Future Directions in Aggression Research: The Contribution of Technology-Integrated Operationalizations

Naturalistic Stimuli of Aggressive Interactions

Rather than directly measuring interpersonal aggressive behavior, some researchers have investigated how people process and evaluate stimuli depicting aggressive interactions by assessing emotional responses, cognitive evaluations, and attentional biases related to aggression. Historically, aggression and anger research has relied on static facial images (e.g., Blair et al., 1999), especially processing of angry morphed faces (see, e.g., Penton-Voak et al., 2013; Stein et al., 2024; Wilkowski & Robinson, 2012), as well as hypothetical stories (Crick, 1995) and static, simple depictions of social scenarios (such as drawings) as social stimuli (Van Dijk et al., 2019); see Figures 1(A)−(C) for examples. This approach could limit participants’ perspective-taking abilities and provide representations that are far removed from reality. Static angry faces, compared to dynamic angry faces, are processed earlier and may lead to attentional bias (Qu et al., 2023). In contrast, dynamic stimuli enhance arousal and are more ecologically valid by simulating real-life scenarios in a more naturalistic and realistic way (Fehr et al., 2014) and better reflect emotional states (Qu et al., 2023). New technologies have facilitated recent advances in this field by enabling the implementation of more realistic and emotionally engaging stimuli that depict violent conflicts or aggressive interactions (e.g., Hernandez Pena et al., 2024b; Nummenmaa et al., 2008; Van Den Stock et al., 2015; Van Dongen et al., 2018); see examples in Figures 1(E) and (F). The majority of these studies have investigated the processing of violent scenes. Many studies thereby investigated passive processing of stimuli depicting aggressive interactions (e.g., strangling, punching, or threatening scenes) versus neutral social interactions (Nummenmaa et al., 2021; Rantanen et al., 2021; Van Dongen et al., 2018). Others explored reactions toward video clips shown from the aggressor’s first-person perspective (Fehr et al., 2014). In some studies, the victim or the perpetrator perspective was compared; see Figure 1(F) (Hernandez Pena et al., 2024b; Nummenmaa et al., 2008; Taubner et al., 2021; Van Den Stock et al., 2015). More directly investigating aggressive reactions, video clips portraying unfair situations, such as bullying, mocking, or accounts of child abuse, have been used to induce anger in adult participants (Seok & Cheong, 2020). Adapted versions have been developed to study peer victimization in children featuring child actors (Troop-Gordon et al., 2018).

Video Games as Measures of Aggression

Numerous studies, reviews, and meta-analyses have examined the potential association between violent video games and aggressive behavior (e.g., see Burkhardt & Lenhard, 2022; Prescott et al., 2018). More recently, video games have also been included to measure aggressive behavior. The use of video games to measure aggression probably started in the early 2000s, when Carnagey and Anderson (2005) employed a race-car video game to assess both rewarded and punished aggression. Subsequent developments have used first-person shooter games (e.g., Chen et al., 2023; Gentile et al., 2016; Kätsyri et al., 2013; Mathiak & Weber, 2006) and driving cars running over pedestrians (e.g., Klasen et al., 2020; Wolf et al., 2018; Zvyagintsev et al., 2016). These virtual violence scenarios allow for semi-natural behavior (Mathiak & Weber, 2006) and may be particularly relevant for measuring appetitive aggression, that is, positive feelings following aggressive behavior (Elbert et al., 2017). Higher immersion and realism in these video games increase aggressive behavior and intentions (Chen et al., 2023; Farrar et al., 2006), likely due to enhanced aggressive affect and cognition and physiological arousal (Chen et al., 2023).

Additionally, there are also some adaptations of existing laboratory tasks becoming more immersive with adding video game elements used in a more laboratory-controlled format. For example, Buades-Rotger et al. (2023) have proposed an adaptation of a Go-Nogo shooting task to measure hostile expectations learning in a more naturalistic, yet controlled setting (see Figure 1(G)). In recent years, there has also been progress in adapting standard game theory paradigms. For example, a novel version of the Chicken Game task has been developed to assess competition, dominance and aggression in a more immersive and dynamic environment by adopting the first-person perspective of a car driving toward another car (Hernandez‐Pena et al., 2024a); see Figure 1(K). In addition, Ahmed et al. (2023) created a video game environment where participants can interact with avatars and experience game theory paradigms (e.g., Prisoner´s dilemma or Dictator game) in a narrative and interactive manner. There are also some improvements in the virtual environment of the well-established Cyberball paradigm to investigate ostracism increasing engagement and customization (Long et al., 2023).

Using video games as lab-based aggression measures, particularly in decision-making scenarios, offers several advantages including enhanced immersion, intuitiveness, and engagement value (Allen et al., 2024). This technology can be successfully combined with real-time neurofeedback. Baqapuri et al. (2021) designed an immersive shooter video game and trained participants to regulate the motor area. Future research could explore the regulation of different target regions (e.g., prefrontal cortex usually associated with self-regulation and cognitive control; Friedman & Robbins, 2022) and whether this reduces aggression in an active and engaging environment.

Video games, however, have also notable disadvantages to consider. For instance, participants are aware that the avatars or other players are not being harmed for real, which fails to meet the common definition of aggression. Some participants might feel guilty while engaging in unjust or immoral behaviors in violent video games; however, this feeling decreases after repeated exposure (Grizzard et al., 2017). Additionally, engaging in “aggressive” behaviors—such as shooting others, running over pedestrians or crashing—can be entertaining and engaging (Gentile et al., 2016); thus, participants may act aggressively not with the intent to cause harm, but for enjoyment or competitiveness. Indeed, some research has shown that competitiveness, rather than violence, may be the factor driving aggressive behavior in violent video games (Adachi & Willoughby, 2011a). Furthermore, individuals with prior experience playing violent video games may introduce a confounding factor that influences both the construct and external validity of these tasks.

