Examining the Proteus Effect in the Context of Healthy Food Choices and Intentions to Eat Healthy: The Role of Avatar Body Size, Avatar Allocation Type, and Visual Perspective

Obesity is a major health problem in western society, leading to increased health risks, like cardiovascular disease, several forms of cancer, and strokes (World Health Organization, 2003). The increasing prevalence of obesity requires innovative approaches to promote healthier lifestyles, one of which involves avatars. Avatars, representations of oneself in digital environments, have the potential to positively change individuals’ health behavior (Rheu et al., 2020). For example, avatars are utilized in virtual reality (VR) weight-loss interventions to reflect the user's future appearance, motivating them to achieve their weight loss goals, or to demonstrate desired behavior that can be mirrored by the user (Horne et al., 2020). Recognizing the potential behavioral effects of avatars is valuable when adopting them in health-related interventions. The Proteus effect, defined as the tendency of avatar users to adapt their behavior to their avatar's appearance, is crucial to consider in these contexts (Clark, 2019; Yee & Bailenson, 2007).

The Proteus effect can be explained with self-perception theory (Bem, 1972), which postulates that people infer their beliefs, attitudes, and behaviors from observing themselves. Translating the principle of self-perception theory to digital environments, avatars could enhance self-perception processes, given that users witness their virtual representation in action. The Proteus effect manifests when beliefs and attitudes about oneself are inferred from the avatar's behavior, which is often associated with stereotypical actions based on the avatar's appearance (Yee et al., 2009). It has been demonstrated for a variety of outcome measures, such as aggressiveness (Ash, 2016; Beyea et al., 2022; Peña et al., 2009), walking speed (Reinhard et al., 2020), math performance (Ratan & Sah, 2015), antisocial behavior (Yoon & Vargas, 2014), and physical activity (Kocur et al., 2020; Li et al., 2014; Lin et al., 2021; Lin & Wu, 2021; Peña et al., 2016; Peña & Kim, 2014), and even physiological measures such as heart rate (Kocur et al., 2021).

According to the Proteus effect, avatars can exert both positive and negative influences on user behavior, depending on whether the avatar's appearance aligns with stereotypes associated with healthy or unhealthy behavior. To illustrate, playing a game with an overweight avatar led to less physical activity in that game than playing with a healthy weight avatar (Peña et al., 2016; Peña & Kim, 2014). Similarly, embodying a muscular avatar (Kocur et al., 2020) or a sweaty avatar (Kocur et al., 2022) enhanced users’ physical performance. Potential negative influences of overweight avatars on health behaviors should be considered, especially since a previous meta-analysis has shown that the effect size of the Proteus effect is relatively larger than related digital media effects (e.g., effects of video gaming on prosocial behavior; Ratan et al., 2019).

Two societal developments underscore the importance of studying behavioral effects of avatars. First, with the development of the Metaverse, it is likely that avatars will be increasingly integrated in our daily lives, gaining a more prominent role in society (Anderson & Rainie, 2022). Moreover, avatars have been increasingly adopted in health-related applications, such as serious games for lifestyle interventions (Baranowski et al., 2008; Dunwell et al., 2015), or within mHealth applications (Taylor et al., 2022). This calls for a better understanding of avatars’ impact on real-world behaviors, including the Proteus effect. Second, recent calls for more inclusivity in gaming (Kafai et al., 2016; Dove, n.d.; Waszkiewicz, 2021) suggest a potential increase in diversity of virtual representations. Recognizing this trend, research should also focus on capturing possible unintended effects resulting from yet underrepresented avatar appearances, such as overweight avatars.

The present study adopts a non-immersive virtual environment (i.e., operated on a flat screen) to study the effect of overweight avatars on health behavior, even though weaker results are expected compared to immersive VR environments (Beyea, Ratan, et al., 2022). Given the widespread accessibility and integration of non-immersive virtual avatars in our daily lives (e.g., social media platforms, gaming), it is crucial to examine avatars and their behavioral effects within non-immersive virtual environments specifically.

When it comes to adopting avatars for obesity treatment and prevention, the avatar's body size is an important appearance characteristic that deserves research attention. Until now, the majority of the studies examined the effect of avatar body size on physical activity. Several studies found that in line with the Proteus effect, avatar users adapt their physical activity levels to the appearance characteristics of their avatar. To illustrate, controlling an overweight avatar resulted in less movements made by its user (Peña et al., 2016; Peña & Kim, 2014), or a worse in-game performance (Li et al., 2014), than playing with a healthy-weight avatar. A study also demonstrated physiological and perceptual responses to the avatar's body shape, as cycling with an athletic avatar in VR resulted in a lower heart rate during cycling and a lower perception of effort (Kocur et al., 2021).

However, research findings are ambiguous when it comes to the effects of avatar body size on eating behavior. In line with the Proteus effect, a study found that among participants who had the intention to lose weight, those who embodied a weight-reduced ideal self in VR consumed less ice cream after the experiment than those who embodied their present self (Kuo et al., 2016). In a different study, participants were instructed to create an avatar reflecting the actual self, the ought self or the ideal self (Sah et al., 2017). As expected, they found that participants who had customized an avatar based on the ought self (compared to the actual self) were more likely to align their health goals with their behavior, as reflected in the healthiness of their (virtual and actual) food choices (Sah et al., 2017). Although hypothesized, no such effect emerged when customizing the ideal self (Sah et al., 2017).

Other studies examining the effect of an overweight (vs. healthy weight) avatar reported null findings. Namely, it was shown that playing a video game with an overweight avatar did not result in differences in post-game eating behavior (i.e., amount of cookies eaten by the participant; Joo & Kim, 2017). Lastly, the Proteus effect was not found in an immersive supermarket environment, in which healthy weight participants had to do grocery shopping with either an overweight or a healthy weight avatar (Verhulst et al., 2018): embodying an overweight avatar did not result in more products bought in the unhealthy product categories than embodying a healthy weight avatar.

Considering the equivocal research findings, the first aim of this study is to contribute to the current body of literature by employing a more robust research methodology. Robustness is achieved through (1) employing multiple studies in the same virtual environment, providing a more rigorous examination under standardized conditions; and (2) utilizing two outcome measures related to healthy eating (i.e., intention to eat healthy and healthiness of food choices in the virtual supermarket), providing a more comprehensive understanding of the impact of overweight avatars.

A noticeable difference between the study that did find, and the studies that did not find, support for the Proteus effect in an eating-behavior context is the extent to which the avatar was connected to its user. The study that found support for the Proteus effect created an avatar based on the physical appearance of its user (Kuo et al., 2016), while the studies reporting null findings used a generic avatar (Joo & Kim, 2017; Verhulst et al., 2018). In other words, the lack of significant findings among studies examining Proteus effects on eating behavior could possibly be explained by this difference.

