The experiment followed a between-groups design with 1 factor with 2 levels: Current Self and Young Self. In the Current Self condition, participants were embodied in a virtual body that looked like their current appearance, in a modern living room (Figure 1). In the Young Self condition, participants saw themselves in their younger body in VR, located in a living room decorated as if it were in their youth (Figure 2). In each case, there was a TV in the room playing adverts and excerpts from programs of the time, followed by the start of a program about the Eurovision Song Contest 1968. In the Current Self condition, participants were then transported to a theater where they saw the present-day appearance of the singer Massiel performing the same song that won the Eurovision Contest in 1968 (“La la la”), celebrating the 50th anniversary, displayed in the VR environment on a large screen at the front of the theater (Figure 3A-3B). In the Young Self condition, participants were transported to a virtual Royal Albert Hall in 1968 and experienced attending the concert performance with the singer Massiel, her backing group, and the orchestra performing at the Eurovision Contest (Figure 3C-3D). In all cases, participants were embodied in their current or younger self virtual bodies from a 1PP (refer to Procedures). Hence, participants saw their virtual body both by looking down toward themselves directly and also as reflected in a virtual mirror. Participants’ real body movements were also mapped to their virtual bodies in real time using VR tracking devices. Participants were randomly allocated to either the Current Self or Young Self condition, and they completed 5 experimental sessions. Further details are given in the Procedures subheading and Table 1.

Thirty participants were recruited from the Department of Medicine (Faculty of Medicine and Health Sciences, University of Barcelona) among those attending cognitive assessments as part of a follow-up study on determinants of healthy aging. There were 7 dropouts: 5 for reasons unknown, 1 because of cardiology advice, and 1 due to the bereavement of a close family member. Finally, there were complete data on 23 participants (11 in the Young Self condition and 12 in the Current Self condition). The resulting distribution of participants by condition and sex is shown in Table 2. All participants were healthy Caucasian women and men aged 65-85 years and were not affected by any illness that prevented regular daily activity such as walking. All were retired.

Additional inclusion criteria included the following: no consumption of alcohol on the day of the experiment (before participation), no use of psychoactive medication, no existing conditions that caused dizziness, no strong susceptibility to motion sickness (eg, while traveling by car, boat, or airplane), no history of epilepsy, no cognitive impairment as assessed with the Mini-Mental State Examination (MMSE) [29-31], no major disabilities (such as requiring a wheelchair or assistive devices), no traumatic events during the age at which they would embody their younger self, and no history of neurological or major neuropsychiatric diagnoses (as determined by their physician). All participants were compensated for their participation with a payment of €100 (€20 for each session; conversion rate: approximately US $104). Multimedia Appendix 1 provides a summary of each demographic, background, and basic health variables and MMSE score for participants across conditions.

The study was carried out at Hospital Clinic, University of Barcelona. The VR experience was presented via the Meta Quest 2 HMD. The device has a display resolution of 1832×1920 pixels per eye and is powered by a Qualcomm Snapdragon XR2 processor with 6 GB of RAM, ensuring smooth performance. The headset displays at a 90-Hz refresh rate for fluid motion and reduced motion blur. It has built-in spatial audio and integrated tracking sensors. The HMD was attached to an MSI Katana laptop (Intel Core i7, RTX3060, and 16 GB of RAM) from which the program was executed. The Quest also has 2 handheld controllers. Both the head and hands are tracked in 6 degrees of freedom. Using this tracking, the motions of the upper virtual body could be matched to the movements of the real body using inverse kinematics, as described by Oliva et al [32].

Participants attended the 5 sessions described earlier in Table 1, with 2 VR exposures spaced approximately 1 week apart. Participants attended the experiment at prearranged times and were randomly allocated to the experimental or control condition. An overview of each session is described next, and a summary is given in Table 1.

