The Proteus Effect: Overview,Reflection,and Recommendations

Abstract

Avatars are an integral element in most virtual environments including virtual reality and video games, and the Proteus effect initially proposed by Yee and Bailenson (2007) is a framework that explains the psychological effects of individuals’ digital self-representation, which theorizes that the traits of avatars would affect individuals’ behavior and cognition accordingly. Although research on the Proteus effect has been fruitful over the years, this line of research also showed inconsistencies over whether the effect is significant, the direction of how avatars affect individuals, as well as its theoretical explanations, which suggests a lack of elucidation on the cognitive processes underlying the Proteus effect. Hence, this paper provides an overview of the Proteus effect research and offers a reflection on the current literature regarding the theoretical basis, empirical evidence, and methodological approaches of the Proteus effect research. Finally, this study provides three recommendations for future research on the Proteus effect.

With the development of gaming and virtual reality (VR) technologies, people have gained the ability to choose and change their digital self-representations in virtual environments, often by using various avatars or virtual characters in VR and games. While applying avatars can enhance individuals’ enjoyment of using these media technologies (Kim et al., 2015; Trepte & Reinecke, 2010), the specific social and psychological effects of such digital self-representation on users have become an intriguing research area. The Proteus effect (Yee & Bailenson, 2007) has been a well-known theoretical perspective that illustrates how avatars affect users’ behavior and cognition, and research on the Proteus effect has been fruitful over the years (Ratan et al., 2020). However, it is noteworthy that previous literature on the effects of avatars showed inconsistent findings regarding whether the effect is significant, the direction of how avatars affect individuals, as well as its theoretical explanations, which suggests a lack of elucidation on the cognitive processes underlying the Proteus effect. Driven by these inconsistencies, this paper provides an overview of the Proteus effect research, its importance in today's media environment, as well as the key theoretical, empirical, and methodological issues in existing literature on the Proteus effect. Finally, based on the circumstances of current research, this study offers some concrete recommendations for future research on the Proteus effect.

The Proteus Effect: An Overview

The Proteus effect was first proposed in Yee and Bailenson's (2007) study. Since avatars serve as a primary identity cue in virtual environments, it was expected that the traits of avatars could affect users’ behavior accordingly in virtual environments, which was the initial theoretical proposition of the Proteus effect. According to Yee and Bailenson (2007), the theoretical basis of the Proteus effect is self-perception theory, which argues that individuals tend to infer their emotions, cognition, and other internal states by observing their behavior and the circumstances where such behavior occurs (Bem, 1972). Specifically, Yee and Bailenson (2009) articulated that in the context of virtual environments, observing one's own appearance can lead to a change in behavior. To test the Proteus effect hypotheses, Yee and Bailenson (2007) conducted two experiments in which they manipulated certain traits of participants’ avatars with the confederate blind to the condition, and the results suggest that participants with attractive avatars walked closer to and disclosed more information to the confederate compared to those with less attractive avatars; also, the height of assigned avatar was positively related to the degree of confidence that an individual behaved in a negotiation task in the VR environment. Based on these results, Yee and Bailenson (2007) concluded that digital self-representations could affect individuals’ behavior in VR environments.

A later study by Yee et al. (2009) extended the scope of the Proteus effect to online games and subsequent face-to-face interactions after the virtual experience. Specifically, they found that in the online game World of Warcraft, taller and more attractive avatars outperformed shorter and less attractive avatars, therefore extending the original scope from VR to online games with avatars as well. Moreover, they also found that individuals who were assigned taller avatars were more aggressive negotiators not only in VR but also in a negotiation task in the real world immediately after the virtual experience, which extends the original theory from behavior in VR environments to subsequent behavior in face-to-face interactions.

