Eliciting social approach-avoidance conflict within a novel experimental paradigm: psychometric and computational evidence with a successful pre-registered replication

Abstract

Affective facial expressions elicit approach-avoidance motivational responses that shape social behavior. Qualitatively, individuals report frequently experiencing competing motivations to approach and avoid other individuals in social contexts (i.e. social approach-avoidance conflict; AAC). Moreover, theoretical frameworks propose that successful navigation of social AAC plays a critical role in adaptive social behavior. However, despite an extensive array of well-validated, non-social AAC paradigms, no research has developed a paradigm that experimentally elicits and reliably quantifies social AAC in humans. To address this issue, we developed and validated a novel social AAC (SAAC) paradigm with an independent replication across two samples. In the SAAC paradigm, morphed facial expressions are used to parametrically modulate the intensity of social reward (happiness), social threat (anger), or social reward-threat conflict (co-occurring happiness and anger). Demonstrating robust AAC effects, social reward-threat conflict uniquely elicited more intermediate approach-avoidance choice selection and slower reaction times compared to social reward and social threat. Furthermore, computational drift diffusion models demonstrated that social AAC was driven by noisier evidence accumulation processes. Together, these findings demonstrated and replicated that our novel SAAC paradigm reliably elicits social AAC, which may provide a more mechanistic understanding of social behavior and its dysregulation in psychopathology.

Introduction

Affective facial expressions convey powerful social signals that activate motivations to approach or avoid other individuals, which play an important role in guiding social behavior (Ambadar et al., 2005; Barrett et al., 2019; Frith, 2009). For example, happy facial expressions serve as social reward signals of social affiliation that typically activate approach motivation, whereas angry facial expressions serve as social threat signals of social exclusion that typically activate avoidance motivation (Chen & Jack, 2017; Nikitin & Freund, 2019; Radke et al., 2018; Stins et al., 2011; Tamir & Hughes, 2018). These types of motivational responses to facial expressions are typically measured using experimental paradigms such as the approach-avoidance task (Heuer et al., 2007). Notably, however, approach motivation and avoidance motivation do not guide social behavior in isolation. Instead, individuals describe frequently experiencing co-occurring motivations to approach and avoid other individuals in social situations (Barker et al., 2019). Moreover, successfully navigating conflict between approach and avoidance motivation in social settings is proposed to play a critical role in adaptively modulating social behavior (for a review, see Barker et al., 2019; Strack & Deutsch, 2004). Therefore, it is important to develop experimental paradigms that can reliably elicit and measure competing social motivations to approach and avoid other individuals.

Historically, social motivation paradigms have largely examined approach-avoidance motivational responses to intense affective facial expressions (Phaf et al., 2014), but these types of facial expressions rarely occur in social contexts (Carroll & Russell, 1997; Matsumoto & Hwang, 2014). Rather, social contexts are largely characterized by facial expressions that display more subtle emotional cues (e.g. a gentle smile) or an admixture of conflicting emotional cues (e.g. a smug smile), which can be reliably and efficiently detected by others (Calvo et al., 2014; Scherer & Ceschi, 2000; Scherer & Ellgring, 2007). For example, facial expressions in social contexts may be ambiguous due to containing varying degrees of social reward intensity (e.g. 50%_Happy) or varying degrees of social threat intensity (e.g. 50%_Angry; Barrett et al., 2019; Matsumoto & Hwang, 2014). Additionally, facial expressions can convey varying degrees of simultaneously co-occurring social reward and social threat (e.g. 50%_Happy + 50%_Angry; Gutiérrez-García & Calvo, 2014, 2016), which are compound facial expressions facilitated by partial independence of musculature controlling the eye and mouth regions of the face (Du et al., 2014; Du & Martinez, 2015). Therefore, it is important to better understand how approach-avoidance motivational responses are modulated in response to varying degrees of social reward, social threat, and social reward-threat conflict.

To address this issue, our past research developed morphed facial expressions that parametrically vary in social reward, social threat, and social reward-threat conflict (Evans et al., 2022; Evans & Britton, 2020). Using these morphed facial expressions, our previous research systematically assessed and replicated the sensitivity of approach and avoidance motivational responses to varying degrees of social signal intensity (Evans et al., 2024). Specifically, individuals exhibit linear increases in approach or avoidance motivation as a linear function of social reward or social threat intensity, respectively. When social reward and social threat vary in co-occurring social reward and social threat (i.e. social reward-threat conflict), individuals exhibited both linear increases in approach motivation and linear decreases in avoidance motivation. In these studies, however, approach motivation and avoid motivation were assessed separately such that approach and avoidance were not competing options. Therefore, it remains an empirical question if social reward-threat conflict facial expressions elicit approach-avoidance conflict (AAC). As noted previously, this is an important issue to address given that AAC occurs frequently in social situations and successful navigation of AAC putatively plays a unique role in social behavior (for reviews, see Barker et al., 2019; Strack & Deutsch, 2004).

Broadly, AAC is a distinct psychological state that occurs when opportunities for reward and threat simultaneously co-occur to activate competing approach and avoidance motivations (Corr, 2013; Lewin, 1935; Miller, 1944). For example, individuals may experience AAC when approaching a reward at the potential cost of incurring punishment or when avoiding a threat at the potential cost of losing a reward. As a result, co-occurring reward and threat presents a more complex gain-loss function that introduces substantial cognitive demands as evidenced by more intermediate approach-avoidance choice proportions, longer decision times, and greater risk aversion (Aupperle et al., 2011; Bublatzky et al., 2017; Diederich, 2003; Garcia-Guerrero et al., 2023; Schlund et al., 2016). Importantly, there are multiple decision-making processes that can contribute to AAC effects such as (1) slower accumulation of evidence, (2) speed-accuracy tradeoff, (3) competing prepotent biases to approach and avoid, and (4) greater duration of non-decision processes such as motor response preparation. As a result, AAC research has recently started to leverage computational approaches such as drift diffusion models (DDMs), which use sequential sampling methods to derive latent parameters that aim to disentangle the aforementioned decision-making processes (Pedersen et al., 2021).

Despite these advances, however, experimental AAC paradigms in humans have exclusively relied on non-social rewards and threats, which as an important gap in the literature. Non-social AAC paradigms typically present individuals with competing opportunities for money gain or loss, exposure to positive or aversive stimuli, or some combination of these non-social contingencies (Kirlic et al., 2017). In contrast, research using animal models has demonstrated that it is important to measure social AAC more specifically, which involves competing opportunities for social rewards and social punishments that produce divergent effects compared to non-social rewards and punishments (for reviews, see Atrooz et al., 2021; Gencturk & Unal, 2024). Similarly, in humans, social rewards and social threats are differentially processed compared to non-social rewards and threats as evidenced by divergent effects on behavior (Barclay & Benard, 2020; Barker et al., 2019; LoBue & Pérez-Edgar, 2014; McDermott & Egwuatu, 2019; Rademacher et al., 2010; Ruff & Fehr, 2014). Further underscoring the importance of this gap in the literature, AAC is a frequent occurrence in social contexts and navigating social AAC is putatively proposed to play a particularly important role in adaptive social behavior (Elliot et al., 2006; Kashdan et al., 2008). In contrast to animal model research, however, there are currently no experimental paradigms that reliably elicit social AAC in humans, which may provide unique mechanistic insights into social behavior.

