the direction of the arrow with corresponding left-or right-hand key-presses, and were encouraged to 'just guess' in case they failed to consciously identify the stimulus. However, to minimize fast-guess responses (which are most likely to be affected by direct motor priming and therefore are not a useful measure of participants' conscious prime perception; e.g., , a question mark ('?') was presented 100 ms after mask offset below the screen center, and participants were asked not to respond before it had appeared. The question mark remained until a response had been given. The next trial then started 2000 ms later.
The prime × Simon task consisted of two practice and 10 experimental blocks, containing 60 trials each; the identification task comprised two practice and two experimental blocks of 70 trials each. During practice, the experimenter stayed in the cubicle to ensure that participants had understood and were able to follow the task instructions. Throughout the experiment, participants initiated each block whenever they felt ready to do so. Including instruction, practice, breaks, and debriefing, the experiment took approximately 30 min.
Data analysis. As in Experiment 1, only correct RTs and error rates following a correct response were analyzed in the prime × Simon task. Data were grouped into six trial types (3 prime compatibilities × 2 location congruencies) and analyzed using mixed ANOVAs with the within-subject factors prime compatibility and location congruency and the betweensubject factor prime duration (16/33 ms). Greenhouse-Geisser corrections were applied where appropriate. Interference effects and RAEs were calculated as in Experiment 1 and were analyzed using t-tests. For the prime identification task, the percentage of correct responses was calculated for each prime duration separately and analyzed with a two-way mixed ANOVA. Follow-up analyses were conducted on 16-and 33-ms primes, using both classical and Bayesian t-tests. .717 (see Figure 6). There was no main effect of group, nor a Group × Duration interaction, However, priming effects were larger in the 33-ms group than in the 16-ms group, F(2, 78) = 12.59, p < .001, MSE = 207.71, η 2 p = .244, although follow-up ANOVAs, conducted for each group separately, confirmed that they were significant in both groups, ps < .001.

Results
Most importantly, as can be seen in Figure 7, the 33-ms group produced an RAE, whereas the 16-ms group did not, reflected in a significant Prime × Location × Group Arguably, this difference -the presence of the RAE with 33-ms primes and its absence with 16-ms primes -might not be due to the difference in these primes' availability to conscious awareness, but rather to the fact that priming effects were significantly larger with 33-than with 16-ms primes. To test this possibility, we conducted a median split of priming effects for each group separately (based on priming effects on location-congruent trials, to avoid confounding priming effects and RAEs), and calculated RAEs (Simon effect on compatible trials minus Simon effect on neutral trials) for each sub-group separately. The results are displayed in Figure 8, which clearly shows that the magnitude of priming effects cannot account for the presence or absence of the RAE: 'large' priming effects in the 16-ms prime group were virtually identical in size to 'small' priming effects in the 33-ms prime group, yet only the latter produced the RAE (highlighted by the dashed oval in Figure 8).
These results strongly suggest that it is prime visibility itself, not magnitude of priming effects, that determines the presence or absence of RAEs.

Discussion
Experiment 2 investigated whether the RAE pattern -an enlarged Simon effect on prime-compatible trials relative to prime-neutral and prime-incompatible trials, or correspondingly, reduced prime benefits on location-incongruent relative to congruent trialsoccurs when primes are successfully masked and thus unavailable to conscious awareness. If it does, then this would suggest that the RAE reflects relatively low-level visual-attentional or motor inhibitory processes directly tied to prime processing. Conversely, if no RAE occurs with subliminal primes (but does occur with similarly masked but supraliminal ones), then this would suggest that it reflects subsequent higher-level or meta-cognitive processes. The present results support the second proposition by demonstrating that the RAE is, indeed, linked to whether or not primes are successfully masked: whereas supraliminal primes produced a robust RAE, this pattern was absent for subliminal primes (in fact, the latter produced a numerically, though not statistically, reversed pattern).
Although this result fits with the prediction of the 'meta-cognitive confusion' or 'aversion' hypothesis, some caveats are in order. First, prime identification performance in a subsequent forced-choice test can only provide an approximation of participants' conscious awareness of the primes during the prime × Simon task. In principle, forced-choice identification might underestimate the level of awareness during the critical task (i.e., participants in the 16-ms group might have been aware of the primes at the time even if they were subsequently unable to successfully identify them). In practice, however, it seems more likely that it overestimates the level of awareness, as during the identification task, participants focus on the primes and actively try to identify them, whereas during the prime × Simon task, they focus on the -subsequently presented and laterally displaced -targets.
Consequently, during the prime × Simon task, 33-ms masked primes might have been less available to conscious awareness than the prime identification performance suggests. It is therefore important to distinguish between the participants' subjective experience of the prime and the stimulus conditions that contribute to this experience: the present experiment can only directly demonstrate a functional role of the latter, not the former, for the RAE.
A second issue, related to the first, is that there is no generally accepted definition of 'consciousness' and related concepts, and consequently few firmly established facts about its function or underlying mechanisms (e.g., Zeman & Coebergh, 2013 conclusively. In order to obtain converging evidence, we therefore turned to a fundamentally different approach: the investigation of individual differences.