In Yee and Bailenson (2007) and Yee et al. (2009), the independent variables of the Proteus Effect research are traits of avatars: in addition to the traits of height and attractiveness in these two initial studies, other avatar traits such as race (e.g., Ash, 2016; Eastin et al., 2009), gender (e.g., Sherrick et al., 2014), and weight (e.g., Peña & Kim, 2014) were also examined by subsequent Proteus effect studies. Since a large number of Proteus effect studies are experimental research, these traits are usually manipulated instead of measured, and some researchers ran pretests to make sure that the manipulation of these traits is valid, as in Yee and Bailenson (2007) as well as Yee and Bailenson (2009).

The dependent variable, however, has witnessed an extension in scope: although the two initial studies only proposed and examined behavior as the independent variable of the Proteus effect (Yee & Bailenson, 2007; Yee et al., 2009), later research has also examined how avatar traits affect cognitive and affective outcomes such as body-related thought (Fox et al., 2013), aggressive affect (Ash, 2016), racial bias (Groom et al., 2009), and exercise attitude and motivation (Li et al., 2014). These outcomes seem to be a subset of dependent variables used in classical media effects research, but are more related to the avatar traits of interest in the Proteus effect research area.

Based on these categories of dependent variables, scholars have applied multiple types of measures including unobtrusive behavioral measurements, self-reported scales, implicit measures, as well as physiological measures. Behavioral outcomes can be measured unobtrusively using behavioral measures without the participants’ awareness, and some frequently used behavioral measures include the interpersonal distance measure in immersive VR (Bailenson et al., 2003) and the measure of confidence as in a monetary negotiation task developed by Yee and Bailenson (2007). Cognitive and affective measurement, however, tend to rely on either self-report scales or implicit measures. As examples of self-reported measures, some scholars who are interested in how avatars affect attitude and behavioral intention used the 1–7 Likert scales of Ajzen and Fishbein (2000). Also, examples of implicit measures include the Implicit Association Test (Greenwald et al., 1998) as a racial bias measure and the word completion task as an aggressive cognition measure (Anderson et al., 2003). In general, implicit measures were used more frequently for concepts that, if measured explicitly, might result in social desirability bias, as in the cases of racial bias and aggressive thought. Variables that are less likely to be affected by social desirability, such as emotions, nonsensitive attitude, and behavioral intention, tend to be measured using self-report scales. Moreover, some scholars utilized physiological measures for the Proteus effect research as well, and the outcomes included electrocardiogram, heart rate, skin conductance responses, and so on (e.g., Ratan & Dawson, 2016).

Since the inception of the Proteus effect research, the focus on avatar traits has been its distinguishing feature, and therefore the contexts of this line of research are often VR or video games where players are allowed to use avatars or even choose or modify their appearance. Researchers have examined virtual environments of both first-person (e.g., Yee & Bailenson, 2007) and third-person point of view (e.g., Ash, 2016), and some even found virtual settings with a hybrid first-and-third-person point of view (Ratan & Dawson, 2016). In a meta-analysis of 46 experimental studies on the Proteus effect, Ratan et al. (2020) found that the overall effect size of these studies ranged from .22 to .26 depending on specific inclusion and exclusion criteria, which was a small but approaching medium effect based on Cohen's rule of thumb (Cohen, 2016).

Why Does the Proteus Effect Matter?

On the one hand, studying the Proteus effect has practical implications, especially given the popularity of games and VR in today's media environment. Statistics show that as of 2021, there are over 171 million VR users in the world, and 57.4 million VR users in the United States, which is roughly 15% of the US population (Zippia, 2022). In terms of games, there are 2.96 billion gamers around the world in 2022 (Newzoo, 2022), and around 66% of Americans play video games regularly (Entertainment Software Association, 2022). Recent research also illustrated that virtual environments could be used for a large variety of meaningful purposes, such as education and surgical training, sports and exercise, social science experiments, arts and cultural heritage, and industrial collaboration (Slater & Sanchez-Vives, 2016). And since avatars are an important part of virtual environments such as VR and games, studying the Proteus effect could broaden our understanding of how the virtual worlds affect individuals, and how to design avatars to promote the above-mentioned positive outcomes. So far, research has examined how virtual environments could be utilized to reduce stereotypes and bias (Banakou et al., 2016), to voice the need for diversity (Tymińska, 2020), and to increase physical activity and exercise-related attitude or behavioral intention (Peña & Kim, 2014; Li et al., 2014), which are all research directions of important practical implications.

