Social media use and positive aspects of caregiving: a longitudinal test of the social compensation hypothesis

Abstract

Objectives

Guided by the social compensation hypothesis, this study examined longitudinal pathways linking multidimensional social media use (SMU) to positive aspects of caregiving (PAC) among family caregivers of people with dementia. Specifically, we tested whether perceived social support (PSS) mediates the association between SMU and PAC and whether offline social network size moderates the SMU-PSS pathway.

Methods

A 3-wave time-lagged panel design was conducted over 6 months. Data were collected from a final analytic sample of 417 family caregivers in Hong Kong and Mainland China. Path analysis within a structural equation modeling framework tested a moderated mediation model in which T1 SMU predicted T3 PAC indirectly via T2 PSS, conditional on T1 offline social network size.

Results

T1 SMU significantly predicted T3 PAC (B = 0.141, p < .001), with a significant indirect effect through T2 PSS (B = 0.019, p = .027), indicating partial mediation. Offline social network size negatively moderated the SMU-PSS association (B = −0.030, p < .01). Simple slope analyses showed that the SMU-PSS relationship was significant among caregivers with smaller offline networks but nonsignificant among those with larger networks.

Discussion

Findings support a compensatory function of SMU: online engagement may enhance positive caregiving experiences partly by strengthening PSS, particularly for caregivers with limited offline social resources. These results inform the development of targeted digital support strategies tailored to caregivers’ baseline social networks.

Informal dementia caregivers worldwide face substantial physical and psychological challenges (Deeken et al., 2019). Given the complex and intensive care demands, family caregivers face sustained stress, making them vulnerable to adverse outcomes such as depression, caregiver burden, social isolation, burnout, and deteriorating health (Collins & Kishita, 2020). While research extensively documents these negative consequences (Cheng, 2017), a growing body of work also highlights the positive aspects of caregiving (PAC), perceived gains such as personal growth and enhanced self-affirmation (Lou et al., 2015; Tarlow et al., 2004). Attending to these positive dimensions offers an alternative lens emphasizing psychological rewards alongside costs (Schulz et al., 2020). Importantly, higher levels of PAC have been associated with better caregiver well-being and may serve as a psychological buffer against caregiving-related stress, thereby supporting caregivers’ capacity to sustain long-term caregiving roles (Lou et al., 2024). Unlike other forms of caregiving, dementia caregiving entails progressive cognitive decline, behavioral symptoms, and societal stigma, frequently forcing caregivers into involuntary social isolation (Won et al., 2024). In this distinctly constrained relational context, identifying accessible resources to cultivate PAC is critical.

Social media offers caregivers novel alternatives to overcome time and space constraints inherent in traditional interactions (Bui et al., 2022). These platforms facilitate emotional and informational exchange, thereby enhancing perceived social support (PSS) (Daynes-Kearney & Gallagher, 2023). Within this context, social media use (SMU) extends beyond leisure activities and functions as a coping strategy that facilitates the accumulation of social capital essential for maintaining psychological well-being (Sun et al., 2023). Nevertheless, it is important to acknowledge the potential limits and harms of digital engagement. For instance, exposure to online misinformation can misguide dementia caregiving behaviors and exacerbate caregiver distress, highlighting a complex duality in its relationship to health outcomes (Wang et al., 2019). The social compensation hypothesis provides a key theoretical framework for understanding how, despite these risks, social media can still yield targeted benefits. This perspective posits that individuals lacking sufficient offline connections, often due to temporal or spatial constraints, are more likely to turn to online interactions to compensate for these deficits (Grieve et al., 2017). Online communication environments provide greater opportunities for self-disclosure and access to social support, thereby enabling individuals with weaker offline social networks to establish socially meaningful relationships and confer particularly pronounced psychological benefits (Moukalled et al., 2021). While social media is widely recognized as a potential source of support, it remains unclear whether online use can compensate for limited offline social networks to foster PAC. Moreover, longitudinal evidence testing this compensatory mechanism is notably lacking.

Utilizing a time-lagged longitudinal design, this study examines the impact of SMU on PAC among informal dementia caregivers, mediated by PSS. Beyond traditional single-item frequency measures, which often obscure the underlying motivations for digital engagement, we employ a multidimensional assessment of SMU to capture a more holistic picture of purposeful digital engagement. Additionally, we test the moderating effect of offline social network size to empirically evaluate the social compensation hypothesis—specifically, whether online interactions provide greater benefits for caregivers with smaller offline networks.