To address this issue, the current study aimed to develop a novel social approach-avoidance conflict (SAAC) experimental paradigm. Based on previous research, we used validated morphed facial expressions that parametrically vary in social reward, social threat, or social reward-threat conflict (see Figure 1; Evans & Britton, 2020; Evans et al., 2024). To measure social AAC, responses to social reward-threat conflict (50%_Happy + 50%_Angry) can be compared to unambiguous social reward (100%_Happy) and unambiguous social threat (100%_Angry), which directly mirrors previous non-social AAC research. Specifically, we hypothesised that social reward-threat conflict faces would elicit comparatively intermediate approach-avoidance choice selections as defined by fewer approach choices compared to unambiguous social reward faces as well as fewer avoid choices compared to unambiguous social threat faces. Similarly, we hypothesised that approach-avoidance choice selection in response to social reward-threat conflict faces would elicit comparatively longer reaction times. Mirroring non-social AAC research (Chu et al., 2023), we further dissociated AAC effects from ambiguity effects by comparing social reward-threat conflict (50%_Happy + 50%_Angry) with ambiguous social reward (50%_Happy) and ambiguous social threat (50%_Angry) signals.

Figure 1.

Parametric modulation of social signals with morphed facial expressions. Legend: Morphed facial expressions were created with facial landmarking and linear interpolation. Neutral, happy, and angry facial expressions were linearly interpolated in 50% increments to parametrically modulate social reward intensity (0%_Happy, 50%_Happy, and 100%_Happy; Top Row), social threat intensity (0%_Angry, 50%_Angry, and 100%_Angry; Middle Row), and social reward-threat conflict intensity (100%_Happy, 50%_Happy + 50%_Angry, and 100%_Angry; Bottom Row). Note: These three actors were not presented in the current study. Instead, the current study exclusively presented facial expressions from twelve Caucasian actors. However, these twelve actors cannot be published per NimStim publishing requirements (Stimulus IDs: 05F, 06F, 07F, 08F, 09F, 10F, 20M, 24M, 30M, 34M, 25M, 32M). Therefore, we present these actors exclusively to depict the orthogonality of face identity across morph types.

In line with recent non-social AAC research (Pedersen et al., 2021), we also implemented drift diffusion modelling (DDM) to disentangle the decision-making processes underlying social AAC effects. Prior to conducting these studies, we did not have a priori hypotheses regarding which specific latent decision-making process contributed to social AAC. Instead, we conducted exploratory DDM analyses in Study 1 to guide pre-registered predictions for Study 2 (https://osf.io/rcjxg). In our pre-registration of Study 2, we predicted that social AAC would be primiarily characterised by smaller drift rates relative to both unambiguous as well as ambiguous social reward and social threat. Additionally, based on the exploratory findings of Study 1, we predicted that social AAC would be characterised by a smaller starting-point bias compared to unambiguous social reward and a smaller decision threshold compared to unambiguous social threat. To establish measurement reliability for conducting individual differences research, we characterised the internal consistency of task effects across both traditional and DDM measures in both studies.

Methods

Participants

Study 1

Participants were 48 undergraduate students recruited from the University of Miami (77.10% Female, 22.90% Male; Age: M = 18.88 years, SD = 1.28; 81.30% Caucasian, 6.30% Black, 8.30% Asian, 4.20% Other/Multi-Racial). We based this sample size on our previous work validating morphed facial expressions (Evans et al., 2024). All participants received course credit.

Study 2 (pre-registered replication)

Participants were 54 community adults from the larger Boston area who were recruited using flyers, word-of-mouth referrals, and online forums (48.10% Female, 46.30% Male, 1.90% Non-Binary, 3.70% Other; Age: M = 22.49 years, SD = 3.45, 31.50% Caucasian, 9.30% Black, 48.10% Asian, 10.20% Other/Multi-Racial). We based this sample size on Study 1 for replication purposes (https://osf.io/rcjxg). All participants were compensated $15/hour.

Stimuli and task

Morphed facial expressions

As in our previous work, morphed facial expressions were created using Morpheus software. Specifically, matching landmarks on facial features were placed on emotional expressions from the same actor, which were used to linearly interpolate these two facial expressions in 50% increments. For example, a landmarked neutral facial expression and a landmarked happy facial expression from the same actor were linearly interpolated to create varying intensities of social reward (i.e. happiness). Using this process, we linearly interpolated pairs of happy, angry, and neutral facial expressions to systematically vary the intensity of social reward-threat conflict signals (i.e. 100%_Happy + 0%_Angry, 50%_Happy + 50%_Angry, and 0%_Happy + 100%_Angry), social reward (i.e. 0%_Happy, 50%_Happy, and 100%_Happy), social threat (i.e. 0%_Angry, 50%_Angry, and 100%_Angry; see Figure 1). As in our previous work, we generated morphed facial expressions for six male actors and six female actors from the NimStim stimulus set (Tottenham et al., 2009).¹

Social approach-avoidance conflict (SAAC) task

The SAAC task was programmed and administered in Eprime 2.0 software. In the SAAC task, a facial expression was presented at the center of the screen with two options (approach/avoid) presented below the facial expression (see Figure 2). On each trial, participants are instructed to choose between approaching or avoiding the presented person as if they encountered the person in a social situation (e.g. at a party). For approach-avoidance choices, participants used the left and right arrow keys on the inset laptop keyboard. Button assignment (left vs. right) for approach and avoidance choices was counterbalanced across participants. For each trial, participants were provided with a 2000ms response window.

Figure 2.

Social approach-avoidance conflict task and drift diffusion models. Legend: In the social approach-avoidance conflict task, participants choose between socially approaching or avoiding a given individual. Drift diffusion models (DDMs) use sequential sampling methods to decompose approach-avoidance choices and reaction time (RT) distributions into latent parameters of underlying decisional processes. The latent drift rate parameter (v) indexes the direction (towards approach or towards avoid) and the rate (larger drift = faster; smaller drift = slower) of evidence accumulation. The latent decision threshold parameter (a) indexes the amount of evidence necessary to choose between approach and avoidance options. The starting point bias parameter (z) indexes the starting point if evidence accumulation is preliminarily biased towards approach or avoidance options. The non-decision time parameter (t) indexes the time required to encode visual information and execute a motor response.

