Action planning and execution cues influence economic partner choice

Prudently choosing who to interact with and who to avoid is an important ability to ensure that we benefit from a cooperative interaction. While the role of others ’ preferences, attributes, and values in partner choice have been established (Rossetti, Hilbe & Hauser, 2022), much less is known about whether the manner in which a potential partner plans and implements a decision provides helpful cues for partner choice. We used a partner choice paradigm in which participants chose who to interact with in the Prisoners ’ Dilemma. Before choosing a cooperation partner, participants were presented with information about the potential partners ’ decision-related actions in another round of the Prisoners ’ Dilemma. They received either information about the potential partners ’ planning during decision making (i.e., decision-time; Experiment 1) or action execution during decision implementation (i.e., movement directness; Experiment 2). Across both games, participants preferred to interact with those who planned actions quickly or executed actions with direct and smooth movements, indicating that they were cooperating confidently and without deliberation. This demonstrates that action cues present in either the planning or implementation of economic decisions influence partner choice. We discuss implications of this finding for human decision-making and perception-action coupling in action understanding.


Introduction
A key driving factor behind humans' success as cooperators is the ability to prudently choose whom to interact with and whom to avoid.This ability to effectively select among potential partners stabilizes cooperation, motivating people to be fair, cooperative and trustworthy (André & Baumard, 2011;Barclay & Willer, 2007;Gächter & Thöni, 2005).By now, the kinds of cues that influence the perception of a (potential) cooperation partner have been well researched.For instance, individuals prefer those who exhibit generosity for the social good (Barclay & Willer, 2007); those who enforce norms by punishing transgressors (Raihani and Bshary (2015), and those who are willing to share both the monetary and effort costs involved in cooperation (Hoffman, Yoeli, & Nowak, 2015;McEllin, Felber, & Michael, 2023;McEllin & Michael, 2022).Moreover, physical attributes, such as facial features (Bonnefon, Hopfensitz, & De Neys, 2017) and voice (O'Connor & Barclay, 2017) or personal values such as political ideologies (Balliet, Tybur, Wu, Antonellis, & Van Lange, 2018) and religious conviction (McCullough, Swartwout, Shaver, Carter, & Sosis, 2016) may be important for deciding who to interact with and who to avoid.
While the role of others' preferences, attributes, and values in partner choice has been established (Rossetti et al., 2022), much less is known about whether the manner in which a potential partner plans and implements a decision provides meaningful cues for partner choice.Across many domains, an actor's (un)certainty when making a decision is reflected in their response times (Greene, 2007;Hyman, 1953;Vickers, 1970).With regards to planning a decision in a cooperative interaction, it has been shown that decision time serves as an indicator of the extent to which an actor is deliberating, and is seen as trustworthy (Jordan, Hoffman, Nowak, & Rand, 2016).Cooperators who plan responses quickly can be described as 'intuitive cooperators' who cooperate without calculating, whereas those who plan their responses slowly can be described as 'deliberative cooperators' who cooperate only after weighing the costs and benefits of cooperation (Rand, Greene, & Nowak, 2012;Bear & Rand, 2016; though the generality of this effect depends on the method by which intuitive processing is induced, Kvarven et al., 2020).Individuals are more likely to trust intuitive rather than deliberative cooperators in economic interactions (Hoffman, Yoeli and Nowak, 2015;Jordan et al., 2016), but the extent to which decision time affects partner choice is unknown.
Concerning action execution, spatial and temporal characteristics of an actor's hand movements may reveal the extent to which they are confident about a decision (Patel, Fleming, & Kilner, 2012;Slepian, Young, Rutchick, & Ambady, 2013).Specifically related to implementing decisions in an economic interaction, the extent to which an actor is conflicted about a decision to cooperate or defect is reflected in the directness of their movement kinematics towards a response option.While actors who are unconflicted about an economic decision execute their decision by moving more smoothly and directly towards their chosen option, those conflicted by an economic decision move less smoothly and less directly towards their chosen option (Kieslich & Hilbig, 2014).
Considering that across different social interactions, the cues present in others' actions guide how we interact with them by informing us about their mental states (Cavallo, Koul, Ansuini, Capozzi, & Becchio, 2016;Krishnan-Barman, Forbes, & Hamilton, 2017;Wolpert, Doya, & Kawato, 2003), the question arises whether action cues with regard to planning and executing decisions may influence important social decisions such as who we decide to interact with.Thus, the current study aimed to investigate whether decision-time as a cue that reflects action planning (Experiment 1) and movement directness during action execution (Experiment 2) inform partner choice in an economic interaction.We used the prisoner's dilemma (Rapoport, Chammah, & Orwant, 1965) a game in which two individuals can either choose to cooperate with their partner for a moderate mutual benefit if their partner also chooses to cooperate or defect for a large individual benefit if their partner chooses to cooperate (while receiving a small payoff if their partner also chooses to defect).This game effectively captures the essence of both small and large scale cooperative problems, thus lending itself well to a task in which participants need to make judgements about the likely cooperativeness of an actor based on action cues.
Participants watched potential partners making a choice in a previous round of the economic game, before expressing their preference to interact with this partner in the future.In Experiment 1a & 2a participants explicitly chose between two potential partners who made the same choice and in Experiment 1b & 2b participants expressed the extent to which they would like to interact with the potential partner in the future.In Experiment 1 potential partners would plan their decisions either quickly or slowly; in Experiment 2, potential partners would execute their decisions either with smooth and direct movements or with indirect and jittery movements.Participants watched a computer screen on which they saw a computerized version of the economic games with the two options presented in the top-left (e.g., cooperate) and topright corners (e.g., defect) of the screen (Kieslich & Hilbig, 2014).The potential partners' decision times could be inferred from four loading dots lighting up over and over in sequence until the decision was made, and the chosen option was highlighted.In trials where participants received information about the potential partners' movements, they saw the partner's mouse cursor starting to move towards the response options with different amounts of jitter (i.e., smoothness) and following trajectories of varying straightness (i.e., directness).
Finally, to directly test which mental states observers infer from decision time and movement directness in the context of an economic interaction, participants in Experiments 1c and 2c rated actors' (non) deliberation about and confidence in their decisions based on the manner in which they planned and executed decisions.We had participants judge actors' (non)deliberation because uncalculating cooperation (i.e., fast response times) has been shown to signal of reliability and trustworthiness (Hoffman et al., 2015;Jordan et al., 2016;Rand et al., 2012).We had participants judge actors' confidence as quick response times and smooth and direct movements reflect an individual's commitment to and certainty about a decision (Kieslich & Hilbig, 2014;Patel et al., 2012;Slepian et al., 2013),