Throughout, we used Bayesian statistical analysis, which is especially appropriate for the exploratory nature of this research. Rather than reaching conclusions (ie, via hypothesis testing), we present probabilities. A particular advantage is that we have many response variables, and with frequentist statistics, every additional statistical test further invalidates the meaning of the “significance level.” We would then have to resort to ad hoc methods such as Bonferroni corrections. In the Bayesian method, all response variables can be treated simultaneously in 1 overall model, and we can derive as many probability statements of interest as needed without impacting their validity. Moreover, unlike simply quoting results such as “t=3.2 and P<.05,” we explicitly state all assumptions and provide the source code online so that anyone can repeat the analysis, change prior distributions or the likelihood function, and examine the responses.

We report posterior means and 95% highest density intervals (HDIs), which can be interpreted as the range within which the true parameter value lies with 95% probability, given the data and model assumptions. In the Bayesian framework, before using the data, parameter values have prior probability distributions, reflecting the current state of knowledge. These distributions are updated by the data to yield the posterior distributions. The HDIs computed from the posterior distributions can be compared to the prior HDIs, and they typically focus on a much narrower range of values. Similarly, we can compute the posterior probability that a parameter is positive (or negative), indicating the probability that it is positively (or negatively) associated with the response variable.

We describe each component of the overall model together with the analysis of each response variable.

The linear predictor for any response variable refers to the right-hand side expression in the statistical model for that variable. For all of our response variables, the linear predictor is of the form:

where individual represents the participant, condition is the main factor (Current Self=0 and Young Self=1), and session is (1,2,…,5). The last term is the interaction between condition and session.

The factors are represented by the following parameters:

uj is the random effect term of the jth participant (j = 1,2,…,n=23)

αj is the main effect of the jth Session (j = 1,2,…,5)

βk is the effect of the kth condition k (k = 1,2), where k=1,2 represent the Current Self and Young Self levels.

γjk,j = 1,2; k = 1,2,…,5,is the interaction effect between session and condition, representing the additional effects of Young Self over the sessions.

For model identifiability, we set:

Hence, parameters are interpreted as offsets from the first participant (an arbitrary reference), the Current Self condition, and Session 1.

The statistical model is equivalent to a 2-way ANOVA model with random effects for participants. Here, the uj account for the effects of the individuals, and the αj,j=2,…,5 account for the change from Session 1 to Sessions 2‐5 under the Current Self condition. The parameter β2 reflects the difference between Young Self and Current Self at Session 1. The γjk allow for interaction between the condition and session; that is, they provide the additional effect of Young Self compared with Current Self in Sessions 2‐5, beyond the effect given by β2.

Hence, β2 represents the initial difference between Young Self and Current Self in Session 1 (ie, before any intervention). The γ2k, k=2,3,…,5 represent the additional effects of being in the Young Self condition across sessions.

As explained at the start of this section, the Bayesian framework assigns prior probability distributions to the parameters, which are then updated by the data to produce posterior distributions:

The likelihood is the distribution of the response variable conditional on the parameters.

Traditional ANOVA assumes a normal distribution for the likelihood. However, Figure S1 in Multimedia Appendix 3 shows the histograms of the subjective variables by condition (Current Self and Young Self), and Figure S2 likewise shows the distributions for the performance measures. In the case of the subjective variables, most of the distributions are skewed, and therefore, skew-normal distributions were used for the likelihoods. Skew-normal distributions have an additional parameter (in addition to the mean and SD) controlling skewness, and when this parameter is 0, the normal distribution is recovered. Therefore, in any case, the normal distribution is a special case of the skew-normal distribution. With respect to the performance variables, these are time or strength measures, which are nonnegative and typically follow a Gamma distribution. The alternative to using these likelihoods would be to attempt ad hoc transformations of the response variables to find one that transforms them to normality, which is, in general, not recommended [52]. We only used a log transformation log(subjective age/age) to ensure that this variable was based on a difference rather than a division, thereby reducing outliers and bringing its distribution into line with the other subjective variables. The full statistical models, including prior distributions, are given in Multimedia Appendix 4.