Importantly, only a few studies have investigated the relationship between violent video games and aggressive behavior as measured by the TAP or the Hot Sauce Paradigm, with findings indicating no significant relationship (Adachi & Willoughby, 2011a; Dowsett & Jackson, 2019; Kneer et al., 2016). Given these considerations, questioning the validity of the aggression measures employed seems prudent. Are laboratory tasks and more natural observations of video playing capturing different types of aggression? Are observations of aggressive behavior in video games truly measuring “real life” aggressive tendencies at all?

One potential avenue for bringing tasks that use video games intended to measure aggression closer to real-life interpersonal aggression is to develop video game environments in which participants can freely interact with each other (similar to Second Life or Roblox, see Section ‘Virtual Reality' and Section ‘Metaverse' for more details on virtual worlds), but in a controlled laboratory setting. This approach would provide opportunities for aggressive behavior while ensuring that participants’ actions result in real (and potentially harmful) consequences administered by the experimenter during or after the task, as in standard lab-based aggression tasks. Such an immersive setting could increase realism and participant engagement, combined with tangible repercussions of aggressive behavior simulating situations similar to those experienced in real life. Another suggestion is to consistently include an optimal control condition with a non-violent video game that matches the violent version in terms of competitiveness, difficulty and pace of action (Adachi & Willoughby, 2011b). This would help confirm whether the observed aggressive behavior in violent video games differs from that in non-violent versions and clarify whether the aggression is due to the violence itself rather than simply increased competitiveness, a greater desire to win, or frustration due to varying levels of difficulty.

Virtual Reality

Virtual Reality (VR; immersive, interactive virtual environment that simulates realistic scenes using a three dimensional display) integrates several features by recording real-time data while, most likely, adopting the first-person perspective of an avatar during social interactions, along with other advantages. There are already well-detailed and recent reviews on the implementation of VR in criminology research (Van Gelder, 2023) and particularly in combination with neuroimaging techniques and the study of body perception and expression in relation to aggression (De Gelder et al., 2023), as well as the assessment and treatment of aggression among forensic and clinical populations (Dellazizzo et al., 2019; Sygel & Wallinius, 2021), specifically concerning posttraumatic stress disorder (Rizzo et al., 2017), intimate partner violence (Johnston et al., 2023) or child abusers (Fromberger et al., 2018).

In this chapter, we want to examine the potential of VR as a tool for research on aggression. Interestingly, as highlighted by the focus of other reviews, most research on VR has been oriented toward applied and therapeutic studies, rather than fundamental aggression research. Consequently, the majority of the reviewed studies employed VR primarily for therapeutic applications. We will primarily focus on VR as an assessment tool for aggression while highlighting its advantages in ecological validity and realism (hereafter referred as mimicking real-life scenarios, ability to evoke emotions, immersion, incorporating interaction) over traditional applications. A major advance in research on criminology and aggression-related topics has been made by the group led by Van Gelder. These authors developed several highly realistic 360° video technologies to construct an immersive VR environment; see Figure 1(I) and 360° Virtual Scenario Method for an example. These videos simulate real-world conflict situations designed to elicit emotions such as anger or sexual arousal within controlled and flexible settings (Herman et al., 2024; Van Gelder, 2023). In an anger condition, represented by a provocative male in a bar, participants reported significantly higher feelings of disgust, anger, and annoyance compared to the control condition with large effect sizes (Herman et al., 2024). In a previous study, comparing written scenarios with VR representations of another bar fight situation, the authors discovered that VR significantly increased realism, subjective feelings related to aggression, such as anger, guilt, fear, and the intention to aggress, providing support for a state-trait model of aggression (Van Gelder et al., 2019, 2022) and physiological changes (Van Gelder et al., 2017). These findings highlight the potential of VR technology to create engaging and impactful simulations that can enhance our understanding of criminal and aggressive behavior. However, further testing is needed to determine whether these measures are related to real-life aggression.

Only a few studies have aimed to directly decrease aggressive behavior, measuring aggression as their primary outcome. The most noteworthy attempt was made by Klein Tuente et al. (2018), who developed a VR-based treatment (VRAPT) in forensic psychiatry to target reactive aggression by simulating interpersonal scenarios. The training sessions focused on emotion recognition, identifying and de-escalating aggressive behavior of others (i.e., avatars), regulating arousal and engaging in interactive virtual role-plays (see Figure 1(H)) (Klein Tuente et al., 2018). The treatment was applied to 128 forensic inpatients (Klein Tuente et al., 2020), with therapists controlling the environments, tailoring experiences using voice morphing and providing real-time physiological feedback. While there was no change in staff-reported or self-reported aggression, self-reported anger, hostility, and impulsivity temporarily decreased, with effects dissipating after 3 months. Most participants were willing to engage in VR therapy, recognizing its benefits alongside other therapeutic approaches (Klein Tuente et al., 2020). Another pilot study has used the VRAPT with prisoners finding a decrease in self-reported aggression and staff-rated observational aggressive behavior (Woicik et al., 2023). Overall, the VRAPT represents a valuable initiative for advancing future research aimed at the study and treatment of aggression, given its high acceptance among participants and patients, adaptability to individual characteristics (e.g., voice morphing) and interactive virtual role-playing which improves immersion.