Predictions
Our first key prediction was that if action planning cues inform partner choice by signalling confidence and/or (non)deliberation, participants should prefer potential partners who planned cooperative decisions quickly over those who planned cooperative decisions slowly.
Our second key prediction was that if action execution cues inform partner choice by signalling confidence and/or (non)deliberation, they should prefer partners who executed cooperative decisions with direct (and smooth) movements over those who executed cooperative decisions with indirect (and jittery) movements.
With respect to actors who chose to defect, our predictions were more open.On one hand, participants may prefer those who defect slowly or with indirect movements in the prisoners' dilemma because this signals lack of confidence and/or conflict about their decision.On the other hand, participants may prefer those who defect with quick or direct movements as their decisions may be more predictable.

Open practices statement
Data, analysis scripts, and experimental materials for all Experiments are available at: https://osf.io/yebvt/.Pre-registrations for Experiments are listed in their respective methods sections.

Experiment 1: Decision time
Experiment 1 aimed to test the effect of decision time upon partner choice in a prisoners' dilemma.Firstly, Experiment 1a explicitly tested whether participants would choose to interact with those who planned decisions quickly over those who planned decisions slowly.Secondly, Experiment 1b varied both decision time (fast, medium, slow) and the partners' decision (i.e., to cooperate or defect), in order to test how both of these factors influence partner preference.Thirdly, Experiment 1c aimed to establish whether decision time in the planning phase of an action signals confidence and/or deliberation about a decision.

Experiment 1a: Method
The pre-registration for this study can be found at: https://aspred icted.org/ka8q9.pdf2.1.1.1.Participants.A g*power analysis using pilot data (from 20 participants) for a Bonferroni corrected (for two comparisons) one sample t-test (d = 0.488, a = 0.025, 1β = 0.95) determined that we should collect 66 participants.Thus, we decided to collect 72 participants (after the replacement of excluded participants) to ensure equal counterbalancing.Using the Prolific recruitment platform (www.prolific.co)we recruited 92 participants (20 removed from the analysis because they did not meet our ore-registered inclusion criteria), with a mean age of 34.92 (SD = 11.94), with 56 identifying as male, 35 identifying as female, and 1 identifying as other.The experiment was conducted in accordance with the Declaration of Helsinki and was approved by the Psychological Research Ethics Board (PREBO) at Central European University, Private University, Vienna, Austria.

