Rethinking Trust in Scientists as a Network Model: A Global Analysis and Implications for Science Communication

Abstract

Trust in scientists is a communicative construct shaped by public perception of scientists. Using network psychometrics on responses from 71,922 individuals in 68 countries, we model trust as an interconnected community system of attitudes and beliefs. Sincerity emerges as the most central node, linking perceptions of competence, integrity, and goodwill. This dynamic structure challenges reflective latent models and emphasizes trust as a product of message framing and credibility cues. Our findings inform science communication theory by reframing trust as an emergent network and offer practical guidance for communicators seeking to design inclusive, trust-enhancing strategies across diverse sociocultural contexts.

Keywords

scientific trust network analysis interdisciplinary science communication public perception of scientists communication theory

Introduction

Public trust in scientists is essential for addressing societal challenges such as climate change, pandemics, and misinformation (Besley et al., 2021; Cologna et al., 2025). Yet, trust is not merely a psychological trait; it is fundamentally shaped through communication. Communication research has long emphasized that public perceptions of scientific credibility arise not only from the information transmitted but also from how scientists are perceived as communicators (McCroskey & Teven, 1999; Vraga et al., 2018). Traits such as competence, integrity, and goodwill (or caring) are central to source credibility, a concept rooted in decades of research on persuasion and trust.

It is important, however, to distinguish between trust, trustworthiness, and credibility. Although trust refers to a willingness to accept vulnerability in relation to others, trustworthiness denotes the qualities attributed to information sources, and credibility describes the persuasiveness of communicators. Building on recent work on epistemic trust (Hendriks et al., 2016; Wilholt, 2013), we emphasize that trusting scientists means not only assuming their benevolence or expertise but also relying on them as providers of reliable knowledge while remaining vigilant against the risk of misinformation.

Most existing measures of trust in scientists rely on reflective latent variable models that assume a universal and static structure (Nadelson et al., 2014; Wolff et al., 2024). These approaches often obscure how trust is contextually interpreted and communicatively constructed. Recent theoretical advances propose more dynamic frameworks in which personality and attitudes are viewed as a complex and ever-changing system of connected nodes (Borsboom et al., 2021). From a communication perspective, this suggests that trust should be examined not as a fixed trait but as a network of beliefs and attitudes influenced by message features, context, and audience interpretation.

In this study, we reconceptualize trust in scientists as an interrelated community network. We apply network psychometrics to global survey data (n = 71,922; 68 countries) to model how different perceptions, such as integrity, benevolence, competence, and openness, cohere into structured meaning systems. This approach responds to recent calls within the science communication literature for models that are both theoretically and practically applicable across diverse contexts (Lewenstein, 2024; Tayeebwa et al., 2022).

By identifying which beliefs are most central to the trust network, we aim to inform both theory and practice. Theoretically, by framing trust as a belief-based community network system, and practically, by guiding how communicators can design trust-enhancing strategies that resonate with public concerns across cultures. Our findings contribute to inclusive science communication by revealing structural variations in trust across cultural contexts, highlighting the need for locally adapted engagement strategies.

Theoretical Framework: Trust as a Communicative Community

Historically, models of reflective latent variables have been used to understand trust in scientists, conceptualizing it as a psychological construct composed of interrelated factors such as competence, benevolence, and integrity, inferred from patterns in individuals’ responses to survey items (Besley et al., 2021; McCroskey & Teven, 1999; Nadelson et al., 2014). While these models have provided important insights, they often miss how trust is expressed, understood, and shaped through communication, particularly within media and social discourse (Weingart & Guenther, 2016).

While some research has indeed treated trust as a relatively stable trait, an important body of scholarship has recognized trust as a more dynamic construct that varies based on the interplay of source characteristics, contextual factors, and recipient attributes (Wintterlin et al., 2022). This perspective acknowledges that trust is not fixed but rather emerges from communicative interactions and can shift depending on situational factors, message framing, and audience characteristics. Our network approach builds upon and extends this dynamic conceptualization by providing a systematic framework for modeling these fluid relationships and examining how different components of trust interact within specific cultural and communicative contexts.