Similar to the original VRAPT, Verhoef et al. (2021) focused on aggressive social information processing, but applied to children and adolescents. A virtual classroom was designed where participants could interact in VR with virtual peers in both proactive and reactive aggressive scenarios. In this pilot study, VR demonstrated good convergent validity, showing moderate correlations with the vignette assessments, while exhibiting greater measurement sensitivity (i.e., being more sensitive to capturing individual differences). Notably, discriminant validity was observed only in the provocative (reactive) but not in the instrumental (proactive) contexts (Verhoef et al., 2021). In a larger sample, the VR assessment elicited increased level of aggressive processing and responses exclusively in provocation scenarios (Verhoef et al., 2022). Interestingly, VR scenarios enhanced predictive validity, explaining an additional 2%−12% of the variance in children´s real-life aggressive behavior and motives beyond what was captured by the vignette assessment in both contexts.

Lobbestael and Cima (2021) developed a VR tool to differentially measure reactive and proactive aggression in young adults. In the reactive scenario, participants were provoked by being hit, with aggression measured by hits against the provoking avatar. The proactive version involved aggression toward an avatar blocking a goal without provocation. While reactive aggression correlated with self-reported aggression, the proactive scenario had limitations, as forced aggression (i.e., hitting the avatar was the only way to achieve the goal) led all participants to hit the avatar. In the same study, the authors tested a revised version using a cheating dartboard game, where participants could choose non-aggressive or aggressive responses (challenging the avatar, starting a fistfight or hitting them with a bottle). Around 80% of participants did not display aggressive behavior in either condition, supporting its ecological validity. Again, behavior in the reactive aggression condition significantly correlated with self-reported aggression and psychopathy traits, but not in the proactive scenario (Lobbestael & Cima, 2021). This revised version appears more suitable because it includes conditions that allow participants to refrain from aggression, mirroring real-life situations more accurately. In a study by Terbeck et al. (2022), participants examined responses to pro-social and provocative interactions with an avatar in VR. Participants were instructed to respond either congruently or incongruently (with an unfriendly punch or a friendly handshake) showing faster responses to provoking avatars.

Another protocol presented the implementation of VR videos depicting experiences of victims of bullying from a first-person perspective, showcasing scenarios such as being insulted, having personal belongings stolen, or experiencing exclusion (Barreda-Ángeles et al., 2021). This pilot study demonstrated the effectiveness of VR in eliciting realistic negative emotional responses to acts of bullying and increasing empathy. When comparing VR to traditional screen presentations, participants reported a greater sense of presence in the VR condition; however, no significant differences were observed in self-reported valence and arousal. Notably, physiological responses were greater in the VR condition compared to the screen version, suggesting increased emotional arousal.

An additional advantage of VR is the option to apply the scenarios as treatment training platform equivalent to exposure therapy. This has been conducted in different settings and populations such as patients with posttraumatic stress disorder (Smeijers et al., 2021; Zinzow et al., 2018). Also, VR has been successfully used to reduce aggressive behavior in children with conduct problems (Alsem et al., 2021), as well as increase empathy in real-life bullying situations (Gu et al., 2023; Ingram et al., 2019).

A recent study on VR and intergroup conflict (Hasler et al., 2021) found that participants in VR perceived harmful actions as less moral and justified, driven by hostile emotions toward in-group actors, compared to those who viewed the same scene on a screen. VR’s ability to enhance immersion and evoke stronger emotions highlights its potential for intergroup violence and its mechanisms.

Some researchers have examined the moderating effect of VR on violent video games. The findings have been mixed, with some studies reporting that VR increases feelings of aggression, hostility and anger compared to screen presentations (Lull & Bushman, 2016; Persky & Blascovich, 2007), while others report a decrease in aggressive behavior and hostile states (Arriaga et al., 2008).

VR as a measurement tool for aggressive behavior offers numerous advantages (see Table 1 for a summary). VR can create artificial environments that closely mimic high-risk real-life situations (Fromberger et al., 2018; Klein Tuente et al., 2018), establishing plausible research settings that are difficult or even impossible to realize in the real world related to aggression and violence (Van Gelder, 2023). This technology provides a visual and auditory perceptual experience comparable to real-world contexts, demonstrating high ecological validity while maintaining a controlled and replicable setting (Rizzo et al., 2017; Van Gelder, 2023). Scenarios are designed in such a way that cover stories are unnecessary, thereby reducing dropout rates or response bias from disbelieving participants (Lobbestael & Cima, 2021). VR can even elicit aggressive responses without endangering others (Fromberger et al., 2018), greatly mitigating ethical and safety concerns (Dellazizzo et al., 2019; Johnston et al., 2023; Van Gelder, 2023); however, this may reduce validity compared to real-life situations where aggressive behavior has consequences. Still, participants tend to respond realistically to virtual representations of real-life events (De Gelder et al., 2023; Dellazizzo et al., 2019) and are less self-conscious about their behavior due to immersive nature of VR (Van Gelder, 2023), which is especially important in aggression research due to issues of social desirability bias.

VR technology, compared to other tools, has consistently been reported to enhance feelings of presence and immersion while effectively triggering emotions like anger (Fromberger et al., 2018; Van Gelder, 2023). Written scenarios are limited in the ability to capture or elicit emotionally laden and visceral aspects due to constraints on contextual information and nonverbal behavior exhibited by characters in the vignettes (Van Gelder et al., 2022). Virtual embodiment through immersive VR facilitates perspective-taking (Gonzalez-Liencres et al., 2020; Johnston et al., 2023; Ventura et al., 2021), which could be especially beneficial for investigating aggression in some psychiatric populations with theory of mind deficits. Also, this technology presents fewer barriers compared to imagination techniques, written scenarios or lab-aggression measures (Fromberger et al., 2018), or self-reported data requiring meta-cognition and retrospection, relying less on participants’ abilities to imagine themselves within those situation effectively (Van Gelder et al., 2022). Specifically, VR can successfully manipulate targeted emotional experiences within environments that represent potential criminal opportunities, facilitating the assessment of decision-making processes while in an emotionally charged state (Herman et al., 2024).