Experiment 3: Aging
After the third decade, the nervous system deteriorates both anatomically and biochemically (for reviews, see Mohan, Mather, Thalamuthu, Baune, & Sachdev, 2016;Rossini, Rossi, Babiloni, & Polich, 2007). Correspondingly, many sensory, cognitive, and motor processes decline from young adulthood to old age (for overviews, see Hofer, Berg, & Era, 2003;Roberts & Allen, 2016). One of the challenges for aging research is therefore to establish whether, against the background of such general decline, any specific functions are 10 Sequential analysis of the present experiment -analogous to Experiment 1 -failed to show prime-triggered CSEs in either group, both Fs < 1.2, both ps > .3. However, due to the small number of valid trials in each condition (12 on average) and the comparatively large variation of trial numbers across conditions (means ranging from 9 to 17), reliable statistical results would require substantially greater numbers of participants (i.e., approximately 2.5 times as many participants in each group, corresponding to Experiment 1). particularly impaired. Perhaps surprisingly, the control of sensorimotor interference appears to remain mostly intact across the lifespan: once older adults' generally slower processing (see Cerella, 1985;Salthouse, 1996) has been taken into account, most interference effects tend be of similar magnitude in young and older adults. Importantly, such age-equivalence occurs not only for simple sensorimotor interference effects like Stroop or flanker effects.
Rather, it also occurs for so-called inhibitory aftereffects, that is, for effects that originate from the reactivation of an automatically activated but then discarded response such as antisaccade delays, NP, IOR, and local switch and N-2 repetition costs (for reviews and metaanalyses, see, e.g., Gamboz, Russo, & Fox, 2002;Rey-Mermet, Gade, & Oberauer, 2018;Verhaeghen, 2011Verhaeghen, , 2014, and spatial NP (Buckolz, Lok, Kajaste, Edgar, & Khan, 2015). Moreover, in tasks in which older adults do appear disproportionately disadvantaged, the age-related deficit takes the form of delayed or reduced aftereffects, such as with IOR onset (Erel & Levy, 2016) and the NCE with subliminal primes , 2012Schlaghecken & Maylor, 2005). In short, in older adults, low-level sensorimotor processes that resemble the back-and-forth of compatible-incongruent trials in the hybrid prime × Simon tasks appear to be mostly intact (or rather, not specifically impaired), and where specific age-related deficits do occur, they result in delayed or reduced sensorimotor interference effects. Thus if the RAE similarly reflects sensorimotor interference, it should correspondingly be age-equivalent or reduced in older compared to young adults.
In contrast, the efficiency of high-level monitoring and decision processes declines with age over and above the effects of general slowing. This shows in older adults' disproportionately enlarged dual-task costs and global (as opposed to local) switch costs, both attributed to an age-related specific difficulty in scheduling two tasks at once or switching the focus of attention in working memory (e.g., Kray & Lindenberger, 2000;Mayr, 2001;Reimers & Maylor, 2005; for reviews and meta-analyses see Jaroslawska & Rhodes, 2019;Verhaeghen, 2011Verhaeghen, , 2014Wasylyshyn, Verhaeghen, & Sliwinski, 2011). It also shows in the types of errors older adults make when interpreting so-called garden-path sentences (Malyutina & den Ouden, 2016) and their disproportionate difficulties in parsing sentences with double negatives (e.g., Yoon et al., 2016), both requiring 'mental backtracking' and reanalyzing already-processed information. Consequently, if the RAE reflects similar highlevel confusion, we should expect it to be disproportionately enlarged in older compared to young adults.

Method
Participants. Twenty-three young and 28 older adults participated in Experiment 3. To obtain groups with similar performance levels, we excluded participants with more than 15% errors and/or fewer than 66% valid correct trials, leaving a final sample of 21 young adults (18-30 years; 7 male; no left-handers) and 19 older adults (64-82 years; 8 male; 3 lefthanders). 11 The former were students from the University of Warwick who took part either for course credit or £6 ($7); the latter were members of a volunteer panel recruited through local newspapers and advertisements who were paid £10 ($12) as a contribution toward their travel expenses. Background cognitive measures were already available for older participants from earlier testing sessions, and were collected from young participants immediately after completing the present experiment. Demographic and background data are listed in Table 4.
As expected (cf. Salthouse, 1991), young adults had poorer vocabulary but better speed and visual acuity than did older adults.
Stimuli. As in SRM11, primes and targets were double arrows (<< and >>), subtending a visual angle of 2.0° × 0.9°. For sufficient numbers of valid trials per condition, in particular 11 We also analyzed the data using more lenient exclusion criteria (error-rate cut-offs of 20%, leaving 23 young and 21 older adults, and of 30%, leaving 23 young and 25 older adults), which produced qualitatively similar though noisier patterns of results.
in older adults, no neutral primes were presented in this experiment. To keep stimulus conditions as similar as possible to another task carried out in the same session (see below), an empty rectangular 'frame' measuring 2.6° × 1.4° was presented between the prime and the target. Primes and frames appeared at the center of the screen, and targets appeared 5.7° to the left or right of the center. Viewing distance was approximately 100 cm.
Procedure. Participants were seated in a comfortable chair with response buttons mounted onto the arm rests. The experimental session lasted approximately one hour, during which participants carried out two different hybrid prime × Simon tasks (with task order counter-balanced across participants). One task addressed the issue of age-related dedifferentiation and de-automatization of inhibitory control and has been published elsewhere (see Maylor et al., 2011). The other, reported here, investigated age-related changes in the RAE. It consisted of six experimental blocks, each containing 72 trials, preceded by one practice block of 32 trials. To help particularly older participants to maintain central fixation, a central 250-ms fixation dot (0.1° × 0.1°) was presented before each prime, followed by a 650-ms blank screen. Next, a prime was presented for 33 ms, followed by the frame for 100 ms, a 50-ms blank, and finally a target, presented for 100 ms. The ITI between target offset and the next fixation dot was 1300 ms. Participants were instructed to maintain central eye fixation throughout, and to respond as quickly and accurately as possible to the direction of the target arrows (i.e., a left-hand key-press to left-pointing arrows, and a right-hand keypress to right-pointing arrows), regardless of the target's location on the screen. All trial types (2 prime compatibilities × 2 target location congruencies × 2 responses) were equiprobable and randomized within each block.

