The time course of social and non-social implicit Sense of Agency

doi:10.21203/rs.3.rs-2097214/v1

Download PDF

Research Article

The time course of social and non-social implicit Sense of Agency

https://doi.org/10.21203/rs.3.rs-2097214/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Sense of Agency (SoA) is the responsibility felt over our actions and their effects. Social Agency is when SoA occurs in a social context. Whether and how SoA and Social Agency differ remains largely unknown. As non-social interaction is fundamentally governed by the laws of physics and material properties, action-effect contiguity is expected. This is not so when interacting with conspecifics, who we know act under their own volition, at a time of their choosing. Therefore, temporal contingency is a characteristic where social and non-social agency may differ. In three experiments, this exploratory study directly compared the time course of implicit SoA in social and non-social cooperative interactions by varying action-effect intervals. Participants were told they were a) interacting via live webcams with another person (social condition), b) interacting with pre-recorded videos (non-social condition) or c) observing two pre-recorded videos (control condition; Experiment 3). Results across all experiments showed greater implicit agency effects for social compared to non-social conditions through Temporal Binding (i.e., the perceived compression of action-effect interval duration; TB), and this to a proportionally comparable degree across all action-effect intervals. Further, in Experiments 1 and 2, there was consistent social TB throughout, whereas non-social TB decreased during the second half. In Experiment 3, the non-social condition did not differ from control, whereas the social condition did, exhibiting enhanced binding. These findings indicate that implicit Social Agency is enhanced, compared to non-social agency, in cooperative interaction. Social Agency also exhibits TB effects at shorter action-effect intervals than non-social agency.

Sense of Agency (SoA) is the responsibility we feel over the effects of our actions (Haggard, 2017; Haggard & Chambon, 2012). For example, SoA is experienced when a bulb lights up upon pressing the corresponding light switch. When we perform such actions by ourselves, the effects of those actions are likely to be predictable and easily attributed to the self, inducing non-social agency. Where SoA gets more complicated, is when actions are performed a) in the presence of another (i.e., a social context), b) aligned with the action(s) of another (i.e., joint action) and/or c) to elicit a response from another (i.e., social effect; Silver, Tatler, Chakravarthi, & Timmermans, 2021). Following the previous example, should a person happen to reach for the same light switch, at the same time as someone else and the light turns on, it becomes far less easy to attribute causality. The SoA that we experience under social circumstances is typically referred to as ‘Social Agency’ (Brandi et al., 2019; Silver et al., 2021). However, it is not known whether social and non-social agency differ. This is because Social Agency remains under examined, rarely directly compared to non-social agency, and generally considered to be a sub-component of SoA (Silver et al., 2021). Further, in the few studies that attempt to assess it, it is generally explained using conflicting terms (see ‘Conflicting views on Social Agency’, Silver et al., 2021). If we are to develop a theoretical understanding of Social Agency, we must first establish whether and how it differs from non-social agency. Specifically, does Social Agency have characteristics distinct from non-social agency?

A key characteristic for experiencing SoA in non-social settings is temporal contiguity between actions and their effects; non-social responses are anticipated in line with our understanding of the laws of physics and material interactions (e.g., electricity; Cravo et al., 2011). This contiguity is probably less important in Social Agency (Pfeiffer et al., 2012; Stephenson et al., 2018) because we recognise the other as an independent agent, acting under their own volition (Pfeiffer et al., 2012). Instead of switching on a light ourselves, if we ask someone else to do it, we would not expect an immediate response or at least expect the person to act at their convenience. Therefore, temporal contingency is an important action-effect characteristic to manipulate to examine whether social and non-social agency differ in general, and specifically whether their time courses differ.

Temporal Binding is the perceived compression of time between a causal action and its effect (Beck et al., 2017; Buehner, 2012; Engbert et al., 2007), and is the most common implicit measure of SoA (Silver et al., 2021). Temporal Binding offers a powerful way to measure the time course of agency effects by scrutinising this measure over different action-effect interval durations. Investigation into the time course of non-social Temporal Binding has revealed divergent results, some studies reporting the strongest binding at the shortest delays (Haggard et al., 2002; Ruess et al., 2018), others reporting binding effects increasing with interval length (Buehner & Humphreys, 2009; Dewey & Knoblich, 2014; Ruess et al., 2017; Wen et al., 2015). These contradictions may be, in part, due to methodological differences: the specific set of interval durations used and if the interval durations were estimated or replicated (Silver et al., 2021).

Only one study has previously examined the time course of Social Agency, but critically used an explicit measure. Explicit and implicit measures of SoA are argued to scrutinise different components (judging and feeling agency, respectively; Moore et al., 2012), which may or may not be correlated (Dewey & Knoblich, 2014; Moore et al., 2012). Pfeiffer et al. (2012) used the eye movements of an on-screen avatar as a proxy for the eye movements of a second, confederate participant. Participants were asked to look at one of two on-screen objects and after a variable latency, the avatar either followed their gaze to the same object or averted their gaze to the opposite object. Here, the participant can be considered to be either initiating acts of joint attention or non-joint attention respectively. Explicit SoA was measured by asking participants to rate how related the action of the other was to theirs. Gaze following was rated as related to participants’ action, whilst gaze aversion was rated as unrelated. Ratings of how related the actions were to each other were highest at 1000 ms and remained consistently high across gaze response latencies of up to 4000 ms, but this was only the case when the avatar (i.e., other) had a choice of response. When the avatar was restricted to establishing joint attention only, relatedness ratings decreased linearly for gaze response latencies beyond 400 ms. This highlights that the agentic status of the ‘other’ in a social interaction is an important component of experiencing Social Agency. Removing the action choice of the ‘other’ diminished perceived volition and therefore agency attributions waned as temporal contiguity decreased. Whilst the findings of Pfeiffer et al. (2012) highlight important points about how agency is assigned to social ‘others’, there was no non-social comparison. Therefore, the question of whether and how explicit Social Agency differs to non-social remains unanswered. Inferences are also limited to how we judge agency over different action-effect contingencies: the study does not answer any questions relating to how we experience (i.e., feel) agency.

Out of the circumstances in which Social Agency arises, by far the least investigated is when an action is performed to elicit (directly or indirectly) a response from another: a social effect. This is unfortunate, as actions with exclusively social effects, such as the gaze cues used by Pfeiffer et al. (2012), arguably offer the most immediate and frequently experienced type of Social Agency to investigate (Silver et al., 2021). Pfister et al. (2014) investigated this type of Social Agency using Temporal Binding as an implicit measure. Participant pairs took turns in fulfilling hierarchical leader and follower roles, where the leader started a trial by performing a keypress at a time of their choice, which generated a tone. This tone sounded after a variable delay of 100–600 ms and acted as a signal for the follower to perform a keypress of their own within 1000 ms. At the end of each trial, both participants either estimated the interval between the leader’s keypress and tone or the interval between the tone and the follower’s keypress. A second experiment was conducted where the follower’s keypress also generated a tone after a variable delay of 100–600 ms and this new interval was also estimated. Estimations by the leader exhibited Temporal Binding for their own keypress-tone interval as would be expected, and interestingly for the tone-follower keypress interval, but only in experiment 2, where this keypress produced its own tone. The follower did not exhibit any Temporal Binding for any of the three intervals estimated. Neither participant exhibited Temporal Binding for the follower action-tone interval. While Temporal Binding appears to be related to the position of the assessed interval in the interaction sequence, whether and how action-effect contingency influenced Temporal Binding was not examined. Binding effects were analysed using interval duration regression residuals of binding scores, rather than raw Temporal Binding. Therefore, the theoretical question of how action-effect contingency influences Temporal Binding remains unanswered.

The leader in Pfister et al.’s (2014) study had an explicitly dominant role, could choose when to act, and their action prompted the action of another. Hence, self and ‘Vicarious’ agency effects (i.e., where agency is experienced for the actions of others; Silver et al., 2021) were expected. The follower’s volition was restricted in multiple ways: having an explicitly subordinate role, acting in response to a signal which had to be within a time limit, and in Experiment 1, their action had no perceivable effect. Even with all these factors, it might be predicted that the causal link between follower action and tone in Experiment 2 would induce Temporal Binding effects. Finding no binding effects, as in this study, would suggest that violating an individual’s agency can override binding effects (i.e., ‘Violated’ agency; Silver et al., 2021). A crucial point to consider, however, is how the tones mediate the interaction in this paradigm. This does two things: it breaks the direct social link between leader action and responding follower action, and it increases the temporal disparity between these actions. The tones, therefore, may have diminished the social context of the interaction. Taken together with the fact that there was no observational non-social control condition, it is prudent not to make wider Social Agency inferences from these findings (Silver et al., 2021), and indeed the authors’ main point is simply that SoA also exists in social interaction.

