Ready to go: Higher sense of agency enhances action readiness and reduces response inhibition

Sense of agency is the subjective feeling of being in control of one's actions and their effects. Many studies have elucidated the cognitive and sensorimotor processes that drive this experience. However, less is known about how sense of agency influences flexible cognitive and motor control. Here, we investigated the effect of sense of agency on subsequent action regulation using a modified Go/No-Go task. In Experiment 1, we modulated participants' sense of agency by varying the occurrence of action outcomes (present vs. absent) both locally on a trial-by-trial basis and globally in terms of the overall probability of action outcomes within a block of trials (high vs. low). Importantly, we investigated how this manipulation influenced participants' responses to subsequent Go, No-Go, or Free-Choice cues. When participants' previous action led to an outcome (i.e., a happy face) compared with no outcome, they responded more accurately and faster to Go cues, reacted less accurately to No- Go cues, as well as made go decisions more frequently and faster to Free-Choice cues. These effects were even stronger when action outcomes occurred more frequently overall in a given block or in several previous trials. Experiment 2 further demonstrated that the effects of action outcome manipulation on subsequent action regulation were independent of the emotional valence of the action outcome (i.e., a happy or an angry face). Our results suggest that a higher sense of agency as induced by the presence of action outcomes enhanced action readiness and suppressed response inhibition. These findings highlight the impact of the control felt on the control used in action regulation, thereby providing new insights into the functional significance of the sense of agency on human behavior.


Introduction
Sense of agency refers to the subjective feeling of controlling one's own actions and their effects (Haggard, 2017;Haggard & Eitam, 2015). Abundant empirical and theoretical studies have revealed the cognitive and sensorimotor processes that drive and modulate sense of agency, including voluntary motor commands, action preparation, selection between action alternatives, and predictable sensory feedback (Haggard, 2017;Malik, Galang, & Finger, 2022). However, research on the opposite direction, i.e., whether and how sense of agency modulates upcoming actions and flexible behavior, is still sparse. Determining how sense of agency potentially affects our readiness to act would be an important step in finding out how our subjective states influence our objective ability to perform goal-directed actions.
Prior studies that aimed to investigate the effects of sense of agency on other cognitive processes typically manipulate the presence and/or the predictability of perceivable action effects to indirectly manipulate participants' sense of agency or feeling of control (Eitam, Kennedy, & Tory Higgins, 2013;Gentsch & Schütz-Bosbach, 2015;Hemed, Bakbani-Elkayam, Teodorescu, Yona, & Eitam, 2020;Hemed, Karsh, Mark-Tavger, & Eitam, 2022;Karsh, Eitam, Mark, & Higgins, 2016;Penton, Wang, Coll, Catmur, & Bird, 2018). The modulation effects of these factors on the sense of agency have been well documented. The comparator model, the probably most widely accepted theory of the sense of agency, states that a sensory prediction is generated from an efference copy of a motor command, and is compared with the actual sensory feedback; a sense of agency arises if they match and diminishes if they mismatch (Blakemore, Wolpert, & Frith, 1998, 2002Frith, Blakemore, & Wolpert, 2000). According to this model, being able to produce intended or predicted effects is thought to induce a high sense of agency; conversely, not being able to produce predicted effects is thought to be associated with little or no sense of agency. In addition, having a higher probability of producing action outcomes is generally considered to increase sense of agency; in contrast, having a lower probability of producing action outcomes is considered to reduce sense of agency. These hypotheses have been supported by many empirical studies using implicit (sensorimotor) and/or explicit measurements (judgements) of agency (Caspar, Desantis, Dienes, Cleeremans, & Haggard, 2016;Ebert & Wegner, 2010;Karsh et al., 2016;Penton et al., 2018;Sato & Yasuda, 2005;Villa, Tidoni, Porciello, & Aglioti, 2021).
Previous relevant studies have largely focused on the effect of sense of agency on action selection (i.e., which action to execute) and the efficiency of action execution (Wen & Imamizu, 2022). For example, participants responded faster when their response was consistently followed by an immediate effect, compared to when no such effect appeared or when it was followed by a time-lagged (300 or 600 ms) effect (Eitam et al., 2013). Moreover, buttons that had a high probability of causing a visual outcome were more likely to be selected and were pressed faster than buttons associated with no chance of causing a perceivable effect, even though these action outcomes were taskirrelevant and valence-neutral (Karsh & Eitam, 2015). Recent studies further found that this facilitation effect was sensitive to the effectiveness of the motor response, that is, how likely the response was to evoke a perceivable effect Tanaka, Watanabe, & Tanaka, 2021). These findings suggest that actions associated with a strong sense of agency or control are preferred and are executed more fluently. More importantly, Hemed et al. (2022) revealed the different effects of sensorimotor and judgement-based aspects of agency on different aspects of responding. Specifically, evaluations of a response's effectiveness were suggested to be driven by at least two different processes: one is based on a sensorimotor (implicit) process and the confirmation of sensorimotor prediction that reinforces response execution (as measured by response speed); the other relies on a conceptual (explicit) judgement of agency and seems to affect response selection (as measured by response frequency).
While fluent action execution is an important part of goal-directed actions, humans need also be able to suppress automatic action tendencies when they are not consistent with their current goals or not appropriate in a given situation. However, it is currently unknown whether sense of agency influences the ability to suppress unwanted or inappropriate actions (i.e., subsequent response inhibition). Response inhibition is considered a hallmark of executive function and cognitive control, and it supports flexible and goal-directed behavior in everchanging environments (Bari & Robbins, 2013;Chambers, Garavan, & Bellgrove, 2009). Previous studies have shown the sense of agency to be modulated under high-level physical effort or cognitive load, although the results are inconclusive. Specifically, when a current trial required a high level of effort, such as an incongruent or error trial, the sense of agency in that trial decreased (Sidarus & Haggard, 2016;Vastano, Pozzo, & Brass, 2017;Wang, Damen, & Aarts, 2018). However, another study found the opposite effect (Van den Bussche, Alves, Murray, & Hughes, 2020). Additionally, it has been shown that the exertion of more effort in the previous trial led to a higher sense of agency in the current trial (Di Costa, Théro, Chambon, & Haggard, 2018;Wang et al., 2018). In contexts requiring high effort (e.g., across a block of trials), some studies found a decrease in the sense of agency (Hon, Poh, & Soon, 2013;Howard, Edwards, & Bayliss, 2016;Potts & Carlson, 2019) while others observed an increase in the sense of agency (Demanet, Muhle-Karbe, Lynn, Blotenberg, & Brass, 2013). These findings on the one hand imply a close relationship between the feeling of being in control and the engagement of cognitive control but on the other hand also suggest a functional distinction. The present study aimed to explore the effect of sense of agency on subsequent action readiness and response inhibition, which might advance our understanding of how the subjective control feeling modulates the actual control engagement in action regulation and in this way, may provide new insights into the functional significance of the sense of agency in goal-directed behavior.
To this end, we adopted a modified Go/No-Go task in which a pair of two trials included two action stimuli. The Go/No-Go paradigm is a widely-used paradigm for measuring action regulation, particularly response inhibition (e.g., Wessel, 2018). In our task, the first action stimulus was always a Go cue (i.e., the inducement trial), while the second action stimulus was a Go, a No-Go, or a Free-Choice cue (i.e., the test trial). Participants were instructed to perform speeded keypress actions to a Go cue, withhold responses to a No-Go cue, or make free choices whether to execute or inhibit a keypress when presented with a Free-Choice cue. Similar to previous studies (Eitam et al., 2013;Hemed et al., 2020), we manipulated participants' sense of agency in the inducement trials by having them perform actions that did or did not result in visual effects. This allowed us to test the immediate effect of the "local" outcome presence on participants' performance in the next trial (i.e., the test trial). We also manipulated the likelihood (high vs. low) of obtaining action outcomes at the block level, to further explore the longer-lasting effect of "global" outcome frequency. We used positively valenced stimuli (i.e., happy face) as action outcomes since several previous studies suggest that positive compared to neutral outcomes might be particularly effective in inducing a high sense of agency (for review, see Kaiser, Buciuman, Gigl, Gentsch, & Schütz-Bosbach, 2021).
We had three hypotheses. First, we expected that the presence of an action outcome in the inducement trial would enhance action readiness in the following test trial, resulting in faster and more accurate responses to Go cues (Kaiser & Schütz-Bosbach, 2019). Similar effects have been observed in previous studies using different paradigms and neutral action outcomes (Eitam et al., 2013;Hemed et al., 2022;Karsh & Eitam, 2015). Second, we hypothesized a detrimental effect of outcome presence on subsequent response inhibition, given that an increase in readiness to act can coincide with a decrease in the ability to inhibit action (Albayay, Castiello, & Parma, 2019;van Peer, Gladwin, & Nieuwenhuys, 2019). External signals and internal decisions have been both found to affect response inhibition (Parkinson & Haggard, 2014). Successful action suppression during No-Go trials was assumed to reflect externally generated inhibition, while a lower frequency of choosing to act and slower reaction times during Free-Choice trials indicate stronger internally generated inhibition (i.e., intentional inhibition; Parkinson & Haggard, 2014). Participants' responses to Free-Choice trials also reflect their "bias" for either action execution or inhibitory tendency. We predicted that the presence of an action outcome would lead to lower accuracy rates during subsequent No-Go trials as well as a higher likelihood of choosing to act and faster choices during subsequent Free-Choice trials, indicating an influence on both types of response inhibition. Last but not least, inspired by the finding that not only the immediate context (i.e., outcome occurrence on trial n-1) but also the more distant context (e.g., number of outcome occurrences on trial n-4 through n-2) influenced participants' response speed , we hypothesized that the global outcome frequency in a given block or several previous trials would further modulate the effect caused by the local outcome presence.