Literature review

Social media use and positive aspects of caregiving

The caregiving experience involves burdens and potential benefits, including PAC, perceived gains such as enhanced self-worth, self-efficacy, and meaning-making (Tarlow et al., 2004). Historically, caregiving research was dominated by a burden-oriented paradigm that emphasized stress and pathology. However, a recent conceptual shift toward a strength-based perspective acknowledges the co-occurrence of distress and psychological rewards (Yu et al., 2018). Empirical studies indicate that higher PAC is associated with lower caregiver burden and fewer depressive symptoms, implying a protective role in caregiver well-being (Xue et al., 2018).

However, progressive functional decline and intensive care demands constrain dementia caregivers’ time, mobility, and offline social engagement. This increases social isolation and diminishes opportunities for positive appraisal, undermining PAC development (Kotwal et al., 2022). From a social compensation perspective, such offline social constraints may motivate caregivers to seek alternative sources of connection and support through online platforms (Newman et al., 2019). In response to these challenges, recent literature has identified SMU as an accessible coping resource that facilitates peer support beyond geographical and temporal constraints (Morini et al., 2025). Online interaction may foster cognitive reframing by normalizing caregiving experiences and enabling the exchange of experiential knowledge, thereby supporting caregivers’ perceptions of competence and meaning (Daynes-Kearney & Gallagher, 2024). In addition, emotional support and validation obtained through social media have been associated with reduced psychological distress and more positive appraisals of the caregiving role (Ke & Lou, 2024). Although prior research has primarily focused on SMU in relation to reduced negative outcomes, emerging evidence suggests that social media engagement may also enhance PAC by reinforcing caregiver identity and perceived value (Haslam et al., 2022). Drawing on these insights, the present study proposes the following hypothesis:

Hypothesis 1: SMU will be positively associated with PAC.

The mediating role of perceived social support

PSS refers to individuals’ subjective perception that support is available and that they are cared for and valued within their social relationships (Vernooij-Dassen et al., 1996). In dementia caregiving, studies suggest that SMU may help maintain or enhance PSS by extending access to supportive ties when offline networks shrink due to care demands, limited mobility, and stigma-related barriers (Yin et al., 2024). Online communities have been shown to provide peer-based emotional validation and informational support from others with similar experiences, which is associated with stronger feelings of connectedness (Marshall et al., 2024).

In turn, PSS is positively associated with PAC (Jiang et al., 2022). The Stress-Buffering Hypothesis posits that perceiving support as available can mitigate stress appraisal and reduce psychological strain (Pluut et al., 2018). Beyond buffering stress, higher perceived support is linked to greater competence and self-efficacy, which may facilitate meaning-making and satisfaction in the caregiving role (Guo et al., 2026). Theoretically, SMU is conceptualized as a resource-building process that fosters PSS, thereby supporting positive cognitive appraisal and meaning-making (Xie et al., 2022). While prior research examined these constructs individually, few tested this mediating pathway longitudinally. Thus, we propose:

Hypothesis 2: SMU will be positively associated with dementia caregivers’ PSS.

Hypothesis 3: PSS will mediate the relationship between SMU and PAC.

The moderating role of offline social network size

While social media generally offers support, the extent to which caregivers benefit from SMU may depend on their existing offline social resources (Pouwels et al., 2022). The Social Compensation Hypothesis proposes that individuals with limited offline connections may derive greater benefits from online interaction by compensating for offline constraints (Grieve et al., 2017). In contrast, the “Rich-Get-Richer” model suggests that individuals with stronger offline networks may be better positioned to benefit from digital engagement (Moshkovitz & Hayat, 2021).

In the context of dementia caregiving, where social isolation is a prevalent and often involuntary consequence of caregiving demands, the compensatory function of social media may be particularly salient. For caregivers with smaller offline social networks, online interaction may serve a more central role in accessing emotional and informational support (Dam et al., 2019). Consequently, the marginal benefit of SMU for PSS is theorized to be greater among caregivers with fewer offline connections. To examine this conditional relationship, the present study proposes:

Hypothesis 4: Offline social network size will moderate the relationship between SMU and PSS, such that the positive association will be stronger for caregivers with smaller offline social networks.