On the other hand, research on the psychological effects of individuals’ self-representation in virtual environments also has theoretical implications. In some ways, the Proteus effect is related to classical entertainment media effects research, where transportation and identification with characters could play an important role in how media affect individuals (e.g., Moyer-Gusé & Nabi, 2010). However, it is noteworthy that some technological affordances of virtual environments provide experiences that are either unique to virtual environments or qualitatively different from other mediated environments, and therefore research on avatar effects could shed light on the unique mechanisms of such experiences and contribute to theory development in this new area.

For example, research on the psychology of entertainment media has examined the concept of immersion or transportation, which could be defined as “the feeling of being ‘lost’ in a theory” or “an integrative melding of attention, imagery, and feelings” (Green et al., 2004, p. 312). With traditional entertainment media, achieving transportation requires a creative and elaborative narrative and readers’ cognitive effort to transport themselves into the narrative world (Van Laer et al., 2014); however, with virtual environments, immersion and transportation become easier due to some modalities (e.g., 360-degree camera) and interactivity with the environment or other avatars. In addition, the virtual embodiment or self-representation is almost a unique phenomenon in virtual environments, as it allows users to choose or change their avatars’ physical appearance at a high level of freedom. This is closely related to the concept of self in social psychology research: scholars have identified multiple dimensions of the concept of self, including the actual or real self (i.e., the status quo), the ideal self (i.e., hopes or aspirations about who we want to be), and the ought self (i.e., individuals’ own or significant others’ sense of who they should be)(Cohen et al., 2019; Strauman, 1996). We would expect that through self-embodiment in an avatar with some positive or socially desirable traits, individuals might experience some “ideal self”; and by interacting with others in virtual environments, they may also form some ideas about “ought self” based on their own perception and their inferred expectation from others; finally, the experienced “ideal self” and “ought self” in virtual environments might also shape individuals’ “actual self” in turn. This unique dynamic of different dimensions of the self-concept is again largely induced by the self-representation affordances of virtual environments. Therefore, research on the Proteus effect can not only answer the “how avatars affect individuals” question, but also examine the extent to which classical social psychology and media effects theories still hold true in mediated communication that takes place in virtual environments and whether new theories should be proposed to fill in any gap, which eventually contributes to theory development in both communication and related fields.

Reflection on Current Literature

So far, the Proteus effect has been extensively applied by many scholars from various fields. These relevant studies, however, have produced heterogeneous results. In this section, this study will reflect on the current literature and introduce some major theoretical, empirical, and methodological issues in previous research on the Proteus effect.

Theoretical Basis: Is Priming an Alternative Interpretation?

As mentioned above, the Proteus effect has been frequently used as a theoretical framework in communication research (e.g., Peña & Kim, 2014; Sherrick et al., 2014; Van Der Heide et al., 2013), and its essential argument is that individuals would think and behave differently according to the appearance of their avatars in virtual environments (Yee & Bailenson, 2007). The theoretical basis of the Proteus effect, as Yee and Bailenson articulated, was self-perception theory (Bem, 1972), which argues that individuals infer their own internal states (e.g., cognition and emotion) from observing their own overt behavior (namely avatar appearance in the context of the Proteus effect) and the surrounding environment.