Apparatus and Stimuli.
Our study was hosted on the Pavlovia online hosting platform and programmed using JavaScript (a combination of PsychoJS and custom written functions).Stimuli size was defined using standard height units (relative to the screen size).
2.1.1.2.1.Payoff matrix -Prisoners' Dilemma.The payoff matrix for the prisoners' dilemma displayed all the possible outcomes in the game and was presented in the middle of the screen (text height = 0.05).For each outcome, the potential partner's payoff was displayed in green and their partners (i.e., from the previous interaction that was presented to L. McEllin et al. the participant) was displayed in blue.The values for the payoff matrix are displayed in Table 1.2.1.1.2.2.Partner response screen.The two response options ('Cooperate' or 'Defect') were displayed in the top left (x = − 0.56, y = 0.4) and top right (x = 0.56, y = 0.4) corners of the screen, with the positioning of the response options being counterbalanced across participants.The decision-making time was represented with four loading dots (text height = 0.05) which repeated in a sequence whilst a potential partner was making their decision.Below the dots, text was displayed indicating that the potential partner was planning their decision.After the decision had been made, a white ellipse (x = 0.2, y = 0.14) around one of the options represented the potential partner's choice.
2.1.1.2.3.Virtual partner behavior.Potential partners were programmed to respond quickly, after anywhere between 2 and 4 s, or respond slowly, after anywhere between 14 and 16 s.
2.1.1.3.Procedure.After clicking our link on Prolific, participants would be informed that they would first play one round of an economic game with a random partner, before being given the opportunity to play several more rounds of that game with a series of partners whom they had the opportunity to choose (in actuality they would choose between potential partners only and then the experiment would end, i.e., they would not play the economic game with the chosen partners).They were told that they would play the economic game with a series of partners who had already given their responses in a previous testing session, believing that their responses would be paired with the responses of others from a later testing session.Participants were instructed to play the game just like they would if playing live, because the stakes were just the same as in a live version of the game, with both their own payoff and their partner's deferred payoff depending on how they had both responded.They were then presented with a description of the prisoners' dilemma game and were asked a series of questions (see SM1 for questions) designed to check their understanding of the game.If they got more than one of these questions wrong the first time, they would be allowed to try again, being informed that they would not be able to proceed with the full experiment should they fail a second time.
If they passed the check, participants would play one round of the prisoners' dilemma, with a random partner.They did not get feedback with regard to the outcome of the game.This was done to give them experience in playing the game before moving on to the partner choice task.
After playing the first round, participants were informed that they would be presented with animations of two potential partners ('Partner A' and 'Partner B') playing the game with another participant, with their task being to choose which potential partner they would rather play the game with (i.e., to choose partner A or B), under the belief that they would have the opportunity to play the game with the chosen partner at the end of the experiment for a bonus of up to 1 GBP (when actually, they never played the game at the end of the experiment, but did receive the full bonus).For one trial, participants would view back-to-back animations of each partner making their decision, with one partner planning their decision quickly (after 2-4 s) and one partner planning their decision slowly (after 14-16 s).Whether partner A or B planned with quickly or slowly was counterbalanced between trials.For each partner, the animation would first display loading dots representing the time it took for the partner to plan their decision, before displaying the choice that the partner had made.The participant would then be prompted to choose either Player A or Player B, by pressing 'a' or 'b' on their keyboard.See Fig. 1 for a graphical overview of the procedure and see SM for video example).
Participants would complete eight 'Cooperation trials' in which they chose between a 'fast cooperator' and a 'slow cooperator' and eight 'Defection trials' in which they chose between a 'fast defector' and a 'slow defector'.We also included eight filler trials in which participants would choose between a cooperator who responded either quickly or slowly and a defector who responded in the other way (this was done to make the aims of the task less salient).Trials were presented in random order.Participants were paid a total of 3GBP for participation.
2.1.1.4.Deception.All our Experiments employed deception.Because we used pre-recorded open-source movement data from a study conducted several years ago we could not contact the original participants from that study to include them as players in the current study.Deception also seemed necessary to us because we wanted to present participants with controlled movements that contain potential cues to mental states but without reflecting an explicit communicative intent (e. g., Sperber & Wilson, 1986) which may not have been possible if we had real partners giving responses in live interactions (in a partner choice setting) as participants may have used their actions to communicate to each other (unless we manipulated their beliefs about whether or not they were being watched, which also would necessitate deception).All participants were fully debriefed after the study and given the opportunity to withdraw their data or contact the experimenter with any concerns they may have.
2.1.1.5.Exclusions.Because we used an online sample, we preregistered several criteria to exclude participants who may have been inattentive throughout the experiment.Firstly, we excluded 18 participants who answered more than one of the four catch questions incorrectly (see SM1 for catch questions).Secondly, we excluded 2 participants who pressed the same button more than 85% of the time (for Experiment 1a, we initially pre-registered this threshold to be 75% but realized that this may have been too strict and result in the unfair exclusion of those with weaker preferences, thereby inflating our effects see SM4 for a robustness check with different exclusion criteria applied).On the trial level, we excluded 15 trials with response times above 30 s.

Experiment 1a: Results
We calculated the preference that each participant had for potential partners deciding quickly or slowly, with 0 representing a complete preference for fast decision-makers, and 1 representing a complete preference for slow decision-makers.A paired t-test comparing preference scores between cooperation trials and defection trials was significant (see Fig.

Experiment 1b -Partner preference
To examine the robustness of the effects in Experiment 1a, we designed Experiment 1b, wherein the variation in decision times was adjusted to be less pronounced compared to Experiment 1a.This modification was implemented to avoid potential demand effects that could arise from overtly noticeable changes suggestive of our underlying hypothesis.

Experiment 1b: Method
The pre-registration for this study can be found at: https://aspredict .Virtual partners would respond with one of three decision times.Fast partners were programmed to respond after 3.5-4.5 s, medium partners were programmed to respond after 7.5-8.5 s, and slow partners were programmed to respond after 11.5-12.5 s.

Procedure.
The procedure was similar to that of Experiment 1a, with participants watching potential partners making decisions in the prisoners' dilemma whilst ongoing decision time was animated, before expressing who they would like to interact with in the future.However, rather than being instructed to explicitly select between two potential partners, participants would first be presented with a potential partner deciding to cooperate or defect in a prisoners' dilemma whilst ongoing decision time was animated, before rating on a scale of 'Not at all' to 'Very much so', how much they would like to interact with this partner in the future.Participants were instructed that the partners that they would interact with at the end of the experiment would be based on these ratings.Participants would complete four cooperate trials and for defect trials for each of the three levels of the action cue (24 trials in total) presented in a random order.See Fig. 3 for a graphical overview of the procedure.