Method

Sampling and data collection

A pilot study (July 4–22, 2024) yielded 215 valid responses from caregivers in Hong Kong and Shanghai, which were excluded from the main longitudinal analysis. The main study employed a six-month longitudinal design spanning from November 1, 2024, to May 19, 2025, with data collected across three waves (T1, T2, and T3) at 2-month intervals in Hong Kong and Mainland China.

Participants included adult family caregivers of people living with dementia who had used social media within the past year. Prior to participation, all respondents were informed of the study’s purpose and provided informed consent. Surveys were distributed via Qualtrics (Hong Kong) and the Wenjuanxing platform (Mainland China). Recruitment employed a multi-channel approach: Hong Kong participants were recruited through self-enrollment and non-governmental organizations (NGOs), while Mainland participants were recruited via caregiver WeChat groups, offline NGO events, and the Wenjuanxing panel service. Respondents who completed the questionnaire and passed the review process received a cash incentive.

To ensure data integrity, five instructional manipulation checks (IMCs) were embedded in the survey. Following data cleaning, the first wave (T1) comprised 1,054 valid responses (Hong Kong: n = 452; Mainland China: n = 602). The retention sample sizes for the second (T2) and third (T3) waves were 518 and 417. Attrition was primarily attributed to the progressive and highly demanding nature of dementia caregiving, which often severely constrains caregivers’ time, energy, and capacity to commit to multi-wave surveys. Notably, the majority of sample attrition is concentrated between Wave 1 and Wave 2, reflecting a common initial attrition shock in longitudinal panels involving chronically stressed populations. Despite this anticipated attrition, our remaining sample size (N = 417) was sufficiently robust to achieve adequate statistical power for the structural equation modeling (SEM). Three cases with non-numeric social network size responses were excluded from moderation analyses.

Measures

Dependent variable: PAC (T3)

Positive caregiving experiences were assessed using the PAC scale developed by Tarlow et al. (2004). We utilized the Chinese version validated by Lou et al. (2015), which has been widely applied in research involving Chinese caregivers. The scale consists of 11 items. Items (e.g., “Made me feel useful”) were rated on a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree). The scale demonstrated excellent internal consistency (Cronbach’s α = 0.91) and sampling adequacy (KMO = 0.92). Factor analysis confirmed a one-factor structure with factor loadings ranging from 0.58 to 0.81. Factor scores were computed via the regression method for subsequent analyses, with higher scores reflecting more positive perceptions.

Independent variable: SMU (T1)

SMU was assessed using the Social Use and Cognitive Use dimensions adapted from Ali-Hassan et al. (2015), where SMU is operationalized as the purposeful engagement in digital platforms to achieve specific relational and informational goals. For dementia caregivers facing involuntary social isolation, capturing their goal‑directed behaviors more accurately reflects how they utilize digital tools for social compensation than measuring screen time. Each dimension consisted of five items, forming a 10-item scale in total (e.g., “I use social media to establish new interpersonal relationships,” “I use social media to share content with others”). All items were measured using a 4-point Likert scale ranging from 1 (strongly disagree) to 4 (strongly agree). This scale showed good reliability (Cronbach’s α = 0.89) and sampling adequacy (KMO = 0.92). Factor analysis supported a two-factor structure with oblique rotation, and factor loadings ranged from 0.45 to 0.87.

Factor scores for both dimensions were calculated using regression methods and summed into an equally weighted composite index. Higher values indicate a greater level of SMU. This aggregated variable was included as an observed variable in subsequent SEM analyses to represent the overall intensity of participants’ SMU.

Mediating variable: PSS (T2)

PSS was assessed using the Multidimensional Scale of Perceived Social Support (Zimet et al., 1988), adapted to evaluate support from both offline and online sources (Wang et al., 2021). The revised version comprised 24 items, assessing respondents’ perceived support from family, friends, and significant others across both offline and online contexts, with 12 items corresponding to each form of support. All items were rated on a 4-point Likert scale, ranging from 1 (strongly disagree) to 4 (strongly agree).

