Authors: Changrun Huang (1Center for Cognitive Neuroscience, Duke University, Durham, NC, USA; 2Department of Psychology and Neuroscience, Duke University, Durham, NC, USA), Tobias Egner (1Center for Cognitive Neuroscience, Duke University, Durham, NC, USA; 2Department of Psychology and Neuroscience, Duke University, Durham, NC, USA)
Categories: Article
Source: Journal of experimental psychology. Human perception and performance
Doi: 10.1037/xhp0001400
Authors: Changrun Huang, Tobias Egner
It is commonly assumed that people can use advance cues to proactively prepare for conflict from distracting stimulus features, yet empirical findings remain inconsistent. We tested the hypothesis that non-arbitrary stimulus-response (SR) mappings are a key determinant of cue effectiveness, as vocal Stroop tasks (with non-arbitrary color-naming responses) have shown reliable cue benefits while manual Stroop tasks (with arbitrary key-press responses) typically have not. Across five experiments, using a spatial Stroop task with non-arbitrary SR mappings, we consistently found no evidence that participants used predictive cues to proactively resolve conflict on incongruent trials. Despite providing optimal preparation conditions (100% valid cues, 2000ms preparation time), cue benefits only emerged on congruent trials when task difficulty was increased substantially (50ms stimulus presentation, Experiment 4), likely reflecting a strategic shortcut rather than enhanced proactive control. These findings demonstrate that non-arbitrary SR mappings are not a sufficient condition for ensuring cue-based engagement of proactive control, and challenge the assumption that this form of control is a readily deployable, domain-general mechanism.
Cognitive control refers to the processes that allow individuals to regulate their thoughts and actions in line with internal goals (e.g., Egner, 2017; Miller & Cohen, 2001). Cognitive control is challenged when task-irrelevant information distracts from, or conflicts with, the processing of task-relevant information. The classic example in the psychology literature for this is the Stroop task (Stroop, 1935). In many modern adaptations of this task, participants are typically asked to name the ink color of printed color words whose meaning can be either congruent with the ink color (e.g., the word BLUE written in blue ink) or incongruent (e.g., the word GREEN written in red)^1^. In the latter case, participants need to override or suppress the processing of conflicting word information to give the correct ink color-based response, whereas in the former, the irrelevant word information can facilitate the correct response. Since word-reading is a highly habitual process, resolving conflict from incongruent words in the Stroop task is considered a canonical act of cognitive control (e.g., Cohen et al., 1990).
A prominent theoretical framework in the field distinguishes between two qualitatively different control mechanisms for resolving reactive control versus proactive control (Braver, 2012). Reactive control refers to a last-minute or emergency recruitment of control in response to an ongoing challenge, for instance, a within-trial refocusing of attention on stimulus color and away from word processing when struggling with an incongruent stimulus in the Stroop task. In contrast, proactive control operates as a preparatory mechanism that preemptively adjusts attentional settings to optimize performance. To wit, if one were to anticipate an incongruent stimulus in the Stroop task, attention toward color and away from word processing could be ramped up prior to stimulus onset. While proactive control should effectively minimize interference, it is believed to come at a significant cognitive cost, requiring active maintenance of goal-relevant representations and a more sustained biasing of attention than reactive control engagement (Braver, 2012).
Given these theoretical distinctions, a critical question in cognitive control research concerns the conditions under which individuals successfully engage proactive control mechanisms when anticipating upcoming conflict. Early investigations appeared to demonstrate that people could readily engage proactive control when the congruency status of the forthcoming trial was conveyed to the participant through predictive cues (Aarts et al., 2008; Correa et al., 2009; Gratton et al., 1992; Logan & Zbrodoff, 1982). However, these studies shared a crucial they employed tasks with small stimulus sets (or low response uncertainty) that, when combined with cues, allowed participants to resolve conflicts through strategic SR shortcuts rather than genuine proactive biasing of attention away from distracter processing (Wühr & Kunde, 2008).
Consider the seminal study by Logan & Zbrodoff (1982), in which participants responded to position words (‘ABOVE’ or ‘BELOW’) presented either above or below central fixation, creating congruent or incongruent spatial relationships. Prior to each word, participants received a cue indicating whether the word would match its spatial position. Results showed faster responses following informative compared to neutral cues. However, the binary response structure (only two options) combined with cue information enabled participants to bypass conflict processing entirely by responding based on task-irrelevant spatial features. Specifically, they could respond directly to spatial location when expecting congruent trials or simply invert their response mapping when expecting incongruent trials. The observation that cue benefits were larger for congruent than incongruent trials supports this interpretation, as response mapping inversion requires additional processing steps.
To overcome this confound, subsequent studies increased response uncertainty by incorporating larger stimulus sets and associated response options by using 4-alternative forced choice tasks. For instance, in a four-choice Stroop task, participants respond to one of four ink colors, each of which can be paired with any of the four corresponding color words. Note that even in high-uncertainty designs, a cue signaling a congruent trial still allows participants to adopt the strategic shortcut of attending to the task-irrelevant dimension (i.e., read the word) as it will reliably indicate the correct response. However, this shortcut is impossible for incongruent trials, as an incongruent cue does not help predict the specific identity of the upcoming ink color. With this predictive shortcut unavailable, the only way for participants to improve performance is to proactively prepare for suppressing the irrelevant word while enhancing attention toward the task-relevant ink color (Bugg et al., 2025). Accordingly, the contrast between cued and uncued incongruent trials constitutes the critical test of cue-based proactive control, as any behavioral benefit observed here can be unambiguously attributed to cue-driven engagement of proactive mechanisms rather than to strategic facilitation or predictive shortcuts. Thus, this particular contrast represents the main effect of theoretical interest in evaluating proactive control.
The evidence for cue benefits on incongruent trials from studies employing high response uncertainty remains inconclusive, however. While some research has demonstrated that individuals handle conflict more effectively when incongruent trials are foreshadowed by cues (Bugg & Smallwood, 2016; Goldfarb & Henik, 2013), other studies suggest that such cues rarely if ever enhance performance on incongruent trials (Jiménez et al., 2021; Wühr & Kunde, 2008; Yang & Pourtois, 2022). This inconsistency is theoretically puzzling given the robust literature demonstrating reliable control adaptation effects when regulation is based on statistical regularities (varying proportions of incongruent trials across blocks, reviewed in Bugg & Crump, 2012; Egner, 2023), a process that appears to operate implicitly (Blais et al., 2012). To understand this apparent contradiction between implicit adaptation and explicit preparation, it is instructive to examine the specific conditions under which cue benefits for incongruent trials have been most reliably demonstrated.