Participants
Twenty-seven healthy participants (15 females; mean age: 26.04 ± 3.48 years; range: 19-35 years), with normal or corrected-to-normal vision, were recruited for Experiment 1. Large effect sizes were observed in reaction times between the "no action effect" and the "immediate action effect" condition in related previous studies (Hemed et al., 2022;Karsh et al., 2016). A post-hoc power analysis for Experiment 1, conducted using the MorePower software (Campbell & Thompson, 2012), indicates that a sample of 27 is adequate for detecting a large (η p 2 = 0.25) effect in a 2 × 2 within-subjects design with a power of 0.80 and α of 0.05. All participants provided written informed consent and received financial compensation for their participation. The procedures were approved by the local ethics committee at the Department of Psychology of LMU Munich in accordance with the Declaration of Helsinki.

Materials and apparatus
A set of 42 colored photographs from the validated NimStim database (Tottenham et al., 2009) was used as action outcomes in this experiment. These photographs show happy facial expressions of 42 actors (18 females, 24 males). Go cues (i.e., a black rectangle), No-Go cues (i.e., a black rectangle with a grey cross), Free-Choice cues (i.e., a grey rectangle with black borders), as well as the photos as action outcomes were presented in the same size of 400 × 514 pixels (width × height) with a grey background on a 24-in. monitor (refresh rate: 60 Hz; resolution: 1920 × 1080 pixels) at a viewing distance of approximately 65 cm. The experiment was implemented using the Presentation software (Neurobehavioral Systems, Inc.).

Design and procedure
A pair of two trials included two action stimuli (see Fig. 1). Unbeknownst to the participants, the first action stimulus was always a Go cue (i.e., the inducement trial), while the second action stimulus was a Go, a No-Go, or a Free-Choice cue (i.e., the test trial), presented with equal probability (i.e., 33.33%). Participants were instructed to (1) press the "down arrow" key using their right index finger as quickly as possible in response to the Go cue; (2) withhold the keypress action in response to the No-Go cue; (3) make a free, spontaneous decision regarding whether to press the key or inhibit the keypress action in response to the Free-Choice cue. Each action stimulus remained on the screen until the participant pressed the button or the stimulus had been presented for 350 ms. An error message was presented for 800 ms if the action response was incorrect (i.e., did not press the key in Go trials or pressed the key in No-Go trials). Such message never appeared in Free-Choice trials, since both pressing the key and not pressing the key were "correct" responses. A central fixation dot was presented for 1300-1700 ms during the inter-stimulus interval.
This experiment adopted a 2 (Outcome Presence: present vs. absent) × 2 (Outcome Probability: high vs. low) within-subjects design (see the table in Fig. 1). The factor "Outcome Presence" was manipulated at the trial-by-trial level. Specifically, for each inducement trial, either a happy face (800 ms) or no visual stimulus at all was presented after participants' keypress action. Thus, participants either had the experience that their action led to a positive effect, or they experienced that their action did not lead to any perceivable outcome. The factor "Outcome Probability" was manipulated at the block level. For each block, participants either had a high (75%) or a low (25%) probability of receiving action outcomes overall. Participants were explicitly informed about the respective outcome probability at the beginning of each block. Specifically, there were two different types of blocks, either having 75% of inducement trials with action outcomes and 25% of inducement trials without action outcomes, or having 25% of inducement trials with action outcomes and 75% of inducement trials without action outcomes. Please note that, unbeknownst to the participants, if a happy face was presented on the inducement trial, participants would also receive a happy face as the outcome of the keypress action in response to the Go cue or the Free-Choice cue during the following test trial (i.e., in case participants chose to act). In this way, we kept the test trial similar to the inducement trial, and thus excluded the possibility that participants treated the inducement trial and the test trial differently and separately over the course of the experiment.
There were 12 blocks for each block type, and 48 pairs of trials for each block. Please note that only when participants made a correct keypress action during the inducement trial, the subsequent test trial would appear, and in this way represented a complete action repetition sequence comprised of a pair of inducement and test trials, which then entered data analysis. Furthermore, only when a correct response was made on the inducement trial was the trial "counted" during data collection. In other words, a block was completed only when participants made correct responses in 48 inducement trials (unbeknownst to Fig. 1. Schema of the modified Go/No-Go task in Experiment 1 and Control Experiment. the participants). Therefore, after deleting the "uncounted" inducement trials (i.e., trials with incorrect responses), always 48 pairs of trials in each block entered further analysis. This step resulted in the exclusion of 8.43 ± 6.25% of trials (Mean ± SD; range: 1.29-24.26% of trials) on average per participant. The uneven proportion of trials with or without action outcomes in each block (e.g., 75% trials vs. 25% trials in blocks having a high probability of obtaining action outcomes) resulted in the unbalanced number of trials in the four different experimental conditions based on the factors "Outcome Presence" and "Outcome Probability". The Go, No-Go, and Free-Choice trials accounted for 33.33% of the test trials per condition, respectively.
To become familiar with the experimental procedures, participants completed two practice sessions (one for each block type; each including 12 pairs of trials) before the actual experiment started. All blocks and trial types were presented in random order. Participants received visual feedback on their accuracy rate and mean reaction time in Go trials as well as the accuracy rate in No-Go trials after each block during the selfpaced inter-block rest.

