2.1 Types of presence

In previous studies, presence has often been regarded as a unitary variable defined as the sensation of ‘being there’ (see [1, 30]. This definition is related to spatial presence, which in the context of VR represents a state of mind characterised by an impression of being amongst landmarks present in a virtual environment (VE) [31]. Because the ability to assess distances and locate oneself relies largely on technical features, these features have often been the focus of research. Studies using 2D screens or VR head-mounted displays (HMDs) have generally shown that technical features such as increased update rate [32, 33], tracking [34], field of view [35], image quality [36], stereoscopy [37, 38] and higher quality of sound [39, 40] increase spatial presence.

However, technical features may not be the only factors to influence the formation of presence. The influence of non-technical, user based factors was posited by Lombard and Ditton [41] who argue that presence is the “perceptual illusion of non-mediation”. Thus, presence may be influenced by both technical and user based factors. To account for this range of factors, Lee et al. [26] and later Tamborini and Bowman [27] divided presence into three distinct but interrelated dimensions: spatial presence, social presence and self presence. As noted above, much of the existent research has been directed to the study of spatial presence.

Social presence refers to users perceiving that they share a space with other conscious beings whose thoughts or emotions they can perceive and with whom they may have dynamic interactions [27, 42, 43]. Social presence seems to be of great importance to user experience in VR [42, 44]. Social presence has been researched substantially in the past, with studies identifying factors ranging from technical to contextual and individual differences being identified as contributing to its formation (see systematic review by Oh et al., [44]). For example, a study by Shin [42] asked participants for subjective appraisals of elements which would make them perceive themselves as present in a VE. The majority referred to presence as the extent to which they could feel they were together with other avatars in the VE. Participants also commented on the importance of feeling connected with other social users present in the VE when referring to presence [42]. In a recent study, Riches et al. [45] showed that in cases where avatars behave in an intelligible way, or in which interaction with avatars is natural, overall presence is greatly enhanced. However, despite emotions having been shown to influence feelings of social connectedness in the real world [46], relatively few studies investigated the role played by emotion in the formation of social presence in VR. One such example is a study by Sajjadi et al. [47] who attributed emotional states to avatars within a VR environment and observed that this had an impact on user levels of social presence. This means that not only users’ felt emotions but also those of other avatars can contribute to the formation of social presence. Still, the underlying process by which social presence is created and the exact role played by emotions in it are not yet fully understood.

Compared to social presence, self presence remains relatively under researched. Cummings and Bailenson [48] excluded social and self presence from their meta-analysis of the effect of immersive technology on presence because there were “simply too few studies in the existing literature that empirically examine the effect of particular immersive features on social and self presence” (p.277). Self presence has been defined as the psychological state in which users regard their virtual self within the VE as their actual self [26, 45]. Biocca [43] takes this point further and talks about the impression of existing as a social being within the VE, which is an extreme form of self presence and is related to social presence. Similar to social presence, the process that determines self presence and the role played by emotion in this process are yet to be fully understood.

These three dimensions of presence—spatial, social and self—should not be regarded as unrelated. It would be expected that they coexist to some extent for presence to be achieved. For example, if the user achieves self presence but other avatars in the VE lack psychological involvement or behavioral interaction, overall presence would be jeopardized. Similarly, if the formation of spatial presence is hampered (e.g. due to a lack of afforded interactivity), this would render the VE less believable as a real space, and thus social and self presence may in turn be hampered.

2.2 Models of presence

Studies have often treated presence as a unitary concept, which may have contributed to the myriad of contrasting theories on the formation of presence. Descriptive models of presence focus on distinguishing the components of presence and put less emphasis on the actual process that creates presence, hence their name. Thus, descriptive models are heavily impacted by technological approaches that strive to improve VR through technical advances. Early evidence of this trend is exemplified by Schubert et al.’s [49] Igroup Presence Questionnaire (IPQ), which identified spatial presence, involvement and realness as the main components of presence. The most prominent descriptive model is the Process model on the formation of spatial presence, proposed by Wirth et al.’s [1].

This model predicts that spatial presence is facilitated by the existence, believability and accuracy of spatial cues present in a VE. This is in turn dependent on whether the user is able to perceive the VE as a plausible space in the first place.

Despite the useful contribution of descriptive models, structural models of presence may offer a better understanding of self presence and social presence because structural models focus on how presence is created in the mind. The user is considered a mediator between bodily interoceptive states and real world exteroceptive signals [50]. Exteroception refers to the perception of the environment via classical sensory modalities, i.e. visual, auditory, tactile etc. Interoception, on the other hand, is the perception of physiological conditions in one’s own body [3, 51]. People create expectations about the stimuli they will encounter, which produce physiological responses that they can perceive (interoception). These internal states are subsequently compared with the actual stimuli that are subsequently perceived (exteroception).