Data analyses.
As before, only RTs and errors following a correct response were analyzed (93.1% of all correct responses, 88.3% of all errors). Data were grouped into four trial types (2 prime compatibilities × 2 location congruencies) and analyzed using mixed ANOVAs with the within-subject factors prime compatibility and location congruency and the between-subject factor age group (young, older). In order to account for the effect of general age-related slowing, RT priming and Simon effects were calculated not as differences, but as ratios 12 ; the RAE was calculated as the difference between Simon effects on prime-compatible and prime-incompatible trials. Follow-up analysis of interference effects and RAEs were conducted using t-tests.

Results
RTs and error rates are depicted in Figure 9, and the corresponding ANOVA results are listed in Table 5. Older adults responded overall 156 ms slower than did young adults but produced virtually identical error rates (both 5.8%). Responses were 38 ms faster and 3.3% points more accurate for compatible than for incompatible primes, and 83 ms faster and 5.6% points more accurate for congruent than for incongruent target locations. Whereas RT priming effects were similar across age groups, RT Simon effects were significantly larger in older than in young adults (106 vs. 60 ms). Prime compatibility and location congruency interacted for both RTs and error rates, and again, as in the previous two experiments, did so in opposite directions: for RTs, the Simon effect was larger with compatible than with incompatible primes (110 vs. 56ms; RAE), whereas for error rates, the reverse was true (4.0% vs. 7.3%). For RTs, but not for error rates, the Prime × Location interaction was modulated by age group, as older adults' RAE (i.e., compatible minus incompatible Simon effect) was more than twice as large as that of young adults (78 vs. 30 ms). 12 For instance, assume that older adults are generally 1.5 times slower than young adults: Task A, taking young adults 300 ms, will take older adults 450 ms, and Task B, taking young adults 400 ms, will take older adults 600 ms. As a result, young adults will have a B-A effect of 100 ms, older adults one of 150 ms -an increase in magnitude that seems to suggest a specific age-related deficit in B-processing where in fact none exists. Using RT ratios avoids this problem (here: 400/300 = 1.3, 600/450 = 1.3, correctly indicating that there is no additional B-processing deficit in older adults; see, e.g., Verhaeghen, 2011). An alternative to calculating proportional effects/ratios is to transform the data using z-scores (see Faust, Balota, Spieler, & Ferraro, 1999;Hedge, Powell, & Sumner, 2018). These two methods led to identical conclusions; we report ratios here but include z-scores in Supplemental Materials.
Given the large overall RT difference between young and older participants, it is essential to investigate whether these interactions remain after general slowing has been taken into account, that is, when effects are calculated as RT ratios. As can be seen in Figure 10, neither priming nor Simon effects differed between young and older adults on trials that were not affected by the RAE (i.e., priming effects on location-congruent trials and Simon effects on prime-incompatible trials, both ts(38) < 1). In contrast, priming effects on incongruent trials were significantly reduced, and Simon effects on prime-compatible trials were significantly enlarged in older compared to young adults, t(38) = 2.71, p = .010, and t(38) = 2.99, p = .005, respectively. Most importantly, after taking general slowing into account, the RAE was still almost twice as large in older than in young participants (0.172 vs. 0.095; t(38) = 3.23, p = .003). 13

Discussion
The aim of Experiment 3 was to directly contrast the sensorimotor interference and high-level confusion hypotheses of the RAE by investigating whether older adults produce an age-equivalent (or even reduced) RAE or a disproportionally enlarged one. The results were clear-cut: far from being age-equivalent, older adults' RAEs were almost double in size compared to those of young adults, even though (a) participants had been selected to ensure similar levels of overall performance, (b) age-related slowing had been partialled out, and (c) interference effects on RAE-unrelated trials (priming effects on location-congruent trials, Simon effects on prime-incompatible trials) were indistinguishable between the two age 13 We also conducted a CSE analysis on ratio effects, which confirmed the presence of domain-specific effects. An interaction between the previous and current trial's location-congruency for priming effects, suggesting at least partial domain-general CSEs, was in fact accounted for by RT differences between these trials: time-course analysis confirmed that priming effects had a negative-going latency slope (decreasing substantially with increasing response latency), and that the interaction was entirely accounted for by the fact that congruent trials were faster (and correspondingly had larger priming effects) following congruent than following incongruent trials, whereas the reverse was true for incongruent trials. CSEs did not differ between age groups but, as with Experiment 2, the numbers of valid trials were rather small for at least some conditions and participants. Consequently, these patterns should be considered suggestive rather than conclusive.
groups. This pattern of results matches the predictions of the high-level confusion hypothesis, and conflicts with the predictions of the sensorimotor interference hypothesis. Therefore, it strongly supports the notion that the RAE does not originate at direct sensorimotor processing stages, but reflects processing difficulties at a higher or subsequent level.
The finding of age-equivalent priming and Simon effects in both RTs and error rates is in line with the notion that aging is not associated with a general executive control deficit.

