Childhood violence exposure and social information processing in young adults: Does relationship with the perpetrator matter?

Abstract

Humans are generally biased to show implicit favoritism for in-group over out-group members, but developmental experiences may alter this process in important ways. Prior work has elucidated associations of family (i.e., in-group) violence exposure in childhood with risk for internalizing symptoms through weakened implicit favoritism for novel in-group members. The present study probes whether childhood violence exposure influences implicit bias and psychopathology differentially depending on the participant’s relationship with the perpetrator (i.e., in-group vs. out-group member) at the time of exposure. We administered a minimal group assignment paradigm and implicit association test to 455 young adults aged 18–25. Young adults who experienced out-group violence in childhood showed stronger implicit in-group favoritism compared to those who experienced in-group or no violence. Implicit out-group favoritism was associated with increased alcohol use. Early-life experiences may shape innate preferences for novel in-group vs. out-group members in ways that have lasting implications for mental health.

Introduction

Experiences of violence during childhood are common, with up to half of U.S. children experiencing direct victimization (e.g., caregiver abuse or assault) or witnessing violence (Finkelhor, Ormrod, & Turner, 2009; Finkelhor, Turner, et al., 2009; Lewis et al., 2019; McLaughlin et al., 2013). Childhood experiences of violence are a strong predictor of transdiagnostic psychopathology across the lifespan (Clark et al., 2010; Cohen et al., 2001; Fowler et al., 2009; Green et al., 2010; McLaughlin et al., 2012), accounting for a significant portion of psychopathology emerging in childhood, adolescence, and adulthood (Green et al., 2010; McLaughlin et al., 2012). Efforts to identify mechanisms that explain this link have highlighted numerous pathways. Here, we examine a novel set of social information processing mechanisms involving implicit preferences for novel in-group vs. out-group members.

One promising area of inquiry involves characterizing how early experiences of violence shape social cognition. Experiencing a sense of belonging in social groups throughout development is critical for well-being, because membership in these groups promotes social connection, identity exploration, mutual support, and safety, among other things (Boyd et al., 2011; Skinner & Meltzoff, 2019; Tajfel & Turner, 1979; Tomasello et al., 2012). These benefits arise from the formation of strong in-group bonds, which necessitate effective use of evolutionarily conserved social categorization strategies. This social categorization mechanism helps us to differentiate between those who are similar to or aligned with us (i.e., “in-groups”) and those who are different from or opposing us (i.e., “out-groups”). Being able to accurately and reliably discern in-group from out-group members provides individuals with a framework for interpreting, predicting, and assessing others’ behaviors based on an implicit understanding of intergroup expectations, norms, and dynamics (Allport, 1954; Meltzoff, 2007; Over, 2016; Swann et al., 2009; Tajfel & Turner, 1979; Tomasello et al., 2012; Wilson & Wilson, 2007). However, early exposure to violence may disrupt these processes, altering learning regarding expectations of intra- and inter-group dynamics in ways that reshape the expression of socio-cognitive biases and related behaviors that underly social connection and emotional well-being throughout development. The present study bridges methods and theory from developmental, social, and clinical psychology to explore these questions.

Childhood Experiences of Violence and Social Information Processing

Childhood violence exposure has been consistently associated with socially relevant information processing biases that facilitate rapid identification of, and response to potential threats at various thresholds of perception and awareness. For example, children exposed to violence are able to identify anger—but not other emotions—in facial expressions more rapidly, with higher accuracy, and with less perceptual input than children who have not experienced violence (Pollak & Sinha, 2002; Pollak et al., 2009), and this threat-related bias persists into adulthood (Gibb et al., 2009). Similarly, children who have previously experienced violence are also more likely to perceive neutral and ambiguous faces as angry compared with those who have not experienced violence (Ardizzi et al., 2015; Pollak et al., 2000).

At a downstream cognitive and behavioral level, these social information processing biases extend to the interpretations children make about the actions of others and their responses based on those interpretations. Specifically, children who experience violence are more likely to perceive ambiguous acts as intentional and malevolent compared with non-exposed youth (hostile attribution bias; Dodge et al., 2015; Dodge et al., 1995). Behaviorally, youth who have experienced violence are more likely to develop aggressive patterns of response to perceived provocation compared with youth who have not experienced violence, even after controlling for baseline aggressive tendencies (Kimonis et al. 2011; Myers et al., 2018). Some evidence suggests that the development of aggressive behavioral responses is at least partially explained by changes in social cognition over time, reflecting increased acceptance of aggression as an allowable response (Guerra et al, 2003; McMahon et al., 2009; Peckins et al., 2018). These patterns are further supported by evidence that children who experience violence exhibit elevated responses in the amygdala and other nodes of the salience network to social cues that signify threat, which may play a crucial role in facilitating the fight or flight response (Hein & Monk, 2017; Jenness et al., 2021; McCrory et al., 2011; McCrory et al., 2013; McLaughlin et al., 2019). Importantly, violence exposure may also erode foundational social expectations such as trust and affiliation. For example, children exposed to community violence exhibit lower levels of prosocial behavior and reduced trust in others (Littman et al., 2020). Together, these findings highlight that violence-exposed youth are more likely to 1) quickly perceive threat in their environment, 2) attribute hostile intent to ambiguous actions, 3) develop more pro-aggression/anti-trust attitudes, and 4) more readily use aggression in certain contexts.

Social Information Processing Biases and Psychopathology Risk

These patterns of social information processing, represented in perception, neural activation, and behavior, may contribute to defensive responses that are likely adaptive in the short term by facilitating the rapid identification of and responses to danger in contexts where threat of harm is high or unpredictable. At the same time, these tendencies may become maladaptive later, especially in safe contexts where the threat of harm is lower. Indeed, these behavioral and neural response patterns are concurrently and prospectively associated with increased risk for many forms of psychopathology, including anxiety, depression, PTSD, psychosis, substance use problems, and externalizing symptoms (Briggs-Gowan et al., 2016; Dodge et al., 1995; Dodge et al., 2015; Dotterer et al., 2017; Marusak et al., 2015; McMahon et al., 2009; Peckins et al., 2018; Reid et al., 2006; Shackman & Pollak, 2014; Swartz et al., 2015).

Social Information Processing Biases as a Mechanism

Experiences of violence in childhood have clear influences on the development of numerous aspects of threat and social information processing. However, it is unknown whether violence exposure may also influence other learned subconscious biases involved in social categorization and establishing group affiliations, such as implicit intergroup biases, that operate somewhere between early subconscious perception and later top-down response processes.

Previous work with youth who have not experienced violence shows that children’s access to and understanding of social categories and group biases vary as a function of their own social experiences and expectations across development (for reviews see Rhodes & Baron, 2019; Skinner & Meltzoff, 2019). For example, infants as young as 6 months old preferentially look at individuals who previously spoke their native language compared to those who speak a foreign language, or their native language spoken with a foreign accent. In addition, children as young as 5 years old preferentially choose as friends individuals who speak their native language (Kinzler et al., 2007). Such social preferences and behaviors may be informed by an underlying assumption among children that perceived similar others are more likely to abide by accepted social norms than perceived dissimilar others (Liberman et al., 2018). Indeed, children aged 3–9 years old expect that members of the same group should not harm one another, while harming members of other groups is considered more acceptable (Rhodes & Chalik, 2013). Importantly for the present work, children as young as 5 years old are primed to rapidly form implicit and explicit biases for new in groups relative to out groups on the basis of minimal information (e.g., using minimal group induction), which further highlights the social significance of these biases for promoting connection and self-preservation (Cvencek et al., 2016; Dunham et al., 2011).