Social Agency is strongest when people cooperate (Silver et al., 2021), and most prior studies of Social Agency have used cooperative paradigms (e.g., Pfeiffer et al., 2012; Pfister et al., 2014; Stephenson et al., 2018; Vogel et al., 2021). In the present exploratory study, therefore, we used a cooperative interaction designed to study the time course of Social Agency and allow direct comparison to non-social agency. In three experiments we kept visual stimuli constant and manipulated the social context of the interaction (i.e., social or non-social). Participants were told they were interacting directly with an interaction partner, through keypresses filmed with a webcam and displayed on the other’s screen and vice versa, in the social conditions. Alternatively, they were told that they were watching pre-recorded videos in non-social conditions. In reality it was always a video recording. Participants initiated an interaction at a time of their choice through a keypress, then watched a direct response (i.e., a keypress by the ‘other’) at various intervals. Temporal Binding was measured trial-wise through participants replicating this action-response interval.

It was an important methodological decision to have real human hands on screen as stimuli. This was so that the social condition could look like a live interaction (similar to Pfister et al.’s 2014 study), but with the advantage of maintaining experimental control. A recent study on Temporal Binding in social contexts (Vogel et al., 2021) manipulated the belief about the Agency of an interaction partner, but always used basic geometric shapes as on-screen stimuli. In the social condition of Experiment 1, the shapes were arranged as a rudimentary face with circles for eyes and a triangle for a nose and participants were told that they were interacting with a confederate; conversely, a non-face-like arrangement of the same shapes indicated that the interaction partner was a computer in the non-social condition; finally, these were reproduced in an observational control condition. As this conflates social appearance and supposed agentive nature of the other, Experiment 2 uncoupled and counterbalanced these factors, but dropped the observational control condition. Both experiments showed a significant enhancement in Temporal Binding for the social condition compared to non-social, which they termed ‘hyper-binding’, and Experiment 2 showed that physical social appearance and supposed agentive nature of the other could both enhance Temporal Binding independently but combining the two did not offer any further enhancement. Authors argued that the results “suggest a prioritization of social information for Temporal Binding regardless of whether this information is introduced by top-down (beliefs) or bottom-up information (stimuli)” (Vogel et al., 2021, p. 1545). However, having shapes represent the social ‘other’ in this study may have limited how much the stimuli were interpreted as social. Our study sought to address this by consistently using videos of real human hands as stimuli throughout, regardless of condition. This ameliorates a) any appearance counterbalancing issues like in Experiment 1 in Vogel et al.’s study (2021), and b) removes the social appearance of stimuli having any contributary effects to any observed social/non-social differences as appearance is consistent. Our rationale was that using videos of a real human hand (indeed the actual hand of the interaction partner) for viewing the responding action to the participant’s action was an indisputably ‘social’ response/appearance. Finally, by designing an interaction with a direct action-effect link between participant and ‘other’ actions (i.e., the responding action of the ‘other’ was the effect of the participant’s action), there was a social effect in the truest sense in our study.

Experiments 1 and 2, analysed together, sought to investigate whether a visual response (i.e., on-screen keypress) was socially derived or not would affect our implicit Sense of Agency, and if this effect was contingent on a specific temporal contiguity of the response. The non-social condition of Experiment 1 required an action from participants. Hence, to rule out any role of action execution on Temporal Binding effects (Wenke & Haggard, 2009), we introduced a passive, observational control condition in Experiment 2, where participants’ initiating action was replaced by an automated tone followed as usual by the on-screen keypress. As these experiments were analysed together, Methods for each are presented sequentially, before presenting a combined Results section.

Experiment 3 sought to include a within-subjects factor to account for individual differences in baseline (i.e., control) interval replication, and to improve upon the observation control condition from Experiment 2. Participants completed both an active experimental condition (social or non-social) and a passive observational control condition. The modified control condition was designed to be solely visual and therefore more comparative to the active conditions than when the tone was used (Experiment 2). Participants in this control condition viewed two on-screen keypresses and replicated the interval between. Taken together, these experiments allowed us to directly compare the time course of social and non-social binding for the same cooperative task, and to incorporate how action preparation/execution may influence Temporal Binding effects.

The time course of Social Agency had not been investigated before with a Temporal Binding measure. Further, previous research for the time course of non-social Temporal Binding had exhibited divergent results. Hence, we only made tentative predictions for the time course of social SoA. Firstly, we predicted that there may be a peak of Temporal Binding effects, similar to those found for an explicit measure by Pfeiffer et al. (2012). Secondly, we predicted this peak would be at longer action-effect interval durations for the social, compared to non-social condition. This was because almost immediate action-effect contiguity is expected in non-social circumstances (Cravo et al., 2011), whereas in social interaction this is not so; we understand that the other chooses when to act (Pfeiffer et al., 2014; Stephenson et al., 2018). Finally, and more generally, as the social condition was cooperative in nature, with the ‘other’ a) always responding and b) with the expected response, it was predicted that this direct social response to the participant’s action would enhance binding, compared to the non-social condition (as in the cooperative continuum theory of Social Agency; Silver et al., 2021).

In Experiment 2, the social condition was largely the same, but the researcher now doubled as the interaction partner (i.e., confederate). A new condition was introduced in Experiment 2, acting as a control: ‘observation’. This was added as even in the non-social condition in Experiment 1, the participant performs a voluntary action, which can, in itself, induce SoA (Haggard, 2017; Haggard & Chambon, 2012). Therefore, in the observation condition, the on-screen interaction was initiated by an automatic auditory tone at a variable time, rather than by the participant’s action. This condition was predicted to exhibit the least Temporal Binding, as no action was required by the participant. The dichotomous (i.e., yes/no) belief rating in Experiment 1 was changed to a rating on a 5-point Likert scale (extremely/very/somewhat/slightly/not at all). This implemented methodology used in previous paradigms (e.g., Caruana, de Lissa, & McArthur, 2017) and incorporated anecdotal participant feedback from Experiment 1, which suggested that there may be some ambivalence in subjective belief. Predictions were the same as in Experiment 1.

Method for Experiment 1

Participants

34 students and staff (20 females; age 18–52 years; mean age = 24.8; AQ scores 8–30, mean AQ = 16.5) from the University of Aberdeen participated for renumeration or course credit. Written consent was obtained from all and the study was approved by the School of Psychology, University of Aberdeen Ethics Committee (PEC/4344/2019/10).

Design

The experiment used a 2 x 20 mixed-design with one between-subjects factor of ‘Socialness’ (‘live’ interaction = social condition, pre-recorded videos = non-social condition) and one within-subjects factor of ‘Interval’ (200–2100 ms in 100 ms increments; 20 conditions total). The Socialness factor was conducted as between-subjects for two reasons: to keep the duration of experimental sessions reasonable (< 1 hr) and to maintain the necessary deception element: the same pre-recorded videos were used in both Socialness conditions, so participating in both could easily lead to the discovery that the social interactions were not ‘live’.

Materials and Apparatus

The experiment took place in two testing cubicles (in adjoining rooms of a research laboratory), each equipped with a desktop PC, LCD monitor, QWERTY keyboard, and webcam. Webcams were mounted so that they each filmed the cubicle’s keyboard and a section of desk so that the interaction partner’s hand was in view (see Fig. 1 to see what was visible to participants on-screen). E-Prime software (Schneider et al., 2002) was used for the delivery of the experimental programme.

Colour videos of the interaction partner’s (i.e., confederate’s) button presses were pre-recorded using an iPhone 11’s integrated camera software (Apple, 2021). Each video showed the partner’s hand with the index finger in a resting position on a keyboard, pressing the ‘B’ key and then returning to the original position. Videos of 11 seconds were edited using VideoPad Video Editor for Mac (NCH Software, 2020), where the key press occurred after exactly 10 seconds, and the video terminated 1 second later. Importantly, the hand was required to remain as still as possible before initiating the move to press the ‘B’ key. This allowed the E-Prime programme (Schneider et al., 2002) to cut sections of the video seamlessly (i.e., without the hand position jumping), to deliver the specific time intervals between participant and on-screen keypresses. Multiple videos were recorded with slightly different hand starting positions to increase believability in the ‘live’ condition that the videos were in fact a genuine live stream from the partner’s webcam.

Participants filled out the Autism Spectrum Quotient (Baron-Cohen, Wheelwright, Skinner, Martin, & Clubley, 2001) questionnaire. This 50-item self-report questionnaire evaluates the degree to which an individual possesses traits associated with Autism Spectrum Disorder, with higher scores (range 0–50) indicating the presence of more autistic traits; any score of 32 + is of clinical relevance (Baron-Cohen et al., 2001). As it has been demonstrated that higher AQ scores can be indicative of lower sensitivity to social stimuli (Bayliss et al., 2005; Freeth et al., 2013), it was desirable to be able to account for this possible modulator to Social Agency effects.

