Museum digital services

As digital technologies have become increasingly integrated into museum exhibitions and public services, museum digitalization has expanded beyond collection digitization and back-end management to support how visitors access information, interpret exhibitions, participate in activities, and extend their museum experiences [32]. Museum digital services therefore encompass not only individual technologies or devices but also a range of digitally mediated services that support visitor access, interpretation, interaction, and participation. These practices broaden access to museum collections and contribute to knowledge preservation, resource use, and visitor experience [33,34].

In actual visits, these digital functions are often distributed across pre-visit information search, on-site guidance and interaction, and post-visit review and sharing, thereby supporting the full visiting process and its extended experience [35]. Visitors may therefore encounter several forms of digital support during the same museum experience rather than relying on a single isolated technology. As a result, visitors usually form an overall judgment of whether the museum’s digital support is useful, easy to use, and smooth as an experience. Lu et al. also noted that museum technology research has gradually moved beyond early concerns with learning and interaction toward variables that are closer to the adoption process, such as intention, behavior, and experience outcomes [36].

In both the Chinese and Russian contexts, museum digital services have continued to develop under the combined influence of policy support and platform building. In China, plans for cultural relic protection and scientific innovation have clearly called for stronger technology application and public service capacity in the museum sector [37,38]. In Russia, public-facing mobile guides and digital access points are provided through cultural sector projects and public platforms [39]. Given the differences in service provision pathways and platform ecosystems between the two countries, restricting the research object to a single technological entry point would make cross-sample comparison more vulnerable to asymmetries in availability and use scenarios. For this reason, this study uses the term museum digital services to cover mini-programs or mobile applications, online exhibitions, and AR- or VR-based guides, and takes visitors’ perceptions of overall digital support as the unit of analysis.

Unified theory of acceptance and use of technology

Technology acceptance refers to an individual’s willingness to adopt and use a technology [40]. In this study, behavioral intention (BI) is referred to as acceptance intention and represents visitors’ intention to use museum digital services in the future. Behavioral theory suggests that individual action is often organized first through intention [41], and intention serves as a link among attitude, social norms, and perceived behavioral control [5]. Therefore, although a gap remains between intention and behavior, meta-analytic evidence shows a strong statistical association between the two [42], and in the absence of longitudinal tracking data or usage logs, using BI as the main outcome variable remains methodologically reasonable [43].

The UTAUT family of models offers a relatively unified framework for explaining technology adoption and has been widely used to analyze adoption intention across different technologies and services [44,45]. It has also demonstrated explanatory value for behavioral intention in previous empirical applications [46]. In 2012, Venkatesh et al. extended the research setting from organizational use to consumer and voluntary use by proposing UTAUT2. Beyond the original core paths, the model added hedonic motivation, price value, and habit, which made it more suitable for explaining intention formation in experience-driven and non-mandatory adoption settings [7].

In public-facing digital service settings, BI is often shaped by both instrumental evaluation and experiential evaluation. More importantly, UTAUT offers a useful basis for explaining acceptance in social and cultural contexts, which makes it suitable for studying the digitalization of public cultural services [8,47]. Recent studies have also applied UTAUT-based and related user-acceptance models to digital museums and cultural heritage services, showing that functional value, ease of use, and experiential factors may contribute to shaping users’ acceptance and usage intentions [13,48,49]. However, these studies have generally focused on a particular technology, digital platform, or national context, leaving the cross-cultural acceptance of broader museum digital support less fully examined.

Based on the specific context of museum digital services, this study assessed the contextual suitability of the UTAUT2 variables. Price value and habit were excluded because of their limited applicability to the present study. Price value reflects the trade-off between the perceived benefits of a technology and its monetary cost [7]. However, most of the museum digital services covered by the questionnaire, including mobile guides, online exhibitions, and AR/VR-based experiences, were not presented as separately priced products, leaving respondents without a consistent basis for evaluating this trade-off. Habit is more applicable to technologies used frequently and under relatively stable conditions [50], whereas museum digital services are often encountered during particular visits and may vary across museums and visiting situations, making stable habitual use difficult to assess. Although some services may involve indirect costs or repeated use, these conditions do not apply consistently across the overall service context examined here.