While self-perception theory can potentially explain why individuals tend to behave consistently with the traits of their avatars, some other scholars have proposed alternative explanations of the same phenomena, and the priming effect (Bargh, 1996) has been identified as a powerful alternative interpretation in the previous literature. To start with, Peña et al. (2009) proposed that an alternative mechanism of the phenomena associated with the Proteus effect is through priming effect (Bargh, 1996), which proposes that situational cues could activate related concepts, stereotypes, or memories that are consistent with the cues without individuals being aware or intended to do so. Peña et al. (2009) argued that the key difference between the Proteus effect and the priming effect lies in that the former examined the effect on behavioral dimensions while the latter on cognitive dimensions. However, this argument might not accurately capture the difference between these two theoretical perspectives. The reason is that, as mentioned earlier in this essay, the categories of dependent variables (behavioral or cognitive) cannot distinguish between the Proteus effect and the priming effect: despite the fact that the initial proposition of the Proteus effect only stated behavior as a dependent variable (Yee & Bailenson, 2007), subsequent research has demonstrated that avatar traits also affected cognitive outcomes such as exercise attitude (Li et al., 2014), body-related thought (Fox et al., 2013), and aggressive thought (Eastin et al., 2009). In addition, the Proteus effect's original theoretical basis, the self-perception theory (Bem, 1972), also included cognitive elements by pointing out that individuals can infer their emotions and cognition from their behavior and environment. For these reasons, it can be inferred that the scope of the Proteus effect has been extended to cognitive outcomes as well, so the major difference between the priming effect and the Proteus effect mechanisms is not in the categories of dependent variables.

Then, can we distinguish the Proteus effect and priming effect in interpreting the findings of avatar effects? In fact, while it can be challenging to completely separate the Proteus effect and priming effect, we can at least partially distinguish the effects of these two by manipulating the level of self-embodiment in virtual environments. This is because the Proteus effect is essentially a theory about digital self-representation, so a distinctive characteristic of this theory is that the avatar should be some sort of virtual self. The priming effect, however, does not need these traits to be a part of the self-concept: simply having these situational cues presented to audiences would be enough to trigger the priming effect. To distinguish between the Proteus effect and the priming effect, one solution might be to manipulate the level of self-embodiment at the same time when manipulating avatar traits in a factorial experiment; and by comparing effect size, we can observe whether the self-embodiment has any effect (i.e., the Proteus effect) on top of the effect of traits as situational cues only (i.e., the priming effect).

One previous experimental study specifically tested the Proteus effect and priming effect using this rationale: in response to Peña et al.'s (2009) argument that priming could also be a potential explanation for the phenomena associated with the Proteus effect, Yee and Bailenson (2009) used a 2 (attractiveness: high vs. low) × 2 (embodiment: high vs. low) between-subjects experimental design to figure out whether the Proteus effect can be alternatively explained by the priming effect. As a result, they found a significant interaction effect between attractiveness and embodiment, such that in high embodiment conditions (where participants were exposed to their avatars in a virtual mirror), the effect of attractiveness is significantly larger on several behavioral outcomes compared to in low embodiment conditions (where participants were shown the video recording of the same avatar). This interaction effect indicates that the priming effect itself cannot fully explain the phenomena associated with the Proteus effect, and the embodiment does play an important role in these outcomes. To be more specific, it can be argued that priming still plays a role in those conditions involving high levels of virtual self-embodiment due to the presence of situational cues in avatars, and therefore the priming effect is always an integral part of the Proteus effect. Nevertheless, the findings of Yee and Bailenson (2009) suggest that the Proteus effect is a unique mechanism at least partially due to the element of self-embodiment, and therefore is not interchangeable with the priming effect in terms of theoretical explanation.

Inconsistent Empirical Findings and Potential Reasons

Another critical issue among current research on the Proteus effect concerns inconsistent empirical findings. Specifically, the Proteus effect proposes that individuals will “take on” the traits of their avatars, which can be reflected in their cognition and behavior (Yee & Bailenson, 2007), and this theoretical proposition was supported by some empirical findings. For example, Peña and Kim (2014) found that when playing tennis video games, females were assigned to use normal-weight self-avatars engaged in significantly more physical activity compared to those assigned obese self-avatars. Similarly, Fox et al. (2013) found that females who were assigned sexualized avatars reported significantly more body-related thoughts compared to those who were assigned non-sexualized avatars. These findings were consistent with the prediction of the Proteus effect, indicating that the traits of avatars could influence individuals’ cognition and behavior accordingly.