Different actor stimuli were used for the social reward, social threat, and social reward-threat conflict conditions to ensure orthogonality between face identity and face emotion (see Figure 1). Each of the nine face morphs was presented 64 times for a total of 576 trials. Task trials were separated by a jittered inter-trial interval with a mean of 500 ms as well as 96 randomly presented null trials (blank screen) to reduce anticipatory effects. To minimise fatigue, the task was presented in four 7-minute blocks with a self-directed break between each block for a total task duration of approximately 28 min.

Data reduction

First, trials in which participants failed to make a response were excluded from all analyses. Non-response rates were low overall (Study 1: M = 2.56%; Range = 0.00% – 11.00%; Study 2: M = 1.39%; Range = 0.00% – 8.00%) and did not consistently differ as a function of variations in social reward-threat conflict, social reward, or social threat (see Supplemental Material). Second, trials with reaction times (RTs) smaller than 200 ms were also excluded in line with common procedures in DDM research (Ratcliff & Tuerlinckx, 2002). In Study 1, we used an a priori exclusion criteria of non-response rates greater than 20% for participant exclusion. In Study 2, we pre-registered the same exclusion criteria (https://osf.io/rcjxg). Based on this exclusion criteria, no participants were excluded in either Study 1 or Study 2.

Drift diffusion model

Broadly, a drift diffusion model (DDM) framework assumes that decision-making in forced-choice tasks is supported by evidence accumulation in favour of one choice until a decisional boundary is crossed and the favoured choice is subsequently selected (for a review, see Myers et al., 2022). As part of the DDM framework, choice selections and RT distributions can be used to model underlying cognitive parameters of (1) drift rate (v), (2) decision threshold (a), (3) starting point bias (z), and (4) non-decision time (t; see Figure 2). First, the drift rate parameter (v) indexes both the choice direction (e.g. avoid = negative drift; approach = positive drift) and rate of evidence accumulation (larger drift = faster; smaller drift = slower). Second, the decision threshold parameter (a) indexes the amount of evidence required for accumulation prior to selecting a choice (larger threshold = more evidence required; smaller threshold = less evidence required). Third, the starting point bias parameter (z) indexes the starting point at which evidence begins to accumulate in favour of approach or avoid choices (e.g. > 0.5 = avoid bias; ~0.5 = no bias, < 0.5 approach bias). Finally, the non-decision time parameter (t) indexes the time required to encode visual information and execute a motor response. Thus, DDMs use sequential sampling methods to identify the degree to which combination of variations in these four latent parameters can optimally account for the distribution of observed data.

We conducted DDMs using the Hierarchical Sequential Sampling Modelling (HSSM) Python toolbox (Wiecki et al., 2013). In brief, the HSSM toolbox utilises hierarchical Bayesian estimation to flexibly estimate DDM parameters based on choice selections and RT distributions. In the current study, we implemented a non-hierarchical approach in which separate DDMs were fit for each participant and level of social signal intensity given that hierarchical approaches did not achieve convergence. DDM models were run in four chains that contained 50,000 samples each with the first 5,000 samples discarded as burn-in. To prevent initial sampling boundary violations, we initialised the non-decision time parameter to be 0.05. Otherwise, we used the default HSSM settings, which implement moderately informative priors based on previous cognitive research studies (Matzke & Wagenmakers, 2009; Wiecki et al., 2013).

To assess convergence of parameter estimates generated by the four separate model chain, we implemented the widely-used Gelman-Rubin statistic (Gelman & Rubin, 1992). For this statistic, an R̂ value of 1.00 is thought to indicate perfect convergence of parameter estimates across DDM chains, whereas R̂ values greater than 1.10 may indicate problematic convergence (Pedersen et al., 2021). In both Study 1 and Study 2, R̂ values were less than 1.01 for all parameter estimates.

To validate the predictive utility of DDMs, we employed a qualitative and a quantitative approach, which both aim to assess the fit between observed distributions (i.e. data points directly measured) and predicted distributions (i.e. data points generated by DDMs). Qualitatively, we performed visual inspection of overlap between observed and predicted distributions across a randomly selected subset of participants and task conditions due to the large amount of data generated (Myers et al., 2022), which overall yielded strong overlap in cases that were inspected. All observed and predicted distribution plots from the current study are publicly available (https://osf.io/rcjxg). Quantitatively, we tested the correlations between mean values derived from observed and predicted distributions, which provides a concise summary statistic of predictive validity (Myers et al., 2022). Across all participants and task conditions, we observed excellent fit overall between observed and predicted mean values for both approach-avoidance choice proportions (all rs > 0.99) and RT (all rs > 0.90; see Supplemental Material).

Data analytic strategy

All analyses were conducted with SPSS software ver. 24.0 (IBM SPSS Statistics for Windows, Version 24.0. Armonk, NY: IBM).

Primary analyses

Our primary analyses aimed to characterise the degree to which measures differed as a function of varying degrees of social reward-threat conflict, social reward, and social threat. To this end, we conducted separate Repeated Measures ANOVA (RM-ANOVA) models to compare modulation as a function of social reward-threat conflict (i.e. 100%_Happy vs. 50%_Happy + 50%_Angry vs. 100%_Angry), social reward (i.e. 0%_Happy vs. 50%_Happy vs. 100%_Happy), or social threat (i.e. 0%_Angry vs. 50%_Angry vs. 100%_Angry). When assumptions of sphericity were violated, we implemented Greenhouse-Geisser correction and report corrected degrees of freedom. For all analyses, the alpha threshold was set at α ≤ 0.05 (two-tailed).

To decompose significant omnibus effects, we utilised two analytic strategies. First, consistent with our previous work (Evans et al., 2024), we used orthogonal polynomial contrasts to examine linear or non-linear modulation as a function of social signal intensity. Second, we also conducted post-hoc comparisons between each pair of morphed facial expressions using paired-samples t-tests.

Additionally, we also conducted a series of confirmatory analyses that aimed to disentangle social AAC effects elicited by co-occurring social reward and social threat (i.e. 50%_Happy + 50%_Angry), from effects elicited by ambiguous social reward (50%_Happy) or ambiguous social threat (50%_Angry). To this end, we directly compared differences between 50%_Happy + 50%_Angry faces and 50%_Happy or 50%_Angry faces. Although non-social AAC research generally disentangles ambiguity by comparing conflicting stimuli with uncertain reward and threat stimuli, it is important to note that neutral faces are also affectively ambiguous (Wieser & Brosch, 2012). Therefore, we also conducted exploratory analyses comparing 50%_Happy + 50%_Angry faces to neutral faces (i.e. 0%_Angry and 0%_Happy).

