Beyond Information: A Dual-Path Strategy for Sustainable Digital-Cultural-Heritage Management Driven by Affective Experience

Abstract

Digital cultural-heritage (DCH) platforms are integral to achieving UN SDG Target 11.4, yet their long-term sustainability is compromised by a systemic vulnerability: the rapid decay of user engagement once the initial “novelty effect” fades. To address the theoretical anomaly of the “null effect” regarding value perception found in prior studies, this paper develops a competitive dual-path model to determine whether information-centric or experience-centric strategies effectively foster sustainable continuance intention. Drawing on the stimulus–organism–response (S-O-R) framework, interactivity is modelled as a high-order managerial investment. A quantitative survey of 407 DCH users was analysed using covariance-based structural equation modelling (CB-SEM). The results resolve the strategic dilemma: while interactivity enhances both pathways, a chi-square difference test $\left(\Delta {\chi }^2\left(1\right)=26.207\right.$, p < 0.001) confirms that affective value exerts a significantly stronger impact on cultural identity than epistemic value, supporting the affective primacy hypothesis. Crucially, cultural identity serves as the essential mediator that translates user experience into “emotional stickiness.” By demonstrating that narrative-driven affective engagement is superior to mere information dissemination, this study provides a validated blueprint for virtual–real symbiosis. The findings offer actionable guidance for managers to build digital resilience and safeguard heritage transmission across generations.

1. Introduction

1.1. Cultural Sustainability in the Digital Era

Cultural sustainability—defined as the preservation, interpretation, and transmission of heritage across generations—now faces unprecedented challenges and opportunities in a rapidly evolving digital landscape. The deepening of the “digital turn” has catalysed the transformation of digital cultural-heritage (DCH) platforms from static archives into dynamic “cultural-experience ecosystems” [1]. In parallel, technological advancements—particularly high-fidelity 3D digital twins and metaverse exhibitions—have fundamentally reshaped how the public accesses and interacts with heritage, positioning DCH platforms as central hubs for destination image formation, cultural education, and engagement [2,3]. This digital evolution unfolds against a backdrop of growing threats to physical heritage sites, which are increasingly vulnerable to coupled natural–social impacts such as climate change and overtourism [4]. These pressures directly challenge the mandate of the United Nations Sustainable Development Goal (SDG) 11.4, which calls for strengthened efforts to protect and safeguard the world’s cultural heritage.

Despite the significant potential of DCH, the long-term viability of many digital initiatives remains precarious. From a sustainability perspective, digital twins serve a dual function: they act as both crucial cultural backups and as “ecological pressure valves”, providing alternatives to physical visitation and thereby aligning with broader environmental mandates [1]. However, DCH platforms frequently struggle to maintain user engagement beyond initial interactions. Extensive research indicates a recurring pattern whereby users exhibit strong initial interest in interactive digital environments, but this engagement declines sharply once the initial novelty effect dissipates [5,6]. This pattern signals a systemic vulnerability and reflects a failure in sustainable resource management, whereby substantial upfront investments do not translate into enduring cultural impacts or economic returns. Against this backdrop, a central challenge emerges for DCH managers: how to convert short-lived, novelty-driven use into sustained engagement that genuinely supports cultural sustainability.

1.2. The Challenge of Sustainable Engagement

In this study, we conceptualise users’ continuance intention [7] not merely as user retention but as an indicator of a DCH project’s long-term viability and its capacity to support cultural transmission [8]. Without sustained use, digital platforms cannot realise their potential to support cultural sustainability. Sustainable heritage management therefore requires a holistic approach that bridges the traditional divide between digital and physical experiences. Rather than treating digital engagement as a substitute for physical visitation, we propose that it can function as a psychological primer (Figure 1) through a mechanism we term virtual–real symbiosis [9].

Building on the cognitive–psychological literature on priming, we conceptualise this primer not as a passive information portal but as a dynamic pre-activation mechanism. Triggered by immersive digital interactions, it builds an initial mental schema of the heritage site. Consistent with the cognitive load theory, this process acts as a “cognitive pressure valve”, reducing wayfinding anxiety and information overload during subsequent on-site visits. By reallocating limited cognitive resources in advance, the primer enables stronger emotional resonance and aesthetic engagement and may help sustain the “anticipation–verification” cycle that supports the resilience of the heritage ecosystem.

In this framework, digital platforms act as a psychological primer (Stage 1) that reduces cognitive load. This, in turn, enables managed visitation (Stage 2), which serves as an ecological pressure valve for on-site heritage environments (Stage 3) and supports the resilience required to advance SDG 11.4 (Stage 4).

1.3. Strategic Dilemmas and Research Gaps

Heritage managers face a fundamental strategic dilemma when allocating resources for DCH development: should they prioritise enhancing cognitive processing (i.e., rational evaluation and information accuracy) or cultivating affective resonance (i.e., emotional immersion and sensory arousal)? While the established stimulus–organism–response (S-O-R) paradigm suggests the coexistence and potential interplay of both cognitive and affective pathways [10], empirical evidence regarding their relative importance in driving sustainable engagement remains fragmented and often contradictory. This lack of clarity complicates the design of effective long-term strategies.