Arguably the most compelling evidence of proactive cognitive control engaged via explicit cues comes from Bugg and Smallwood (2016). Using a four-choices vocal Stroop task with congruency cues, they showed that with a deterministic cue (100% valid) and 2000 ms preparation time, participants responded more quickly when forewarned of an impending incongruent stimulus. This proactive control effect was replicated by (Jiménez et al., 2021, Experiment 7) when using a similar set up with a vocal Stroop task. Surprisingly, however, this effect disappeared when the Stroop task required manual key presses instead of vocal responses (Jiménez et al., 2021, Experiment 6). This discrepancy is theoretically puzzling, as one might expect that if individuals can proactively heighten control over anticipated conflicts, this capability should hold across response modalities. To explain this pattern, Jiménez et al. (2021) proposed that the engagement of cue-based proactive control is contingent on the nature of the stimulus-response (S-R) mapping within a given conflict task. They hypothesized that cue benefits on incongruent trials may emerge in conflict tasks with non-arbitrary S-R mappings (like vocal color naming), but not in those with arbitrary S-R mappings (like learned key presses for colors).
The theoretical basis of the S-R mapping hypothesis can be formalized using the Dimensional Overlap framework (Kornblum & Hasbroucq, 1990; Zhang et al., 1999). This framework classifies conflicts based on the intrinsic similarity between sets of stimulus and response features (but see De Houwer, 2003a for a different definition), where high similarity (e.g., between a stimulus set of color words and vocal color word responses) translates into a dimensional overlap. Such overlaps, and thus sources of conflict, can occur between the relevant stimulus, irrelevant stimulus, and response dimensions. According to this framework, a common source of conflict shared between vocal and manual Stroop tasks occurs at the stimulus level. This stimulus conflict arises from the overlap between the relevant and irrelevant stimulus dimensions due to the inherent semantic association (e.g., the word BLUE semantically conflicts with red ink). The crucial distinction for the S-R mapping hypothesis lies in the nature of the response conflict. In the manual Stroop task, ink colors (the relevant stimulus dimension) are arbitrarily mapped to key presses that have no intrinsic relation to color. Here, response conflict arises from newly learned S-R association but it is not caused by direct dimensional overlap (see Table 1; classified as Ensemble 4 in Kornblum’s taxonomy). Conversely, in the vocal Stroop task, the vocal response (e.g., saying “Red”) is intrinsically and semantically linked to the color word (e.g., “Blue”), creating a response conflict that involves direct Dimensional Overlap (classified as Ensemble 8). Crucially, the absence of Dimensional Overlap in the manual Stroop task does not imply an absence of response conflict, as studies clearly show that significant interference also arises from newly learned S-R mappings (Augustinova et al., 2019; De Houwer, 2003b; Kinoshita et al., 2017). Rather, the central hypothesis is that the qualitatively different sources of conflict (conflict arising from Dimensional Overlap versus conflict that is associatively learned) are the critical factor determining the efficacy of proactive control cues.
This qualitative distinction in response conflict may translate into different leverage points for proactive control. In vocal Stroop tasks, the irrelevant word may directly activate a competing articulatory motor program within the same verbal domain as the required response (Augustinova et al., 2019). Cues that signal upcoming incongruent trials may enable participants to proactively bias their response selection processes by suppressing the automatically activated verbal response corresponding to the word while prioritizing the selection of the verbal response corresponding to the ink color. In manual Stroop tasks, however, the response dimension lacks natural overlap with either stimulus dimension, as the mapping between colors and key presses is newly instructed rather than intrinsic. This creates an indirect relationship between the word stimulus and the motor response, introducing an additional translation process that may make this task version less amenable to proactive control strategies.
Taken together, the findings from Jiménez et al. (2021) suggest a compelling but as-yet untested hypothesis. This hypothesis posits that the arbitrary versus non-arbitrary nature of the S-R mapping is a critical factor for engaging cue-based proactive control. The present study was designed to provide the first direct, systematic test of this idea. To disentangle the S-R mapping hypothesis from the effect of response modality, we opted to test it in a manual task that, unlike the manual color-naming Stroop, involves non-arbitrary responses. Specifically, we employed a variant of the spatial Stroop task where spatial response locations have direct overlap with task-relevant (arrow direction) and irrelevant (stimulus position) stimulus dimensions (Viviani, Visalli, Finos, et al., 2023; Viviani, Visalli, Montefinese, et al., 2023). We combined this task with a typical congruency cuing procedure, where in the cued condition, a cue indicated the congruency of each forthcoming trial, whereas in the uncued condition, cues did not convey congruency information (e.g., Bugg & Smallwood, 2016). Critically, this approach allows us to evaluate whether non-arbitrary mappings can yield the cue benefits that have proven elusive in the numerous prior studies employing arbitrary manual response mappings (e.g., Jiménez et al., 2021 Experiment 5, 6 and 9; Wühr & Kunde, 2008; Yang & Pourtois, 2022). We note that while monetary incentives can enhance cue utilization (Bugg et al., 2015; Chiew & Braver, 2016), we here focused on intrinsic task factors to better understand why vocal Stroop tasks appear to naturally yield cue benefits while manual versions typically do not.
In Experiment 1 we tested the S-R mapping hypothesis via two complementary experimental approaches. First, we used a variant of the spatial Stroop task that has non-arbitrary response mappings (Viviani, Visalli, Finos, et al., 2023; Viviani, Visalli, Montefinese, et al., 2023). In this task, participants are required to respond to the direction of arrow stimuli using spatially corresponding keys, while ignoring spatially congruent or incongruent stimulus locations. This set-up preserves the nonarbitrary S-R dimensional overlap that characterizes vocal Stroop tasks while using manual responses, and it allowed us to isolate the role of S-R mapping from response modality per se. Second, to further disentangle stimulus conflict from response conflict, we applied De Houwer’s (2003) two-to-one response mapping method, where participants use one key to respond to two stimulus categories. This manipulation creates conditions where stimulus conflict can occur with or without response conflict, allowing us to test whether congruency cues may selectively facilitate resolution of response conflict (where dimensional overlap is most relevant) versus stimulus conflict. We hypothesized that participants would demonstrate proactive control benefits from congruency cues in the spatial Stroop task and that these benefits would be specifically evident for response conflict rather than stimulus conflict.
We report how we determined our sample size, all data exclusions, all manipulations, and all measures across all five experiments, and we follow JARS (Appelbaum et al., 2018). Sample sizes for all experiments were determined based on precedent in the literature, with several previous studies demonstrating congruency cue benefits employing samples of 24 participants (Bugg & Smallwood, 2016; Jiménez et al., 2021). To ensure adequate statistical power while accounting for potential attrition, we recruited a minimum of 30 participants per experiment. Data exclusion criteria and procedures are detailed in the respective Method sections for each experiment.
Data collection for all experiments occurred between 2024 and 2025. All participants provided informed consent prior to participation. All studies were approved by the Duke University Campus Institutional Review Board (Protocol # 2025-0187).
All data, analysis code, and research materials are available on https://osf.io/65wnh (Huang & Egner, 2025). Data were analyzed using R, version 4.4.2 (R Core Team, 2024) and the package afex, version 1.4.1 (Singmann et al., 2024). None of these studies were preregistered. Our findings are constrained to participants recruited from online platforms (Prolific) who were predominantly from Western, English-speaking populations.
