Was it Real or Virtual? Confirming the Occurrence and Explaining Causes of Memory Source Confusion between Reality and Virtual Reality

Source confusion occurs when individuals attribute a memory to the wrong source (e.g., confusing a picture with an experienced event). Virtual Reality (VR) represents a new source of memories particularly prone to being confused with reality. While previous research identified causes of source confusion between reality and other sources (e.g., imagination, pictures), there is currently no understanding of what characteristics specific to VR (e.g., immersion, presence) could influence source confusion. Through a laboratory study (n=29), we 1) confirm the existence of VR source confusion with current technology, and 2) present a quantitative and qualitative exploration of factors influencing VR source confusion. Building on the Source Monitoring Framework, we identify VR characteristics and assumptions about VR capabilities (e.g., poor rendering) that are used to distinguish virtual from real memories. From these insights, we reflect on how the increasing realism of VR could leave users vulnerable to memory errors and perceptual manipulations.


INTRODUCTION
Memory is a fundamental cognitive function.Long-term memory, in particular, has a signifcant impact on our lives.It shapes our personality, our social interactions, and our future choices and behaviors [7].Remembering an event can be seen as a process in three steps: frst, we acquire the information (referred to as encoding); then, we retain it over time (storage); and fnally, we bring it back when needed (retrieval) [4,47,67].
However, this process is not infallible.Our memory is prone to several types of limitations: forgetting (when we cannot remember), distorting (when we remember inaccurately), and persistence (when we cannot forget) [62,63].One type of memory distortion that can be at the origin of false memories (i.e., when we strongly believe in a memory of an event that never occurred) is source confusion [41,42,62,64].Also known as misattribution, source confusion occurs when we attribute a memory to the wrong source [33,62].For instance, we might see something in a picture and later think we witnessed this event.This phenomenon can result in minor inconveniences, like being uncertain about whether we closed the door or just imagined doing so, but it can also lead to more signifcant negative consequences [62].For instance, confusing something heard in the street with something heard on TV can lead to the spread of misinformation [41], or even to false eyewitness testimonies [84].There are numerous examples of source confusion in daily life, and it concerns every type of memory source.
Virtual Reality (VR), which fnds applications in everyday life and diverse domains (e.g., entertainment, training, marketing, or therapy), represents a new source of memories which can be concerned by source confusion.Someone might experience something in VR and later think it happened in reality.Source confusion between reality and other sources like imagination or pictures has long been studied to understand how, when, and why it occurs.In the Source Monitoring Framework (SMF), Johnson et al. identify the type of cues that are used to identify a memory's source [33,34].The SMF is a general framework that could theoretically be applied to every type of source.However, only a few studies focused on source confusion between VR and reality (that we refer to as VR source confusion) [28,29,60].VR difers from other sources in several aspects that might infuence source confusion.Contrary to other media like pictures, VR ofers control over what users perceive, providing more immersive experiences with more sensory and perceptual information, and a greater level of media richness [66].A unique characteristic of VR is its capacity to elicit a sense of presence, defned as the subjective feeling of being in a place ("being there") while the physical body is situated in another place [19,70,83].There is currently no understanding of how these specifc aspects of VR might infuence source confusion.
In this paper, we present a mixed-method study, based on a study initially conducted by Hofman et al. in 2001 [29] that we updated using current technology (Oculus Quest 2), and modifed to incorporate questionnaires to explore which aspects of VR experiences could infuence VR source confusion.During the study, participants (n=29) were asked to observe 10 real and 10 virtual objects.Later, they took a source identifcation test to identify, for each object, if they saw it in reality, in VR, or if did not see it during the study.
The results of a quantitative analysis confrm that VR source confusion occurs and suggest that the technological progress made over the last 20 years might increase the frequency of VR source confusion (from 10% of virtual objects confused for real ones in Hofman et al. 's study [29] to 20% in our study).The results do not show a correlation between the level of presence (i.e., SUS Mean [73]) and the occurrence of VR source confusion.In line with the SMF, the results of a qualitative analysis show that identifying the source of a memory is a decision process based on a set of cues that are then compared with prior knowledge and assumptions.We identify these cues for the specifc case of VR vs reality source identifcation.We found that VR source confusion occurs: 1) when cues are incorrect, 2) when cues are missing, 3) when cues are not specifc to one source, and 4) when an individual holds incorrect assumptions about reality or VR technology capabilities.
From these insights, we join the discussion started by Slater et al. about the ethics of realism [71].Technology is still far from the point where users would not be able to distinguish if they are in VR or reality.However, source confusion between VR and reality can already occur today, and the impacts of memory errors should not be underestimated [41,84].Therefore, we should not wait to reach the Ultimate Display [77] to start addressing the ethical, psychological, and societal implications of blending between virtual and real experiences, particularly regarding the risks for memory manipulations [9].We encourage further research to develop protection mechanisms to limit VR source confusion.
This work has two main contributions: 1) a laboratory study confrming the existence of source confusion between VR and reality and demonstrating its potential increase, and 2) a quantitative and qualitative exploration of the causes of VR source confusion, building upon the Source Monitoring Framework, identifying causes of VR source confusion and characteristics of VR that could amplify the phenomenon.Readers can access all supplementary material, including fgures, raw data, and coding details, on the Open Science Framework (OSF) platform 1 .

RELATED WORK
In this section, we present studies that contribute to understanding how VR is remembered and impacts memories, as well as why source confusion occurs.

Impact of Immersion and Presence on Memory
One aspect that diferentiates VR from other media is the level of immersion it ofers, and the sense of presence it creates.In their review published in 2019, Smith et al. present studies that seek to understand the impact of immersion on episodic memory performances [74].Several studies show that more visual details (e.g., colors, textures, larger feld of view) lead to better memory performances [54,74,82].However, other studies show that the impact of visual fdelity depends on the type of visual details that are added [44,45,74], and the efect on memory could be subtle or dependent on the context [50,74].Multimodal sensory information, such as audio, tactile, or olfactory information, might support the encoding process and improve recall performances [2,16,74].While some studies suggest that active interaction and navigation in the VR environment support memory performances [26,30,31,74], these fndings have not always been observed [24,61], showing that it might depend on the type of task and context.In 2021, Kisker et al. showed that memories from VR experiences are more vivid than memories originating from screen experiences, and that retrieval of these memories requires less efort [35].Schöne et al. also observed that memories of immersive 360° videos are retrieved more efciently and more successfully than 2D videos [65].Overall, the diferent aspects of immersion could have a positive impact on memory performances [37,74] but it is not fully clear.
While immersion corresponds to objective characteristics, presence is a subjective feeling, that is infuenced by immersion and other human factors [69].The sense of presence elicited by VR can lead to realistic behavior [53,68], physiological and psychological responses [12,36,46], which could potentially lead to the creation of memories similar to those formed by real-life experiences.According to Witmer and Singer, the level of presence partly depends on how much attention is given to the virtual world compared to reality [83].In their survey, Smith et al. argue that knowing that attention leads to better episodic memory performances, a high level of presence in VR could therefore also have a positive impact on memory performances [74].However, they also argue that if presence is related to the amount of "task-irrelevant information" in the environment (e.g.ambient sounds), it could be distracting and lead to poor memory performances.The results of two studies did not show a correlation between the level of presence and memory recall performances [10,76].In 2017, Makowski et al. found that presence was related to better memory performance.However, this study tested how many details participants could remember from a movie, which did not involve VR [43].While these studies are focused on the impact of immersion and presence on recall performances, in this paper, we focus on memory source identifcation, a task that involves diferent cognitive processes.