Relational Status to the Perpetrator as a Potential Modulating Factor

In a prior study, we used minimal group assignment and implicit association tests, experimental methods common in social psychology, to test whether childhood violence exposure increased psychopathology risk through changes in implicit intergroup biases (Kasparek et al., 2023). We found that young children who experienced in-group violence (i.e., intrafamilial violence) showed weaker in-group favoritism for novel in-groups at a later timepoint compared to children who had not experienced violence. Weaker in-group favoritism, in turn, predicted increases in internalizing symptoms over a 2-year period and mediated the association between in-group violence exposure and prospective internalizing symptoms. These results prompted additional questions, however, related to whether the child’s relationship to the perpetrator(s) of violence at the time of exposure may influence these associations. Consider, for example, a case in which a child has primarily experienced violence from figures that humans are evolutionarily primed to classify as in-group members (e.g., family members). Given that humans expect in-group members to provide safety and abide by prosocial norms (Liberman et al., 2018), the violation of these expectations inherent in a violent interaction may alter the development or expression of implicit favoritism for in-group members. In contrast, considering expectations that out-group members are more likely to violate prosocial norms (Baron & Dunham, 2015; Liberman et al., 2018), violence perpetrated by out-group members may have a different impact on implicit in-group favoritism (e.g., by strengthening associations of the out-group with negative concepts).

Given the significance of highly attuned intergroup classification biases for promoting affiliations that facilitate belonging, prosperity, and well-being—and for avoiding those that may entangle individuals in conflict or harm—identifying experiences that shape the development and expression of these biases is crucial. Scholars of developmental psychopathology have long highlighted the importance of the child-perpetrator relationship as a potential moderator of how violence exposure influences social cognitive development (Dodge, 2006). In parallel, developmental theorists have advocated for a social identity approach to understanding how group attitudes and intergroup biases unfold across development (Verkuyten, 2021). This perspective emphasizes the interplay between psychological processes, such as implicit bias, and the broader social, cultural, and political contexts in which they develop. Importantly, group identities are not limited to social categorization—they also carry stereotypical expectations and behavioral consequences that meaningfully shape real-world decisions and functioning. Prior research has shown that these biases shift across development as a function of experience and expectation (for reviews, see Rhodes & Baron, 2019; Skinner & Meltzoff, 2019).

Young adulthood represents a particularly salient period for examining these processes. It is marked by identity formation, increasing autonomy, the expansion of social networks beyond early caregiving environments, and the emergence of novel in-group/out-group dynamics (Arnett, 2000). It is also a developmental window when lifelong patterns of interpersonal functioning and mental health often take shape, and when many psychiatric disorders—especially alcohol use disorders—first emerge at scale (Kessler et al., 2005). Studying implicit bias during this period offers a unique opportunity to assess whether violence-related biases persist as individuals transition into new social roles, and to evaluate how these biases may influence functioning and risk.

The Present Study

Together, these streams of evidence from clinical, social, and developmental psychology show that implicit intergroup biases emerge early in development and inform social interactions and affiliations and related behavioral responses throughout development. Thus, in addition to attempting to replicate findings of weaker in-group favoritism among young adults who experienced in-group violence as children (Kasparek et al., 2023), the present study examines whether in- vs. out-group status of the perpetrator of violence modulates patterns of implicit intergroup bias following experiences of violence in childhood. A child’s relationship to someone who subsequently behaved violently toward them or whom they saw behaving violently toward someone else would likely be a crucial factor to consider regarding intergroup biases given that expectations may be confirmed or violated in ways that produce new learning. Given the salience of threatening experiences and their widespread impact on learning across development as previously discussed, we may reasonably expect evolutionary biases scaffolding social categorization of and identification with novel in- and out-group members to be altered by such experiences.

We thus sought to test several hypotheses in the present study. We hypothesized that childhood experiences of violence would not be associated with differences in implicit bias for novel groups, and that implicit bias for novel groups would instead vary as a function of whether the perpetrator was an in- or out-group member at the time of exposure. Specifically, we hypothesized that experiencing out-group violence in childhood would be associated with stronger implicit favoritism for novel in-groups in young adulthood. Conversely, we hypothesized that any experience of in-group violence in childhood would result in weaker implicit favoritism for novel in-groups. Finally, we hypothesized that weaker implicit favoritism for novel in-groups would be associated with transdiagnostic psychopathology, and mediate associations of childhood violence exposure with psychopathology.

Transparency and Openness

Preregistration

Hypotheses, study design elements, methods, and analytic plan for this study were pre-registered on Open Science Framework (https://osf.io/hm8uy). The pre-registration was submitted after data were collected but before data were examined or any outcomes were observed. We note that the hypotheses presented in this manuscript are directional and thus differ from the non-directional hypotheses that were pre-registered.

Data, Materials, Code, and Online Resources

We do not have permission to redistribute some materials used in the present study, such as the FACEs affect stimuli and iatgen Qualtrics IAT code and analysis scripts. Those materials are easily available online, however. The cleaned final dataset, variable descriptions, and R code used to analyze the data are available through Harvard Dataverse (https://doi.org/10.7910/DVN/CCMJTJ) and also linked through Open Science Framework (https://osf.io/hm8uy). In addition, we include several figures and tables, as well as additional methods and results in the Supplemental Material. Some supplemental figures provide visual representations of the IAT task conditions so readers can visualize examples of the exact stimuli that participants saw in various conditions. Supplemental methods and results, as well as other supplemental figures and tables, describe and illustrate effects from supplemental analyses performed to clarify results from pre-registered analyses.

Reporting

We report how we determined our sample size, all data exclusions, all manipulations, and all measures in the study.

Ethical Approval

The Institutional Review Board of Harvard University approved all procedures. This study was not carried out in accordance with the current provisions of the World Medical Association Declaration of Helsinki as the study procedures were pre-registered after data collection had concluded.

Method

Participants and Procedures

Participants were recruited via Prolific Academic, an academically oriented, research-specific online platform for running behavioral studies. Prolific was chosen because it is shown to have an optimally honest and experiment-naïve user network compared with other platforms, such as Amazon’s Mechanical Turk, and higher data quality (i.e., fewer failed attention checks) compared to comparable research-focused platforms, such as CrowdFlower (Peer et al., 2017). To decide the final sample size needed to detect a small effect (f ²=.02), we ran a power simulation using the WebPower package in R (Zhang & Mai, 2021). This analysis showed that a sample of approximately 500 participants is necessary to be powered at 80% (1–β=0.8) with α=0.05. We ran 1,250 participants through a screener to assess violence exposure prior to age 18 (i.e., childhood violence exposure) and whether participants experienced or witnessed in- or out-group violence exposure as described in more detail below. Using our pre-registered definitions of in- and out-group violence exposure (https://osf.io/hm8uy), 928 participants met criteria for violence exposure and 322 had no exposure. Of those who experienced violence, 660 met criteria for in-group violence exposure and 268 met criteria for out-group violence experiences.