With VR behavioral data is collected in real time, revealing precise information underlying violent and aggression dynamics (Rizzo et al., 2017) and allowing for the identification of event chains linking predictors to outcomes in detail—contrary to standard survey methods (Van Gelder, 2023), facilitating testing of prevalent aggression theories. Moreover, this technology can be combined with physiological data recording and even biofeedback. Advanced VR googles can capture ancillary movements such as eye-tracking, pupil dilatation or reaction times (Herman et al., 2024). This multidata approach can improve validation of integrative aggression theories, such as the General Aggression Model (Anderson & Bushman, 2002) or I³ Model (Finkel & Hall, 2018), because it combines personal characteristics cognitive, emotional and biological responses with situational variables. Another advantage is the possibility to conduct VR experiments within prisons, forensic and clinical institutions, reaching populations that are often difficult to access—such as incarcerated offenders (Van Gelder, 2023). In most studies, participants report high levels of retention rates, motivation, engagement and acceptability when using VR technology to measure aggression (Klein Tuente et al., 2020; Verhoef et al., 2021, 2022; Woicik et al., 2023).

There are also some disadvantages associated with VR technology, consult Table 1 for a summary. Due to the higher levels of immersion, absorption and emotional involvement, participants might experience intense negative emotional responses, which have been linked to negative rumination, anxiety and distress (Lavoie et al., 2021; Lundin et al., 2023), raising concerns about lasting harm to subjects that experimenters must consider carefully (Fromberger et al., 2018). Another limitation is that these experiments often rely on measuring behavioral intentions rather than actual behavior;participants typically report what they would do in specific situations rather than demonstrating those actions directly (Van Gelder, 2023). Thus, one main limitation specifically in aggression research is that participants are aware there are no real-world consequences, which significantly restricts the assumption that they believe that their behavior might cause harm to others or that the others want to avoid such potential harm.

Although it can easily be combined with physiological and eye-tracking recordings, VR has limitations regarding its implementation with neuroimaging techniques. Previous studies indicate difficulties integrating VR within magnetic resonance imaging (MRI) systems due to physical motion, suggesting the use of non-intrusive eye-tracking system as a solution (Qian et al., 2021). Elevated motion can also pose challenges for other neuroimaging techniques such as electroencephalography (EEG) or functional near-infrared spectroscopy (fNRIs), restricting the investigation of neural correlates of aggression using VR. In relation to this head motion, participants may experience “cybersickness”—symptoms like dizziness and nausea—which limits its suitability for all individuals and patients (González Moraga et al., 2022; Lavoie et al., 2021; Lundin et al., 2023; Smeijers et al., 2021; Van Gelder, 2023; Woicik et al., 2023).

How can VR studies be adapted to investigate aggression?

VR technology today is limited in its realism concerning the complexity of real-life interactions (Van Gelder, 2023). Most studies used pre-programmed tasks (e.g., Lobbestael & Cima, 2021; Van Gelder et al., 2019), while only a few have incorporated interactions with real humans (i.e., therapists) (Klein Tuente et al., 2018, 2020). Expanding the use of real interactions within VR environments could enhance ecological validity by involving actual people interacting in VR scenarios with limited potential harmful consequences in a laboratory setting. Even without real consequences, participants might still experience emotional distress during social virtual interactions if they know that they are playing against other real individuals. This research is particularly relevant given the exponential growth of virtual world technologies, such as “the metaverse”; see section ‘Social Virtual World' for more details on social media, metaverse and cyberbullying (Cheng, 2023).

As previously described, a few studies have used VR to investigate reactive and proactive aggression (Lobbestael & Cima, 2021; Verhoef et al., 2021, 2022). All have found support for measuring reactive aggression but not proactive aggression. There is a gap in the literature regarding improving VR assessments for measuring proactive aggression by incorporating non-provoking conditions alongside incentives. One potential approach could involve realistic and engaging settings (e.g., a competitive card or resource-based games) in which participants play against other individuals represented by avatars, while introducing an unfair rule allowing them to earn more points or money by interfering with others’ performance; driven by instrumental motivation (Zhu et al., 2019). Another potential approach could be the use of high monetary incentives (real or fictional) to elicit proactive aggression; however, this raises substantial ethical concerns. As discussed in the ethical considerations section below, participant well-being must always take precedence over experimental aims. Some individuals might experience distress or guilt after realizing that they acted aggressively for personal gain instead of in response to provocation, as in reactive aggression.

Audiovisual Recordings and Wearable Devices

A significant discrepancy between lab-based aggression measures and real-life aggressive behavior is evident. Consequently, aggression research can benefit from being examined in authentic, real-world contexts. The challenge of capturing genuine instances of aggression in natural settings is well-known; however, this approach may facilitate a more comprehensive understanding of the precursors and dynamics of aggression. The use of audiovisual recording for the study of children’s aggressive interactions is a well-established practice (e.g., see Pepler & Craig, 1995). However, the use of fixed cameras has been criticized due to the fact that aggression occurs primarily in settings that are not easily accessible, thereby necessitating the use of mobility devices (Wettstein & Jakob, 2010). One approach to approximating real-world aggression scenarios (Wettstein & Jakob, 2010) and the environmental conditions (Wettstein & Scherzinger, 2015) is the use of camera-glasses. Despite its potential as a novel assessment technique and the authors’ assertion of low reactivity and high compliance, further applications of this technique have been limited probably due to legal and ethical considerations and change of behavior because of lack of intimacy (Wettstein & Scherzinger, 2015).