Avatar--self connection is defined as the psychological connection that an individual develops with a digital representation, perceiving the avatar as a reflection of themselves (Jin, 2010). Previous studies have identified several constructs that could increase the likelihood of the occurrence of the Proteus effect, which are related to the avatar--self connection, such as the possibility to decide on the avatar's appearance (Ratan & Sah, 2015), self-relevance (i.e., the extent to which an avatar user thinks the avatar is relevant to him/herself; Ratan & Dawson, 2016), self-presence (i.e., the extent to which some aspect of individuals’ media use is relevant to their body-schema, emotions, and identity; Ratan & Hasler, 2009), identification (Praetorius & Görlich, 2020), and the extent to which embodiment (i.e., the perception of the virtual body as being one's own physical body; Kilteni et al., 2012) is experienced (Gorisse et al., 2019).

The importance of the avatar--self connection in a context unrelated to health behavior is demonstrated by Yee and Bailenson (2009). They found that the occurence of the Proteus effect depended on how the researchers defined the avatar for participants. More specifically, if participants were led to believe that it was a representation of oneself, the Proteus effect was observed, while the Proteus effect was not observed with the exact same avatar framed as someone else. While research attention has been devoted to identifying constructs that could potentially affect the Proteus effect, considerably less research has actually investigated the conditions underlying the Proteus effect. The second aim of the present study is to fill this research gap, by exploring the role of two moderators---avatar choice (Study 1) and visual perspective (Study 2)---which are likely to enhance the avatar--self connection, and thus the Proteus effect.

The first experiment was designed to examine whether being able to choose one's own avatar, where the avatars varied on appearance characteristics (i.e., clothing style, hair style, and skin color), could lead to a stronger Proteus effect. It is argued that the Proteus effect is stronger when users perceive a strong integration between their own self-perception and the avatar's characteristics (Ratan & Dawson, 2016), and that through self-assigning the avatar, this integration can be facilitated. A prior study has already shown that extended self-presence (i.e., the extent to which a mediated identity is important to the individual; Ratan & Hasler, 2010) is stronger when controlling an avatar customized by its user, compared to one assigned to its user (Ratan & Dawson, 2016).

In the process of avatar customization, which differs from choosing an avatar in terms of the possibility to change specific avatar features (as in the present study), people often portray their avatar as an idealized self (Bessière et al., 2007; Ducheneaut et al., 2009; Messinger et al., 2008). For example, a study has shown that World of Warcraft players created their avatar more similar to their ideal self than to their actual self (Bessière et al., 2007). In relation to avatar body size, Ducheneaut et al. (2009) found that those with a high BMI customized avatars that significantly differed from themselves in terms of attractiveness, weight, and physical fitness. An interesting and yet unstudied perspective is whether having the option to choose between different overweight (i.e., typically not idealized) avatars leads to an enhanced avatar--self connection, compared to being assigned to an overweight avatar. The act of choosing between various overweight avatars (instead of being assigned to one) could possibly cancel out the effects that the visual discrepancies between the self (i.e., healthy weight) and the avatar (i.e., overweight) have on the avatar--self connection. In other words, while it is expected that the avatar--self connection is weaker when controlling an overweight avatar, this effect could be mitigated when choosing for this avatar. To sum up, the Study 1 hypotheses were:

The same effects are expected for unhealthy food choices in a virtual supermarket:

A 2 (avatar body size: overweight vs. healthy weight) by 2 (avatar allocation type: self-assigned vs. experimenter-assigned) between-subjects design was employed to study the main effect of avatar body size and moderating effect of avatar allocation type. Thereby, both food choices in the virtual environment and intentions to eat healthy for the next two weeks were included as dependent variables. Data were collected over a five-month period (from April to September 2020), and the study received ethical approval from the Research Ethics and Data Management Committee of Tilburg University under identification code 2020/027a.

Dutch participants were recruited through the university's human subject pool in exchange for one research credit, and via the researchers’ personal networks in exchange for €5,-. After registration, an inclusion questionnaire had to be filled out to assess the participant's eligibility for participation. Only participants with a BMI ranging from 18.5 up to and including 30 (i.e., non-obese) were included, to ensure that they would experience a discrepancy between the overweight avatar body size and their own body size. Furthermore, participants with an adjusted diet were excluded (i.e., participants with an eating disorder, diabetes, a medical diet, a religious diet, or a vegan diet), as this could be a confounding factor when measuring the effects of avatar appearance on food choices. From the included participants, 45 were removed from the analyses (see Figure 1).

The experimental manipulations and the shopping task of this study were implemented in the VirtuMart, a virtual 3D supermarket created by L.N. van der Laan with Unity3D, which was designed after the example of a Dutch supermarket chain. The VirtuMart is a tool often used to investigate food purchasing behavior (Blom et al., 2021; Hoenink et al., 2021; Smit et al., 2021). The VirtuMart includes almost 1,200 premium-brand and budget-brand products, where 19% of the products are considered healthy according to the Dutch dietary guidelines (Hoenink et al., 2021). For this study, the screen-based virtual supermarket was used, which could be downloaded and installed by participants on a laptop/computer, with no need for other technological equipment. Given that shopping behavior in virtual supermarkets is a good proxy of real-world shopping behavior (van Herpen et al., 2016; Waterlander et al., 2015), this outcome measure aligns with the objective to study behavioral responses to the Proteus effect.

Participants used a virtual avatar to complete the shopping task. They perceived the avatar from a first-person perspective (1PP, i.e., through the eyes of the avatar). When entering the virtual supermarket, participants were told that, in order to practice with the controls, they had to walk up to the mirror that was placed before the entrance of the supermarket. The main purpose of this was to ensure that participants had a clear picture of what their avatar looked like, without being obtrusive about the importance of the avatar's appearance.

Participants used their keyboard to navigate through the virtual supermarket with their avatar, and the viewpoint could be changed by moving the mouse. Clicking on products resulted in adding these to the shopping basket, which was visually represented in the supermarket, next to the avatar. The selected products could be viewed by pressing the “escape” key and navigating to “Show my shopping basket” or by changing the viewpoint to the shopping basket in the supermarket. The shopping list could be viewed by pressing the “L” key. Participants could exit the supermarket by walking to the checkout, or by pressing the “escape” key and navigating to “Go to checkout.”