In science communication, trust is increasingly viewed as a construct shaped by the framing of messages, the cues associated with the source, and the degree of audience involvement (Schröder et al., 2025). The source credibility model (McCroskey & Teven, 1999) suggests that perceptions of a communicator’s expertise, integrity, and goodwill influence the persuasiveness of their message. Newer models go a step further, focusing on how the traits of the communicator interact with how different audiences interpret them, especially in politically divided or skeptical contexts (Vraga et al., 2018).

Network theory presents a promising alternative for exploring these relationships. Instead of viewing trust as a fixed trait, network psychometrics frames it as a fluid system of linked attitudes and belief clusters, where each element affects and is affected by the others. This reflects the intricate relationship between individual perceptions and beliefs (Borsboom et al., 2021). The method has been successfully applied across diverse domains, including the validation of clinical instruments such as the Eating Disorder Examination Questionnaire (Laskowski et al., 2023), personality research examining the structure of openness to experience (Christensen et al., 2020), and cultural studies measuring cross-national differences in values and attitudes (De Benedictis et al., 2023). These examples demonstrate the versatility of network psychometric frameworks for uncovering community structures in complex psychological and social constructs. Within this framework, edges represent undirected partial correlations between beliefs, capturing their unique associations after conditioning on all other nodes. Importantly, these edges should not be interpreted as causal relations but as statistical dependencies that highlight potential pathways for further inquiry. This perspective allows researchers to identify which elements are most influential in shaping the network (Bavelas, 1948).

From the standpoint of science communication, this network perspective offers significant theoretical and practical advantages over traditional latent variable models. Theoretically, it provides a more nuanced framework for understanding trust as a dynamic system that emerges from communication and varies across situations by mapping the specific interconnections between individual attitudes and beliefs across theoretical dimensions. Unlike reflective models that treat trust dimensions as separate constructs connected only through factor correlations, network analysis reveals how specific perceptions may directly influence other trust components across dimensional boundaries. This network structure offers practical benefits for communicators.

By identifying the most central and interconnected nodes within the network, it highlights precise strategic points for intervention where targeted messaging could strengthen multiple trust dimensions simultaneously. In addition, by examining how these network structures vary across cultural contexts, this approach helps develop more culturally attuned and inclusive communication strategies that acknowledge the unique ways trust is constructed in different local settings, ultimately making cross-cultural scientific engagement more meaningful and effective (Putnick & Bornstein, 2016).

Method

Participants and Sampling

This study used data from a large-scale, preregistered, and pretested cross-sectional survey of 71,922 participants across 68 countries (Cologna et al., 2025). The survey was conducted between November 2022 and August 2023 as part of the TISP Many Labs project (“Trust in Science and Science-Related Populism”), an international, multidisciplinary consortium of 241 researchers from 179 institutions. The original research utilized a standardized questionnaire, translated into 37 languages, to investigate public trust in scientists and their role in society. Samples were weighted to reflect national distributions of age, gender, education, and population size. In most countries, data collection was conducted in collaboration with local research partners who ensured consistent application of the definition of “scientists” across cultural contexts to enhance cross-national comparability. Most respondents had higher education (59.0%), and 48.2% identified as female. The mean age was 43.5 years (SD = 15.1), with a median of 42.0 years and a range from 18 to 100. A full description of the sample for each country is provided in Supplemental Table S1.

Instrument

Trust in scientists was measured using a validated 12-item scale developed by Cologna et al. (2025), based on Besley et al.’s (2021) four-dimensional framework of competence, integrity, benevolence, and openness, commonly associated with communicator trustworthiness. For each item, participants provided their responses on a 5-point scale, ranging from 1 (very [inexpert]) to 5 (very [expert]). Each item was semantically pretested to ensure conceptual alignment across contexts. To maintain clarity and consistency throughout the manuscript, we will use abbreviated labels derived from core keywords listed in Table 2.

