In social gerontology literature, the link between social support and depression is well established. Buffering Hypothesis (Cohen and Wills, 1985) is the most influential theoretical perspective on social support, which hypothesizes that social support reduces the effects of stressful life event on health through either the supportive actions of others or the belief that support is available (Lakey and Cohen, 2000). In other words, the existence of one’s social network, as well as substantive interactions generated among social ties, can buffer people from negative and stressful events. Therefore, social support acts to reduce the chances of stressors and negative events that provoke depression.

A sizable number of aging studies have established the relationship between social support and geriatric depression (Verstraten et al., 2005; Lee et al., 2012; Su et al., 2012). Given that a person’s social network decreases over time in late life (Antonucci, 1991), older adults receive less emotional support from social networking. This could increase the risk of depression. Barg et al. (2006) found that high adequacy of social support that older adults receive corresponds to low depressive scores. In a study with forty elderly aged 60 years and above, Patil et al. (2014) found a significant negative correlation between perceived social support and depression.

Besides the direct effect on geriatric depression, social support would also influence geriatric depression through loneliness. Loneliness refers to the subjective feeling state of being alone, separated, or apart from others (Tomaka et al., 2006). It is one of the strongest predictors of depression according to Singh and Misra (2009). Strong association between loneliness and geriatric depression was further confirmed in a large scale study (Aylaz et al., 2012). The longitudinal study of Cacioppo et al. (2006) supported that loneliness is a significant risk factor for depression among the aging population. Cohen-Mansfield and Parpura-Gill (2007) proposed a framework, Model of Depression and Loneliness, to examine factors influencing loneliness and depressed affect in older population. According to this framework, the reduced social contacts that occurs with age will influence loneliness, which, in turn, will affect depression. The feeling of loneliness is common for elderly people in their late life because of the lack of close family ties (e.g., living alone), impaired social support, and loss of mobility in social activities. The negative effect of social support on loneliness among older adults has found in many studies, especially on the Asian contexts (Kang et al., 2018; Chen et al., 2019). Lonely people suffer from more depressive symptoms, as they have been reported to be less happy, less satisfied, and more pessimistic (Singh and Kiran, 2013). As a result, loneliness may play a mediation role on the relationship between social support and depression among older adults.

In video games, social support is likely to be affected by different play mode. Many researchers have examined the social support derived from video games with multiple players (Trepte et al., 2012; Zhang and Kaufman, 2016a). In contrast with those in single-player games, players in multiple-player games need to learn and apply their social skills to achieve game goals. While interacting with others, players enhance their in-game social networking and interaction in these virtual social communities. This phenomenon is more obvious in games designed specifically for engaging cooperation and mutual assistance (Ewoldsen et al., 2012). In these collaborative games, players in a team assist each other and share successes and failures together. These social interactions strengthen group cohesion and interdependent bonds among teammates (Isbister, 2010; Banks, 2012), and foster strong feelings of virtual support and new friendships (Smyth, 2007). These virtual support and relationships in multiple-players gaming may affect and extend players’ pre-existing relationships in real life (Williams, 2006). Gentile and Gentile (2008) argued that social skills learned and practiced in the gaming environment could be generalized into the real context, which may promote prosocial behavior and enhance social support in daily life. Trepte et al. (2012) has presented a large-scale study to examine the theoretical framework of how social interactions in online multiple-player gaming affected offline social support. In their study, players’ social proximity and mutual familiarity during online interactions were supported to foster both online bridging social capital (people feel informed and inspired by each other) and bonding social capital (emotional support and understanding), while both social capital dimensions are positively related to offline social support.

The theoretical framework of gaming and social support provided above can be applied into the context of exergaming and older adults. It is reasonable to posit that exergames in multiple-player mode lead to higher level of social support than those in single-player mode, and the higher level of social support further results in greater reduction of depression among older adults.