Therefore, on the basis of the context-adapted UTAUT2 model, this study retained performance expectancy (PE), effort expectancy (EE), social influence (SI), facilitating conditions (FC), and hedonic motivation (HM) to explain visitors’ acceptance intention (BI) toward museum digital services. Performance expectancy (PE) reflects visitors’ judgment of whether digital services can generate benefits for the museum visit [12], and existing studies suggest that when users believe a technology can deliver clear benefits, their acceptance intention tends to become stronger [51]. Effort expectancy (EE) refers to perceived ease of use, that is, visitors’ perception of the learning and operational costs involved in using digital technology. Easy-to-use interface design can reduce technical barriers and strengthen visitors’ willingness to use digital services, particularly in heritage visits and on-site guidance settings where immediate interaction is required [52]. Social influence (SI) concerns the extent to which normative cues, such as peer suggestions, social media discussions, or the atmosphere on site, shape intention. The use of museum digital services is not a purely private technical act, but often takes place in visiting situations with clear social visibility. Related studies also show a positive relationship between social influence and behavioral intention [26,48], which makes this factor meaningful in cross-sample comparison. Facilitating conditions (FC) concern whether external resources such as network access, devices, on-site guidance, and staff support are available. They affect not only whether the service can be used, but also how visitors assess usage risk and time costs. In consumer settings, differences in facilitating conditions are more likely to influence intention and use [13]. HM reflects the enjoyment, fun, and sense of immersion experienced during use. Studies of museum VR and related digital experiences have identified enjoyment and immersion as important factors in users’ evaluations, satisfaction, and continued engagement [49]. The retention of HM is particularly relevant because museum digital services combine informational functions with cultural, aesthetic, and experiential engagement.

Hofstede’s cultural dimensions

Individuals’ interpretations of social norms, judgments of uncertainty in technology use, and evaluations of long-term benefits are often influenced by their cultural value orientations. Srite and Karahanna incorporated individual-level cultural values into a technology acceptance framework and showed that individuals’ endorsement of particular cultural values can alter the way factors such as social influence affect behavioral intention [15]. Subsequent research integrating cultural dimensions with UTAUT2 has also suggested that cultural values do not necessarily determine technology acceptance directly; rather, they may change the strength of the relationships between different antecedents and intention formation [30].

Hofstede conceptualized national culture along six dimensions, providing a structured framework for examining the relationship between culture and technology acceptance [53]. However, Hofstede’s national culture scores primarily reflect aggregate societal tendencies and cannot be assumed to represent the actual value orientations of every individual within a country. Directly inferring individual-level cultural characteristics from national-level cultural patterns may result in an ecological fallacy [54]. Therefore, this study does not equate culture with nationality but uses the CVSCALE to measure the cultural values actually endorsed by individual respondents [17]. In this study, the Chinese and Russian samples represent different social and digital-service contexts, whereas collectivism orientation, uncertainty avoidance, and long-term orientation are treated as individual-level cultural value variables.

Beyond Hofstede’s framework, cross-cultural technology-use research has also drawn on Schwartz’s theory of cultural value orientations to explain users’ technology-related judgments and behaviors across different cultural settings through alternative value structures [55]. This line of research indicates that the cultural influences on technology use can be examined through different theoretical frameworks and further suggests that the selection of cultural dimensions should correspond to the specific research context and the mechanisms under investigation.

Of Hofstede’s six dimensions, this study selected collectivism orientation (COL), uncertainty avoidance (UAI), and long-term orientation (LTO) because they have relatively clear theoretical links to the technology acceptance relationships examined in this study. COL is associated with the group-reference and social-coordination mechanisms reflected in social influence. UAI is related to the reduction of uncertainty and the availability of external support reflected in effort expectancy and facilitating conditions. LTO is associated with the long-term functional value reflected in performance expectancy and the extended experiential value that hedonic motivation may carry in the museum context.