Sample size was determined based on precedent in the literature. Several previous studies demonstrating congruency cue benefits employed samples of 24 participants (Bugg & Smallwood, 2016; Jiménez et al., 2021). To ensure adequate statistical power for detecting similar effects while accounting for potential attrition, we recruited a minimum of 30 participants per experiment. The effect of interest for the current study was cue benefits for incongruent trials in a task with non-arbitrary stimulus-response mapping; therefore, we used Bugg and Smallwood (2016) as our reference for expected effect size (d = 0.92). A priori power analysis using the pwr package in R showed that our sample sizes provided > 99% power to detect incongruent cue benefits (α = .05, two-tailed paired t-tests) based on the expected effect size (d = 0.92) from Bugg and Smallwood’s similar conditions (100% valid cues, 2,000ms preparation, non-arbitrary mapping). This sample size provided sufficient power across all experiments given identical design parameters (100% valid cues, 2,000ms preparation, non-arbitrary mapping). In Experiment 1, a total of 33 participants were recruited via the Prolific platform and received monetary compensation for their participation. All participants reported having normal or corrected-to-normal vision with no color blindness. Prior to the experiment, participants provided informed consent. One participant was excluded for their mean accuracy and mean reaction time exceeding ±2.5 SD from the overall sample means, resulting in a final sample of 32 participants (16 women; age 21–60 years, M = 37.94, SD = 10.30). The study was approved by the Duke University Campus Institutional Review Board (Protocol # 2025-0187).
The current experiment used a well-validated spatial Stroop task that has been shown to reliably elicit robust congruency effect (Puccioni & Vallesi, 2012; Viviani, Visalli, Finos, et al., 2023; Viviani, Visalli, Montefinese, et al., 2023). The experiment was created using jsPsych (de Leeuw et al., 2023). All stimuli were superimposed on a light gray background (RGB: 192/192/192).
As illustrated in Figure 1, each trial began with a symbolic cue displayed for 1500 ms, followed by a 500 ms presentation of a partial black square outline (110 × 110 pixels) with a central fixation cross (33 × 33 pixels). This sequence created a consistent 2000 ms cue-to-stimulus interval (CSI). Subsequently, an arrow (24 × 24 pixels) appeared in one of four quadrants (lower left, upper left, lower right, upper right) within the square and remained visible until response, followed by a 300 ms inter-trial interval (blank screen). Participants were instructed to respond to the arrow’s direction while ignoring its’ spatial position, using horizontally aligned response keys that corresponded to the arrow’s horizontal position (i.e., left vs. right of center).
Prior to the presentation of the arrow target stimulus, participants viewed one of two types of symbolic cues (48-pixel font size): either an informative cue (= or ≠) or a non-informative cue (o). In the informative-cue blocks, the cues predicted the upcoming Stroop stimulus congruency with 100% validity. Specifically, an equal sign (=) indicated that the arrow’s position would match its pointing direction (congruent trial), while a not-equal sign (≠) indicated a mismatch (incongruent trial). In the non-informative-cue blocks, a circle symbol (o) provided no predictive information about congruency. The response mapping employed a two-to-one design that we assumed to be sufficiently non-arbitrary due to its general spatial correspondence. Specifically, participants pressed the ‘d’ key (left side of keyboard) with their left index finger when the arrow pointed to the top-left or bottom-left, and the ‘k’ key (right side of keyboard) with their right index finger when the arrow pointed to the top-right or bottom-right. This configuration was intended to create dimensional overlap between the response (left or right key press), the arrow’s relevant directional information, and its’ irrelevant spatial location (as detailed in Table 1).
The task contained three trial types with different conflict congruent trials (50% of trials; e.g., arrow ↖ at top-left), same-response incongruent trials (25% of trials; e.g., arrow ↖ at bottom-left, creating stimulus conflict without response conflict), and different-response incongruent trials (25% of trials; e.g., arrow ↖ at top-right, creating both stimulus and response conflicts). The trials were pseudo-randomly intermixed within each block, with a restriction imposed to control for congruency sequence effects^2^.
The main experiment followed a counterbalanced design, with participants beginning with either the informative or non-informative blocks. Each condition (informative and non-informative) consisted of two sub-blocks of 96 trials, for a total of 192 trials per informative and non-informative condition. Within each of these 192 trials, there were 96 congruent trials, 48 same-response incongruent trials, and 48 different-response incongruent trials. Before starting each condition, participants completed a 48-trial practice block with trial-by-trial feedback. To advance from practice to test blocks, participants needed to achieve at least 75% accuracy. Detailed instruction of the corresponding cue was provided right before respective practices. For the informative condition, participants were told that the symbolic cues would reliably predict the match or nonmatch between the arrow’s pointing direction and position, while for the non-informative condition, participants were told that the cues provided no predictive information. They were encouraged to utilize the predictive information from informative cues to optimize their performance. The entire experimental session, including instructions and practice blocks, took approximately 25–30 minutes to complete.
The study followed a 2 (Cue Block: informative vs. non-informative) × 3 (Trial Type: congruent, same-response incongruent, different-response incongruent) factorial design. To evaluate whether congruency cues facilitated participants’ performance, we analyzed mean response times (RTs) and error rates using repeated-measures ANOVAs. The within-subject factors were Cue Block (informative vs. non-informative) and Trial Type (congruent, same-response incongruent, different-response incongruent). For RT analyses, we excluded incorrect trials and responses outside the 200–3000 ms range. Additionally, we removed trials with RTs exceeding ±2.5 SD from each participant’s mean. Bayes factors (BFs) were calculated using the BayesFactor package in R (Morey et al., 2018) with default Jeffreys-Zellner-Siow (JZS) priors (Cauchy scale r = 0.707 for t-tests, r = 0.5 for ANOVA effects; Rouder et al., 2012). We report BF₁₀ for evidence favoring the alternative hypothesis and BF₀₁ for evidence favoring the null hypothesis. Following Wagenmakers et al. (2018), BF > 3 indicates substantial evidence, BF > 10 indicates strong evidence, and BF > 30 indicates very strong evidence. Extreme values are reported as BF > 1000 or BF < 0.001.
Mean RTs for each condition are plotted in Figure 2A (left panel). The RT analysis revealed a significant main effect of Trial Type, F(2, 62) = 78.89, p < .001, ηp² = .72, BF₁₀ > 1000. Post-hoc comparisons indicated that responses in the congruent condition (M = 519 ms) were significantly faster than those in the same-response incongruent condition (M = 569 ms, t(31) = 7.36, p < .001, d = 0.43, BF₁₀ > 1000) and the different-response incongruent condition (M = 655 ms, t(31) = 9.41, p < .001, d = 1.00, BF₁₀ > 1000). Furthermore, responses in the same-response incongruent condition were faster than those in the different-response incongruent condition (t(31) = 8.83, p < .001, d = 0.60, BF₁₀ > 1000). However, contrary to our expectation, the main effect of Cue (F(1, 31) = 0.21, p = .646, ηp² = .01) and the interaction between Cue Block and Trial Type (F(2, 62) = 2.40, p = .104, ηp² = .07) were not significant, thus revealing no cue benefit. Bayes factors provided strong evidence for the absence of an interaction effect (BF₀₁ = 18.98) and strong evidence for the absence of a main effect of Cue Block (BF₀₁ = 33.27).