Technology and Memory Distortions
The potential negative impact of technologies on memories, particularly regarding memory distortions, is a growing concern [8,9,14,18,56].If a person is exposed to a suggestion, for example through doctored pictures [13,23,27,52,56,81], confusing the source of this suggestion with what was originally experienced can lead to the creation of a false memory [42,[62][63][64]66].In 2002, Wade et al. exposed participants to manipulated pictures of them as a child in a hot air balloon, which led to the induction of false memories for half of the participants [81].
Segovia et al. argue that VR has greater media richness than narratives and pictures, potentially making it a more powerful tool for memory suggestions.In a study, they showed that exposing preschool children to a virtual experience (swimming with whales in VR) can lead to the creation of a false memory of this event [66].Although this work is focused on the efect of VR on young children, who are more vulnerable to suggestions, it is a hint toward thinking that VR could be used to induce false memories to adults too, for more subtle events than swimming with whales.Other works showed that revisiting altered reconstruction of past events with VR can have an impact on the perception of the past, and on users' feelings [9,15,22].

Source Confusion
Since source confusion can be responsible for memory distortions, cognitive science researchers tried to understand why, when, and how this phenomenon occurs.

Reality
Monitoring.Research about source confusion has long been focused on confusion between internal sources (e.g., thoughts, imagined events), and external sources (i.e., what has been perceived from reality) [1,21,25,34,38].Johnson et al. explain that memories are not always directly labeled with their source [33].They call "Reality Monitoring" the decision process of identifying if a memory comes from an internal or external source [34], and explain that memories that have a lot of sensory characteristics (e.g., auditory information), semantic content (i.e., meaningful information or details) and contextual information (i.e., spatial and temporal details), have more chances to be identifed as real than imagined, potentially leading to source confusion.
They note that this decision process takes into account other factors, such as prior knowledge and assumptions [34].An event that is judged as highly improbable, such as the memory of a fying whale, is more likely to be identifed as imagined, regardless of the amount of sensory and semantic details of this memory.They also point out that a memory is never completely dissociated from other memories like prior events and context, which can help to identify the source of a memory.

Source
Monitoring.Source confusion can occur with all kinds of sources, such as narration [23,39], images [25,27,40,81], or even videos [20,51,52].In the Source Monitoring Framework (SMF), Johnson et al. categorize cues that are used to identify the source of a memory: sensory and perceptual information, contextual information, semantic details, afective information, and cognitive operations.The outcome of the source monitoring decision depends on the type and amount of cues and how unique they are [33].While the SMF is a general framework that can theoretically be applied to all kinds of sources, there is currently a lack of understanding about how this framework applies to the specifc case of VR vs Reality source identifcation.

Source confusion between VR and reality.
Few researchers started to study source confusion between VR and reality.In 1995, Hofman et al. did a frst study to assess if participants could distinguish between VR, real and imagined events, and to understand the diferent characteristics of memories from these sources [28].They suggest that if there is a correlation between the level of presence during a VR experience and source confusion, then the Memory Characteristics Questionnaire [32] could be used as a way to measure presence, or at least the quality of the VR experience [28].They found that memories from real events present more clarity, better spatial location, and certainty in the identifcation of the source than memories from VR and imagined events.Memories from VR events felt more intense and had more colored and vivid details than memories from imagined events.The results of this study do not show a correlation between source confusion and the level of presence.
In 2001, Hofman et al. conducted a new study, showing that VR source confusion occurs [29].By asking participants to rate the quality of their memories of real and virtual objects on a set of diferent aspects, they found that memories of real objects are rated higher than memories of virtual ones on the following aspects: visual detail, touching, size, physical texture, and light refectivity.However, the goal of this study was not focused on investigating the causes of VR source confusion.Also, these two studies were conducted using deprecated technology (Virtual Research VR8 HMD: feld of view 60° vertical, 60° horizontal, resolution 640 X 480 per-eye, refresh rate 60 Hz).Technological progress in terms of resolution, feld of view, and refresh rate could have an impact on source confusion (e.g., Oculus Quest 2: feld of view 93° vertical, 97° horizontal, resolution 1832x1920 per-eye, refresh rate 120Hz).
More recently, Rubo et al. conducted a study to compare source confusion between real, VR, and desktop screen events, and to evaluate the efect of manipulation on source confusion [60].No evidence of the impact of manipulation on source confusion was found.The results indicate that source confusion is more frequent between VR and reality than between desktop screens and reality, meaning that VR exposes users to more risks of source confusion than previous media.However, which specifc aspects of VR experiences enhance source confusion is still unclear.In 2023, Mizuho et al. presented words to participants in diferent environments (reality, VR -high visual fdelity, VR -low visual fdelity) [49].Their results do not show an impact of visual fdelity on errors when identifying the context in which the words were shown.However, in this study, alterations in visual fdelity were limited to the environment (i.e., the room) and did not extend to the content to be memorized (i.e., words displayed on a desktop screen).
In this paper, we build on Hofman et al. 's questions to investigate the impact of presence on source confusion [28] and explore the causes of VR source confusion, using recent technology (i.e., Oculus Quest 2).

METHOD
We conducted a laboratory study, using a procedure inspired by Hofman et al.'s 2001 study [29], which we refer to as Study H. Similar to Study H, participants observed 10 objects in reality and 10 in VR.One and two weeks later, they took a source identifcation test.We adapted Study H by using recent technology (i.e., Oculus Quest 2), incorporating measures of presence, and adding an open questionnaire to investigate the reasoning behind source identifcation.