Power distance, masculinity/femininity, and indulgence/restraint were not included because power distance and masculinity/femininity lack direct and theoretically distinct links to the specific acceptance relationships examined in the present model. Meta-analytic evidence also indicates that their moderating effects in technology acceptance models are generally relatively weak [16]. Indulgence/restraint is not included in the original CVSCALE adopted in this study [17], and its measurement at the individual level would require a separately developed scale [56]. Moreover, this dimension reflects a general orientation toward gratification and self-restraint, whereas hedonic motivation concerns the enjoyment derived from using a specific technology. A clear and independent mechanism linking indulgence/restraint to technology-specific hedonic motivation would therefore need to be established. Based on these theoretical and measurement considerations, the present study ultimately included COL, UAI, and LTO.

The moderating logic of collectivism (COL)

Collectivism emphasizes group relationships, shared goals, and consistency between the individual and the group. Individuals with stronger collectivist orientations are more likely to rely on social reference points and group norms when forming judgments and to regard the opinions and behavior of others as important decision cues [53]. In museum visits involving shared participation, the use of digital services is often socially visible. Visitors may judge whether a digital service is worth using based on recommendations from companions, the observable behavior of other visitors, and discussions on social media. Cross-cultural research has shown that differences in individualism–collectivism orientation can influence technology acceptance behavior [57]. A recent study of e-government AI assistants in China also identified social influence as an important predictor of usage intention [28], providing additional evidence of the relevance of social cues in Chinese digital service contexts. For visitors with stronger collectivist orientations, group norms and the behavior of others are likely to carry greater referential value. When companions, social networks, or other visitors express positive attitudes toward a digital service, these social cues are more likely to be incorporated into the individual’s acceptance judgment, thereby strengthening the positive effect of social influence on acceptance intention. Accordingly, the following hypothesis is proposed:

1. H2a: Collectivism (COL) positively moderates the effect of social influence (SI) on acceptance intention (BI); the higher the level of COL, the stronger the effect of SI on BI.

The moderating logic of uncertainty avoidance (UAI)

Uncertainty avoidance reflects the extent to which individuals feel uncomfortable with ambiguity, unpredictability, and potential risk. Meta-analytic evidence indicates that UAI is related to ease-of-use judgments and technology acceptance pathways, although the direction and strength of its moderating effects may vary across technologies and model relationships [16]. In museum settings, visitors often need to understand unfamiliar interfaces and interaction procedures within a limited amount of time. Lower learning difficulty and clearer operating procedures can make the use process more predictable. For visitors with higher levels of UAI, reducing ambiguity and uncertainty during use may be particularly important. Effort expectancy may therefore become a more important basis for forming acceptance intention, strengthening its positive effect on BI.

Individuals with higher levels of UAI may also place greater importance on external support and safeguards that reduce operational risks. Stable devices and network connections, clear instructions, and on-site assistance from staff can reduce potential problems during the use of museum digital services. Previous research has shown that UAI can moderate the relationship between facilitating conditions and behavioral intention [58]. When visitors attach greater importance to the controllability and reliability of the use process, facilitating conditions are more likely to translate into acceptance intention. Therefore, the following hypotheses are proposed:

1. H3a: Uncertainty avoidance (UAI) positively moderates the effect of effort expectancy (EE) on acceptance intention (BI); the higher the level of UAI, the stronger the positive effect of EE on BI.
2. H3b: Uncertainty avoidance (UAI) positively moderates the effect of facilitating conditions (FC) on acceptance intention (BI); the higher the level of UAI, the stronger the positive effect of FC on BI.

The moderating logic of long-term orientation (LTO)

Long-term orientation emphasizes future returns, sustained investment, and the long-term consequences of present actions [53]. Individuals with stronger long-term orientations tend to place greater value on the sustained benefits that technology may provide and may be more willing to invest the learning and adaptation effort required to obtain those benefits. Recent research has found that long-term orientation positively predicts mobile technology adoption intention [59], suggesting that when the value of a technology emerges gradually through continued use and learning, long-term value orientation may become an important basis for acceptance.