Psychometric analyses

In line with our previous work (Evans et al., 2024), we also characterised the internal consistency of linear and non-linear SAAC task effects. Specifically, we computed scalar values derived using dot multiplication between values observed for each morph and the polynomial contrasts used in our primary analyses (see Supplemental Material). Internal consistency of polynomial task effects was estimated by computing split-half reliability with Spearman-Brown correction.

Results

Primary analyses

Social reward-threat conflict

Standard Task Measures.

For proportion of approach-avoidance choices (0 = Approach, 1 = Avoid), we observed and replicated a significant effect of social reward-threat conflict (Study 1: F(_1.75, 82.08) = 106.15, p < 0.001, η²_p = 0.69; Study 2: F(_{2, 106}) = 163.18, p < 0.001, η²_p = 0.76; see Figure 3). Demonstrating social AAC, participants exhibited more intermediate proportion of approach-avoidance choices to 50%_Happy + 50%_Angry faces compared to both 100%_Happy and 100%_Angry faces (all pairwise ps < 0.001; see Figure 3 and Table 1). For approach-avoidance RTs, we observed and replicated a significant effect of social reward-threat conflict (Study 1: F(_{2, 94}) = 27.67, p < 0.001, η²_p = 0.37; Study 2: F(_{2, 106}) = 13.68, p < 0.001, η²_p = 0.21; see Figure 4). Consistent with AAC effects, participants exhibited significantly slower RTs to 50%_Happy + 50%_Angry faces compared to both 100%_Happy faces and 100%_Angry faces (all pairwise ps < 0.001; see Table 2). In contrast, RTs did not differ between 100%_Happy and 100%_Angry faces (all pairwise ps > 0.19).

Figure 3.

Social approach-avoidance conflict task choice results. Legend: Demonstrating evidence of social approach-avoidance conflict (AAC), social reward-threat conflict faces (50%_Happy + 50%_Angry) elicited more intermediate approach and avoidance choice distributions compared to both unambiguous social reward (100%_Happy) and social threat (100%_Anger). Overall, the number approach choices also increased as a linear function of social reward (happiness) intensity, whereas the number of avoidance choices increased as a linear function of social threat intensity (anger). Note: *** = p < 0.001; H = Happiness (0%, 50%, or 100% intensity), A = Anger (0%, 50%, or 100% intensity).

Table 1.

Morph-related comparison of approach-avoidance choice proportions.

Morph Type	Sample	Morph A	Morph B	Difference	95% CI	p-value	Cohen’s D
Conflict	Study 1	100Happy + 0Angry 0.32 (0.33)	50Happy + 50Angry 0.67 (0.29)	−0.36 (0.28)	−0.44 – −0.28	<0.001	−1.27
Conflict	Study 1	50Happy + 50Angry 0.67 (0.29)	0Happy + 100Angry 0.92 (0.16)	−0.24 (0.24)	−0.31 – −0.17	<0.001	−1.01
Conflict	Study 1	100Happy + 0Angry 0.32 (0.33)	0Happy + 100Angry 0.92 (0.16)	−0.60 (0.33)	−0.70 – −0.50	<0.001	−1.80
Conflict	Study 2	100Happy + 0Angry 0.24 (0.27)	50Happy + 50Angry 0.56 (0.33)	−0.33 (0.25)	−0.40 – −0.26	<0.001	−1.29
Conflict	Study 2	50Happy + 50Angry 0.56 (0.33)	0Happy + 100Angry 0.90 (0.16)	−0.34 (0.27)	−0.41 – −0.26	<0.001	−1.23
Conflict	Study 2	100Happy + 0Angry 0.24 (0.27)	0Happy + 100Angry 0.90 (0.16)	−0.66 (0.28)	−0.74 – −0.59	<0.001	−2.34
Reward	Study 1	0Happy 0.60 (0.31)	50Happy 0.27 (0.26)	0.33 (0.28)	0.24 – 0.41	<0.001	1.17
Reward	Study 1	50Happy 0.27 (0.26)	100Happy 0.25 (0.27)	0.02 (0.15)	−0.02 – 0.07	0.28	0.16
Reward	Study 1	100Happy 0.25 (0.27)	0Happy 0.60 (0.31)	−0.35 (0.32)	−0.44 – −0.26	<0.001	−1.10
Reward	Study 2	0Happy 0.53 (0.33)	50Happy 0.27 (0.27)	0.25 (0.22)	0.19 – 0.31	<0.001	1.14
Reward	Study 2	50Happy 0.27 (0.27)	100Happy 0.20 (0.23)	0.07 (0.17)	0.03 – 0.12	0.002	0.44
Reward	Study 2	100Happy 0.20 (0.23)	0Happy 0.53 (0.33)	−0.33 (0.28)	−0.41 – −0.25	<0.001	−1.14
Threat	Study 1	0Angry 0.64 (0.32)	50Angry 0.85 (0.22)	−0.21 (0.21)	−0.27 – −0.15	<0.001	−0.98
Threat	Study 1	50Angry 0.85 (0.22)	100Angry 0.93 (0.17)	−0.08 (0.11)	−0.11 – −0.05	<0.001	−0.74
Threat	Study 1	100Angry 0.93 (0.17)	0Angry 0.64 (0.32)	0.29 (0.27)	0.21 –; 0.37	<0.001	1.17
Threat	Study 2	0Angry 0.62 (0.33)	50Angry 0.78 (0.24)	−0.15 (0.17)	−0.20 – −0.11	<0.001	−0.91
Threat	Study 2	50Angry 0.78 (0.24)	100Angry 0.91 (0.15)	−0.13 (0.15)	−0.17 – −0.09	<0.001	−0.87
Threat	Study 2	100Angry 0.91 (0.15)	0Angry 0.62 (0.33)	0.29 (0.28)	0.21 – 0.36	<0.001	1.02

Legend: Values represent the proportion of approach-avoidance choices (approach = 0, avoid = 1). Means and standard deviations are presented for face morphs separately as well as differences for each pairwise comparison.

Figure 4.

Social approach-avoidance conflict task reaction time results. Legend: Demonstrating evidence of social approach-avoidance conflict (AAC), social reward-threat conflict faces (50%_Happy + 50%_Angry) elicited slower reaction time (RT) during decision making compared to both unambiguous social reward (100%_Happy) and social threat (100%_Anger). Additionally, RT also decreased as a linear function of both social reward (happiness) intensity as well as social threat (anger) intensity. Note: *** = p < 0.001; H = Happiness (0%, 50%, or 100% intensity), A = Anger (0%, 50%, or 100% intensity).

Table 2.

Morph-related comparisons of reaction times.