The analysis of error rates mirrored the findings from the RT analysis, revealing a significant main effect of Trial Type (F(2, 62) = 33.59, p < .001, ηp² = .52, BF₁₀ > 1000), a nonsignificant main effect of Cue Block (F(1, 31) = 0.56, p = .461, ηp² = .02, BF₀₁ = 26.91) and nonsignificant interaction (F(2, 62) = 0.67, p = .506, ηp² = .02, BF₀₁ = 75.63). Post-hoc comparisons for Trial Type indicated that participants committed significantly more errors in the different-response incongruent condition (M = 6.02%) compared to the congruent condition (M = 0.36%, t(31) = 6.28, p < .001, d = 1.50, BF₁₀ > 1000) and the same-response incongruent condition (M = 0.94%, t(31) = 5.88, p < .001, d = 1.23, BF₁₀ > 1000). No significant difference was found between the congruent and same-response incongruent conditions (p = .509, BF₀₁ = 2.17, d = 0.35).
In Experiment 1, we combined 100% valid congruency cues with and a 1 SR mapping in a spatial Stroop task with non-arbitrary responses. Consistent with previous studies, we observed robust congruency effects in this task (Puccioni & Vallesi, 2012; Viviani, Visalli, Finos, et al., 2023; Viviani, Visalli, Montefinese, et al., 2023). Extending previous findings with this task, we also found that congruency effects can be observed both for stimulus-based and for response-based conflicts, as the reaction time difference between congruent and same-response incongruent trials was robust but only about the third of the magnitude of the difference between congruent and different-response incongruent trials (cf., De Houwer, 2003). Contrary to our main hypothesis, however, despite employing a conflict task with non-arbitrary response mapping, we found no evidence of cue benefits for either congruent or incongruent trials. Bayes factor analyses strongly supported the absence of both a main effect of Cue Block and the interaction effect between Cue Block and Trial Type.
While these results suggest that conflict involving dimensional overlap between irrelevant stimulus features and responses is not a sufficient condition for observing cue benefits, several alternative explanations warrant examination before drawing definitive conclusions. One possible explanation for the absence of cue benefits in Experiment 1 may be our use of a 1 response mapping method. Although this approach was initially implemented to isolate whether the cue specifically facilitates conflict resolution during the stimulus encoding versus response selection phase, it may have inadvertently compromised the non-arbitrary nature of the response mapping. Specifically, by not directly mapping the arrow’s direction to its corresponding spatial location, we potentially diminished the task’s inherently non-arbitrary characteristics, thereby reducing the cue’s potential effectiveness.
Furthermore, combining cues with the 1 response mapping may have limited the cues’ utility. For instance, in same-response incongruent trials, participants might anticipate conflict following an incongruent cue, yet experience a relatively small amount of conflict since the incongruency manifests only at the stimulus level rather than at the stimulus and response level. This discrepancy between expected and experienced conflict could potentially discourage the use of the cue. To address these methodological concerns, Experiment 2 abandoned the 1 mapping approach, using a four-key response mapping instead, while maintaining the cue manipulation. This modification ensures a more direct, non-arbitrary response mapping, allowing for a more definitive assessment of whether cue benefits emerge under genuinely non-arbitrary manual response-mapping conditions.
In Experiment 2, we abandoned the 1 mapping approach and instead implemented a direct, one-to-one response configuration which should provide a cleaner and more definitive test of the S-R mapping hypothesis. Participants now responded to four distinct arrow directions using four spatially corresponding keys (lower left, upper left, lower right, upper right) that spatially aligned with both the arrows’ orientations and positions. This design ensures a truly non-arbitrary stimulus-response mapping. If cues function primarily to resolve conflict in stimulus-response dimensional overlap, we should observe significant performance improvements in the informative cue block compared to the non-informative cue block. Conversely, if conflict involving stimulus-response dimensional overlap is not the determining factor in cue effectiveness, we would expect comparable performance between these conditions.
A total of 40 participants were recruited via the Prolific platform and received monetary compensation for their participation. One participant was excluded for their mean accuracy and mean reaction time exceeding ±2.5 SD from the overall sample means, resulting in a final sample of 39 participants (19 women; age 20–60 years, M = 36.51, SD = 12.01).
The stimuli, design, and procedure were identical to those of Experiment 2, with the critical modification that participants now responded to all four arrow directions using four spatially corresponding keys (See Figure 1). Specifically, participants responded to arrows pointing to the top-left using their left middle finger on the ‘e’ key, to the bottom-left using their left index finger on the ‘d’ key, to the top-right using their right middle finger on the ‘o’ key, and to the bottom-right using their right index finger on the ‘k’ key. This response configuration ensured spatial correspondence between stimulus directions, stimulus locations, and response locations. The study conformed to a 2 (Cue Block: informative vs. non-informative) × 2 (Trial Type: congruent vs. incongruent) factorial design^3^. Each cue block condition (informative and non-informative) contained a total of 192 trials, composed of 96 congruent and 96 incongruent trials. Participants also completed a 48-trial practice block before each condition. The entire session lasted approximately 25–30 minutes.
The repeated measures ANOVA on mean RTs revealed a significant main effect of Trial Type (F(1, 38) = 176.69, p < .001, ηp² = .82, BF₁₀ > 1000), a nonsignificant main effect of Cue Block (F(1, 38) = 1.26, p = .268, ηp² = .03, BF₀₁ < 0.001), and a nonsignificant interaction (F(1, 38) = 1.56, p = .219, ηp² = .04, BF₀₁ = 10.21). This pattern of results is displayed in Figure 2A (middle panel). Similarly, the error rate analysis revealed a significant main effect of Trial Type (F(1, 38) = 59.87, p < .001, ηp² = .61, BF₁₀ > 1000), a nonsignificant main effect of Cue Block (F(1, 38) = 0.37, p = .545, ηp² = .01, BF₀₁ = 42.93), and a nonsignificant interaction (F(1, 38) = 0.00, p = .945, ηp² = .00, BF₀₁ = 38.15). Both RT and error rates were substantially heightened in incongruent (MRT = 687 ms, Merror = 3.66%) relative to congruent (MRT = 529 ms, Merror = 0.41%) trials.
Despite implementing a strictly non-arbitrary response mapping in Experiment 2, and obtaining very robust congruency effect, the results failed to demonstrate any evidence of cue benefits. Instead, Bayes factor analysis provided strong evidence (BF₀₁ > 10) supporting the null interaction effect. These findings suggest that the non-arbitrary nature of response mapping in most manual conflict tasks may not the determining factor in participants’ utilization of cues for conflict preparation. However, a potentially important difference between the procedures of the above Experiments and successful demonstrations of cue benefits in the prior literature using the vocal Stroop task (e.g.,Bugg & Smallwood, 2016) pertains to the time available between providing a response on trial N and being presented with the cue of trial N+1. We addressed this discrepancy in Experiment 3.