A total of 666 participants across the three violence exposure groups were invited to participate in the full study procedures based on the following pre-registered inclusion criteria: English-speaking, non-colorblind, had at least a 95% satisfaction rating on Prolific Academic, between 18 and 25 years old, provided correct task completion codes, and passed all attention checks. Participants who completed the full study answered questionnaires about psychopathology symptoms over the past month, underwent a minimal group assignment induction, and completed a related implicit association test as described below. Additional pre-registered inclusion criteria were applied to the 666 participants who were invited to take part in the full study, including that they completed the full procedure, entered the correct task completion codes, passed all attention checks, and correctly identified the team they were assigned to and the team’s color. 4 participants were also excluded based on poor performance on the IAT (see Methods for more details). Several inclusion and exclusion criteria were not pre-registered but were incorporated into the data cleaning pipeline post hoc to ensure the highest data quality for hypothesis testing. These include the exclusion of participants with <95% approval rating on Prolific and those with poor IAT performance (i.e., >10% trials with RTs <300ms) as well as individual trials from the IAT with response times <300ms.

After filtering by these criteria, the final sample included 455 participants (M_age=22 years and 3.6 months, SD=2 years, 45.9% female). Of these 455 participants, 33.4% (n=152) did not experience violence, 34.3% (n=156) experienced any in-group violence, and 32.3% (n=147) experienced only out-group violence. Of those who experienced any in-group violence, 69 (44%) experienced only in-group violence and the other 87 participants experienced mixed in- and out-group violence exposure. The race/ethnicity of the sample was 58.5% White (non-Hispanic), 14.1% Asian, 11.4% Hispanic or Latino, 9.0% Black or African American, 6.4% Multiracial, and 0.2% American Indian/Alaskan Native; 0.4% of participants chose not to indicate their race/ethnicity. Participants were also asked about the highest level of education any of their primary caregivers had attained. Approximately half of the sample had at least one caregiver with a bachelor’s or higher degree (48.3%). 78.6% of the sample had a caregiver who has attended at least some college, while 20.2% had at least one caregiver with a high school diploma/GED equivalent or less educational experience. 1.1% of the sample chose not to report the educational status of their parents. All participants reported living in the United States. The Institutional Review Board of Harvard University approved all procedures. Written informed consent was obtained from participants at the start of the study. All participants were compensated for completing the screener, and those who were invited to the full study were compensated an added sum upon completion of the full procedure.

Materials and Measures

Violence Exposure

Participants were initially screened for exposure to violence that occurred in childhood—defined as the period from birth to when they turned 18 years old—using 10 items selected from the Juvenile Victimization Questionnaire – Second Revision, Screener Sum Version (adult retrospective self-report). The JVQ is psychometrically-sound (α=.80) and widely used to assess exposure to various forms of violence that occurred in childhood (Finkelhor et al., 2005). Specifically, respondents were asked to endorse which among 10 given violent scenarios they had experienced or witnessed. Scenarios included experiencing violence perpetrated by family members or strangers (e.g., “Has someone ever hit or attacked you on purpose using an object or weapon?”), witnessing violence between family members (e.g., “Have you ever seen a parent or caregiver get pushed, slapped, hit, punched, or beat up by another parent, caregiver, or their boyfriend, girlfriend, or partner?”), and witnessing community violence (e.g., “Have you ever seen someone else get hit or attacked you on purpose using an object or weapon?”). Henceforth, we will refer to “experiences” or “exposure” to violence but please note that these terms reflect violence that was witnessed and/or experienced first-hand.

Relational Status of The Perpetrator

For each violent scenario participants endorsed experiencing in childhood, they were asked to indicate whether the person(s) who committed the violent act(s) described in each scenario were a member of their immediate family or an intimate partner. This operationalization for in-group violence was chosen based off definitions used in earlier work from our group (Kasparek et al, 2023). In addition, we opted to use a definition consistent with evolutionary theory to test these hypotheses using the most conservative approach. Families constitute the most proximate and evolutionarily conserved in-group because family groups are traditionally made up of close genetic relatives, and much evolutionary theory suggests that humans show the greatest degree of collaborative behavior, a hallmark of in-group relationships, with genetic relatives (Emlen, 1995; Hamilton, 1964). Even in cases of blended families, adoptive families, families with same-sex caregivers, and single-parent families, the family group is still the most proximate and significant in-group, particularly for early developmental needs and consequences (Lansford et al., 2001). In addition, romantic partners constitute another evolutionarily significant in-group given that pair-bonding and romantic love are universal mechanisms that promote cooperation and long-term commitment, which are essential for shared parenting and survival (Fletcher et al., 2015). Out-groups, by contrast, are simply groups with which one does not identify, though the reason for this lack of affiliation can vary from circumstantial (e.g., strangers who have never met) to more intentional based on experience (e.g., an “enemy”).

Based on their responses to these follow-up questions, participants were sorted into the control group (i.e., no violence exposure) or one of the two violence exposure sub-groups based on our pre-registered criteria. Specifically, anyone reporting any instance of violence perpetrated by members of their immediate family or an intimate partner were categorized as having experienced “in-group violence”. Of the 156 participants who met this criterion, 69 (44%) experienced only in-group violence and the other 87 participants experienced mixed in- and out-group violence exposure. We opted not to analyze in-group-only and mixed-violence groups separately. Our central hypothesis centered on whether exposure to in-group violence disrupts the formation of positive in-group schemas, making it important to retain anyone who experienced in-group violence—regardless of additional out-group experiences—as part of the in-group exposure group. This strategy aligns with our theoretical interest in whether any violations by in-group members are sufficient to disrupt core assumptions about safety and trust. Participants reporting only experiences of violence perpetrated by people who were not family or intimate partners were categorized as having experienced “out-group violence.”

Psychopathology

Participants who were invited to participate in the full study completed a short battery of well-validated and widely-used questionnaires assessing past-month symptoms of psychopathology, including the GAD-7 for anxiety symptoms (Spitzer et al., 2006); PHQ-9 (minus the suicidal ideation question) for depression symptoms (Kroenke et al., 2001); the Alcohol Use Disorders Identification Test (AUDIT) to assess problems related to alcohol use (Saunders et al., 1993); and the Brief Hypervigilance Scale to index hypervigilant tendencies in the wake of possible trauma (Bernstein et al., 2015). In addition, they completed the Buss-Perry Aggression Questionnaire to assess aggressive response styles and traits associated with aggressive tendencies across four profiles: physical aggression, verbal aggression, anger, and hostility (Buss & Perry, 1992). All psychopathology measures demonstrate adequate to excellent internal consistency, test-retest reliability, and convergent validity in original validation studies. Internal consistency for each measure in the current sample ranged from good to excellent: PHQ-9 (α=.89), GAD-7 (α=.92), AUDIT (α=.86), BHS (α=.85), and BPAQ (α=.91). The order in which participants completed these questionnaires was randomized to avoid any potential order effects. Psychopathology symptom scores were calculated by summing participant scores on each respective measure of psychopathology, consistent with common operationalizations of psychopathology in the field.