Design.
This Experiment employed a 2 × 3 within-subjects design.We manipulated the virtual partner's decision time, with three levels (fast, medium, slow) and the virtual partner's choice, i.e., whether they choose to cooperate or defect.Fig. 1.Graphical overview of one trial of our Experimental paradigm for Experiment 1a and 2a (not to scale).Firstly, participants view the payoff matrix (far left panel).Secondly participants watch partner A and then partner B (middle two panels) either planning (decision-time animated with loading dots; Exp.1a) or planning and executing (movements animated with a moving cursor; Exp.2a) a decision to cooperate or defect (decision is animated with a circle around the chosen option).Thirdly, participants choose whether they would like to interact with partner A or B at the end of the Experiment.Fig. 2. Point plots for Experiments 1a (action planning) and 2a (action execution).Y-axis shows participants' preference for fast/direct over slow/indirect potential partners (0 = a complete preference for those who planned decisions quickly or executed decisions directly; 1 = a complete preference for those who planned quickly or executed indirectly).Individual points represent one participant and error bars represent 95% within-subject confidence intervals.
2.2.1.5.Exclusions.Because we used an online sample, we preregistered several criteria to exclude participants who may have been inattentive throughout the Experiment.Out of the original 164 participants, we excluded 10 participants who answered more than one of the four catch questions incorrectly (see SM1 for catch questions).Secondly, we excluded 1 participant who responded within 5 points of the same value more than 75% of the time (i.e., clicked on the same location for the majority of trials as it was likely they were clicking without moving their mouse).This resulted in a final sample of 153 participants.On the trial level, we excluded 20 trials with response times above 30 s or below 500 ms.
To investigate the interaction, we carried out separate within subjects ANOVAs for cooperate and defect trials.For cooperate trials, a within subjects ANOVA revealed a main effect of decision time, F (2,304) = 47.14, p < .001,ηp2 = 0.24.Post-hoc t-tests (Bonferonni corrected for six comparisons) revealed that preference for cooperators decreased as a function of decision time, with fast cooperators being preferred over medium speed cooperators, t(152) = 6.17, p < .001,d = 0.5 and slow cooperators, t( 152

Experiment 1c: Does planning time influence the perception of confidence and/or deliberation?
To gain a more comprehensive understanding of the mechanisms underlying the effect of decision-time upon partner choice, Experiment 1c investigated the influence of decision time on the perception of an actor's level of deliberation and confidence.

Experiment 1c: Method
The pre-registration for this study can be found at: https://aspredict ed.org/1SJ_H6B  2.3.1.1.Participants.Taking the smallest effect size from Experiment 1b, a g*power analysis for a Bonferroni corrected (for three comparisons as we did not predict an interaction between movement directness and trial type) one sample t-test (d = 0.41, a = 0.017, 1β = 0.95) determined that we should collect 100 participants.Using the Prolific recruitment platform (www.prolific.co)we recruited 100 participants (5 removed from the analysis because they did not meet our ore-registered inclusion criteria), with a mean age of 31.38 (SD = 10.67), with 70 identifying as male, 28 identifying as female, and 2 identifying as other.
Same as Experiment 1a.
2.3.1.2.Procedure.This procedure was similar to that of Experiment 1b but with a few differences.Firstly, participants were not instructed that they would be playing the prisoners' dilemma with the observed actors, but just that they would rate the actors on several dimensions.Specifically, they would be presented with an actor choosing either to cooperate or defect before rating on a scale from Not at all to Very much either: 1) 'How confident was the actor in their decision?'or 2) 'How carefully did the actor think about their decision?'(we phrased the question this way to ask about deliberation, which is not a commonly used word thus may be misunderstood, in everyday terms).Participants would complete two cooperate trials and two defect trials for each of the three levels of the action cue for each question (24 trials in total) presented in a random order.See Fig. 3 for a graphical overview of the procedure.

Design.
This study employed a 2 × 3 within-subjects design and had two dependent variables.We manipulated the virtual partner's decision time, with three levels (fast, medium, slow) and virtual partner's choice, i.e., whether they choose to cooperate or defect.Our dependent variables were deliberation ratings and confidence ratings.
2.3.1.4.Exclusions.Because we used an online sample, we preregistered several criteria to exclude participants who may have been inattentive throughout the experiment.Out of the original 100 participants, we excluded 5 participants who answered more than one catch question incorrectly resulting in a final sample of 95 participants.On the trial level, we excluded 15 trials with response times above 30 s or below 500 ms.

Experiment 1c: Results
We carried out two separate ANOVAs to investigate the effects of choice and decision time upon deliberation (deliberation ratings were reverse coded so that 0 represents deliberation and 1 represents (non) deliberation) and confidence ratings.See Fig. 5 left panel.

Experiment 1: Summary
Experiment 1 investigated the effects of decision time as an action planning cue upon partner choice in the Prisoners' Dilemma.Overall, we demonstrate that participants prefer to interact with those who cooperate quickly compared to slowly, with this effect generalizing across two contexts: one in which participants are forced to choose between fast and slow actors and one in which participants can freely express how much they would like to interact with a particular actor (whilst we varied decision time).Moreover, a separate validation Experiment demonstrates that decision time may reflect (non)deliberation and confidence, thus may explain why participants prefer those who cooperate quickly over those who cooperate slowly.