General Discussion
The present study set out to investigate an apparent paradox in cognitive control. On the one hand, studies of between-trial CSEs typically suggest that separate, domain-specific mechanisms control the activation and inhibition processes associated with different types of distractors (Akçay & Hazeltine, 2011;Egner et al., 2007;Forster & Cho, 2014;Funes et al., 2010;Kim et al., 2012;SRM11;Wendt et al., 2006; for reviews, see Braem et al., 2014;Duthoo et al., 2014aDuthoo et al., , 2014bEgner, 2008Egner, , 2017 In particular, the results suggest that this improvement depends on 'conscious' or high-level mechanisms. In a masked prime task, the prime is subliminal and not available to conscious perception. Consequently, participants are not aware of the need to inhibit a response to it (and automatic inhibition of subliminally triggered motor activations is indeed impaired in older adults; see Maylor et al., 2011;Schlaghecken, Birak et al., 2011, 2012Schlaghecken & Maylor, 2005). In the present experiment, in contrast, as well as in the Kubo-Kawai and Kawai (2010) study, the relevant stimuli (primes and nogo-targets, respectively) were fully visible, and participants were in fact explicitly instructed not to respond to them. It is tempting to argue that such intentional inhibition, mediated by dorsal fronto-median cortex (e.g., Ficarella & Battelli, 2017), provides a top-down 'boost' to lowlevel sensorimotor inhibition, and that this might be particularly true for older adults, who have been shown to 'shift' activation from posterior to more anterior cortical areas, especially under increased cognitive load (e.g., Ansado, Monchi, Ennabil, Faure, & Joanette, 2012). However, there is as yet not enough evidence available in support of this conjecture, and further research is needed to explore this issue more directly.
that RTs show under-additive within-trial interactions between distractor domains (see Table   2), suggesting shared control mechanisms. The question of whether control mechanisms are domain-specific or domain-general is of central importance in the field (e.g., Braem et al., 2014;Duthoo et al., 2014aDuthoo et al., , 2014bEgner, 2008Egner, , 2017, affecting not only theories of cognitive control, but also the understanding of normal versus impaired control processes in developmental and clinical psychology. It is therefore surprising that relatively few studies have addressed within-and between-trial interactions simultaneously (see the two right-most columns of Tables 1 and 2). In particular, as far as we are aware, no study to date has addressed this issue in a hybrid task free of repetition and contingency confounds.
Experiment 1 confirmed that the 'paradoxical' interference pattern occurs within a single, confound-free design. In a hybrid prime × Simon task, trial-by-trial modulations of interference effects were domain-specific (i.e., a trial's prime compatibility affected priming effects, but not Simon effects, on the next trial, whereas a trial's target-location congruency affected subsequent Simon but not subsequent priming effects), suggesting that distractorrelated processing and control occurs within separate, domain-specific sensorimotor pathways. At the same time, interference effects interacted within each trial, such that Simon effects were larger with compatible than with neutral or incompatible primes, or correspondingly, prime benefits were smaller with incongruent than with congruent target locations. The inclusion of neutral primes allowed us to establish that this interaction resulted from a selective response slowing on prime-compatible location-incongruent trials (see Figure 3). We argued that neither sensorimotor interactions nor attentional 'enhancements' of sensorimotor processing produce the slowing on prime-compatible location-incongruent trials, but that it arises instead from difficulties beyond sensorimotor processing stages that are unique to this particular trial type.
The feature that sets compatible-incongruent trials apart from other trial types is that they contain a back-and-forth sequence of response triggers (see Figure 1). Such a sequence might not affect activation and/or inhibition within each (domain-specific) sensorimotor channel. However, to a high-level or meta-cognitive action monitoring system, it presents as a series of not merely conflicting ("go right -no, go left") but self-contradicting instructions ("go right -no, go left -no, go right after all"). We argued that similar to awareness of overt response errors (Hajcak & Foti, 2008;Hochman et al., 2017), awareness of these selfcontradictions is likely to cause confusion and a reluctance to reactivate any of the previously discarded responses.
If this high-level interpretation of the RAE is correct, then minimizing access to highlevel or 'aware' processing should minimize the RAE, and maximizing high-level confusability should maximize it. Experiments 2 and 3 directly investigated these predictions.
Using a masked prime × Simon paradigm, Experiment 2 confirmed that the RAE is associated with conscious awareness. Whereas supraliminal masked primes (i.e., primes that could be identified with above-chance accuracy) produced an RAE, this effect disappeared with subliminal primes (i.e., primes that were not available to conscious awareness, as evidenced by chance-level identification performance). This result is in line with the highlevel confusion account, and conflicts with low-level sensorimotor inhibition and automatic attentional shift accounts: similar to supraliminal stimuli (though perhaps more strongly dependent on top-down contingencies), subliminal stimuli can trigger automatic sensorimotor processes (e.g., D'Ostilio, Collette, Phillips, & Garraux, 2012;Eimer & Schlaghecken, 1998;McBride, Sumner, & Husain, 2018;Praamstra & Seiss, 2005; and shifts of visual spatial attention (e.g., Palmer & Mattler, 2013;Reuss, Pohl, Kiesel, & Kunde, 2011; for a review, see Ansorge, Kunde, & Kiefer, 2014). In contrast, stimuli or events of which we are not aware seem to have little if any effect on such processes as shifting internal attention in working memory or setting up a new task-set (e.g., Janczyk & Reuss, 2016;, or indeed delaying response execution following an error (Klein et al., 2007). Consequently, if the RAE reflects such automatic sensorimotor or attentional processes, it should have been found regardless of conscious prime awareness.
The fact that it was not suggests that the RAE is not triggered by the compatible-incongruent trials' features as such (that is, not by low-level sensorimotor processing of these features), but rather by the participant's awareness of the self-contradictory nature of these features and/or their associated sensorimotor processes.
However, not only are there several caveats attached to subliminal stimulation studies (as discussed above), it is also clear that evidence from absence necessarily remains inconclusive. Experiment 3 therefore sought to establish additional evidence for the highlevel confusion hypothesis in a context in which the high-level confusion account predicts enlarged RAEs, whereas low-level interference accounts predict unaltered or reduced RAEs.
Thus, we compared performance in a hybrid (non-masked) prime × Simon task between young and older adults: if the RAE reflects an aftereffect of sensorimotor inhibition, it should be age-equivalent or reduced in older compared to young adults, whereas if it reflects highlevel confusion, it should be disproportionately enlarged in older adults. Again, results were unambiguous: although older adults produced overall longer RTs, their priming and Simon effects were of equivalent magnitude to those of young adults, confirming that sensorimotor inhibition is not selectively impaired by aging. The same appeared to be true for CSEs, which were of similar magnitude and equally domain-specific for young and older adults -however, as noted above, the relatively low number of trials per condition coupled with the relatively small number of participants renders this result suggestive rather than conclusive. 15 Importantly, however, despite the overall similarity of performance, older adults produced disproportionately increased RAEs, in line with the high-level confusion account and in direct contradiction of low-level sensorimotor accounts.