Data analysis

We adopted a network psychometric approach to examine how public trust in scientists is structured and how consistent this structure is across 68 countries (N = 71,922).

Our analysis began with Exploratory Graph Analysis (EGA), a method that identifies the number and composition of dimensions in multivariate data (Golino & Epskamp, 2017). EGA estimates a Gaussian graphical model (GGM), where each item is represented as a node and the edges represent partial correlations between items after controlling for all others. This means edges capture the unique associations among items. To detect communities (i.e., clusters of items that form dimensions), we applied three algorithms: Louvain, Leiden, and Walktrap (Christensen et al., 2023). The model with the lowest Total Entropy Fit Index (TEFI)¹ was selected as the best solution. To assess the stability and uncertainty of the TEFI values, we performed bootstrap analyses with 1,000 resamples for all models (theoretical and empirical), calculating 95% confidence intervals using percentile methods. This approach allows for robust comparison between models by quantifying the precision of the TEFI estimates and determining whether observed differences are statistically meaningful. To evaluate the stability of these solutions, we used a bootstrap procedure (Christensen & Golino, 2021), which repeatedly resamples the data to assess whether the same community structure reappears consistently.

This item-level approach is fundamental to EGA’s exploratory purpose because it allows the network structure to emerge without imposing predefined dimensions. By modeling partial correlations among individual items, we could test whether the public’s perceptions of trust in scientists followed the theorized dimensions or revealed alternative patterns. Accordingly, we analyzed raw item responses rather than composite mean scores, which would assume unidimensionality within each theoretical dimension and obscure potential cross-dimensional connections.

Next, we tested for measurement invariance across countries, following Dinić et al. (2025) and Jamison et al. (2024). Two levels of invariance were examined. Configural invariance was tested by examining whether the same community structure emerged consistently across all 68 countries through EGA with bootstrap validation. For each country, we verified that: (1) the number of dimensions, (2) the same items loaded primarily onto their theoretically expected communities, and (3) the community structure remained stable across bootstrap replications, with items replicating in their assigned communities at rates exceeding the 75% threshold recommended by Christensen and Golino (2021). Metric invariance tested whether the connections (edge weights) between items were comparable across cultures and it was assessed through permutation tests with 500 iterations, in which country labels were randomly shuffled and edge weights recalculated. A false discovery rate (FDR) correction (Benjamini & Hochberg, 1995) with α = 0.05 was applied across all 2,278 pairwise country comparisons.

¹TEFI is an information-theoretic index of model fit; lower values indicate a structure that more efficiently explains the observed associations (Golino et al., 2021).

Following the alignment approach (Muthén & Asparouhov, 2014), approximate metric invariance was considered supported when fewer than 25% of edge weights showed significant differences. We also examined partial metric invariance, requiring at least two invariant items per community across comparisons to ensure meaningful cross-country interpretation (Cieciuch & Davidov, 2016).

Finally, after confirming measurement invariance, we computed four centrality indices: strength, closeness, betweenness, and expected influence. Strength reflects the sum of the absolute values of a node’s edge weights, indicating its total connectivity. Expected influence is the sum of the raw edge weights (accounting for sign), capturing the net influence of a node, especially in networks with negative correlations. Closeness measures the inverse of the average shortest path to all other nodes, indicating how efficiently a node can reach or be reached by others. Betweenness captures how often a node serves as a bridge on the shortest paths between other items, reflecting its role as a connector (Opsahl et al., 2010; Rhemtulla et al., 2016).