The revised scale exhibited excellent internal consistency (Cronbach’s α = 0.96) and sampling adequacy (KMO = 0.96). Factor analysis confirmed a three-factor structure with oblique rotation; all items loaded clearly onto their respective factors (loadings ranged from 0.50 to 0.80), cumulatively explaining over 70% of the variance. To reduce model complexity and to maintain consistency with the analytical approach used for other study variables, factor scores for the three dimensions were calculated separately using regression methods and then summed. The resulting composite score was included as an observed variable in subsequent analyses to represent the overall level of PSS among participants. The reliability and validity statistics for PAC, SMU, and PSS are summarized in Table 1.

Table 1.

Reliability and validity analysis statistics.

Variable type	Variable	Dimension	Items	Cronbach’s α	KMO value
Independent variable	Social media use	Total	10	0.89	0.92
		Social use	5	0.84
		Cognitive use	5	0.76
Mediator	Perceived social support	Total	24	0.96	0.96
		Dimension 1	10	0.94
		Dimension 2	8	0.92
		Dimension 3	6	0.88
Dependent variable	PAC	Total	11	0.91	0.92
		PAC	11	0.91

Note. N = 414. KMO = Kaiser-Meyer-Olkin, PAC = positive aspects of caregiving.

Moderating variable: social network size (T1)

Social network size was assessed at Wave 1. Participants were asked to quantify the number of individuals in their support network. This network can include a range of individuals who provide emotional, practical (e.g., help with transportation, household chores, shopping), medical, or financial support. Participants provided a direct numerical answer. The responses were treated as a continuous variable. To ensure data quality, the data were cleaned by removing extreme outliers and non-numeric entries prior to the analysis.

Covariates

This study included several individual-level characteristics as control variables, all measured at Wave 1. Core demographic covariates comprised the respondent’s age (in years), gender (coded as 0 = male, 1 = female), marital status (coded as 1 = unmarried, 2 = married, 3 = partnered, 4 = divorced/separated, and 5 = widowed) and weekly caregiving hours (from 1 = Less than one hour to 8 = 70 hours or more). Additionally, socioeconomic status was controlled for using education level, measured on a 7-point scale (from 1 = no formal education to 7 = Master’s degree or higher), and family monthly income, also on a 7-point scale (from 1 = less than ¥2,999 to 7 = ¥30,000 or more). Finally, self-rated health was included, assessed on a scale from 0 to 100. Categorical variables were dummy-coded.

Statistical analysis

Data analyses were performed using Stata/MP 18.0, encompassing descriptive statistics, correlation analysis, and path analysis. A three-wave time-lagged panel mediation model was employed to investigate the longitudinal associations among SMU, PSS, and PAC. This design is fundamental for improving causal inference over cross-sectional approaches by establishing a clear temporal sequence among the core variables (Cook et al., 2002).

The model was specified to strictly follow this temporal ordering, testing the primary hypothesized path: SMU (T1)- PSS (T2)-PAC (T3). In this structure, the independent variable (measured at Wave 1) precedes the mediator (measured at Wave 2), which in turn precedes the dependent variable (measured at Wave 3). This temporal separation of measurements minimizes common method bias and fulfills a critical prerequisite for drawing causal-like conclusions.

To test for moderated mediation, an interaction term between social media usage and social network size (both measured at T1) was created and included as a predictor of the mediator, PSS, at T2. While more complex panel models, such as the CLPM, examine reciprocal dynamics over time, our focused, theory-driven approach is specifically designed to test a directional X-M-Y causal chain. As argued by scholars such as Cole and Maxwell (2003), this three-wave time-lagged design serves as a standard and valuable paradigm for longitudinal mediation analysis, providing stronger evidence than a study where all variables are measured concurrently. Considering the characteristics of the dataset, a complete case analysis was employed. Baseline variables measured at T1 that were significantly associated with sample attrition were further incorporated as covariates in all models to address potential attrition bias.