RAE in Context
Together, the present results demonstrate that an apparently fundamental paradox in cognitive control -namely, control processes appearing to be simultaneously domain-specific (as evidenced by a lack of between-trial interactions) and domain-general (as evidenced by the presence of an under-additive within-trial interaction) -is not paradoxical after all. The evidence suggests that whereas interference effects and CSEs primarily reflect sensorimotor processes within separate, domain-specific pathways, the apparent domain-general withintrial interaction in reality reflects confusion at subsequent (higher-level or meta-cognitive) processing levels, induced by a specific self-contradictory sequence of events.
The extent to which this explanation applies to situations other than the specific conditions of an arrow-based prime-Simon task is of course an empirical question that requires further investigation. However, our account seems to fit results from various other hybrid interference paradigms. For instance, in studies using hybrid flanker × Simon paradigms (see Table 2, top), flanker effects are typically smaller on location-incongruent than on location-congruent trials. Whereas in the present experiments, the RAE was driven by a delay on prime-compatible location-incongruent trials, it seems likely that in flanker × Simon paradigms, it is driven by a delay on flanker-incompatible location-congruent trials, because location processing is faster than identity processing (Cespón, Galdo-Álvarez, & Díaz, 2013;Finkenbeiner & Heathcote, 2016). Consequently, in these tasks, the first 'instruction' would come from the stimulus location (left), the second from the flankers (no, Future research will have to systematically vary task parameters in order to map out under which conditions aging effects do or do not occur. right), and the final one from the target itself (no, left after all). As none of these studies included neutral trials, though, it is not possible to directly test this prediction at present.
However, neutral trials were included by Stoffels and van der Molen (1988) in a hybrid flanker × auditory accessory Simon task, and here, the data do indeed suggest that the withintrial distractor interaction is primarily driven by delayed responses on flanker-incompatible location-congruent trials (at least when trial presentation was mixed rather than blocked: see their Figure 2). Using a hybrid flanker × letter-Stroop task, Rey-Mermet and Gade (2016; Exp1c) also employed neutral trials, and found that responses were selectively delayed on flanker-compatible Stroop-incongruent trials. Applying the same logic as before, this could be interpreted as suggesting that in this experiment, the flanker information (i.e., whether or not the color of the central target letter matches the color of the flanking letters) was processed earlier than the Stroop information (i.e., whether the color of the central target letter matches the meaning of the word composed of target and flanking letters).
Obviously, these analyses are not meant to imply that the RAE can explain any and all hybrid interference-task results. For instance, although IOR-based tasks ( Table 2, bottom) typically produce reduced interference effects at cued relative to uncued locations, this pattern might only bear a superficial similarity to the RAE. In fact, interpreting it in terms of inhibited processing at cued locations (rather than in terms of high-level confusion) seems to be more parsimonious as well as more intuitively plausible. It is also worth noting that combinations with a Stroop task often do not produce any within-trial interactions (see Table   1), possibly because of differences in the time course of interference effects (Hommel, 1997).
The apparent discrepancy between this general trend and the Rey-Mermet and Gade (2016) results discussed above might be related to differences in the time course of word and color processing when both features form a coherent whole (as in standard Stroop tasks) compared to when they are separated by coloring different letters of the word differently (as in the Rey-Mermet & Gade, 2016, study). Future research, ideally using electrophysiological measures, will need to investigate this issue directly.
Overall, however, it seems clear that for a substantial part of the literature, apparent domain-general within-trial interactions might instead result from confusion at subsequent (higher-level or meta-cognitive) processing levels. Such 'awareness-induced slowing' is reminiscent of at least two other phenomena: post-error slowing and 'analysis paralysis'.
Post-error slowing refers to the finding that responses are often slower when they follow an error than when they follow a correct response (for a review, see Danielmeier & Ullsperger, 2005). Two aspects of post-error slowing are of particular interest in the present context. At an even broader level, the awareness-related slowing evident in both the RAE and post-error slowing is reminiscent of 'analysis paralysis', the break-down of decision-making processes through overthinking. In sports, this is known as 'choking' (an expert athlete's sudden failure to perform a highly trained skill when under pressure; Baumeister, 1984): a major contributing factor to choking is the conscious monitoring of sensorimotor processes, 16 We were able to confirm the latter in the data from Experiment 3: although, as noted above, the two age groups produced similar error rates, older adults showed significantly enlarged post-error slowing, even after accounting for overall RT differences (post-error to post-correct RT ratio = 1.113 [SD = 0.11] for older and 1.041 [SD = 0.76] for young adults, t(31.03) = 2.31, p = .028). However, further analyses revealed that posterror slowing and the RAE were not correlated in either young or older adults (both rs < 0.16, both ps > .51), suggesting that despite the conceptual similarities between the two measures, they do not reflect the same underlying mechanism. It would be interesting to investigate this issue more closely in a larger sample and with a task that simultaneously assesses error awareness.
which is assumed to interfere with their normally smooth, automatic execution (for recent reviews, see Mesagno & Beckmann, 2017;Roberts, Jackson, & Grundy, 2019). The same phenomenon also occurs in everyday tasks, from the detrimental effect of an internal focus on motor skill learning (for reviews, see Kim, Jimenez-Diaz, & Chen, 2017;Wulf, 2013) to older adults' increased but, importantly, non-functional use of attentional resources in maintaining postural and gait stability (see reviews by Li, Bherer, Mirelman, Maidan, & Hausdorff, 2018;Mak, Young, Chan, & Wong, 2018).