For analysis of the parameters, we concentrate on the HDIs of their posterior distributions. The HDI is analogous to a traditional CI, but it is an interval with a posterior probability. The 95% HDI is the smallest interval that contains 95% of the posterior distribution of the parameter. This is in contrast to a CI in frequentist statistics, which is such that if there were a large number of identical repetitions of the experiment, then 95% of the computed CIs would contain the true value of the parameter. However, any specific computed CI (ie, the one based on the actual available data) does not have 95% probability of containing the true value. In the frequentist interpretation, this specific computed interval either contains the population value of the parameter or does not; hence, only a probability of 0 or 1 could be assigned. Additionally, for each parameter, we provide the probability that it is positive or negative (ie, positively or negatively associated with the response variable).

All these results are provided in Tables S1 and S2 in Multimedia Appendix 5, which give summaries of the posterior distributions of all the parameters. Here, we present a graphical summary for each response variable, including box plots of the raw data by session and condition, and HDIs represented as horizontal line segments (Figures 4 and 5). At the end of each line segment representing the HDI, the probability of the parameter being positive is shown.

The experiment was approved by the University of Barcelona Bioethics Commission (Comissió Bioètica of Universitat de Barcelona) and conducted in accordance with institutional ethical guidelines and international standards for the protection of human participants. Ethical measures included obtaining written informed consent, ensuring the right to withdraw at any point during or after the experiment, and maintaining confidentiality. Participants were informed that their data were strictly private and that the researchers would only access an anonymous ID number rather than any identifying information. After the final phase of the experiment, participants were debriefed about the purpose of the study. Because the intervention involved viewing and embodying a self-representation at a different life stage, participants were screened for factors that could increase distress (including major neuropsychiatric diagnoses and traumatic experiences during the target age period). Participants were reminded that they could pause or discontinue the VR exposure at any time without penalty.

Table 5 summarizes participants’ demographic characteristics, background, and basic health variables, including ratings of subjective age and cognitive impairment. Overall, the groups were comparable, with no notable differences observed. Participants generally rated their health as good, and MMSE scores were above the standard 23/24 cutoff (overall mean 28.7, SD 0.78), indicating normal cognitive functioning [53].

Figure 4A shows the box plots for the presence variables. There appears to be no difference in presence scores between the conditions, except for the variable “visited” in Session 3. In Session 2, the scores for “others” are low, whereas in Session 3, they are high. This is because there were no others in Session 2; thus, these data are consistent.

Figure 4B shows the results for the recognition, body ownership, and agency variables shown in Table 3. While in Session 2, people recognized their current selves, they did not recognize their younger selves. However, this was rectified in Session 3. For the Young Self condition, the body ownership scores for the variables mybody and mirror were low in Session 2, but this was rectified in Session 3. Agency-related scores (mymovements) were high in all conditions and sessions, reflecting the operation of the tracking system and the program.

This experiment was not about body ownership as such, and therefore, we do not make inferences about these variables to the broader population. Rather, we report them as manipulation checks. Participants’ self-reported ownership and agency ratings were generally consistent with the intended embodiment manipulation, particularly by the second VR exposure (Session 3), when recognition of the younger avatar improved.

The questionnaire included the following question: “How much dizziness or nausea did you feel while in the virtual reality?” on a 7-point scale ranging from 1 (“Not at all”) to 7 (“Very much so”). For both sessions and conditions, the median score was 1, with the IQR of 0. There were 2 outliers, a score of 7 in Session 2 (the first VR exposure) and a score of 3 in Session 3.