For the shopping task, participants were instructed to buy products from the following product categories in an amount they would need for the coming 24 hours: bread or bread substitutes, sandwich spreads, drinks, desserts, and snacks. These categories were employed because within each of these categories, both healthy and unhealthy options were available. Within the product categories, participants were free to choose healthy or unhealthy alternatives.

The factor avatar allocation type consisted of two levels: self-assigned and experimenter-assigned. In the self-assigned condition, participants were exposed to either six healthy-weight or six overweight avatars in the inclusion questionnaire, and they had to choose one of the avatars for the shopping task. They were provided with the following instructions: “An avatar is a virtual representation. Choose which avatar you would like to use to fulfil the supermarket task, by clicking on this avatar.” Thereby, 94% of both males and females chose an avatar corresponding to their self-reported gender. In the experimenter-assigned condition, participants were also exposed to six healthy-weight or overweight avatars, but they were only able to look at the avatars without the ability to choose one of them. In this condition, the experimenter picked an avatar corresponding to the gender of the participant. Furthermore, the experimenter ensured that the avatars that were chosen in the choice condition were also equally used in the no-choice condition.

The study was communicated as if it were two separate studies, named “Shelf location and shopping behavior” and “Personality characteristics and choice for organic products,” to cover the link between the shopping task and the subsequent questions in the questionnaire. After registration for the study, participants received the inclusion questionnaire with questions to assess their eligibility for participation and questions related to the avatar allocation manipulation. When participants were eligible for the study, they received an email, including their participant number, the link to the questionnaire, and instructions on how to install the virtual supermarket. First, participants had to download the virtual supermarket as instructed. Afterwards, they had to open the questionnaire, in which they agreed to the informed consent and filled out their participant number, some demographic questions, and cover questions about their personality. Next, it was explained how to navigate to the virtual supermarket, in which they first received the instructions for the shopping task, and then they entered the virtual supermarket.

Upon supermarket entrance, participants had to perform the practice task by walking towards the virtual mirror to see their virtual representation, and then continued performing the shopping task. After completion, they were redirected to the survey to answer questions measuring self-presence, presence, avatar characteristics, autonomy, diet and weight importance, intentions to eat healthy, perceived susceptibility, perceived severity, self-awareness, beliefs about obese persons, and manipulation checks. After completion, participants received a debriefing in the survey, in which the true purpose of the study was revealed.

Food choice healthiness was measured by taking the mean Nutri-Score of the participant's purchased products. The Nutri-Score is a point system that assesses the healthiness of a product by allocating positive points for energy, total sugar, saturated fatty acids and sodium, and negative points to fruit, vegetables, nuts, fiber and protein (Chantal & Hercberg, 2017). The scale that is derived from this runs from −15 (most healthy product) to +40 (least healthy product) and is divided into five categories (A to E), representing the Nutri-Scores. The five-point scale was used as a measure of healthy food choices, whereby 1 indicates least healthy and 5 indicates healthiest (M = 2.58, SD = 0.59).

Participants indicated to what extent they agreed to the statements “I'm planning to eat healthily for the next two weeks” and “I expect that I will eat healthily for the next two weeks” on a seven-point Likert scale. The two items formed a reliable scale (α = .85, M = 5.40, SD = 1.12).

Participants had to indicate for the following statements to what extent they agreed with it (1 = totally disagree, 7 = totally agree): “The virtual avatar had more bodyweight than I have,” “The virtual avatar was overweight,” and “The virtual avatar's body weight resembled my own bodyweight.” The latter statement was reverse-coded, and the scale was reliable (α = .84, M = 4.60, SD = 1.85).

Participants were asked whether they received the possibility to choose an avatar by answering the question with yes or no. They also had to answer the statements “I had the feeling that I could decide for myself what my avatar looked like” (M = 2.43, SD = 1.76) and “I believe that there were sufficient avatars available to choose from” (M = 2.34, SD = 1.66) on a seven-point Likert scale.

To explore whether participants who were shopping with an overweight avatar indeed experienced their avatar heavier-weighted than the participants shopping with a healthy-weight avatar, an independent two-sample t-test was performed, including the avatar body size (healthy weight vs. overweight) as independent variable, and the manipulation check questions (e.g., the virtual avatar had a higher bodyweight than I have) as dependent variable. The manipulation check revealed that there was a significant difference between the healthy weight and overweight condition in terms of their assessment of the weight of the avatar, t (132.54) = −19.34, p < .001, 95% CI [−3.39, −2.84]. Those in the overweight conditions assessed their avatar as being more overweight (M = 6.29, SD = 0.96) than those in the healthy-weight conditions (M = 3.12, SD = 0.96), which indicates that the manipulation of avatar body size was successful.

A chi-square test was used in order to analyze whether participants correctly recalled how they were allocated to the avatar. The chi-square test showed a significant association between the answer to this question and the condition the participant was assigned to, χ² (1) = 47.86, p < .001. In the experimenter-assigned condition, 81% answered correctly that they could not choose the avatar (vs. 19% answering incorrectly that they could choose an avatar). In the self-assigned condition, 78% answered correctly that they could choose an avatar (vs. 22% answering incorrectly that they could not choose an avatar). A second manipulation check was performed to assess whether respondents in the self-assigned condition experienced more freedom in terms of avatar choice than those in the experimenter-assigned condition. The results showed that there was a significant difference between the experimenter-assigned and self-assigned condition in terms of the assessment of whether one was able to decide what the avatar looks like, t (127.54) = −4.68, p < .001, 95% CI [−1.88, −0.76]. In the self-assigned condition, participants had a stronger feeling that they could decide what their avatar looked like (M = 3.10, SD = 1.80) than in the experimenter-assigned condition (M = 1.79, SD = 1.47). Hence, the manipulation of avatar allocation type was considered successful.

Hypotheses were tested using ANOVAs, including avatar body size as independent variable, avatar allocation type as moderator, and intentions to eat healthy or food choice healthiness as dependent variables.^1,2 Results are summarized in Table 1, and plotted in Appendix A.

There was no significant main effect of avatar body size on intentions to eat healthy and food choice healthiness, which led to the rejection of H1a and H2a. Additionally, the interaction effect of avatar body size and avatar allocation type on intentions to eat healthy and food choice healthiness was not significant. Therefore, H1b and H2b were rejected.

Since these results of Study 1 did not provide any evidence for the Proteus effect in a non-immersive virtual supermarket setting, some exploratory analyses were performed that provide more insights into the experimental manipulations of this study. The first analysis tested whether, according to the Proteus effect, stereotypical thoughts about overweight were indeed more strongly elicited when embodying an overweight avatar, compared to a healthy-weight avatar. The second analysis focused on whether the avatar--self connection is indeed enhanced when being able to self-assign, instead of being assigned to an avatar.