2.3 Interoceptive Predictive Coding Model (IPCM)

Based on this concept of interoception, Seth et al. [3] formulated the Interoceptive Predictive Coding Model (IPCM) of Conscious Presence. This structural model stipulates that presence felt in the real world or in a VE is a product of the matching between predictions of interoceptive (emotional) states and actual states created when encountering external stimuli. Thus, users predict the type and intensity of emotion that a certain experience would elicit and then match the felt emotion with this prediction. Although there will always be a degree of mismatch, the IPCM postulates that presence is the result of a good match between these two processes. If the mismatch is greater, in order to achieve presence the user must exert additional effort to suppress information from the external stimuli that is incompatible with their expectations (See Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Figure from Seth et al. [3] that illustrates the fundamental architecture of the IPCM model.

“Both agency and presence components comprise state and error units; state units generate control signals (Aout, Pout) and make predictions [Apred, Ppred, Apred(p)] about the consequent incoming signals (Ain, Pin); error units compare predictions with afferents, generating error signals [Aerr, Perr, Aerr(p)].

https://doi.org/10.1371/journal.pone.0280390.g001

Applied to VR, the IPCM postulates that prior to experiencing a VE, the user would predict the emotions that the VE would elicit [3]. For example, before trying a scary VR experience, the user would predict high fear and low relaxation. This expectation would be accompanied by neurophysiological reactions in the user’s body [3]. If the intensity of the predicted fear is lower or higher than the actual fear felt during the VR experience, the user has to suppress this mismatch to achieve presence. This idea of suppression of expectations that are incompatible with the VR experience has been postulated before [50, 52]. Sheridan, in his estimation theory [52], talks about the necessity for continuous suppression of disbelief, which determines the formation of presence through a continuously updated interior model of the environment. However, the above-mentioned study focused on the prediction of events occurring in the environment [52]. In contrast, the IPCM emphasizes the prediction of internal states, thus placing the user at the centre of the presence formation process.

The IPCM is supported by neurophysiological and neuroimaging studies investigating the role of brain regions such as the anterior insular cortex (AIC). Most importantly, the AIC has been shown to be responsible for the comparison between predicted and actual interoceptive signals [53, 54]. This is what the IPCM describes as the process by which emotions contribute to the formation of presence. One study used VR to simulate a roller coaster experience while brain activity was measured using functional magnetic resonance imagery (fMRI). Several brain areas were activated more when participants reported strong presence, amongst which was the AIC, compared to when they did not, although this effect was only observed in a sample of children, but not adults. This partly supports the IPCM’s assumption that there exists a relationship between presence and predicted and felt emotions [3]. Moreover, hypoactivity of the AIC has been found in patients with depersonalization disorder (DPD). This condition is characterized through a loss of subjective sense of reality of self and of the world [55, 56]. This is further corroborating evidence that predictive coding may have an influence on presence. Nonetheless, more neuroimaging and neurophysiological studies are needed to clarify the exact role of emotion in the creation of presence at the neural level.

2.4 Presence and emotion

The relationship between presence and emotion has mostly been studied by means of correlations, so the directionality of any causation is impossible to derive [24]. For example, Baños et al. [24] exposed users to relaxing and joyful VR environments and found that presence and emotion intensity were positively correlated. Nonetheless, the authors acknowledge that this correlation does not elucidate whether higher presence leads to more intense emotions, or if this effect manifests in the opposite direction. Other experiments in VR exposure therapy have reported significant correlations between the level of presence and intensity of emotions [57–59]. According to Frijda’s [60] law of apparent reality, emotions are elicited in situations that are being perceived as real, and their intensity is directly proportional to the degree to which this is the case.

Several studies have assumed that presence drives emotion and not the other way around [50, 61, 62]. Still, some research points to an inverse effect; that is, emotions may drive presence. For example, while investigating the use of VR for treatment of arachnophobia, it was found that presence was enhanced when a spider was introduced in the VE [22]. A study by Baños et al. [23] independently manipulated technical features by presenting the same VE via an HMD, a computer monitor and a video wall. Emotional valence of the VE was also controlled: in one scenario it was sad while in the other neutral. Results showed an interaction effect between technical features of the system, presence and emotion. Technical features were found to affect presence only in the neutral condition and not in the emotion condition. Higher levels of presence were however found in the emotion condition, in line with previous research [22]. Similarly, VR has been used before as a distraction tool in patients with burn wounds, showing that it is more powerful than conventional media in reducing patient anxiety during painful medical procedures (see systematic review by Morris et al. [63]). Overall, these results suggest that emotional content could represent a stronger driver of presence than technical features.