Long-term orientation may also shape how visitors evaluate the functional benefits of technology. Meta-analytic evidence indicates that long-term orientation positively moderates the relationship between perceived usefulness and behavioral intention [16]. Perceived usefulness and performance expectancy both concern users’ evaluations of the functional benefits provided by technology. In the context of museum digital services, performance expectancy involves not only whether a service improves information access and navigation during the current visit, but also whether it supports deeper cultural understanding and extends the learning experience. Visitors with stronger long-term orientations may attach greater importance to these sustained benefits, making performance expectancy more likely to translate into acceptance intention.

The moderating role of LTO in the relationship between hedonic motivation and acceptance intention is more specific to the museum context. Previous research has shown that the usability and interactivity of museum technologies can strengthen visitors’ cognitive engagement and subsequently influence their learning experiences [60]. Enjoyment generated by digital services therefore need not be limited to immediate entertainment; it may also serve as an experiential resource that supports deeper exploration, continued engagement, and cultural understanding. For visitors with stronger long-term orientations, enjoyment and immersion may be more readily translated into acceptance intention when they also contribute to further learning and sustained engagement. Accordingly, the following hypotheses are proposed:

1. H4a: Long-term orientation (LTO) positively moderates the effect of performance expectancy (PE) on acceptance intention (BI); the higher the level of LTO, the stronger the positive effect of PE on BI.
2. H4b: Long-term orientation (LTO) positively moderates the effect of hedonic motivation (HM) on acceptance intention (BI); the higher the level of LTO, the stronger the positive effect of HM on BI.

The proposed research model is shown in Fig 1.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. The research model.

https://doi.org/10.1371/journal.pone.0357563.g001

Questionnaire development and measurement

This study used a structured questionnaire to measure museum visitors’ acceptance intention toward digital services and its influencing factors in the Chinese and Russian samples. The items for the technology acceptance constructs (PE, EE, SI, FC, HM, and BI) were adapted from the established UTAUT2 scale [7] and were semantically adjusted to fit the context of museum digital services. For example, the technology referred to in the item wording was specified as services such as mobile guides, AR/VR experiences, and online exhibitions. The Hofstede-related cultural dimensions—collectivism orientation (COL), uncertainty avoidance (UAI), and long-term orientation (LTO)—were measured at the individual level using the CVSCALE [17] and contextualized for museum digital-service use.

To ensure consistency across language versions, the questionnaire was translated into Chinese and Russian by the first author, and the Russian version was subsequently reviewed by a Russian-language expert to ensure semantic accuracy and clarity. The item wording was revised accordingly. A small-scale pretest involving 20 participants was conducted before formal data collection to examine the clarity and comprehensibility of the questionnaire items. The questionnaire contained two parts: the first part collected respondents’ background information, and the second part included the items for each construct. All items were measured on a seven-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree). This study was conducted in accordance with the Declaration of Helsinki and was approved by the Medical Ethics Committee of Jiangnan University (Approval No. JNU202506RB067). Before participation, all participants were informed of the study purpose, the voluntary nature of participation, the anonymous handling of data, and the use of the data solely for research purposes. Electronic informed consent was obtained from participants in the Chinese online survey, whereas oral informed consent was obtained from participants in the Russian offline survey. No personally identifiable information was collected.

Data collection

This study used a combination of online and offline methods for data collection. The Chinese questionnaire was distributed mainly through WeChat and Xiaohongshu from 10 October 2025–20 October 2025, resulting in 350 returned questionnaires. The Russian questionnaire was collected mainly through offline intercept surveys conducted around Palace Square and university areas in Saint Petersburg from 10 October 2025–25 October 2025, resulting in 350 returned questionnaires.