Synopsis and Outlook
The present study has provided evidence that in sensorimotor control tasks, a selfcontradictory sequence of events -requiring the reactivation of a recently discarded response alternative -slows response execution, provided these events are available to conscious awareness, an effect disproportionately enlarged in older adults. We have argued that this slowing does not reflect a direct inhibitory aftereffect (i.e., the response takes longer to reach execution threshold because it has been suppressed below baseline), but instead reflects a reluctance to reactivate the discarded response (akin to the aversion to repeat an erroneous response). Future studies using electrophysiological measures should investigate this issue directly.
As mentioned before, the RAE appears to share a general "avoid the discarded" feature with error aversion and numerous putative inhibitory aftereffects like the NCE, IOR, NP, N-2 repetition costs, and spontaneous alternation. It would be particularly interesting to explore whether there are common processing principles -or even a common mechanismunderlying such "been there, done that" phenomena (see Phillmore & Klein, 2019), for instance by using computational modelling to investigate underlying latent variables. Several models already exist for standard interference tasks (specifically the Flanker task, see Evans & Servant, 2020), though to the best of our knowledge, no model has yet been developed to provide an account of performance in (two-or four-alternative) hybrid interference tasks. We hope that the data from the present study (https://osf.io/2643d/) will be useful to such future modelling developments.  (1) whether congruency sequence effects (CSEs) were investigated ('-' = not investigated), and if so, whether they were found not to be present at all (0), domain-specific (S), or domain-general (G), and (2) if CSEs were investigated, whether there was a control of possible confounds. * See Table 2 for similar paradigms with reactivation aversion-like effects  (4) if CSEs were investigated, whether there was a control of possible confounds, and if so, how this was done. a The corresponding experiment with color-word Stroop stimuli did not produce a within-trial interaction b Also report electrophysiological and/or haemodynamic measures for each interference condition (studies that report electrophysiological/haemodynamic measures, but do not do so separately for each interference condition, are not marked here) c The corresponding experiment with letters instead of arrows (Exp1) did not produce a within-trial interaction d According to 24paret_all_rt_no_stat.txt, available at https://osf.io/2v857/ e The corresponding Positive Compatibility Effect condition (i.e., short prime-target ISI) did not produce a within-trial interaction f Exp3 & suppl. Exp3: previous flanker congruency interacted with NCE; effect of previous prime compatibility not investigated g Other studies from this lab, using the same paradigm, did not obtain within-trial interactions (Bensmann, Zink, Arning, Beste, & Stock, 2019;Stock, Friedrich, & Beste, 2016) h The corresponding experiment with letters instead of arrows (Exp1) did not produce a within-trial interaction i Primes, like the targets, were presented laterally displaced, hence domains were mixed rather than clearly separate in this study j non-significant trend (significant in 1-tailed test); note that the authors refer to the prime as a "flanker-like cue" k 'Part' = partial: previous Stroop congruency affected Stroop effect, but previous location congruency did not affect Simon effects l IOR = inhibition of return m Review and reanalysis of 13 published experiments * See Table 1 for similar paradigms without reactivation aversion-like effects   (Raven, Raven, & Court, 1988); maximum score = 33; data missing for one young participant b Processing speed based on the Digit Symbol Substitution task (Wechsler, 1981); data missing for one young participant c Visual acuity as measured by the number of lines read correctly from the Near Vision Test Card (Schneider, 2002) viewed at a distance of 16 inches whilst wearing corrective glasses, with scores ranging from 1 (16/160lowest acuity) to 9 (16/16 -highest acuity)  Figure 1. Hybrid prime × Simon task with neutral primes: Participants respond to the pointing direction of the target arrow with a spatially corresponding key-press, while trying to ignore the preceding prime (compatible, neutral, or incompatible with respect to the required response) and the target's location (congruent or incongruent with the response hand). Letters along the arrows indicate the sequence of response activations triggered by first the prime, then the target location, and lastly the target identity (R = right hand, L = left hand).