2.5 Testing the IPCM

The IPCM, like other presence models, does not distinguish between different components of presence in VR, preferring to simplify it into a unitary concept. This is problematic because, as mentioned, predictive coding can be expected to apply mainly to social and self presence and less to the formation of spatial presence. There is a need to test how these components of presence are achieved individually and to what extent emotions and expectations contribute to their creation.

Secondly, emotions differ in characteristics such as arousal, with feelings of fear and anxiety characterised by high levels of arousal increasing with sense of presence [64] and emotions with low arousal such as relaxation not showing such an association to presence [24, 57]. This suggests that the relation between presence and emotion could be moderated by the type of emotion despite a similar level of arousal (e.g., happiness and fear) or by the level of arousal (e.g., fear and relaxation) [65].

Thirdly, distinct highly arousing emotions could manifest differently. Fear, unlike happiness, is an important evolutionary survival function that triggers fight or flight responses [66]. Due to its vital importance, fear can be elicited via two pathways at neural level: a perceptual (e.g. by seeing fear-inducing cues) and a conceptual pathway (e.g. information on an incoming fear-inducing stimulus). Perceptual fear-related cues are known to evoke instant physiological and behavioural reactions (e.g. seeing a snake). Similarly, conceptual information about fear-inducing stimuli can determine a weaker, yet significant physiological activation [67] (e.g. being informed that you will encounter a snake). This is supported by the Emotional Processing Theory which suggests that information about an imminent stimulus (priming) pertaining to an emotion in itself induces that emotion to an extent [68]. This process is also the basis for the prediction and subsequent matching of emotions on which the IPCM is based. In the present paper we will refer to such information provided to participants as priming. However, it remains unclear whether such effects would be present for other emotions such as happiness which, although arousing, do not have such strong importance for survival. To this end, the applicability of the IPCM will be tested with both fear and happiness. As a further comparison, the model will also be tested with relaxation as it is a low-arousal emotion that is well documented [24, 50].

Finally, the IPCM does not account for individual differences derived from personality traits. It has been shown that personality traits such as neuroticism, conscientiousness and extraversion have a significant moderating effect on emotion regulation, notably suppression [58]. In addition, it seems that personality traits play a decisive role in how users engage with factors determining presence [32, 35, 69–71]. For example, willingness to try, openness to new experiences and introversion are important factors which may moderate presence within a VE [29].

4.1 Experimental design

The experiment adopted a between groups design with seven participant groups. Three groups were primed with perceptual stimuli (PP) while another three received conceptual priming (CP). The seventh group received no priming and acted as the control group. Thus, the seven experimental conditions were as follows: PP + relaxation, PP + happiness, PP + fear, CP + relaxation, CP + happiness, CP + fear and the control group. From hereon we will distinguish between ‘predicted’ emotions, referring to the intensity of a given emotion that the user expected they would feel within the VR experience prior to being exposed to it, and ‘felt’ emotions, which represent the retrospective subjective appreciation of the intensity of emotions they felt in said VE. Within the experimental design, the predictor variables were the type of priming (conceptual or perceptual) and the scores for predicted and felt emotions (relaxation/happiness/fear). Additional predictor variables were computed as differences between predicted emotion scores and felt emotion scores for the three emotions. The five personality traits measured via the Big Five Inventory (BFI) were also used as predictors [72]. The outcome variables were spatial presence, social presence and self-presence, measured via subjective questionnaires, plus two overall presence measures used and standardized by previous literature [73, 74]. The two measures assessing overall presence were the UCL Presence Questionnaire [73] and item number two from a questionnaire used by Egan [74]. The study took place in the VR/motion capture lab at the University of [REDACTED].

4.2 Emotion measures

In order to test the IPCM, we computed scores from predicted and felt emotion intensity scores. Thus, predicted and felt emotion intensities had to be assessed in a similar manner and on a similar scale. We opted to collect self-report measures of emotion rather than, for example, physiological measures because the IPCM is based on subjective expectations of emotions. Previous studies such as that of Riva et al. [75] successfully utilised concise measures of emotion consisting of one question. Similarly, a recent VR study investigating the role of emotions in VR presence formation has utilised the same emotion measures (see. [76]). Thus, we replicated Riva et al.’s [75] questions assessing participant levels of relaxation, happiness and fear and used three items. We chose a 1–10 scale for the emotion questions because they were administered verbally by the experimenter and it was previously shown that a 10-point scale is easiest to respond verbally to [77]. The same scale was used by Baños et al. [23] in their study where they also assessed participant emotion within VR.