To create composite scores for the multi-dimensional constructs while addressing the limitations of simple summation, a two-step procedure was used (Anderson & Gerbing, 1988). First, exploratory factor analysis with principal axis factoring and Promax rotation was conducted for each scale (e.g., SMU, PSS) to confirm their dimensional structure and calculate factor scores for each dimension using the regression method. Second, for each construct, these factor scores were summed to create a single composite index (e.g., an overall SMU score). This hybrid composite scoring approach was deliberately chosen over a full latent structural model to maintain an optimal parameter-to-sample-size ratio, given the large number of original items (45 items across three focal variables), thereby preventing non-convergence and unstable parameter estimates (Bandalos, 2002). Prior to computing these composite scores, a confirmatory factor analysis (CFA) including all core variables was evaluated. The measurement model demonstrated satisfactory overall fit (χ²/df = 2.35, CFI >0.90, RMSEA = 0.05, and SRMR = 0.05). Furthermore, to avoid underestimating the relationships among variables, we constructed a full‑latent SEM using the item‑to‑construct balance parceling technique as a robustness check. The model exhibited excellent fit (χ²/df = 0.997, CFI = 1.000, TLI = 1.000, RMSEA < 0.001, SRMR = 0.017). All structural paths remained significant and highly consistent with those derived from the primary composite‑based analysis. Detailed parameter estimates for this full-latent model are provided in Supplementary Table 1.

These composite indices were then used as observed variables in a path analysis to test the time-lagged mediation model [X(T1)-M(T2)-Y(T3)]. Model parameters were estimated using the maximum likelihood (ML) method. The significance of the indirect effect was tested using bias-corrected bootstrapping with 5,000 resamples.

Results

Participant characteristics and attrition analysis

A total of 1054 participants were recruited at the baseline survey (Time 1). After three waves of data collection, excluding invalid questionnaires and cases with failed longitudinal matching, the final sample for model analysis consisted of 417 participants, representing a retention rate of 39.56%.

The baseline comparison between retained and attrited samples is presented in Table 2.

Table 2.

Comparison of baseline characteristics between retained and attrited samples.

Variables	Total sample (N = 1,054)	Retained group (n = 417)	Attrited group (n = 637)	Test (t/χ2)	p
Demographic characteristics
Age (years), M (SD)	47.78 (16.34)	51.13 (13.87)	45.58 (17.43)	−5.47	<.001
Gender	—	—	—	0.28	.597
Marital status	—	—	—	14.13	.007
Education level	—	—	—	74.96	<.001
Income	—	—	—	11.56	.072
Self-rated health, M (SD)	72.71 (18.95)	69.28 (17.84)	74.95 (19.34)	4.80	<.001

Note. M = mean, SD = standard deviation.

To examine the potential for systematic bias from sample attrition, an attrition analysis was conducted using baseline (Time 1) data. The sample was divided into a “Retained Group” (completed all three survey waves, n = 417) and a “Attrited Group” (withdrew at Time 2 or Time 3, n = 637). Independent sample t-tests were used for group comparisons on continuous variables, while chi-square tests were used for categorical variables.

The results indicated no statistically significant differences between the two groups in terms of gender (${\mathrm{\chi }}^2$ = 0.28, p = .597) and income level (${\mathrm{\chi }}^2$ = 11.56, p = .072), suggesting a relatively balanced distribution of these characteristics. However, significant differences were observed in other key variables. Demographically, the retained group was significantly older than the attrited group (t = −5.47, p < .001), and significant distributional differences were found in education levels (${\mathrm{\chi }}^2$ = 74.96, p < .001), marital status (${\mathrm{\chi }}^2$ = 14.13, p = .007), and self-rated health (t = 4.80, p < .001). Given that these disparities suggest a non-random pattern of attrition, we included age, education level, marital status, and self-rated health as covariates in all subsequent multivariate statistical models to minimize the potential impact of attrition bias on the study’s internal validity.

Descriptive analysis

Table 3 presents the descriptive statistics for the demographic characteristics and key study variables of the final analytic sample (N = 417). Regarding the covariates, the participants had an average age of 51.10 years (SD = 13.87) and reported a mean self-rated health score of 69.28 (SD = 17.84). The descriptive data for other demographic variables, including gender, education, marital status, and income, are also displayed in the table. Turning to the focal variables of the study, participants reported an average score of 25.00 (SD = 6.79) for SMU. In terms of social resources, the mean score for PSS was 66.65 (SD = 15.69), while the average social network size was 3.08 (SD = 6.19). Regarding the outcome variable, participants reported a mean level of 39.95 (SD = 8.22) for PAC. Overall, the range and standard deviations indicate sufficient variability across all key measures for subsequent path analysis.