Procedure

Participants were randomly assigned to either the ‘live’ condition (i.e., they were told that they would be interacting with a confederate who was introduced as another participant; social condition) or the ‘pre-recorded videos’ condition (where they were told that they would be watching pre-recorded videos that would be played in response to their action; non-social condition). Participants in the social condition were instructed not to move the position of the keyboard and to keep their hand as still as possible when not pressing the ‘B’ key. This was explained as being important to reduce ambiguity of when they were choosing to press the key or observing the likewise keypress, but was in fact a cover story to justify why the hand viewed on-screen was (perhaps unusually) still. The computerised experimental task was completed either in pairs (where the confederate was taken to the adjoining laboratory) or individually, according to condition.

On each trial, the participant pressed ‘B’ on their keyboard at the time of their choice. The (partner’s) hand on-screen would press ‘B’ in response after a variable time interval (200–2100 ms in 100 ms increments; randomised across trials). The on-screen hand was visible from trial onset to 1000 ms after the hand pressed ‘B’. Following each trial, the participant was asked to replicate the interval between their keypress and the on-screen keypress: they were asked to press and hold the spacebar for the same length of time (see Fig. 1 for trial sequence). There was then a 500 ms blank screen, after which the next trial began.

The experiment started with a practice block of 10 trials where only the interval initiation (participant action) and response (observed action) were presented. The participant completed a further 20 trials where this sequence was followed by the replication task. Experimental blocks (4 blocks of 20 trials; each ‘Interval’ used once per block; 80 total) with the full sequence were then presented; each block required participant action to commence, allowing for breaks to be taken as desired.

After completing the experimental task, participants completed the AQ (Baron-Cohen et al., 2001). Participants in the social condition were also asked, post-debrief, if they had believed they were interacting with the confederate (yes/no).

Method for Experiment 2

Participants

49 students and staff (29 females; age 19–49 years; mean age 25.6; AQ scores 6–33, mean AQ = 17.9) from the University of Aberdeen participated for renumeration or course credit. Written consent was obtained, and the study was approved by the School of Psychology, University of Aberdeen Ethics Committee (PEC/4344/2019/10).

Design

The design was the same as in Experiment 1, except for the addition of a new observation control condition of the between-subjects Socialness factor (now 3 conditions: social/non-social/observation, resulting in a 3 x 20 mixed design.

Materials and Apparatus

The materials and apparatus were identical to those used in Experiment 1, except that a new set of videos were generated of the researcher’s hand.

Procedure

The procedure diverged from Experiment 1 in three ways: 1) in the new observation condition, instead of the participant pressing the ‘B’ key of the keyboard to initiate the interaction within each trial, an auditory tone (150 ms, 600 Hz) would sound after a variable interval (500 ms – 2500 ms) from the start of the trial and then the on-screen hand would press at a variable interval (same as Exp 1; 200 – 2100 ms in 100 ms increments; randomised across trials) from onset of the tone, 2) the belief rating was now taken on a 5-point Likert scale (extremely/very/somewhat/slightly/not at all) and 3) the researcher now acted as the confederate in the social condition.

To improve believability of the deceptive interaction, in the new social condition the researcher pre-launched the live webcam feeds before participants entered the laboratory. Then, upon bringing participants into the laboratory, the researcher explained the webcam set-up for viewing each other’s hands and demonstrated this by waving their hand in front of one camera, letting the participant view this action on the other cubicle’s screen. Importantly, this genuinely live feed was left open in a background window whilst E-Prime was running the experiment programme. Also, participation in the social condition was now completed in adjacent testing cubicles of the same laboratory, so that researcher and participant were seated side-by-side, divided only by a screen (see Fig. 2a).

In the non-social condition (as in Experiment 1) participants were not observed by the researcher, who was still present in the lab, but worked on other tasks at a workstation in the opposite corner of the room and only interacted with the participant when prompted that they were finished. In the social condition, the researcher consistently pretended to ‘interact’ with the participant, delivering sporadic commentary on the task, asking if the participant would like a break between blocks and importantly, keeping their hand in a similar position to the resting position on the videos. This was so that when the E-Prime programme terminated, the researcher’s hand was present at a convincing position on the webcam live feed.

Statistical Analysis for Experiments 1 & 2

Outlier detection and exclusion criterion for participant-wise and trial-wise measures.

The data was pooled across Experiments 1 and 2. First, we excluded participants for whom the correlation coefficient between interval estimations and interval durations lay more than 2.5 absolute deviations from the group median correlation coefficient (MAD; b = 1.4826; Leys et al., 2013) (2 participants were excluded). This was to ensure participants included in the analysis had actively engaged in the task. Second, individual trials where responses were more than 2.5 absolute deviations from each interval duration median response were excluded (MAD; b = 1.4826; Leys et al., 2013) (8.4% of trials were excluded). This was to remove any response outliers. Finally, if any participant had more than 25% of their trials removed by the trial-wise criterion, they were excluded entirely from analyses (further 5 participants excluded). Participants to be included in analysis after all exclusions were: 36 for the social condition, and 28 for the non-social condition.

All analyses were done on percentage Temporal Binding, calculated as per Eq. 1 below. As Temporal Binding is the perceived compression of time between events, a binding score below zero represents active binding (i.e., underestimation of interval duration), zero represents accurate interval estimation, and a score above zero represents an overestimation of interval duration. Using percentage binding, rather than a measure of Temporal Binding in milliseconds (i.e. $interva{l}_{estimate}- interva{l}_{actual})$, allows comparison across all interval durations, in that it accounts for numerically larger differences merely due to the to-be-estimated intervals being longer, and associated larger variance for longer durations.

$${Binding}_{percentage}= \frac{interva{l}_{estimate}- interva{l}_{actual}}{interva{l}_{actual}} \times 100$$

The data were analysed with a Mixed Linear Model (MLM; Kliegl et al., 2012), for the following reasons: the uneven distribution of data across conditions of multiple independent variables (both between- and within-participants), the combination of both continuous and categorical variables within the design, the desire to account for between-participant variance for repeated measures factors, and different methodologies (i.e., Experiment 1 and Experiment 2). Experiment 2 had an additional between-subjects condition for the Socialness factor (observation; 12 participants, with 10 remaining after all exclusions); while combining data from Experiments 1 and 2 in the same MLM, the observation condition was excluded from the main analysis (see Supplementary Material for a separate MLM run only on data from Experiment 2, including observation condition to scrutinise the role of action preparation/performance on binding effects).

Introducing Belief and Block as exploratory fixed effects.

In addition to the fixed effects associated with experiment design (2 Socialness conditions x 20 Interval durations), we introduced two additional factors in the model. First, we recoded the social condition into participants who said they believed they were interacting live with someone else (N = 22), and those who said they didn’t (N = 14). This grouping was obtained by using the yes/no answer to the ‘belief’ question in Experiment 1 and a Likert rating of 3–5 vs 1–2 in Experiment 2. This resulted in the 2-level between-subjects factor Socialness being replaced by a between-subjects Socialness/Belief factor with 3 levels (non-social/social_no/social_yes). Second, in order to explore how differences in Sense of Agency developed over the experimental blocks, we introduced a factor Block Number with 2 levels (block 1/block 2). Due to a programming error, the distribution of interval durations was not exactly the same across all four blocks, but was identically balanced within the first two blocks and the last two blocks; these were pooled into block 1 and block 2 respectively.

The fixed effects included in the MLM were as follows. A categorical fixed effect of ‘Socialness/Belief’. Contrasts were coded for this factor to compare non-social to all social participants (regardless of belief; Socialness) and the social believers to social non-believers (Belief). An ordered categorical fixed effect of Block Number (‘BlockNr’; 1, 2), coded using orthogonal polynomial contrast, and a continuous fixed effect of Interval (200 ms – 2100 ms), which was centred and scaled using the base R scale function (‘Interval_Scaled’; R Core Team, 2019). The MLM was built using the lmer function of the lme4 package (Bates, Mächler, Bolker, & Walker, 2015) of R statistical programming language (R Core Team, 2019), in the R Studio integrated development environment (RStudio Team, 2019); ggplot2 (Wickham, 2016) was used for plotting data.

The strategy adopted when building the MLM was to begin with the maximal random effects structure justified by the design (i.e., include all random slopes and intercepts for within-subjects fixed effects, with interactions), then simplify element-by-element until the model converged (Barr et al., 2013). ‘Subject’ and ‘Methodology’ (Experiment 1/Experiment 2) were included in the MLM as random effects. As both Experiment 1 and 2 included a between-subject factor, it was important to include Subject to account for any individual differences between groups. To justifiably pool data from the two experiments, it was also essential to account for any difference in binding effects that might be attributed to the slightly different methodologies used. This maximal model did not converge, therefore we systematically simplified it.