With moderate to high probabilities over the sessions, those in the Young Self condition reported lower subjective age than those in the Current Self condition. With moderate probabilities, the Young Self group had greater AARC Positive scores than the Current Self group in Sessions 4‐5. This means that they tended to perceive more positive changes with age. With moderate probabilities, the Young Self group had greater well-being scale scores than the Current Self group in Sessions 2 and 4. This scale indicates a greater sense of well-being. In Sessions 2 and 4, the Young Self group had lower TMT Part A times than the Current Self group, with high probability. This refers to the time required to complete the TMT. The number of mistakes made in the TMT Part B was lower in the Young Self than in the Current Self condition in Sessions 2 and 4, with moderate probabilities. The right-hand grip strength measured for the Young Self group was higher in Session 4 than that of the Current Self group, with moderate probability. The left-hand grip strength was greater in the Young Self group overall, with high probabilities.

The statistical analysis did not find any salient differences between the groups for Philadelphia total (a measure of morale), AARC Negative (reporting more negative changes with age), TMT Part B time (the time to complete a TMT), and balancemean (time standing on 1 leg).

It is interesting to see that differences between the 2 groups were most often found in Sessions 2 (the first VR session) and 4 (1 week after the VR exposures). However, it is important to note that all of the effects had vanished by Session 5, except for AARC Positive (with moderate probability), TMT Part B errors (with moderate probability), and left-hand grip strength (with high probabilities).

This study tested whether a VR analog of the counterclockwise paradigm, combining immersion in an autobiographical past context with embodiment in a younger-looking virtual body, produces measurable improvements in age-related subjective and performance outcomes relative to a Current Self VR control. We conducted a between-groups experiment in which a group of older adults was embodied in a virtual body that either resembled themselves as they are today (Current Self) or as they were in their teenage years or early 20s (Young Self). The Current Self group spent time in a modern virtual living room, where they saw their virtual body when looking down at themselves and when it was reflected in a mirror, and which moved synchronously with their own movements. They were then transported to the audience of a concert, where the modern-day Massiel performed the song that won her the Eurovision Song Contest in 1968. The Young Self group followed the same procedures, except they were embodied in their young body, in a 1960s living room, and were transported into the audience of a VR depiction of the original 1968 Massiel performance. Hence, they were either transported back in time with respect to their appearance and location (Young Self) or remained as they are now with respect to their appearance and the setting (Current Self). The experiment was conducted over 5 sessions (1: initial contact, 2: first VR exposure, 3: second VR exposure, 4: one week later, and 5: about two weeks later) with a number of subjective response measures and performance measures on each occasion.

Relative to the control, the Young Self condition showed the clearest evidence of benefit for subjective age and well-being, alongside improved TMT performance and higher grip strength, with effects most evident at Session 2 and at 1-week follow-up (Session 4). Several outcomes showed little evidence of between-group differences (eg, Philadelphia morale, AARC Negative, TMT Part B completion time, and balance), and most effects diminished by Session 5, but with persistent results observed for AARC Positive, TMT errors, and left-hand grip strength.

VR offers a potentially powerful method for delivering RT. Rather than looking at old photographs, newsreels, or home movies, participants can be directly immersed in scenes from the past. The systematic review of 22 studies by Ng et al [54] shows, most importantly, that across a wide range of interventions, there was little evidence of contraindications such as simulator sickness, and some positive outcomes for cognition. For example, Niki et al [55] carried out a study in Japan where VR was used among 10 participants aged at least 75 years who either saw scenes captured from a 360-degree camera (“live action”) or computer graphics–generated scenes themed from 1955 to 1980. They found that the method reduced anxiety and that there was some preference for the live action method, without contraindications such as simulator sickness. Likewise, a scoping review by Restout et al [56] found that participants positively appraised the use of 360-degree videos in this domain. Tominari et al [57] conducted a study in which 52 individuals with mild cognitive impairment participated in 8 weeks of RT, using either tablet-based panoramas (ie, nonimmersive) or conventional photos. Both groups demonstrated significant improvements in cognitive functions, with the tablet group showing higher morale. Similarly, Huang and Yang [58] carried out a pilot study involving 20 participants with dementia, where VR RT sessions—featuring scenes such as historical residences, music, and photographs—were held twice weekly for 3 months. While the VR intervention did not lead to significant cognitive changes immediately after the treatment, it did result in improvements in depressive symptoms, with evidence of decline over the next 6 months.