Manipulation check
As a manipulation check for the sense of agency, we conducted a control experiment that used the same design as Experiment 1 but included explicit judgements of agency (sample size: 22; 13 females; mean age: 25.14 ± 2.97 years; range: 22-33 years; with normal or corrected-to-normal vision). Specifically, participants were required to rate how much control they felt over the face on a visual analogue scale (i.e., 0-no control to 100-total control) at the end of a subset of inducement trials (1/4 trials per block; randomly selected; followed by Go, No-Go, or Free-Choice test trials with equal probability) and also at the end of each block. In total, participants rated their sense of agency for 72 inducement trials in each of the four conditions, and also for 12 blocks with high probability of getting action outcomes and 12 blocks with low probability of getting action outcomes.

Statistical analysis
2.1.5.1. Behavioral performance in experiment 1. Descriptive statistics are provided for reaction times in Go trials and Free-Choice trials, accuracy rates in Go trials and No-Go trials, as well as response rates (i.e., the relative frequency at which participants chose to press the button) in Free-Choice trials (see Supplementary Table 1). Notably, only test trials were analyzed since inducement trials were used to manipulate participants' sense of agency only. For the analysis of reaction times in Go trials, incorrect trials (i.e., did not press the key in test trials) were removed (15.02 ± 8.83% of trials on average per participant). For the analysis of reaction times in Free-Go trials, trials without keypress response were removed (35.61 ± 15.40% of trials on average per participant). None of the participants always chose not to act or always chose to act in Free-Choice trials (minimum response rate: 0.10; maximal response rate: 0.96; see Supplementary Table 2). Separate twoway repeated measures analysis of variances (ANOVA) with two withinsubjects factors (Outcome Presence and Outcome Probability) were conducted on these dependent variables. Data were averaged for each of the four conditions, yielding a single value per condition for each participant, which then entered the ANOVA analysis. The number of trials per condition for each analysis is summarized in Supplementary  Table 3. Additionally, a three-way repeated measures ANOVA including a third within-subjects factor (Trial Type: Go vs. No-Go trials) was performed on accuracy rates, in order to explore the effect of sense of agency on motor tendency (i.e., a preference for go or no-go responses). Partial eta-squared (η p 2 ) was calculated to reflect the effect size of the Ftests. Post-hoc pairwise comparisons were conducted and the Holm correction method was applied for multiple comparisons when there was a significant interaction between factors. Unlike the p-value in frequentist hypothesis testing, the Bayes Factors (e.g., the BF 10 value) in Bayesian hypothesis testing can indicate how much more likely the alternative hypothesis is than the null hypothesis . Therefore, we also reported BF 10 values from the corresponding Bayesian repeated measures ANOVAs. A BF 10 between 1.00 and 3.00 indicates anecdotal evidence, a BF 10 between 3.00 and 10.00 indicates moderate evidence, and a BF 10 >10.00 indicates strong evidence for the alternative hypothesis; in contrast, a BF 10 between 0.33 and 1.00 indicates anecdotal evidence, a BF 10 between 0.10 and 0.33 indicates moderate evidence, and a BF 10 smaller than 0.10 indicates strong evidence for the null hypothesis . All analyses were performed in the JASP software (version 0.17.0.0; JASP Team, 2023). In addition, following the analyses by Hemed et al. (2020), we performed a second set of analyses based on a 2 (Outcome Presence: present vs. absent) × 2 (Previous Outcome Frequency: 3-4 times vs 0 times) within-subjects design. The factor "Previous Outcome Frequency" refers to the number of times the positive outcome had occurred in the previous two pairs of trials. Notably, the number of times the positive outcome had occurred in the previous two pairs of trials can be 0, 1, 2, 3, or 4; whereas, we selected two extreme cases only for comparison (3-4 times vs. 0 times), in order to show the potential effect more clearly. In addition, we binned the trials with 3 and 4 outcome events in the previous two pairs of trials together due to the limited trial number. Descriptive statistics for all dependent variables are summarized in Supplementary Table 4. Separate two-way repeated measures ANOVAs with two within-subjects factors (Outcome Presence and Previous Outcome Frequency) were conducted on the dependent variables, and a three-way repeated measures ANOVA including a third within-subjects factor (Trial Type: Go vs. No-Go trials) was performed on accuracy rates. The number of trials per condition for each analysis is summarized in Supplementary Table 5.

Agency ratings in control experiment. Two-way repeated measures ANOVAs with two within-subjects factors (Outcome Presence and
Outcome Probability or Outcome Presence and Previous Outcome Frequency) were conducted on agency ratings obtained at the end of inducement trials. A paired samples t-test was conducted on agency ratings obtained at the end of each block. The effect size was estimated by Cohen's d.
For reaction times in Free-Choice trials, the two-way repeated measures ANOVA showed significant main effects of Outcome Presence, F(1,26) = 53.46, p < .001, η p 2 = 0.67, BF 10 > 10 6 , and Outcome Probability, F(1,26) = 23.87, p < .001, η p 2 = 0.48, BF 10 > 10 3 . There was also a significant interaction between Outcome Presence and Outcome Probability, F(1,26) = 13.78, p < .001, η p 2 = 0.35, BF 10 > 10 2 . Further analyses showed that, participants responded faster to Free-Choice cues when their previous action led to a positive outcome compared with no outcome (both ps < 0.001). This effect was stronger in blocks having a high probability of obtaining positive outcomes compared with blocks having a low probability of obtaining positive outcomes (see Fig. 2B).
For accuracy rates in Go trials, the two-way repeated measures ANOVA showed significant main effects of Outcome Presence, F(1,26) = 55.53, p < .001, η p 2 = 0.68, BF 10 > 10 10 , and Outcome Probability, F (1,26) = 66.94, p < .001, η p 2 = 0.72, BF 10 > 10 9 . There was also a significant interaction between Outcome Presence and Outcome Probability, F(1,26) = 43.60, p < .001, η p 2 = 0.63, BF 10 > 10 5 . Further analyses showed that, participants responded more accurately to Go cues when their previous action led to a positive outcome compared with no outcome (both ps < 0.001). This effect was stronger in blocks having a high probability of obtaining positive outcomes compared with blocks having a low probability of obtaining positive outcomes (see Fig. 2C).
For accuracy rates in No-Go trials, the two-way repeated measures ANOVA showed significant main effects of Outcome Presence, F(1,26) = 122.40, p < .001, η p 2 = 0.83, BF 10 > 10 9 , and Outcome Probability, F (1,26) = 14.10, p < .001, η p 2 = 0.35, BF 10 > 10 2 . There was also a significant interaction between Outcome Presence and Outcome Probability, F(1,26) = 7.51, p = .011, η p 2 = 0.22, BF 10 = 25.21. Further analyses showed that, participants responded less accurately to No-Go cues when their previous action led to a positive outcome compared with no outcome (both ps < 0.001). This effect was stronger in blocks having a high probability of obtaining positive outcomes compared with blocks having a low probability of obtaining positive outcomes (see Fig. 2D).
Moreover, the analysis of accuracy rates in both Go and No-Go trials, using a three-way repeated measures ANOVA, showed a significant interaction between Outcome Presence, Outcome Probability, and Trial Type, F(1,26) = 29.23, p < .001, η p 2 = 0.53, BF 10 > 10 7 . As we were interested in the performance differences between Go and No-Go trials in each experimental condition, we further analyzed the interaction by focusing on the effect of Trial Type for each level of the other two factors (i.e., Outcome Presence and Outcome Probability). The results showed that, (1) in blocks having a high probability of obtaining positive outcomes, participants' accuracy rates were significantly higher in Go trials compared with No-Go trials when their previous action led to a positive outcome (p < .001); however, participants' accuracy rates did not differ significantly between Go trials and No-Go trials when their previous action led to no outcome (p = .560); (2) in blocks having a low probability of obtaining positive outcomes, participants accuracy rates were significantly higher in Go trials compared with No-Go trials when their previous action led to a positive outcome (p < .001) but also when their previous action led to no outcome (p < .001; see Supplementary  Fig. 1A).
For response rates (i.e., the relative frequency at which participants chose to press the button) in Free-Choice trials, the two-way repeated measures ANOVA showed significant main effects of Outcome Presence, F(1,26) = 111.94, p < .001, η p 2 = 0.81, BF 10 > 10 11 , and Outcome Probability, F(1,26) = 30.77, p < .001, η p 2 = 0.54, BF 10 > 10 6 . There was also a significant interaction between Outcome Presence and Outcome Probability, F(1,26) = 32.97, p < .001, η p 2 = 0.56, BF 10 > 10 4 . Further analyses showed that, participants responded more frequently to Free-Choice cues when their previous action led to a positive outcome compared with no outcome (both ps < 0.001). This effect was stronger in blocks having a high probability of obtaining positive outcomes compared with blocks having a low probability of obtaining positive outcomes (see Fig. 2E).
In summary, when participants' previous action led to a positive outcome compared with no outcome, they responded faster and more accurately to Go cues, responded less accurately to No-Go cues, as well as responded faster and more frequently to Free-Choice cues. These effects were stronger when the global probability of getting positive outcomes in a given block was high compared with low. In addition, participants had higher accuracy rates in Go trials compared with No-Go trials (i.e., a preference for go over no-go responses) when their previous action led to a positive outcome or when their previous action unsurprisingly did not lead to any outcome. In contrast, participants had comparable accuracy rates in Go and No-Go trials (i.e., no obvious preference for go over no-go responses) when the global probability of getting positive outcomes in a given block was high but their previous action surprisingly did not lead to any outcome.