Figure 3. Response times (RTs; lines) in milliseconds (ms) and percentage error rates (bars)
in Experiment 1 as a function of prime compatibility, plotted separately for horizontal and vertical trials, and for congruent (black) and incongruent (white) target locations. Error bars represent ±1 SEM after removal of between-subject variability (see Cousineau, 2005).    in Experiment 2 as a function of prime compatibility, plotted separately for the two prime duration groups, and for congruent (black) and incongruent (white) target locations (note that the RT scale differs from Figures 3 and 9). Error bars represent ±1 SEM after removal of between-subject variability (Cousineau, 2005), calculated for each group separately. Figure 8. Reactivation aversion effect (RAE: Simon effect on compatible trials minus Simon effect on incompatible trials) in milliseconds (ms) as a function of priming effect magnitude, plotted separately for participants in the 16-ms prime group (squares) and 33-ms prime group (diamonds). Error bars represent ±1 SEM. The dashed oval marks the relevant comparison between RAEs of 16-ms participants with large, and 33-ms participants with small, priming effects. Figure 9. Mean correct response times (RTs; lines) in milliseconds (ms) and percentage error rates (bars) on prime-compatible and prime-incompatible trials in Experiment 3, plotted separately for young and older adults, and for location-congruent (black) and locationincongruent (white) trials. Error bars represent ±1 SEM after removal of between-subject variability (Cousineau, 2005), calculated separately for each age group. Figure 10. Interference effects (response time [RT] ratios) for young (dark gray) and older (light gray) participants in Experiment 3. Priming effects are plotted separately for trials with congruent/incongruent target locations; Simon effects are plotted separately for trials with compatible/incompatible primes. Priming and Simon effects are expressed as RT ratios (e.g., priming = incompatible RT/compatible RT; for convenience, 1 is subtracted from the result so that 0 indicates that values are the same); RAE (reactivation aversion effect) = compatibletrial Simon effect minus incompatible-trial Simon effect. Error bars represent ±1 SEM.

Experiment 1 -Additional Error Rate Analyses
Interference effects in general, and CSEs in particular, are often discussed more prominently in terms of response time (RT) differences than in terms of error rate differences. One reason for this might be that error rates are subject to floor effects (after all, it is not possible to produce fewer than zero errors, whereas it is almost always possible to respond more quickly). Another is that high error rates, more so than long RTs, are typically considered evidence that a participant was unable or unwilling to properly complete the task and thus should be excluded from analysis. However, despite these limitations error rates are still likely to provide valuable information about information processing and cognitive control. Below, we therefore present the complete error rate analysis, analogous to the RT analysis reported in the main paper: error rates were analyzed using a 2 (orientation) × 3 (prime compatibility) × 2 (location congruency) × 3 (previous prime compatibility) × 2 (previous location congruency) within-subject ANOVA. Results are listed in Table S1, and correspondingly indicated below in square brackets. The complete error rate pattern is depicted in Figure S1. Figure S1. Error rates in Experiment 1 as a function of trial orientation, previous (Prev.), and current prime compatibility (x-axis: C/Comp = compatible; N/Neut = neutral; I/Incomp = incompatible), previous location congruency (solid lines: pC = previous congruent; dashed lines: pI = previous incongruent), and current location congruency (black: Cong = congruent; white: Incong = incongruent).