Khirallah Abd El Fatah et al [15] used personalized immersive VR reminiscence scenarios created using participants’ photo albums and videos, along with background information such as songs and friends gathered during interviews. These scenarios were compared with traditional RT and a control group that received standard care in older adult homes. While both the traditional and VR RT enhanced cognitive function and psychological well-being, the VR approach demonstrated greater efficacy.

While much RT research is mainly empirical, Lau et al [59] introduced a conceptual model (Immersion, Interaction, Imagination, and Impression) to help choose the content suitable for VR RT scenarios. This involved carrying out a survey conducted to identify suitable reminiscence materials for different age groups in Hong Kong, followed by the development of a prototype application. The prototype was then tested in a pilot study with occupational therapists, who provided positive feedback on its potential to offer a more engaging therapeutic experience compared to traditional methods. They noted that it helped elicit positive attitudes and motivation among older adults with dementia and improve their condition. The content was chosen to represent culturally significant scenes from various periods of Hong Kong’s past, tailored to specific age groups.

The most important aspect of our paradigm is that we do not use VR only to reproduce the appearance of the past, but we also change the body of the participant to represent the corresponding younger age. This was impossible in the original counterclockwise study, but it is straightforward in VR. As we have seen, other uses of VR in RT immerse people in representations of the past, but the issue of how they themselves appear is not considered. In this sense, we have two findings: (1) that RT in VR can produce several indicators of at least short-term improvement, and (2) the young virtual body can play an important role. It could be argued that our control condition should instead have been a condition without any virtual body self-representation. However, the fact that we found, on some variables, a difference between the Young and Current Self representations addresses this objection, since we have shown that changing the body has an effect. A study without any virtual body representation would not add additional information.

It might be thought that age may impact the propensity of older adults to achieve body ownership; in particular, they might not tolerate the small differences between proprioception and the seen movements of the virtual body that are inevitable in a VR setup. However, Martinelli et al [60] found that as people age and their proprioceptive sense declines, they rely more on the visual sense, so that movements of the virtual body would override such inaccuracies.

In relation to the impact of the representation of people at a different age, Reinhard et al [61] studied the effect on walking speed of embodying younger adults in older bodies and found that the speed after embodiment was lower. However, we did not find the converse effect, that embodying older adults in their younger body would increase their walking speed.

Our findings on subjective age and handgrip strength are also consistent with experimental and observational work on subjective age and physical functioning. In an experimental study, Stephan et al [62] provided older adults with positive, age-referenced feedback after an initial handgrip test; those who were led to believe that their performance was good for their age subsequently reported a younger subjective age and showed stronger handgrip on a second trial compared with controls who did not receive such feedback. In a complementary experiment, Swift et al [63] showed that framing a handgrip test as a comparison with younger adults (“Are they half as strong as they used to be?”) impaired older adults’ handgrip strength and persistence relative to a neutral comparison condition, indicating that age-related social comparison cues can immediately weaken physical performance. However, our results suggest that the Young Self condition functioned as a positive prime, successfully counteracting typical age-related decline by supporting the perceived physical capability of the participants through their rejuvenated virtual identity as their younger self and through exposure to an event from their youth.