However, there are also many studies on the psychological effects of avatarsof which the findings were non-significant or even in the opposite direction of the prediction of the Proteus effect. To start with, some empirical studies have shown that avatar traits did not significantly affect users’ cognition or behavior (e.g., Ash, 2016; Guegan et al., 2020), suggesting that the Proteus effect may not always occur in virtual settings. Moreover, some studies even found that avatars affect individuals in an opposite direction as the Proteus effect: for example, Van Der Heide et al. (2013) found that compared to females who used attractive avatars, those who used unattractive ones were indeed reported by both the conversational partner and naïve observers to have engaged in significantly more positive relational communication in terms of similarity/depth and immediacy/affect dimensions. This is the opposite of what the Proteus effect would predict—that individuals with attractive avatars would engage in more positive relational communication with their partner compared to those with less attractive avatars. In addition, in another study (McCain et al., 2018), the researchers assigned half of the female participants to use the avatar of Kim Kardashian while the other half a generic female avatar and their results showed that participants who were assigned Kim Kardashian avatar did not take on her luxury purchase behavior and in fact reported lower narcissism compared to those assigned a generic avatar, which again is inconsistent with what the Proteus effect would predict. In summary, these findings suggest that the Proteus effect may not always hold true, or at least not always strong enough that can counterbalance other theoretical mechanisms that would predict avatar effects in other directions.

Given these inconsistent findings, we may wonder why avatar traits would affect individuals in different ways. According to previous studies, one potential reason might be related to individuals’ cognitive processes that guide their behavior in reaction to avatars. Going back to classical social psychology theories, one limitation associated with the Proteus effect perspective is that it is essentially similar to social learning theory (Bandura & Walters, 1977), which proposes that people learn from their social interactions and environments (although in the case of the Proteus effect, people actually learn from their own avatars instead of others); however, a common criticism of social learning theory was that it ignores the cognitive process in one's learning: in other words, people's cognitive mechanisms matter in this learning process and may affect the learning outcomes, which is the essential idea of the later social cognitive theory (Bandura, 2001). Another analogy would be from the mass communication field, namelythe change of paradigm from strong and direct media effects to uses and gratifications as well as moderate media effects, indicating increasing attention on media users’ cognitive processes in the media effects research paradigm. Similarly, one inherent limitation of the Proteus effect would be that it is similar to a social learning and direct media effect perspective, and it largely ignores individuals’ cognitive elaboration throughout the learning process. For this reason, if we take individuals' cognition into consideration, we will have some explanations for why the findings of some studies are inconsistent with the predictions of the Proteus effect.

Specifically, previous literature suggested some theoretical bases that could lead to the opposite direction of outcomes from what the Proteus effect would predict. For example, Van Der Heide et al. (2013) mentioned that a competing mechanism for the Proteus effect would be behavioral compensation, which is drawn from computer-mediated communication research (e.g., Tong & Walther, 2015). Specifically, Van Der Heide et al. (2013) stated that in the virtual environment, if individuals perceive that their avatars are not attractive, they would expect their confederate to have a negative perception of themselves. As a result, these individuals make efforts to behave in a manner that minimizes or compensates for the negative expectations and perceptions of their confederates. By contrast, if individuals perceive that their avatar is attractive, they would not engage in these extra efforts in terms of self-presentation because they would not expect any negative perception or expectation from their partner. Based on this theoretical mechanism, Van Der Heide et al. (2013) proposed two competing hypotheses, one from the Proteus effect perspective and the other from the behavioral compensation perspective, and their results, as mentioned earlier, supported the behavioral compensation hypothesis instead of the Proteus effect. In addition, as described above, McCain et al. (2018) found that participants who were assigned to use the avatar of Kim Kardashian did not take on her luxury purchase behavior and in fact reported a lower level of narcissism compared to those assigned to a generic avatar, and the researchers also provided some explanation on why this would happen: individuals tend to dissociate themselves from negative or undesirable feedback about oneself, therefore there could be a boundary condition of the Proteus effect in that people might only take the desired traits from avatars and intentionally distance themselves from undesired traits. These mechanisms all shed light on the cognitive processes that could drive individuals to behave differently from what the Proteus effect would predict.