Participants
We recruited 32 participants from the local mailing list.One could not go through the full study because of a technical issue, and one did not complete any of the following questionnaires, ending up with n=30 (13 identifed as female, 17 as male, age: M= 28.2, SD = 7 yrs).They provided informed consent, and their data was anonymized.Participants were not informed that the study was about memory.They received a 15 euros compensation.4 participants had never tried VR before, 24 only a few times, and 2 often use VR.

Environment
The real environment was a room (Figure 1) containing a table (A), a chair, a box (B), and panels covering the windows (C).The table was used to put the tablet for questionnaires, the tennis ball and racket, and the VR HMD.The objects participants had to observe were placed on a box (B).Participants had enough space to move freely (D).
A standalone app was developed using Unity (2021.3.25f1),deployed on an Oculus Quest 2. Participants could interact in the VR environment (i.e., grab the ball and racket) using controllers.The VR environment (Figure 1) was modeled after the real room, with identical dimensions and orientation.Similar to Study H, we used 30 objects (Appendix A, Figure 3), each available both in reality and in VR.We randomly divided these objects into three sets, each containing 10 objects.Each set was assigned to a specifc source for each participant (i.e., real, virtual, or new -not presented).The source of each set was rotated across participants to ensure that every object was shown an equal number of times in each source across the entire study.Each participant observed a total of 20 objects: 10 in VR and 10 in reality.

Objects
Similar to Study H, we used everyday life objects (Appendix A, Figure 3).We selected these objects from the standardized and validated database OpenVirtualObjects [78], which is tailored for VR-based research.We selected objects with high recognizability for which we could easily fnd a real equivalent.The objects and the environment were rendered using basic rendering proposed by Unity (baked light, no shadows) to evaluate VR source confusion with basic settings of current technology.

Study Procedure
The study was 90 minutes long, including a 30-minute training session.During training, the experimenter provided detailed instructions and safety guidelines, and participants practiced the tasks in both conditions (VR and reality).Participants were told that the study aimed to investigate perception of both real and virtual environments, without mentioning the following memory questionnaire.Once the study started, the experimenter remained hidden behind the panels, and task instructions were conveyed automatically through a speaker.
Participants completed the following task (engagement + observation + questionnaire) 10 times in reality and 10 times in VR (Figure 2).As done in Study H, the order of the objects, and therefore the order of the condition, was random to avoid patterns that would infuence memorization.
3.4.1 Engagement.Participants were instructed to put the VR HMD if needed (depending on the current condition).In Study H, participants were exposed to VR only for 7 seconds per object.To increase immersion and potentially induce more variation of the sense of presence, participants engaged in a game (i.e., bouncing a ball on a tennis racket) for 30 seconds before seeing each object, in both conditions.This duration was chosen to be long enough to create engagement while keeping the study shorter than 90 minutes.Then, participants were instructed to place the racket and the ball back on the table and to keep facing the table for 8 seconds.During this time, the experimenter placed the next object on the box (or made it appear on the virtual box), without being seen.Unlike Study H, participants were asked to not touch the objects.It is worth noting that investigating the efect of touch on VR source confusion is an important area for future study, though it was not our primary focus.Participants were then asked to verbally name the object out loud (in English or French) to ensure accurate identifcation because the source identifcation test relies on the names of the objects.We extended the observation time to 15 seconds (instead of 7 seconds for Study H), since in our study, participants had to physically move towards the object and verbally name it.
3.4.3Qestionnaire.Then, participants were instructed to remove the VR HMD if needed and complete a 10-item questionnaire.Simultaneously, the experimenter removed the object from the box.This questionnaire was an additional component compared to Study H, which we included to investigate the impact of presence.
Due to time constraints during the study, we had to be selective with the number of questions (each questionnaire was flled 20 times per participant).We opted to focus on assessing the sense of presence since it is often seen as a distinguishing factor of VR from other forms of media, using the SUS questionnaire [73], chosen for its brevity (6 questions) and widespread use in assessing presence [19] (Appendix B).The SUS was completed after each VR exposure (10 times in total).An adapted version of the SUS was completed after each reality exposure (Appendix C).Although we did not use the data from this "reality SUS", we included it to maintain the balance between VR and reality experiences, to prevent potential bias in the results of the source identifcation test.To further gauge participants' level of enjoyment and potential boredom, we introduced four additional questions (Appendix B).

Follow-Up Questionnaires
One and two weeks after the study, participants received questionnaires to complete online.

Source identification tests.
The source identifcation test was the same as used in Study H (Appendix D).For each of the 30 objects, participants had to determine if the object was real, virtual, or if they did not see it at all during the study.They were also asked to indicate their level of confdence in their answers using a 6-point scale with 20% intervals from 0 % (completely guessing) to 100% (completely confdent).This test was done both at week 1 (W1) and week 2 (W2).
3.5.2Everyday Memory Qestionnaire (EMQ).We added the Everyday Memory Questionnaire [59] to the W2 questionnaire, which consists of 13 self-assessment questions regarding participants' memory abilities, specifcally focusing on how frequently memory lapses occur in their daily lives.We incorporated this questionnaire 1) to identify any outliers among the participants in terms of their memory abilities, and 2) to examine whether the outcomes of the self-assessment of everyday memory abilities are related to the results of the source identifcation test.

Source monitoring questionnaire.
In Study H, participants completed a VR Memory Characteristics Questionnaire (VRMCQ) to rate the qualities (e.g., visual details, shape, texture, thoughts) associated with their memories of each object (e.g.: "How much visual detail does your memory for the bowl have?")[29], adapted from the Memory Characteristics Questionnaire (MCQ) proposed by Johnson et al. [32].To understand the reasoning behind participants' source identifcation process more openly, we opted for a diferent approach, using open-ended questions instead of a quantitative questionnaire.This allowed us to capture participants' reasoning without constraining it.The questionnaire (referred to as source monitoring questionnaire, done at W2) was tailored to each participant based on their responses at W1. Participants were given the name of objects that they incorrectly or correctly identifed as real or virtual the week before, for which they were highly confdent (80% or above) or not confdent (40% or below) in their answers.They were asked to articulate the underlying reasons that infuenced their source identifcation decision (Appendix E).Additionally, a general question asked once per participant allowed them to give any additional insights or feedback they deemed relevant.

Analysis
3.6.1 Qantitive analysis.We use an exploratory approach.We run correlation and multiple linear regression on the quantitative data to explore which factors could infuence the results of the source identifcation test.Further, we conducted repeated measures ANOVA and t-tests to compare the numbers of errors across error types and time.We used Greenhouse-Geisser correction to account for the violation in sphericity and Bonferroni-corrected post-hoc tests to compare individual diferences.We conducted the analyses in JASP 0.17.2.1.
3.6.2Qalitative analysis.The answers collected from the source monitoring questionnaire were frst classifed within the categories of the SMF by the frst author.The answers were color-coded to keep track of the associated questions (real or virtual object, correct or incorrect identifcation, high or low confdence).The answers were then coded and grouped by themes within the SMF.Two other authors then checked the codes asynchronously.Then, all three authors discussed the codes during two synchronous collaborative sessions (two hours each), until reaching an agreement.