Beyond these lab manipulations, Wang et al [64] found that community-dwelling older adults who felt younger than their chronological age demonstrated better performance on several physical fitness indicators, including grip strength. In a geriatric outpatient sample from Northern India, Bhattarai et al [65] reported that each additional kilogram of handgrip strength was associated with a higher likelihood of feeling younger than one’s peers. Large cross-sectional data from the Successful Aging Evaluation study similarly show that feeling younger than one’s chronological age is associated with better physical and mental health across the adult lifespan [66]. Taken together, this literature suggests that subjective age is modifiable and closely linked to markers of physical functioning. Against this backdrop, the present results—in which older adults were embodied in a younger virtual body—tended to reduce subjective age and enhance left-hand (and to some extent right-hand) grip strength relative to the Current Self condition and can be interpreted as a proof of concept for the current intervention. We suggest that VR-based reminiscence and embodiment paradigms tap into similar subjective-age mechanisms that have been demonstrated in non-VR experimental and epidemiological work.

Although not in the context of RT, there is related work in which people were embodied in bodies of different ages. For example, Hershfield et al [67] embodied younger adults in older bodies, and this impacted their behavior with respect to saving behavior. Tammy and Wu [68] embodied older adults (aged at least 60 years) who did not engage in vigorous activity in younger virtual bodies, finding that this led to greater perceived exercise exertion. Vahle and Tomasik [69] carried out studies with young and older adults to examine the effect of embodiment in differently aged bodies on cognitive and physical functioning. They found that younger participants embodying older avatars experienced performance declines in both cognitive and physical tasks compared with those with younger avatars. However, and in contrast to some of our findings, they did not find any significant performance differences in physical or cognitive domains for older participants embodying younger avatars vs older avatars.

The most important limitation of the study is the small sample size. This occurred for two reasons: (1) the difficulty in recruiting from this population, and (2) the relatively large drop-out rate due to illness or other factors more likely to occur among older participants. However, the Bayesian analysis is ideal for this. We start with prior distributions with high variance. Each additional data point updates the probability distributions of the parameters of the statistical model. Of course, the greater the number of participants, the narrower the posterior HDIs may become. If the data had no effect on the response variables, then the posterior distributions would not change much from the priors. However, in this experiment, if we compare the HDIs of the prior distributions (Table S2 in Multimedia Appendix 5) with the HDIs of the corresponding posterior distributions (Tables S1 and S2 in Multimedia Appendix 5; Figures 5 and 6), we can see that they are much narrower than the priors. Hence, clearly, the data had an important effect.

Beyond sampling limitations, other procedural factors may have further constrained the study’s findings, and these should be addressed in future work. Due to technical limitations in the avatar reconstruction pipeline, the young avatars were often less realistic, having been derived from low-resolution, black-and-white historical photographs that remained suboptimal despite artificial intelligence (AI)–based enhancement. This discrepancy may have weakened participants’ sense of body ownership and inadvertently confounded condition assignment with avatar quality. AI-based enhancement may also produce unintended perceptual effects, such as eliciting an uncanny valley response [70]. Although this is unlikely in our manipulation, since the virtual body representations were still far from photorealistic, such effects may become more relevant as future enhancements increase realism. Future studies should therefore treat AI enhancement as a perceptual manipulation that requires validation, for example, by conducting brief pilot tests that include ratings of avatar eeriness and likability.

Additional problems included the need for experimenter-prompted verbal guidance during some VR sessions. Since the tasks were audio-guided, participants frequently required additional assistance, leading to interruptions that may have reduced presence for some participants. Future setups should incorporate in-scene virtual agents capable of offering guidance and companionship within the virtual environment, thereby preserving presence while minimizing experimenter intervention.

Hardware constraints may have also affected participant experience. For example, the headsets tethered to the computer limited mobility, and battery depletion or incompatibility with corrective glasses likely reduced comfort and visual fidelity, particularly among participants with pre-existing vision impairments. Furthermore, extending the duration of VR exposure would give participants more opportunity to acclimate to both the virtual environment and their embodied avatars, mitigating the disorienting effects of technological novelty in this age group. Finally, repeated administration of certain neuropsychological tests may have introduced practice effects or ceiling limitations, obscuring potential improvements. Future protocols should diversify cognitive measures to ensure sensitivity to genuine change.