Methodology: Merits, Flaws, and Limitations

The Proteus effect is by its nature a theoretical perspective of media effects. Therefore, to establish causal relationships, experimental research methods are frequently used for the Proteus effect research. In this section, this study will review some strengths of previous research on the Proteus effect and summarize some common flaws in this line of research.

There are several methodological strengths in the existing literature on the Proteus effect. To start with, since avatar traits are the independent variables of the Proteus effect research, in experimental research design, it is important to ensure that this manipulation is successful. In this regard, most previous studies have selected or designed the traits that were valid enough to pass manipulation checks (e.g., Fox et al., 2013; Yee & Bailenson, 2007), which is a good sign in terms of internal validity. Random assignment is also a practice that was adopted in most previous experimental studies on the Proteus effect. Furthermore, some researchers made additional efforts to eliminate the influence of potential confounding variables: for example, in Yee and Bailenson (2007), considering that the perception and expectation from the confederate could be a confound, the researchers managed to set the confederate blind to the condition (the attractiveness and height of the participants’ avatar). In this way, internal validity could be enhanced.

However, there are also limitations in terms of methodology in existing research on the Proteus effect. To start with, while some researchers made efforts to control confounding variables, many experimental studies still had confounding variables that might have impaired their internal validity. For example, in Peña et al.'s (2009) Study 1, the researchers randomly assigned participants to use either white-cloaked avatars or black-cloaked avatars in a three-avatar group where all avatars wore cloaks of the same color, and participants were instructed to have a group discussion on a given task. While this study found that participants who used black-cloaked avatars showed more aggressive attitudes and intentions, it is hard to determine whether the effect came from seeing one's own avatar in black or seeing a group of individuals in black cloaks. Such confounding variables would threaten the internal validity of research findings even in experimental studies, not to mention that there are some survey studies without manipulation and random assignment of avatars (e.g., Stavropoulos et al., 2021), where it would be even harder to rule out third variables and establish causal relationships. Moreover, it is noteworthy that many previous studies had a small sample size, meaning that there might not have been enough statistical power to draw the conclusions. Also, most Proteus effect studies used convenient samples of undergraduate students, which limits the generalizability of these findings.

Recommendations

Based on the reflection on current research on the Proteus effect in terms of theoretical basis, empirical evidence, and methodological issues, this paper makes some recommendations for future Proteus effect research regarding how to address these problems. These recommendations include elucidating the cognitive process related to the Proteus effect, differentiating the Proteus effect from the priming effect, and methodological improvement for the Proteus effect research.

Elucidating the Cognitive Process Related to the Proteus Effect

Section “Inconsistent Empirical Findings and Potential Reasons” shows the Proteus effect is essentially like social learning theory (Bandura & Walters, 1977) and the direct media effects paradigm, while largely ignoring the cognitive processes that take place in this learning process. The previous section also discussed the competing theoretical perspectives that could lead to effects in the opposite direction of the predictions of the Proteus effect. To further this line of research, future research should make efforts to understand the cognitive mechanisms related to how avatars affect individuals, on which this study also proposes some concepts and mechanisms that could be employed to better understand such cognitive processes.

To start with, individuals’ evaluations of the avatar traits can play a role in how these traits affect them. As in the example of McCain et al.'s (2018) study, the traits of narcissism and luxury purchase behavior associated with Kim Kardashian are probably not desirable for most participants, and as a result, participants did not take on such traits and behavior. By contrast, in the study of Yee and Bailenson (2007), since height and attractiveness are generally considered as socially desirable traits, participants were more likely to take on these traits to. Based on these findings, future studies could add users’ evaluations toward the avatar traits as a moderating variable and see whether such evaluations affect the strength and direction of the Proteus effect.