Table 3.

Descriptive statistics of the research variables.

Variable	Observations	Mean	SD	Min	Max
Positive aspects of caregiving	417	39.95	8.22	12	55
Social media use	417	25.00	6.79	10	40
Perceived social support	417	66.65	15.69	24	96
Social network size	414	3.08	6.19	0	100
Covariates
Age	417	51.10	13.87	20	82
Gender	417
Male	120 (28.78%)
Female	297 (71.22%)
Education	417	4.38	1.57	2	7
Marriage	417
Single	129 (30.94%)
Married	265 (63.55%)
Partner	3 (0.72%)
Divorced/separated	15 (3.60%)
Widowed	5 (1.20%)
Income	417	3.16	1.61	1	6
Weekly hours of caregiving	417	4.96	2.06	1	8
Self-rated health	417	69.28	17.84	3	100

Note. SD = standard deviation.

Longitudinal path analysis and hypothesis testing

To examine the hypothesized longitudinal moderated mediation model, a path analysis was conducted using Stata. The model specified T1 SMU as the independent variable, T2 PSS as the mediator, and T3 PAC as the dependent variable. Additionally, T1 social network size was modeled as a moderator of the first-stage relationship (SMU-PSS). Control variables including gender, age, education level, and other covariates were included in the estimation.

The model fit was evaluated using standard goodness-of-fit indices (Hu & Bentler, 1999). The results indicated an excellent fit to the data: ${\mathrm{\chi }}^2$ (2) = 0.58, p = .750; RMSEA = 0.000 (90% CI [0.000, 0.067]); CFI = 1.000; TLI = 1.176; SRMR = 0.003. The nonsignificant chi-square value and perfect CFI/RMSEA scores suggest that the hypothesized theoretical model fits the observed data structure exceptionally well.

The standardized and unstandardized path coefficients are presented in Table 4.

Table 4.

Results of longitudinal path analysis.

Predictor	Dependent variable	b	SE	β	z-value
First stage: mediator model	Perceived social support
SMU		0.256	0.049	0.281	5.24***
SNS		0.009	0.011	0.036	0.78
Interaction (SMU $\times$ SNS)		−0.030	0.010	−0.135	−2.97**
Controls
Self-rated health		0.010	0.004	0.115	2.48*
Female (Other controls included)		0.193	0.158	0.057	1.22
Second stage: outcome model	Positive aspects of caregiving
Perceived social support		0.075	0.031	0.120	2.46*
T1 SMU (Direct)		0.141	0.032	0.247	4.43***
Controls
Self-rated health		0.009	0.003	0.159	3.38***
Female (Other controls included)		0.022	0.099	0.010	0.22

Note. N = 414. b = unstandardized coefficients, $\mathrm{\beta }$ = standardized coefficients, SMU = social media use, SNS = social network size. Control variables (age, education, etc.) were included in the model but omitted from the table for brevity.

^*** p < .001.

^** p < .01.

^* p < .05.

First, regarding the main causal chain, SMU significantly and positively predicted PSS (p < .001). Subsequently, PSS significantly and positively predicted PAC (p = .014). Additionally, a significant direct effect of SMU on PAC was observed (p < .001), indicating a partial mediation model. Specifically, SMU exerted a significant indirect effect on PAC through PSS (p = .027). Calculation of the mediation proportion showed that this indirect pathway accounted for 11.9% of the total effect.

The interaction effect between SMU and social network size on PSS was statistically significant (p = .002). To interpret this interaction, a simple slope analysis was performed. The positive association between SMU and PSS was significant among individuals with low levels of social network size (p < .001). However, for individuals with high levels of social network size, this relationship was nonsignificant (p = .462), as shown in Figure 1.

Figure 1.

Interaction effect on social network size. Slopes are significant at low levels but non-significant at high levels.