Although the present experiment was not designed to investigate body ownership as such, the intervention necessarily relies on inducing a degree of body ownership in a self-resembling virtual body, which raises ethical and safety considerations. For example, experimental work suggests that immersive VR can transiently increase dissociative experiences, including depersonalization and derealization (DPDR) symptoms, particularly among individuals with higher preexisting dissociation tendencies [71,72]. However, recent cross-sectional survey findings indicate that most current VR experiences are unlikely to cause long-term DPDR symptoms for the majority of users [73]. In addition, avatar-based exposures can influence self-perception and body image attitudes [74]. A recent systematic review highlights how digital bodies might expose users to privacy and identity issues [75]. To mitigate potential risks, our study design incorporated several procedural safeguards, including rigorous participant screening for neuropsychiatric diagnoses and prior trauma, time-limited exposures, and a controlled laboratory environment. However, since we did not use standardized assessments for DPDR or body image changes, we cannot exclude the possibility of subtle or transient altered self-experiences. Future research should integrate validated state measures of dissociation, monitor for adverse psychological reactions both immediately postexposure and during follow-up, and establish prespecified stopping rules and support pathways for participants in distress.

Furthermore, although we manipulated embodied age between groups, our study does not directly test whether switching from a younger to an older or current virtual self (or vice versa) produces asymmetric emotional responses. However, participants in the Young Self condition necessarily transitioned from viewing and moving as a younger-looking self in VR to re-encountering their current real body when the headset was removed. This “re-entry” could plausibly produce a contrast effect, heightening AARC in a manner analogous to temporal self-comparison processes often implicated in nostalgia-related reactions [76]. However, since many of the responses (especially the reduction in subjective age) lasted at least 1 week after the VR exposure, this was unlikely to have been a problem. Future research should examine whether effects differ depending on the direction of embodied-age change and test “re-entry” procedures (eg, a brief grounding and debriefing phase or a gradual transition back to a current-self representation) designed to minimize potential negative contrast effects [77].

Overall, our experimental study has shown that a VR version of counterclockwise, where people are placed back in the past with a virtual body that looks like their younger self, can have a number of positive effects on responses that indicate self-improvement to overcome some effects of aging. However, most of these effects do not last beyond about 3 weeks. In a way, this is not surprising, because in the original counterclockwise experiment, the participants spent almost an entire week in the house decorated to look like the 1950s, in their 1950s clothes and with 1950s TV and magazines, whereas in our experiment, there were just 2 short exposures. However, our results hold promise for future work with multiple exposures, each with a different scenario.

Apart from the audience in the theater, the participants were alone in the VR scenarios. In earlier pilot studies, some participants suggested that it would have been useful if they had been accompanied by virtual human characters representing people from their past (in the Young Self) and present (in the Current Self) scenarios. This is possible, in that we would require photographs of others to construct their virtual bodies. However, for the Young Self condition, there may be ethical or data protection problems, for example, in representing people who were no longer alive and therefore unable to give permission, and for older adults, it would be quite likely that significant others from their youth (such as parents) would no longer be living, although some might be (siblings and friends). This may further complicate the process of recruitment, since not only would we need to find participants, but also those who had significant others able to permit the use of their photographs at a younger age. While this would make recruitment more difficult, it is likely to be worthwhile.

More broadly, these findings suggest that VR-based interventions may be able to operationalize “age mindset” manipulations in a scalable and controllable way, allowing systematic testing of intervention “dose” (duration, repetition, and variety of scenarios) and mechanistic ingredients (past-context cues vs self-representation). If replicated in larger studies, this approach could provide a method for low-burden behavioral interventions targeting subjective age and related functional outcomes, with potential extensions to populations where reminiscence-based approaches are already used (eg, mild cognitive impairment or depressive symptoms), while maintaining careful attention to usability and safety in older adults.