The questions measuring emotions felt during the VR experience were phrased as: ‘For the following emotions, how intensely did you feel them during the VR experience, 10 being the highest and 1 being the lowest?’. Below the question, relaxation, happiness and fear were listed, each with a scroll bar set by default in the middle of the scale, at 5. The scroll bar could be dragged to indicate the level of emotion felt. To measure predicted emotions before VR the same questions were rephrased: (‘Rate how intensely you think you will feel the following emotions after the VR experience, 10 being the highest and 1 being the lowest’). The items about emotion predictions were administered so as to later compute the difference between predicted and felt emotions.

4.3 Presence measures

To assess presence, we used a battery of questionnaires. We administered the UCL Presence Questionnaire as it is a well-established measure of presence [73]. It also has been successfully used in investigating the relation between emotion and presence in VR [75]. In addition, three custom items were created to assess spatial presence, social presence and self-presence respectively, as described by Lee et al. [26]. Thirdly, an item from Egan et al.’s [74] presence questionnaire was used since this study also correlated presence with emotional states. This item was the only one in Egan et al.’s questionnaire which referred to levels of presence. We used this range of measures of presence because previous studies have shown that correlations of presence with personality traits vary greatly depending on the questionnaire being used [78]. By using multiple measures we intended to verify whether results would indeed be inconsistent as past research would suggest. All presence questionnaires, including our custom one, were marked on a 7 point rating scale, which is what the authors proposing them suggested [73, 74]. From here onward our three custom measures of presence will be referred to as spatial, social and self-presence. For simplicity and clarity, we will refer to the UCL Presence Questionnaire as UCL Presence. Because it was used by Egan [74] as the sole measure for presence, we will refer to the second question from the questionnaire used by them as Overall Presence (see Table 1). In this study, none of the measures mentioned above were combined. Instead, we will assess and discuss them separately. As with measuring emotions, questionnaire measures were also preferred to physiological measures of presence because the latter have produced contradictory results in several studies [74, 79, 80].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. The UCL presence questionnaire [73], the custom items assessing spatial, social and self-presence and the second question (Q2) from Egan and colleagues [74].

https://doi.org/10.1371/journal.pone.0280390.t001

4.4 Personality measure

The Big Five Inventory (BFI) which contains 44 items [72] was used to measure participants’ personality traits. This was chosen because it was used in previous studies to investigate the effect of personality traits on emotion suppression and regulation [58], which may have an important effect in the presence creation process [3, 81]. Additionally, Kober [78, 82] used the BFI and showed that it correlates with presence in VR. Other studies investigating personality of participants in non-VR computer mediated environments have also used the BFI successfully [29, 83–85]. Five main personality traits were investigated: extraversion vs introversion, agreeableness vs antagonism, conscientiousness vs lack of direction, neuroticism vs emotional stability, and openness vs closedness to experience.

4.5 Audio files

For the perceptual priming conditions, three pieces of classical music were chosen for their ability to induce the three desired emotions. For relaxation, Chopin’s ‘Nocturne, op. 15: 6 in G minor’ was chosen [86]. For happiness, Beethoven’s ‘Symphony No. 6, 3rd movement’ was chosen [87, 88]. Lastly, Holst’s ‘Mars, the bringer of War, op. 32: 1’ from ‘The Planets’ was selected to induce fear [89]. Each piece was played for 90 seconds, similarly to other studies that induced emotions using the same pieces [87, 88].

For the conceptual priming conditions, three ‘reviews’ were written about the VR experience we used, portraying it respectively as relaxing, happy or fearful. The text of the three reviews was kept the same but key words related to emotions were changed accordingly (e.g. fearful to relaxing; unfriendly to friendly). The reviews were pilot tested on a sample of five people who estimated that the reviews reflected similar levels of relaxation, happiness and fear respectively. The reviews were then read aloud by a female voice and recorded using a high quality microphone. This was done to ensure that the music and reviews were presented via the same sensory modality, i.e. auditory.