QUANTITATIVE RESULTS 4.1 Data
There are three types of errors possible for the source identifcation test: forgetting (identifying a real or virtual object as new), false recognition (identifying a new object as real or virtual), and source confusion (identifying a real object as virtual or vice versa).To keep the paper concise, we focus our analysis on the occurrence of source confusion.
The scores collected for each participant are: • The SUS Mean: average of the answers to the 10 VR SUS questionnaires (Appendix B). • The SUS Count: determined by counting how many of the six SUS questions received responses equal to or greater than 6 [72,80].Then, the count values from the 10 questionnaires are averaged to get one participant's score (scale ranging from 0 to 6).• SUS Q1 to Q6: average of answers to questions 1 to 6 on the 10 questionnaires after VR exposure (Appendix B). • Fun Score: calculated from Q7 and Q9.Bored score: calculated from Q8 and Q10 (Appendix B). • Overall Errors: total number of incorrect answers for each participant during the source identifcation test.Source Confusion, Forgetting and False Recognition are the number of each respective type of error.• Overall confdence: average of reported confdence on the 30 items (scale from 0 to 100%).• Confdence in source confusion: average of reported confdence on the items of the Source Identifcation test where the participant made a source confusion error (scale from 0 to 100%).• Everyday Memory Questionnaire: average of the 13 items of the EMQ [59], assessing their everyday memory abilities on a scale ranging from 1 (rare memory failures in everyday life) to 5 (frequent memory failures).
4.1.1Oulier and sample size.The boxplot of the number of overall errors shows one data point as being an outlier (18 errors out of 30) (Figure 4).We decided to exclude this data point for the quantitative analysis.Additionally, one participant did not complete the W2 source identifcation test and source monitoring questionnaire, resulting in a sample size of n=29 for W1 data and n=28 participants for W2 data.We mainly focus on results from W1, since the answers from W2 might have been biased by the frst questionnaire.Unless specifed diferently, we are referring to the results of W1.

Occurence of Source Confusion
In this section, we provide an overview of the source identifcation results and draw comparisons with the fndings from Hofman et al. 's 2001 study (Study H).
The results of our study are aligned with the results found in Study H, providing further evidence that participants sometimes  On average, one week after exposure, participants confused the source of 23% of the 20 objects that were presented (Figure 5).Participants were more prone to source confusion (59% of the overall errors at W1) than forgetting (20%) and false recognition (21%) (Figure 6).Contrary to Study H, source confusion was the main type of error (Figure 6).The post-hoc test (n=28) for the efect of time showed that errors increased from W1 to W2 ( = −4.268,< .001,= −0.357).On the other hand, the post-hoc tests for types of errors showed that source confusion was more prevalent than forgetting ( = −7.087,< .001,= −1.809)and false recognition ( = −6.450,< .001,= −1.646)while the diference was not  Additionally, it is interesting to note that source confusion was more frequent for the objects presented in reality than in VR (20% of the virtual objects were confused for real ones at W1, while 27% of the real objects were confused for virtual ones (Figure 7)).However, a 2x2 repeated measures ANOVA (n=28) with the factors time (W1, W2) and source of the objects (real, virtual) did not show a signifcant efect of the source of the objects on the number of source confusion errors (F(1, 27) = 3.249, p = 0.083, 2 = 0.107).

Exploration of Factors Infuencing VR Source Confusion
In this section, we present a quantitative exploration of the potential factors that could infuence VR source confusion.Given the exploratory nature of our investigation and since we conducted several tests, we do not place emphasis on the signifcance of pvalues and do not report them.Our primary aim is to provide initial insights and assess whether the diferent collected scores (feeling of presence, including which aspects thereof, enjoyment, boredom, and general memory abilities (subsection 4.1)) could potentially be factors infuencing source confusion.Pearson's correlation shows a correlation between the EMQ score and source confusion (r = +0.445), the EMQ score accounting for 19.8% of the variance in source confusion error (moderate to large efect size) (Figure 8).This result makes sense since the EMQ is a general memory ability questionnaire that is likely to correlate with many memory tasks.It suggests that individuals who perceive their memory as frequently failing are more likely to experience source confusion between VR and reality, and highlights the importance of considering memory ability when examining VR source confusion.
Pearson's correlation showed there was a weak correlation between the SUS Mean and source confusion (r = +0.301), the SUS Mean accounting for 9% of the variance in source confusion (moderate efect size).The SUS Count does not show a strong correlation with source confusion (r = +0.279,small efect size) (Figure 8).
The SUS questionnaire is commonly used to assess the sense of presence.However, it is not entirely clear how it is constructed, and the specifc roles and underlying constructs of each question remain unclear.Notably, there has been no factor analysis conducted to date [70].The various questions within the SUS may measure diferent aspects, which could potentially have varying efects on source confusion.For these reasons, we are interested in examining the correlations between source confusion and each individual question of the SUS (Figure 8).
There was a correlation between the answers to item 5 of the SUS (SUS Q5) and source confusion (r = +0.434), the SUS Q5 accounting for 18.8% of the variance in source confusion (moderate to large efect size) (Figure 8).This suggests that participants who rated Q5 higher tend to make more source confusion errors.The correlations between source confusion and SUS Q1, SUS Q2, SUS Q4 are SUS Q6 are negligible (Figure 8).Pearson's correlation also shows Q3 as accounting for 13.1% of the variance in source confusion (r = +0.363,moderate efect size) (Figure 8).The moderate efect size suggests that the aspect measured by Q3 could also partly explain source confusion.

Multiple linear regression.
We further investigated predictors of source confusion, by conducting a multiple linear regression with the EMQ score and each item of the SUS as predictors, using backward entry.
Multiple linear regression using backward data entry shows that the model using SUS Q5 and EMQ scores as predictors can predict source confusion (F(2,25) = 5.64, R2 = 0.311).31.1% of the variance in source confusion is explained by the model using the predictors Q5 and the EMQ scores.Q5 was positively associated with source confusion ( = 0.347) and EMQ was also positively associated with source confusion ( = 0.360).
The regression analysis employed here used a stepwise approach, which, while useful for identifying potential predictors, may introduce some bias in efect estimation.It is important to note that the emphasis here is on exploratory analysis rather than hypothesis testing and that the efect size should be interpreted with caution.