Experiment 2: Movement directness
Experiment 2 aimed to test the effect of movement directness (and smoothness) upon partner choice in a prisoners' dilemma.Experiment 2a directly tested whether participants would choose to interact with those who executed decisions with direct (and smooth) movements over those who executed decisions with indirect (and jittery) movements.Experiment 2b varied both movement directness (direct, medium, indirect) and the partners' decision (i.e., to cooperate or defect), in order to test how both of these factors influence partner preference.Experiment 2c aimed to test whether directness (and smoothness) in the execution phase of an action signals confidence and/or deliberation about a decision.

Experiment 2a: Method
The pre-registration for this study can be found at: https://aspred icted.org/ec9cf.pdf3.1.1.1.Participants.Using the same power calculation as in Experiment 1a, we aimed for a final sample of 72 participants after exclusions.In total we recruited 93 participants (21 excluded), 59 of which identified as male, and 34 of which identified as female.The sample had a mean age of 27.93 (SD = 7.95).
3.1.1.2.2.Virtual partner response screen:.Like in Experiment 1a, the response options ('Cooperate' and 'Defect') were displayed in the top left and right (counterbalanced across participants) and the decision-making phase of the potential partner's response was represented by loading dots in the center of the screen.However, this time the decision-making time was fixed and followed by an animation of the action execution phase of the potential partner's response, with a white dot (x = 0.05, y = 0.05) representing the potential partner's mouse movements towards one of the response options.After the decision was implemented (i.e., the dot had stopped on one of the response options) a white ellipse around the choice was also displayed.

Virtual partner behavior.
With respect to the potential partner's movements, our two kinematic parameters of interest were maximum deviation (in pixels) from the most direct path to the chosen option (as an index of directness towards the chosen option, thus relative attraction to the other option) and jitter.Potential partners were programmed either to move towards the response option with a trajectory that was direct (i.e., low max deviation from the most direct path and thus little attraction to the other option) and smooth (i.e., no jitter) or to move towards the response option with an indirect (i.e., high max deviation from the most direct path, thus substantial attraction to the other option) and jittery trajectory.
To create these trajectories, we took the open-source movement data from Kieslich & Hilbig, 2014 (http://journal.sjdm.org/14/14808/data.csv) which contained movement data from participants responding to mouse tracked versions of economic games (i.e., prisoner's dilemma, stag-hunt, and chicken) in which the response options were displayed in the top corners of the screen.For each trial (1726 in total) there was a movement trajectory (x and y coordinates) that was normalized to have 101 even time steps (frames).To isolate trajectories that were suitable for animations of direct and indirect movements we took several processing steps to ensure strict experimental control of the cues present in the movements.
Firstly, to ensure that our trajectories varied only with regards to jitter and max deviation, we controlled for several other factors.We removed any long pauses from the trajectories.We removed movements that hovered around the home button (i.e., movements that hadn't travelled 20% of the path by frame 50) as in these trials participants likely made their decision before starting to move, meaning that their decision process never leaked into the kinematics (i.e., against the instructions in Kieslich & Hilbig, 2014).We removed movements that hovered around an option without making a choice (i.e., movements that had travelled 80% of the path by frame 50), as this may act as an unwanted signal of uncertainty.We removed long pauses during a movement (i.e., where movements were around zero velocity for more than 15 frames) as this also may act as an unwanted signal of uncertainty (e.g., reading or equipment related issues).We removed movements with long paths (with a maximum deviation of more than 1000 pixels, or a total path length of more than 4000 pixels), as these movements may have represented participants choosing one option before having a change of heart and moving towards another option as this may have also reflected participants unintentionally moving towards the wrong option before correcting their mistake.We removed movements that contained excessive direction changes on either the x or y axes (>2SD of the mean) as this may have reflected indecisiveness.
Secondly, we quantified maximum deviation (maximum distance in pixels from the most direct path towards the response option) and jitter (RMSE of smoothed velocity profile subtracted from unsmoothed velocity profile), before splitting the data into quartiles based on both parameters (we log transformed both parameters to avoid a skewed final sample).
Thirdly, we took the trajectories that fell in the first quartile for both max deviation and jitter as our pool of 'direct trajectories' and those that fell in the fourth quartile for both maximum deviation and jitter as our pool or 'indirect trajectories'.For each pool, we took a random sample of 16 trajectories, that fell within 1.5SD with respect to the mean max deviation and 1SD with respect to the mean jitter of that pool.See Fig. 6 for trajectories and kinematic parameters.Finally, we randomly paired the trajectories from each pool, generating a list of 16 pairs of movement trajectories (one direct and one indirect trajectory per pair), ensuring that the differences between the pairs with respect to max deviation and jitter were not correlated (i.e., the difference in jitter between each trajectory in a pair did not increase as a function of the difference in max deviation between each trajectory in a pair).This trial list was fixed throughout the experiment, so that all participants viewed the same 16 pairs of movements..1.1.3. Procedure.This was the same as in Experiment 1a, except for the stimuli, where the decision-making time was held constant for both partners within a trial (i.e., decision time was a random value between 7 and 9 s and was always the same for both partners) and animations of the virtual partners' movements towards the response options were added (see SM for video example).