Similar to social support, play mode in video games may affect loneliness through social interaction and communication among players. Li and Counts (2007) argued that MP games could actually increase communication between teammates via digital and traditional channels. Further, such game mode could create feelings of co-presence and make players feel more connected to each other (Lazzaro, 2004). The co-presence and connection to others could therefore reduce loneliness. A systematic review from Li et al. (2018) identified that playing MP exergames could improve older adults’ social well-being, through increasing social bonding with their peers and grandchildren. In the study from Kahlbaugh et al. (2011), older residents received personal visits from students who played a bowling game on Nintendo Wii console with them for 10 weeks. Results showed that older residents felt less lonely than they had at the start of the study. The decrease in loneliness was perhaps not due to playing the Wii itself, but by the interactions between older participants and students. In a large sample survey study, Lee and Ishii-Kuntz (1987) indicated that doing something together with other people reduced loneliness among older adults.

As a result, compared to SP exergames, the social interactions and communication in MP exergames lead to lower level of loneliness, and thus results in greater reduction of depression among older adults.

This study sought to build a conceptual model to address the influencing mechanism between exergame play mode and depression in old age. The model is rooted in cognitive-behavioral theories and conceptualizes depression as resulting from an interaction of social events and cognitive processes. It utilizes two key concepts, social support and loneliness, to clarify how play mode can facilitate the depression treatment in the exergaming context. Figure 1 shows the proposed conceptual model in this study. Based on the model, five hypotheses were proposed in the specific context of exergames and subthreshold depression in late life:

Participants were recruited from two senior activity centers located in the western and northern parts of Singapore, respectively. The center managers assisted in delivering recruitment message and approached potential participants. Generally, older adults were included in the study if they were aged 55 and above, and diagnosed with subthreshold depression in the screening section. Patient Health Questionnaire – 9 (PHQ-9; Spitzer et al., 1994) was used as a screening tool for admission into the study. Followed the scale instructions from Kroenke and Spitzer (2002), the total PHQ-9 score below “4” indicated none depression, whereas “5” to “14” indicates mild to moderate depression, and “15” to “27” indicates moderately severe to severe depression. Therefore, only participants who scored 5–14 in PHQ-9 were included in the final pool. Given that the particular interventions, participants were asked to self-report their previous experience on exergaming system. Only those with no experience or little experience (less than a few hours performance) were included in the study. Further, through self-reporting and information obtained from center managers, participants were excluded if they had the following conditions which prevent them from performing the exergame correctly and safely: (a) have serious cognitive disorders (e.g., Parkinson disease or Alzheimer’s disease), (b) intellectual disability and physical limitations (severe mobile, visual, or hearing problems), (c) other depressive disorders (e.g., bipolar, dysthymic, schizophrenia), or (d) have received antidepressant medications (or other forms of depression therapies) within the last 3 months.

Based on a systematic review of Li et al. (2015) which examined the overall effect size of exergames on depression, a general effect size of 0.25 was used in the current study to determine the sample size. Through the G^∗Power software with a priori analysis method, a sample of 54 participants is required in order to achieve the according general effect size with a power of 0.9 (α = 0.05). What’s more, the results of this systematic review shown that majority of previous studies used a small sample size of less than 50 participants. As a result, the current study targeted on 60 participants (30 per condition) which is slightly higher than recommended number.

In randomization process, each involved participant was labeled with a unique ID after screening, and an online tool¹ was used to generate two experimental conditions with random participant IDs. In order to avoid the interaction, these two experimental conditions in the same center were conducted in different time slots.

The ethical approval of this study was obtained from the Institutional Review Board at the university where the research team was originally from (IRB-2014-07-039). Participation in this pilot study was voluntary, and informed consent was obtained for all participants before the study. Each participant was awarded SGD$15 equivalent of shopping voucher after having completed all sessions.

To manipulate play mode, two different exergame conditions – SP and MP – were involved in the study. Nintendo Wii Tennis was chosen as the exergame for both conditions because of two reasons. First, according to previous literatures (Gao and Mandryk, 2011; Theng et al., 2012), older adults were recommended to perform physical activities with simple and single movements because of their impaired physical and cognitive conditions. The Wii Tennis game fulfils these safety criteria and has already been used as exercise intervention for older adults in previous studies (Rosenberg et al., 2010; Maillot et al., 2012). Second, the tennis game contains both SP and MP modes, which could ensure equal comparison between two conditions.