Mobile devices can also assess other parameters. Proactive and reactive types of aggression differ in physiological correlates including heart rate, skin conductance and respiratory sinus arrhythmia (for a specific review, see Fanti et al., 2024). Consequently, some research in the field of aggression has concentrated on using wearable devices (such as sensor wristbands) to monitor physiological data (including heart rate and electrodermal activity) and physical activity (such as motion using an accelerometer). Biosensor technology can be used to anticipate aggression episodes. Previous studies have demonstrated efficacy of predicting naturalistic aggressive behavior in challenging populations, including children and youth with autism spectrum disorder (Goodwin et al., 2019; Imbiriba et al., 2023; Ozdenizci et al., 2018) and patients in psychiatric mental health institutions (De Looff et al., 2019). The analysis of biosignals prior to the onset of aggression has the potential to mitigate the occurrence and impact of aggression in clinical populations. Further research should focus on identifying risk factors, as well as providing more evidence on how reliable this data is in predicting aggression.

Ter Harmsel et al. (2023) used the Sense-IT app for intervention with forensic outpatients with high levels of aggression, finding decreased in self-report aggression, increased interceptive awareness and physiological signals of tension. Thus, this app can not only be used for intervention but also holds promise for acquiring relevant research data on the daily-life associations between physiological data, anger and incidents of aggression measured in combination with EMA.

Ecological Momentary Assessment

As mentioned before, aggression research is mainly investigated to date using lab-based tasks or survey methods. However, this is restricted to the measurement time and context. Ecological Momentary Assessment (EMA), also known as experience sampling method, allows for real-time data collection that captures individuals’ behaviors, thoughts and emotions in their natural environments (Shiffman et al., 2008). This methodology involves repeated assessments of participants’ experiences as they unfold in daily life, thereby minimizing recall bias, enhancing ecological validity, capturing day-to-day variation and improving reliability by collecting multiple data points per item (Borah et al., 2021; Murray et al., 2022; Shiffman et al., 2008). EMA aims to document micro-processes and contingencies between the behavioral, cognitive and emotional dynamics in real-world contexts (Russell & Gajos, 2020; Shiffman et al., 2008) using various technologies, including smartphones and electronic diaries (Borah et al., 2021; Shiffman et al., 2008). The behavioral, emotional, and experience data collected by EMA can be effectively combined with other data simultaneously acquired with wearable sensors, such as physiological data, geographic or motion data (Borah et al., 2021; Russell & Gajos, 2020). By integrating these different data sources, researchers can build a more comprehensive picture of the factors influencing aggressive behavior in real time, which facilitates testing theories combining biological and environmental factors like the General Aggression Model (Anderson & Bushman, 2002).

The application of EMA in aggression research offers unique opportunities while posing specific challenges (see Table 1 for an overview). Aggression trait measures were not originally designed to capture momentary experiences, and there is no guarantee that these measures work as intended with EMA aimed at capturing “real-time” processes, despite adaptations (Borah et al., 2021). Some adaptations may include rewording items to reflect state-like rather than trait-like experiences and adjusting the focus to capture momentary incidents, as serious aggressive events are often too infrequent to be assessed within the short time frames typical of EMA studies (Murray et al., 2022). There is a critical need for psychometrically validated measures specifically designed to assess aggression within this sampling method. The most notable attempt has been made by Borah et al. (2021) and Murray et al. (2022) by creating and validating the Aggression-ES scale, which is tailored to measure aggression in momentary assessments. This scale includes 12 items targeting both proactive and reactive aggression, as well as physical and social/indirect forms of aggression and focuses on common “everyday” manifestations of this behavior due to the lack of probability of serious incidents within the EMA timeframe and the goal of greatest universal applicability (Borah et al., 2021).

Disadvantages of EMA include the potential burden on participants (Borah et al., 2021), which is especially relevant when working with at-risk or under-resourced youth (Russell & Gajos, 2020). Consequently, brief scales are sometimes preferred (Murray et al., 2022) and some studies may benefit from tailoring the assessment protocols to better align with participants’ daily lives to enhance compliance (Russell & Gajos, 2020). Interestingly, using the Aggression-ES scale along with other emotion scales, Murray et al. (2023) found that higher levels of stress and overall negative affect (particularly feelings of being upset) at a given notification predicted a greater likelihood that participants would miss the subsequent prompt. In other words, the emotional state of the participant might influence the likelihood of answering a prompt and, therefore, affecting the data quality. This finding is very relevant for future studies, as it highlights the importance of considering non-random compliance in EMA research. Understanding these dynamics can inform strategies to improve engagement in smartphone-based data collection applications, ultimately improving the quality and reliability of EMA findings. The use of event-contingent, instead of prompt-contingent sampling, in which participants report only when they have engaged in aggressive behavior, might reduce unnecessary data collection and reduce the burden of many EMA assessments. However, this method also presents two major challenges: the need for clear definitions of the aggressive events and the risk that participants may underreport the instances or precursors of aggressive behavior (Murray et al., 2022; Shiffman et al., 2008). Although EMA minimizes issues associated with retrospective recall, it does not guarantee that participants can accurately report their behaviors, emotions, and experiences (Borah et al., 2021). Participants may still be susceptible to response bias, or may struggle to accurately assess their emotions. This is particularly relevant for individuals with alexithymia, who may have difficulty identifying and articulating their emotions. Therefore, researchers must take these factors into account to ensure the validity of data collected in EMA studies.