Results based on the factors "outcome presence" and "Previous
Outcome Frequency". For reaction times in Go trials, the two-way repeated measures ANOVA showed significant main effects of Outcome Presence, F(1,26) = 48.22, p < .001, η p 2 = 0.65, BF 10 > 10 7 , and Previous Outcome Frequency, F(1,26) = 14.43, p < .001, η p 2 = 0.36, BF 10 > 10 3 . There was also a significant interaction between Outcome Presence and Previous Outcome Frequency, F(1,26) = 18.61, p < .001, η p 2 = 0.42, BF 10 > 10 3 . Further analyses showed that, participants responded faster to Go cues when their previous action led to a positive outcome compared with no outcome (both ps < 0.01). This effect was stronger when positive outcomes had occurred three to four times compared with zero times in the previous two pairs of trials (see Fig. 3A).
For reaction times in Free-Choice trials, the two-way repeated measures ANOVA showed significant main effects of Outcome Presence, Further analyses showed that, participants responded faster to Free-Choice cues when their previous action led to a positive outcome compared with no outcome (both ps < 0.05). This effect was stronger when positive outcomes had occurred three to four times compared with zero times in the previous two pairs of trials (see Fig. 3B).
For accuracy rates in Go trials, the two-way repeated measures ANOVA showed significant main effects of Outcome Presence, Further analyses showed that, if positive outcomes had occurred three to four times in the previous two pairs of trials, participants responded more accurately to Go cues when their previous action led to a positive outcome compared with no outcome (p < .001). Whereas, if positive outcomes had occurred zero times in the previous two pairs of trials, their accuracy rates in Go trials did not differ significantly when their previous action led to a positive outcome compared with no outcome (p = .148; see Further analyses showed that, participants responded less accurately to No-Go cues when their previous action led to a positive outcome compared with no outcome (both ps < 0.001). This effect was stronger when positive outcomes had occurred three to four times compared with zero times in the previous two pairs of trials (see Fig. 3D).
Moreover, the analysis of accuracy rates in both Go and No-Go trials, using a three-way repeated measures ANOVA, showed a significant interaction between Outcome Presence, Previous Outcome Frequency, and Trial Type, F(1,26) = 45.48, p < .001, η p 2 = 0.64, BF 10 > 10 13 . As we were interested in the performance differences between Go and No-Go trials in each experimental condition, we further analyzed the interaction by focusing on the effect of Trial Type for each level of the other two factors (i.e., Outcome Presence and Previous Outcome Frequency). The results showed that, (1) when positive outcomes had occurred three to four times in the previous two pairs of trials, participants' accuracy rates were significantly higher in Go trials compared with No-Go trials when their previous action led to a positive outcome (p < .001); however, participants' accuracy rates were significantly lower in Go trials compared with No-Go trials when their previous action led to no outcome (p = .043); (2) when positive outcomes had occurred zero times in the previous two pairs of trials, participants' accuracy rates were significantly higher in Go trials compared with No-Go trials when their previous action led to a positive outcome (p < .001) but also when their previous action led to no outcome (p < .001; see Supplementary Fig. 1B).
For response rates (i.e., the relative frequency at which participants chose to press the button) in Free-Choice trials, the two-way repeated Further analyses showed that, participants responded more frequently to Free-Choice cues when their previous action led to a positive outcome compared with no outcome (both ps < 0.01). This effect was stronger when positive outcomes had occurred three to four times compared with zero times in the previous two pairs of trials (see Fig. 3E).
In summary, when participants' previous action led to a positive outcome compared with no outcome, they responded faster and more accurately to Go cues, responded less accurately to No-Go cues, as well as responded faster and more frequently to Free-Choice cues. These effects were stronger when positive outcomes had occurred three to four times compared with zero times in the previous two pairs of trials. In addition, participants had higher accuracy rates in Go trials compared with No-Go trials (i.e., a preference for go over no-go responses) when their previous action led to a positive outcome or when their previous action unsurprisingly did not lead to any outcome. In contrast, participants had lower accuracy rates in Go trials compared with No-Go trials (i.e., a preference for no-go over go responses) when positive outcomes had occurred three to four times in the previous two pairs of trials but their previous action surprisingly did not lead to any outcome. Participants also reported higher agency ratings when they had a high compared with low probability of obtaining action outcomes in a given block (see Fig. 4A). However, the interaction between Outcome Presence and Outcome Probability was not significant, F(1,21) = 0.07, p = .788, η p 2 < 0.01, BF 10 = 1.02.
For agency ratings obtained at the end of each block, the paired samples t-test showed that participants reported higher agency ratings for blocks having a high probability of obtaining action outcomes compared with blocks having a low probability of obtaining action outcomes, t(21) = 4.79, p < .001, Cohen's d = 1.02, BF 10 > 10 2 (see Fig. 4B).
In summary, participants had a higher sense of agency (i.e., feeling more in control) when their previous action led to a positive outcome compared with no outcome and also when the global outcome probability in a given block was high compared with low.