Despite both Experiments 2 and 3 employing non-arbitrary response mappings yet yielding no cue benefits, a critical timing difference between our experiments and those of Bugg and Smallwood (2016) warrants consideration. As noted by Jiménez et al., (2021), trials in Bugg and Smallwood’s study involved manual coding of vocal responses by the experimenter, creating a temporal lag after each response. To accommodate this, Bugg and Smallwood (2016) calibrated the inter-trial interval based on varying cue-to-stimulus intervals, ensuring that the combined response-to-cue interval (RCI) and cue-to-stimulus interval (CSI) consistently totaled 3100 ms. Correspondingly, Jiménez et al., (2021, Experiment 7) found that a vocal Stroop task with a 750 ms CSI combined with a 2250 ms RCI produced significant cue benefits, whereas a condition with only the 750 ms CSI without an extended RCI failed to elicit cue benefits (Experiment 10). These findings suggest that task pace or inter-trial transition duration may critically influence the engagement of proactive control mechanisms when utilizing cues.
To rule out inter-trial timing interval as a potential explanation for our null findings, Experiment 3a extended the RCI to 1000 ms to approximate the timing employed by Bugg and Smallwood (2016). Building on this, Experiment 3b implemented an even longer RCI of 2250 ms (following Jiménez et al., 2021) to further examine whether a slower response pace facilitates effective utilization of cue information.
We recruited two separate groups of participants via the Prolific platform for Experiment 3a and 3b, with all participants receiving monetary compensation. For Experiment 3a, 34 participants were initially recruited, with two excluded (one for accuracy, one for reaction time exceeding ±2.5 SD from sample means), resulting in a final sample of 32 participants (17 women; age 20–39 years, M = 27.91, SD = 5.19). For Experiment 3b, from an initial sample of 33 participants, three were excluded (one for accuracy, two for reaction time exceeding the same threshold), yielding a final sample of 30 participants (18 women; age 18–40 years, M = 31.37, SD = 6.62).
The tasks employed in Experiment 3a and 3b were identical to that of Experiment 2, with the critical modification that the post-response blank interval (the RCI) was extended to 1000 ms in Experiment 3a and to 2250 ms in Experiment 3b.
Experiment 3a. Mean reaction times for each condition are plotted in Figure 2A (right panel). The repeated measures ANOVA on mean RTs returned a significant main effect of Trial Type (F(1, 31) = 136.85, p < .001, ηp² = .82, BF₁₀ > 1000), a nonsignificant main effect of Cue Block (F(1, 31) = 0.03, p = .875, ηp² < .01, BF₀₁ = 2.97), and a nonsignificant interaction (F(1, 31) = 0.02, p = .886, ηp² < .01, BF₀₁ = 19.77). Similarly, the error rate analysis revealed a significant main effect of Trial Type (F(1, 31) = 35.98, p < .001, ηp² = .54, BF₁₀ > 1000), a nonsignificant main effect of Cue Block (F(1, 31) = 0.98, p = .329, ηp² < .01, BF₀₁ = 18.22), and a nonsignificant interaction (F(1, 31) = 0.77, p = .388, ηp² < .01, BF₀₁ = 20.59). Performance was worse in incongruent (MRT = 886 ms, Merror = 1.74%) than in congruent (MRT = 715 ms, Merror = 0.20%) trials.
Experiment 3b. Mean reaction times for each condition are plotted in Figure 2B (left panel). The repeated measures ANOVA on mean RTs returned a significant main effect of Trial Type (F(1, 29) = 137.00, p < .001, ηp² = .83, BF₁₀ > 1000), a nonsignificant main effect of Cue Block (F(1, 29) = 0.58, p = .451, ηp² = .02, BF₀₁ = 0.04), and a nonsignificant interaction (F(1, 29) = 0.40, p = .535, ηp² = .01, BF₀₁ = 21.04). Similarly, the error rate analysis revealed a significant main effect of Trial Type (F(1, 29) = 21.39, p < .001, ηp² = .42, BF₁₀ > 1000), a nonsignificant main effect of Cue Block (F(1, 29) = 0.71, p = .407, ηp² < .01, BF₀₁ = 30.17), and a nonsignificant interaction (F(1, 29) = 0.04, p = .835, ηp² < .01, BF₀₁ = 32.43). Performance was worse in incongruent (MRT = 729 ms, Merror = 2.20%) than in congruent (MRT = 588 ms, Merror = 0.31%) trials.
The results of Experiment 3a and 3b are consistent with those from Experiments 1 and 2, revealing no evidence of cue benefits even with slower task pace. Bayes factors provided strong evidence for the null interaction effect (both BF₀₁ > 10). Across experiments, our findings demonstrate that conflict involving non-arbitrary stimulus-response dimensional overlap does not inherently guarantee the effective engagement of proactive control through cue utilization, including under timing conditions that prior, vocal Stroop studies suggest are optimal.
Leaving aside the factor of response modality, what reason may there be for participants not employing the congruency cues to enhance their performance? A possible determinant of cue usage concerns participants’ motivation. Although participants may fully comprehend and perceive the cue, it might offer minimal value if they feel like they can perform the task adequately without it. Moreover, using the cue likely imposes cognitive effort, against which the cue utility must be weighed. According to the “expected value of control” framework (M. Botvinick & Braver, 2015; Shenhav et al., 2013), the preemptive engagement of cognitive control in preparation for upcoming conflict is cognitively demanding, taxing limited processing resources, and consequently would only be recruited if deemed necessary for adequate task performance.
One approach to enhance the value of congruency cues is by increasing the difficulty of stimulus processing (Bugg et al., 2015). Under more challenging conditions, the utility of the advance information provided by the cue should increases, thus encouraging cue usage. To test this hypothesis in the context of the current task protocol, Experiment 4 increased task difficulty by reducing stimulus presentation time.
Experiment 4 aimed to increase the utility of cue information, thereby enhancing participants’ motivation to use it in preparation for upcoming task trials. To this end, we increased task difficulty by restricting stimulus presentation time to just 50 ms, immediately followed by a post-stimulus mask designed to curtail lingering post-perceptual processing. This manipulation creates conditions under which advance information (and preparation) should become more beneficial, potentially making the informational value of the cue more salient and increasing the likelihood that participants will engage proactive control mechanisms to optimize performance.
A total of 33 participants were recruited via the Prolific platform and received monetary compensation for their participation. One participant was excluded for the mean accuracy and the mean reaction time exceeded ±2.5 SD from the overall sample means, resulting in a final sample of 32 participants (11 women; age 20–39 years, M = 27.41, SD = 4.91).
The task was identical to that of Experiment 3a, with a critical modification to increase task the arrow stimulus was presented for only 50 ms, immediately followed by a visual mask. The mask consisted of superimposed arrows pointing in all four directions (24 pixels × 24 pixels), terminating the perceptual processing of the target stimulus. Participants were required to respond within 1500 ms from the onset of the mask, maintaining time pressure while ensuring the task demanded rapid stimulus encoding.