Regarding the first analysis, a seven-point Likert scale was adopted to assess participants' stereotypical beliefs about whether the avatar would eat “healthy” (1) or “unhealthy” (7), and whether the avatar would eat “much” (1) or “little” (7). The results showed that there was a significant difference between the overweight condition compared to the healthy-weight condition on the assessment of the healthiness of the food the avatar eats, t (111.41) = −7.04, p < .001. Those in the overweight condition were more likely to attribute unhealthy eating to their avatar (M = 5.03, SD = 1.79) than those in the healthy-weight condition (M = 3.07, SD = 1.32). Similarly, a significant effect was revealed for the amount that the avatar would eat, t (105.93) = 4.90, p < .001. Those in the overweight conditions indicated that the avatar would eat a larger amount of food (M = 2.84, SD = 1.53) than those in the healthy weight condition (M = 3.97, SD = 1.04). Hence, it can be inferred that participants indeed held stereotypical beliefs, suggesting that the absence of the Proteus effect was not due to the manipulation failure.

Regarding the second analysis, it was argued that self-assigning an avatar would lead to a stronger avatar--self connection compared to being assigned to an avatar, which could foster the Proteus effect. Proto self-presence (i.e., the extent to which an avatar is integrated into body schema) and extended self-presence (i.e., the degree to which the avatar is important to the individual) were used as proxies to test whether this connection was indeed stronger among those choosing an avatar themselves, using the self-presence questionnaire (Ratan & Hasler, 2009). Two items were dropped from the extended self-presence scale, as these items only applied to customization of avatars, and not to allocation (i.e., ”How much effort did you put into making your avatar's sex clear to others?” and “How much effort did you put into making your avatar's race clear to others?”). The proto self-presence subscale (e.g., “When using the virtual environment, how much do you feel your avatar is a part of your body?”) had a good reliability (α = .84, M = 2.64, SD = 0.79). The extended self-presence scale (e.g., “How much do you care about how your avatar looks?”) also had a good reliability (α = .80, M = 2.39, SD = 0.81). When merging the subscales, the scale was reliable (α = .84, M = 2.56, SD = 0.64). All the items were measured on a five-point Likert scale.

Proto-self presence did not increase after experiencing a self-assigned avatar (M = 2.54, SD = 0.78), compared to an experimenter-assigned avatar (M = 2.75, SD = 0.80), t (134.94) = 1.51, p = .133, 95% CI [−0.06, 0.47]. For extended self-presence, a significant difference was revealed. Those with a self-assigned avatar experienced higher extended self-presence (M = 2.54, SD = 0.78) than those with an experimenter-assigned avatar (M = 2.25, SD = 0.80), t (133.33) = −2.09, p = .039, 95% CI [−0.55, −0.01]. Although the effect size is small (Cohen's D = 0.36), it can be concluded that indeed, a self-assigned avatar increases the avatar--self connection, compared to being assigned to an avatar. An additional mediation analysis was performed to investigate whether the effect of avatar body size and avatar allocation type on intentions to eat healthy and food choice healthiness was mediated by extended self-presence, but the mediation analysis did not yield significant results (see Appendix B for the findings).

The first aim of this study was to assess whether, in line with the Proteus effect, grocery shopping with an overweight avatar would lead to more unhealthy food choices in the virtual supermarket and less intentions to eat healthy. However, this study did not provide evidence for the Proteus effect, as there was no effect of avatar body size on these dependent variables. The second aim was to assess whether the freedom to choose an avatar, compared to being assigned to an avatar, would strengthen the Proteus effect. Although extended self-presence was higher when self-assigning an avatar compared to being assigned to an avatar, no moderation effect was found of avatar allocation type, indicating that self-assigning did not affect the occurrence of a Proteus effect. There was also no main effect of avatar body size on the dependent variables. Further discussion of the null findings is provided in the general discussion.

In Study 2, it was examined whether the Proteus effect depends on the visual perspective from which the user perceives the avatar. The assumed theoretical basis for the potential influence of visual perspective on the strength of the Proteus effect is related to the avatar--self connection. The first-person perspective (1PP), in contrast to the third-person perspective (3PP) mirrors the way the world is naturally perceived and interacted with, which could contribute to a cognitive and emotional bridge between the user and avatar. A stronger sense of “being” the avatar is felt when controlling an avatar from the 1PP (e.g., Debarba et al., 2015; Scoresby & Shelton, 2011); hence it is likely that the avatar--self connection becomes stronger in the 1PP compared to the 3PP. Since there are no studies that the authors are aware of which have directly compared 1PP and 3PP effects when studying the Proteus effect, it is unclear whether these perspectives impact the Proteus effect differently.

Nonetheless, even in the absence of a direct comparative analysis between these two perspectives, they separately yielded significant results in previous studies (e.g., Kuo et al., 2016; Li et al., 2014; Yee & Bailenson, 2009). Until now, no studies employed a non-immersive 1PP setup (as the current study does) to investigate the Proteus effect, but only an immersive 1PP VR setup, whereby the physical body is replaced for an avatar body in the virtual environment. Regarding the literature, the majority of the 1PP immersive studies found support for the Proteus effect (e.g., Kuo et al., 2016; Yee & Bailenson, 2009), except for Verhulst et al. (2018).

When looking at studies employing a 3PP, there is also variation in types of setups. A number of studies used Wii Fit games to study the Proteus effect (Li et al., 2014; Peña et al., 2016; Peña & Kim, 2014). These studies all found (partial) support for the Proteus effect. Other studies used a 3PP in a non-immersive gaming setup (e.g., Ash, 2016; Joo & Kim, 2017; Peña et al., 2009), leading to varying results. For example, Joo and Kim (2017) used a 3PP in a non-immersive gaming setup, whereby no Proteus effect was found on healthy eating behavior. Studies did find effects of avatar color (through skin tone: Ash, 2016; through clothing: Peña et al., 2009) on aggressiveness. These varying results could be an indication that the Proteus effect is not effective for all types of behaviors.

In conclusion, it is difficult to draw inferences about the influence of perspective on the Proteus effect from previous studies, since a wide variety of methods is used, and previous studies did not directly compare the 1PP and 3PP. Since the avatar--self connection is considered stronger in a 1PP, it is expected that the Proteus effect is stronger in a 1PP. The hypotheses for this study are as follows:

The same effects are expected for unhealthy food choices in a virtual supermarket:

The hypotheses were tested using a 2 (avatar body size: overweight vs. healthy weight) by 2 (visual perspective: 1PP vs. 3PP) between-subjects design, with healthiness of food choices and intentions to eat healthy for the next two weeks as dependent variables. Data collection ran simultaneously with Study 1 (from April to September 2020), and the study received ethical approval. Recruitment strategies and inclusion criteria were identical to those of Study 1 (Figure 1).