The simplification strategy was to 1) drop interactions in slopes of random effects that accounted for the least amount of variance, 2) drop slopes of random effects, again starting with the slopes that accounted for the least amount of variance and then, if necessary 3) drop the intercept of the random effect that accounted for the least amount of variance. Once the maximal random effect structure that would converge was found, simplification continued until all non-significantly contributing elements were removed (tested by comparing model versions through analysis of variance; ANOVA) until the optimal model was reached. Lmer (Bates et al., 2015) syntax of the optimal model is shown below:

Optimal_model = lmer (Percentage_Binding ~ 1 + Interval_Scaled * Socialness/Belief * BlockNr + (1 + Interval_Scaled | Subject), data = Exp12, control = lmerControl (optimizer = "bobyqa"))

Results for Experiments 1 & 2

The output of the MLM to predict Temporal Binding from Interval, Socialness/Belief condition contrasts and Block Number is shown in Table 1 and depicted in Fig. 3. Crucially, there was a significant main effect of Socialness, with more binding (%) in the social condition (M = -38.4, 95% CI [-39.7, -37.1]; independent of belief), compared to the non-social condition (M = -19.4, 95% CI [-21.3, -17.5]), b = 20.09, SE = 4.42, t = 4.54, p < .001 (to rule out any contributory effects of differing participant demographics to this between-subject result, exploratory analyses were carried out regarding Age and AQ scores of participants: no significant results were found; see Supplementary Material). However, there was no significant difference in binding between participants who believed the social interaction (M = -38.4, 95% CI [-40.0, -36.8]) and those that did not (M = -38.5, 95% CI [-40.7, -36.2]), b = − .04, SE = 5.93, t = − .01, p = .994. This effect of Socialness was qualified by interactions with Interval (b = -5.91, SE = 2.50, t = -2.36, p = .021), which we explored through curve fitting (see below), and with Block Number (b = -6.77, SE = 1.20, t = -5.64, p < .001), with the effect of Socialness being smaller in block 1 (see below).

There was a significant main effect of Interval on binding, b = -18.20, SE = 1.28, t = -14.24, p = < .001, with percentage binding increasing for longer durations. There was a significant linear main effect of Block Number on binding, b = -3.74, SE = .61, t = -6.14, p = < .001, with responses in block 2 (M = -27.4, 95% CI [-29.1, 25.7]) exhibiting less binding overall than block 1 (M = -33.1, 95% CI [-34.6, -31.5]).

Table 1: Results of the MLM to predict percentage binding for Experiments 1 and 2.

Fixed Effect	b	SE	df	t	p
Intercept	-31.73	2.26	60.89	-14.05	< .001
Interval	-18.20	1.28	60.94	-14.24	< .001
Socialness	20.09	4.42	60.94	4.54	< .001
Belief	− .04	5.93	60.85	− .01	.994
Block	3.74	.61	4868.24	6.14	< .001
Interval*Socialness	-5.91	2.50	61.13	-2.36	.021
Interval*Belief	-3.66	3.35	60.80	-1.09	.280
Interval*Block	-2.39	.61	4868.83	-3.92	< .001
Socialness*Block	6.77	1.20	4868.59	5.64	< .001
Belief*Block	2.53	1.59	4867.98	1.59	.113
Interval SocialnessBlock	− .92	1.20	4869.70	− .77	.442
IntervalBeliefBlock	.70	1.59	4868.70	.44	.661

To investigate the interaction between Socialness and Block Number, additional analyses were carried out. Follow up MLMs were run to test the simple effects of Socialness (social or non-social) and Block Number. Within block 1, binding was significantly greater for the social, compared to non-social condition, b = -15.10, SE = 4.42, t = -3.42, p = .001 (M = -39.3, 95% CI [-41.1, -37.5] and M = -24.8, 95% CI [-27.4, -22.3] respectively), this was further enhanced in block 2, b = -25.07, SE = 4.42, t = -5.67, p = < .001 (M = -37.5, 95% CI [-39.4, -35.6] and M = -13.8, 95% CI [-16.5, -11.0] respectively). As shown in Fig. 3, this increasing difference between social and non-social was due to a decrease in binding in the non-social condition from block 1 to block 2, b = 11.68, SE = 1.27, t = 9.21, p = < .001, while there was no difference in the social condition from block 1 to block 2, b = 1.71, SE = 1.10, t = 1.55, p = .121.

To examine the time course of binding differences between social and non-social conditions (Socialness/Belief*Interval interaction), we subjected the data to curve-fitting; exponential curves were fit to percent binding as a function of Interval duration for each participant. The parameters of these curve fits were then compared across conditions using analyses of variance (ANOVAs). The exponential function (Rashal & Yeshurun, 2014; Scolari et al., 2007; Soo et al., 2018) used to fit the data was:

$$y\left(x\right)= \propto \left({1-e}^{\left(-s\left(x-t\right)\right)}\right)$$

where $y$ is the interval-replication percentage binding, $\propto$ is the asymptote, $s$ is the scaling factor, $x$ is the interval duration and $t$ is the x-intercept of the curve (see Figure 4 for a visualisation of the model). Critical interval duration ($x$_c) was defined as the interval duration at which percentage binding reached 90% of the asymptote, and was calculated using the following equation (3) for each Socialness/Belief condition:

$${x}_{c}=t-\text{l}\text{o}\text{g}\left(0.1\right)/s$$

Any curves with poor fits (r² ≤ .3) were excluded from the subsequent analysis; in total 8 participant curves (3 non-social, 2 social_no, 3 social_yes) were excluded. Exponential curve fits to the mean data (averaged over participants) for each condition are shown in Fig. 5, panel A. Analyses of variance were then carried out across the between-subject Socialness/Belief conditions on the asymptotes, x-intercepts and critical interval durations.

A one-way between-subjects ANOVA was carried out on the asymptotes of the three Socialness/Belief conditions. There was a significant effect of condition, F(2, 53) = 5.43, p = .007, η²_p = .17. Bonferroni corrected comparisons revealed that the difference between non-social (M = -40.0, 95% CI [-49.4, -30.5]) and social non-believers (Social No in Fig. 5; M = -60.8, 95% CI [-71.7, -50.0]) was significant, t(53) = 3.00, p = .012, d = 1.0. No other comparisons were significant: the difference between non-social and social believers (Social Yes in Fig. 5; M = -54.4, 95% CI [-62.5, -46.3]) was p = 0.06, the difference between social non-believers and social believers was p = 1. Similarly, a one-way between-subjects ANOVA on the x-intercepts of the Socialness/Belief conditions indicated a significant effect of condition, F(2, 53) = 7.20, p = .002, η²_p = .21. Bonferroni corrected planned comparisons revealed that the difference between non-social (M = 489.7, 95% CI [403.5, 576.0]) and social believers (Social Yes in Fig. 5; M = 272.0, 95% CI [200.3, 343.7]) was significant, t(53) = 3.64, p = .002, d = 1.2. No other comparisons were significant: the difference between non-social and social non-believers (Social No in Fig. 5; M = 326.3, 95% CI [ 177.4, 475.2]) was p = 0.07, the difference between social non-believers and social believers was p ≈ 1. Finally, a one-way between-subjects ANOVA on the critical interval durations of the Socialness/Belief conditions did not reveal any significant differences, F(2, 53) = 0.51, p = .604, η²_p = .02. All three parameters are plotted in Fig. 5.

Discussion for Experiments 1 & 2

As predicted, Experiments 1 and 2 demonstrated Social Agency in that we observed enhanced Temporal Binding effects in the social compared to non-social condition. However, contrary to prediction (Pfeiffer et al., 2012) there was no ‘peak’ of binding at any interval; instead, percentage binding evened out to an asymptote, never returning towards more accurate replication at any interval duration. Also contrary to prediction, the social condition exhibited more binding at shorter interval durations than non-social. Whilst the social enhancement was robust across blocks 1 and 2 of the experiment, the Temporal Binding effects in the non-social condition waned, increasing the social/non-social difference even further.

Binding effects are implicit: under-replicating intervals goes contrary to task instruction, and in principle over time participants should learn to accurately reproduce those intervals. However, the implicit social effect holds up during the entire experiment, with social binding no different between blocks. Non-social binding effects, however, diminish as the experiment proceeds, with participants getting overall more accurate at the task in block 2, compared to block 1. This would suggest that there may be some learning component in non-social participants, where task demands begin to override binding effects with task habituation. This non-social finding also argues against fatigue influences (e.g., boredom or rushing to complete the experiment) as non-social participants’ replication durations are in fact more accurate and hence longer in block 2.

While our results demonstrate a clear social binding enhancement, they do not reveal its specific underlying causes. Nevertheless, it was unexpected that believing (or not) the social interaction had no effect on binding. This aspect has previously been overlooked in the field as, generally, participants who do not believe in deceptive interactions are simply excluded (Beyer et al., 2017; Pfeiffer et al., 2012; Stephenson et al., 2018; Vogel et al., 2021). Participants who did not believe the social interaction would be predicted to behave the same as those in the non-social condition, since believing the social relevance of on-screen stimuli has been found to be a crucial aspect to socially driven responses (Caruana et al., 2017; Gobel et al., 2018). However, two factors may explain our contradicting results: first, self-report methods cannot be wholly relied upon due to demand characteristics (Nichols & Maner, 2008) and social desirability (Dodaj, 2012). When being told that they had been deceived, it is reasonable to consider that some of the participants who reported not believing the interaction did so out of embarrassment or not wanting to seem gullible, whilst genuinely believing. Second, belief may evolve over the experiment and many “non-believing” participants may have started off, or completed most of, the experiment as believing the cover story, only gradually catching on to the live videos being oddly similar to one another. Importantly, it is unclear if there is a threshold degree of certainty/belief that one is interacting with a conspecific beyond which one experiences Social Agency. Future experiments should continue to consider the role of belief in binding effects, rather than excluding non-believing participants.