4.6 Apparatus

An Oculus Rift HMD was used for presenting the VR environment. This headset has a maximum refresh rate of 90Hz and uses OLED panels with a pixel density of 1080x1200 per eye. To enhance sound quality, a pair of Sennheiser HD 380 Pro headphones was used instead of the built in Oculus Rift headphones. The HMD was powered by an Alienware Area 51 desktop computer running Windows 10, with a 3.4 GHz Intel Core i7, 16 gigabytes of RAM, and an Nvidia 1080 Ti graphics card with 11 gigabytes of GDDR5 memory.

4.7 Virtual reality experience

A VR experience from the Oculus Store named ‘Senza Peso’ [90] (see Fig 2) was chosen. This is an Italian opera-like experience dominated by intense visuals and vocals. Participants see themselves in a Venetian gondola that floats along a canal, surrounded by a surreal environment populated by characters placed on either side of the canal. Along the journey covered by the gondola, there are several environments which are visually stimulating and include fire and water cascades (see Fig 2). Additionally, several characters appear on the side of the water at certain times and they are performing actions such as searching with a lantern, or lighting a fire. This was also important since it was intended for the VE to elicit social presence as well. ‘Senza Peso’ was piloted together with a range of other candidate VR experiences prior to the current study with 36 participants (nine males, 27 females) over the age of 18 (mean age = 21, range = 18–44). They were asked to rate elicited emotions. ‘Senza Peso’ was interpreted by equal numbers of participants as relaxing, happy or fearful and it was therefore chosen for the main study in order to control for one emotion being evoked more compared to the others, which could have biased our results.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Screenshots taken from the user’s perspective in the VE.

The experience presented a surreal journey in a Venetian gondola. Figure b) shows a starry sky mirrored in a lake on which the gondola is floating. Figures a, c and d) show the gondola at various stages of the journey. The full experience is a VR version of a mini-opera by Cory Strassburger and Alain Vasquez [92]. It is available free from the Oculus Store [90].

https://doi.org/10.1371/journal.pone.0280390.g002

The experience lasted for approximately 7 minutes. This has been shown to be the ideal duration for a passive VR experience as the level of presence is expected to reach maximum levels around the 7 minute mark, after which boredom can ensue [19]. Because assessment of presence was done after the experience, the seven minute timing was important so as to avoid a recency effect whereby participants would report boredom although they had reported high presence during the experience. In order not to add agency as a potentially confounding factor, no active input or interaction by the participant was facilitated. The experience facilitated six degrees of freedom (three-axis rotational tracking plus three-axis positional tracking) for head movements. Participants did not have a perceptible body in the experience as this would have introduced new confounding variables in the experimental design (see [91]).

4.8 Procedure

Participants were met by the experimenter and were asked to read the information sheet and complete the consent form. Participants were randomly assigned to one of the seven experimental groups. The study adopted a between-subjects design. A pre-questionnaire on demographics was then completed by the participants. This was followed by the assessment of their initial levels of relaxation, happiness and fear, i.e. before any intervention. They were also asked to indicate their expectations of the emotions they would feel in the VR experience. This was done to allow assessment of potential biases resulting from participants’ existing knowledge of VR which could have influenced their a priori expectations of emotions.

Next, participants were fitted with the Oculus Rift and listened to the priming material while a Unity default infinite horizon was played to control for visual input. The only difference between the seven experimental conditions was the auditory content that participants were exposed to. Participants in the three perceptual priming experimental conditions listened to music either eliciting relaxation, happiness or fear. The other three conceptual priming experimental groups, in contrast, listened to the priming material eliciting relaxation, happiness or fear. Participants in the control condition only experienced the same infinite horizon for the same amount of time, but did not receive any priming prior to the VR experience. The VR infinite horizon also served to familiarize the participants with the Oculus Rift head tracking and allowed them to accommodate to any VE-specific distance compression [93]. The latter manifests as an underestimation of distance in VR and may have the potential to affect the formation of spatial presence due to erroneous perception of distances between the user and features in the VE [26, 41]. For participants in the conceptual conditions, the review was played portraying the experience as being relaxing, happy or frightening respectively. For participants in the perceptual conditions, the music selected to induce the corresponding emotion was presented (see Fig 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Visual representation of the study procedure in chronological order.

https://doi.org/10.1371/journal.pone.0280390.g003

After the priming phase, levels of felt and expected emotions were measured a second time for the three emotions, using the same questions as before. This second measurement of emotion expectations was conducted in order to test the effect of the priming interventions. Next, the ‘Senza Peso’ VR experience was presented. Before commencing the experience, participants were verbally encouraged to make use of the head tracking and look around as much as they liked and to anything in the VE capturing their attention. After this VR experience, the headset was removed and the levels of felt emotion were measured, followed by the battery of presence questionnaires. After the experimental phase, participants were administered the BFI via the Qualtrics platform and were debriefed.