Morph Type	Sample	Morph A	Morph B	Difference	95% CI	p-value	Cohen’s D
Conflict	Study 1	100Happy + 0Angry 807.05 (95.67)	50Happy + 50Angry 882.83 (127.44)	−75.78 (85.16)	−100.51 – −51.06	<0.001	−0.89
Conflict	Study 1	50Happy + 50Angry 882.83 (127.44)	0Happy + 100Angry 788.12 (120.53)	94.71 (95.96)	66.85 – 122.51	<0.001	0.99
Conflict	Study 1	100Happy + 0Angry 807.05 (95.67)	0Happy + 100Angry 788.12 (120.53)	18.93 (98.42)	−9.65 – 47.51	0.19	0.19
Conflict	Study 2	100Happy + 0Angry 810.19 (112.39)	50Happy + 50Angry 858.39 (118.23)	−48.20 (85.75)	−71.60 – −24.79	<0.001	−0.56
Conflict	Study 2	50Happy + 50Angry 858.39 (118.23)	0Happy + 100Angry 793.61 (103.50)	64.78 (87.96)	40.77 – 88.85	<0.001	0.74
Conflict	Study 2	100Happy + 0Angry 810.19 (112.39)	0Happy + 100Angry 793.61 (103.50)	16.58 (108.34)	−12.99 – 46.15	0.27	0.15
Reward	Study 1	0Happy 863.45 (130.19)	50Happy 846.36 (137.19)	17.09 (87.11)	−8.20 – 42.38	0.18	0.20
Reward	Study 1	50Happy 846.36 (137.19)	100Happy 804.46 (109.08)	41.89 (70.63)	21.38 – 62.40	<0.001	0.59
Reward	Study 1	100Happy 804.46 (109.08)	0Happy 846.36 (137.19)	−58.98 (98.71)	−87.64 – −30.32	<0.001	−0.60
Reward	Study 2	0Happy 856.12 (136.06)	50Happy 839.69 (117.06)	16.43 (87.61)	−7.48 – 40.34	0.17	0.19
Reward	Study 2	50Happy 839.69 (117.06)	100Happy 802.64 (103.21)	37.04 (56.09)	21.73 – 52.36	<0.001	0.66
Reward	Study 2	100Happy 802.64 (103.21)	0Happy 856.12 (136.06)	−53.47 (107.23)	−24.21 – 82.74	<0.001	−0.50
Threat	Study 1	0Angry 850.94 (120.14)	50Angry 808.42 (123.22)	42.52 (65.99)	23.36 – 61.68	<0.001	0.64
Threat	Study 1	50Angry 808.42 (123.22)	100Angry 757.17 (107.28)	51.25 (81.48)	27.59 – 74.91	<0.001	0.63
Threat	Study 1	100Angry 757.17 (107.28)	0Angry 850.94 (120.14)	−93.77 (91.85)	−120.44 – −67.10	<0.001	−1.02
Threat	Study 2	0Angry 838.63 (133.12)	50Angry 824.60 (133.66)	14.03 (67.73)	−4.46 – 32.51	0.13	0.21
Threat	Study 2	50Angry 824.60 (133.66)	100Angry 767.04 (106.12)	57.57 (71.66)	38.01 – 77.13	<0.001	0.80
Threat	Study 2	100Angry 767.04 (106.12)	0Angry 838.63 (133.12)	−71.60 (90.64)	−96.33 – −46.86	<0.001	−0.79

Legend: Means and standard deviations are presented for face morphs separately as well as the differences for each pairwise comparison.

DDM Parameters.

For the drift rate parameter (v), we demonstrated and replicated a significant effect of social reward-threat conflict (Study 1: F(_{1.43, 67.08}) = 121.94, p < 0.001, η²_p = 0.72; Study 2: F(_{2, 106}) = 190.83, p < 0.001, η²_p = 0.78; see Figure 5 and Table 3). Consistent with AAC effects, participants exhibited a weaker drift rate for 50%_Happy + 50%_Angry faces compared to both 100%_Happy faces and 100%_Angry faces (all pairwise ps < 0.001; see Figure 5 and Table 3). In contrast, we did not observe reliable AAC effects for the decision threshold parameter, starting point bias parameter, or non-decision time parameter (see Supplemental Material)

Figure 5.

Noisier evidence accumulation underlies social approach-avoidance conflict. Legend: Social reward-threat conflict faces (50%_Happy + 50%_Angry) elicited a smaller absolute magnitude of drift rate (v parameter) compared to both unambiguous social reward (100%_Happy) and social threat (100%_Anger). In terms of absolute magnitude, drift rates also increased as a linear function of both social reward (happiness) intensity as well as social threat (anger) intensity. Note: *** = p < 0.001; H = Happiness (0%, 50%, or 100% intensity), A = Anger (0%, 50%, or 100% intensity).

Table 3.

Morph-related comparisons of latent drift rate parameters.

Morph Type	Sample	Morph A	Morph B	Difference	95% CI	p-value	Cohen’s D
Conflict	Study 1	100Happy + 0Angry −1.19 (1.76)	50Happy + 50Angry 0.79 (1.33)	−1.97 (1.45)	−2.39 – −1.55	<0.001	−1.36
Conflict	Study 1	50Happy + 50Angry 0.79 (1.33)	0Happy + 100Angry 2.48 (1.08)	−1.70 (1.25)	−2.06 – −1.33	<0.001	−1.36
Conflict	Study 1	100Happy + 0Angry −1.19 (1.76)	0Happy + 100Angry 2.48 (1.08)	3.67 (2.08)	3.07 – 4.27	<0.001	−1.77
Conflict	Study 2	100Happy + 0Angry −1.81 (1.61)	50Happy + 50Angry 0.32 (1.64)	−2.13 (1.58)	−2.56 – −1.70	<0.001	−1.35
Conflict	Study 2	50Happy + 50Angry 0.32 (1.64)	0Happy + 100Angry 2.55 (1.16)	−2.23 (1.44)	−2.62 – −1.84	<0.001	−1.55
Conflict	Study 2	100Happy + 0Angry −1.81 (1.61)	0Happy + 100Angry 2.55 (1.16)	4.36 (1.88)	3.85 – 4.89	<0.001	−2.32
Reward	Study 1	0Happy 0.43 (1.51)	50Happy −1.25 (1.34)	1.69 (1.47)	1.26 – 2.11	<0.001	1.15
Reward	Study 1	50Happy −1.25 (1.34)	100Happy −1.40 (1.43)	0.15 (1.02)	−0.15 – 0.44	0.33	0.14
Reward	Study 1	100Happy −1.40 (1.43)	0Happy 0.43 (1.51)	−1.83 (1.92)	−2.39 – −1.27	<0.001	−0.95
Reward	Study 2	0Happy 1.53 (1.44)	50Happy 0.61 (1.08)	1.53 (1.44)	1.14 – 1.92	<0.001	1.06
Reward	Study 2	50Happy 0.61 (1.08)	100Happy −2.14 (1.83)	0.61 (1.08)	0.31 – 0.90	<0.001	0.56
Reward	Study 2	100Happy −2.14 (1.83)	0Happy 1.53 (1.44)	−2.14 (1.84)	−2.64 – −1.64	<0.001	−1.17
Threat	Study 1	0Angry 0.48 (1.38)	50Angry 1.76 (1.22)	−1.28 (1.13)	−1.61 – −0.95	<0.001	−1.13
Threat	Study 1	50Angry 1.76 (1.22)	100Angry 2.61 (1.14)	−0.85 (0.93)	−1.12 – −0.58	<0.001	−0.91
Threat	Study 1	100Angry 2.61 (1.14)	0Angry 0.48 (1.38)	2.13 (1.32)	1.66 – 2.60	<0.001	1.32
Threat	Study 2	0Angry 0.63 (1.84)	50Angry 1.57 (1.45)	−0.94 (1.06)	−1.23 – −0.65	<0.001	−0.88
Threat	Study 2	50Angry 1.57 (1.45)	100Angry 2.72 (1.27)	−1.15 (1.28)	−1.50 – −0.80	<0.001	−0.90
Threat	Study 2	100Angry 2.72 (1.27)	0Angry 0.63 (1.84)	2.09 (1.87)	0.80 – 1.50	<0.001	1.12