Of particular concern is a theoretical anomaly reported by recent studies, such as Deng et al. [11], which find that perceived value exerts only a weak or non-significant effect on cultural identity—a phenomenon we term the “null effect.” From a sustainability perspective, this missing link is problematic because cultural identity forms the bedrock of heritage resilience and long-term advocacy. We argue that this anomaly likely stems from a methodological limitation common in prior research: conceptualising “value” as a unidimensional construct. Such an approach obscures the subtle yet potentially competitive interplay between distinct cognitive and affective mechanisms [12].

To address this theoretical inconsistency and interrogate the reported null effect, we decompose perceived value into its constituent dimensions within a structural model. Specifically, we ask whether, in sustainability-oriented digital cultural-heritage (DCH) management, affective value exerts a stronger influence on cultural identity than epistemic value, thereby constituting the dominant pathway for sustaining engagement. We propose that the null effect reported in prior studies may be an artefact of construct aggregation—a phenomenon we term cognitive dilution. When perceived value is operationalised as a single dimension, utilitarian components (e.g., efficiency and ease of use) can dominate the variance. Yet, utilitarian value may be insufficient to trigger the deep, self-referential processing implicated in identity formation. By failing to separate affective (“hot”) and cognitive (“cold”) pathways, prior models may have attenuated the estimated effect of emotion, increasing the risk of a Type II error in explaining cultural identity. We draw on the elaboration likelihood model (ELM) to separate these pathways and re-examine the link between value perceptions and cultural identity.

The remainder of the paper is structured as follows: Section 2 develops the theoretical framework; Section 3 outlines the methodology; Section 4 reports the empirical results; and Section 5 discusses the shift towards soul-centric management and its implications for advancing SDG 11.4.

We postulate that the “null effect” observed in prior studies is likely an artefact of construct aggregation, a phenomenon we term “Cognitive Dilution”. When scholars measure “Perceived Value” as a unidimensional construct, the utilitarian components (efficiency, ease of use) often dominate the variance. However, utilitarian value rarely triggers the deep, self-referential processing required for identity formation. By failing to disentangle the “hot” (affective) pathway from the “cold” (cognitive) pathway, previous models statistically diluted the impact of emotion, leading to a Type II error regarding the drivers of cultural identity. Our study employs the ELM framework to structurally separate these pathways, thereby recovering the missing link.

2. Literature Review and Hypothesis Development

2.1. The Stimulus: Interactivity in DCH Design

In DCH design, interactivity distinguishes dynamic digital environments from static displays [13]. To maintain theoretical clarity and avoid conflating external stimuli with internal experiential states (e.g., flow), we define interactivity as a set of design-controllable environmental cues. Within the S-O-R framework, interactivity is conceptualised as a managerial investment variable—an exogenous stimulus (S) that can be strategically adjusted to shape user responses [10].

Drawing on recent work on the “social heritage metaverse” [14], we model interactivity as a reflective second-order construct. To assess its measurement properties, we apply the criteria proposed by Jarvis et al. [15]. This specification follows the premise that interactivity in digital cultural-heritage (DCH) contexts reflects a holistic user-agency empowerment strategy. Its three dimensions—technological interaction (TI), social interaction (SI), and personalisation (PERS)—are not independent causal drivers; instead, they manifest the underlying strategy. Empirically, strong covariation among the first-order dimensions, together with standardised loadings exceeding 0.70 (p < 0.001), supports the reflective specification (

Table 1. Evaluation of the reflective measurement model: Factor loadings and reliability statistics.

Discriminant Criteria	Reflective Logic	Formative Logic	Application in This Study
Causal Direction	Construct → indicators (the latent construct gives rise to its indicators).	Indicators → construct (indicators compose and define the construct).	Reflective supported. Interactivity is conceptualised as a latent user-agency empowerment trait operationalised as a management strategy, which manifests through technological interaction (TI), social interaction (SI), and personalisation (PERS), rather than as an additive bundle of discrete functional features.
Indicator Interchangeability	Indicators are expected to share a common conceptual theme and be conceptually interchangeable under the same underlying meaning.	Indicators capture non-overlapping facets and are therefore not substitutable.	Reflective supported. TI, SI, and PERS share a common user-agency empowerment theme and represent alternative expressions of the same underlying empowerment logic, indicating substantial conceptual overlap among the first-order dimensions.
Indicator Covariation	Substantial interindicator correlations are expected.	Correlations among indicators are not required.	Reflective supported. CB-SEM-based confirmatory factor analysis (CFA) indicates strong covariation among TI, SI, and PERS; standardised second-order loadings all exceed 0.70 (p < 0.001), demonstrating high internal consistency of the reflective specification.
Antecedents and Consequences	Indicators are assumed to share common antecedents and consequences.	Indicators may exhibit distinct antecedents and consequences.	Reflective supported. Empirically, TI, SI, and PERS exhibit parallel structural-path patterns to both hedonic interface engagement (HIE) and cultural identity (CID) in the CB-SEM structural model, supporting the assumption of shared theoretical antecedents and aligned downstream effects as manifestations of a single latent empowerment trait.