Participants were under the assumption that they signed up for two separate studies (i.e., “Supermarket cues and behavior in a virtual supermarket” and “Personality and snacking-choice”), to prevent participants from becoming suspicious of the study goals and the link between the shopping task and the questionnaire. After registration for the study, participants received the inclusion questionnaire, in which they had to answer questions to assess their eligibility for the study. Furthermore, they were asked which of the provided hair and skin colors represented their color the most, to enhance the feeling of having a self-representation in the virtual supermarket. Based on this information, an avatar with matched skin color, hair color, and gender was selected by the experimenters. When participants were eligible for the study, they received an email, including their participant number that was linked to the matched avatar, the link to the questionnaire, and instructions on how to install the virtual supermarket. These instructions were identical to the instructions of Study 1. After completion of the supermarket task, participants answered questions measuring presence, avatar characteristics, diet and weight importance, intentions to eat healthy, perceived susceptibility and perceived severity, body size perception, body satisfaction, beliefs about obese persons, immersion, and manipulation checks. Lastly, participants received a debriefing in the survey, in which the true purpose of the study was revealed.

Measurements of the intentions to eat healthy (M = 5.30, SD = 1.26) and healthiness of food choice (M = 2.74, SD = 0.66) were identical to those of Study 1. A question was added to be able to conduct a manipulation check on visual perspective. Participants were asked to indicate from which perspective they saw the avatar (answer options: 1PP, 3PP, or don't know). Definitions of the perspectives were provided with the question.

A manipulation check was conducted to compare the assessment of the avatar's overweight between participants in the overweight condition and participants in the healthy-weight condition. A t-test showed that the manipulation was successful, t (127.98) = −17.30, p < .001, as those in the overweight condition perceived their avatar as more overweight (M = 6.32, SD = 1.05) than those in the healthy-weight condition (M = 3.10, SD = 1.07).

Another manipulation check was done to assess whether participants correctly remembered from which perspective they saw the avatar. The chi-square test showed a significant effect, χ² (2) = 25.41, p < .001, indicating that the manipulation check was successful. From those in the 1PP condition, 66% correctly answered that they saw their avatar from the 1PP, whereas 23% answered that they saw the avatar from the 3PP, and 11% indicated they did not know the answer. In the 3PP condition, 63% answered this question correctly, 23% answered it incorrectly, and 13% did not know.

Hypotheses were tested using two ANOVAs, of which the results can be found in Table 2.^3,4 For intentions to eat healthy, the ANOVA showed that there was no main effect of avatar body size, as hypothesized in H3a. Hence, H3a was rejected.

For food choice healthiness, the ANOVA revealed no significant effect of avatar body size. Furthermore, the interaction effect of avatar body size and visual perspective on food choice healthiness was non-significant. Therefore, H4a and H4b are rejected (results are plotted in Appendix C).

The aim of Study 2 was to investigate whether visual perspective could affect the strength of the Proteus effect, such that controlling an avatar from a 1PP would lead to a stronger Proteus effect, as the avatar--self connection would become stronger, compared to a 3PP. Two unexpected results were found in this study: (1) an effect of avatar body size was found in the 3PP instead of 1PP, and (2) the direction of the effect was opposite to the Proteus effect. In other words, in the 3PP, intentions to eat healthy were stronger when controlling an overweight compared to a healthy-weight avatar.

An explanation for observing an effect in the opposite direction could be that those controlling an overweight avatar might have experienced the increased body size as a fear-eliciting stimulus, rather than a stereotype-eliciting stimulus. Thereby, the elicited effects might stem from fear appeal theory rather than from the Proteus effect. Some evidence for this explanation can be found in the work of Ahn (2015) and Owen et al. (2018), who found that exposure to self-similar avatars that suffered from the consequences of unhealthy behavior (i.e., soft-drink consumption, sun exposure) led to increased perceived susceptibility to those consequences. A difference between these previous studies and the present study is that the present study did not link the unhealthy behavior explicitly to the negative health consequences. The results from the current study add to previous findings that even without explicitly linking the unhealthy behavior to the negative health consequences, avatar characteristics (and more specifically, avatar body size) may be interpreted as a fear appeal (Witte & Allen, 2000), leading to behavior aimed at mitigating negative health consequences associated with the avatar's appearance. To further explore this notion, the third study aimed to investigate whether explicitly posing a health threat, in the form of a textual message, could potentially enhance the positive effect of avatar body size on intentions to eat healthy.

Another unexpected finding of Study 2 is that the positive effect of avatar body size on intentions to eat healthy occurred only in the 3PP and not in the 1PP. An explanation for this outcome could be found in studies that examined effects of visual perspective using imagination tasks. For example, imagining oneself eating candies from a 3PP led to a lower number of candies eaten in a taste-test than imaging oneself from a 1PP (Christian et al., 2016), which arguably is due to the emphasis on an unfavorable self-concept. In the current study, it could be that viewing their avatar from 3PP led to similar effects. Another explanation could be that participants in the 3PP were consistently exposed to their avatar, whereas those in the 1PP condition had to look down to see their avatar's body. Although all participants were instructed to view their avatar in the mirror at the beginning of the supermarket task, it is likely that those in the 1PP condition were exposed to their avatar's looks at a lower intensity, which could have inhibited effects. Analyses showed that this did not lead to differences in perceptions of the avatar's body size between the 1PP and 3PP.

The results of Study 2 led to the question why those controlling an overweight avatar demonstrated stronger intentions to eat healthy than those controlling a healthy weight avatar in the 3PP. The findings suggest that an overweight virtual avatar may have functioned as a fear stimulus (instead of a Proteus effect eliciting stimulus), leading to behavioral intentions that would prevent someone becoming overweight. Specifically, the overweight avatar may have served as a fear-inducing visual depiction of the health threat. In the third experiment we integrated a second fear-inducing stimulus in the form of a health message (in all conditions), to examine whether explicitly linking the behavior (i.e., unhealthy eating) to the health threat (i.e., overweight) would facilitate avatar-induced fear appeal effects. Based on the findings of Study 2, it was expected that controlling an overweight avatar would lead to stronger intentions to eat healthy than controlling a healthy-weight avatar.