Results based on the factors "outcome presence" and "Previous
Outcome Frequency". For agency ratings obtained at the end of inducement trials, the two-way repeated measures ANOVA showed significant main effects of Outcome Presence, F(1,21) = 20.34, p < .001, η p 2 = 0.49, BF 10 > 10 2 , and Previous Outcome Frequency, F(1,21) = 29.14, p < .001, η p 2 = 0.58, BF 10 > 10 2 . Participants reported higher agency ratings when their previous action led to a positive outcome compared with no outcome. Participants also reported higher agency ratings when positive outcomes had occurred three to four times compared with zero times in the previous two pairs of trials (see Fig. 4C). However, the interaction between Outcome Presence and Previous Outcome Frequency was not significant, F(1,21) = 3.34, p = .082, η p 2 = 0.14, BF 10 = 2.94.
In summary, participants had a higher sense of agency (i.e., feeling more in control) when their previous action led to a positive outcome compared with no outcome and also when their action had resulted in positive outcomes frequently in several previous trials.

Discussion
Results of agency ratings in Control Experiment suggest that our experimental manipulation affected participants' sense of agency as intended. Specifically, participants had a higher sense of agency (i.e., feeling more in control) when their previous action led to a positive outcome (i.e., a happy face) compared with no outcome and also when positive action outcomes occurred frequently in a given block or in several previous trials. In other words, both the "local" occurrence of an action outcome and the "global" high probability of getting action outcomes led to an enhancement of the sense of agency. These effects are in line with previous literature. For example, higher feeling of control was reported when participants' action was always followed by a visual effect compared with no effect (Karsh et al., 2016) and also when the likelihood of an outcome following an action was increased (Penton et al., 2018).
Importantly, results of behavioral performance in Experiment 1 indicate that the "local" occurrence of a positive outcome following an action influenced participants' subsequent motor responses, manifested by higher accuracy rates and faster reaction times to Go cues, lower accuracy rates to No-Go cues, as well as higher response rates and faster reaction times to Free-Choice cues. In other words, the "local" outcome occurrence enhanced action readiness and suppressed response inhibition. Furthermore, these effects were stronger when positive action outcomes occurred frequently in a given block or in several previous trials, suggesting a modulation effect of the "global" outcome probability.
Together our results of behavioral performance and agency ratings both suggest that the sense of agency does play a functional role in action regulation. Specifically, the immediate, trial-by-trial enhancement of sense of agency caused by the "local" presence of an outcome led to an enhancement of action readiness and at the same time reduced response inhibition in subsequent trials. These effects were further amplified by a more longer-lasting enhancement of the sense of agency as induced by the "global" outcome frequency in a given block or in several previous trials. Under the assumption that No-Go and Free-Choice responses are reflective of internally and externally generated inhibition respectively (Parkinson & Haggard, 2014), our results indicate that when participants felt more in control over the keypress action, they not only performed worse at canceling the prepotent keypress action that is inappropriate for No-Go cues but also consciously preferred to repeat this keypress response when the Free-Choice cue appears. In addition, participants were more accurate in performing keypress actions than withholding them (i.e., a preference for go over no-go responses) in conditions with "local" outcome presence, and this tendency for motor execution was enhanced by the "global" outcome frequency in a given block or in several previous trials. This finding thus suggests an increased readiness for action at the cost of inhibitory control when sense of agency was relatively high. In contrast, participants did not have an obvious preference for go or no-go responses (based on the factor "Outcome Probability") or even had a preference for no-go over go responses (based on factors "Previous Outcome Frequency"), in the condition with high "global" outcome frequency but surprising outcome absence at the "local" level. This finding may indicate that a reduced sense of agency shifts participants' tendency from preparedness for action towards increased inhibitory control or general passivity.
Overall, these effects may also be explained by motivational factors related to the positively valenced action outcomes used in Experiment 1. In other words, the prospect of a positive outcome on a given trial might have induced the strong readiness for action, whereas the absence of a (rewarding) positive outcome led to the opposite effects. Therefore, in a second experiment, we further investigated the potential role of the emotional valence of action outcomes on our findings. To this end, we used the same design as in Experiment 1 but added a new condition in which the action outcome was comprised of a negatively valenced stimulus (i.e., an angry face). In the interest of limiting the overall length of the experiment, we presented Go and No-Go cues only in this version, and thus, did not include Free-Choice cues. If the presence of a positive outcome and the presence of a negative outcome (compared with no outcome) both resulted in similar effects on subsequent action regulation, it can be speculated that the effectiveness of the motor action in terms of producing a perceivable effect per se is a driving factor of our results. Otherwise, the outcome valence might be the key determinant of the observed effects.

Participants
A group of 30 healthy participants was recruited for Experiment 2. None of these participants took part in Experiment 1. Two participants were excluded from data analysis, because data collection was either not completed or data analysis revealed extreme values in behavioral performance (> 3 standard deviations from sample mean), leaving a final sample of 28 (18 females; mean age: 27.07 ± 7.30 years; range: 18-48 years). A post-hoc power analysis for Experiment 2, conducted using the MorePower software (Campbell & Thompson, 2012), indicates that a sample of 28 is adequate for detecting a large (η p 2 = 0.16) effect of the 3level within-subjects factor with a power of 0.80 and α of 0.05. Other experimental requirements were identical to Experiment 1.

Materials and apparatus
In addition to the photographs of happy facial expressions used in Experiment 1, another set of 42 colored photographs showing angry facial expressions was selected from the same database (Tottenham et al., 2009). The action stimuli (the Go cue and the No-Go cue) and the apparatus were identical to Experiment 1.

Design and procedure
This experiment adopted a 3 (Outcome Presence: absent vs. present positive vs. present negative) × 3 (Outcome Probability: high probability of no outcome vs. high probability of positive outcome vs. high probability of negative outcome) within-subjects design (see Fig. 5). The factor "Outcome Presence" was manipulated at the trial-by-trial level. Specifically, for each inducement trial, either a happy face, a negative face, or no visual stimulus at all was presented after participants' keypress action. The factor "Outcome Probability" was manipulated at the block level. Specifically, there were three different types of blocks (4 blocks for each block type, and 48 pairs of trials for each block), randomly presented. In each of the three block types, one of the possible action outcomes (positive/negative/no outcomes) was more likely to occur (50% of inducement trials), while the other two types of action outcomes were less likely to occur (25% of inducement trials each).
Consistent with Experiment 1, only when a correct response was made on the inducement trial was the trial "counted" during data collection. This step resulted in the exclusion of 11.81 ± 6.66% of trials (Mean ± SD; range: 1.37-25.29% of trials) on average per participant. The uneven proportion of trials with positive, negative or no outcomes in each block (e.g., 50% trials vs. 25% trials vs. 25% trials in blocks having a high probability of obtaining positive outcomes) resulted in the unbalanced number of trials in the nine different experimental conditions based on the factors "Outcome Presence" and "Outcome Probability". The Go and No-Go trials accounted for 50% of the test trials per condition, respectively. To become familiar with the experimental procedures, participants completed three practice sessions (one for each block type; each including 8 pairs of trials) before the actual experiment started. Other experimental procedures were identical to Experiment 1.