QUALITATIVE RESULTS
In this section, we investigate the reasoning behind virtual vs real source identifcation, from the answers to the source monitoring questionnaire.We investigate how the Source Monitoring Framework (SMF, see subsubsection 2.3.2) [33] applies to virtual vs real source identifcation, giving further understanding of what could cause VR source confusion.

Source Identifcation Process
Consistently with the SMF, participants overall reported a decision process and strategies to retrieve the source of the objects, based on diferent cues.When observing the object, participants collected several types of information that got encoded in their memory: the external information captured by the senses (e.g., what they saw or heard), and the internal information they generated while seeing the object (e.g., thoughts, emotions) (Figure 9).Later, during the source identifcation test, some of the previously stored information was retrieved.Participants recalled visual memories of seeing the object (e.g., "I try to remember visually the things I saw and try to see if it seems real or virtual"), as well as other memories of the cognitive operations that occurred when seeing the object, such as the thought they had when seeing the object, or the feeling it gave them (e.g., "I remember specifc thoughts as I discovered these objects [...].These thoughts came back as I was answering and helped me being confdent about the object being real", "I remember my feeling when seeing the objects").One type of information might be more prominent than another (e.g., "I remember this thought better than the actual pear"), but both internal and external information result in cues, that are used to determine the source of the object (Figure 9).These cues are then compared to a preconception of what could be real and what could be rendered in VR (Figure 9). the cues are aligned with assumptions about reality, the source is identifed as real (e.g., "I remember it to be non-centered on the storage unit, which rings the 'real' bell"), and same for VR (e.g., "I remember a banana with a very smooth skin, what I identify as a VR image").The assumptions that are presented in the SMF are mainly about the real world (what could possibly be real) [33].In this study, we found that for the specifc case of VR vs real source identifcation, the assumptions are not only about the real world but also about what VR technologies are capable of (what could possibly be rendered or modeled).

Cues Used to Distinguish VR from Reality
From the answers to the source monitoring questionnaire, we identify the cues that are associated with VR and reality, within the SMF categories (Figure 9).

5.2.1
Lighting.The memory of the refectivity of light on the objects was associated with reality (e.g., "[the object] refectivity made it particularly distinctive from the virtual objects").Objects remembered with poor refectivity were identifed as virtual (e.g., "I remember [the cutter] was virtual because it did not shine like a real cutter, it was not refective").The memory of an "artifcial light" or the "lack of shadows" was also associated with VR.Not remembering artifcial lighting efects resulted in confusing a virtual object for a real one (e.g., "If the light on an object looked artifcial, I chose virtual.For the USB key, I do not remember this, so I chose virtual").

General rendering quality.
Remembering the "blurry" and "pixelized" aspects of an object was associated with VR (e.g., "I had a vague memory of a blurry lemon.When the VR headset slid a little bit on my head, the outlines of the objects were a little blurry.This is why I assumed I saw the lemon in VR").On the other hand, noticing the good rendering quality of the virtual objects also gave cues associated with VR (e.g., "I remembered the USB key being pretty well rendered").

Sensory feedback.
The memory of additional sensory feedback (i.e., sound) was associated with reality (e.g., "The sound of the real objects being put on the storage unit diferentiated the real and virtual environment").The absence of additional sensory cues was mistakenly associated with VR (e.g., "[For the banana,] I don't remember any smell so I 100% assumed it was virtual").

Textures.
Objects remembered with a smooth and artifcial texture were associated with VR (e.g., "I remember a banana with a very smooth skin, what I identify as a VR image", "For the hat, I vaguely remember a somewhat artifcial texture").The lack of texture was also associated with VR (e.g., "the lemon had no texture").The memory of an artifcial texture also resulted in confusing a real object for a virtual one (e.g., "When trying to remember the object, I visualize it with a rendered texture, like in VR").

Level of details.
Memories of objects having a lot of details were associated with reality (e.g., "I also remembered the objects very precisely, with sharp details, unlike with the objects I remembered from the virtual room").A poor level of detail was associated with VR (e.g., "It was too simple compared to real grapes", "things in which I can't remember details, my brain assumed them to be virtual").

Shapes.
Remembering shapes that do not match the expected shape of an object was associated with VR (e.g., "The shape of the object seemed so odd in my mind that I was sure it was virtual").On the contrary, the memory of an object where nothing looked unusual was identifed as real (e.g., "I remembered they looked correct, not oddly shaped").A memory of a shape that does not match the expectation of reality resulted in confusing a real object for a virtual one (e.g., "That shape of lemon is not very common, and I usually only see that shape in movies or cartoons").

Colors.
Remembering colors that do not match the expectations of reality was associated with VR ("I remember it having an odd color").An incorrect expectation of real world colors resulted in confusing a real object for a virtual one (e.g., "I remember that they were green scissors, and usually the real scissors that I am familiar with are orange or red").

Functional properties.
Remembering that an object was missing specifc details linked the function of the object was associated with VR (e.g., "the stick was lacking the inside electronics, and could therefore not be real").Remembering that some of these specifc details were not correctly modeled, making the object unfunctional, was also associated with VR (e.g., "I use cutters frequently, so the angle of the tip being weird stuck in my memory").On the other hand, remembering the presence of functional properties was associated with reality (e.g., "I have a clear image of the electric wire hanging").The presence of branding was also associated with reality (e.g., "For the lighter, I had noticed the brand.The virtual objects were detailed but didn't have any branding.").

5.2.9
Recognizability.The memory of having trouble identifying the object when it was presented was associated with VR (e.g., "They looked strange and I had trouble identifying them").On the contrary, the memory of directly being able to identify the object was associated with reality (e.g., "It was a nice old black book.And I recognized it immediately as such.So I ticked it as real").This trouble in recognizing virtual objects could come from unrealistic models and poor rendering quality.

Emotional responses.
The memory of having an emotional response when seeing the object was associated with reality (e.g., "I remembered I thought they looked great and would have liked to taste them, which is something I would not have thought if they were in VR").On the contrary, the absence of memory for emotional response, such as fear or the drive to interact with the object was associated with VR (e.g., "[the cutter] is related to danger, but I don't remember having that feeling, so I was sure it wasn't real", "I never felt the drive to open a book during the test.Therefore either it was virtual, or it was not there at all").