As explained by the Extended Parallel Process Model (EPPM), when people are confronted with a threatening health message, both perceived susceptibility (i.e., the degree to which you feel vulnerable to the threat) and perceived severity (i.e., the degree to which you assess the threat as something severe) are expected to determine their behavioral response to the threat (Witte & Allen, 2000). More specifically, if susceptibility and severity are perceived as high, someone is more likely to adopt preventive behavior. These variables are therefore included in Study 3 as potential mechanisms that could explain the effect found in Study 2. The link between appearance-focused interventions and health behavior including these mediators has only been marginally investigated before (for studies using similar variables, see Ahn (2015) and Owen et al. (2018)). In line with the findings from Study 2, it is expected that the positive effect of controlling an overweight versus a healthy weight avatar on intentions to eat healthy, mediated by perceived susceptibility and severity, only occurs in the third-person perspective, and not in the first-person perspective. Specifically, it is expected that:

The design, experimental manipulations, recruitment strategies, and inclusion criteria were identical to those of Study 2.

The procedure was identical to Study 2, except for the health message that was shown on the screen before participants started the task in the virtual supermarket environment.

The measures of the dependent variables food choice healthiness (M = 2.74, SD = 0.64) and intentions to eat healthy (M = 5.02, SD = 1.28) were identical to Studies 1 and 2. The mediators perceived susceptibility and perceived severity were measured using the Risk Behavior Diagnosis Scale (Gore et al., 2005; Witte et al., 1996). The measures were adapted to the examined risk (i.e., becoming overweight), and items were measured on a seven-point Likert scale. Specifically, perceived susceptibility was assessed by the statements “I am at risk for overweight,” “It is possible that I will develop overweight,” and “I am susceptible to becoming overweight” (M = 2.36, SD = 1.49; α = .92). Perceived severity was measured with the items “Overweight is a serious threat,” “Overweight is harmful,” and “Overweight is a severe threat” (M = 6.01, SD = 1.05; α = .93).

An additional question was asked to check whether participants had seen the health message. A total of 76% indicated they read the health message and remembered the content, 20% read the message but did not remember the content, and 4% did not read the health message.

The same manipulation checks were performed as in Study 2. An independent samples t-test showed that participants controlling an overweight avatar indeed perceived their avatar as being more overweight (M = 6.39, SD = 0.88) than did those controlling a healthy-weight avatar (M = 3.21, SD = 1.15), t (123.41) = −18.01, p < .001, indicating that the manipulation was successful. Furthermore, a chi-square test revealed that, for the 1PP versus the 3PP condition, participants answered the question “From which perspective did you see the avatar?” significantly different, χ² (2) = 45.95, p < .001. From those in the 1PP condition, 80% answered the question correctly, while 13% thought they were viewing their avatar from a 3PP, and 7% did not know. From those in the 3PP condition, 63% correctly answered the perspective question, while 23% wrongly thought they viewed the avatar from a 1PP, and 14% indicated they did not know the right answer.^5,6

Two moderated mediation analyses were run with Hayes Process Macro for R, one with intentions to eat healthy and one with food choice healthiness as the dependent variable. Avatar body size served as independent variable, visual perspective as moderator, and perceived severity and perceived susceptibility as mediators. Bootstrapping was used for both analyses since the dependent variables were not normally distributed. The number of bootstrap samples were set to 10,000. Unstandardized regression coefficients, p-values, and bootstrapped confidence intervals are reported in Tables 3 and 4. For the sake of completeness, Appendix D shows the means and standard deviations of the mediators and dependent variables for each experimental group.

It was hypothesized that controlling an overweight avatar would lead to a higher perceived susceptibility (H5a) and a higher perceived severity (H5b), but only in the 3PP and not in the 1PP. However, these interaction effects were not significant, leading to the rejection of H5a and H5b (Table 3). The interaction effects on perceived susceptibility and perceived severity are plotted in Appendix E.

The effects of perceived susceptibility (H6a) and perceived severity (H6b) on intentions to eat healthy were both not significant. Thus, both H6a and H6b are rejected, indicating that the effect of avatar body size on intentions to eat healthy is neither mediated by perceived susceptibility nor by perceived severity.

The moderated mediation analysis was repeated with food choice healthiness as dependent variable (Table 4).⁷ There was no interaction effect between avatar body size and visual perspective on perceived susceptibility (H5a) and perceived severity (H5b). Hence, H5a and H5b were rejected.

In H7, positive effects of (a) perceived susceptibility and (b) perceived severity on food choice healthiness were hypothesized. However, these effects were both not significant; therefore, H7a and H7b are rejected.

All the analyses were repeated while including only the participants who remembered the health message; the results can be found in Appendix F. A significant and positive effect was found of avatar body size on perceived severity, b = 1.22, p = .048, for the moderated mediation on food choice healthiness. More specifically, those controlling an overweight avatar reported a higher perceived severity (M = 6.20, SD = 0.71) than those controlling a healthy weight avatar (M = 5.86, SD = 1.17). The remaining effects were not significant.

Although Study 2 provided initial support that effects in line with fear appeal theory would occur when controlling an overweight avatar, this was not confirmed when looking at the results for the whole group in Study 3. However, among those who saw the health message, a higher perceived severity was reported when controlling an overweight avatar compared to when controlling a healthy weight avatar, indicating that psychological processes related to fear-appeal theory may play a role when a more direct link is provided between the unhealthy eating and the development of obesity. Possible explanations for these different findings are given in the general discussion.

According to the Proteus effect, avatar users in a virtual environment behave according to the stereotypes elicited by their avatar's appearance. The first aim of the present research was to test if the Proteus effect occurs in the context of healthy food choices in a virtual supermarket environment, when manipulating avatar body size. The second aim was to investigate the effect of two moderators (avatar allocation type in Study 1, visual perspective in Study 2) that were assumed to influence the strength of the Proteus effect. Study 3 further explored the unexpected findings of Study 2, testing whether controlling an overweight avatar could lead to stronger intentions to eat healthy, since the overweight avatar may have functioned as a fear-inducing stimulus. In particular, fear appeal literature would suggest that the elicitation of fear leads to desired health behavior (e.g., eating healthily), while the Proteus effect would suggest that elicitation of stereotypical thoughts leads to undesired behavior (e.g., eating unhealthily), when controlling an overweight avatar. The studies presented in this paper yield valuable insights for avatar-behavior research in several ways. Firstly, a scrutinized approach was used to demonstrate the Proteus effect, using the same virtual environment across three studies. The methodological consistency enhances the robustness of our findings, providing reliable evidence for the absence of the Proteus effect. Secondly, this is the first study that demonstrated fear appeal effects (Rogers & Deckner, 1975; Witte et al., 1996) instead of the Proteus effect (Yee & Bailenson, 2009) as a response to controlling an overweight avatar. Findings from the present study show that perceived severity is stronger when controlling an overweight avatar, which could be a trigger to study avatar effects in the context of fear appeals. The coexistence of fear appeal and self-perception theory as explanatory frameworks for avatar-behavior effects could lead to new research directions, for example, aiming at understanding under which circumstances each of the effects manifests.