To examine cross-country variability in the structural organization of trust, we summarized the distribution of each centrality index across the 68 national networks. We also estimated network loadings (Christensen et al., 2023) to quantify the strength of association between each item and its corresponding community. Conceptually, network loadings are analogous to factor loadings in traditional factor analysis: both quantify how strongly an item is associated with a given dimension. However, although factor loadings reflect the linear relation between an observed indicator and an underlying latent factor, network loadings are derived from the partial correlation structure of the network and represent how tightly an item is embedded within its community of interconnected nodes. Following the updated empirical guidelines provided by Christensen et al. (2023), revised network loadings of approximately 0.20, 0.35, and 0.50 can be interpreted as reflecting small, moderate, and large associations, respectively. These thresholds correspond roughly to simulated population loadings of 0.40, 0.55, and 0.70 in traditional factor models. To evaluate the stability of these associations, we averaged the absolute network loadings values by item and country, providing a measure of how consistently each indicator contributes to the trust structure across cultural contexts.

Together, these indices help identify which attitudes and beliefs about scientists (e.g., sincerity, competence, integrity) are most central, most bridging, or most efficient in shaping the overall perception of trust.

To account for survey design and sampling differences in the global data set, we repeated the analyses using a weighted correlation matrix, incorporating the survey weights at the item level before estimating the network. Country-level and sample-size weights were computed in the original data set (Cologna et al., 2025), and we utilized the pre-calculated global weight to construct the weighted correlation matrix. Results from the weighted analyses, including network structure, community composition, stability (bootstrap), and centrality indices, were highly similar to the unweighted analyses. For clarity, the main text reports the unweighted analyses.

All analyses were conducted in R 4.4.235 using the EGAnet package 2.3.036 for network estimation and invariance testing. Data and code are available at https://osf.io/uvknh

Results

Community Detection Comparison via TEFI

We applied three community detection algorithms (Louvain, Leiden, and Walktrap) to the trust in scientists’ network. Both Louvain and Leiden identified the same three communities (with sizes of 3, 3, and 6 items), and produced identical TEFI scores of −5.05. Walktrap also returned a three-community solution (with 3, 5, and 4 items), but its structure diverged slightly, grouping two items originally related to benevolence (“Concern” and “Improve”) into the integrity community and exhibiting poor model fit with a TEFI of −4.83, as lower TEFI values reflect better fit. Supplemental Figure S1 presents a comparison of the community structures identified by the Louvain/Leiden and Walktrap algorithms.

To further evaluate these empirical solutions, we compared them against three theoretical models using the TEFI, where lower values indicate a better fit. Bootstrap analysis with 5,000 resamples revealed distinct TEFI distributions for each model: The theoretical 2-community model (competence + character traits) produced the lowest TEFI score (−7.92, 95% CI = [−7.94, −7.90]). The theoretical 3-community model (competence; integrity, benevolence + openness) yielded a TEFI of −5.16 (95% CI = [−5.18, −5.13]), while the theoretical 4-community model yielded a positive TEFI (0.19, 95% CI = [0.13, 0.25]), indicating overfitting and unnecessary complexity relative to the data. The nonoverlapping confidence intervals confirm that these models represent statistically distinct levels of fit to the data. Table 1 summarizes the results.

Table 1.

TEFI Values With Bootstrap Confidence Intervals (1,000 Resamples).

Model	Type	N Communities	Mean TEFI	SD	95% CI	Description
Theoretical-2D	Theoretical	2	−7.92	0.01	[−7.94, −7.90]	2 broad factors
Theoretical-3D	Theoretical	3	−5.16	0.01	[−5.18, −5.13]	3 groups per literature model
Theoretical-4D	Theoretical	4	0.19	0.03	[0.13, 0.25]	4 constructs as originally formulated
Louvain	Empirical	3	−5.05	0.01	[−5.07, −5.03]	Merged constructs Benevolence + Openness
Leiden	Empirical	3	−5.05	0.02	[−5.08, −5.03]	Identical to Louvain
Walktrap	Empirical	3	−4.82	0.01	[−4.85, −4.80]	Merged but item 11 separated

Note: TEFI values calculated with 1,000 bootstrap resamples. Confidence intervals represent the 2.5th and 97.5th percentiles of the bootstrap distribution.