Results
Error rates were higher on vertical than on horizontal trials [1a], specifically when trials were location incongruent [3b]. Both the priming effect (higher error rates on prime-incompatible than on prime-neutral and on prime-compatible trials) and the Simon effect (higher error rates on location-incongruent than on location-congruent trials) were significant [2a] [3a]. Prime compatibility and target location congruency interacted [4a], with larger Simon effects on prime-incompatible than on prime-neutral or prime-compatible trials. This effect, too, was more pronounced on vertical than on horizontal trials [4b]. Furthermore, error rates were higher following a compatible than following a neutral or incompatible trial [5a], particularly on horizontal trials [5b], and higher following a congruent than following an incongruent trial [6a]. Both priming effects and Simon effects showed domain-specific CSEs, with larger priming effects following a compatible than following an incompatible trial [8a], particularly on horizontal trials [8b], and larger Simon effects following a congruent than following an incongruent trial [9a]. Priming effects were also larger following a location-congruent than following a location-incongruent trial [10a], whereas Simon effects were enlarged following a prime-compatible trial in the horizontal condition, but enlarged following a prime-incompatible trial in the vertical condition [11b]. Two aspects are particularly noteworthy about these error rate results. First, as noted in the main text, the Prime × Location interaction is in the opposite direction to the one observed for RTs (i.e., instead of the RAE pattern of enhanced Simon effects on compatible relative to neutral and incompatible trials, error rates instead show enhanced Simon effects on primeincompatible trials). Second, unlike RTs, error rates also show evidence of domain-general CSEs, at least for priming effects. We believe that both of these patterns reflect the same mechanism, namely, the execution of very fast, prime-related responses on incompatible incongruent trials. Inspection of error rates as a function of trial type and response latency (see Figure S2) confirms this. For this analysis, correct and incorrect responses for each trial type (2 previous location congruency × 2 current location congruency × 3 current prime compatibility) were sorted into latency quintiles for each participant individually, then separate error rates were calculated for each participant, trial type, and quintile. Finally, these error rates were averaged across participants and plotted against the equally averaged mean RT of their corresponding quintile. As can be seen from Figure S2, the majority of errors occurred in the fastest quintile and with incompatible incongruent trials. On these trials, prime direction is opposite to target direction, but matches target location -that is, a (incorrect) motor response triggered by the prime will not only not be stopped by the (equally incorrect) target location but, if anything, will be facilitated by it. Furthermore, the faster the response, the more likely it should be that it was triggered by the prime (and therefore, that it would result in an error on incompatible trials). Consequently, the incompatible-trial error rate should be higher on (faster) previouscongruent than on (slower) previous-incongruent trials, which is exactly what was observed in the present data. Figure S2. Error rates in Experiment 1 as a function of overall (correct and incorrect) mean response time (RT), plotted separately for current-trial prime compatibility (compatible/Comp = circles; neutral/Neut = squares; incompatible/Incomp = triangles) and location congruency (congruent/Cong = black; incongruent/Incong = white), and for previous-trial location congruency (pC/previous congruent = solid lines; pI/previous incongruent = dashed lines).
It should be noted, though, that the same argument holds for compatible trials, on which such prime-related responses produce the correct outcome (as on these trials, by definition, prime and target match). It is therefore prudent to investigate whether the RAE is a by-product of these asymmetrical error rate effects. To this purpose, we re-calculated RT latency quintiles as above, but for correct responses only. If the RAE is an artefact of fast errors, it should be largest in the fastest quintile and diminish with increasing latency. Inspection of Figure S3 shows that the opposite is the case: in the fastest quintile, the RAE is absent, in the slowest quintile (where error rates increase again, see Figure S2), it appears distorted, but in the midlatency range (Quintiles 2-4), where errors are almost entirely absent, a clear RAE pattern is present.
Together, these results suggest that (a) error rates on incompatible incongruent trials are an index of inhibitory control, namely, the ability to suppress a strong prime-triggered motor activation, but that (b) response errors are not responsible for the RAE as observed in RTs. Figure S3. Mean correct response time (RT) in Experiment 1 on location-congruent (black) and location-incongruent (white) trials as a function of prime compatibility (C = compatible, N = neutral, I = incompatible) and latency quintile (from Q1 = fastest 20% of responses to Q5 = slowest 20%).

Experiment 1 -Visual Attention Hypothesis
The visual attention hypothesis assumes that directional arrow primes automatically shift the focus of visual attention to the indicated location. As a consequence, targets appearing at this location (i.e., inside the attentional focus) will be identified more quickly, whereas targets appearing on the opposite side (i.e., away from the attentional focus) will be identified more slowly, relative to trials with neutral primes, in which visual attention is not focused on a potential target location (see Figure S4). Figure S4. Possible shifts of visual spatial attention (gray spotlights) as a result of prime arrow direction; following neutral primes, visual spatial attention is assumed to be relatively unfocussed.
Running head: REACTIVATION AVERSION EFFECT AND SENSORIMOTOR CONTROL 82

Experiment 1 -Effects with 95% Confidence Intervals
For completeness, we present in Tables S2 and S3 all effects (i.e., RT and error rate differences) in Experiment 1 with their corresponding 95% Confidence Intervals. For details on how effects were calculated, see main text.

Experiment 2 -Effects with 95% Confidence Intervals
For completeness, we present in Table S4 all effects (i.e., RT and error rate differences) in Experiment 2 with their corresponding 95% Confidence Intervals. For details on how effects were calculated, see main text. Furthermore, as the analysis of error rate effects in Experiment 1 indicates that the RAE does not manifest in error rates, we do not provide "RAE" effects for error rates in the tables below.

Experiment 3 -Effects with 95% Confidence Intervals
For completeness, we present in Table S5 all effects (i.e., RT ratios and error rate differences) in Experiment 3 together with their corresponding 95% Confidence Intervals. For details on how effects were calculated, see main text. Furthermore, as the analysis of error rate effects in Experiment 1 indicates that the RAE does not manifest in error rates, we do not provide "RAE" effects for error rates in the table below.

Experiment 3 -Z-Scored Results
In order to account for age-related overall slowing, Faust, Balota, Spieler, and Ferraro (1999; see also Hedge, Powell, & Sumner, 2018) suggest to z-score the data prior to analysis. In the present Experiment 3, this method produced the same pattern of results as the analysis of RT ratios (reported in the main text). As can be seen in Figure S5 (left panel), the only substantial difference between young and older participants occurred for prime-compatible locationincongruent trials. Pairwise comparison of interference effects (see Figure S5, right panel) confirmed that relative to young adults, older adults produced significantly reduced priming effects on location-incongruent trials or correspondingly, significantly enlarged Simon effects on prime-compatible trials, whereas their priming effects on trials on location-congruent and their Simon effects on prime-incompatible trials (i.e., those effects unaffected by the RAE) were indistinguishable from those of their younger counterparts. Consequently, older adults produced a disproportionately enlarged RAE relative to young adults (see also three-way interaction in Table S6).  Note. Significant p-values highlighted in bold.