).

Technological interactivity (TI) captures users’ navigational control and the system’s functional responsiveness. In complex digital-twin environments, effective TI reduces operational friction and provides a foundation of agency for meaningful exploration [16,17]. Social interactivity (SI) denotes the capacity for reciprocal communication among users and between users and institutions. As noted by Chiu et al. [18], community citizenship behaviours can foster affective connectedness, transforming solitary browsing into a communal experience and enabling deeper metasocial engagement. Personalisation (PERS) refers to the system’s ability to tailor content and presentation to individual user profiles. Functioning as a “silent dialogue”, personalisation can enhance perceived warmth and trust [19], increasing the relevance and impact of the virtual museum experience [20].

We propose that this multidimensional interactive environment activates two value-processing pathways—epistemic and affective—through which users evaluate DCH experiences. Accordingly, we hypothesise the following:

H1a. 

Interactivity positively predicts epistemic value (EV).

H1b. 

Interactivity positively predicts affective value (AV).

2.2. The Organism: Cognitive vs. Affective Value

To open the “black box” of value formation at the organism (O) stage and address inconsistencies in prior work—particularly the null effect of aggregate value on identity reported by Deng et al. [11]—we draw on the elaboration likelihood model (ELM) [21] to distinguish two information-processing routes. To minimise overlap with perceived usefulness (a utilitarian TAM construct), we draw on Sheth et al.’s (1991) consumption value theory [22] and define cognitive value specifically as epistemic value. Here, epistemic value refers to the platform’s capacity to satisfy curiosity and novelty-seeking, rather than to facilitate task efficiency [23]. This distinction matters because a system may be easy to use (high utilitarian value; cf. Zeithaml [24]) while still failing to elicit deeper cultural understanding. This refinement helps address the cognitive-dilution problem reported in prior work, in which epistemic value is conflated with utilitarian usefulness. By contrast, affective value (AV) aligns more closely with the ELM peripheral route and with the experience–economy framework [25]. This pathway emphasises emotional immersion and sensory arousal over purely utilitarian considerations. Recent work in VR settings suggests that high-fidelity visuals and perceived authenticity can trigger emotional responses with limited cognitive elaboration [26].

2.3. Cultural Identity and Affective Primacy

Cultural identity (CI) refers to the profound sense of belonging, attachment, and pride associated with a specific cultural group [27]. From a sustainability perspective, CI is a critical mediator because it provides the intrinsic motivational foundation for enduring support of heritage initiatives, transforming passive consumption into active advocacy. While both cognitive and affective value dimensions are theorised to contribute to identity formation, recent empirical evidence suggests that they do not operate symmetrically.

Huang and Mohamad [28] identified a “null effect” in which narrative involvement failed to trigger parasocial interactions, echoing the findings of Deng et al. [11] in low-arousal contexts. These anomalies highlight a critical limitation of purely cognitive approaches: in the absence of deep emotional connection, unidirectional cognitive narratives—no matter how elaborate—may fail to build the relational bonds necessary for strong cultural identification.

Building on Zajonc’s [29] affective primacy hypothesis, we argue that, in immersive DCH contexts, the immediacy of sensory immersion often eclipses deliberate cognitive analysis. Users may feel resonance with a cultural symbol or narrative before they fully think through its historical details or utilitarian benefits. This suggests that affective pathways are more direct and impactful in fostering cultural identity within highly interactive digital environments. Thus, we propose not only that both pathways are active but that the affective route is dominant:

H2a. 

Epistemic value has a positive effect on cultural identity.

H2b. 

Affective value has a positive effect on cultural identity.

H2c (Affective Primacy Hypothesis).

The effect of affective value on cultural identity is significantly stronger than that of epistemic value.

2.4. The Response: Behavioural Outcomes

At the downstream end of the S-O-R chain, we focus on how identity-aligned experiences translate into satisfaction and continuance. In cultural consumption contexts, self-congruity (identity) typically precedes gratification (satisfaction): when users perceive that a DCH experience affirms “who they are” or resonates with their cultural background, self-verification processes are activated, deepening overall satisfaction [30]. This satisfaction, in turn, drives intentions to continue using the platform.

Furthermore, Li et al. [8] argue that satisfaction derived from culturally congruent DCH experiences fosters “emotional stickiness.” This stickiness acts as a psychological resilience mechanism against the rapid decay of the “novelty effect” frequently observed in AR/VR applications [5,6], thereby supporting long-term viability. Accordingly, we hypothesise the following:

H3. 

Cultural identity has a positive effect on satisfaction (SAT).

H4. 

Satisfaction has a positive effect on continuance intention (CONT).

2.5. Theoretical Clarification: Comparing Indirect Pathways (H5)

To further validate the dominance of the affective route and provide a statistical basis for resource allocation, we explicitly compare the strength of the two indirect mediation chains from interactivity to continuance intention. This comparison aims to assess whether the “affective chain” functions not merely as an alternative but as the primary driver of sustainability in DCH management (Figure 2).

H5. 

The specific indirect effect of the affective chain (interactivity → affective value → CI → SAT → CONT) is significantly stronger than that of the cognitive chain (interactivity → epistemic value → CI → SAT → CONT).