Legend: Means and standard deviations are presented for face morphs separately as well as the differences for each pairwise comparison.

Social reward

Standard Task Measures.

For approach-avoidance choices, we replicated a significant effect of social reward intensity (Study 1: F(_{1.37, 64.29}) = 54.26, p < 0.001, η²_p = 0.54; Study 2: F(_{1.50, 79.59}) = 59.37, p < 0.001, η²_p = 0.53), which was driven by a linear increase in approach choices as social reward intensity increased (Study 1 Linear Contrast: F(_{1, 47}) = 58.02, p < 0.001; Study 2 Linear Contrast: F(_{1, 53}) = 70.42, p < 0.001; see Figure 3). For approach-avoidance RTs, we replicated a significant effect of social reward intensity (Study 1: F(_{1.76, 82.71}) = 11.88, p < 0.001, η²_p = 0.20; Study 2: F(_{1.41, 74.59}) = 10.89, p < 0.001, η²_p = 0.17), which was driven by a linear decrease in RTs as social reward intensity increased (Study 1 Linear Contrast: F(_{1, 47}) = 17.14, p < 0.001; Study 2 Linear Contrast: F(_{1, 53}) = 13.43, p < 0.001; see Figure 4).

Drift Diffusion Parameters.

For the drift rate parameter (v), we replicated a significant effect of social reward intensity (Study 1: F(_{1.38, 64.87}) = 43.30, p < 0.001, η²_p = 0.48; Study 2: F(_{1.50, 79.37}) = 43.30, p < 0.001, η²_p = 0.53), which was driven by a linear shift towards negative drift rates as social reward intensity increased (Study 1 Linear Contrast: F(_{1, 47}) = 43.56, p < 0.001; Study 2 Linear Contrast: F(_{1, 53}) = 73.36, p < 0.001; see Figure 5). In contrast, we did not observe reliable effects of social reward intensity on the decision threshold, starting point bias, or non-decision time parameters (see Supplemental Material).

Social threat

Standard Task Measures.

For approach-avoidance choices, we replicated a significant effect of social threat intensity (Study 1: F(_{1.16, 54.69}) = 47.90, p < 0.001, η²_p = 0.51; Study 2: (F(_{1.20, 79.59}) = 63.80, p < 0.001, η²_p = 0.49), which was driven by a linear increase in avoidance choices as social threat intensity increased (Study 1 Linear Contrast: F(_{1, 47}) = 52.16, p < 0.001; Study 2 Linear Contrast: F(_{1, 53}) = 55.92, p < 0.001; see Figure 3 and Table 1). For approach-avoidance RTs, we observed and replicated a significant effect of social threat intensity (Study 1: F(_{1.77, 82.96}) = 32.68, p < 0.001, η²_p = 0.41; Study 2: F(_{1.75, 92.77}) = 26.00, p < 0.001, η²_p = 0.33), which was driven by a linear decrease in RTs as social threat intensity increased (Study 1 Linear Contrast: F(_{1, 47}) = 50.03, p < 0.001; Study 2 Linear Contrast: F(_{1, 53}) = 33.69, p < 0.001; see Figure 4 and Table 2).

Drift Diffusion Parameters.

For the drift rate parameter (v), we observed and replicated a significant effect of social threat intensity (Study 1: F(_{1.40, 65.74}) = 69.54, p < 0.001, η²_p = 0.60; Study 2: F(_{1.36, 71.86}) = 56.72, p < 0.001, η²_p = 0.52), which was driven by a linear increase in avoidance-biased drift rate across morphs (Study 1 Linear Contrast: F(_{1, 47}) = 50.03, p < 0.001; Study 2 Linear Contrast: F(_{1, 53}) = 33.69, p < 0.001; see Figure 5 and Table 3). For the decision threshold parameter (a), we observed and replicated a significant effect of social threat intensity (Study 1: F(_{2, 94}) = 27.80, p < 0.001; Study 2: F(1.74, 92.11) = 11.79, p < 0.001; see Figure S1 and Table S1). In contrast, we did not observe reliable effects of social threat intensity on the initial starting point bias or non-decision time parameters (see Supplemental Material).

Disentangling social AAC from social ambiguity

For the proportion of approach-avoidance choices (0 = Approach, 1 = Avoid), 50%_Happy + 50%_Angry faces elicited a smaller proportion of approach choices compared to 50%_Happy faces (Study 1: t(47) = 11.66, p < 0.001, 95% CI = [0.33 – 0.47], d = 1.68; Study 2: t(53) = 8.00, p < 0.001, 95% CI = [0.22–0.36], d = 1.08) as well as a smaller proportion of avoidance choices compared to 50%_Angry faces (Study 1: t(47) = −4.93, p < 0.001, 95% CI = [−0.25 – −0.10], d = −0.71; Study 2: t(53) = −6.24, p < 0.001, 95% CI = [−0.28 – −0.14], d = −0.85). For approach-avoidance RTs, participants exhibited significantly longer RTs to 50%_Happy + 50%_Angry faces compared to 50%_Happy faces in Study 1 (t(47) = 2.36, p = 0.02, 95% CI = [5.43–67.52], d = 0.34), which was not statistically significant in Study 2 (t(53) = 1.41, p = 0.16, 95% CI = (−7.82–45.22), d = 0.19). Additionally, 50%_Happy + 50%_Angry faces elicited significantly longer RTs to compared to 50%_Angry faces in both studies (Study 1: t(47) = 4.91, p < 0.001, 95% CI = [43.94 – 104.87], d = 0.71; Study 2: t(53) = 2.56, p = 0.01, 95% CI = [7.25 – 60.33], d = 0.35). For the drift rate parameter, 50%_Happy + 50%_Angry faces elicited a significantly larger (i.e. closer to 0) drift rate compared to 50%_Happy faces (Study 1: t(47) = 11.01, p < 0.001, 95% CI = [1.66–2.41], d = 1.59; Study 2: t(53) = 8.43, p < 0.001, 95% CI = [1.23–2.00], d = 1.15) as well as a significantly smaller drift rate compared to 50%_Angry faces (Study 1: t(47) = −4.56, p < 0.001, 95% CI = [−1.40 – −0.55], d = −0.66; Study 2: t(53) = −6.02, p < 0.001, 95% CI = [−1.66 – −0.83], d = −0.82).