Future research should focus on examining risk factors such as emotional lability (i.e., variability in emotional states over time) or the co-occurrence of precursors (e.g., provocation and anger) that cannot be examined with traditional survey data (Borah et al., 2021; Russell & Gajos, 2020). Different precursors (e.g., intoxication, lapses in self-control, or hostile states) acquired in natural settings and in the flow of daily life, rather than in artificial environments, may have a stronger predictive effect on aggressive behavior. This research also has the potential to investigate how daily events, behaviors, cognitions, and emotions converge to influence long-term changes in traits (Borah et al., 2021). Using EMA, researchers can investigate how within-person aggressive behavior varies both between individuals and across contexts, such as work, school, home or leisure time (Russell & Gajos, 2020). A multimodal approach combining physiological, location, behavioral and emotional monitoring holds great promise for investigating the precursors and maintenance of aggression and better testing integrative and comprehensive aggression models (Anderson & Bushman, 2002; Baskin-Sommers et al., 2024; Finkel & Hall, 2018).

Cyberbullying

Access to social media and its widespread use have facilitated new avenues for online aggression, with frequency and problematic usage of social platforms strongly associated with cyberbullying (Craig et al., 2020). Cyberbullying is usually defined as “an aggressive, intentional act carried out by a group or an individual, using electronic forms of contact, repeatedly and over time against a victim who cannot easily defend him or herself” (Smith et al., 2008). Instances of cyberbullying include sending mean messages or sharing inappropriate images of others (Giumetti & Kowalski, 2022) (see Figures 2(A) and (C) for examples). The nature of social media increases the likelihood of this harmful behavior due to its greater accessibility to potential targets, the ability to edit or delete posts and the expanded anonymity and popularity it provides (Giumetti & Kowalski, 2022). Compared to traditional bullying, which is limited by time and location, cyberbullying is distinguished by its ability to occur anytime and anywhere, the potential for content to be shared widely via bystanders without the aggressor needing to repeat the behavior and the complexities of power dynamics that can arise from anonymity and superior technological skills (Corcoran et al., 2015).OPEN IN VIEWER

Much of the existing research on cyberbullying relies on non-experimental methods such as interviews and primarily self-reported data (Chun et al., 2020; Zhang et al., 2022), which raises concerns about social desirability bias (Nagar & Talwar, 2021). Experimental research, however, can measure actual bullying behavior in controlled environments (Giumetti & Kowalski, 2022; Nagar & Talwar, 2021; Pieschl et al., 2013). Only a limited number of studies have used experimental settings to investigate cyberbullying. The most common method is experimental vignettes controlling for variables related to cyberbullying (Nagar & Talwar, 2021). For example, Jungert et al. (2021) used a written vignette task to depict either direct bullying or cyberbullying, exploring bystanders’ roles. Similarly, Pieschl et al. (2013) used vignettes prompting participants to imagine themselves in various cyberbullying scenarios involving harassment or outing, differing by media type (text vs. video), with distinct coping strategies based on the type of cyberbullying and media. The described studies allow for exploration of individual responses to the same situations; something not achievable with self-reported retrospective data alone. Hypothetical vignettes are a popular method in social science, but more innovative methods observing actual behaviors reflecting cyberbullying are still limited.

Some of these experimental advances include the use of communication platforms like simulated social media websites as experimental paradigms to facilitate and measure cyberbullying behavior and changes in cyber-bystander responses within a controlled virtual environment; a method that remains underutilized (Nagar & Talwar, 2021). Some studies have successfully employed tools within a controlled virtual setting such as Social Network Software Systems to directly measure behaviors or facilitate behavior change. For a comprehensive systematic review on 17 articles utilizing this tool or computer games, please refer to Nagar and Talwar (2021). In short, simulated social networks—mimicking platforms like Facebook—allow researchers to measure behaviors such as forwarding offensive messages, flagging, liking/disliking or deleting them, replying to posts, block or report users (e.g., see Figures 2(B) and (C)) and observing interactions in group chats. Researchers maintain full control over platform features and paradigms, manipulating different variables such as audience (DiFranzo et al., 2018) (see Figure 2(D)), while ensuring participant safety; avatars can be created instead of using real photos on simulated accounts, promoting a sense of integrity while still simulating interactions with others (Nagar & Talwar, 2021). Lastly, a recent study by McLoughlin et al. (2020) explored the neural underpinnings of cyberbullying by examining brain activation in bystanders witnessing cyberbullying behavior (see an example in Figure 2(A)). To illustrate power imbalances, the cyberbullying comment received more “likes” than the post itself.

Future studies could concentrate on cyber-aggression (including a wider behavioral range) rather than focusing on only cyberbullying behaviors to broaden our understanding of harmful online behaviors that can occur without the need for repetition and power imbalance, unlike traditional bullying definitions, while still recognizing that even a single incident can have significant psychological effects on the victim (Corcoran et al., 2015; Zych et al., 2016). Cyber-aggression is defined as “intentional harm delivered by the use of electronic means to a person or a group of people irrespective of their age, who perceive(s) such acts as offensive, derogatory, harmful or unwanted” (Grigg, 2010), including behaviors such as bullying, harassment, abuse or hostility on the Internet. This approach will also broaden the scope by not only examining cyber-aggressive behaviors within peer groups but also addressing targeting of celebrities, vulnerable individuals or groups, and school staff (Corcoran et al., 2015).

Machine Learning Techniques for Identifying Written Aggression on Social Media

Advances in technology have greatly increased the availability of rich datasets that integrate different types of data, requiring the development of novel analytical methods to effectively process and interpret these data. The use of new methods is also becoming increasingly common in the study of aggression and violence on social media and the internet. There is a growing and urgent need for effective approaches to cyberbullying detection. Particularly in social media environments characterized by anonymity, harmful consequences of online aggression are observed (Al-Harigy et al., 2022). The field of cyberbullying and abusive language detection has witnessed considerable advancement through the integration of artificial intelligence and machine learning techniques, including natural language processing and deep learning. Several comprehensive reviews have compared various machine learning classifiers and algorithms (Al-Garadi et al., 2019; Al-Harigy et al., 2022; Han et al., 2024; Muneer & Fati, 2020; Vidgen & Derczynski, 2020) and others have addressed the challenges and limitations associated with automated detection and non-supervised techniques (Elsafoury et al., 2021; Farag et al., 2019; Rosa et al., 2019).