Finally, the conditional indirect effects of SMU on PAC via PSS were examined. The indirect effect was statistically significant for individuals with low levels of social network size (p = .024, 95% CI [0.004, 0.063]). In contrast, the indirect effect was not significant for those with high levels of social network size (p = .481, 95% CI [−0.008, 0.017]). The index of moderated mediation was marginally significant (p = .056), suggesting that the mediation mechanism is conditional upon the level of social network size. Additionally, a sensitivity analysis using full information maximum likelihood (FIML) to account for all baseline participants (N = 1,054) yielded consistent structural paths and significance levels, confirming the robustness of our complete-case estimates to sample attrition. Complete FIML estimation results are available in Supplementary Table 2.

Discussion

This study employed a three-wave longitudinal design to elucidate the mechanisms and boundary conditions linking SMU to PAC among informal caregivers of people with dementia. The results revealed that initial SMU indirectly promoted future PAC by enhancing caregivers’ PSS. This finding aligns with a growing body of evidence suggesting that SMU can bolster PSS, promote social participation, and reduce social isolation among informal dementia caregivers, all of which contribute to an improved caregiving experience (Dam et al., 2017). Our findings indicate that this beneficial process is contingent on caregivers’ offline social resources. Specifically, the translation of SMU into positive outcomes was more pronounced among caregivers with smaller offline social networks compared to those with larger networks. This pattern may reflect the unique stress-coping processes and resilience-generation mechanisms inherent to informal dementia caregiving, a dynamic that warrants further verification in future research.

The results confirmed that SMU serves as a significant distal predictor of PAC, with PSS playing a partial mediating role. Through Conservation of Resources theory, individuals strive to obtain and protect resources to prevent stress (Hobfoll, 1989). Given the high-burden nature of dementia care, caregivers living in a chronically stressful environment tend to utilize social media as a coping strategy for resource acquisition (Ferrero-Sereno et al., 2025). PSS acts as a functional “mediating resource” obtained through this strategy. When caregivers perceive higher levels of support, they report greater confidence in their caregiving abilities and lower perceived burden, which in turn enhances self-efficacy and a sense of gain, ultimately fostering PAC (Nemcikova et al., 2023). The observation of partial mediation (11.9%) suggests that SMU may also enhance PAC through other pathways, such as providing a space for temporary mental disengagement, allowing caregivers to obtain “psychological respite” from the intense caregiving environment.

Another key contribution of this study lies in clarifying a theoretically meaningful boundary condition, namely the buffering role of offline social network size, which challenges the traditional “Rich-get-Richer” hypothesis (Kraut et al., 2002). We found that SMU significantly boosted PSS for caregivers with smaller offline social networks but had no significant effect for those with larger networks. This can be explained by the Social Compensation Hypothesis. For dementia caregivers with a paucity of offline networks who face a high risk of social isolation, social media acts as a critical medium that connects them to the outside world (Liao et al., 2024). It transcends spatiotemporal barriers, providing an accessible channel for social connection that compensates for the lack of face-to-face interaction (Wellman, 2001). In contrast, for caregivers with extensive offline social networks (e.g., strong family support or stable community ties), their need for support is likely already met through face-to-face interactions. For these “resource-rich” caregivers, the additional support derived from social media may yield diminishing marginal returns, a phenomenon consistent with prior empirical studies (Pouwels et al., 2021, 2022).

This study enriches the caregiving literature by clarifying how digital engagement operates differently within the distinct relational context of dementia caregiving. While digital engagement in general populations often yields a “Rich-get-Richer” effect, the progressive and stigmatizing nature of dementia caregiving frequently leads to involuntary social isolation. In such highly constrained contexts, the results extend the boundary conditions of the Social Compensation Hypothesis by revealing that digital platforms function as a critical compensatory lifeline rather than merely an additive resource. Furthermore, there have been prior inconsistencies regarding social media’s impact on caregivers (Lumini et al., 2025). Our findings suggest that such inconsistencies may partly reflect the neglect of baseline offline resources and longitudinal compensatory dynamics. By demonstrating that the positive benefits of social media are most prominent among the socially isolated, this study provides a more nuanced perspective on the role of digital technology within the Stress Process Model of dementia caregiving.