3. Materials and Methods

3.1. Measurement and Scale Development

To ensure construct validity, all measurement items were adapted from established scales and carefully tailored to the DCH context. To minimise translation-induced bias, scale development followed Brislin’s [31] rigorous back-translation procedure.

Interactivity (S): Modelled as a second-order construct using items adapted from Liu and Shrum [16] for technological interactivity (TI), McMillan and Hwang [17] for social interactivity (SI), and Komiak and Benbasat [19] together with Sepe et al. [12] for personalisation (PERS).

Perceived Value and Identity (O): Operationalised as follows: epistemic value (EV) was adapted from Zeithaml [24], affective value (AV) from Pine and Gilmore [25], and cultural identity (CI) from Deng et al. [11] and Breakwell [32].

Response Variables (R): Measured using established scales: satisfaction (SAT) was adapted from Oliver [30], and continuance intention (CONT) from Bhattacherjee [7].

All items were measured using five-point Likert scales, with wording adjusted to reflect specific DCH platform usage scenarios; see Appendix A for an overview of the constructs and their item content.

3.2. Data Collection and Sample Profile

We administered a cross-sectional survey to adults (aged ≥18 years) who had visited at least one DCH platform in the past six months. Data were collected between March and May 2025 via online DCH communities and social-media channels. Of the 530 questionnaires returned, 407 responses were retained after data screening, yielding an effective response rate of $76.8\%$. Most respondents were based in China (79.4%), situating the analysis within a specific cultural context.

The final sample size (N = 407) exceeds commonly recommended lower bounds for complex covariance-based structural-equation modelling (CB-SEM). Consistent with Wolf et al. [33] and Westland [34], this sample size is sufficient to provide adequate statistical power and solution propriety. The demographic profile reflected typical users of immersive DCH platforms: $69.1\%$ of respondents were aged 19–34 years, consistent with younger cohorts that most actively engage with digital heritage experiences (see

Table A2. Cultural-interest profiles (Q8).

Category		Responses Percentage of Responses (%)	Percentage of Cases (%)
	N
Chinese culture	323	28.6%	79.4%
Southeast-Asian culture	184	16.3%	45.2%
East-Asian culture	143	12.7%	35.1%
Western culture	253	22.4%	62.2%
Middle-Eastern culture	104	9.2%	25.6%
African culture	27	2.4%	6.6%
Latin-American culture	35	3.1%	8.6%
Other	60	5.3%	14.7%
Total	1129	100.0%	277.4%

Note: Multiple responses were allowed; therefore, the percentage of cases exceeds 100%.

for cultural-interest profiles). As summarised in

Table 2. Demographic characteristics of respondents (N = 407).

Item	Category	n	Percentage (%)
Gender	Male	174	42.8
	Female	233	57.2
Age	Under 18	8	2
	19–24 years	76	18.7
	25–34 years	205	50.4
	35–44 years	95	23.3
	45 years and above	23	5.7
Education	High school or below	70	17.2
	Currently in college	223	54.8
	Bachelor’s degree	109	26.8
	Postgraduate or above	5	1.2
Device used	VR headset	259	63.6
	Desktop computer	27	6.6
	Smartphone or tablet	121	29.7
Visits	1 time	46	11.3
	2–3 times	153	37.6
	4–6 times	150	36.9
	7 times or more	58	14.3

, the observed distributions align with the user base commonly reported for immersive DCH platforms.

Although the sample skews towards younger cohorts (19–34 years; $69.1\%$), this pattern likely reflects the digital vanguard of heritage consumption. From the perspective of cultural sustainability (SDG 11.4), understanding engagement mechanisms in younger cohorts is important, as they are likely to be primary agents of intergenerational transmission of digital cultural memory over the coming decades.

3.3. Data-Analysis Strategy

Data were analysed in IBM SPSS 26.0 and AMOS 26.0 using the two-step approach recommended by Anderson and Gerbing [35]. We used covariance-based structural equation modelling (CB-SEM), rather than PLS-SEM, because our primary aim was theory testing rather than prediction [36]. Specifically, testing the affective primacy hypothesis (H2c) requires nested model comparisons using a chi-square difference test ($\Delta {\chi }^2$), which is naturally supported within CB-SEM and can be evaluated alongside global fit indices. In addition, the sample size (N = 407) is sufficient for maximum likelihood estimation (

Table 3. Criteria for selecting CB-SEM over PLS-SEM: A comparative assessment (adapted from Hair et al. [36]).

Evaluation (Criteria)	CB-SEM (AMOS)	PLS-SEM (SmartPLS)	Alignment in This Study (Verdict)
Research Goal	Theory verification, confirmatory modelling, and model comparison using global fit criteria.	Prediction-oriented modelling and variance explanation.	Aligned. The study aims to verify a theory-driven dual-path structural model, including a reflective second-order construct and nested $\Delta {\chi }^2$ model comparisons, which require CB-SEM.
Sample Size and Distribution	Requires adequate sample size and approximate multivariate normality for ML estimation.	Less restrictive distributional assumptions; suitable for small to moderate samples.	Aligned. N = 407; Mardia’s coefficient indicates acceptable multivariate normality, supporting ML-based CBSEM estimation.
Model Fit Indices	Emphasises strict global goodness-offit (e.g., ${\chi }^2/df$, RMSEA, CFI, TLI, GFI) and nested model testing.	Emphasises predictive relevance; global fit indices are not the primary evaluation criteria.	Aligned. The study requires global model fit assessment and $\Delta {\chi }^2$ nested model testing to compare constrained and unconstrained models.