Statistical analysis
Descriptive statistics are provided for reaction times in Go trials as well as accuracy rates in Go trials and No-Go trials (see Supplementary  Table 6 and 7). For the analysis of reaction times in Go trials, incorrect trials were removed (24.58 ± 11.07% of trials on average per participant). Consistent with Experiment 1, two sets of two-way repeated measures ANOVAs were conducted, either based on the factors "Outcome Presence" and "Outcome Probability" or the factors "Outcome Presence" and "Previous Outcome Frequency". The factor "Previous Outcome Frequency" refers to the number of times the type of outcome in the inducement trial had occurred in the previous two pairs of trials (2-4 times vs. 0 times). We binned the trials with 2, 3, and 4 outcome events in the previous two pairs of trials together due to the limited trial number. The number of trials per condition for each analysis is summarized in Supplementary Table 8 and 9. Additionally, three-way repeated measures ANOVAs including a third within-subjects factor (Trial Type: Go vs. No-Go trials) were performed on accuracy rates, in order to explore the effect of sense of agency on the motor tendency. The Greenhouse-Geisser correction was applied in case of violations of the sphericity assumption. For the sake of brevity, the uncorrected degrees of freedom were reported. Other statistical procedures were identical to Experiment 1.

Results based on the factors "outcome presence" and "outcome probability"
For reaction times in Go trials, the two-way repeated measures ANOVA showed a significant main effect of Outcome Presence, F(2,54) = 72.87, p < .001, η p 2 = 0.73, BF 10 > 10 13 . Further analyses showed that, participants responded faster to Go cues when their previous action led to a positive or negative outcome compared with no outcome (both ps < 0.001), while their reaction times in Go trials did not differ significantly when their previous action led to a positive outcome compared with a For accuracy rates in Go trials, the two-way repeated measures ANOVA showed a significant main effect of Outcome Presence, F(2,54) = 99.68, p < .001, η p 2 = 0.79, BF 10 > 10 15 . The main effect of Outcome Probability was not significant, F(2,54) = 2.33, p = .107, η p 2 = 0.08, BF 10 = 1.50. However, there was a significant interaction between Outcome Presence and Outcome Probability, F(4,108) = 3.55, p = .017, η p 2 = 0.12, BF 10 = 5.77. Further analyses showed that, in any of the three types of blocks (i.e., having a high probability of obtaining positive/negative/no outcomes), participants responded more accurately to Go cues when their previous action led to a positive or negative outcome compared with no outcome (all ps < 0.001). Whereas, in none of the three types of blocks, participants' accuracy rates in Go trials differ significantly when their previous action led to a positive outcome compared with a negative outcome (all ps > 0.05; see Fig. 6B).
For accuracy rates in No-Go trials, the two-way repeated measures ANOVA showed a significant main effect of Outcome Presence, F(2,54) = 61.56, p < .001, η p 2 = 0.70, BF 10 > 10 11 . Further analyses showed that, participants responded less accurately in No-Go trials when their previous action led to a positive or negative outcome compared with no outcome (both ps < 0.001), while their accuracy rates in No-Go trials did not differ significantly when their previous action led to a positive outcome compared with a negative outcome (p = .760; see Fig. 6C). There was also a significant main effect of Outcome Probability, F(2,54) = 6.28, p = .004, η p 2 = 0.19, BF 10 = 4.83. Further analyses showed that, participants responded more accurately in No-Go trials in blocks having a high probability of obtaining positive outcomes, when compared with blocks having a high probability of obtaining no outcome (p < .001), and when compared with blocks having a high probability of obtaining negative outcomes (p = .045). Participants' accuracy rates in No-Go trials did not differ significantly in blocks having a high probability of obtaining negative outcomes compared with blocks having a high probability of obtaining no outcome (p = .147). However, the interaction between Outcome Presence and Outcome Probability was not significant, F(4,108) = 1.62, p = .175, η p 2 = 0.06, BF 10 = 0.95.
Moreover, the analysis of accuracy rates in both Go and No-Go trials, using a three-way repeated measures ANOVA, showed that the interaction between Outcome Presence, Outcome Probability, and Trial Type was not significant, F(4,108) = 1.86, p = .123, η p 2 = 0.06. However, the BF 10 was 6.62, indicating moderate evidence for the interaction effect. As we were interested in the performance differences between Go and No-Go trials in each experimental condition, we further analyzed the interaction by focusing on the effect of Trial Type for each level of the other two factors (i.e., Outcome Presence and Outcome Probability). The results showed that, in any of the three types of blocks (i.e., having a high probability of obtaining positive/negative/no outcomes), (1) participants' accuracy rates were significantly higher in Go trials compared with No-Go trials when their previous action led to a positive or negative outcome (all ps < 0.001); (2) in contrast, participants' accuracy rates were significantly lower in Go trials compared with No-Go trials when their previous action led to no outcome (all ps < 0.001; see Supplementary Fig. 2A).
In summary, when participants' previous action led to an outcome (positive or negative) compared with no outcome, they responded faster and more accurately to Go cues, as well as responded less accurately to No-Go cues. However, their reaction times in Go trials, as well as accuracy rates in Go trials and No-Go trials, did not differ significantly when their previous action led to a positive outcome compared with a negative outcome. In addition, participants had higher accuracy rates in Go trials compared with No-Go trials (i.e., a preference for go over no-go responses) when their previous action led to an outcome, irrespective of the emotional valence (i.e., positive or negative). In contrast, participants had lower accuracy rates in Go trials compared with No-Go trials (i.e., a preference for no-go over go responses) when their previous action led to no outcome. Further analyses showed that, regardless of the frequency of the same type of outcome in the previous two pairs of trials (i.e., two to four times or zero times), participants responded faster to Go cues when their previous action led to a positive or a negative outcome compared with no outcome (all ps < 0.001); whereas, their reaction times in Go trials did not differ significantly when their previous action led to a positive outcome compared with a negative outcome (both ps > 0.05; see Fig. 7A).