In summary, natural language processing enables researchers to analyze large volumes of user-generated content to identifying instances of aggressive language through techniques with pre-defined rules such as sentiment analysis and topic classification (Al-Harigy et al., 2022). This approach enables the examination of patterns of online aggression across a range of social media platforms, including Twitter, Facebook, Instagram, Formspring and Reddit. The majority of research efforts in the field of cyberbullying detection have concentrated on the analysis of individual social media posts, such as a single tweet or a blog response from a single user, or the aggregation of multiple texts from different users within a conversation. While this approach has proven effective in identifying instances of aggressive language, it often fails to consider the broader context and dynamics that characterize online interactions (Cheng et al., 2021). Meanwhile, deep learning models can handle diverse data types simultaneously, including text, images and metadata. These models can recognize complex patterns considering the relationship between the elements (Al-Harigy et al., 2022), rather than relying solely on predefined lists. This approach enhances detection accuracy by capturing subtle communication forms and contextual features, including emotional cues (Al-Harigy et al., 2022), that traditional approaches may overlook.

Despite these advancements, several challenges remain in this field (see Table 1 for an overview). The lack of consensus on a definition of cyber-aggression hinders both the detection of such incidents and the generalization of models across different studies (Elsafoury et al., 2021; Rosa et al., 2019; Vidgen & Derczynski, 2020). The inconsistent use of terms such as “cyberbullying,” “cyber-aggression,” or “cyber-grooming” across the literature underscores the need for clearer, consensus-driven definitions and standardized operational frameworks (Mladenović et al., 2022). Establishing such definitions within digital aggression contexts would enhance construct validity and conceptual clarity, improve measurement precision and automatic detection, as well as facilitate comparability and replication across studies. Manual data labeling practices frequently exhibit low inter-annotator agreement scores along with class imbalances and insufficient linguistic diversity, which can lead to overfitting and unreliable models (Elsafoury et al., 2021; Farag et al., 2019; Vidgen & Derczynski, 2020). For a repository containing training datasets related to abusive language, refer to https://hatespeechdata.com (Vidgen & Derczynski, 2020). The primary issue is that violent social media data is considerably more intricate than it is often represented in these models. The data on aggression dynamics generated on social media is frequently produced in real time and often comprises interactions among multiple users. Such content often comprises slang expressions (Elsafoury et al., 2021), abbreviations (Al-Garadi et al., 2019), neologisms, emotional expressions and emojis (Al-Harigy et al., 2022), with many posts being informal and lacking grammatical accuracy (Cheng et al., 2021). By focusing solely on individual messages rather than the broader context of conversational flows and sessions, traditional methods may overlook critical patterns related to repetition and power dynamics that are inherent to cyber-aggressive behavior (Cheng et al., 2021). A further limitation to date is the inability of current methods to detect instances of cyber-aggression occurring in real time on different online platforms and social media networks, which is essential for prevention strategies (Han et al., 2024; Muneer & Fati, 2020; Rosa et al., 2019).

Future directions focusing on understanding how aggression propagates through social media sessions require a more comprehensive analysis that considers not just isolated texts but also multimodal data, including images, timestamps and user interactions over time and more complex and dynamic structures (Cheng et al., 2021; Vidgen & Derczynski, 2020), including emojis that may help to determine the context of the sentence and to investigate the sentiment analysis, not merely the words (Al-Harigy et al., 2022). In addition, future research could explore language and conversational parameters such as melody, speed, or content in social media videos on platforms such as TikTok or Instagram to further understand violent interactions online.

Future studies should explore how connections between users change over time, thereby providing a more realistic insight into aggressive user interactions. In a recent study, Terizi et al. (2021) proposed a novel framework for evaluating the propagation of aggression in social networks. This framework uses opinion dynamics models to simulate how aggressive behavior spreads among users. This shift toward session-based detection (i.e., modeling the hierarchical structure of user interaction networks and temporal patterns of social media sessions) has the potential to facilitate more effective identification of power dynamics and repetition of cyberbullying instances, as well as the development of prevention strategies (Cheng et al., 2021). Furthermore, future research should examine pre-trained models like “Bidirectional Encoder Representations from Transformers” on datasets containing slang or informal language to potentially improve its performance in detecting nuanced forms of aggression (Al-Harigy et al., 2022; Elsafoury et al., 2021). Integrating these advancements into practical tools, such as real-time monitoring systems or third-party anti-bullying applications, could greatly enhance efforts to combat cyber-aggression successfully across diverse online environments (Cheng et al., 2021; Farag et al., 2019; Muneer & Fati, 2020; Rosa et al., 2019). For example, Raj et al. (2022) proposed a deep learning model designed to detect cyberbullying in real-time tweets across multiple languages with an impressive accuracy of 98%. This research could expand the scope beyond text-based detection to images and emojis and other social media platforms.