Contrary to expectations, none of the studies found support for the Proteus effect, as hypothesized in H1a, H2a, H3a, and H4a. In domains other than eating behavior, the Proteus effect has been demonstrated rather consistently with different experimental setups, for example, immersive virtual environments (Yee et al., 2009), Wii-Fit (Li et al., 2014; Peña et al., 2016; Peña & Kim, 2014), and non-immersive games (Ash, 2016; Joo & Kim, 2017; Peña et al., 2009), though effects were shown to be stronger in immersive setups (Beyea, Ratan, et al., 2022). Also, the Proteus effect has been demonstrated across different outcome measures, for example, aggressiveness (Ash, 2016; Peña et al., 2009), walking speed (Reinhard et al., 2020), physical activity (Li et al., 2014; Peña et al., 2016; Peña & Kim, 2014), math performance (Ratan & Sah, 2015), and antisocial behavior (Yoon & Vargas, 2014). Although in the current studies, participants did hold eating-related stereotypes about overweight avatars (i.e., eating too much, eating unhealthily), these stereotypes were not translated into behavior in line with the stereotypes (i.e., making less healthy food choices, having less healthy eating intentions). A few studies that examined the Proteus effect for eating behavior reported null findings as well, in both an immersive VR setup (Verhulst et al., 2018) and a non-immersive setup (Joo & Kim, 2017). Our findings, combined with these earlier studies on the Proteus effect for eating behavior, suggest that the Proteus effect might not be applicable to all behavior types, at least not for eating behavior.

In the realm of health behavior, studies examining physical activity repeatedly demonstrated the Proteus effect. The distinct results retrieved in physical activity versus eating behavior studies may be explicable by comparing their methodological setup. In previous studies examining physical activity, users performed the real-life behavior in sync with their avatars, that is, the avatar's movements corresponded to the user's movements (e.g., Li et al., 2014). This synchronization could have enhanced cognitive fluency, defined as the ease with which avatar information is processed (Sah et al., 2021). High cognitive fluency, for example, facilitated by synchronized movements, potentially enables users to seamlessly integrate avatar information into their self-concept (Sah et al., 2021), increasing the likelihood for a Proteus effect. However, this explanation contradicts findings of Verhulst et al. (2018), who also synchronized movements and did not find a Proteus effect when manipulating avatar body size.

Nevertheless, if cognitive fluency is indeed an important precondition for the Proteus effect, future studies could focus on how this fluent processing can be stimulated. In the present studies, participants in the overweight condition assessed their avatar as significantly less similar in terms of appearance than participants in the healthy-weight condition. Although appearance similarity is proposed as a way to enhance cognitive fluency (Sah et al., 2021), many Proteus effect studies successfully incorporated avatars with non-similar appearance (e.g., fictitous World of Warcraft characters; Stavropoulos et al., 2021). Hence, for setups in which appearance dissimilarity is integrated, researchers could investigate the impact of fluent processing by means of other aspects, such as identity aspects (e.g., personality), technological aspects (i.e., amount of control over the avatar, avatar choice), and task-related aspects (e.g., familiarity with the avatar). A previous study already demonstrated that familiarity with the avatar, operationalized as a task in which participants had to identify their own avatar among several other avatars, increased the perceived embodiment, which in turn moderated the Proteus effect (Beyea, Van Der Heide, et al., 2022). Other ways in which fluent processing could be facilitated should be further examined.

Another explanation for the null findings could be that the Proteus effect is more likely to occur when the avatar's appearance characteristics are associated with positive traits instead of negative traits. From a literature review, Praetorius and Görlich (2020) concluded that an avatar portraying undesirable characteristics can be a boundary condition for the Proteus effect. To illustrate, McCain et al. (2018) discovered that virtually embodying a narcissistic character did not lead to the adoption of narcissistic traits. However, contradictory findings were reported in a different study, where avatars with negative traits elicited more aggressive intentions (Peña et al., 2009). Similarly, the Proteus effect was found in studies that included overweight avatars to negatively manipulate the user's physical activity level (Li et al., 2014; Peña et al., 2016; Peña & Kim, 2014), which can also be considered a negative trait. Given these conflicting results, it remains unclear whether the null findings of the present study are indeed attributable to the negative traits portraited by the avatar. Future research should focus on studying specific avatar traits, to provide a more comprehensive understanding of the specific avatar characteristics that could lead to the Proteus effect.

Study 1 could not support the moderating role of avatar allocation type on the effect of avatar body size on intentions to eat healthy (H1b) and food choice healthiness (H2b). The current study did show that self-presence, which is related to the avatar-self connection, was higher for those being able to choose an avatar, compared to those being allocated to an avatar. Previous studies also found that psychological processes, such as identification with the avatar (Koulouris et al., 2020) and presence in the game (Bailey et al., 2009), were affected by avatar customization (i.e., creating one's own avatar). Contrastingly, Beyea, Van Der Heide, et al. (2022), who investigated the Proteus effect in text-based communication, did not find an effect of choosing the avatar on the experience of embodiment (operationalized as perceiving the avatar as a digital representation of oneself). Thus, research is inconclusive on whether psychological processes related to the avatar--self connection are affected by avatar creation and allocation. This emphasizes the need for more research on the relationship between avatar creation processes (e.g., customizing or choosing an avatar) and behavior. Another interesting research angle would be how specific conditions under which the avatar can be created, such as the number of avatar choices or amount of time spent on customizing the avatar, could affect behavioral changes.

In Study 2, moderation effects were hypothesized for visual perspective (H3b and H4b), such that the effect of avatar body size on intentions to eat healthy and food choice healthiness would be stronger in the 1PP compared to the 3PP, as the avatar--self connection is hypothesized to be stronger in the 1PP. However, Study 2 did not provide evidence for this.

A potential explanation is that the effect of visual perspective may be more likely to occur in immersive setups, where the sense of presence (i.e., the experience of being in a virtual environment) is more salient. When being fully immersed in a 3D virtual environment, it is more likely that someone will experience a stronger avatar--self connection in the 1PP compared to the 3PP, as this is also the way in which individuals perceive the real world. Studies indeed found that the Proteus effect was only demonstrated under those perceiving a greater embodiment (Ash, 2016) or higher spatial presence (Reinhard et al., 2020). Currently, no studies have directly compared a 1PP and 3PP in the context of the Proteus effect, nor have they investigated the effects of a 1PP in a non-immersive setup. Thus, for this study, it is unknown whether the null findings are derived because of lacking differences between 1PP and 3PP or because of a low level of immersion. Future research should therefore employ an experimental design in which the interaction between visual perspectives and technological setups are examined.