Exploratory Comparisons to Neutral Faces.

Given that 0%_Happy and 0%_Angry faces are both neutral facial expressions, we averaged responses to these neutral faces for comparisons with 50%_Happy + 50%_Angry faces. For the proportion of approach-avoidance choices, 50%_Happy + 50%_Angry faces did not significantly differ compared to neutral faces (Study 1: t(47) = 1.57, p = 0.12, 95% CI = [−0.02–0.12], d = 0.23; Study 2: t(53) = −0.28, p = 0.78, 95% CI = [−0.09 – 0.06], d = −0.03). For approach-avoidance RTs, 50%_Happy + 50%_Angry faces elicited significantly longer RTs to compared to neutral faces in Study 1 (t(47) = 2.35, p = 0.02, 95% CI = [3.66 – 47.61], d = 0.34), which was in the same direction, but not statistically significant in Study 2 (t(53) = 1.03, p = 0.31, 95% CI = [−10.37 – 32.40], d = 0.14). For the drift rate parameter, 50%_Happy + 50%_Angry faces did not elicit differences in parameter estimates compared to neutral faces (Study 1: t(47) = 1.69, p = 0.10, 95% CI = [−0.06 – 0.72], d = 0.24; Study 2: t(53) = −0.11, p = 0.61, 95% CI = [−0.57 – 0.37], d = −0.07).

Given the exploratory nature of this analysis and somewhat inconsistent direction of effects across studies, we repeated analyses in a larger sample combining participants from both Study 1 and Study 2 (n = 102). For approach-avoidance choice selection, 50%_Happy + 50%_Angry faces did not elicit significant differences compared to neutral faces (t(101) = 0.80, p = 0.42, 95% CI = [−0.03 – 0.07], d = 0.08). For approach-avoidance RTs, 50%_Happy + 50%_Angry faces elicited significantly longer RTs compared to neutral faces (t(101) = 2.35, p = 0.02, 95% CI = [2.76 – 33.04], d = 0.23). For the drift rate parameter, 50%_Happy + 50%_Angry faces did not elicit significant differences compared to neutral faces (t(101) = 0.62, p = 0.54, 95% CI = [−0.21–0.39], d = 0.06).

Psychometric analyses

Overall, our primary analyses demonstrated that task effects were most robust and replicable for approach-avoidance choices, drift rate (v) and RT. Therefore, we characterised the psychometric properties for task effects in these three primary measures. Across all three types of morphed faces (social reward-threat conflict, social reward, and social threat), linear task effects on approach-avoidance choices and drift rate (v) exhibited good or excellent internal consistency in both Study 1 (all rs ≥ 0.90) and Study 2 (all rs ≥ 0.86). Similarly, non-linear task effects on RT exhibited internal consistency that approached or reached acceptable levels in both Study 1 (all ρs ≥ 0.66) and Study 2 (all ρs ≥ 0.66).

Discussion

Overall, our results demonstrated social reward-threat conflict faces robustly and reliably elicit social AAC effects. First, social reward-threat conflict (50%_Happy + 50%_Angry) elicited more intermediate approach-avoidance choice proportions and longer RTs during decision-making compared to both unambiguous social reward faces (i.e. 100%_Happy) and social threat faces (i.e. 100%_Angry). Moreover, social reward-threat conflict faces also exhibited more intermediate approach-avoidance choices and longer RTs compared to ambiguous social reward faces (i.e. 50%_Happy) and social threat faces (i.e. 50%_Angry). Together, these results demonstrate the methodological rigor of this approach by disentangling social AAC effects from broader social ambiguity effects. Although controlling for reward and threat ambiguity in this manner is directly aligned with non-social AAC research, it should be noted that social reward-threat conflict faces elicited less consistent differences compared to neutral faces that are ambiguous due to a lack of social reward or social threat. Second, computational DDM results demonstrated that social AAC effects were largely driven by noisier evidence accumulation as evidenced by unique effects of social AAC on drift rate parameters. Third, our psychometric analyses demonstrated that SAAC task effects generally exhibited adequate internal consistency for rigorous translational research characterizing individual differences. Together, these results demonstrate that our novel SAAC paradigm can robustly and reliably elicit social AAC, which has not been experimentally instantiated and measured in humans to date.

As hypothesised, social reward-threat conflict faces (50%_Happy + 50%_Angry) elicited social AAC as exhibited by more intermediate approach-avoidance choice proportions and longer RTs during decision making compared to unambiguous social reward faces (100%_Happy) and social threat faces (100%_Angry). This pattern of results is highly consistent with the large body of AAC effects observed in non-social AAC research (Aupperle et al., 2011; Bublatzky et al., 2017; Diederich, 2003; Garcia-Guerrero et al., 2023; Schlund et al., 2016). Moreover, our results also demonstrated that social AAC effects could not simply be attributed to the ambiguity of morphed facial expressions more generally. Specifically, social reward-threat conflict faces also elicited more intermediate approach-avoidance choice proportions and longer RTs compared to ambiguous social reward (50%_Happy) and social threat (50%_Angry). Together, these results demonstrate that social reward-threat conflict robustly elicits social AAC effects, which dovetails with research using non-social AAC experimental paradigms.

However, it is important to note that less consistent differences were observed when comparing social reward-threat conflict faces with neutral faces that are ambiguous due to a lack of affect, rather than ambiguous affect (Wieser & Brosch, 2012). Specifically, social reward-threat conflict faces did not significantly differ in approach-avoidance choice selection compared to neutral faces regardless of the sample size examined. Although social reward-threat conflict faces elicited longer RTs compared to neutral faces in both Study 1 and a larger sample size when combining samples, this effect was not statistically significant in Study 2. Overall, this pattern of results suggests that neutral faces may introduce greater ambiguity during decision-making than ambiguous social reward or ambiguous social threat. This pattern of results is notable given that non-social AAC research typically relies on ambiguous rewards and threats to disentangle AAC effects from ambiguity effects, rather than utilising neutral stimuli that are not associated with a clear outcome contingency (e.g. Chu et al., 2023). Thus, it may be important for social AAC research moving forward to consider neutral facial expressions as a unique control condition to disentangle social AAC effects from social ambiguity effects. However, it is also important to note that our previous research has established that social reward-threat conflict faces are rated as equally happy and angry (i.e. conflicting social signals) and that social reward-threat conflict faces are rated as both more happy and more angry compared to neutral faces (i.e. more potent conflict; Evans et al., 2024). Therefore, we argue that social reward-threat conflict faces are nevertheless best suited to elicit social AAC effects more specifically, whereas neutral faces may be better suited towards understanding social ambiguity effects more generally. Future research using additional modalities such as neuroimaging may be useful to more precisely disentangle the mechanisms underlying motivational responses to conflicting and neutral facial expressions.