Exclusions.
We recruited 93 participants in total.We excluded 21 participants who answered more than one catch question incorrectly but no participants who pressed the same button more than 85% of the L. McEllin et al. time.On the trial level, we excluded 11 trials with response times over 30 s.

Experiment 2b: Partner preference
To examine the robustness of the effects in Experiment 2a, we designed Experiment 2b, wherein the variation in movement directness (and smoothness) was adjusted to be less pronounced compared to Experiment 2a.This modification was implemented to avoid potential demand effects that could arise from overtly noticeable changes suggestive of our underlying hypothesis.

Experiment 2b: Method
The pre-registration for this study can be found at: https://aspredict ed.org/LTQ_5WH 3.2.1.1.Participants.Using the same power calculation as in Experiment 1b, we aimed to collect 164 participants.Using the Prolific recruitment platform (www.prolific.co)we recruited 181 participants (7 removed from the analysis because they did not meet our pre-registered inclusion criteria and 17 removed due to a Prolific errorsee exclusion section), with a mean age of 29.87 (SD = 9.83), with 99 identifying as male, 82 identifying as female, and 0 identifying as other.Virtual partners would respond with one of three kinds of Movement trajectory (Direct, Medium, Indirect).To create these trajectories, we used the same open-source movement data from Kieslich and Hilbig (2014) as in Experiment 2a, applied the same pre-processing steps as before (i.e., removing trials with excessive hovering, direction changes or path length and removing long pauses from the movements), and quantified movement directness using the same parameters as before (i.e., Maximum Deviation from the optimal trajectory and movement jitter).However, our method of sampling trajectories based on movement directness differed from that of Experiment 2a.We first split the data into eight quantiles for MD and jitter, before sampling eight movements that fell in the first, fifth and seventh quantile with respect to both of these parameters (as choosing these three quantiles resulted in approximately equal distances between the three categories with respect to our kinematic parameters whilst also  allowing us to remove extreme values from the eight quantile, see Fig. 7 for trajectories).We sampled eight movements that fell within 2 SD of the mean values for each of these quantiles, resulting in eight direct movements, eight medium movements and eight indirect movements.

Procedure.
The procedure was the same as Experiment 1b, except we kept decision-time constant (as in Experiment 2a) and manipulated movement directness.
3.2.1.4.Design.This Experiment employed a 2 × 3 within-subjects design.We manipulated the virtual partner's movement directness, with three levels (direct, medium, indirect) and the virtual partner's choice, i.e., whether they choose to cooperate or defect.

Exclusions.
For Experiment 2b we recruited 181 participants in total.Firstly, we had to recruit an extra 17 participants because an error with Prolific's exclusion function, meant that 17 of our participants had also participated in Experiment 1b thus were not naïve to our research question.Out of the 164 participants who had only participated in Experiment 2b, we excluded 7 participants who answered more than one catch question incorrectly and 0 participants who pressed the same point on the scale more than 75% of the time.On the trial level, we excluded 17 trials with response times above 30 s or below 500 ms.
For defect trials, a within subjects ANOVA revealed no main effect of decision time, F(2,312) = 0.2, p = .1,ηp2 = 0.001.Post-hoc t-tests (Bonferroni corrected for six comparisons) revealed that preference for defectors did not depend on movement directness, with no difference between direct cooperators and medium directness cooperators, t( 156

Experiment 2c: Deliberation and confidence
To gain a more comprehensive understanding of the mechanisms underlying the effect of movement directness upon partner choice, Experiment 2c investigated the influence of movement directness on the perception of an actor's level of deliberation and confidence.

Experiment 2c: Method
The pre-registration for this study can be found at: https://aspredict ed.org/HZR_WT2 3.4.1.1.Participants.Using the same power calculation as in Experiment 1c, we aimed to collect 100 participants.Using the Prolific recruitment platform (www.prolific.co)we recruited 100 participants (4 removed from the analysis because they did not meet our ore-registered inclusion criteria), with a mean age of 31.93 (SD = 10.05), with 52 identifying as male, 47 identifying as female, and 1 identifying as other.3.4.1.4.Design.This employed a 2 × 3 within-subjects design with two dependent variables.We had two different questions (deliberation, confidence) We manipulated the virtual partner's movement directness, with three levels (direct, medium, indirect and virtual partner's choice, i.e., whether they choose to cooperate or defect.Our dependent variables were deliberation ratings and confidence ratings.

Apparatus and
3.4.1.5.Exclusions.For Experiment 2c we recruited 100 participants in total.We excluded 4 participants who answered more than one catch question incorrectly.On the trial level, we excluded 10 trials with response times above 30 s or below 500 ms.We had to remove one participant from our analysis of the confidence questions as multiple trials from the same level of the design met our exclusion criteria (i.e., meaning for that participant we did not have at least one response for every level of the design).

Experiment 2c: Results
We carried out two separate ANOVAs to investigate the effects of choice and movement directness upon deliberation (deliberation ratings were reverse coded so that 0 represents deliberation and 1 represents (non)deliberation) and confidence ratings.See Fig. 5 right panels.