Results based on the factors "outcome presence" and "Previous
For accuracy rates in Go trials, the two-way repeated measures ANOVA showed significant main effects of Outcome Presence, F(2,54) = 57.08, p < .001, η p 2 = 0.68, BF 10 > 10 15 , and Previous Outcome Frequency, F(1,27) = 34.45, p < .001, η p 2 = 0.56, BF 10 > 10 10 . There was also a significant interaction between Outcome Presence and Previous Outcome Frequency, F(2,54) = 29.58, p < .001, η p 2 = 0.52, BF 10 > 10 8 . Further analyses showed that, regardless of the frequency of the same type of outcome in the previous two pairs of trials (i.e., two to four times or zero times), participants responded more accurately to Go cues when their previous action led to a positive or a negative outcome compared with no outcome (all ps < 0.01); whereas, their accuracy rates in Go trials did not differ significantly when their previous action led to a positive outcome compared with a negative outcome (both ps > 0.05; see Fig. 7B).
For accuracy rates in No-Go trials, the two-way repeated measures ANOVA showed a significant main effect of Outcome Presence, F(2,54) = 48.49, p < .001, η p 2 = 0.64, BF 10 > 10 9 . Further analyses showed that, participants responded less accurately in No-Go trials when their previous action led to a positive or negative outcome compared with no outcome (both ps < 0.001); whereas, their accuracy rates in No-Go trials did not differ significantly when their previous action led to a positive outcome compared with a negative outcome (p = .959). There was also a significant main effect of Previous Outcome Frequency, F(1,27) = 17.42, p < .001, η p 2 = 0.39, BF 10 = 16.54. Further analyses showed that, participants responded less accurately in No-Go trials when the same type of outcome had occurred two to four times compared with zero times in the previous two pairs of trials (p < .001; see Fig. 7C). However, the interaction between Outcome Presence and Previous Outcome Frequency was not significant, F(2,54) = 0.80, p = .453, η p 2 = 0.03, BF 10 = 0.84.
Moreover, the analysis of accuracy rates in both Go and No-Go trials, using a three-way repeated measures ANOVA, showed a significant interaction between Outcome Presence, Previous Outcome Frequency, and Trial Type, F(2,54) = 17.56, p < .001, η p 2 = 0.39, BF 10 > 10 8 . As we were interested in the performance differences between Go and No-Go trials in each experimental condition, we further analyzed the interaction by focusing on the effect of Trial Type for each level of the other two factors (i.e., Outcome Presence and Previous Outcome Frequency). The results showed that, (1) regardless of the frequency of the same type of outcome in the previous two pairs of trials (i.e., two to four times or zero times), when participants' previous action led to a positive or negative outcome, their accuracy rates were significantly higher in Go trials compared with No-Go trials (all ps < 0.001); (2) in contrast, when participants' previous action did not lead to any outcome and the same type of outcome had occurred two to four times (i.e., outcomes never occurred) in the previous two pairs of trials, their accuracy rates did not differ significantly between Go trials and No-Go trials (p = .949); (3) furthermore, when participants' previous action did not lead to any outcome but the same type of outcome had occurred zero times (i.e., outcomes had occurred two to four times) in the previous two pairs of trials, their accuracy rates were significantly lower in Go trials compared with No-Go trials (p < .001; see Supplementary Fig. 2B).
In summary, when participants' previous action led to an outcome (positive or negative) compared with no outcome, they responded faster and more accurately to Go cues, as well as responded less accurately to No-Go cues. However, their reaction times in Go trials, as well as accuracy rates in Go trials and No-Go trials, did not differ significantly when their previous action led to a positive outcome compared with a negative outcome. In addition, participants had higher accuracy rates in Go trials compared with No-Go trials (i.e., a preference for go over no-go responses) when their previous action led to an outcome, irrespective of the emotional valence (i.e., positive or negative). In contrast, participants had comparable accuracy rates in Go and No-Go trials (i.e., no obvious preference for go over no-go responses) when action outcomes had never occurred in the previous two pairs of trials and their previous action unsurprisingly did not lead to any outcome. Furthermore, participants had lower accuracy rates in Go trials compared with No-Go trials (i.e., a preference for no-go over go responses) when outcomes had occurred two to four times in the previous two pairs of trials but their previous action surprisingly did not lead to any outcome.

Discussion
Consistent with Experiment 1, results in Experiment 2 indicate that the "local" outcome presence led to an enhancement of action readiness and at the same time reduced response inhibition, manifested by higher accuracy rates and faster reaction times to Go cues as well as lower accuracy rates to No-Go cues. In addition, when participants' previous keypress action led to an outcome, they were more accurate in repeating that action than withholding it (i.e., a preference for go over no-go responses); in contrast, when participants' previous action did not result in any perceivable outcome, they had no obvious preference for go or nogo responses or even had a preference for no-go over go responses. This also supports our previous finding indicating that a reduced sense of agency might have shifted participants' tendency from preparedness for action towards increased inhibitory control or general passivity. Importantly, participants' performance did not differ significantly when their previous action led to a positive outcome compared with a negative outcome. That is, participants' action readiness and response inhibition were not modulated by the emotional valence of action outcomes, indicating that our findings are not driven by the rewarding properties of the outcome per se.
Interestingly, the modulation of our results by the "global" outcome frequency in Experiment 2 was much less prominent than in Experiment 1. For instance, in Experiment 1, the facilitation effect of "local" outcome presence on reaction times in Go trials was stronger in blocks having a high probability of getting positive outcomes (i.e., a significant interaction effect), while such facilitation effect did not differ significantly among different types of blocks in Experiment 2 (i.e., no interaction effect). This is probably because the "global" context (i.e., the relative frequency of a particular type of outcome) within a given block in Experiment 2 (50% vs. 25% vs. 25%) was less obvious than in Experiment 1 (75% vs. 25%). The longer-lasting enhancement of the sense of agency caused by the "global" outcome frequency might be much weaker in Experiment 2 than Experiment 1, and thus had less impact on participants' subsequent motor responses.