Minimal Group Assignment

To create and interrogate novel group affiliations and preferences, we used a minimal group assignment induction procedure and associated implicit association task, consistent with prior work (Kasparek et al., 2023; Lazerus et al., 2016). To induce minimal groups, participants were asked to complete the Ten-Item Personality Inventory or TIPI (Gosling et al., 2003). They were told that their responses would determine their team assignment based on a personality match to ensure team collaboration and cohesion. Participants were then randomly assigned to the Eagles or the Rattlers team and instructed that they could identify Eagles team members by the blue shirts they were wearing, and that the Rattlers could be identified by their green shirts. Next, participants read a vignette explaining that the goal of the entire study was for them to work with their team to complete a problem-solving task. They were then told that “science has shown that in order to perform well on the task, they must get to know the team members on both teams.”

Implicit In-Group Favoritism

An online version of the Implicit Association Test (IAT) developed by Carpenter and colleagues (2019) for use with Qualtrics was then used to assess participants’ implicit in-group favoritism. The IAT is a timed, computerized sorting task in which the participant is instructed to sort stimuli as quickly as possible using two response buttons on a keyboard. The fundamental principal of the IAT is that it is easier for a participant to sort stimuli that are paired based on a previously held mental association (e.g., congruent) than when stimuli are paired in a way that does not match previously held associations (e.g., incongruent). Thus, participants will respond faster to congruent trials. For example, in the context of the present study, most people would have in-group members mentally linked with positive words or concepts, and out-group members with negative words or concepts, and would therefore sort stimuli into those category pairs rather quickly. Conversely, participants would find it more difficult to sort stimuli into pairs that are not mentally associated as readily (i.e., in-group with negative concepts and out-group with positive concepts) and would therefore respond more slowly. A greater difference in speeds between the two trial types suggests a stronger association between the items in the easier task, or in the case of the present study, stronger in-group favoritism.

IAT Procedure.

Participants were asked to sort team members based on their team affiliation, which was signaled via the shirt-color of each stimulus (blue for Eagles and green for Rattlers), consistent with prior work (Kasparek et al., 2023; Lazerus et al., 2016). The IAT consisted of seven blocks, each making up sets of trials. In each trial, participants saw an image of a new Rattlers or Eagles team member with a neutral affective expression on the screen. The face stimuli used in the task were adapted from the FACES stim set, created by Ebner and colleagues (2010), the same images used by Lazerus and colleagues (2016). There was an even split between male and female identities and only “young” adult stimuli were used to ensure they were as close in age as possible to the age of participants in the present study. The average perceived age of the young adult stimuli was 28.5 years old (Ebner et al., 2010). Stimuli represented “targets” (e.g., Rattlers or Eagles team members) or the category (e.g., positive-negative). When stimuli appeared, participants were told to sort them as rapidly as possible by pressing the “E” key on their computer keyboard with their left index finger, or the “I” key with their right index finger. The key components of the IAT are the combined blocks, in which participants sorted stimuli according to contrasting dimensions (see Supplemental Figure 1). The congruent block (e.g., in-group + positive words; out-group + negative words) is presumed to align with participants’ associations, while the incongruent block (e.g., out-group + positive words; in-group + negative words) is presumed to conflict with their associations.

Throughout the procedure, participants underwent practice trials for each trial-type to familiarize themselves with the task (see Supplemental Figure 2). These practice trials helped minimize potential order effects and ensure participants understood the sorting procedure. To further minimize order effects, the order of presentation for different conditions was counterbalanced across participants. Left/right starting positions for targets and categories were also counterbalanced, with variations in the initial combined block presentation. Identities presented were counterbalanced such that a given participant was equally likely to see each identity as either an Eagles or Rattlers team member. In addition, this version of the IAT corrected participants when they paired a given stimulus with the incorrect label. This process helps ensure that the participants were learning the correct pairings over time, resulting in a more valid estimate of their implicit bias by the end of the task. The full IAT procedure took approximately twelve minutes to complete and included practice trials familiarizing participants with target stimuli (e.g., Rattlers or Eagles team members), category stimuli (e.g., positive, or negative words), and combined blocks. The procedure also incorporated a 40-trial practice block where participants practice sorting categories in a reversed position to minimize interference from prior learning. The speed with which participants responded across trials was measured in milliseconds.

Internal Consistency.

The De Houwer and De Bruyker (2007) procedure was used to test internal consistency of the IAT used in the present study. First, trials were sorted by target/category and then scored separately based on odd/even trials (and correlating the two, with a split-half spearman-brown correction). Using this method, the internal consistency of the task as administered in the present study was .84. For comparison, IAT reliabilities tend to fall between .70 and .90 (Hofmann et al., 2005)

IAT D-Score.

Consistent with scoring conventions (Greenwald et al, 2003), we dropped individual trials with response times >10,000ms and <300ms. In addition, we eliminated participants with >10% trials <300ms. As a result of these data cleaning steps, 4 participants and an additional .49% of task trials were excluded from analyses. Trials from combined test blocks (e.g., pairing good/bad words, Rattlers/Eagles team names, and pictures of Rattlers/Eagles team members) were used to create a test score (D; Greenwald et al., 2003). Specifically, the difference in average speed across these combined test blocks was divided by the pooled standard deviation for that pair of blocks for each participant. Positive D indicates stronger associations of the in-group with good (i.e., in-group favoritism). Negative D-Scores indicate stronger association of the out-group with good (i.e., out-group favoritism). A D-Score of 0 indicates no difference in associations of either group with good or bad (i.e., ambiguous or no preference).

Explicit Intergroup Favoritism

At the end of the study, participants were asked 6 questions aimed at determining their explicit preferences for their in-group compared to the out-group: I [value/like/feel connected to] the [Eagles/Rattlers]. Participants used a sliding scale ranging from 0 (completely disagree) to 100 (completely agree) to indicate their alignment with each of the 6 statements. In-group and out-group favoritism scores were calculated for each participant based on whether they were assigned to the Eagles or Rattlers team. This was done by averaging their ratings from the 3 questions they answered about their in-group and the 3 questions that they answered about the out-group, respectively. Cronbach’s alpha for in-group and out-group explicit favoritism scores in the present sample was .91 and .86 respectively, reflecting good to excellent internal consistency.

Demographics

Demographic questions were included at the end of the study to prevent any unintended priming effects. Participants were asked to indicate their sex assigned at birth, gender identity, race/ethnicity, nationality, languages spoken, and age. Finally, we collected information to help determine each participant’s socioeconomic status, such as the highest level of education attained by their primary caregiver(s), whether their childhood home was owned or being bought vs. rented, and variables to help us to determine how crowded their childhood home was.

Manipulation & Attention Checks

We included a series of manipulation checks to help ensure the validity of our data. We posed several questions throughout the study requiring participants to identify which team they had been assigned to and the color associated with that team as a check to ensure they were paying attention and remembered this important detail throughout.

We also included attention checks in both the screener and the main study. These attention checks were embedded within questionnaires throughout the study and instructed participants to select a certain response from the range of Likert-response options for a given question. For example, in the middle of a psychopathology questionnaire with response options ranging from strongly disagree to strongly agree, a participant might be instructed to “Please select Neither Agree nor Disagree.” These attention checks ensured that participants were adequately engaging with study materials. Submissions from participants who failed any attention checks were rejected and their data were not included in analyses.