Simply being in close proximity to a conspecific (i.e., Social Agency context; Silver et al., 2021) cannot play a significant role in the social binding enhancement found in these experiments. The contribution to the MLM of the random effect (i.e., Methodology) which would account for the methodological difference between Experiment 1, where a confederate went into an adjacent laboratory, and Experiment 2, where the researcher acted as confederate from the adjacent PC station within the same laboratory, was not significant and was removed through model simplification.

The unexpected finding that percentage binding approaches an asymptote, in all conditions, as action-effect interval durations increase (i.e., raw millisecond binding continues to increase as intervals increase) supports a theory of Temporal Binding where the underlying mechanism is the dynamic modulation of the ‘internal clock’ (Buonomano, 2007; Gibbon, 1977; Treisman, 1963), a neural pacing signal influenced by stimulation and motor activity (Wenke & Haggard, 2009). A slowing of this ‘internal clock’ would translate to a consistent percentage underestimation of time intervals, as in these experiments.

The overestimation of the shortest interval durations may be due to action preparation and/or execution confounds, which can influence the ‘internal clock’ mentioned above. Even when no participant action is part of the original interval, the reproduction of temporal intervals not only involves evaluation of the original interval, but also a motor production component which signals the start and end of each interval replication (Gilden et al., 1995). This motor component includes an inherent delay (Gilden et al., 1995), which would explain the ubiquitous overestimation of the shortest intervals. A consistent motor delay would have the greatest impact on the percentage binding of shorter intervals, with the delay effect reducing in influence as interval durations increased, which is what our results show.

Experiment 3 sought to address questions raised in the first two experiments. Firstly, it is possible that individual differences in baseline time perception could account for between-subject differences in Temporal Binding observed in the previous experiments. We therefore introduced a new observational within-participant control condition in Experiment 3. This allowed us to determine a ‘baseline’ of Temporal Binding for each participant and remove some between-participant variance. In the ‘observation’ condition of Experiment 3, we ensured that stimuli were consistently visual throughout all conditions by presenting two consecutive keypresses on screen. The same interaction paradigm was used as in Experiment 2, with some modifications: we used fewer ‘Interval’ conditions while increasing the highest tested interval to 2600 ms. The former was so that even with introducing a within-subject control, testing sessions remained around 1 hr. The latter was intended to help examine whether there was any decrease in binding at longer intervals.

Method

Participants

35 students and staff (24 females; age 17–37 years; mean age 23.2; AQ scores 9–28, mean AQ = 17.5) from the University of Aberdeen participated for renumeration or course credit. Written consent was obtained from all participants and the study was approved by the School of Psychology, University of Aberdeen Ethics Committee (PEC/4504/2020/5).

Design

Experiment 3 had a 2x2x7 mixed design, with a between-subjects factor of ‘Socialness’ (social/non-social) and two within-subjects factors of ‘Participant Role’ (action/observation) and ‘Interval’ (200–2600 ms in 400 ms increments; 7 conditions total).

Materials and Apparatus

Materials and apparatus used were identical to Experiments 1 and 2, except for the following: the control condition showed two parallel videos of a hand and keyboard, where two keypresses would happen consecutively (see Fig. 6). First, the left video would show a hand pressing the ‘B’ key on that keyboard (beginning 500 ms – 2500 ms from trial onset). Next, after a variable delay (200–2600 ms, in 400 ms increments, after the onset of the keypress), the right video would show a hand pressing ‘B’ on that keyboard. Video would terminate as before, 1000 ms after second ‘B’ press. Participants then completed the replication task as before, this time for the interval between the two on-screen keypresses.

Procedure

Procedure generally followed that for Experiments 1 and 2. The main difference was that now the AQ was completed between blocks implementing the two ‘role’ conditions (i.e., action and observation). It was reasoned that this extended break from replicating intervals would reduce carryover effects from whichever condition was carried out first to the second; condition order was counterbalanced across participants. There were a few further minor changes: participants now completed a shorter block of practice trials (5 trials interaction only, 5 full trials with interaction followed by replication task) for each ‘role’ condition. The practice trials were reduced because participants were completing two conditions, each with its own practice blocks. Also, feedback from participants in Experiment 2 stated that the initial number of practice trials was excessive and hindered task engagement. The videos used in the non-social condition and the observation control condition were now not of the hand of the researcher present. This allowed greater believability for participants in the ‘live’ interaction social condition as the hand seen on-screen in that condition was the hand of the researcher present, whilst the observation control condition showed a different hand altogether. Finally, with the number of tested intervals reduced, each of the 7 interval durations was tested 3 times per block, for 4 blocks, leading to 84 trials per condition. This meant each participant completed 168 trials total (either social action and observation conditions or non-social action and observation conditions).

Statistical Analysis for Experiment 3

Outlier detection and exclusion criteria were the same as in Experiments 1 and 2. Overall, one participant was excluded (correlation coefficient 2.5 MAD criterion) and 1.9% of trials were excluded (trial-wise 2.5 MAD interval duration criterion). After all exclusions, there were 15 participants in the social condition, and 18 in the non-social condition, to be included in analysis.

Percentage Temporal Binding was also calculated as for Experiments 1 and 2. However, Experiment 3 had a within-subjects control condition. Hence, to adjust for between-participant variability, analyses were carried out on the difference in percentage Temporal Binding between the experimental conditions (social/non-social) and the observation condition. Specifically, we subtracted, for each interval duration, the mean reproduced interval in the observation condition from each trial in the social/non-social condition for that interval duration participant-wise.

The Fixed effects structured of the MLM was the same as Experiments 1 and 2, apart from the following changes. Firstly, as the control and experimental conditions were counterbalanced, a fixed effect of Position was added, which accounted for whether the experimental condition came before or after the control for each participant. Secondly, Belief was removed as a fixed effect for two reasons: only one participant from the social condition of Experiment 3 reported not believing the ‘social’ interaction, and from the MLM results of Experiments 1 and 2, Belief did not significantly contribute to the model. The non-believing participant therefore remained in the MLM. The only random effect included in the model was Subject, to account for any individual differences in the between-subjects Socialness and Position factors. The starting point of the MLM was once again the maximal model and the same simplification strategy as in Experiments 1 and 2 was carried out. Lmer (Bates et al., 2015) syntax of optimal model reached for Experiment 3 was as follows:

Optimal Model = lmer (percent binding (control subtracted) ~ 1 + Interval (scaled) * Socialness * Block * Position + (1 + Interval (scaled) + Block | Subject), data = Exp3, control = lmerControl(optimizer = "bobyqa"))

Results

The output for the MLM to predict Temporal Binding (%) difference from factors Interval, Socialness, Block Number and Position is shown in Table 2. Figure 7 shows that in line with Experiments 1 and 2 results, and crucial to our hypotheses, there was a significant effect of Socialness, with greater binding difference in the social condition (M = -13.4, 95% CI [-15.2, -11.5]), compared to the non-social condition (M = -.9, 95% CI [-2.7, 1.0]), b = 12.73, SE = 5.73, t = 2.22, p = .034. There was a significant main effect of Interval on binding difference, b = 5.97, SE = 1.53, t = 3.90, p = <.001, with the 200 ms interval duration driving this effect (due to raw overestimation of this interval). There was a significant main effect of Block Number on binding difference, b = 3.59, SE = 1.25, t = 2.88, p = .007, with responses in block 2 (M = -4.1, 95% CI [-6.0, -2.3]) exhibiting less binding difference overall than block 1 (M = -8.9, 95% CI [-10.8, -7.0]). All these effects interacted with Position in a significant 4-way interaction (b = -6.91, SE = 3.21, t = -2.15, p = .032), binding difference diminished across blocks, and across the whole experiment binding difference diminished from the first to second condition completed, the degree of which depended on interval duration and condition (see Figure 10 in Supplementary Materials for a visualisation of this interaction).

Table 2: Results of MLM to predict percentage binding difference scores (experimental – control conditions) for Experiment 3.