Stability of the Network Structure

To evaluate the stability of the community structure identified by the Leiden algorithm, we conducted a nonparametric bootstrap exploratory graph analysis (bootEGA) with 5,000 resamples, using the Leiden community detection algorithm. The bootstrapped solution confirmed the three-community structure, with excellent item stability across replications. Figure 1 shows the original network structure on the left and the replication frequencies for each item on the right. All items consistently clustered with their assigned community across all bootstrap iterations (replication = 1), indicating high structural stability.

Figure 1.

Network Structure of the Trust in Scientists Data Estimated With Exploratory Graph Analysis (EGA; Left) and Item Stability Via Bootstrap EGA (Right).

Considering the exceptionally high item stability observed in the bootstrap analysis using bootEGA(), along with the favorable TEFI scores, our findings offer strong empirical support for adopting the three-community network structure identified by both the Louvain and Leiden algorithms. These two methods produced the same community arrangement and identical TEFI values (−5.05), confirming their statistical robustness. We chose to move forward with the Leiden algorithm due to its additional theoretical and practical strengths, which ensure well-connected communities, enhance reproducibility across samples, and address known limitations of the Louvain method (Christensen et al., 2023). This decision is also consistent with our EGA() results, which indicate that benevolence and openness form a cohesive community that matches the “Goodwill” theoretical dimension. Goodwill refers to the perception that a communicator has the listener’s best interests at heart, encompassing traits such as empathy, caring, and responsiveness (McCroskey & Teven, 1999). Our empirical structure mirrors this conceptualization, with items such as “Concerned,” “Improve,” “Open,” and “Transparent” clustering together consistently.

Table 2 outlines the complete set of items and their respective community assignments, providing empirical clarity on how beliefs about scientists’ Competence, Integrity, and Goodwill systematically cluster within the trust network.

Table 2.

Items, Descriptions, and Community Assignments of the Trust in Scientists Scale.

Measurement Invariance Across Countries

To evaluate measurement invariance across countries, we conducted a multi-step analysis using the invariance() function from the EGAnet package, which employs the Leiden algorithm and Glasso model, following best practices outlined in recent literature (Jamison et al., 2024). Configural invariance was evaluated using bootEGA with 500 resamples. The median dimensionality was three (CI = 3–3), with structural consistency values of 1.00, 0.998, and 0.998 for the three dimensions, and average item stability exceeding 0.998. These results support a highly stable three-community solution at the global level. While country-level analyses revealed variability (35 countries with three communities, 32 with two, and one with four), these differences highlight the contextual nature of trust. We therefore interpret the global invariance results as evidence of a general structure, while also underscoring the importance of examining national divergences, as presented in the Supplementary Materials.

Next, the approximate metric invariance test, based on permutation testing of network loadings, also demonstrated support, with only 9% of item-level comparisons significant, well below the conventional 25% threshold (Muthén & Asparouhov, 2014). This indicates that most items contribute equally to their respective communities across groups. However, partial invariance was not supported, as no groups had at least two invariant items in each community (Cieciuch & Davidov, 2016), meaning some item-level differences still exist. Table 3 summarizes the results.

Table 3.

Summary of Measurement Invariance Tests.