General discussion
Using a modified Go/No-Go task, the present study investigated the influence of sense of agency on subsequent action regulation. In Experiment 1, we manipulated participants' sense of agency by varying the presence of a visual outcome (i.e., a happy face) at the trial-by-trial level and the overall outcome probability at the block level (high vs. low) for a given motor action, and measured participants' responses to the subsequent Go, No-Go, or Free-Choice cue. Additionally, we conducted a control experiment including subjective judgements of agency as a manipulation check. As predicted, the immediate enhancement of sense of agency caused by the "local" outcome presence enhanced participants' readiness to act while at the same time suppressed their ability to cancel a prepotent action when required to do so. In addition, these effects were further amplified by a longer-lasting enhancement of the sense of agency caused by the "global" outcome probability in a given block or in several previous trials. In Experiment 2, we further manipulated the emotional valence of the action outcome (i.e., a happy or an angry face), and found no difference of positive versus negative outcomes on subsequent action regulation but similar effects.
One of our main findings is that participants' readiness or preparedness to perform a keypress action varies as a function of the sense of agency over that action. Both experiments demonstrate that participants responded more accurately and faster to the subsequent Go cue that signals a keypress action when the previous keypress action produced a visible effect. In addition, the control experiment confirmed that participants felt a higher sense of agency when their action led to a visible effect compared with no effect. These results support that the feeling of control over the motor action is updated on a short, trial-totrial timescale and can have an instant impact on the following motor responses . Our finding is also in line with previous empirical work showing that actions followed by predictable perceptual changes are reinforced. For example, two-month-old infants moved their limb more frequently when their movement caused a movement of the mobile tethered to that particular limb (Watanabe & Taga, 2006, 2011. Similarly, for adult participants, a motor response that leads to a perceptual change in the environment increases the speed and the frequency of that response Karsh et al., 2020;Tanaka et al., 2021). Moreover, the reliability of action outcomes has been found to boost the readiness potential, reflecting enhanced neural activities prior to the action . Together, it is reasonable to conclude that feeling in control over a specific action appears to reinforce that action and in this way increases the tendency to repeat it, while feeling out of control over one action discourages repetition of that action. Consistently, the comparison results of accuracy rates in Go and No-Go trials reveal that participants' motor tendency was shifted from readiness to action towards increased inhibitory control when their sense of agency decreased. Specifically, participants exhibited a preference for go over no-go responses in conditions with "local" outcome occurrence. In contrast, no obvious preference between go and no-go responses or even a preference for no-go over go responses was observed in the condition with high outcome frequency at the block level (i.e., global) but surprising outcome absence at the trial level (i.e., local).
The present study provides, to the best of our knowledge, the first evidence that a higher sense of agency results in reduced response inhibition. Specifically, when participants felt more in control over the keypress action as induced by the occurrence of a perceivable outcome, they responded less accurately to the subsequent No-Go cue that required canceling the prepotent keypress action. Given that response inhibition is regarded as one of the central components of cognitive control (Botvinick & Braver, 2015), our finding suggests that inhibitory control deployed over one specific action increases as the feeling of control over that action decreases. This interpretation is in line with the proposal that using cognitive control needs to be distinguished from the feeling of being in control, i.e., the sense of agency (Potts & Carlson, 2019). Additionally, losing control of an already controlled visual object has been found to attract attention (Wen & Haggard, 2018). Our finding extends this earlier report by suggesting that detecting diminished control may also act as a trigger to executive function/cognitive control that aims to reassert control. Effectively controlling the environment is a psychological and biological imperative for survival (Leotti, Iyengar, & Ochsner, 2010); therefore, our observation of the sense of agency as input for adaptive adjustment of cognitive control highlights its critical role in human survival.
Importantly, results in Experiment 1 indicate that a sense of agency modulates not only the response inhibition triggered by external No-Go cues but also the response inhibition generated from internal decisions (also known as intentional inhibition). Specifically, after experiencing that the keypress action did not produce any visible effect, participants were less likely to choose to perform that action again in response to Free-Choice cues; and even when they chose to press the button, the reaction times were slower. These results indicate that participants tend to cancel the about-to-be-executed action when feeling out of control. It has been shown that subliminal masked prime presented prior to the Free-Choice cue can influence participants' decision to act or inhibit, suggesting a modulation effect of unconscious processing on intentional inhibitory control (Parkinson & Haggard, 2014). In Experiment 1, although the action-outcome contingency can be consciously perceived, it was neither explicitly associated with the sense of agency or the feeling of control nor did it provide any information about participants' performance. In other words, the sense of agency was task-irrelevant and thus most probably processed implicitly. Therefore, our results support that the free decisions to act or not to act can be influenced by implicit processing regarding our degree of control over the external environment.
Notably, there might be a link between the long reaction times in Go/ Free-Choice trials and higher accuracy rates in No-Go trials in the condition with high "global" outcome probability but surprising "local" outcome absence. That is, when the previous keypress action surprisingly did not produce any outcome, participants might hesitate to press the key again. Such hesitation could be associated with no action in the subsequent phase, resulting in a low response accuracy for the Go cue and a high response accuracy for the No-Go cue. In other words, the putative "better" response inhibition might be a result of a general response slowing or down-regulation of motor preparation, which is another side of the same coin. Dissociating the effects of sense of agency on action readiness and response inhibition, disentangling the precise nature of the underlying inhibitory mechanisms, as well as testing the specific time course of these processes would be worthwhile endeavors for future studies.
Interestingly, the effects of the "local" outcome presence on subsequent action regulation are modulated by the "global" outcome probability. Specifically, results in Experiment 1 showed that the enhancement of action readiness and the suppression of response inhibition after obtaining action outcomes compared with no outcome were amplified when participants had a high probability of obtaining action outcomes in a given block and also when action outcomes occurred frequently in several previous trials. Notably, this modulation effect was much less prominent in Experiment 2, probably because the longerlasting enhancement of the sense of agency induced in Experiment 2 was not as strong as that in Experiment 1 due to the relative frequency rates of the different outcomes. The modulating effect of the global context has also been reported by Hemed et al. (2020), who found that reaction times on trial n was sensitive to the outcome presence on trial n-1 and the magnitude of this effect was modulated by the outcome frequency in trial n-4 to n-2. This finding suggests that participants' action regulation is sensitive to both the immediate, local context and the more distant, global context regarding the effectiveness of an action (i.e., whether the action can lead to an effect and how reliable it is). This is also consistent with the recent view that the brain carefully monitors one's control over external objects and is highly sensitive to any change in the degree of control (Wen et al., 2021;Wen & Haggard, 2018).
The observed effects of the sense of agency on response inhibition and action readiness could be explained within a larger framework of "motivation from control". This framework posits that the sense of agency, i.e., the feeling of control over external objects, is a form of internal reward and serves as an important source of motivation (Nafcha, Higgins, & Eitam, 2016). This view has been supported by accumulative behavioral and neuroimaging evidence. For example, a high probability of obtaining action outcomes, which contributes to an increase in the sense of agency, facilitates the likelihood and speed of an action being selected (Karsh et al., 2020;Penton et al., 2018). In addition, high action-effect contingency (Behne, Scheich, & Brechmann, 2008) and high perceived control (Lorenz et al., 2015;Tricomi, Delgado, & Fiez, 2004) are associated with increased activity in brain areas (e.g., striatum) involved in reward processing. From this perspective, the enhanced action readiness and reduced inhibitory tendency after feeling more in control in the present study may reflect a rise in motivation. Moreover, the "motivation from control" framework emphasizes the degree of control the organism has over the environment but not the value (desired or undesired) of the outcome (Nafcha et al., 2016). In other words, anything that happens after a motor response, no matter whether it is positive, negative or neutral, can be judged as feedback about successful control and hence can motivate action. Notably, the value of the outcome is another important source of motivation, known as "motivation from outcome". It can generally trump the "motivation from control", if the outcome itself provides information about the relation between the current state and the desired goal (Nafcha et al., 2016). However, participants' task goal (i.e., more accurate and faster responses) was independent of the emotional valence of action outcomes in the present study; therefore, "motivation from control" probably played a dominant role. This may explain why the emotional valence of action outcomes did not modulate participants' action regulation in the present study. From the participants' view, both the happy and the angry face presented immediately after their keypress response indicated that they were effective in controlling the environment and hence seemed to trigger comparable degrees of motivation for future actions.
One limitation of the present study should be noted. The present study cannot isolate the relative contributions of the conceptual (explicit) judgement of agency and sensorimotor (implicit) predictability to the effects of the present study. Our control experiment showed that participants reported higher agency ratings (i.e., feeling more in control) when their previous action led to an outcome compared with no outcome and also when action outcomes occurred relatively frequently in a given block or in several previous trials. Thus, it is reasonable to speculate that higher-level aspects of the sense of agency (explicit judgement of agency) impacted our results. However, our results cannot be fully explained by changes in the explicit judgement of agency as we observed similar effects across the two main experiments in absence of any explicit judgements. More importantly, Hemed et al. (2022) have elegantly shown that fulfilled sensorimotor prediction facilitated response speed (i.e., action efficiency), while conceptual judgement in contrast mainly affected response frequency in terms of which action to choose (i.e., response selection). Therefore, it is highly likely that conceptual and sensorimotor aspects of the sense of agency both contributed to our results. Further studies are needed to disentangle the relative contributions of these factors on action regulation, for example by replacing the no-effect condition with a sensory-unpredictable condition .
In conclusion, a higher sense of agency in the current trial, as induced by the presence of an outcome for a given action, facilitated action readiness but at the same time disrupted response inhibition in a subsequent trial. These effects were further amplified by a longer-lasting enhancement of the sense of agency as induced by high outcome frequency in a given block or in several previous trials. Furthermore, these effects were independent of the emotional valence of the keypress outcome. We explain these effects within the framework of motivation from control. Our findings highlight the impact of the control felt on the control used in action regulation and hereby provide new insights into the functional significance of the sense of agency on human behavior.

Declaration of Competing Interest
None.