4.9 Participants

A total of 126 participants (51 males, 75 females) over the age of 18 (mean age = 26.33, range = 18–65, SD = 8.71) were recruited via opportunity sampling. Each of the seven experimental groups consisted of 18 participants with counterbalanced numbers of males and females. Participants were randomly assigned to one of the experimental conditions. They were a mixture of undergraduate (39), masters (35) and PhD (19) students as well as university staff members (30) and others, who were university alumni (three). All participants had normal or corrected to normal vision and hearing. In terms of previous experience with VR, 42 declared they had ‘none at all’, 63 had ‘a little’, 17 had ‘a moderate amount’, while only one had ‘a lot’ and three ‘a great deal’. There was also no significant difference between the seven groups in terms of their age (p = .751), education (p = .916) and experience with VR (p = .426). No participants withdrew from the study before its end.

The sample size was calculated with an a priori power analysis for a multiple linear regression analysis, fixed model, by using G*Power 3.1 [94]. For the estimation we used a Cohen’s f² of 0.15 (for a medium effect size), a level of power of 0.80, nine predictors (induced emotion, type of priming, the three measured emotions and five personality traits), and an α-level of 0.05. This analysis returned a total necessary sample size of 114, i.e., 16 participants per group. The study was approved by the Psychology Research Ethics Committee (PREC) of the University of Bath (ethics code: PREC 18–301). All participants were adults and gave written informed consent prior to participating and were debriefed accordingly at the end of the study. Participants were also informed they were free to withdraw from the study at any point, but none chose to do so.

6.1 Spatial presence

For spatial presence, as expected, no significant regression was found (F(3, 111) = 1.153, p = .321). Only agreeability acted as a significant predictor. This yielded a significant regression equation (F(1, 124) = 7.044, p = .009), with an R² of.054. Spatial presence was equal to 5.328 +.051 (p = .009) (computed fear score). This suggests that computed emotion scores cannot predict spatial presence; cf. social and self presence in H1. Agreeability predicting spatial presence partially supports H4.

6.2 Social presence

In the case of social presence, a significant regression equation was found (F(14, 111) = 1.80, p = .046), with an R²of.185. Two significantly contributing factors were found, namely the computed fear score (p = .026) and agreeableness score (p = .034). Next, we examined whether there was any interaction effect between fear and agreeableness scores. To correct for collinearity between the interaction scores (calculated as the product of computed fear scores and agreeableness scores) and the original scores for fear and agreeableness, we centered the data for all these predictors and checked for correlations. No correlation was found among the centered predictors (p > = .338), thus showing lack of multicollinearity. A regression with agreeableness, computed fear score and the interaction variable was run. This yielded a significant regression equation (F(3, 122) = 5.778, p = .001), with an R² of.124. Participants’ predicted social presence was equal to 4.732 +.169 (p = .006) (computed fear score) +.073 (p = .002) (agreeability score). No interaction effects were found between agreeableness and fear scores (p = .894). Computed fear scores predicting social presence partially supports H1 and H2. The lack of an effect of perceptual or conceptual priming means that H3 is rejected. H4 is partially supported as only agreeableness predicted social presence, but not openness and introversion.

6.3 Self presence

For self presence, a non-significant regression equation was found (F(14, 111) = 1.127, p = .343), with an R²of.124. However, two significantly contributing factors were found, namely computed happiness scores (p = .036) and computed fear scores (p = .036). This prompted a regression analysis with just these two predictors, as for social presence. No interaction variable was calculated between happiness and fear computed scores as we did not hypothesize any interaction between them. The analysis showed a significant regression equation (F(2, 123) = 3.213, p =.044), with an R² of.050. Participants’ predicted self presence was equal to 4.909 +.177 (p =.041) (computed happiness score) +.136 (p =.023) (computed fear score). These results suggest that self presence is predicted by both computed fear and computed happiness scores, unlike spatial and social presence. No personality traits predicted self presence. The computed happiness and fear scores predicting self presence supports H1. The fact that computed happiness and computed fear, but not computed relaxation, predicted self presence supports H2. The lack of an effect of perceptual or conceptual priming means that H3 is rejected. H4 is not supported as no personality traits predicted self presence.