To ensure data quality, this study applied strict screening criteria. First, responses were excluded if participants reported no previous experience with museum digital services, as such experience was considered necessary for evaluating acceptance intention toward these services (21 Chinese responses and 34 Russian responses). Second, invalid responses with excessively short completion times (less than 100 s) or incomplete answers were removed (13 Chinese responses and 9 Russian responses). The final valid sample included 316 Chinese questionnaires (effective rate = 90%) and 307 Russian questionnaires (effective rate = 88%). The final valid sample size in each group (N > 300) was considered sufficient for structural equation modeling (SEM) and supported stable parameter estimation.

The final sample included 316 respondents from China and 307 respondents from Russia. Men accounted for 53.48% of the Chinese sample and 58.96% of the Russian sample, while women accounted for 46.52% and 41.04%, respectively. Respondents aged 18–35 constituted the largest age group in both samples, representing 67.41% of the Chinese sample and 73.62% of the Russian sample. Participants aged 36–55 accounted for 19.94% and 18.24%, while those older than 55 accounted for 12.66% and 8.14%, respectively. Regarding education, 34.18% of the Chinese respondents had a high school education or below, 39.56% had university-level education, and 26.27% had postgraduate education or above. In the Russian sample, the corresponding proportions were 11.73%, 59.61%, and 28.66%. Respondents with university or postgraduate education accounted for 65.82% of the Chinese sample and 88.27% of the Russian sample. Detailed demographic characteristics are presented in Table 1.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Participants’ demographic characteristics.

https://doi.org/10.1371/journal.pone.0357563.t001

Data analysis

Data processing and descriptive statistical analyses were conducted using SPSS 26.0, and partial least squares structural equation modeling (PLS-SEM) was used to test the proposed model. The measurement model was assessed using indicator loadings, Cronbach’s alpha, composite reliability (CR), average variance extracted (AVE), and the heterotrait–monotrait ratio (HTMT). The structural model was evaluated using variance inflation factors (VIF), path coefficients, the coefficient of determination (R²), predictive relevance (Q²), and effect sizes (f²). Direct and interaction effects were assessed using 5,000 bootstrap resamples and 95% confidence intervals. Values of f² equal to 0.02, 0.15, and 0.35 were interpreted as small, medium, and large effects, respectively [61].

Moderating effects were tested using a two-stage construct-score approach. Five interaction terms—SI × COL, EE × UAI, FC × UAI, PE × LTO, and HM × LTO—were created using standardized predictor and moderator scores. Significant interaction effects were further examined through simple-slope analyses at the moderator mean and at one standard deviation below and above the mean. Before comparing the Chinese and Russian samples, measurement invariance was assessed using the measurement invariance of composites (MICOM) procedure with 5,000 permutations [62]. Permutation-based multigroup analysis and PLS-MGA were subsequently used to compare the direct and interaction effects across the two samples [63,64]. Between-sample differences were evaluated using two-sided permutation p-values below 0.05 and PLS-MGA p-values below 0.05 or above 0.95.

Measurement model

The reliability and convergent validity of the measurement model were assessed using indicator loadings, Cronbach’s alpha, composite reliability (CR), and average variance extracted (AVE). As shown in Tables 2 and 3, all indicator loadings exceeded the recommended threshold of 0.70, ranging from 0.815 to 0.923 in the Chinese sample and from 0.800 to 0.925 in the Russian sample. Cronbach’s alpha ranged from 0.828 to 0.902 in the Chinese sample and from 0.797 to 0.901 in the Russian sample, while CR values ranged from 0.890 to 0.938 and from 0.880 to 0.938, respectively. The AVE values ranged from 0.744 to 0.835 in the Chinese sample and from 0.711 to 0.834 in the Russian sample. All reliability and convergent-validity indicators met the recommended criteria, indicating satisfactory internal consistency and convergent validity in both samples.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Measurement model assessment for the Chinese sample (n = 316).

https://doi.org/10.1371/journal.pone.0357563.t002

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Measurement model assessment for the Russian sample (n = 307).