).

Reliability and Validity. Reliability was evaluated using Cronbach’s $\alpha$ and composite reliability (CR). Discriminant validity was rigorously examined using both the traditional Fornell–Larcker criterion [37] and the more robust Heterotrait–Monotrait (HTMT) ratio of correlations [38], ensuring that all HTMT values remained below the conservative threshold of 0.85.

Structural Model & Hypothesis Testing. Structural paths were estimated using maximum likelihood (ML) estimation. To rigorously test the affective primacy hypothesis (H2c), we conducted a chi-square difference test ($\Delta {\chi }^2$) comparing an unconstrained model with a constrained model in which the paths from CV and AV to CI were set to be equal.

Mediation Analysis. The comparison of indirect pathways (H5) was assessed using bias-corrected bootstrapping with 5000 resamples to generate 95% confidence intervals.

Common method bias (CMB) was addressed both procedurally and statistically, following the recommendations of Podsakoff et al. [39]. Procedurally, we ensured respondent anonymity, separated independent and dependent variables in the survey flow, and randomised item order to reduce consistency motifs. We also embedded attention checks in the form of instructional manipulation checks [40] to filter out inattentive responses and ensure data quality. Harman’s single-factor test indicated that the first factor accounted for 30.83% of the total variance, well below the 50% threshold [39]. Model fit was evaluated using the strict cutoff criteria recommended by Hu et al. [41], while also acknowledging the sensitivity of fit indexes to lack of measurement invariance as highlighted by Chen [42].

4. Results

4.1. Measurement-Model Assessment

The measurement model was evaluated using confirmatory factor analysis (CFA) in IBM AMOS 26.0. The model demonstrated excellent fit to the data (${\chi }^2/df$ = 1.17, CFI = 0.994, TLI = 0.992, RMSEA = 0.020, SRMR = 0.029), meeting and exceeding the stringent cutoff criteria recommended by Hu et al. [41] and remaining well within acceptable ranges even when considering the sensitivity of these indexes to measurement non-invariance discussed by Chen [42]. Given these exceptionally robust indices (particularly RMSEA = 0.020), we explicitly confirmed that no theoretically unsubstantiated modification indices were applied to improve fit, thereby ensuring the model’s parsimony.

Reliability and convergent validity. To assess the psychometric properties of the scales, we examined standardised factor loadings, composite reliability (CR) and average variance extracted (AVE). As shown in

Table 4. Reliability and convergent validity.

Construct	Item	Loading	Cronbach’s $\mathit{\alpha }$	CR	AVE
	TI1	0.805	0.859	0.859	0.670
TI	TI2	0.832
	TI3	0.819
	SI1	0.760	0.850	0.857	0.667
SI	SI2	0.892
	SI3	0.793
	PERS1	0.890	0.878	0.880	0.710
PERS	PERS2	0.810
	PERS3	0.826
	EV1	0.822	0.862	0.862	0.676
EV	EV2	0.836
	EV3	0.808
	AV1	0.850	0.864	0.866	0.684
AV	AV2	0.861
	AV3	0.767
	CI1	0.831	0.883	0.888	0.727
CI	CI2	0.890
	CI3	0.835
	SAT1	0.841	0.895	0.897	0.743
SAT	SAT2	0.894
	SAT3	0.850
	CONT1	0.863	0.911	0.912	0.775
CONT	CONT2	0.922
	CONT3	0.855

Note: CR = composite reliability; AVE = average variance extracted. All factor loadings are significant at p < 0.001 (two-tailed).

, all standardised factor loadings were statistically significant (p < 0.001) and ranged from 0.760 to 0.922, indicating strong indicator reliability. Composite reliability (CR) values for all constructs ranged from 0.857 to 0.912, well above the recommended threshold of 0.70. Furthermore, average variance extracted (AVE) values ranged from 0.667 to 0.775, exceeding the 0.50 benchmark and confirming satisfactory convergent validity for all latent constructs.

Second-order construct validation. The second-order CFA confirmed that the higher-order “interactivity” construct is strongly reflected in its three first-order subdimensions—technological interactivity, social interactivity and personalisation (see

Table A3. Second-order CFA for interactivity.

Path (Second Order → First Order)	Std. $\mathit{\beta }$	S.E.	C.R. (t-Value)	p-Value	Conclusion
Interactivity → Technical Interactivity (TI)	0.750	-	-	***	Supported
Interactivity → Social Interactivity (SI)	0.710	0.083	8.542	***	Supported
Interactivity → Personalisation (PERS)	0.786	0.109	8.996	***	Supported
Model Fit	${\chi }^2/\mathrm{df}$ = 1.709	$\mathrm{CFI}$ = 0.992	RMSEA = 0.042	Target Coeff = 1.000

Note: Std. $\beta$ = standardised path coefficient; C.R. = critical ratio. The path from interactivity to TI was fixed to 1.0 for model identification; therefore, standard error and critical ratio are not reported. *** p < 0.001.