Context.
The memory of an object cannot be completely isolated.Contextual cues were also used to identify the source.However, responses related to context are by nature very specifc to this study.Remembering the thoughts about the origin of the object, wondering how it got there and what the experimenter did, was associated with reality (e.g., "I mentally attached these objects to the thoughts that someone had to physically carry them to do the experiment").
The memory of the room around the object was also used to identify the source, both for reality (e.g., "I basically had the image in my head of the real room around the object"), and VR (e.g., "Sometimes I also remembered the virtual room around them").The memory of an object that was not perfectly centered on the box was associated with reality (e.g., "I remembered it to be non-centered or aligned on the storage unit, which rings the "real" bell", "All the VR objects were centered").
The context of the order of appearance of objects was cited, both for real objects (e.g., "I think it appeared shortly after the real apple") and virtual objects (e.g., "The lemon was the frst virtual fruit, the other ones were real before so it was remarkable").

Causes of VR Source Confusion
From the data and in conjunction with the SMF, we observe four causes of source confusion, listed below.It is worth noting that source confusion can persist, even when being told the correct source (e.g., "I ticked it as real.And I still see it as real by the way", "They looked real in my memory.They still do"), and participants were sometimes surprised by their mistakes ("I'm really surprised by this [error]").

Incorrect cue.
If a cue is incorrect, then the resulting source identifcation reasoning is likely to be incorrect.For instance, a participant had seen a real banana but remembered it as a "plastic banana", leading them to think it was virtual.The distortion of the cue might be caused by memories of diferent objects getting mixed up.If a virtual object is later seen in reality, the memories of both objects might be confused (e.g., "I have a tape dispenser on my desk.By association, I could have believed that this virtual tape dispenser was the real one").

Lack of cues.
Not remembering enough cues to deduce the source can also result in VR source confusion (e.g., "I knew that I had named it, but I have no visual recollection of it").In such cases, participants had to guess the source (e.g., "I wasn't sure if it was virtual or real.I guessed it") and rely on assumptions about the study (e.g., "I thought about the possible real things"), which can lead to mistakes.The lack of cues could be caused by a weak encoding due to poor attention during the study (e.g., "I paid less attention, therefore I could not have a clear image").It could also be caused by a natural forgetting of the amount of information that is stored over time.

5.3.3
Cues not specific to one source.Some cues, like the memory of the thoughts about the origin of the object, are specifc to one source and allow to identify the source without any doubt (e.g., "building a story about an object (is this the researcher's property or did they buy it for the experiment, how did they remove the label, does it show evidence of use, etc.) only happens with real-life objects").
However, the cues are not always specifc to one source (e.g., "the elements of the keys that I remember were specifc neither to reality nor VR").In such cases, participants either guessed or relied on assumptions about VR and reality.For example, a participant remembered that the book did not have a title ("I remember not seeing any title on the cover, and it feels like a virtual thing to not have a title on the cover").This cue is not decisive, since both a real and a virtual book could have a homogeneous cover.The participant incorrectly assumed that a real book would always have a title, resulting in source confusion.
Additionally, a participant points out that "plastic objects with fat surfaces such as the mouse and the hair dryer were difcult to remember if they were real or virtual.Because those shiny, artifcial, smooth surfaces are easier to simulate in 3D".While the memory of a smooth texture is associated with VR, this cue is not specifc to one source, since a real object could also have a smooth texture.

Incorrect assumptions.
Current knowledge and assumptions are an essential part of the source identifcation process.Particularly, incorrect assumptions about what can be possibly rendered and modeled in VR resulted in source confusion (Figure 9).Objects that participants assumed to be easy to render or modeled were incorrectly identifed as virtual (e.g., "I thought they were pretty easy to render and were more likely to be virtual"), meaning that underestimating VR capabilities could result in more VR source confusion.

DISCUSSION
Our perspectives join the discussion initiated by Slater et al. in 2020.
In "The Ethics of Realism in Virtual and Augmented Reality", they raise ethical concerns about the growing realism of VR, and how it might impact users in the long term, mentioning the creation of false memories.Among others, they raise these questions: "Can people already today be confused between reality and virtual reality?", "Does greater realism lead to greater confusion between the real and the virtual?" [71].

VR Source Confusion can Already Occur Today
The idea of a VR technology so advanced that it is indistinguishable from reality itself captivates imaginations and fantasies.Back in the 1960s, Ivan Sutherland envisioned the Ultimate Display, a visionary concept of a display that would transcend the boundaries of reality [77].Despite rapid technological advancement, passing the "Visual Turing Test" [3,75], where users would no longer be able to tell if they are in the real or virtual world, remains a monumental technological challenge.It would require substantial investments and major innovations in both hardware (e.g., weightless HMD, sensory feedback, resolution exceeding human vision) and software (e.g., realistic rendering and graphics, reduced latency).However, the results of our study confrm that VR can already be confused with reality despite the limitations of current technology in terms of realism and hardware capabilities.The confusion does not occur during the experience (at the direct perception level), as envisioned by the Ultimate Display, but when thinking back to the experience (at the memory level).Furthermore, source confusion occurred more frequently in our study (participants confused the source of 23% of the objects that were presented) than in Study H (18%) (Figure 5).This diference might be attributed to diferences in the study settings.For example, in Study H, participants were asked to touch the objects, potentially providing additional sensory cues that could aid in later accurately recalling the source.However, it is reasonable to think that such diferences would have impacted the confusion of the real and virtual objects equally, but the proportion of real objects that were confused for virtual ones is similar in our study (27%) as in Study H (26%) (Figure 7).This result suggests that improvements in VR technology made over the past two decades (i.e., improved feld of view, refresh rate, resolution) might result in increasing VR source confusion.In 2021, Rubo et al. found a lower rate of VR source confusion using HTC Vive Pro (16% of errors in the real vs VR source identifcation tests) [60], but these results cannot be compared to ours due to major diferences in the study design.For instance, participants were informed of the exact purpose of the study, and the source monitoring test was done directly after seeing the objects, which can explain fewer source confusion errors.