Metaverse

As social virtual environments become increasingly prominent, there is a critical need for further research in the field of cyber-aggression. The Metaverse is an emerging virtual space that integrates various technologies, creating a shared environment where users can interact through avatars, transcending physical limitations of time and space (Wang et al., 2023). Virtual environments and immersive games, including platforms like Second Life, Habbo, Roblox and VRChat, serve as precursors to the Metaverse, offering some insights into potential issues (Dwivedi et al., 2022). Aggressive behaviors in the Metaverse include online abuse, sexual harassment, cyberbullying, hate speech and discrimination (Dwivedi et al., 2022, 2023). The unique characteristics of this virtual world, particularly online disinhibition, can lead individuals to engage in deviant actions more freely than they might in face-to-face interactions (XinYing et al., 2024). A crucial question is how the Metaverse facilitates aggressive and deviant behavior (Dwivedi et al., 2022). XinYing et al. (2024) investigated the influence of various features on deviant behaviors among users of Metaverse platforms. The authors found that technical features, such as immersive experiences and invisibility, alongside social factors like homophily, social ties and social identity significantly impact users’ deviant behavior. In other words, deviant behavior among users can be contagious and tends to increase under conditions of anonymity, while adopting a broad social identity can reduce such behaviors. As previously discussed in the VR section, the immersive nature of the Metaverse may intensify the psychological impact of these aggressive acts on victims, making them feel real despite occurring in a virtual context (Dwivedi et al., 2023).

One critical area for future investigation is the comparison between virtual aggression and real-life aggression. Researchers should explore whether the triggers, manifestations and impacts of aggressive behaviors are consistent across these contexts. Do real-life and virtual aggression share similar risk factors? How does anonymity afforded by avatars exacerbate antisocial behaviors like cyber-bullying and harassment? How do immersive experiences influence users’ emotional responses and behavioral choices within the Metaverse? Gaining further knowledge about these influencing factors can inform effective prevention strategies aimed at reducing cyber-aggressive behavior in emerging digital spaces such as Meta.

Ethical Considerations

The generalizability of aggression research is inherently limited due largely to ethical constraints limiting harmful behavioral operationalization. As result, the majority of the studies include low-harm scenarios where participants perceive their actions only minimally harmful, which may lead them to believe that others might similarly be low motivated to avoid these “harmful” actions. Consequently, behaviors exhibited in lab-based aggression paradigms appear restricted and do not adequately represent the multi-dimensional nature of aggression. To partly address this limitation, it is necessary to investigate manifestations of aggression in daily life across various contexts, including home, work/school and digital environments. VR also offers a promising avenue for simulating different scenarios where participants can exhibit a higher range of aggressive behaviors (or at least show their intentions in that context). However, researchers must remain mindful that lab-based findings under-represent the broader spectrum of aggressive behaviors and should exercise caution when generalizing these results to real-world situations.

The integration of these emerging technologies into aggression research raises important ethical considerations that must be explicitly addressed in both study design and implementation. Highly immersive technologies such as VR, naturalistic stimuli and interactive video games can evoke strong temporary emotions (e.g., anger, fear, distress). While this emotional engagement is valuable for ecological validity, it also increases the risk of short- or long-term discomfort, particularly in vulnerable or clinical populations. Participants may experience lingering rumination, guilt or anxiety after exposure to realistic aggressive scenarios (Lavoie et al., 2021). Historical precedents in social psychology, such as the Milgram obedience experiments (Milgram, 1963), have shown that participants can experience lasting distress when they realize their own capacity for potential harm, even in a controlled research setting (Perry, 2013). Aggression researchers must therefore balance the pursuit of ecological realism with the paramount responsibility to protect participants’ psychological well-being. This is particularly relevant for proactive-aggression paradigms, in which harm may be inflicted for instrumental reasons, which could increase post-experimental rumination or guilt. Hyperscanning also presents unique ethical challenges, as researchers cannot fully control participants’ spontaneous behavior. One participant may act more aggressively than expected, which could cause distress to the other participant(s), or even put a strain on their relationship if they know each other.

To safeguard participants, researchers should provide transparent information about potential adverse effects, ensure that consent is fully informed and understood, allow participants to skip tasks or questions, and make it clear that they can withdraw at any time without penalty. Continuous monitoring is recommended during immersive tasks, particularly for high-arousal tasks or vulnerable groups. After participation, structured debriefing procedures should be used to help individuals process their reactions, understand the experimental context and prevent lingering guilt or distress. When appropriate, participants should have access to psychological support or follow-up contact.

Social-media and digital-platform research presents further ethical challenges due to the blurring of boundaries between public and private spaces. Users may not perceive their online content as public, even when it is technically accessible. Thus, researchers must prioritize transparency, user privacy and harm minimization when analyzing or disseminating online behavioral data. Ethical guidelines should include robust safeguards such as anonymization, encryption and data minimization, in line with data protection frameworks like the General Data Protection Regulation (GDPR; European Commission, 2016) and HIPAA (U.S. Department of Health and Human Services, 2013).

Finally, studies involving vulnerable populations, such as forensic or clinical populations or minor participants, require particular caution. These groups face an increased risk of distress, coercion and data misuse. In such contexts, independent ethical oversight, careful risk-benefit evaluation and comprehensive debriefing are essential to upholding participant well-being.

Conclusion

Effective assessment of aggression and violence remains one of the major challenges in the field. Advances in the operationalization and measurement of aggressive behavior are critical to improve our understanding and validation of existing theories and to inform prevention strategies and targeted treatments for aggressive individuals in forensic and clinical populations. These advanced assessment methods are essential for obtaining more accurate data, particularly from violent or high-risk patients, who are often difficult to reach in aggression research. By improving our measures and translating findings into real-world situations, we can more effectively predict and address aggressive behavior in real life.

We propose a shift in operationalization toward technology-integrated approaches that conceptualize aggression as a dynamic, context-dependent process. These tools aim to improve ecological and construct validity, with support from the integration of multimodal and multitemporal data and addressing emerging forms of aggression. Future research should focus specifically on digital environments, such as social media and the metaverse, where new forms of technology-mediated interpersonal aggression, such as cyber-aggression, may arise and existing theories may not apply adequately.