). Specifically, personalisation exhibited the strongest loading ($\beta$ = 0.786), followed by technological interactivity ($\beta$ = 0.750) and social interactivity ($\beta$ = 0.710). This hierarchical structure justifies treating interactivity as a holistic managerial stimulus in the structural model.

Discriminant validity. To ensure that the constructs were empirically distinct, we employed the Fornell–Larcker criterion and the heterotrait–monotrait (HTMT) ratio. As shown in

Table A4. Discriminant Validity: Square root of AVE and latent variable correlations.

Construct	Mean	SD	1	2	3	4	5	6	7	8
1. TI	3.68	1.05	0.819
2. SI	4.09	0.84	0.548	0.817
3. PERS	4.13	0.86	0.597	0.590	0.843
4. EV	3.94	0.98	0.161	0.221	0.102	0.822
5. AV	3.89	0.95	0.433	0.505	0.406	0.244	0.827
6. CI	3.83	1.06	0.236	0.304	0.265	0.213	0.535	0.853
7. SAT	4.04	0.97	0.272	0.267	0.198	0.203	0.416	0.471	0.862
8. CONT	3.93	0.99	0.303	0.220	0.245	0.115	0.323	0.256	0.501	0.880

Note: Bold diagonal elements represent the square root of AVE; values below the diagonal represent latent variable correlations.

, the square root of the AVE for each construct exceeded its inter-construct correlations. Additionally, all HTMT values were below the conservative threshold of 0.85, confirming robust discriminant validity. It is worth noting that since the second-order construct ‘interactivity’ is reflected by three first-order dimensions, the higher-order structure is just-identified. Consequently, the target coefficient is 1.0, indicating that the second-order conceptualisation captures the covariance among dimensions without loss of information compared with the first-order correlated model.

4.2. Structural Model and Hypothesis Testing

The structural model (Figure 3) exhibited a good fit (${\chi }^2/\mathrm{df}$ = 1.32, CFI = 0.987, RMSEA = 0.028).

Explanatory Power $\left({\mathrm{R}}^2\right)$: The model explained substantial variance in key constructs: $37.0\%$ for affective value, $30.1\%$ for cultural identity, and $24.4\%$ for continuance intention. This indicates that the proposed dual-path mechanism accounts for a meaningful proportion of user engagement.

Path Analysis (H1–H4): As detailed in

Table 5. Structural model path coefficients and hypothesis testing results.

Hypothesis	Path	Std. $\mathit{\beta }$	S.E.	C.R.	p-Value	Result
H1a	Interactivity → Epistemic Value (EV)	0.239	0.085	3.802	***	Supported
H1b	Interactivity → Affective Value (AV)	0.608	0.094	8.496	***	Supported
H2a	Epistemic Value (EV) → Cultural Identity (CI)	0.108	0.057	2.154	0.031	Supported
H2b	Affective Value (AV) → Cultural Identity (CI)	0.522	0.065	9.494	***	Supported
H3	Cultural Identity (CI) → Satisfaction (SAT)	0.481	0.045	8.848	***	Supported
H4	Satisfaction (SAT) → Continuance Intention (CONT)	0.494	0.056	9.326	***	Supported

Note: Std. $\beta$ = Standardised path coefficient; C.R. = Critical ratio. *** p < 0.001.

, interactivity acted as a powerful stimulus, significantly enhancing both affective value ($\beta$ = 0.608, p < 0.001) and epistemic value ($\beta$ = 0.239, p < 0.001). In turn, cultural identity functioned as a pivotal driver of satisfaction ($\beta$ = 0.481, p < 0.001; supporting H3), which subsequently fueled continuance intention ($\beta$ = 0.494, p < 0.001; supporting H4).

These results clarify the cyclical nature of the engagement mechanism. Continuance intention functions not merely as an endpoint but as the behavioural trigger for the reinforcement loop depicted in the virtual–real symbiosis model (Figure 1; green dotted line). By strengthening intentions to return, this outcome reactivates the psychological primer, helping sustain an iterative cycle between digital stimulation and on-site visitation.

4.3. Multi-Factor Competitive Analysis: Establishing Affective Primacy

To rigorously address the inconsistency in prior research—specifically the “null effect” anomaly noted by Deng et al. [11]— we conducted targeted statistical comparisons (see

Table 6. Tests of affective primacy (comparison of direct and indirect effects).