Analysis Plan

All analyses controlled for age and sex, and analyses with implicit bias as an outcome controlled for participants’ randomly assigned group as well. Our childhood SES proxy variable, highest level of education attained by a primary caregiver, differed between violence exposure subgroups, F(2,447)=4.93, p=.01. Specifically, those who reported not experiencing violence in childhood reported higher average educational attainment among primary caregivers compared with the in-group exposure subgroup (difference=.46, p=.01) and the out-group exposure subgroup at a trend level (difference=.45, p=.09). As such, analyses including violence exposure or perpetrator status as a primary predictor were run with SES as a covariate as a sensitivity analysis. These additional sensitivity analyses were not pre-registered. Results were unchanged when controlling for SES (see Supplemental Tables 1–4).

Linear regression was used for analyses in which the dependent variable was normally distributed. In cases where the dependent variable was non-normally distributed (e.g., zero-inflated and/or right-skewed) count data, we examined residuals and consulted the Bayesian Information Criterion (BIC) to select the proper distribution family and related regression type. Specifically, for analyses in which depression, anxiety, hypervigilance, or alcohol use disorder (AUD) symptoms were the dependent variable, negative binomial or zero-inflated negative binomial regression were used according to whichever model had the lowest BIC score, indicating better model fit. The decision to use BIC for model selection represents a deviation from the pre-registered analytic plan, which specified the use of likelihood ratio tests. This adjustment was made to take advantage of BIC’s broader applicability across models and its more conservative penalization in larger samples. Results from models estimated using zero-inflated negative binomial regression feature coefficient estimates and p-values for both the conditional model, and the zero-inflated model. Estimates for the conditional model describe the expected count of the dependent variable given the predictors; estimates for the zero-inflation model describe the log odds of excess zeros in the dependent variable, given the predictors, compared to what would be expected from a standard negative binomial distribution. Odds ratios (ORs) are provided for statistically significant zero-inflation estimates to facilitate easier interpretation of effects. ORs above one indicate increased odds of zeros in the dependent variable with each unit increase of a given predictor, while ORs below one indicate decreased odds of zeros in the dependent variable. We applied False Discovery Rate (FDR) correction for multiple comparisons and report corrected p-values when needed; however, this procedure was not specified in our pre-registration.

Results

Descriptive Statistics

Means, standard deviations, and intercorrelations for variables of interest are presented in Table 1.

Table 1.

Descriptive Statistics & Intercorrelations

	No Violence (n = 152)		In-Group Violence (n = 156)		Out-Group Violence (n = 147)
	M(%)	SD	M(%)	SD	M(%)	SD
1. Sex (Female)	53%	.50	51%	.50	34%	.48
2. Age	22.28	1.95	22.29	2.10	22.45	1.91
3. Highest Caregiver Education	4.27	1.48	3.71	1.47	3.91	1.44
4. Number of Violent Experiences	0	0	3.92	1.86	2.12	1.12
5. Team Affiliation (Eagles)	47%	.50	51%	.50	50%	.50
6. IAT D-score	.27	.44	.29	.46	.40	.47
7. Explicit In-Group Favoritism	59.55	21.24	59.42	21.43	59.45	20.29
8. Explicit Out-Group Favoritism	39.72	18.29	41.44	19.22	40.28	17.90
9. Depression Symptoms (PHQ-9)	6.63	5.53	10.06	5.85	8.67	5.52
10. Anxiety Symptoms (GAD-7)	6.09	5.04	8.96	5.46	7.77	5.17
11. Hypervigilance Symptoms (BHS)	4.95	4.02	7.90	4.78	6.52	4.54
12. Alcohol Use Disorder Symptoms (AUDIT)	4.07	5.54	3.58	4.42	3.40	3.73
13. Aggression Symptoms (BPAS)	62.31	17.37	77.79	17.44	71.36	17.66

Correlations
	1.	2.	3.	4.	5.	6.	7.	8.	9.	10.	11.	12.	13.
1.	-
2.	−0.04	-
3.	−0.12**	0.13**	-
4.	0.11*	0.04	−0.19***	-
5.	−0.09	0.02	0.05	0.01	-
6.	0.05	0.01	0.08	−0.01	−0.06	-
7.	−0.04	0	0	−0.01	0.06	0.02	-
8.	0.02	−0.04	−0.02	0.03	−0.04	−0.05	0.39***	-
9.	−0.18***	0	−0.07	0.25**	0.03	−0.05	−0.07	−0.06	-
10.	−0.21***	0	−0.06	0.22**	−0.01	−0.06	0.02	−0.04	0.8***	-
11.	−0.13**	−0.04	−0.07	0.29**	0.04	−0.02	−0.02	−0.05	0.52***	0.61***	-
12.	−0.03	0.09*	0.16***	−0.02	0.01	−0.09	0.02	−0.05	0.07	0.1*	−0.04	-
13.	0.03	−0.05	−0.13**	0.32***	0.04	−0.05	−0.07	−0.1*	0.48***	0.46***	0.44***	0.13**	-

Note:

^*p < .05;

^**p < .01,

^***p < .001

Violence Exposure and Psychopathology

Childhood violence exposure was associated with greater symptoms of depression (conditional: b= .26, p < .001; zero-inflation: b=−1.79, p < .001, OR=.17); anxiety (conditional: b=.27, p < .001; zero-inflation: b=−1.32, p=.004, OR=.27); hypervigilance (b=.42, p < .001); and aggression (b=.18, p < .001), but not with symptoms of AUD (conditional: b=−.17, p=.20; zero-inflation: b=−.08, p=.83) (see Figure 1).

Figure 1.

Associations between violence exposure and a) depression, b) anxiety, c) alcohol use, d) hypervigilance, and e) aggression symptoms. Each point in the shaded region represents a single observation. The shaded grey area and the curve along the outer edges represent a smoothed density curve showing the full distribution of the data. The black line in the middle of the shaded region is the mean and the white band above and below the black line signifies the 95% confidence interval.

Implicit and Explicit Bias for Novel Groups

Across conditions, participants exhibited moderate-to-strong implicit bias favoring their novel in-group members after minimal group assignment (D-score=0.32, t(454)=14.92, p < .001, 95% CI=[.27, .36]; Cohen’s d=0.70). Participants similarly exhibited very strong explicit bias favoring their in-group (M=59.31 95% CI=[57.37, 61.24]) relative to the outgroup (M=40.43, 95% CI=[38.73, 42.13]) across conditions, t(454)=18.45, p < .001, 95% CI=[16.30, 21.45]; Cohen’s d=0.95.

Violence Exposure and Implicit and Explicit Bias

Childhood violence exposure, without considering the relational status of the perpetrator to the participant, was not associated with implicit bias for one’s randomly assigned group (ß=.07, p=.11) nor explicit bias favoring the in-group (ß=−.004, p=.93) or the out-group (ß=.02, p=.62).

Relational Status of the Perpetrator and Implicit and Explicit Bias

The relational status of the perpetrator at the time of childhood violence exposure was associated with implicit bias (see Figure 2). Specifically, participants who experienced solely out-group violence in childhood exhibited more favoritism for their novel in-group relative to those who experienced no childhood violence (ß=−.13, p=.01) and those who experienced any in-group violence in childhood (ß=−.11, p=.04); there was no significant difference in implicit bias between participants who experienced any in-group violence and those who did not experience violence (ß=.02, p=.69).

Figure 2.

Association of violence exposure group (i.e., relational status of the perpetrator) with a) implicit bias and explicit bias for novel b) in-groups and c) out-groups. Each point in the shaded region represents a single observation. The shaded grey area and the curve along the outer edges represent a smoothed density curve showing the full distribution of the data. The black line in the middle of the shaded region is the mean and the white band above and below the black line signifies the 95% confidence interval.