In addition, the types of avatars and user–avatar relationships could also matter in how avatars affect individuals. Research in the game studies field has shown that depending on game genres and settings, there exist multiple types of avatars and player–avatar relationships. For instance, Kromand (2007) identified two continua forming four axes that categorize avatars in video games: first, avatars can be categorized based on the continuum of closed–open, depending on whether an avatar's personality traits are pregenerated by game developers or gained from players’ choices; second, the continuum of central–acentral identification refers to whether the player identifies with the avatar from an emotional first-person perspective or as a third person. Likewise, player–avatar relationships can also be classified into different categories (Banks, 2015; Banks & Bowman, 2016). According to Banks and Bowman (2016), there exist both parasocial (one-way, nondialectical) player–avatar relationships and social (two-way, dialectical) ones, and such relationships generally fall into four types: avatar-as-object, avatar-as-me, avatar-as-symbiote, and avatar-as-social other. This typology was supported by subsequent empirical findings (e.g., Tymińska, 2020). Furthermore, depending on the types of avatars and player–avatar relationships, the avatar-based storytelling practices also vary (Tymińska, 2016). Given these previous findings on the heterogeneity of avatars and player–avatar relationships, we should probably not assume that avatars would affect users in a consistent way through similar mechanisms across all game/VR settings. Hence, future scholars should consider the role of avatar types and player–avatar relationships in shaping users’ cognition and behavior, as they may either interact with avatar effects or lead to different theoretical mechanisms underlying the Proteus effect.

Relatedly, the specific context of the virtual environments also matters. Two previous studies (Banakou et al., 2016; Groom et al., 2009) are good examples of how context shapes the effect of avatars on individuals. Both these studies examined whether assigning individuals to a Black avatar could affect implicit racial bias, but their results were in opposite directions to each other: Groom et al. (2009) found that individuals who were assigned to use black avatars showed significantly more implicit racial bias than those who were assigned to use white avatars, while Banakou et al. (2016) found that embodying white individuals in a black avatar significantly decreased their implicit racial bias, reflected in a before–after change. In discussing the inconsistency between their results and those of the previous study (Groom et al., 2009), Banakou et al. (2016) proposed that the context of the virtual environment might have played a role: they argued that Groom et al. (2009) used a job interview scenario that itself might prime stereotype activation and racial bias, whereas in Banakou et al.'s (2016) study the context is a Tai Chi class in VR which is less likely to result in racial bias. This informs us that the context of virtual environments might change individuals’ mindsets, which in turn shapes the extent to which people “learn” traits, cognitions, and behaviors from their avatars.

Lastly, individual differences might also influence what and how people take on from the avatars. Some relevant individual differences could be personality traits. For example, openness to experience traits might drive individuals to actively experience and learn from the avatars, and self-monitoring traits can make individuals more willing to adapt their behavior according to the environment, therefore reinforcing the Proteus effect. In addition to these individual difference factors that can predict the strength of the Proteus effect in general, some individual difference factors would be more closely related to certain specific avatar traits. For example, individuals with a higher level of trait empathy might report a decrease in implicit bias to a greater extent after being embodied in a black avatar compared to those who have low level of trait empathy. In addition to these dispositional personality traits, it is also likely that individuals’ situational states in terms of mood or emotions could also affect the extent to which the Proteus effect takes place, which can be associated with mechanisms of mood management theory (Zillmann, 1988) or emotion regulation (Bartsch et al., 2008) perspectives.

Differentiating Proteus Effect With Priming Effect

As discussed in the “Theoretical Basis: Is Priming an Alternative Interpretation?” section, the priming effects has been brought up by some scholars as an alternative interpretation of the phenomena associated with the Proteus effect, because it can be argued that avatar traits are situational cues that can activate related concepts, memories, and stereotypes in one's mind without individuals’ awareness or intention (Bargh, 1996). Although priming seems to be an integral part of how avatars affect individuals, the Proteus effect researchers should make efforts to highlight the distinctive characteristics of avatars as self-representation, instead of using situational cues only without appropriate self-embodiment.