Fixed Effect		b		SE		df		t		p
Intercept	-5.82		2.87		28.90		-2.03		.052
Interval	5.97		1.53		28.79		3.90		<.001
Socialness	12.73		5.73		28.90		2.22		.034
Block	3.59		1.25		29.33		2.88		.007
Position	-10.88		5.73		28.90		-1.90		.068
Interval*Socialness	-3.99		3.06		28.79		-1.30		.203
Interval*Block	-.54		.80		2565.41		-.67		.501
Socialness*Block	-3.62		2.49		29.33		-1.45		.157
Interval*Position	5.89		3.06		28.79		1.92		.065
Socialness*Position	2.28		11.47		28.90		.20		.844
Block*Position	4.20		2.49		29.33		1.69		.102
IntervalSocialnessBlock	.84		1.61		2565.41		.52		.602
IntervalSocialnessPosition	4.20		6.13		28.79		.69		.499
IntervalBlockPosition	-3.31		1.61		2565.41		-2.06		.039
SocialnessBlockPosition	1.28		4.98		29.33		.26		.798
IntervalSocialnessBlock*Position	-6.91		3.21		2565.41		-2.15		.032

Discussion

Experiment 3 confirmed the findings of Experiments 1 and 2, demonstrating more Temporal Binding when participants were told they were interacting with another person in real time as compared to videos across all tested intervals. This experiment introduced a within-participant control condition, which allowed for controlling for individual differences in baseline replication. This control condition also showed that whereas a social context elicited reliable binding (especially in the earlier blocks), non-social Temporal Binding never significantly differed from baseline observational replication in either block.

Overall, the difference in percentage binding was greatest for the shortest interval duration and decreased as interval durations increased. This was driven by the overestimation of the shortest interval (200 ms), which was also observed in Experiments 1 and 2 (reasons for which are discussed in the Exp 1 & 2 discussion section). This did not impact the effect of Socialness.

Whereas in Experiments 1 and 2 the difference between social and non-social conditions increased across blocks due to a decrease in non-social binding in the later blocks, if anything Experiment 3 shows a slight (non-significant) decrease in difference between the social and non-social conditions, due to a decrease in social binding difference.

Importantly, the main effects of Socialness and Block Number in Experiment 3 are qualified by a 4-way interaction with Position and Interval. In short, the least percentage binding difference was found in a) the non-social condition, b) the longer intervals, c) block 2 and d) the second half of the experiment. The fact that the binding difference decreases in the social condition may initially seem like it contradicts the results from Experiments 1 and 2, where percentage binding remained constant throughout. However, Experiment 3 uses a different measure (within-subject percentage binding difference) from that used in Experiments 1 and 2 (percentage binding), so a comparison between the two sets of results must be appropriately cautious. The fact that the main effect of Socialness is observed with both measures, despite changes to the experimental design and sample size, is testament to the robustness of the social/non-social effect.

Three experiments used an action-effect interval replication task to investigate whether and how our social Sense of Agency differs from a generic (non-social) Sense of Agency. Specifically, we tested if participants’ implicit Sense of Agency depended on whether they interacted (or were told they interacted), in real time, with a social partner or with a video (social and non-social condition, respectively). The three experiments consistently showed greater Temporal Binding effects in social, compared to non-social, conditions, despite the stimulus remaining exactly the same. Experiments 1 and 2 used a between-subjects design and found that participants in the social condition demonstrated substantially more binding than those in the non-social condition, regardless of the action-effect interval duration. This difference was also reflected in Temporal Binding occurring at shorter action-effect interval durations in the social condition than in the non-social condition. Experiment 3 replicated this difference between social and non-social Temporal Binding, but this time using a within-participant control condition in which participants merely observed two keypresses without any action initiation. This showed that the social condition represented an enhancement of Temporal Binding with respect to the observation condition, whereas the non-social condition’s binding did not differ from observation. The results are in line with similar previous results (Pfister et al., 2014; Stephenson et al., 2018; Vogel et al., 2021) indicating that the experience of social interaction can elicit so-called ‘hyper-binding’ (Vogel et al., 2021; but see discussion below), in the sense that Sense of Agency in social interaction is above and beyond that observed in observational tasks and especially non-social contexts. Furthermore, these findings are in line with our recent proposal that Social Agency can be considered to vary along a continuum with cooperation as a determining factor, where cooperative social interactions enhance agency, compared to non-social interactions (Silver et al., 2021). While the current study uses a minimalist paradigm, the (explained as) live social interaction effectively creates a cooperative context in that both agents must cooperate to move the trial forward.

The paradigm improved upon previous similar research in the field. First, in contrast to most Sense of Agency studies, we used a wide range of intervals to systematically investigate not only the social/non-social difference, but also the time course of binding effects, something only previously investigated using an explicit measure (Pfeiffer et al., 2012). Furthermore, in contrast to, for instance, Pfister et al. (2014), we included a non-social condition that was similar in all respects to the social condition except for what participants were told. Third, the effect of the participant’s action was a response from the ‘other’, creating a direct social effect in the social condition without a tone mediating between a person’s and the other’s action (Silver et al., 2021). Finally, in contrast to Vogel et al (2021), our study used credible social stimuli that depended less crucially on participants’ beliefs and contained direct comparisons between non-social interaction and an observation (control) condition.

The exact underlying mechanisms that cause this consistent enhancement of Temporal Binding in social contexts are largely unknown. The fact that, in Experiments 1 and 2, percentage Temporal Binding as a function of interval duration forms a negative exponential curve, bottoming out at an asymptote, supports our argument that cooperative interaction induces a modulation of the ‘internal clock’ (Buonomano, 2007; Gibbon, 1977; Treisman, 1963). An individual whose internal clock has a slower oscillatory component, would count fewer oscillations over a defined period compared to an individual with a faster oscillation, effectively reducing the perceived time that passed (which in our experiments would equate to an underestimation of interval duration, that is, enhanced binding). This slower oscillation would equate to a consistent percentage reduction in perceived time, irrespective of duration.

There is a suggestion in the literature of a specific, dedicated clock mechanism for Temporal Binding effects (Fereday & Buehner, 2017). If this were the case, what could drive the observed social interaction induced increase in underestimation? One possibility is that physiological arousal plays a role in the binding enhancement. Humans are a pro-social species. Cooperative social, compared to non-social interaction, could decrease arousal, and therefore have physiological outcomes, which might impact time perception. This may seem initially counter-intuitive because interacting with a conspecific could be posited to be more arousing than interacting with a computer, but this is not always the case. When a human ‘other’ consistently performs a pro-social response to our actions (as in our paradigm), we use cognitive shortcuts such as mentalising (Freeman, 2016; Frith & Frith, 2006), to both reduce cognitive load (Freeman, 2016; Frith & Frith, 2006) and speed up cognitive processing (Itier et al., 2006; Rousselet et al., 2008); we have pro-social expectations of human interaction partners, which we do not have for artificial, non-volitional partners. For example, in everyday life we ubiquitously perform eye movements intended to lead the gaze of others to objects or locations of interest (i.e., gaze cues; de Hamilton, 2016) because we expect a conspecific to follow these, the pro-social response.

A resultant decrease in heart rate or electrodermal activity (e.g., galvanic skin response; Nichols & Champness, 1971) could influence the oscillatory component of the internal clock and thus lead to underestimation. Removing the other discussed contributing influences on time perception, which have stronger influence for shorter interval durations (e.g., a motor delay component of interval replication), this is consistent with what we found throughout all three experiments in this study (i.e., percentage binding approached asymptote as interval duration increased).

Nonetheless, while our results generally confirm the social/non-social effect, they also show notable differences with respect to previous studies. First, whereas we found an overall increase of Sense of Agency in the social condition, we did not find a difference in the time course of Sense of Agency in the way Pfeiffer et al. (2012) did. They found a shift in the peak of the time course (the time point at which the feeling of causing the other agent’s gaze shift was the highest) towards longer intervals when the other agent could choose whether to imitate the participant’s action or not, as compared to where they always had to imitate the participant’s action. Furthermore, in the latter condition they found a marked decrease in explicit agency when intervals became longer. We did not observe such a decrease in Temporal Binding, as the percentage binding in the social condition remained flat and negative (reaching an asymptote). Pfeiffer and colleagues used gaze shifts instead of button presses and it is possible that difference between our findings and theirs, in terms of time course, could be a consequence of these differences in methods and the measures used for interval-estimation (explicit Sense of Agency ratings vs interval estimation by reproduction). It is nonetheless surprising that our results show Temporal Binding up to 2100 ms (Experiments 1 & 2) and even 2600 ms (Experiment 3), whereas Pfeiffer et al. (2012) found a consistent linear decline in explicit agency from 800 ms to 4000 ms, when the other agent could only imitate the participant, as was the case in our three experiments. Future experiments should therefore examine whether increasing the number of response options, and hence increasing the agency of the responding agent, might change the time course in paradigms such as ours. In addition, an explicit measure should be used to make results directly comparable to Pfeiffer et al. (2012).