Impact of Presence and Realism on VR Source Confusion
Our results do not show any meaningful relation between the level of presence (measured with the SUS) and the occurrence of source confusion.However, the correlation we observe between SUS Q5 and source confusion errors is particularly noteworthy when considering the memory-related aspect of this question.SUS Q5: "Consider your memory of being in the virtual room.How similar in terms of the structure of the memory is this to the structure of the memory of other places you have been today?By 'structure of the memory' consider things like the extent to which you have a visual memory of the virtual environment, whether that memory is in color, the extent to which the memory seems vivid or realistic, its size, location in your imagination, the extent to which it is panoramic in your imagination, and other such structural elements." Q5 stands out from the SUS, as it is not focused on the feeling of "being there", unlike, for instance, Q1 ("Please rate your sense of being in the virtual room... "), or Q6 ("During the time of your experience, did you often think to yourself that you were actually in the virtual room?") (Appendix B).Q5 being a good predictor for source confusion suggests that people who tend to have a vivid, realistic, and panoramic memory of the virtual environment, are more prone to VR source confusion.This result is aligned with Reality Monitoring, where Johnson et al. explain that people who have a vivid imagination are more prone to source confusion between reality and imagined events [34].
From our qualitative analysis, we found that in the specifc case of VR vs real source identifcation, the cues that are used to distinguish the source are mostly related to visual and sensory fdelity (e.g., textures, lighting, level of details) (Figure 9).These cues are then compared with assumptions about what could possibly be rendered and modeled in VR.While Mizuho et al. did not fnd evidence for the impact of visual fdelity of the environment on source confusion, their study did not test the impact of visual fdelity of the content to be memorized itself (e.g., objects instead of written words) [49].From our results, and in conjunction with the SMF [33,34] and prior research [28,29,60,66], we argue that gradually enhancing the realism of VR experiences (in terms of rendering, modeling and sensory feedback) would gradually amplify VR source confusion, for two reasons: 1) Increasing realism would provide users with fewer distinguishing cues for source identifcation.
2) Increasing realism would disrupt the assumptions about VR capabilities.Prior knowledge and assumptions about VR technology capabilities are a central part of the source identifcation process (e.g., low level of details identifed as virtual, high level of details identifed as real, based on the assumption that it is difcult to render details in VR).Users underestimating what can be rendered in VR could be more prone to VR source confusion.
VR inherently tries to mimic reality and is getting better and better with the constant pursuit of more realism.This paper provides another argument for why it might be reasonable to consider what level of realism is desirable.

Risks of Perceptual Manipulations
Source confusion is a common memory error that is part of our everyday experiences.It often happens genuinely, and we may not even realize its efects.It can, however, also lead to signifcant consequences, such as false eyewitness testimony or misinformation [41,42,62,84].Beyond inadvertent occurrence, VR source confusion could also be leveraged to deliberately manipulate users.Bonnail et al. identifed several types of memory manipulations that could be enabled or amplifed by Extended Reality (XR) [9].Source confusion is part of the memory limitations that could be leveraged with malicious intent.They give a speculative example involving a VR social media platform where users can revisit 3D reconstructions of past events.Similar to current social media flters, these 3D reconstructions could be altered.Later on, users might confuse the real event with the altered reconstruction of that event, resulting in a distorted view of the past, making them vulnerable to attacks.The cues that usually enable users to identify a memory as virtual could intentionally be removed to increase the potential for VR source confusion.
In 2020, Mhaidli et al. identifed potential manipulative advertising techniques that could be enabled by XR [48].They identify "misleading experiences" and "distortion of reality" as two potential mechanisms for XR advertising, pointing toward the power ofered by greater immersion and extreme realism.VR source confusion could be leveraged for manipulative advertising purposes.For example, advertisers could expose users to a product in VR, leading them to later believe they experienced it in reality, potentially infuencing their future purchasing decisions.On a positive note, VR source confusion could also be leveraged for positive purposes, beneftting users.By blurring the lines between real and virtual experiences, VR source confusion could be used for therapeutic applications.
Given that memories shape our personality, social interactions, and future choices and behaviors [5][6][7], the impacts of source confusion and memory manipulations should not be underestimated.
Understanding and addressing the implications of VR source confusion is crucial for ensuring responsible and ethical development and use of this technology.While Slater et al. initiated the discussion on the ethics of realism in XR [71], our paper provides additional empirical evidence of how increasing realism brings new risks and challenges.In light of the recent rise in research projects exploring the side efects of XR at CHI [9,17,55,57,79] and HCI-related security conferences [11,58], we believe it is timely to engage this discussion within our community, where HCI researchers have the necessary skills and understanding of technology, social concerns, and human behavior.We do not argue for the abandonment of the vision of realistic and everyday usable XR but rather want to refect on how this technology can deliver all its benefts in an ethical and safe manner.

LIMITATIONS AND FUTURE WORK 7.1 Measure of VR Source Confusion
Our modifcations to the original study by Hofman et al. in 2001 could have infuenced source confusion.The inclusion of an engagement task and questionnaires before and after viewing the objects may have divided participants' attention, potentially leading to a less efective encoding of information, and therefore more source confusion.Asking participants to observe the objects without touching them might also have amplifed VR source confusion.The specifc conditions under which we observed source confusion do not fully represent the complexity of natural settings.Examining VR source confusion in more intricate and ecologically valid scenarios, beyond observing objects, would be of great interest.

Exploring Causes of VR Source Confusion
The cues that we identify from the source monitoring questionnaire are not exhaustive.A diferent study design could uncover additional cues (e.g., haptic feedback, implausible elements, or remembering the moment of entering VR were not mentioned due to the study settings but could be important cues).There is also a potential bias, as we asked participants to explain their reasoning after telling them the correct source.This might have infuenced their thought process after the fact.We encourage future research to further understand factors of source confusion.
In our study, every participant viewed the same environment with no variation in realism.Future studies should assess, through quantitative analysis, how diferent levels of realism (e.g., textures, lighting, resolution, level of details, sensory feedback) impact VR source confusion.
Furthermore, this questionnaire only captured the conscious aspect of source identifcation.Irrational or unconscious factors might infuence the process (e.g., participants sometimes answered "I don't exactly know the reason.I just felt that way", "I honestly don't know why I answered it was real.Maybe an intuition?").Also, some participants might have answered the source identifcation test intuitively, without conscious reasoning, and might have thought about reasons for their choices after the fact, when asked in the source monitoring questionnaire.
In this paper, we explored the impact of the sense of presence using the SUS questionnaire but did not fnd conclusive results.Presence is inherently subjective and involves several underlying components (e.g., plausibility, place, and body illusion, as proposed by Slater et al. [68,70]).We chose the SUS questionnaire for its widespread use and brevity, and we relied on the interpersonal variation of presence.When discussing presence, it would be more precise to refer to "the SUS score".Using alternative questionnaires may provide more insights and diferent results.Moreover, emphasis could be placed on other VR-specifc factors that we did not focus on, such as embodiment.
Future work should also further compare source confusion between VR and reality and source confusion between desktop screen and reality, building on Rubo et al. 's study [60], to better quantify VR source confusion, and to provide more understanding of the specifc factors of VR that infuence source confusion.

Design Guidelines to Mitigate the Risks
Identifying the factors impacting VR source confusion is essential to mitigate the risks it implies.For instance, if future studies reveal that using realistic colors and detailed textures signifcantly increases VR source confusion, then deliberate design choices like incorporating unnatural colors and basic textures in VR environments could be used to reduce confusion when preferred by users, or in contexts sensitive to memory manipulations (e.g., court settings)..
Researchers could then go further by designing a "Virtual Reality Check", a set of elements that could be integrated into the design of VR experiences to provide more cues to diferentiate VR from reality, inspired by the spinning top concept from the movie Inception2 , which is used to distinguish dreams from reality.