6.4 Overall presence

For presence as measured by Q2 in Egan et al.’s questionnaire [74], we found a significant regression equation (F(14, 111) = 2.768, p <.001), with an R² of.259. Two significant contributors were found for the model, namely computed happiness score (p =.046) and openness (p =.033). Next, we examined whether there was any interaction effect between happiness and openness scores. To correct for collinearity between the interaction scores (calculated as the product of happiness scores and openness scores) and the original scores for happiness and openness, we centered the data for all these predictors and checked for correlations. No correlation was found among the centered predictors (p > =.161), indicating a lack of multicollinearity. A significant regression equation was found (F(3, 122) = 9.252, p <.001), with an R² of.259. Participants’ presence was equal to 6.428 +.028 (p = 0.49) (openness score) +.166 (p <.001) (computed happiness score)—.018 (p =.020) (openness and happiness score interaction variable). An interaction was found between computed openness and happiness scores (see Fig 4). The regression lines represent the data for low, medium and high openness. The higher R² value obtained for people scoring low on openness indicates that this group was more susceptible to differences between predicted and felt emotions. For those scoring medium or high openness this effect was not so pronounced, suggesting more sensitivity to large discrepancies. Results for Q2 suggest that happiness is a strong predictor of presence in general. Furthermore, the interaction with openness shows that personality traits can moderate not only presence, but the relationship between presence and emotion. The fact that computed happiness, but not computed relaxation, predicted Overall presence partially supports H2. H4 is also partially supported because openness predicted Overall presence.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Interaction between openness and happiness scores when predicting presence.

The blue line represents the best linear fit to presence scores for the low openness group. The blue circles represent that data points for the low openness group. The red line represents the best linear fit to presence scores for the medium openness group. The red squares represent the data points for the medium openness group. The green line represents the best linear fit to presence scores for the high openness group. The green triangles represent the data points for the high openness group. R2 values for the different levels of openness can be found in the legend.

https://doi.org/10.1371/journal.pone.0280390.g004

A non-significant regression equation was found for the UCL presence questionnaire measure (F(14, 111) = 1.274, p =.235), with an R² of.138. Only the computed fear scores acted as a significant predictor, which partially supports H2. This yielded a significant regression equation (F(1, 124) = 5.228, p =.024), with an R² of.040. Presence as measured by the UCL questionnaire was equal to 4.887 +.088 (p =.024) (computed fear score), suggesting that computed fear scores predicted presence. The UCL presence questionnaire thus showed different results to presence as assessed by Egan et al.’s Q2 (see Table 2). More precisely, computed fear scores seem to be a good predictor of presence as measured by the UCL questionnaire, and not computed happiness, as was the case with presence measured via Q2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Table showing the regression results for the presence measures.

https://doi.org/10.1371/journal.pone.0280390.t002

6.5 Testing IPCM predictions

We investigated whether overestimating and underestimating emotion equally lead to a decrease in presence, as suggested by Seth et al. in their model [3]. For this analysis we chose Egan et al.’s Q2 since it represented the most general measure of presence [74]. We ran Bonferroni corrected independent samples t-tests between those who underpredicted, predicted correctly, and overpredicted emotion intensity. For Q2, we found that the underpredicting group (Mean = 5.951, SD = 1.139) achieved significantly less presence than those who overpredicted emotion (Mean = 6.708, SD =.581); t(87) = −4.032, p <.001, d = −.319. There was no significant difference between the underpredicting group and the group who predicted correctly (Mean = 6.513, SD =.960), or between the group who predicted correctly and the overpredicting group (p > =.066). Results were similar for fear, in that the only significant difference found was between the underpredicting group (Mean = 6.7419, SD =.575) and the overpredicting group (Mean = 6.216, SD = 1.075); t(89) = 2.535, p =.039, d =.011. For relaxation, no differences were found between any of the groups (p > =.051) (see Fig 5). These results further support the IPCM and H1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Graphs at the top show mean scores of presence as measured by Q2 for participants who overestimated, estimated correctly and underestimated levels of relaxation, happiness and fear.

Graphs at the bottom represent mean distributions of the gaussian psychometric functions for presence, as measured by Q2 for relaxation, happiness and fear. The solid blue line represents the line of best fit for these data points. It represents the normal distribution fitted to the data as predicted by the IPCM. The dashed blue line represents the actual peak in presence and the black dashed line represents the peak in presence as hypothesized by the IPCM.

https://doi.org/10.1371/journal.pone.0280390.g005

6.6 Emotion intensity

Paired-samples t-tests were conducted to compare the felt emotion scores (third measurement) for relaxation, happiness and fear for the entire sample size. There was a significant difference between the scores for relaxation (Mean = 5.73, SD = 2.37) and happiness (Mean = 7.23, SD = 1.99;t(125) = −7.387, p <.001, d = −.658, between relaxation and fear (Mean = 2.68, SD = 2.47); t(125) = 7.923, p <.001, d =.706, and between happiness and fear; t(125) = 13.631, p <.001, d = 1.214. Thus, happiness scores were overall highest, followed by relaxation and lastly by fear.