Test Type	Comparison (Path/Chain)	$\mathit{\beta }/\mathbf{\Delta }{\mathit{\chi }}^2$	95% BCaCI	p-Value	Conclusion
H2c: Path Comparison	AV → CI (Affective Path)	0.522	-	-	Supported
	EV → CI (Cognitive Path)	0.108	-	-	(Affective > Cognitive)
	Difference Test ($\Delta {\chi }^2,\mathrm{df}=1$)	26.207	-	0.000	Difference highly significant
H5: Chain Comparison	Affective Chain (INT → AV → CI → SAT → CONT)	0.102	[0.055, 0.175]	0.000	Supported
	Cognitive Chain (INT → EV → CI → SAT → CONT)	0.008	[0.001, 0.027]	0.025	(Affective > Cognitive)
	Effect Difference (Affective Cognitive)	0.094	[0.049, 0.167]	0.000	Significant

Note: BCaCI = bias-corrected and accelerated confidence interval. Indirect effects are based on bias-corrected bootstrapping estimates. Path comparison ($\Delta {\chi }^2$ test): We compared an unconstrained model with a constrained model in which the paths from epistemic value and affective value to cultural identity were set to be equal. The chi-square difference test revealed that the constrained model fit significantly worse.

).

Path Comparison ($\Delta {\chi }^2$ Test): We compared an unconstrained model with a constrained model where the paths from epistemic value and affective value to cultural identity were set to be equal. The chi-square difference test revealed that the constrained model fit significantly worse ($\Delta {\chi }^2\left(1\right)$ = 26.207, p < 0.001). This result provides strong statistical evidence that affective value exerts a significantly stronger impact ($\beta$ = 0.522) on cultural identity than epistemic value ($\beta$ = 0.108), thereby supporting H2c.

Indirect Effect Comparison (H5): Using bias-corrected bootstrapping (5000 resamples), the analysis confirmed that the affective chain (interactivity → AV → CI → SAT → CONT) had a significantly stronger indirect effect ${\beta }_{indirect}$ = 0.102 compared with the cognitive chain (${\beta }_{\mathrm{indirect}}$ = 0.008). The confidence interval of the difference excluded zero ($\Delta \beta$ = 0.094, 95% CI [0.049, 0.167]), confirming that the mechanism of virtual–real symbiosis was primarily emotion-driven.

Affective primacy is supported when (a) AV → CI > EV → CI (H2c) and (b) the affective chain’s indirect effect exceeds the cognitive chain (H5 Effect Difference). Indirect effects and confidence intervals are based on 5000 bootstrap resamples (bias-corrected and accelerated, BCa).

4.4. Robustness Check: Competing-Model Analysis

To ensure the robustness of our findings, we compared the proposed full-mediation model against a partial-mediation alternative (see

Table A5. Comparison of competing structural models.

Model	Description	${\mathit{\chi }}^2$($\mathbf{df}$)	CFI	RMSEA	AIC	Model Comparison ($\mathbf{\Delta }{\mathit{\chi }}^2$)
M0	Baseline Model (Full Mediation)	319.465 (243)	0.987	0.028	433.465	-
M2	Partial Mediation (Adds value → SAT path)	299.710 (241)	0.990	0.025	417.710	19.755 (p < 0.001)

Note: CFI = Comparative Fit Index; RMSEA = Root Mean Square Error of Approximation; $\mathrm{AIC}$ = Akaike Information Criterion. A significant $\Delta {\chi }^2$ indicates that the added path improves model fit. p < 0.001.

) (which added direct paths from value dimensions to satisfaction). While the partial model showed a slightly better fit, the dominance of the affective pathway (AV vs. EV) remained consistent and highly significant across all specifications. This confirms that the core mechanism for building deep cultural identity remains the affective pathway verified in the main model.

5. Discussion

5.1. From Archive-Centric to Soul-Centric Management

The results indicate a shift in digital heritage management from an “archive-centric” approach to “soul-centric” management. Although technical interactivity and epistemic value lower the entry barrier, affective value shows a substantially larger association with sustained engagement ($\beta$ = 0.522) than epistemic value ($\beta$ = 0.108), suggesting that emotional stickiness is central to long-term engagement. Users in immersive DCH contexts therefore appear to seek not only historical knowledge but also an affective connection to the past.

Together, these findings imply that the long-term sustainability of digital heritage depends on an affective infrastructure—a psychological micro-foundation that supports progress towards SDG 11.4. We summarise this hierarchy—from cognitive utility to affective resonance—as the “hierarchy of digital heritage engagement” (Figure 4).

5.2. Theoretical Implications

First, substantiating the virtual–real symbiosis mechanism.

Our results support a reinforcement hypothesis rather than a substitution hypothesis. We articulate the “virtual–real symbiosis” mechanism [9] as operating on three interrelated levels:

(i)

Experience Level: This study proposes shifting the conceptualisation of digital museums from static ‘containers of information’ to ‘affective amplifiers’. Digital platforms function as experiential extensions. As Luther et al. [2] argue regarding digital twins, these technologies provide “impossible viewpoints” that cannot be accessed in situ, fueling curiosity rather than replacing it.

(ii)

Behavioural Level: High-definition immersion creates a productive tension of “seeing but not being there.” This experiential gap transforms informational satisfaction into a craving for embodied verification—a desire to confirm virtual impressions physically.

(iii)

Value Level: While epistemic value offers rational justification, Affective Value builds identity-based motivation. This aligns with the mandates of SDG 11.4, suggesting that robust digital engagement underpins the economic resilience of the broader heritage ecosystem.