https://doi.org/10.1371/journal.pone.0357563.t003

Discriminant validity was assessed using the heterotrait–monotrait ratio (HTMT). As shown in Table 4, all HTMT values were below the recommended threshold of 0.85. The highest HTMT value was 0.674 in the Chinese sample and 0.612 in the Russian sample, indicating satisfactory discriminant validity across all constructs in both samples.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. HTMT results for discriminant validity in the Chinese and Russian samples.

https://doi.org/10.1371/journal.pone.0357563.t004

Measurement invariance

Before conducting the multigroup comparisons, measurement invariance between the Chinese and Russian samples was assessed using the MICOM procedure. Configural invariance was established because the same measurement items, data treatment procedures, model specifications, and estimation settings were used in both samples. Based on 5,000 permutations, the original correlations for PE, SI, HM, COL, UAI, LTO, and BI exceeded their respective 5% quantiles, indicating compositional invariance, whereas EE and FC did not meet this criterion (Table 5, Panel A).

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Results of the MICOM assessment across the Chinese and Russian samples.

https://doi.org/10.1371/journal.pone.0357563.t005

The subsequent tests of construct means and variances showed no significant differences for PE, COL, and BI, indicating full measurement invariance for these constructs (Table 5, Panels B and C). SI, HM, UAI, and LTO achieved compositional invariance but showed at least one significant difference in their means or variances; these constructs therefore achieved partial measurement invariance. Measurement invariance was not established for EE and FC because they failed to meet the compositional invariance criterion. Accordingly, partial or full measurement invariance provided an adequate basis for the subsequent between-sample comparisons of structural relationships involving PE, SI, HM, COL, UAI, LTO, and BI. By contrast, comparisons involving EE or FC should be regarded as exploratory and interpreted with caution because compositional invariance was not established for these constructs.

Structural model and direct effects

Before testing the structural paths, the explanatory power, predictive relevance, and collinearity of the model were assessed. The R² values for BI were 0.777 in the Chinese sample and 0.786 in the Russian sample, with adjusted R² values of 0.767 and 0.777, respectively, indicating that the model explained a substantial proportion of the variance in acceptance intention in both samples. The corresponding Q² values were 0.597 and 0.590, both above zero, supporting the predictive relevance of the model (Table 6, Panel A). The VIF values ranged from 1.227 to 1.628 in the Chinese sample and from 1.302 to 1.930 in the Russian sample, indicating no substantial collinearity concerns (Table 6, Panel B). The structural model was therefore considered suitable for the subsequent assessment of the direct and moderating effects.

Download:

• PNG
  larger image
• TIFF
  original image

Table 6. Structural model assessment in the Chinese and Russian samples.

https://doi.org/10.1371/journal.pone.0357563.t006

All five direct paths were positive in both the Chinese and Russian samples, and none of the 95% confidence intervals included zero; therefore, H1a–H1e were supported in both samples (Table 7). In the Chinese sample, hedonic motivation had the largest path coefficient (β = 0.307, 95% CI [0.210, 0.382], f² = 0.270), representing a medium effect. In the Russian sample, facilitating conditions (β = 0.316, 95% CI [0.231, 0.398], f² = 0.282) and effort expectancy (β = 0.308, 95% CI [0.233, 0.387], f² = 0.310) had the largest path coefficients, both with medium effect sizes, while the remaining direct effects were small.

Download:

• PNG
  larger image
• TIFF
  original image

Table 7. Direct effects in the Chinese and Russian samples.

https://doi.org/10.1371/journal.pone.0357563.t007

Moderating effects and simple-slope analysis

All five interaction effects were positive in both samples, and none of their 95% confidence intervals included zero. H2a, H3a, H3b, H4a, and H4b were therefore supported in both samples (Table 8). In the Chinese sample, SI × COL, FC × UAI, and PE × LTO showed relatively larger effects, although their f² values remained small. In the Russian sample, FC × UAI showed the largest interaction effect (β = 0.221, 95% CI [0.145, 0.290], f² = 0.207), representing a medium effect, while the remaining interaction effects were small.