Second, Vogel et al. (2021) not only found Temporal Binding in their non-social conditions, but also found substantial Temporal Binding in their observation-only conditions. However, neither of their experiments directly compared an observation condition with a non-social condition with visual human-resemblance (i.e., dissociating belief and appearance conditions; computer interaction partner with face-like stimuli arrangement), therefore we cannot directly compare their results to our comparison between non-social and observation conditions where appearance was consistently social (i.e., real hands). In our Experiment 3, there was no difference in Temporal Binding between the non-social and observation condition. This is surprising, in that the non-social condition should in principle elicit more Temporal Binding compared to observation, since it is akin to a standard cause-effect sequence initiated by participant action, which has been widely shown to lead to both explicit and implicit Sense of Agency(Moore et al., 2012; Moore & Obhi, 2012; Wen, 2019). It could be that our Experiment 3 showed more than expected binding in the observation condition as there was still action involvement in the participants’ responses to evaluate the interval, leading to similar binding as in the non-social condition, and more ‘hyper-binding’ (term coined by Vogel et al., 2021) in the social condition. Alternatively, it could be that the underestimation observed in non-social and observation conditions that we infer as Temporal Binding is an artefact of the interval replication method, and in fact the social binding reflects normal Temporal Binding and not ‘hyper-binding’, with the nature of the non-social condition somehow inhibiting Sense of Agency — for instance due to participants seeing their action as causing the onset of a video in which a hand pushes a button, rather than feeling they directly caused a hand to press a button.

Third, in Experiments 1 and 2, we found no effect of belief: it did not matter whether participants believed that they were interacting live with a conspecific or not. Of course, this could simply reflect desire, especially among psychology students, to say they suspected the deceptive manipulation, or it may reflect disbelief that participants developed by the end of the experiment, but not necessarily at the outset.

In all three experiments, in the social condition, the other’s response was never uncertain other than in terms of temporal contiguity (as in many SoA paradigms, e.g., Pfister et al., 2014). Hence, the effect of including non-cooperative actions on binding cannot be determined from this study (nor was it designed to test this question). However, the cooperation continuum theory of Social Agency would predict diminished binding effects on such trials compared to non-social interaction (Silver et al., 2021). Previous research suggests that expending more cognitive resources predicting the actions of an uncertain or uncooperative partner interferes with agency, causing diminished implicit binding (Beyer et al., 2017, 2018).

In our experiments, the participant always initiated the interaction and did so at a time of their choosing. Previous research has demonstrated that acting in a similar manner in a social interaction with another participant who is labelled as a ‘follower’, induces vicarious Social Agency, where agency is felt not only for one’s own action, but also the subsequent action of the other (Pfister et al., 2014; Silver et al., 2021). Although the role of the ‘leader’ is implicit in the current study, it would be reasonable to assume that acting first and volitionally, in an interaction with a conspecific partner, would still have a similar influence over binding effects. This could contribute to the enhanced social, compared to non-social binding effects found in this study. It would be of great interest to reverse the interval initiation and completion actions in this study, having the participant act in response to the on-screen keypress. This would be predicted to greatly inhibit agency effects in the social condition (as in the ‘follower’ role of Pfister et al., 2014), but there should be no such negative influence on binding in the non-social condition (Silver et al., 2021). Responding to non-social prompts is common in everyday life (e.g., stopping at a red traffic light whilst driving), and since the action prompt is performed by a non-volitional entity, it should not negatively influence agency (Silver et al., 2021).

In conclusion, the present study found consistent evidence that (implied) social interactions induced enhanced Temporal Binding effects compared to non-social interactions, over a wide range of intervals between one’s actions and the other’s response. These findings are in line with the continuum theory of Social Agency (Silver et al., 2021), which argues that (implied) cooperative social interactions, as in this study, enhance agency compared to non-social interactions. We suggest that this enhancement derives from a modulation of the ‘internal clock’ model of time perception.

Competing interests: The authors have no competing interests to declare that are relevant to the content of this article.

Funding: This work was supported by the UKRI Biotechnology and Biological Sciences Research Council (BBSRC) grant number BB/M010996/1 to Crystal Silver (“Mechanisms of Social Agency”), and by a Carnegie Trust Research Incentive Grant to Bert Timmermans and Ramakrishna Chakravarthi ("Experiencing myself through you: Self-Agency in social interaction" - RIG008270).

Ethics approval: This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the School of Psychology, University of Aberdeen Ethics Committee (PEC/4344/2019/10).

Author contributions: C.A.S., B.T. and R.C. contributed to planning the experiments. C.A.S. and B.T. programmed the experiments and collected the data. C.A.S. and R.C. analysed and visualized the data. C.A.S. prepared the drafts and final version. C.A.S., B.T., R.C., and B.W.T. contributed critical revisions.

Consent to participate: Informed consent was obtained from all individual participants included in the study.

Data availability statement: Data is deposited on OSF - https://osf.io/usqjy/.