As described previously, longer RTs associated with social AAC can reflect multiple distinct decision-making processes, which we aimed to elucidate within a DDM framework. Our DDM analyses demonstrated and replicated that social AAC was characterised by noisier evidence accumulation (i.e. smaller drift rate). Specifically, social reward-threat conflict elicited a smaller drift rate compared to both unambiguous social reward and social threat faces as well as ambiguous social reward and social threat faces. Notably, these findings dovetail with DDM research using non-social stimuli and outcomes, which demonstrate that non-social AAC is also characterised by noisier evidence accumulation (Chu et al., 2023; Le Duc et al., 2025). From a functional perspective, these findings are consistent with a large body of literature proposing that AAC serves to inhibit goal-directed behavior when opportunities for both reward and punishment are simultaneously present (Barker et al., 2019; Gray & McNaughton, 2003). Taken together, these findings inform a more mechanistic prediction regarding how social AAC inhibits goal-directed behavior in social contexts. Specifically, social AAC may inhibit goal-directed behavior to provide a longer window of time for noisier evidence accumulation to sufficiently reduce outcome uncertainty in a social situation that contains opportunities for both social affiliation and social rejection. This mechanistic prediction is consistent with findings that social anxiety, which is characterised by disrupted goal-directed behavior in unfamiliar social situations, is associated with both greater behavioral inhibition as well as more frequent experiences of social AAC (Barker et al., 2019; Kashdan et al., 2008). Thus, although speculative, we propose that noisier evidence accumulation during social AAC may be an underlying mechanism that facilitates behavioral inhibition when social affiliation and social rejection are similarly likely to occur.

In contrast, we observed inconsistent evidence for social AAC effects in other DDM parameters despite some support for our pre-registered predictions regarding these parameters. In Study 2, as predicted, we observed that social reward-threat conflict faces were characterised by a smaller starting point bias compared to unambiguous social reward as well as a smaller decision threshold compared to unambiguous social threat. However, it is important to highlight that these findings were not consistent with traditional patterns of AAC. Specifically, social reward-threat conflict faces were not also conjointly characterised by a smaller decision threshold compared to unambiguous social reward or a smaller starting point bias compared to unambiguous social threat. Regarding the decision threshold parameter, individuals may exclusively exhibit more response caution (larger decision threshold) to unambiguous social threats to avoid probable negative outcomes such as social rejection. This interpretation is consistent with previous DDM research demonstrating that stimuli associated with more certain negative outcomes robustly elicit larger decision thresholds, which may maximise the probability of successful threat avoidance (Chu et al., 2023). Regarding the initial starting point bias parameter, it may be premature to interpret this finding further given that Study 1 only demonstrated a marginal effect, which raises concerns regarding robust replication. Perhaps more fundamentally, approach-avoidance decisions in the current study were not paired with post-choice selection outcomes (i.e. choice-dependent rewards or punishments), which obfuscates interpretation of the starting point bias parameter more generally. Lastly, we did not predict or observe social AAC effects in the non-decision time parameter, which suggests that slower RTs in response to social reward-threat conflict faces are not attributable to basic visual encoding or motor response execution. Together, the lack of consistent social AAC effects in these DDM parameters suggests that social AAC may be primarily driven by noisier evidence accumulation (drift rate) when social reward and social threat simultaneously co-occur.

Finally, given concerns regarding replication in the field of psychology, it is critical to develop and validate experimental paradigms that produce psychometrically-sound measures for use in translational research (Parsons et al., 2019). Overall, social AAC effects in the current study generally demonstrated psychometric properties ranging from approaching acceptable to excellent, which is an important prerequisite for reliably characterizing individual differences in SAAC task effects. As described previously, social AAC occurs frequently in daily life and individual differences in successfully navigating social AAC can account for variability in social behaviors (for reviews, see Barker et al., 2019; Strack & Deutsch, 2004). Importantly, it is necessary to develop rigorous measures of SAAC specifically, in addition to non-social AAC more generally, given that social reward and social threat exert unique influences on behavior compared to non-social rewards and threats (Barclay & Benard, 2020; LoBue & Pérez-Edgar, 2014; McDermott & Egwuatu, 2019; Rademacher et al., 2010; Ruff & Fehr, 2014). Therefore, the SAAC task provides an important first step towards mechanistically assaying how individual differences in navigating social AAC processes contributes to variability in social behavior. is important to develop reliable experimental paradigms that elicit social AAC towards ultimately better understanding individual differences in social behavior.

Although these findings offer important implications, it is necessary to note several limitations of the current research. First, we relied on a validated morphed face stimulus set from our previous work, which is comprised exclusively of Caucasian faces to minimise potential confounds of racial identity on social AAC effects (Paulus & Wentura, 2014). Therefore, it will be important for future research to develop more racially representative morphed face stimulus sets and empirically establish the degree to which racial identity potentially moderates social AAC effects. Second, we created morphed facial expressions using linear interpolation, which has been validated in our previous research (Evans et al., 2024). Although some research demonstrates that linear interpolation adequately captures changes in facial action units that are analogous to facial landmark changes measured in spontaneous facial expressions of emotion (Korolkova, 2018), other research demonstrates that linear interpolation does not comprehensively capture more complex, non-linear spatiotemporal changes in facial action units (Krumhuber et al., 2023). Therefore, future research will be necessary to further validate SAAC effects using morphed facial expressions derived with more sophisticated methodologies. Finally, we did not conduct a parameter recovery simulation study to further validate DMMs that generated latent decision-making parameters. However, we note that our primary DDM parameter results replicated across independent datasets, which itself provides a robust form of model validation. Nevertheless, a successful parameter recovery simulation study would provide greater confidence that DDM models produce valid representations of underlying latent parameter estimates.

Despite these limitations, the current studies nevertheless offer an important advancement by validating a novel SAAC paradigm that experimentally and reliably elicits social AAC. As noted previously, social AAC is common in social contexts and navigation of social AAC facilitates adaptive social behavior. Together, these results demonstrate that the SAAC paradigm may ultimately be useful towards providing mechanistic insights into variability in social behavior.

Supplementary Material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/02699931.2025.2533382.