Download:

• PNG
  larger image
• TIFF
  original image

Table 8. Moderating effects of individual-level cultural values.

https://doi.org/10.1371/journal.pone.0357563.t008

Simple-slope analyses further clarified the interaction patterns. In the Chinese sample, the effects of SI, EE, FC, and PE on BI were not significant when the corresponding cultural value was low, but became significant at the mean level and increased further at the high level. The effect of HM on BI remained significant across all three levels of LTO. In the Russian sample, the effects of SI on BI and EE on BI were significant across all three moderator levels, whereas the effects of FC, PE, and HM on BI were not significant when the corresponding cultural value was low. At the mean and high levels of the moderators, all conditional effects were positive and significant (Table 9).

Download:

• PNG
  larger image
• TIFF
  original image

Table 9. Results of the simple-slope analyses.

https://doi.org/10.1371/journal.pone.0357563.t009

The corresponding interaction patterns are shown in Figs 2 and 3.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Simple-slope plots of the moderating effects in the Chinese sample.

https://doi.org/10.1371/journal.pone.0357563.g002

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Simple-slope plots of the moderating effects in the Russian sample.

https://doi.org/10.1371/journal.pone.0357563.g003

Multigroup analysis

Permutation-based multigroup analysis and PLS-MGA yielded consistent significance decisions. The effects of EE on BI and FC on BI were stronger in the Russian sample, whereas the effect of HM on BI was stronger in the Chinese sample. No significant between-sample differences were found for PE → BI or SI → BI (Table 10). Because compositional invariance was not established for EE and FC in the MICOM assessment, the between-sample differences involving these constructs should be interpreted cautiously. In contrast, compositional invariance was established for HM, providing a firmer measurement basis for the corresponding between-sample difference. None of the five interaction paths differed significantly between the Chinese and Russian samples (Table 10). Thus, although COL, UAI, and LTO strengthened their corresponding UTAUT2 relationships within both samples, the strength of these moderating effects did not differ significantly between the two samples.

Download:

• PNG
  larger image
• TIFF
  original image

Table 10. Multigroup comparison of direct and interaction effects.

https://doi.org/10.1371/journal.pone.0357563.t010

Contributions and implications

Theoretically, this study applies a context-adapted UTAUT2 model to museum digital services and examines mobile applications, online exhibitions, and AR/VR-based guides as forms of overall digital support. Rather than using nationality as a direct proxy for culture, it measures collectivism orientation, uncertainty avoidance, and long-term orientation at the individual level. This provides a more specific account of how cultural values shape visitors’ reliance on social cues, operating conditions, and experiential value. The study also combines MICOM with PLS-MGA to formally compare the two samples and distinguish numerical differences in path coefficients from statistically supported between-group differences. This approach strengthens the rigor of the cross-sample analysis and provides clearer evidence of both shared mechanisms and sample-specific patterns in museum digital service acceptance.

In practical terms, museums can optimize digital services according to visitors’ primary concerns. Greater interactivity, immersion, and narrative engagement may benefit visitors who place more value on experience. Clearer operating procedures and reliable devices, connectivity, and on-site assistance may be more important for visitors who are concerned with ease of use and support. Museums may also increase the social visibility of digital services through demonstrations, shared participation, and shareable experiences, while making their value for continued learning, cultural understanding, and extended engagement more explicit.

Limitations and future research

This study leaves several opportunities for further research. First, the samples were drawn from China and Russia, and different data collection approaches were used in the two countries. This may have affected sample comparability to some extent. Future studies could adopt more closely aligned survey procedures and include additional countries, regions, and museum types.

Second, EE and FC did not establish compositional invariance in Step 2 of MICOM. Between-group comparisons involving these constructs should therefore be interpreted with caution. Future research could refine the relevant measurement items and repeat the comparison after stronger measurement invariance has been established.

In addition, the study primarily relied on cross-sectional survey data and acceptance intention and examined different forms of museum digital services within an overall service context. Future research could incorporate actual usage data, longitudinal surveys, or on-site observation and examine specific digital service types separately.