Baron-Cohen, S., Wheelwright, S., Skinner, R., Martin, J., & Clubley, E. (2001). The Autism-Spectrum Quotient (AQ): Evidence from Asperger Syndrome/High-Functioning Autism, Males and Females, Scientists and Mathematicians. In Journal of Autism and Developmental Disorders (Vol. 31, Issue 1).
Barr, D. J., Levy, R., Scheepers, C., & Tily, H. J. (2013). Random effects structure for confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language, 68(3), 255–278. https://doi.org/10.1016/j.jml.2012.11.001
Bates, D., Mächler, M., Bolker, B. M., & Walker, S. C. (2015a). Fitting linear mixed-effects models using lme4. Journal of Statistical Software, 67(1). https://doi.org/10.18637/jss.v067.i01
Bates, D., Mächler, M., Bolker, B. M., & Walker, S. C. (2015b). Fitting linear mixed-effects models using lme4. Journal of Statistical Software, 67(1). https://doi.org/10.18637/jss.v067.i01
Bayliss, A. P., Pellegrino, G. Di, & Tipper, S. P. (2005). Sex differences in eye gaze and symbolic cueing of attention. Quarterly Journal of Experimental Psychology Section A: Human Experimental Psychology, 58(4), 631–650. https://doi.org/10.1080/02724980443000124
Beck, B., Di Costa, S., & Haggard, P. (2017). Having control over the external world increases the implicit sense of agency. Cognition, 162, 54–60. https://doi.org/10.1016/j.cognition.2017.02.002
Beyer, F., Sidarus, N., Bonicalzi, S., & Haggard, P. (2017). Beyond self-serving bias: Diffusion of responsibility reduces sense of agency and outcome monitoring. Social Cognitive and Affective Neuroscience, 12(1), 138–145. https://doi.org/10.1093/scan/nsw160
Beyer, F., Sidarus, N., Fleming, S., & Haggard, P. (2018). Losing control in social situations: How the presence of others affects neural processes related to sense of agency. ENeuro, 5(1). https://doi.org/10.1523/ENEURO.0336-17.2018
Brandi, M. L., Kaifel, D., Bolis, Di., & Schilbach, L. (2019). The Interactive Self - A Review on Simulating Social Interactions to Understand the Mechanisms of Social Agency. In i-com (Vol. 18, Issue 1, pp. 17–31). Walter de Gruyter GmbH. https://doi.org/10.1515/icom-2018-0018
Buehner, M. J. (2012). Temporal Binding. In Attention and Time (p. 14). Oxford University Press. https://doi.org/DOI:10.1093/acprof:oso/9780199563456.003.0015
Buehner, M. J., & Humphreys, G. R. (2009). Causal Binding of Actions to Their Effects. Psychological Science, 20(10), 1211–1228.
Buonomano, D. v. (2007). The biology of time across different scales. Nature Chemical Biology, 3(10), 594–597.
Caruana, N., de Lissa, P., & McArthur, G. (2017a). Beliefs about human agency influence the neural processing of gaze during joint attention. Social Neuroscience, 12(2), 194–206. https://doi.org/10.1080/17470919.2016.1160953
Caruana, N., de Lissa, P., & McArthur, G. (2017b). Beliefs about human agency influence the neural processing of gaze during joint attention. Social Neuroscience, 12(2), 194–206. https://doi.org/10.1080/17470919.2016.1160953
Cravo, A. M., Claessens, P. M. E., & Baldo, M. V. C. (2011). The relation between action, predictability and temporal contiguity in temporal binding. Acta Psychologica. https://doi.org/10.1016/j.actpsy.2010.11.005
de Hamilton, A. F. C. (2016). Gazing at me: The importance of social meaning in understanding direct-gaze cues. In Philosophical Transactions of the Royal Society B: Biological Sciences (Vol. 371, Issue 1686). Royal Society of London. https://doi.org/10.1098/rstb.2015.0080
Dewey, J. A., & Knoblich, G. (2014a). Do implicit and explicit measures of the sense of agency measure the same thing? PLoS ONE, 9(10). https://doi.org/10.1371/journal.pone.0110118
Dewey, J. A., & Knoblich, G. (2014b). Do implicit and explicit measures of the sense of agency measure the same thing? PLoS ONE, 9(10). https://doi.org/10.1371/journal.pone.0110118
Dodaj, A. (2012). Social desirability and self-reports: Testing a content and response-style model of socially desirable responding. Europe’s Journal of Psychology, 8(4), 651–666. https://doi.org/10.5964/ejop.v8i4.462
Engbert, K., Wohlschläger, A., Thomas, R., & Haggard, P. (2007). Agency, Subjective Time, and Other Minds. Journal of Experimental Psychology: Human Perception and Performance, 33(6), 1261–1268. https://doi.org/10.1037/0096-1523.33.6.1261
Fereday, R., & Buehner, M. J. (2017). Temporal Binding and Internal Clocks: No Evidence for General Pacemaker Slowing. Journal of Experimental Psychology: Human Perception and Performance, 43(5), 971–985. https://doi.org/10.1037/xhp0000370.supp
Freeman, C. (2016). What is Mentalizing? An Overview. British Journal of Psychotherapy, 32(2), 189–201. https://doi.org/10.1111/bjp.12220
Freeth, M., Foulsham, T., & Kingstone, A. (2013). What Affects Social Attention? Social Presence, Eye Contact and Autistic Traits. PLoS ONE, 8(1). https://doi.org/10.1371/journal.pone.0053286
Frith, C. D., & Frith, U. (2006). The Neural Basis of Mentalizing. In Neuron (Vol. 50, Issue 4, pp. 531–534). https://doi.org/10.1016/j.neuron.2006.05.001
Gibbon, J. (1977). Scalar Expectancy Theory and Weber’s Law in Animal Timing. In Psychological Review (Vol. 84, Issue 3).
Gilden, D. L., Thornton, T., & Mallon, M. W. (1995). 1/f Noise in Human Cognition. Science, 1837–1839.
Gobel, M. S., Tufft, M. R. A., & Richardson, D. C. (2018). Social Beliefs and Visual Attention: How the Social Relevance of a Cue Influences Spatial Orienting. Cognitive Science, 42, 161–185. https://doi.org/10.1111/cogs.12529
Haggard, P. (2017). Sense of agency in the human brain. Nature Reviews Neuroscience, 18(4), 197–208. https://doi.org/10.1038/nrn.2017.14
Haggard, P., & Chambon, V. (2012). Sense of agency. Current Biology, 22(10), R390–R392. https://doi.org/10.1016/j.cub.2012.02.040
Haggard, P., Clark, S., & Kalogeras, J. (2002). Voluntary action and conscious awareness. Nature Neuroscience, 5(4), 382–385. https://doi.org/10.1038/nn827
Itier, R. J., Latinus, M., & Taylor, M. J. (2006). Face, eye and object early processing: What is the face specificity? NeuroImage, 29(2), 667–676. https://doi.org/10.1016/j.neuroimage.2005.07.041
Leys, C., Ley, C., Klein, O., Bernard, P., & Licata, L. (2013). Detecting outliers: Do not use standard deviation around the mean, use absolute deviation around the median. Journal of Experimental Social Psychology, 49(4), 764–766. https://doi.org/10.1016/j.jesp.2013.03.013
Moore, J. W., Middleton, D., Haggard, P., & Fletcher, P. C. (2012). Exploring implicit and explicit aspects of sense of agency. Consciousness and Cognition, 21(4), 1748–1753. https://doi.org/10.1016/j.concog.2012.10.005
Moore, J. W., & Obhi, S. S. (2012). Intentional binding and the sense of agency: A review. In Consciousness and Cognition. https://doi.org/10.1016/j.concog.2011.12.002
Nichols, A. L., & Maner, J. K. (2008). The good-subject effect: Investigating participant demand characteristics. Journal of General Psychology, 135(2), 151–165. https://doi.org/10.3200/GENP.135.2.151-166
Nichols~, K. A., & Champness’, B. G. (1971). Eye Gaze and the GSR. In JOURNAL OF EXPERIMENTAL SOCIAL PSYCHOLOGY (Vol. 7).
Pfeiffer, U. J., Schilbach, L., Jording, M., Timmermans, B., Bente, G., & Vogeley, K. (2012). Eyes on the mind: Investigating the influence of gaze dynamics on the perception of others in real-time social interaction. Frontiers in Psychology, 3(DEC). https://doi.org/10.3389/fpsyg.2012.00537
Pfeiffer, U. J., Schilbach, L., Timmermans, B., Kuzmanovic, B., Georgescu, A. L., Bente, G., & Vogeley, K. (2014). Why we interact: On the functional role of the striatum in the subjective experience of social interaction. NeuroImage. https://doi.org/10.1016/j.neuroimage.2014.06.061
Pfister, R., Obhi, S. S., Rieger, M., & Wenke, D. (2014). Action and perception in social contexts: intentional binding for social action effects. Frontiers in Human Neuroscience, 8. https://doi.org/10.3389/fnhum.2014.00667
R Core Team. (2019). R: A language and environment for statistical computing. R Foundation for Statistical Computing.
Rashal, E., & Yeshurun, Y. (2014). Contrast dissimilarity effects on crowding are not simply another case of target saliency. Journal of Vision, 14(6). https://doi.org/10.1167/14.6.9
Rousselet, G. A., Husk, J. S., Bennett, P. J., & Sekuler, A. B. (2008). Time course and robustness of ERP object and face differences. Journal of Vision, 8(12). https://doi.org/10.1167/8.12.3
RStudio Team. (2019). RStudio: Integrated Development for R. RStudio Inc.
Ruess, M., Thomaschke, R., & Kiesel, A. (2017). The time course of intentional binding. Attention, Perception, and Psychophysics, 79(4), 1123–1131. https://doi.org/10.3758/s13414-017-1292-y
Ruess, M., Thomaschke, R., & Kiesel, A. (2018). The Time Course of Intentional Binding for Late Effects. In Timing and Time Perception (Vol. 6, Issue 1, pp. 54–70). Brill Academic Publishers. https://doi.org/10.1163/22134468-00002099
Schneider, W., Eschman, A., & Zuccolotto, A. (2002). E-Prime User’s Guide. Psychology Software Tools Inc.
Scolari, M., Kohnen, A., Barton, B., & Awh, E. (2007). Spatial attention, preview, and popout: Which factors influence critical spacing in crowded displays? Journal of Vision, 7(2). https://doi.org/10.1167/7.2.7
Silver, C. A., Tatler, B. W., Chakravarthi, R., & Timmermans, B. (2021a). Social Agency as a continuum. Psychonomic Bulletin & Review, 28, 434–453. https://doi.org/10.3758/s13423-020-01845-1/Published
Silver, C. A., Tatler, B. W., Chakravarthi, R., & Timmermans, B. (2021b). Social Agency as a continuum. Psychonomic Bulletin & Review, 28, 434–453. https://doi.org/10.3758/s13423-020-01845-1/Published
Soo, L., Chakravarthi, R., & Andersen, S. K. (2018). Critical resolution: A superior measure of crowding. Vision Research, 153, 13–23. https://doi.org/10.1016/j.visres.2018.08.005
Stephenson, L. J., Edwards, S. G., Howard, E. E., & Bayliss, A. P. (2018a). Eyes that bind us: Gaze leading induces an implicit sense of agency. Cognition, 172, 124–133. https://doi.org/10.1016/j.cognition.2017.12.011
Stephenson, L. J., Edwards, S. G., Howard, E. E., & Bayliss, A. P. (2018b). Eyes that bind us: Gaze leading induces an implicit sense of agency. Cognition, 172, 124–133. https://doi.org/10.1016/j.cognition.2017.12.011
Treisman, M. (1963). TEMPORAL DISCRIMINATION AND THE INDIFFERENCE INTERVAL : IMPLICATIONS FOR A MODEL OF THE “INTERNAL CLOCK’’’ 1.” Psychological Monographs: General and Applied, 77(13), 1–31.
Vogel, D. H. v, Jording, M., Esser, C., Weiss, P. H., & Vogeley, K. (2021). Temporal binding is enhanced in social contexts. Psychonomic Bulletin & Review. https://doi.org/10.3758/s13423-021-01928-7/Published
Wen, W. (2019). Does delay in feedback diminish sense of agency? A review. In Consciousness and Cognition. https://doi.org/10.1016/j.concog.2019.05.007
Wen, W., Yamashita, A., & Asama, H. (2015). The influence of goals on sense of control. Consciousness and Cognition, 37, 83–90. https://doi.org/10.1016/j.concog.2015.08.012
Wenke, D., & Haggard, P. (2009). How voluntary actions modulate time perception. Experimental Brain Research, 196(3), 311–318. https://doi.org/10.1007/s00221-009-1848-8
Wickham, H. (2016). ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag.

No competing interests reported.

SupplementaryMaterials.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

The time course of social and non-social implicit Sense of Agency

Status:

Version 1

Abstract

Figures

Introduction

Experiments 1 And 2

Method for Experiment 1

Method for Experiment 2

Experiment 3

Method

Statistical Analysis for Experiment 3

Results

Discussion

General Discussion

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1