On this aspect, the level of self-representation and self-embodiment matters in how much we can distinguish between the Proteus effects and priming effect, and future researchers can consider adopting factors that could lead to a higher level of avatar self-representation in virtual environments. For instance, it is likely that first-person perspectives would elicit more feelings of self-embodiment compared to third-person perspectives, and the perceived realism in virtual environments can also play a similar role. User-avatar similarity, wishful identification with avatars, and level of avatar customization might also increase the feeling of self-representation. In fact, some previous studies have considered these concepts related to self-representation. For example, Ratan and Dawson (2016) proposed a new concept, avatar self-relevance, which they defined as “the extent to which the user of a mediated representation that appears and/or behaves like the user is closely connected to the self” (p. 1067) and suggested that avatar self-relevance could potentially moderate avatar effects. Also, Ash (2016) found that embodiment moderated the relationship between avatar race and in-game behavior. To further build on this line of research, future scholars should consider using various technological affordances in VR or games to increase users’ feelings of self-representation in avatars, instead of just showing an avatar and telling participants something like “look, this is you” as manipulation. The higher level of self-representation that the experiment provides, the more confident that researchers can be in arguing that the outcomes are due to the Proteus effect instead of the pure priming effect.

Methodological Improvement for Proteus Effect Research

Section “Methodology: Merits, Flaws, and Limitations” reviewed some flaws and limitations of the research methods that previous Proteus effect studies applied, and here this study will give some concrete recommendations on potential methodological improvements that can help to address these limitations.

To start with, researchers should use more representative samples with larger sample sizes. As mentioned in the previous section, some methodological flaws in the Proteus effect research include that most previous studies used convenient samples of undergraduate students, and many of them had a small sample size (<20 individuals in each experimental condition). Such issues can limit the generalizability of Proteus effect research, and therefore future researchers should attempt to have a larger and/or nationally representative sample to increase external validity.

Moreover, future scholars can also consider conducting longitudinal studies to examine the long-term effects of avatars on individuals. Although Yee et al. (2009) indicated that the Proteus effects could persist in subsequent face-to-face interactions following the VR experience, they measured the outcomes immediately after participants left the VR setting, and the same practice was adopted in many other previous studies (e.g., Groom et al., 2009). A few scholars have conducted longitudinal research (e.g., Banakou et al., 2016), but overall there has been limited research about the Proteus effect with a longitudinal design. Therefore, more research should be done to figure out how long the Proteus effect persists, and whether continuous usage of avatars could reinforce the Proteus effect and produce long-lasting influence. Future studiescould also combine the Proteus effect research with the media prescription research perspective (Prestin & Nabi, 2020) to elucidate the mechanisms of long-term avatar effects.

Conclusion

With the development of games and VR, understanding how virtual avatars affect individuals has become increasingly important. This study provides an overview of the Proteus effect research in terms of its development as well as commonly used variables and measures. Then, this essay highlighted the theoretical and practical implications of studying the effects of avatars in today's media environment. Following that, this study summarized some common problems in the previous literature, specifically the inconsistent theoretical explanations of the Proteus effect, the heterogeneous and sometimes conflicting findings among previous studies, as well as methodological issues related to confounding variables and sample size.

Based on these issues in previous research, this paper proposes three recommendations for future research. First, scholars should make efforts to elucidate the cognitive processes underlying the psychological effects of avatars, especially how the effect is affected by various situational and dispositional factors such as users’ evaluations of the avatars’ traits, the type of avatars and player–avatar relationships, the context of the virtual setting, as well as individual differences. Second, to distinguish the Proteus effect from the priming effect, researchers should aim to have a higher level of users’ self-embodiment in the research design to increase users’ feelings of self-representation. Finally, future studies should use a larger sample size to gain adequate statistical power and use more representative samples to increase generalizability. Longitudinal studies would also be beneficial to examine the durability of the Proteus effect, as well as whether frequent use of avatars over a long time produces any long-lasting effect on individuals.