The RTs are plotted in Figure 2B (middle panel). A repeated measures ANOVA on mean RTs revealed a significant main effect of Trial Type, F(1, 31) = 232.59, p < .001, ηp² = .88, BF₁₀ > 1000, a significant main effect of Cue Block, F(1, 31) = 8.28, p = .007, ηp² = .21, BF₁₀ > 1000 and a significant interaction, F(1, 31) = 12.32, p = .001, ηp² = .28, BF₁₀ > 1000. Simple effects analyses showed that, for congruent trials, participants responded significantly faster when the cue was informative compared to when it was non-informative (Minform = 425 ms vs. Mnoninform = 478 ms; t(31) = 4.12, p < .001, d = 0.62, BF₁₀ = 106.75)^4^. No such effect was observed for incongruent trials, though the numerical pattern was in line with that expected if participants were using the cues (Minform = 563 ms vs. Mnoninform = 582 ms; p > .174, d = 0.19, BF₀₁ = 2.21).
The error rate analysis revealed a significant main effect of Trial Type, F(1, 31) = 60.78, p < .001, ηp² = .66, BF₁₀ > 1000. More errors were made in incongruent trials (M = 16.29%) than congruent trials (M = 4.46%). In contrast, the main effect of Cue Block (F(1, 31) = 1.73, p = .198, ηp² = .05, BF₀₁ = 1.66) was not significant. Moreover, the interaction between Trial Type and Cue Block was nonsignificant (F(1, 31) = 0.33, p = .569, ηp² = .01, BF₀₁ = 17.06), as reflected by a substantial Bayes factor in favor of the null hypothesis. We note that the absence of a cueing effect for the congruent trials in the error rates does not appear to reflect a speed-accuracy trade-off. A speed-accuracy trade-off would predict an increase in errors accompanying the faster reaction times in the cued condition, a pattern we did not observe. A more plausible explanation for the lack of a cueing effect on congruent trials may be a ceiling effect on accuracy, as performance on congruent trials was very high (over 95% correct), leaving little room for improvement.
The results of Experiment 4 revealed, for the first time in this series of experiments, that participants utilized the cue to improve task performance. However, this effect was reliable only in congruent trials. Incongruent trials showed the expected numerical trend but it failed to reach statistical significance.
Error rates for incongruent trials were substantially higher compared to previous versions of the spatial Stroop task, confirming that our manipulation successfully increased task difficulty. This heightened difficulty likely enhanced the strategic value of cue utilization, as participants appeared to leverage advance information to optimize performance under increased cognitive demands. The pattern of selective benefits for congruent trials suggests that participants may have adopted a strategic shortcut. For instance, when forewarned by a congruent cue, they likely exploited this information by relaxing top-down control and basing their response on the arrow’s irrelevant location rather than its task-relevant direction. Crucially, this cuing advantage did not reliably extend to incongruent trials where, as argued in the introduction, participants would have had to genuinely engage proactive control to produce a cue benefit. Nevertheless, the fact that participants made any sort of use of the cue in Experiment 4 seemed encouraging and motivated us to attempt to enhance the utility of the cues in an additional, different way in Experiment 5, namely, by manipulating the frequencies of congruent/incongruent trials.
To further enhance the potential for the congruency cue to be employed by participants, we presented a mostly congruent stimulus list (75% congruent, 25% incongruent) in Experiment 5. This design arguably encourages the use of proactive control in response to incongruent cues in two, related ways. First, a predominantly congruent condition is likely to induce a lower baseline level of cognitive control, thereby providing ample capacity for participants to recruit additional proactive control in response to the informative cue (i.e., it reduces the possibility of ceiling effects). Second, in a context where incongruent trials are relatively infrequent, a cue signaling an upcoming conflict becomes particularly valuable as it can preempt a large performance cost due to the low baseline control levels. This aligns with previous research demonstrating that the benefit of explicit congruency cues is more pronounced when the likelihood of encountering conflict is low (Goldfarb & Henik, 2013 Experiment 2).
A total of 33 participants were recruited via the Prolific platform and received monetary compensation for their participation. Three participant was excluded for the mean accuracy (N=1) and the mean reaction time (N=2) exceeded ±2.5 SD from the overall sample means, resulting in a final sample of 30 participants (18 women; age 18–40 years, M = 31.37, SD = 6.62). For reaction times (RTs) analysis, four additional participants were excluded due to having no correct responses for any incongruent trials in either the informative or non-informative blocks.
The task was identical to that of Experiment 4, retaining the high-difficulty design parameters (i.e., 50 ms stimulus presentation followed by a mask). The critical additional manipulation was the trial proportions. Each 192-trial block consisted of 25% incongruent trials (48 trials) and 75% congruent trials (144 trials). The two trial types were randomly intermixed As in the previous experiments, participants also completed a practice block of 48 trials before each cue block. The total session lasted approximately 25–30 minutes.
The RTs are plotted in Figure 2B (right panel). Repeated measures ANOVAs were conducted on mean RTs and mean error rates. The RT analysis revealed a significant main effect of Trial Type, F(1, 25) = 128.79, p < .001, ηp² = .84, BF₁₀ > 1000. Responses were slower in incongruent trials (M = 619 ms) than congruent trials (M = 437 ms). The main effect of Cue Block (F(1, 25) = 1.39, p = .250, ηp² = .05, BF₀₁ < 0.001) and the interaction were nonsignificant (F(1, 25) = 0.37, p = .550, ηp² = .01, BF₀₁ = 0.28).
The error rate analysis also revealed a significant main effect of Trial Type, F(1, 29) = 34.68, p < .001, ηp² = .54. Responses are more error-prone in incongruent trials (M = 40.66%) than congruent trials (M = 4.38%). The main effect of Cue Block (F(1, 29) = 1.85, p = .184, ηp² = .06, BF₀₁ = 1.66) and the interaction were nonsignificant (F(1, 29) = 0.09, p = .771, ηp² < .01, BF₀₁ = 17.06).
The results of Experiment 5 provided no evidence of cue benefits, despite increased task difficulty through brief stimulus presentation (replicating the timing of Experiment 4) and a predominantly congruent context (75% congruent trials). The predominantly congruent environment was designed to encourage a less effortful baseline cognitive control setting, theoretically providing ample capacity for participants to recruit additional proactive control in response to incongruent cues. Additionally, in a context where incongruent trials constitute only 25% of all trials, the cue signaling an upcoming conflict trial should become particularly valuable for optimizing performance. The substantially elevated error rates in incongruent trials (M = 40.66%) confirmed that the task successfully created challenging conditions under which advance preparation should be beneficial. The fact that no cue benefits were observed on incongruent trial regardless casts some doubt on the idea that cue value per se is a reliable determinant of congruency cue use. Perhaps surprisingly, participants also did not show a cue benefit on congruent trials, unlike in Experiment 4. However, it is plausible that this could reflect the fact that in a 75% congruent list, using the location as a response shortcut may become the default strategy in both the uncued and cued conditions, and therefore no additional benefits can be observed in the latter.