6.7 Discomfort and emotion/presence reports

Because post-experience emotion scores are emotional appraisals of the VR app, we took into account that negative experience due to cybersickness, which can cause sensations of dizziness or nausea, could impact these scores. Discomfort could be due to motion sickness, or general dissatisfaction with the system hardware [50, 74]. To exclude this possibility we ran a correlation analysis between presence scores and general enjoyment as measured by Q9 in Egan’s [74] questionnaire which referred to general discomfort while using the system (“I did not feel any discomfort while using the application”). Results showed no significant correlation between discomfort and reported levels of emotion during the VR experience for relaxation, happiness or fear (p >.051). No significant correlation was found between general comfort and presence measured by any of our measures (p >.382). Hence, physical discomfort or lack of satisfaction with the VR hardware did not affect our results.

6.8 Summary of hypotheses validation

Our results showed that, as hypothesized in H1, the IPCM predicted the formation of social and self presence but not spatial presence. Moreover, H2 was supported because the IPCM predicted presence based on computed happiness and computed fear but not computed relaxation. The lack of an effect of perceptual or conceptual priming on any of the emotions or components of presence means that H3 is rejected. Lastly, results showed that some personality traits, namely openness and agreeableness moderate the relationship between emotion and presence, and so we partially accept H4. A schematic presenting the significant interactions found can be seen in Fig 6.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Graphical representation of the revised IPCM model, as informed by the regression results.

The left row presents the three emotions tested. The middle row presents the types of presence that were verified and interactions between factors impacting overall presence. The right column presents the BFI [72] personality traits tested. Arrows signify significant interactions and items that are in grey did not show any significant impact on presence measures.

https://doi.org/10.1371/journal.pone.0280390.g006

7.1 Limitations

Although not affecting our results for H1, H2 and H4, priming participants with conceptual or perceptual stimuli did not show an effect, thus not supporting H3. One possible explanation for this is our use of audio-only stimuli for the priming. VR is a rich, multisensory medium and thus the priming we attempted via a unisensory stimulus may not have had a large effect. Future research could attempt priming using multisensory stimuli, preferably presented in VR. This could clarify whether the use of unisensory priming was indeed the reason why priming did not succeed. Still, it is worth noting that because the computed emotion scores were calculated using the scores collected after the priming phase and those collected after the VR experience, the lack of a priming effect has no impact on our other findings.

An important limitation related to social presence in this study is the lack of user interaction with the virtual avatars present in the experience. This is because social presence is not only determined by the mere presence of other sentient beings, but also interactions with them which would lead to more complex inferences about their thoughts or emotions [44]. This means that in the present study, social presence may not have been felt the same way as it would be in VR applications which do facilitate interaction with other social beings within the VE [101]. This lack of interaction was intentional, so as to not introduce potential confounding variables to the design due to how users chose to interact or not with the VE. Still, this does not mean that the present results on social presence are not valid, but that they strictly apply to VEs where there exists no interaction with other social agents. Further research should focus on social presence and employ more sophisticated VEs where such interactions would be possible.

Using self-report measures for both emotion and presence could also be a limitation as judging the intensity of felt emotions could be considered when reporting presence and vice-versa. As observed by IJsselsteijn [102] an emotional response towards the VE (e.g. a fear or happy response) can be interpreted by the participant as a sign of a higher level of presence, possibly due to the existence of specific internal feedback mechanisms.

Measuring either emotions or presence via an objective measure could be a solution and seems promising, but has so far been unsuccessful in the context of VR presence [74, 79]. Additionally, while such measures may indeed measure arousal, it is impossible to estimate emotional valence (e.g. whether the user is happy or fearful) from these measures. Nonetheless, self report measures themselves pose limitations, such as being prone to demand characteristics; which do not affect physiological measures to the same extent [103]. Demand characteristics could be especially relevant if users are primed with conceptual material, like we did in the present study. This means that perhaps physiological measures could be used in conjunction with self report ones to validate obtained emotion scores.

Apart from the UCL Presence Questionnaire [73], the present study used custom, individual items for the three types of presence examined and a single item from Egan’s [74] study to measure overall presence. Future studies should use more recent and well-validated measures which were not available at the time of our study, such as the Multimodal Presence Scale (MPS) [104], which has subscales for spatial, social and self presence.