It could also be that the effects of gameplay features are more complex than suggested in this paper. A previous study found that visual perspective did not lead to differences in reported presence when the game players chose their avatars (Lim & Reeves, 2009). When the game players did not choose their avatars, the 1PP led to higher presence than the 3PP. Although these results do not match the null findings of Study 2, it is likely that combinations of gameplay features may also lead to distinct psychological effects, which in turn could affect the Proteus effect. Future studies could examine these interaction effects in order to establish clear conditions under which the Proteus effect occurs.

In Study 2, a significant interaction effect of avatar body size and visual perspective was found on intentions to eat healthy, but in the opposite direction than the Proteus effect. Namely, those controlling an overweight avatar had stronger intentions to eat healthy than those controlling a healthy-weight avatar, but only in the 3PP. This finding could be an indication that the overweight avatar is perceived as a feared future self, stimulating individuals to change behavior to prevent becoming this version of the self (e.g., Ahn, 2018). To build further on this finding, Study 3 facilitated the interpretation of the avatar as a feared self with a health message. This message explicitly linked the health threat (i.e., obesity) to the health behavior (i.e., healthy eating). Furthermore, perceived susceptibility and perceived severity were included as potential mechanisms of the effects found in Study 2. No effect was found of avatar body size on intentions to eat healthy or product choice healthiness, and neither perceived susceptibility nor perceived severity functioned as mechanisms in this relationship. However, effects of avatar body size on perceived severity were marginally significant, and when repeating the analyses among only those who remembered the health message, the effect was significant. Although these effects did not translate into behavior, the present study shows that the perceived harm of becoming overweight can be heightened when controlling an overweight avatar, supporting the assumption that fear appeal theory could explain avatar appearance effects.

Some previous studies also point in the direction that fear appeal mechanisms could play a role in explaining the relationship between a self-relevant fear appeal and behavior. For example, exposure to an overweight virtual self (as a consequence of soft drink consumption) in combination with a pamphlet indicating the risks of obesity, led to thoughts of consuming soft drinks in the near future, which in turn led to higher perceived susceptibility, and this decreased soft drink consumption after two weeks (Ahn, 2015). In an experiment using a non-immersive setup, scholars found that those being exposed to an aged self had higher susceptibility of getting skin damage and stronger intentions to use sun protection, compared to those not being exposed to an aged self (Owen et al., 2018). Perceived susceptibility and perceived severity are still understudied in the context of virtual avatars, and further studies are needed to explore and differentiate between fear appeal effects and the Proteus effect. Thereby, other process indicators such as self-reported fear and skin conductance measures could be measured.

Questions that remain are when fear appeal effects occur (instead of the Proteus effect), which could be examined by future studies by integrating these two research perspectives. Taking this a step further, a possible explanation for the null findings on eating behavior retrieved in the current study could be that the Proteus effect and fear appeal effects occur simultaneously, and therefore cancel each other out. Dual processing might play a role here, whereby processing the avatar body either unconsciously or consciously could lead to different mechanisms and in turn, to different behaviors.

Several limitations should be taken into account when interpreting the study results. First off, a noticeable difference between the 1PP and the 3PP conditions is that participants were less intensively exposed to their avatars in the 1PP. This was partly accounted for by showing the participants their avatars in a mirror before the shopping task. Furthermore, data showed that the overweight avatar body was not perceived differently between the 1PP and the 3PP. Nevertheless, it could still be the case that regular exposure to the overweight avatar led to more frequent reminders of choosing the healthy products, and future studies should examine the role of exposure time to ensure that this was not a confounding factor in the present study.

Second, the present study considered only one variable, self-presence, as a proxy for the avatar--self connection. Especially since there are many related concepts that could affect the avatar--self connection (e.g., self-relevance, identification, embodiment), it would have been relevant to include additional mediators to be able to understand the potential effects. Future studies should include these mediators to provide a more comprehensive understanding of how the Proteus effect works.

Third, since the research was not conducted in a controlled lab setting, it is unknown whether individuals fully engaged in the study without being distracted. Given that the Proteus effect does not last long (Yee et al., 2009), and that some participants required more than one hour to partake in the study, it is plausible that the time between the supermarket task and questionnaire completion was too big to observe a Proteus effect. Therefore, future studies should employ a controlled research environment to effectively control for the duration of participation.

Although support has been found for the Proteus effect in many different setups and for a variety of behaviors, the current study did not find the Proteus effect on healthy eating behavior in a non-immersive virtual supermarket, when manipulating avatar body size. This implicates that the Proteus effect is a complex phenomenon to understand, which may be context-dependent and specific to certain behaviors. Findings were more in line with fear appeal literature than with the Proteus effect, showing that when controlling an overweight avatar, perceived severity (Study 3) and intentions to eat healthy (Study 2) were stronger rather than weaker. This study is the first study that brought these two research perspectives together, and future studies should attempt to disentangle these two types of effects by exploring the circumstances under which those effects occur.

The authors do not have any competing interests to declare.

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

The data that support the findings of this study are available from the corresponding author upon reasonable request.

The authors would like to thank Luc Wolfs, Natasja van Hummel, Roy Schuurkes, Sophie de Graaf, Teddy Eliëns, Daphne Buskoop, Fay van Tussenbroek, and Shahin Pazoki for their valuable contributions to the data collection process.

Study 1¹² showed that those with a self-assigned avatar experienced higher extended self-presence (M = 2.54, SD = 0.78) than those with an experimenter-assigned avatar (M = 2.25, SD = 0.80), t (133.33) = −2.09, p = .039, 95% CI [−0.55, −0.01]. To see whether extended self-presence played a role in the hypotheses that were proposed in Study 1, two moderated mediations were run, with avatar body size as independent variable, avatar allocation type as moderator, extended self-presence as mediator, and either intentions to eat healthy or healthiness of food choice as dependent variable. The results can be found in²Tables B1 and B2.

The analyses of Study 3 were conducted on the part of the sample that was able to remember having seen the health message. For this analysis, the 33 participants who did not see the health message were excluded from the analyses. The results of the two moderated mediations are reported in table F1 (intentions to eat healthy) and table F2 (food choice healthiness). Table F3 shows the means and standard deviations per experimental condition for the sub-sample that saw the health message.