We consistently observed null effects of congruency cues on incongruent trial performance across all experiments, suggesting that participants either were not willing or not capable of using cues to engage proactive control. To further contextualize these null findings, we sought to establish whether participants did nevertheless display control adjustment effects unrelated to the use of explicit cues. This is important for establishing that it is specifically the use of explicit trial-level cues that was ineffective in modulating attentional control, rather than a complete absence of attentional adaptation altogether that characterizes the above data sets. To rule out the latter possibility, we examined the congruency sequence effect (CSE), whereby congruency effects are reduced following incongruent relative to congruent trials (Gratton et al., 1992; reviewed in Egner, 2007). The CSE is thought to reflect participants adjusting cognitive control in line with their experience on the most recent trial, upregulating attention in response to incongruent trials, and downregulating attention in responses to congruent ones (Botvinick et al., 2001; Gratton et al., 1992; reviewed in Braem et al., 2019). Importantly, this form of adaptive control emerges naturally from conflict experience rather than volitional attempts to use explicit congruency cues (e.g., van Gaal et al., 2010), providing a complementary measure of participants’ use of adaptive control (Egner, 2014).
The CSE was calculated as the difference in congruency effects between trials following congruent versus incongruent CSE = (cI - cC) - (iI - iC), where cI = incongruent after congruent, cC = congruent after congruent, iI = incongruent after incongruent, and iC = congruent after incongruent. Positive CSE values indicate conflict adaptation, reflecting smaller congruency effects following incongruent trials. For this analysis, we excluded the first trial of each block (as these lacked a preceding trial) and trials following incorrect responses (to prevent confounding with post-error slowing effects).
To quantify the CSE in each experiment, we calculated individual participant CSE scores and conducted one-sample t-tests against zero. The results demonstrated significant conflict adaptation effects in all five experiments (see Figure 3): Experiment 1 (M = 29 ms, t(31) = 5.86, p < .001, d = 1.04), Experiment 2 (M = 46 ms, t(38) = 7.62, p < .001, d = 1.22), Experiment 3a (M = 49 ms, t(31) = 5.25, p < .001, d = 0.93), Experiment 3b (M = 43 ms, t(29) = 5.70, p < .001, d = 1.04), Experiment 4 (M = 36 ms, t(31) = 6.07, p < .001, d = 1.07), and Experiment 5 (M = 54 ms, t(22) = 2.88, p = .009, d = 0.60). Across all experiments, the CSE was robust and substantial (M = 42 ms, t(187) = 12.04, p < .001, d = 0.88). A one-way ANOVA comparing CSE magnitude across experiments revealed no significant differences (F(5, 182) = 1.01, p = .416, ηp² = .03), indicating consistent conflict adaptation regardless of the specific experimental manipulation.
Obtaining a robust CSE in all experiments demonstrates that participants did in fact modulate their attention, presumably implicitly, in response to trial events in the current experiments, even though they showed no evidence of such modulation in response to informative trial-level cues. This dissociation suggests that the absence of cue benefits cannot be attributed to a general deficit in control modulation but is in fact specific to the use of explicit cues for proactive control.
Across five experiments, we systematically investigated whether non-arbitrary stimulus-response (SR) mappings would enable the cue-based proactive control that has consistently failed to emerge in prior studies using arbitrary response mappings in manual conflict tasks (Jiménez et al., 2021 Experiment 6; Wühr & Kunde, 2008; Yang & Pourtois, 2022). Contrary to the S-R mapping hypothesis, the spatial Stroop task with non-arbitrary response mappings consistently provided no evidence that congruency cues enhance conflict resolution. Additionally, Experiment 3 and 5 further demonstrated that neither manipulating inter-trial transition speed nor implementing a mostly congruent list influenced participants’ use of cues. Cue benefits emerged only when the expected value of advance information was substantially increased through brief stimulus presentation (Experiment 4). However, these benefits were restricted to congruent trials, where benefits can be obtained by relaxing rather than strengthening attentional control. Finally, despite a pervasive lack of evidence for cue-triggered attentional adjustments, there was reliable evidence for a more implicit form of control adjustment, in the shape of the CSE, throughout all of the experiments. Taken together, the stark contrast between the robust, experience-driven CSE and the complete absence of cue-driven proactive control challenges the prevailing assumption that individuals can readily use explicit cues to prepare for cognitive conflict.
The consistent absence of cue benefits suggests that S-R dimensional overlap, while a notable difference between typical vocal and manual Stroop tasks, may not be the critical determinant for congruency cue efficacy. An alternative explanation for the discrepancy between our findings and those of (Bugg & Smallwood, 2016) could be the difference in response modality (i.e., vocal naming vs. manual key presses). However, this explanation is more descriptive rather than explanatory, as it does not specify why a particular motor system should be a boundary condition for proactive control. Proactive control, as conceptualized in the cognitive control literature, represents a domain-general mechanism for anticipatory adjustment of attentional settings (Braver, 2012). The upregulation of attention towards the task-relevant dimension and/or the suppression of task-irrelevant processing should, in principle, operate independently of the specific motor system used to execute the response. If participants can successfully implement top-down attentional modulation based on cues in vocal tasks, no compelling theoretical reason exists for why identical attentional adjustments would fail when responses are executed manually.
A domain-general conception of proactive control further predicts that attentional tuning should operate consistently across different task contexts, regardless of whether conflict is semantic or spatial in nature. Yet across multiple experiments employing fully predictive (100% valid) congruency cues with ample preparation time (2000 ms CSI)—conditions that previously yielded cue benefits (e.g., Bugg & Smallwood, 2016)—we observed no performance improvements attributable to the cues. The sole exception occurred in Experiment 4, where brief stimulus presentation increased task difficulty and produced a significant cue benefit, albeit exclusively in congruent trials. However, cue benefits found in congruent trials might not reflect an genuine upregulation of proactive control, as participants can instead employ strategic shortcuts, such as allowing their responses to be directly driven by the distractors (Wühr & Kunde, 2008). Our findings imply that participants resorted to such a strategy only when task demands became sufficiently challenging. Intriguingly, participants refrained from using this type of strategy in the other experiments, despite its’ potential to facilitate responses on 50% of trials. A plausible explanation is that the cognitive cost of switching attention between task-irrelevant dimensions (for congruent trials) and task-relevant dimensions (for incongruent trials) outweighed the strategic benefits.
One may argue that the absence of cue benefits on incongruent trials in manual Stroop tasks, in contrast to vocal Stroop tasks, is simply due to a floor effect, as prior studies have shown that congruency effects are typically smaller in manual than in vocal versions (Augustinova et al., 2019; Kinoshita et al., 2017). A smaller congruency effect might leave little room for cues to exert an influence. However, this explanation is unlikely to be sufficient for two reasons. First, the congruency effects across our experiments were robust and substantial, ranging from 130 to 170 ms. Second, the fact that we also observed a strong CSE in this same data demonstrate that the effect was large enough to be modulated. Therefore, the absence of a cueing effect in our study cannot be attributed to a lack of room for modulation.