In the context of rapid digital advancement, these findings offer tentative implications for support strategies targeting informal dementia caregivers. First, from a mechanistic perspective, the results suggest that SMU may not automatically translate into PAC; its potential benefits appear to depend partly on whether it successfully fosters a perception of social support. Second, from a practice perspective, digital support programs could benefit from a targeted rather than a “one-size-fits-all” approach (Zhai et al., 2023). Healthcare professionals and social workers might consider assessing dementia caregivers’ baseline social network size. For informal caregivers with small social networks, facilitating digital literacy and encouraging participation in online communities could serve as a potential strategy to compensate for offline resource deficits. Conversely, for caregivers with robust offline networks, simply encouraging more SMU might yield limited additional benefits; instead, leveraging their existing family and community ties may be more effective in enhancing their positive caregiving experiences.

Several limitations of the present study should be acknowledged. First, regarding the measurement strategy, this study utilized aggregated scores derived from factor analysis as manifest variables in the model. While this approach effectively simplified the model structure, it may have obscured the nuances and internal complexity of specific sub-dimensions. Second, although the time-lagged panel design offers a significant advantage in establishing temporal precedence, the potential for causal inference remains constrained by the observational nature of the data. Consequently, residual bias from unmeasured time-varying confounding factors cannot be entirely ruled out. Third, the reliance on self-reported measures introduces potential susceptibility to common method bias and recall bias. Despite our rigorous quality control procedures, these inherent limitations cannot be completely eliminated. Fourth, a relatively high attrition rate was observed across the three waves of data collection. Although participant dropout is a recognized challenge in longitudinal studies involving high-stress populations, this attrition raises concerns regarding systematic attrition bias. Specifically, the compensatory effects of social media observed here may be particularly reflective of older, physically vulnerable caregivers. Therefore, caution should be exercised when generalizing these findings to the younger, healthier populations, or most overwhelmed and vulnerable segments of the dementia caregiving population, as well as to caregivers in different cultural contexts or those managing other chronic conditions where digital norms may differ. Furthermore, while our measure captures purposeful social media engagement, it does not distinguish objective usage intensity, specific platform types, or other nuanced purposes of use. Finally, our study did not directly measure the role of societal stigma, which may act as a crucial underlying driver pushing caregivers toward online rather than offline support. To address these limitations, future research could benefit from granular tracking and content analysis to elucidate how different platforms, specific digital content, and usage intensities distinctively impact caregiving experiences. Moreover, there is a pressing need for focused research on multiply marginalized subpopulations, such as urban–rural disparities, paying particular attention to intersecting vulnerabilities and the interplay between caregiving stigma and offline network constraints. Such endeavors will provide the necessary theoretical and empirical grounding for developing tailored empowerment strategies for these groups.

Conclusion

This study suggests that SMU significantly enhances PAC among informal dementia caregivers by bolstering PSS, a relationship that tends to be more pronounced among individuals with smaller social network sizes. These findings provide empirical grounds for support strategies targeting informal caregivers and underscore the critical importance of incorporating social network size into intervention assessments. Against the backdrop of rapid digitalization and social media proliferation, fostering online support for socially isolated caregivers represents a key strategy. Specifically, practical interventions could adopt a blended approach: informal, peer-based grassroots communities can be leveraged to provide essential emotional validation, while formalized, medicalized digital programs should be integrated to ensure access to reliable health information. Ultimately, incorporating caregivers’ baseline social network sizes into intervention assessments is critical for effectively improving their psychological well-being in high-stress environments.

Supplementary material

Supplementary material is available at The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences online.

Funding

This work was supported by the National Natural Science Foundation of China (grant no. 72174118) and the Hong Kong University Grants Committee General Research Fund (GRF) (grant no. 17603923).

Conflicts of interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

The data generated and analyzed during the current study are available from the corresponding author on reasonable request. This study was not preregistered.

Author contributions

Mingyang Yang (Data curation, Formal analysis, Methodology, Visualization, Writing—original draft, Writing—review & editing), Fan Yang (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Supervision, Writing—review & editing), Shaowei Liu (Data curation, Investigation, Methodology, Writing—review & editing), Junyao Zheng (Data curation, Validation, Writing—review & editing), Wenjie Li (Data curation, Validation, Writing—review & editing), Yiyan Zhou (Data curation, Writing—review & editing), Shihan Lu (Data curation, Writing—review & editing), Sicheng Zhong (Data curation, Writing—review & editing), Xiaobo Ke (Data curation, Writing—review & editing), and Vivian W.Q. Lou (Funding acquisition, Supervision, Writing—review & editing)