Relational status of the perpetrator was not associated with explicit bias favoring novel in-group members (out-group vs. no violence: ß=−.0004, p=.995; out-group vs. any in-group violence: ß=−.01, p=.87; no violence vs. any in-group violence: ß=−.01, p=.88) or out-group members (out-group vs. no violence: ß=−.01, p=.88; out-group vs. any in-group violence: ß=.03, p=.61; no violence vs. any in-group violence: ß=.04, p=.50) (see Figure 2).

Implicit Bias and Psychopathology

Implicit bias was associated with greater AUD symptoms (conditional: b=−.33, p=.02; zero-inflation: b=−.28, p=.48), such that participants exhibiting more ambiguous preferences or out-group favoritism exhibited greater AUD symptoms (see Figure 3). Implicit bias was not associated with depression (conditional: b=−.08, p=.22; zero-inflation: b=−.13, p=.78); anxiety (conditional: b=−.05, p=.46; zero-inflation: b=.93, p=.13); hypervigilance (conditional: b=−.06, p=.40; zero-inflation: b=−1.98, p=.33); or aggression (b=−.03, p=.32) symptoms. The negative association of implicit bias with AUD symptoms remained significant after controlling for violence exposure as a sensitivity check (conditional: b=−.33, p=.02; zero-inflation: b=−.31, p=.43).

Figure 3.

Association of implicit bias D-Score with Alcohol Use Disorder (AUD) symptoms. Each point represents a single observation. The solid black line represents the line of best fit based on predictions from the zero-inflated negative binomial model. The shaded grey region shows the 95% CI.

Violence Exposure, Relational Status of the Perpetrator, Implicit Bias, and Psychopathology

Given that the mediator variable (implicit bias) and outcome variables (psychopathology symptoms) reflect processes measured at the same timepoint, we present the following results as exploratory tests of indirect effects to be further studied using appropriate longitudinal designs. With that in mind, we identified a moderated mediation, such that the association of childhood violence severity with AUD symptoms was mediated by implicit bias among participants who experienced solely out-group violence in childhood (b=−.13, 95% CI=[−.45, −.01]), but not among those who experienced any in-group violence (b=−.01, 95% CI=[−.13, .05]) (see Figure 4). This was due to a significant moderation of the a-path and trend-level moderation of the b-path by perpetrator status. Specifically, relational status of the perpetrator moderated the association of violence severity with implicit bias (ß=−.30, p=.01), such that more experiences of violence were associated with increased in-group favoritism among participants who experienced out-group violence in childhood (b=.08, 95% CI=[.02, .13], p=.01); there was not a significant association of the number of experiences of violence with implicit bias among participants who experienced any in-group violence (b=−.02, 95% CI=[−.06, .01], p=.15) (see Figure 5). Relational status of the perpetrator moderated the association of implicit bias with AUD symptoms at a trend level (conditional: b=.59, p=.07; zero-inflation: b =.60, p=.51). Ambiguous preferences or greater out-group favoritism were associated with more AUD symptoms among participants who experienced solely out-group violence in childhood (b=−1.87, 95% CI=−3.68, −.05], p=.04); there was not a significant association of implicit bias with AUD symptoms among participants who experienced any in-group violence (b=−.24, 95% CI=[−2.04, 1.56], p=.79) (see Figure 5).

Figure 4.

Moderated mediation model showing differential mediation effect of childhood violence severity on alcohol use disorder symptoms through implicit bias depending on whether the violence came from an in-group or out-group source.

Note: *p < .05, ^†p < .1

Figure 5.

Association of a) the number of violent experiences with implicit bias and b) implicit bias with AUD symptoms by in- and out-group perpetrator status. Each point represents a single observation. Lines and related 95% CIs represent the predicted fit by perpetrator status based on the interaction model. Note: * p < .05

Supplemental Analyses & Results

In-Group Only Violence Exposure Isolation & Comparison

We ran supplemental analyses that were not pre-registered to clarify our results related to relational status of the perpetrator. Specifically, we reanalyzed the IAT data separating participants who experienced solely in-group violence from those who experienced mixed in- and out-group violence. We found that participants who experienced solely out-group violence exhibited more favoritism for their novel in-group relative to those who experienced no childhood violence (ß=−.13, p=.01); participants who experience solely out-group violence showed greater in-group favoritism than those who experienced violence from both in- and out-group sources at a trend-level (ß=−.10, p=.06). There was no difference in implicit bias between participants who experienced solely out-group violence and solely in-group violence (ß=.03, p=.61), mixed violence and no violence (ß=−.01, p=.92), or solely in-group violence (ß=.02, p=.74). Finally, there was no difference between participants who experienced solely in-group violence and no violence (ß=−.03, p=.61) (see Supplemental Figure 3). However, these results should be interpreted with caution given that the sample sizes for the solely in-group and mixed violence exposure subgroups are small relative to the no exposure and solely out-group exposure groups, and relative to what our power analysis indicated was required for our pre-registered analyses.

Decomposed D-Scores Elucidating Implicit Valence Preferences

To better understand the nature of participants’ implicit evaluations, we conducted additional analyses. Standard IAT scores capture whether participants respond more quickly when the in-group is paired with positive words and the out-group with negative words, compared to the opposite pairings. While this indicates the relative strength of in-group versus out-group associations, it cannot reveal whether a more neutral score reflects participants viewing both groups positively, both groups negatively, or a combination of positive and negative associations that cancel each other out.

To address this limitation, we created “decomposed” IAT scores (see O’Shea et al., 2020). This method separates response patterns into distinct indices for how positively or negatively participants implicitly evaluate their in-group and their out-group. For each participant, we first identified the trials in which their in-group was paired with positive words, their in-group was paired with negative words, their out-group was paired with positive words, and their out-group was paired with negative words. We then calculated the average RT for each trial type. Longer RTs indicate that the pairing was less intuitive or more difficult for the participant to process, whereas shorter RTs suggest the pairing was more intuitive. See Supplement for details.

Decomposed D-Scores revealed a significant association of relational status of the perpetrator with implicit evaluations of in-group and out-group. For in-group valence, participants exposed to out-group violence exhibited the strongest implicit positivity toward their in-group (M=.46, SE=.05), compared to those exposed to in-group violence (M=.31, SE=.05) or no violence (M=.29, SE=.05), F(2, 443)=3.95, p=.02 (see Supplemental Figure 4a). For out-group valence, participants with out-group violence exposure exhibited the strongest implicit negativity toward the out-group (M=−.42, SE=.04), compared to those with in-group violence exposure (M=−.31, SE=.04) and no violence (M=−.27, SE=.04), F(2, 443)=3.26, p=.04 (See Supplemental Figure 4b).

Discussion

Here we found that the relationship of the perpetrator to the participant at the time of childhood violence exposure predicted differential patterns of expression of implicit bias for novel groups in young adulthood. Specifically, participants who experienced only out-group violence showed more favoritism for novel in-group members than those who experienced any in-group violence and those who experienced no violence. There were no differences in explicit bias between these groups. Weaker in-group favoritism was associated with greater alcohol use disorder (AUD) symptoms over the past month, controlling for the effect of violence exposure. Violence exposure was associated with increased AUD symptoms through implicit bias, but only among participants who experienced solely out-group violence. These findings, together with prior work from our group (Kasparek et al., 2023), suggest that implicit bias is a possible mechanism linking childhood violence exposure with psychopathology. In addition, they encourage further exploration of specific contextual features of early violent experiences, such as perpetrator group, as potential modulating factors of associations of violence exposure with implicit bias expression and psychopathology risk.