Our conclusion that a non-arbitrary response mapping is insufficient to engage cue-based proactive control can be further supported by considering the distinction between same-hand and different-hand incongruent trials. If this conclusion holds, one would expect cue benefits to be absent for both subtypes of incongruent those involving only stimulus–stimulus overlap (analogous to same-hand conflict) and those involving both stimulus–stimulus and stimulus–response overlap (different-hand conflict). Although we were unable to conduct this specific same-hand versus different-hand analysis in Experiments 2–5 (see Footnote 3), the results from Experiment 1 speak directly to this prediction. In that experiment, we observed a robust reaction time difference between same-response and different-response incongruent trials, confirming that they reflect distinct levels of conflict (see also Burca et al., 2021; Martinon et al., 2024). Crucially, however, cue benefits were absent for both trial types. While the 1 mapping in Experiment 1 may have partially diluted the non-arbitrary nature of the task, the consistent lack of cue benefits across both conflict types aligns with the conclusion that a non-arbitrary S–R mapping is not sufficient to recruit cue-based proactive control.
To account for the general lack of cue utilization, two main possibilities arise. First, volitionally engaging proactive control based merely on abstract cue information might be inherently difficult (Braem et al., 2024). This difficulty becomes apparent when contrasted with the consistently observed CSE in our experiments, which demonstrates that participants adaptively adjusted cognitive control based on recent conflict experience (i.e., conflict adaptation, Botvinick et al., 2001; Gratton et al., 1992; reviewed in Egner, 2007). While experience-driven control seems to emerge naturally from conflict encounters, translating abstract symbolic cue information into attentional strategies may be a substantially more demanding (Yang & Pourtois, 2022). This challenge may in part stem from the ambiguous nature of proactive control implementation. When participants see an incongruent cue, they must determine how to “focus harder” on the task-relevant dimension, yet there may be no clear, universal strategy for achieving this attentional state, and it requires deliberate configuration of attentional settings without the benefit of immediate sensory feedback. This interpretation aligns with visual attention research demonstrating that distractor suppression occurs effectively through learned experience but not through top-down, volitional control (Gaspelin & Luck, 2018; Wang & Theeuwes, 2018; Huang et al., 2021, 2023, 2025; for a review, see Luck et al., 2021; Theeuwes et al., 2022). Thus, the abstract nature of cue information may create an implementation gap that prevents effective proactive control engagement.
Second, given proactive control may be perceived as cognitively costly, participants may tend to avoid this effortful strategy unless sufficiently motivated (Shenhav et al., 2013, 2017). Extant studies have shown that extrinsic incentives, such as monetary rewards, can successfully motivate participants to adopt proactive control strategies and utilize predictive cues more effectively (Bugg et al., 2015, Experiment 4; Chiew & Braver, 2016). According to the expected value of control framework, control allocation is guided by a cost-benefit analysis. Under standard task conditions, the perceived cognitive effort of interpreting the cue, translating it into a control strategy, and maintaining a heightened attentional focus might outweigh the anticipated benefit, especially if reactive control adjustments following stimulus presentation are deemed sufficient to provide the correct response most of the time. Thus, participants may default to a less demanding, more reactive control mode. The results of Experiment 4 lend partial support to this interpretation. Increasing task difficulty (via brief stimulus exposure) likely increased the perceived value of utilizing the cue. However, this might not have sufficiently motivated participants to engage in proactive control over conflict. Instead, participants appeared to rely on a strategic shortcut during congruent trials. This outcome suggests that either the difficulty manipulation was insufficient to offset the cognitive cost of proactive control or that it failed to engage the dopaminergic, reward-based neural pathways that facilitate effortful control allocation (Westbrook et al., 2020).
While external incentives like monetary rewards can motivate proactive control engagement (Bugg et al., 2015, Experiment 4; Chiew & Braver, 2016), we deliberately avoided such manipulations. Instead, we attempted to increase the intrinsic value of informative cues through task difficulty and incongruency probability manipulations. This approach addresses two theoretical concerns. First, it reveals the natural propensity for proactive control without immediate external incentives, arguably more faithfully reflecting most real-world cognitive control scenarios. Second, monetary incentives would not resolve our core theoretical why vocal but not manual Stroop tasks show cue benefits, given that the foundational vocal Stroop studies (Bugg & Smallwood, 2016; Jiménez et al., 2021) succeeded without monetary incentives.
A potential concern is that the current study’s sample sizes were based on the assumption of a large effect size (d=0.92) reported by Bugg and Smallwood (2016). Given the mixed findings in the broader literature, it is likely that the true effect size may be smaller, suggesting our study could be underpowered. While this is a reasonable concern, we believe our findings are robust for two main reasons. First, the null effect of cueing on incongruent trials was consistently observed across five experiments with different mappings, timings, and difficulty levels. Second, this interpretation is further strengthened by our Bayesian analyses, which consistently provided substantial evidence in favor of the null hypothesis across most experiments.
Across our experiments, we employed a 2000 ms CSI based on the findings of Bugg and Smallwood (2016). However, the precise temporal requirements for effective cue utilization remain unclear. The study by Bugg and Smallwood (2016) indicated that engaging proactive control via congruency cues in their vocal Stroop task necessitated a substantial CSI of at least 2000 ms. They observed that the beneficial effects of the cue disappeared when the CSI was shorter (500, 1000, or 1500 ms), suggesting a lengthy duration is needed to translate the cue information into a preparatory control state. However, this finding contrasts notably with results reported by Jiménez et al. (2021). Using a similar vocal Stroop paradigm, they successfully observed a cue benefit with a much shorter CSI of only 750 ms in Experiment 7. Critically, the effectiveness of this shorter preparation window in Jimenez et al.’s study was contingent upon the presence of a relatively long RSI of 2250 ms. When this long RSI was removed (while keeping the 750 ms CSI), the cue benefit vanished in their Experiment 10. These contrasting findings suggest that sufficient time between trials, potentially allowing for the decay of processing from the previous trial or providing adequate unoccupied time for cognitive reconfiguration, might also be a critical factor in enabling participants to effectively prepare for upcoming conflict based on a cue (Bugg et al., in prep). This temporal puzzle represents an important avenue for future research to clarify the boundary conditions under which congruency cues can effectively engage proactive control mechanisms.
Beyond these temporal dynamics, a definitive test of the S-R mapping hypothesis requires addressing the limitations of cross-study comparisons. The findings of the present study rely on comparisons with prior studies using arbitrary mappings. While the consistent null effects observed here mitigate the concern that uncontrolled variables obscured a true effect, a more rigorous test would require manipulating mapping arbitrariness orthogonally within a single experiment. Furthermore, given that non-arbitrary S-R mapping alone was insufficient to produce the cue benefits consistently observed in vocal tasks, a fundamental puzzle why does the vocal Stroop task appear uniquely amenable to cue-based proactive control? Future research should aim to isolate the specific properties of the vocal modality that facilitate this form of cognitive control.
To conclude, our findings demonstrate that non-arbitrary SR mappings are not a sufficient condition for effective cue utilization in conflict tasks. Despite creating optimal conditions for cue usage to emerge, we found no evidence of proactive control engagement via explicit cues, suggesting that its successful application may be highly circumscribed, with the vocal Stroop task representing a potential special case rather than the norm. These findings challenge the prevailing assumption that cue-based proactive control is a readily deployable, domain-general mechanism.