CONCLUSION
Through a laboratory study, we confrmed that individuals are not always able to distinguish if a memory originates from reality or from a VR experience.We identifed cues that are used to distinguish VR from reality, and assumptions about VR capabilities that result in VR source confusion.We contribute to the discussion initiated by Slater et al. [71] about the ethics of realism.Increasing realism would remove cues and disrupt the assumptions about VR capabilities, leaving users more vulnerable to VR source confusion, and more generally perceptual manipulations.While a lot of efort is given to constantly improving the realism of VR experiences, we argue that it becomes crucial to start implementing preventive measures to mitigate the risks of VR source confusion and other possible threats.
Disclaimer: The authors used Large Language Models such as ChatGPT to rephrase parts of the paper.The input was always a self-formulated paragraph with the request to improve phrasing and correct grammatical and spelling mistakes.

C QUESTIONNAIRE AFTER EACH REALITY EXPOSURE (ADAPTED SUS + ENJOYMENT AND BOREDOM)
(1) Please rate your sense of being in the room, on the following scale from 1 to 7, where 7 represents your typical experience of being completely aware of a place.I had a sense of being in the room: 1 (not at all) 2 3 4 5 6 7 (very much) (2) To what extent were there times during the experience when you were aware of being in the room?There were times during the experience when I was aware of being in the room... 1 (At no time) 2 3 4 5 6 7 (almost all the time) (3) When you think back to the experience, do you think of the room more as images that you saw or more as somewhere that you visited?The room seems to me to be more like... 1 (Images that I saw) 2 3 4 5 6 7 (Somewhere that I visited) (4) During the time of the experience, which was the strongest on the whole, your sense of being aware of the room or of being elsewhere?I had a stronger sense of... 1 (Being elsewhere) 2 3 4 5 6 7 (Being aware of the room) (5) Consider your memory of being in the room.How similar in terms of the structure of the memory is this to the structure of the memory of other places you have been today?By 'structure of the memory' consider things like the extent to which you have a visual memory of the virtual environment, whether that memory is in colour, the extent to which the memory seems vivid or realistic, its size, location in your imagination, the extent to which it is panoramic in your imagination, and other such structural elements.I think of the room as a place in a way similar to other places that I've been today... 1 (Not at all) 2 3 4 5 6 7 (Very much so) (6) During the time of your experience, were you frequently aware of being in the room?During the experience I was frequently aware of being in the room... 1 (Strongly disagree) 2 3 4 5 6 7 (Strongly agree) (7) to (10) 10: Correlation heatmap generated with JASP 0.17.2.1.Value indicated is Pearson's r. 0.1 < r < 0.3 is a small efect, 0.3 < r < 0.5 is a moderate efect, r > 0.5 is a large efect.Elements with a black border are also presented in Figure 8.

Figure 1 :Figure 2 :
Figure 1: Setup of the environment in Virtual Reality (VR) and Reality (R), containing a table (A), a box (B), panels (C), and a space for the participant (D).All the fgures of this paper, can be reused, under CC BY.

Figure 3 :
Figure 3: 30 objects in Virtual Reality (VR) and Reality (R).VR models imported from the OpenVirtualObjects database [78].List of objects in Appendix A. CC BY.

3. 4 . 2
Observation.Participants were instructed to observe the object on the box for 15 seconds.They were free to move around it.

Figure 8 :
Figure 8: Correlation heatmap generated with JASP 0.17.2.1.Value indicated is Pearson's r. 0.1 < r < 0.3 is a small efect, 0.3 < r < 0.5 is a moderate efect, r > 0.5 is a large efect.CC BY. 4.3.1 Correlations.We started the analysis with a larger exploration of correlations on all the variables (source identifcation errors, presence score, EMQ score, level of enjoyment, boredom, and confdence).The full correlation heatmap can be found in Figure 10.Pearson's correlation shows a correlation between the EMQ score and source confusion (r = +0.445), the EMQ score accounting for 19.8% of the variance in source confusion error (moderate to large efect size) (Figure8).This result makes sense since the EMQ is a general memory ability questionnaire that is likely to correlate with many memory tasks.It suggests that individuals who perceive their memory as frequently failing are more likely to experience source confusion between VR and reality, and highlights the importance of considering memory ability when examining VR source confusion.Pearson's correlation showed there was a weak correlation between the SUS Mean and source confusion (r = +0.301), the SUS Mean accounting for 9% of the variance in source confusion (moderate efect size).The SUS Count does not show a strong correlation with source confusion (r = +0.279,small efect size) (Figure8).The SUS questionnaire is commonly used to assess the sense of presence.However, it is not entirely clear how it is constructed, and the specifc roles and underlying constructs of each question remain unclear.Notably, there has been no factor analysis conducted to date[70].The various questions within the SUS may measure diferent aspects, which could potentially have varying efects on source

Figure 9 :
Figure9: Source identifcations process, including list of cues that were used by participants to distinguish VR from reality, and the assumptions associated with each type of cue.CC BY.
identical to Appendix B with "task" instead of "VR task" Last week, during the study, you were asked to observe some objects, sometimes in reality (physical objects), and sometimes in Virtual Reality (virtual objects, seen through the VR headset).In the following questionnaire, you will be asked to identify, for each object, if you saw it in reality, in Virtual Reality (VR) or if you did not see it during the study.Then, you will have to indicate how confdent you are in the accuracy of your response.Try to give spontaneous answers to these questions.You cannot go back on your answer once you click "next".The name of each object is given both in English and in French.If you don't know the meaning of a word, you are allowed to search for it in a dictionary."For each of the 30 objects: "[Name of the object in English and in French]: -This object was real (I saw it in the real world, without the headset) -This object was virtual (I saw it in VR, with the headset down over my eyes) -I did not see this object during the study How confdent are you in the accuracy of your response?possible questions, for each combination of real/virtual objects, correct/incorrect answer, high/low level of confdence, tailored for each participant: "Last week, you identifed the [list of objects tailored to participants] to be [real/virtual], and that was [correct/incorrect].It was actually [real/virtual] objects.You were [level of confdence]% confdent in your answers.What were the reasons that made you think that these objects were [real/virtual]?What were the reasons that you were [highly/not] confdent in your answers (if there are any)?"Additional question: "Do you have other comments about what helped you to identify the source of the objects, or about how you feel about this questionnaire in general?" "8