In SP condition, participants were asked to perform the exergames individually and played against a virtual player. Due to practical considerations on schedule and arrangement, participants came with a group of 15 people in one session and took turns to play the exergames. Despite the group setting, each participant was instructed to play on their own and without any physical assistance from others during the gaming. The total time for a group was around 3 h in one session. In the MP condition, two participants formed a team and cooperated in the same game to play against two virtual players. Similar to SP condition, all participants in came with a group of 14 people (seven pairs of participants) and performed the exergames in one session. To balance the performance time with the SP condition, the total time for MP condition was around 1.5 h for each session. The MP mode in the study applied collaborative games but not competitive ones. Compared to collaborative mode with the shared objective, competitive mode with the personal objective may elicit anxiety and fear of failure (Wu et al., 2015). These emotions elicited in competitive games may have negative effects on depression.

Participants from both conditions performed the corresponding exergame interventions once a week for 6 consecutive weeks. During the intervention period, one or two student assistants coordinated each exergame session. They assisted in setting up the exergame and solved the technical problems occurred during the game playing.

Participants were asked to fill in two self-reported surveys at both pre-study (before the first exergame section) and post-study period (after the last exergame section). The basic demographic information of participants was collected in pre-study survey, including their age, gender, education level, living conditions, and physical health conditions. The three key variables, including subthreshold depression, social support, loneliness, were collected at both pre-study and post-study period. Trained student assistants from the research team assisted participants who had literacy and visual problems in filling the questionnaires. Both English and Chinese versions of questionnaire were used in the study. Most of the measurements were adopted from existing sources to ensure the validity. Participants selected the language version that matched their preference.

Descriptive statistics of key variables were first conducted for both pre- and post-study data, following by a series of mixed analysis of variance (mixed ANOVA) to examine the interaction effects between time (pre-study vs. post-study) and play mode (SP game vs. MP game). Time was input as a within-group variable while play mode as a between-group variable. Both the descriptive statistics and mixed ANOVA tests were performed in IBM SPSS version 23.

Path analysis was then applied to test the conceptual model proposed between play mode and subthreshold depression. Path analysis is one of the most common techniques used in structural equation modeling (SEM). SEM is a family of statistical methods with a special aim to develop and test theoretical models. SEM evaluates the relationships between observed and latent variables (Hoyle, 1995). In path analysis, goodness-of-fit is first evaluated, followed by the estimation of path coefficients. Path coefficients, also known as standardized regression coefficients, determine the effects of exogenous (predictor) variables on endogenous (predicted) variables.

Different from common SEM in cross-sectional designs, the conceptual model in the current study involved a categorical exogenous variable (play mode). Bollen (1989) clarified that the inclusion of categorical exogenous variables did not violate the assumption of multivariate normality underlying the commonly used maximum likelihood method of estimation. MacCallum and Austin (2000) further indicated that one can use SEM to model the relationships of experimentally manipulated independent variables (such as dummy variables), to other variables, including covariates, mediators, and outcomes. All path analyses were conducted in Mplus version 6.0 (Muthén and Muthén, 1998).

A total of 52 participants completed both pre- and post-study surveys. The mean age of included participants was 72.12 (SD = 8.65), and 39 (75%) of them were female participants. Descriptive analysis was conducted across the two conditions on demographic characteristics and key psychosocial variables at pre-study period. Table 1 illustrates the detailed results. Findings from t-test and Chi-square tests indicated that participants from the two conditions did not have significant differences in demographic characteristics, as well as physical and cognitive status. Furthermore, no significant group difference emerged during the pre-study period in depression [t(50) = 0.56, p = 0.582], social support [t(50) = 1.24, p = 0.221], and loneliness [t(50) = 1.62, p = 0.111]. As a result, participants in SP and MP can be considered as equal in demographic, physical, and psychosocial conditions. Table 2 indicates key psychological outcomes across two conditions at both pre-study and post-study period. Findings indicates significant bivariate correlations among subthreshold depression, social support and loneliness after the intervention period.