Experiment 2: Summary
Experiment 2 investigated the effects of movement directness (and smoothness) as an action execution cue upon partner choice in the Prisoners' Dilemma.We demonstrate that participants prefer to interact with those who implement cooperative decisions with direct and smooth movements compared to those who implement cooperative decisions with indirect and jittery movements.We generalize this effect across two contexts: one in which participants are forced to choose between actors with direct (and smooth) movements and actors with indirect (and jittery) movements, and one in which participants can freely express how much they would like to interact with a particular actor (whilst we varied movement directness and smoothness).Moreover, a separate validation Experiment demonstrates that movement directness (and smoothness) may act as a signal of confidence (and not deliberation), thus may explain why participants prefer those who cooperate with direct movements over those who cooperate with indirect movements.

General discussion
This study investigated whether people select potential partners in an economic interaction based on cues present in the planning of the partners' previous prosocial or selfish decisions, and in the actions used to execute these decisions.In Experiment 1a and 2a, participants chose either between two partners who cooperated or two partners who defected in the Prisoners' Dilemma (see SM3 for a generalization of this study to the Stag Hunt), and in Experiment 1b and 2b participants freely expressed their preferences for interacting with a potential partner in the future.The partners' responses contained action planning cues., i.e., decision time (Experiment 1), or action execution cues,.i.e., movement directness (Experiment 2).Finally, in line with previous literature around intuitive cooperation (e.g., Hoffman et al., 2015;Rand et al., 2012) and confidence in action execution (e.g., Patel et al., 2012;Slepian et al., 2013), participants in Experiment 1c and 2c were able to judge an actor's level of deliberation and confidence from cues present in their action planning and execution.
Confirming our key predictions, participants exhibited strong preferences for partners who planned cooperative decisions with quickly planned or directly executed actions compared to partners with slowly planned or indirectly executed actions This was the case both when being forced to choose between partners (i.e., Experiment 1a and 2a) and when freely expressing their preference for partners (i.e., Experiment 1b and 2b).Moreover, confirming our assumptions that action cues carry informative signals about decision making, we found that action planning cues carry signals to both confidence and (non)deliberation (i.e., decision speed scales to ratings of confidence and deliberation) and action execution cues carry signals to confidence (i.e., movement directness scales to ratings of confidence) but not deliberation.Taken together, the above demonstrates that individuals estimate confidence and deliberation behind a cooperative action from the way in which this action was planned and executed, using these estimates to inform their partner choice, preferring those whose cooperative actions reflect confidence (in both planning and execution) and (non)deliberation (in planning only) rather than uncertainty and deliberation.
Why would people prefer actors whose cooperative actions signal confidence and (non)deliberation?Action cues to confident and (non) deliberative cooperation may signal that a potential partner is likely to reliably cooperate in the future, whereas action cues to deliberative cooperation and uncertainty may signal that a potential partner is likely to cooperate conditionally.This is in line with the finding that people are more cooperative towards those who cooperate without considering the costs and benefits of the interaction (Hoffman et al., 2015;Jordan et al., 2016).However, although we found that individuals have a strong preference for intuitive or confident cooperators, considering that in many cases intuitive responders can also have a strong penchant for defection, or that many unconditional cooperators show patterns of 'fast-deliberation' (Jagau & van Veelen, 2017;Montero-Porras, Lenaerts, Gallotti, & Grujic, 2022) it may be interesting to explore under what conditions individuals prefer to interact with deliberative or unconfident cooperators.
For defection, our findings were mixed.When being allowed to freely express their preferences for both cooperators and defectors, participants showed a general aversion to defectors, and did not discriminate based on action cues.However, when being forced to choose between two defectors, participants preferred those who defected either deliberatively or with uncertainty.Thus, in situations where interacting with defectors cannot be avoided, action cues may still be informative, with those who defect confidently and/or intuitively being perceived as likely to behave similarly in the future, compared to those who defect uncertainly or deliberatively and whose behavior may change under the right circumstances.
Overall, we demonstrate that in addition to cues related to interaction history (Barclay & Willer, 2007;Hardy & Van Vugt, 2006) or traits (Balliet et al., 2018;Bonnefon et al., 2017) the ways in which people plan and implement prosocial or selfish choices carry important cues with respect to partner choice.Much as action parameters render instrumental or social intentions visible and thereby facilitate effective coordination in a joint action (Cavallo et al., 2016;McEllin, Sebanz, & Knoblich, 2018;Sebanz, Bekkering, & Knoblich, 2006), action parameters may render hidden mental states associated with decision-making visible (i.e., confidence and deliberation), thus facilitating effective cooperation in economic interactions.This points to the fact that much of human's cooperative success, beyond action coordination (Sebanz et al., 2006), can be attributed to the perception-action coupling that allows us to understand each other's actions as they unfold.In addition to allowing us to predict and understand the motor and social intentions behind an action (Jacob & Jeannerod, 2005;McEllin et al., 2018), perception-action coupling allows us to observe a partner's decision process as it unfolds, providing us with additional information about aspects of their decision-making that are not observable in the outcome of their actions (i.e., interaction history).Such an ability may have been crucial for cultivating and maintaining mutually beneficial partnerships with those who are likely to continue cooperating, whilst avoiding those who may act selfishly and free ride.Moreover, these additional cooperative (and competitive) advantages may have been one of the contributing factors to the evolutionary selection of the mechanisms that support action understanding through perception-action coupling.