Violence Exposure, Relational Status of The Perpetrator, and Bias for Novel Groups

Minimal group assignment produced heightened implicit and explicit in-group favoritism, such that participants were more likely to implicitly associate their randomly assigned group with positive concepts (“good”) compared to the out-group and displayed greater explicit preferences for their in-group as well. These findings are consistent with prior work in children and adults (Cvencek et al., 2011; Cvencek et al., 2016; Dunham et al., 2011; Dunham & Emory, 2014) and suggest that the minimal group assignment induction was successful in the present study.

Childhood violence exposure, combined across in-group and out-group exposure, was not associated with implicit favoritism. However, when examined by relational status of the perpetrator to the participant at the time of violence exposure, significant differences in implicit bias emerged. Specifically, participants who experienced only out-group violence in childhood showed stronger favoritism for their novel in-group members compared to participants who experienced no violence or any instance of in-group violence in childhood. The direction of this finding was consistent with what we hypothesized; however, we did not find a significant association of in-group violence with reduced in-group favoritism compared to the no exposure group, as hypothesized. Neither violence exposure nor relational status of the perpetrator of violence were associated with explicit bias; thus, we can reasonably conclude that this was a purely implicit effect below the level of conscious awareness.

As previously hypothesized (Kasparek et al., 2023), the relational status of the perpetrator of violence to the participant is associated with differential expressions of implicit bias for novel groups, even into young adulthood. Stronger affiliations of in-groups with positive concepts and out-groups with negative concepts after out-group violence exposure may ultimately serve as a protective mechanism scaffolding stronger in-group bonds. Indeed, as previously discussed, strong in-group bonds confer many benefits, including fostering a sense of belonging, offering safety, and supporting wellbeing (Baumeister & Leary, 1995; Boyd et al., 2011; Spoor & Kelly, 2004; Skinner & Meltzoff, 2019; Tajfel & Turner, 1979; Tomasello et al., 2012). Moreover, stronger associations of out-group members with negative concepts may subserve avoidance of out-group members who may represent threat or activate related schemas. This is consistent with prior work showing heightened hypervigilance-related symptoms among people who experience extra-familial violence (Smith et al., 2019). Though this hypervigilance may be adaptive in certain contexts, prior work has also shown that worries about danger and threat of harm following experiences of violence are an important mechanism linking interpersonal violence with risk for psychopathology and future revictimization (Jaffe et al., 2019). Future research is needed to better characterize whether a stronger desire to seek in-group safety or a stronger desire to avoid out-group violence is driving this effect to further clarify this risk pathway.

We did not find a significant difference in implicit bias between participants who experienced any instance of in-group violence in childhood and those who did not experience violence. The absence of any distinct effect of in-group violence on implicit bias may signify confusion regarding who is likely to be safe or trustworthy due to prior experiences that violate expectations of safety from in-group members. More specifically, it may be that those who experience in-group violence “lose” their implicit sense that in-group members are preferable or safe. Though this may be adaptative when participants were in a family environment where violence was likely, it may be maladaptive later in life as they try to build new relationships and form a new sense of family.

Decomposing D-Scores to Unpack Bias Trends

Given these findings, we conducted supplementary analyses decomposing IAT D-scores to better clarify the motivational processes underlying these biases. Out-group violence was associated with enhanced implicit in-group positivity and stronger implicit out-group negativity, suggesting that early out-group threat may strengthen both implicit affiliative and defensive motivational processes. Conversely, participants who experienced in-group violence or no violence exhibited muted decomposed scores in both directions and did not differ from one another, even though both groups showed small, positive overall D-scores. This pattern suggests that a single experience of in-group violence may be insufficient to produce meaningful changes in implicit intergroup motivation and that minimal overall bias in these two groups reflects generally weak implicit responses rather than differences in underlying motivational processes.

Other Possible Explanations

This discrepancy may be attributable to methodological differences between the present and prior studies. The composition of the in-group violence samples between the two studies is one key difference. In the prior study, most participants experienced or witnessed intra-familial violence only. In the present study, in-group violence was operationalized as intra-familial but also included intimate partner violence given the older age of our participants and social re-orientation toward peers in adolescence and adulthood. In addition, we included any participant who experienced at least one instance of in-group violence in the in-group violence exposure subgroup for analysis, regardless of whether they experienced violence from other sources as well. As such, 56% (n=87) of the participants who were categorized as experiencing any in-group violence in the present study also experienced out-group violence. Thus, it may be that past experiences of violence from mixed sources diminished the effect of in-group violence on implicit bias in the present study.

To address this question, we conducted supplementary analysis comparing those who experienced solely in-group violence to those who experienced mixed violence, solely out-group violence, and no violence. Participants who experienced solely out-group violence exhibited stronger implicit in-group favoritism compared to those with mixed violence exposure (trend-level) and those with no violence exposure. However, implicit bias did not differ in any of the other contrasts – most notably, there was no difference between the solely in-group and mixed violence exposure groups. These findings provide preliminary evidence that the inclusion of mixed-exposure participants in the broader in-group violence category may not be the primary reason for the lack of significant effects in this group. Nonetheless, given the relatively small subgroup sizes, future work should aim to compare implicit bias expressions for novel group members among comparable sample sizes of participants who experienced solely out-group violence, mixed violence, and solely in-group violence. In addition, different definitions of in-group violence should be explored to discern whether potential effects of in-group violence vary by perceptions of relational closeness, for example.

A second major difference in these samples is age. Participants in the prior sample were between five and six years old at the time in-group violence exposure was assessed. In addition, implicit bias for novel groups and baseline psychopathology were measured at ages seven and eight, and psychopathology was reassessed at ages nine and ten. In the present sample, violence exposure could have occurred anytime between birth and age 18 years old, and implicit bias and psychopathology were assessed concurrently among participants ranging in age from 18 to 25 years old. Thus, the samples reflect associations of violence exposure, implicit bias, and psychopathology risk at fundamentally different stages of development. Considering this context, it may be that intrafamilial in-group violence is particularly impactful at younger developmental stages, perhaps because younger children are dependent on family for safety and survival needs. Intrafamilial in-group violence may no longer be as salient a factor impacting expressions of implicit bias in young adulthood when people have greater agency over choosing in-groups and moving between them as needed to achieve optimal levels of support and safety. Similarly, perhaps early experiences of out-group violence remain salient motivators of implicit bias throughout life given that humans are already pre-disposed to view out-group members more negatively, trust out-group members less, and expect out-group members to violate social norms to a greater extent than in-group members (Liberman et al., 2018). In this sense, prior experiences of out-group violence may thus confirm and strengthen hypervigilance about out-groups throughout life. Future work is needed to better characterize the developmental unfolding of implicit bias in the context of violence exposure from in- and out-group sources, and to better map alterations in bias with risk for psychopathology.