The results of the mixed ANOVA did not show any significant interaction effects between time and play mode on social support [F(1,50) = 2.710, p = 0.106, η² = 0.051], loneliness [F(1,50) = 0.005, p = 0.944, η² = 0.00], or subthreshold depression [F(1,50) = 0.802, p = 0.375, η² = 0.016]. Figure 3 demonstrated the plots of the interaction effect among the three psychological variables. Despite the MP condition seemed to release more increase in social support and more reduction in subthreshold depression over the 6 weeks when compared to SP condition (Inspected from the descriptive statistics in Table 2), these changes were not statistically significant.

Nevertheless, there were strong significant main effects of time on all the three key psychosocial attributes, including social support [F(1,50) = 24.152, p < 0.001, η² = 0.326], loneliness [F(1,50) = 25.027, p < 0.001, η² = 0.334], and subthreshold depression [F(1,50) = 117.431, p < 0.001, η² = 0.701]. The findings indicated that the older adults had improvements on social support, loneliness, and subthreshold depression after the 6-week exergames playing. As a result, the strong effects of exergames were supported as a psychosocial intervention for older adults with subthreshold depression.

In path analysis, goodness-of-fit was first assessed through multiple fit statistics, including Chi-square test, Comparative Fit Index (CFI), Root Mean Square Error of Approximation (RMSEA), and Standardized Root Mean Square Residual (SRMR). Chi-square test (χ²) is the most basic fit statistic that compares predicted covariance matrix with observed matrix (Bentler, 2004). To reduce the effects of sample size, normed Chi-square (χ²/df) was calculated. A value of normed Chi-square smaller than 3 indicated an acceptable fit. CFI assesses the relative improvement in fit of the researcher’s model compared with a baseline model (Hu and Bentler, 1990). Values greater than roughly 0.90 may indicate reasonably good fit of the proposed model. RMSEA measures error of approximation (Steiger and Lind, 1980) with a value less or equal to 0.05 considered as good fit and 0.05–0.08 considered as fair fit. SRMR is a measure of the mean absolute value of the covariance residuals. A recommended cut-off point for SRMR is smaller than 0.08 (Hu and Bentler, 1990).

Based on the above criteria, the proposal model resulted in a good fit to data. Table 3 shows the goodness-of-fit indices and model fit results. Estimations of path coefficient and hypotheses testing were conducted after determining the final model. Figure 4 shows the results of the path analysis in the final model. During the hypotheses testing, significant negative effects of exergame play mode were observed on loneliness among older adults (β = −0.25, SE = 0.12, p = 0.042), thereby supporting H1b. However, H1a was not supported with no significant play mode effects on social support (β = 0.10, SE = 0.14, p = 0.488). Consequently, compared to SP exergames, MP exergames led to more reduction in loneliness among older adults, but not in social support. A strong and significant predictive effect from loneliness was also observed on subthreshold depression among older adults (β = 0.59, SE = 0.09, p < 0.001). Meanwhile, subthreshold depression was also significantly affected by social support, with β = −0.20, SE = 0.10, p = 0.047. Thus, both H3 and H4 were supported. However, the direct effect of exergame play mode was not significant on subthreshold depression among older adults (β = −0.08, SE = 0.10, p = 0.434), thereby not supporting H1c. Lastly, H5 was supported by a significant path from social support to loneliness (β = −0.32, SE = 0.12, p = 0.008), which highlights the potential effect of social support on loneliness during exergame playing.

Following the guideline by Stride et al. (2015), the mediation tests were further conducted on Mplus with Andrew Hayes’ PROCESS analysis. The mediation results indicated that play mode in exergames has a (approaching) significant indirect effect on subthreshold depression via medication of loneliness, with β = −0.15, SE = 0.08, p = 0.055. But the mediation role of social support was not supported in the relationship between play mode and subthreshold depression, with a non-significant indirect effect of β = −0.02, SE = 0.03, p = 0.511. Consequently, the results from the model examination supported H6b but not H6a.

A significant large R² of 0.51 (p < 0.001) in subthreshold depression was reported during the post-study period. This result demonstrated that 51% of the unique variance of subthreshold depression was explained by independent and mediating variables mentioned above. Generally, the findings from path analysis support the final model in which four out of the six path coefficients are statistically significant. To sum up, compared to SP exergames, MP exergames reduced subthreshold depression via the effect on loneliness. Although social support was not supported to be a significant mediator of play mode, it independently affected loneliness and subthreshold depression among older adults.