Second, validating affective primacy in heritage adoption. By rigorously testing H2c and H5, we provide context-specific evidence for the affective primacy hypothesis [29]. The clear dominance of the affective chain challenges traditional TAM-inspired models that prioritise functional utility [23]. Instead, our results point towards a “feeling-first, meaning-second” adoption logic, where cultural identity formation is structurally an affective process grounded in attachment [27].

5.3. Practical and Ethical Implications for Sustainable Management

To operationalise these theoretical insights in sustainable management practice, DCH managers must fundamentally reorient their strategies from information delivery to emotion curation. Drawing on our findings, we propose three strategic shifts for practice.

(1)

From archiving to narrative staging: A staged affective-arousal strategy. Because narrative involvement alone can yield a null effect when decoupled from emotion [28], managers should shift from “displaying data” to staging narratives. Design could follow a staged affective–arousal strategy, beginning with low-intensity cognitive scaffolding to establish an initial mental schema and progressively introducing higher-intensity emotional challenges. Prioritising sensory cues—such as dynamic lighting and immersive soundscapes—can create an affective hook that leverages affective primacy.

(2)

Deploying AI-driven narrative agents as empathetic guides. Although our model foregrounds affective experience, implementation may be constrained in contexts where human staffing is infeasible. Managers could deploy AI-driven narrative agents. Rather than relying on static information boards, DCH platforms could incorporate AI-enabled non-player characters (NPCs) that infer user affect and adapt narrative tone, functioning as empathetic guides rather than didactic lecturers [43]. However, Suiçmez et al. [44] report that 73.4% of visitors have not yet experienced AI museum applications, suggesting a prolonged novelty phase. Moreover, robust data governance is essential to prevent algorithmic bias from distorting cultural authenticity [45].

(3)

Emotional ROI and sustainability metrics. Cultural institutions should complement conventional behavioural indicators (for example, page views and session duration) with metrics of “emotional retention,” such as reported pride, belonging, and emotional resonance. In line with Li et al. [8], our results indicate that these emotion-–behaviour links are stronger predictors of long-term recommendation intentions than satisfaction alone. Monitoring “emotional stickiness” therefore offers a more sensitive and accurate indicator of a project’s long-term sustainability. From the perspective of sustainable resource allocation, investments in emotionally rich interaction design yield a higher return on investment (ROI) than simply layering additional high-definition digital technologies. We argue that affective stickiness functions as a “renewable psychological resource” that sustains intergenerational cultural transmission at a lower marginal cost.

5.4. Limitations and Future Research

Despite the methodological rigour, several limitations delineate avenues for future research. First, the cross-sectional design precludes causal inference about the long-term evolution of user behaviour; future work could adopt longitudinal designs to track how the balance between epistemic and affective value shifts over time. Second, the sample skews towards younger cohorts (19–34 years; 69.1%; Table 2), which may limit generalisability.

The observed null effect of epistemic value on identity should be interpreted in light of cultural context. Given that most respondents were from China (79.4%), the pattern may reflect characteristics of high-context communication [46], in which identity formation can rely more on implicit affective ties than on explicit information exchange. This framing may help explain why affective experience outweighed informational gains in shaping identity-related outcomes. Future research could test the model in lower-context settings (e.g., Western Europe) to evaluate this cultural boundary condition.

Although this concentration limits generalisability to older cohorts, it is strategically relevant for sustainability because digital natives are likely to be key actors in advancing SDG 11.4 and strengthening the intergenerational resilience of cultural heritage. Finally, future models could explicitly incorporate variables related to generative AI. As generative AI (GenAI) shifts personalisation from static recommendation to real-time content creation, future research should examine how personalised narrative curation shapes perceived authenticity and whether it disrupts or strengthens the affective pathway.

6. Conclusions

In the era of the “digital turn,” this study provides empirical evidence on the psychological mechanisms that structure user engagement with digital heritage experiences. By integrating the S-O-R framework [10] with a dual-pathway competitive model, we demonstrate that although technical interactivity simultaneously engages cognition and emotion, it is affective value that forms the critical bridge to loyalty. The validated structural model (Figure 3) supports the theoretical hierarchy summarised in Figure 4. This alignment indicates that epistemic value (the foundation) facilitates initial entry, whereas emotional stickiness (the apex) underpins the sustained commitment needed for heritage resilience.

The optimal relationship between digital and physical museums is not substitution but virtual–real symbiosis. Digital platforms extend cognitive reach and activate emotional investment, which can subsequently be realised in the physical realm. By jointly strengthening cognitive understanding and affective attachment, digital heritage initiatives contribute not only to the protection of assets but also to the cultivation of resilient cultural identities—a foundational requirement for the long-term success of the UN Sustainable Development Goal (SDG) 11.4.

Ultimately, this study suggests that the sustainability of digital cultural heritage does not lie in the preservation of data bytes alone but in the sustained resonance of cultural memory. By pivoting from an information-centric to an emotion-centric management strategy, DCH platforms can transcend their role as mere archives. They become resilient “emotional infrastructures” that safeguard heritage against the twin threats of physical decay and public amnesia, thereby fulfilling the deepest mandate of SDG 11.4.