We did, however, show a significant interaction between the number of violent experiences and relational status of the perpetrator of violence in predicting implicit bias. Specifically, for participants who experienced out-group violence, more prior experiences of violence predicted stronger in-group favoritism (see Figure 5), consistent with our hypotheses. Violence severity and perpetrator status did not significantly interact in the prediction of implicit bias among those who experienced any in-group violence, though the slope trended in the hypothesized direction of weakened in-group favoritism. Thus, out-group violence may result in enhanced favoritism for novel in-group members regardless of the level of prior exposure. Given that humans are evolutionarily primed to more readily associate out-groups with negative concepts, even one experience of out-group violence may strengthen implicit in-group preferences given stereotype-congruency. Conversely, exposure to in-group violence may only weaken in-group favoritism at high rates of prior exposure. This, too, fits with an evolutionary perspective given that humans are primed to robustly favor in-groups, even when there are few similarities, as previously discussed. Such an engrained bias may be resilient to dampening effects, especially given incongruency between our implicitly held beliefs about in-groups and negative or harmful experiences with evolutionarily conserved in-groups (i.e., family). Thus, future work is needed to determine if the effect of in- and out-group violent experiences on implicit bias adheres to a dose-response vs. mere-exposure pattern, respectively, as proposed.

Implicit Bias and Psychopathology

We found mixed support for our hypothesis that implicit bias would be associated with transdiagnostic risk for psychopathology. Specifically, weaker in-group favoritism (i.e., ambiguous group preference or out-group favoritism) was associated with greater alcohol use disorder symptoms (AUD), but not depression, anxiety, hypervigilance, or aggression symptoms. Ambiguous implicit preferences for novel groups or an implicit preference favoring the out-group is a departure from the normative, evolutionarily conserved propensity to favor novel-ingroup members after minimal group-assignment, which is considered protective (Cvencek et al., 2016; Dunham et al., 2011). As previously suggested (Kasparek et al., 2023), it may be that participants who express out-group favoritism or ambiguous preferences for novel groups have a harder time forming or maintaining strong in-group bonds. Given that in-group bonds are necessary for safety, support, and emotional wellbeing, the absence of such bonds may contribute to conditions that give rise to psychopathology. Indeed, we show in prior work that reduced in-group favoritism predicted increased internalizing symptoms over time in young children (Kasparek et al., 2023).

Given that this study was cross-sectional, increased AUD symptoms among this subgroup could reflect several processes: 1) participant efforts to cope with a relative lack of strong in-group bonds following violence exposure; or 2) preferences for the out-group may indicate a less risk-averse profile that would be more likely to facilitate heightened use of alcohol and other substances. Importantly, we additionally found that implicit bias and perpetrator status interacted to predict AUD symptoms. Specifically, greater out-group favoritism was associated with more AUD symptoms among participants who experienced solely out-group violence in childhood but not among those who experienced any in-group violence. This finding suggests that those who experience out-group violence and nevertheless favor the out-group (or don’t show strong favoritism for the in-group) use alcohol to a greater extent. Favoring the out-group, especially after prior threatening experiences from out-group members, may be a proxy for increased risk-taking behavior or a lack of caution, which may relate to increased substance use. Future longitudinal work is needed to disentangle if out-group favoritism precedes AUD symptoms to clarify this effect. Future work should also measure trait disinhibition and other variables indexing risk-taking proclivities to test if out-group favoritism is a proxy for risk-taking.

We did not find associations of implicit bias with other forms of psychopathology assessed despite uncovering these associations in prior work (Kasparek et al., 2023). There may be several reasons for this discrepancy. Reduced in-group favoritism was associated with increased internalizing psychopathology over time in a sample of children in prior work (Kasparek et al., 2023). As previously mentioned, the age of the sample in the present study is much older. As such, perhaps alterations in implicit bias owing to childhood violence are sufficient to increase risk for mental health difficulties among pre-adolescents, but not sufficient to produce differences in psychopathology manifesting in or persisting into young adulthood. In addition, the present study was cross-sectional. Differences in implicit bias may be helpful in predicting change in psychopathology over time but may not be a sensitive correlate of concurrent psychopathology, particularly among young adults. Future work should assess whether implicit bias for novel groups in young adults predicts changes in psychopathology over time.

We found preliminary support for our hypothesis that implicit bias would mediate the association of violence exposure with psychopathology. We identified a moderated mediation, such that implicit bias mediated the association of violence with AUD symptoms only among participants who experienced solely out-group violence. Specifically, more experiences of violence were associated with increased in-group favoritism among participants who experienced out-group violence, and those who experienced out-group violence but nevertheless exhibited greater out-group favoritism reported greater AUD symptoms. Importantly, however, this mediation is cross-sectional. Though participants reported on perpetrator status and violent experiences occurring prior to the study, both implicit bias and alcohol use disorder symptoms were measured concurrently. As such, interpretations are advanced with caution, and future longitudinal work is needed to substantiate implicit bias as a potential mechanism of psychopathology risk.

As discussed, participants who experience out-group violence may wish to integrate more deeply with in-groups or avoid out-groups to a greater extent as a safety mechanism. Weaker in-group favoritism or out-group favoritism after out-group violence may reflect difficulty learning from past experiences, which may result in lower levels of inhibition or higher levels of risk-approach. In turn, alcohol and other substances may more readily be used, perhaps to cope with past violent experiences or a relative lack of strong in-group bonds. This interpretation is plausible given the finding that participants who experienced out-group violence and showed more out-group favoritism in the present study reported greater alcohol use.

Limitations & Future Directions

The present study has several strengths, including a large and diverse sample, methodological consistency with prior research in the assessment of implicit intergroup bias, and rigorous pre-registration of hypotheses, variable definitions, and analytic plans. However, several limitations warrant consideration. Most notably, the study’s cross-sectional design limits causal inference. Although participants retrospectively reported violence exposure occurring prior to 18 years of age, both implicit bias and psychopathology symptoms were assessed concurrently. As such, the directionality of relationships among childhood violence exposure, implicit biases, and psychopathology remains uncertain. Future longitudinal research is essential to clarify the temporal ordering and directionality of these associations.

Additionally, the operationalization of perpetrator status poses limitations. Specifically, the “out-group violence” category included only participants who experienced violence exclusively from out-group sources, while the “in-group violence” category included participants who experienced violence from in-group sources but did not exclude those with mixed violence exposure (in-group and out-group sources). Though our supplementary analyses suggest minimal bias differences between mixed-exposure and exclusively in-group exposure groups, future studies should explicitly compare implicit biases across comparable groups that have experienced solely out-group violence, solely in-group violence, and mixed sources of violence exposure. Further, alternative operationalizations of in-group violence, such as perceptions of relational closeness or betrayal, warrant exploration.

Conclusion

This study underscores the significance of perpetrator relational status in shaping implicit biases and associated psychopathology risk following childhood violence exposure. Out-group violence, in particular, robustly predicts stronger implicit favoritism toward novel in-groups, reflecting potentially adaptive threat responses involving both increased in-group affiliation and heightened out-group negativity. Deviations from this patterns, such as out-group favoritism after out-group violence exposure, are associated with elevated alcohol use disorder symptoms. These findings enhance our understanding of how contextual and developmental factors shape implicit biases and their links to psychopathology. Such insights are critical for identifying at-risk individuals and tailoring interventions to address distinct patterns of implicit bias associated with different violence exposure histories.