Future directions
In the current study we investigated how action cues inform whether one should engage with a partner, but what about cooperative behavior once engaged in an interaction?Could cues from a partners' cooperative or selfish actions when engaged in an economic interaction also help us decide how to best interact with that partner?For instance, interaction history may tell an actor that their partner is cooperative, but that partner may start to execute cooperative actions deliberatively or with uncertainty rather than intuitively or with confidence (e.g., their decision-time may suddenly begin to increase).This may provide the actor with a useful cue that their partner is considering defection, allowing the actor to deal with this issue proactively, for instance by acting conservatively to avoid being exploited or punish and curtail transgressions early.Future research should investigate the influence of action cues on cooperative behavior in a repeated interaction, using for instance a repeated Prisoners' Dilemma.
Our results also revealed a graded relationship between action cues and partner preference (and ratings of deliberation and confidence)that is -the faster an action is planned or the more directly an action is executed, the stronger the preference for that partner, perhaps because the magnitude of the action cue signals the extent of an actor's confidence or deliberation.However, because we wanted to ensure that participants were exposed to all levels of our design (i.e., a complete within-groups design), and because our primary question pertained to whether or not action cues inform partner choice rather than measuring the extent to which these cues inform partner choice, we manipulated our action cues as a factor (i.e., with two or three levels) rather than continuously.Future studies could try and map the specific relationship between action cues and partner choice (or perceived confidence and deliberation).
Looking beyond economic games, are the effects of action cues upon decision making specific to economic interactions?Considering that uncertainty with regards to perceptual or moral decision-making may be observed in decision-times (Greene, 2007;Vickers, 1970) and movement parameters (Patel et al., 2012;Spivey, Grosjean, & Knoblich, 2005) we doubt this is the case.Future research should aim to investigate how such cues help us coordinate everyday decisions (e.g., what to have for dinner) by understanding an actor's general preferences, or how these cues can help us to coordinate larger decisions (e.g., who to vote for) by understanding an actor's moral preferences.

Potential limitations
Would our findings generalize to non-western, and particularly nonindustrialized societies (Henrich, Heine, & Norenzayan, 2010)?On one hand there is evidence that the tendency to make inferences about an actor's mental state can be observed across multiple small-and largescale non-western societies (Barrett, Todd, Miller, & Blythe, 2005) hinting that the ability to make inferences about social preferences using action cues may be universal.On the other hand, there are many dimensions on which cultures (or individuals for that matter) differcultural differences in phenomena such as context sensitivity (Masuda & Nisbett, 2001), risk perception, (Weber & Hsee, 1998) or intuitive cooperation (Nishi, Christakis, & Rand, 2017) may all influence the effect of action cues upon partner choice, thus warranting further investigation.More specifically, cultural differences in the extent to which individuals reason dialectically means that many cultures may tolerate or even value conflictedness in decision-making (Hamamura, Heine, & Paulhus, 2008;Peng & Nisbett, 1999), which leads to interesting predictions for future cross-cultural research.For instance, individuals from cultures who value dialectical thinking may view individuals who exhibit conflict as more trustworthy or at least be less averse to these individuals, and may view individuals who are highly confident or nondeliberative as overly impulsive or even foolish.

Conclusion
The current study demonstrates that action cues present in either the planning (i.e., decision time) or execution (i.e., movement directness) of economic choices influence partner choice by reflecting how confident or conflicted a potential partner is when making cooperative decisions.This bears on our understanding of: a) human economic behavior; b) human decision-making and decision-coordination in general; and c) the utility (both in our evolutionary past and today) of a coupling between perception and action that allows us to recycle the representations within our own motor repertoire for the purposes of action understanding.

Declaration of Competing Interest
We have no conflicts of interest to disclose.

Fig. 3 .
Fig. 3. Graphical overview of one trial of our Experimental paradigm for Experiment 1b-c and 2b-c (not to scale).Firstly, participants view the payoff matrix (far left panel).Secondly participants watch the potential partner (middle panel) either planning (decision-time animated with loading dots; Exp.1b-c) or planning and executing (movements animated with a moving cursor; Exp.2b-c) a decision to cooperate or defect (decision is animated with a circle around the chosen option).Thirdly, participants either express how much they would like to interact with that partner at the end of the Experiment (Experiment 1b & 2b) or rate how confident/deliberating that partner was about their choice (Experiment 1c & 2c).

Fig. 4 .
Fig. 4. Point plots for Experiments 1b (action planning) and 2b (action execution).X-axis shows magnitude of the action cue, Y-axis shows mean preference (0 = 'Not at all' and 1 = 'Very much'), and colors index decision type (blue = cooperate trials, red = defect trials).Individual points represent one participant and error bars represent 95% within-subject confidence intervals.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) ) = 12.85, p < .001,d = 1.13 and slow responses, t(94) = 13.76,p < .001,d = 1.71, and medium responses yielding higher confidence ratings than slow responses t(94) = 7.34, p < .001,d = 0.58.

Fig. 6 .
Fig. 6.Topplots of raw kinematics of selected trajectories for Experiment 2a.Bottomhistograms showing distribution of kinematic parameter values across direct and indirect trials for Experiment 2a.

Fig. 7 .
Fig. 7. Top-plots of raw kinematics of selected trajectories for Experiment 2bc.Bottom-histograms showing distribution of kinematic parameter values across direct, middle and indirect trials for Experiment 2b-c.