Step	Criterion	Result	Notes
Configural Invariance	Same community structure across groups	Supported	Network structure is consistent across bootstrap resamples
Approximate	≤ 25% significant	Supported (9%	Indicates item
(Metric) Invariance	item-level loadings	significant)	contributions are generally equivalent
Partial Invariance	≥ 2 invariant items per group	Not Supported	No group met the minimum criterion

This pattern (support for configural and approximate invariance but failure of partial invariance) suggests that while the overall structure and relative item loadings are stable, minor item-specific deviations remain, preventing claims of partial metric invariance at the item level (Jamison et al., 2024). In practice, this means we can proceed with confidence in cross-country comparisons of community structures, but we should interpret the results with caution when evaluating specific item loadings. By selecting the best-performing algorithm and running item-level stability analyses for each country, we found that Competence is the most stable community, consistently retaining its original item structure (Expert, Qualified, and Intelligent) in 63 out of 68 countries. In contrast, the Integrity and Goodwill communities were preserved in only 23 and 5 countries, respectively. The exact structure identified in the global analysis was replicated only in Costa Rica, Indonesia, and Russia (see Summary S1).

Centrality Analysis of the Trust in Scientists’ Network

Using absolute scores for Betweenness, Closeness, and Strength (see Figure 2 for values by item), we identified the following hierarchy of item centrality in the network:

Figure 2.

Strength, Closeness, Betweenness, and Expected Influence Centrality in the Trust in Scientists Network.

Strength and Expected Influence

In the global network, strength and expected influence values were identical due to the absence of negative partial correlations. The Sincere item (“How sincere or insincere are most scientists?”) emerged as the most central node, with a strength value of 2.09, indicating the strongest direct connections to other items and suggesting a highly influential role in shaping public trust. Other influential items included Improve (0.81), Otherint (0.76), and Honest (0.51). Nodes such as Expert, Open, Intellig, and Otherviews showed relatively low strength (below −1.0), implying weaker direct associations. Country-level results (available in the Supplementary Materials) revealed only one case (Czech Republic) with negative partial correlations, all connected to the Honest node. All country-specific strength and expected influence values are provided in Supplementary Materials 1.

Closeness

Sincere again tops the list with 1.99, meaning it is, on average, the closest (via shortest paths) to all other nodes, reinforcing its central position. Honest (1.41) and Improve (0.64) also score above average. Open and Otherviews are farthest away (lowest closeness), highlighting their peripheral roles.

Betweenness

Sincere leads with 2.35, and Honest follows at 1.28, suggesting these items frequently occur on shortest paths, acting as bridges between other clusters. Nodes like Otherint (0.74) and Improve (0.20) play moderate intermediary roles. Ethical (1.14) and Open (0.87) appear less often on critical paths, indicating a lower bridging function.

Sincere is the most central node in this trust in science network. Its high strength (2.09) suggests that it has the strongest cumulative connections with other items, implying it is the most influential hub in activating the network. Its betweenness value of 2.35 indicates that it frequently serves as a bridge on the shortest paths connecting other items, suggesting it plays a pivotal connecting role within the network. In addition, its high closeness (1.99) indicates that it is, on average, closer to all other items, reinforcing that it can efficiently disseminate or absorb information within the network. Beyond demonstrating structural dimensionality, the networks reveal the centrality of sincerity as a bridging belief, suggesting it may be especially consequential for interventions aiming to strengthen trust.

Cross-Country Variability in Centrality and Network Loadings

To assess the cross-cultural stability of the trust network, we examined the variability of node centrality indices and network loadings across the 68 national samples (see Figure 3). The boxplots illustrate substantial consistency in the relative ordering of nodes, with Sincere showing the highest centrality across all measures and countries, followed by Improve, Honest, and Otherint. Median values for Sincere remained well above zero in all four indices, indicating that its prominence as the structural hub of the network is robust across cultural contexts. The dispersion of centrality values was greatest for Improve and Concerned, suggesting that these beliefs may play more context-dependent roles in national trust structures.

Figure 3.

Distribution of Node Centrality Indices and Network Loadings (EGL) Across 68 Countries.

The lower panel of Figure 3 summarizes the distribution of network loadings across countries, which quantify each item’s association with its community. The mean absolute network loading values ranged from approximately 0.20 to 0.35, corresponding to small-to-moderate associations (Christensen et al., 2023). Items such as Sincere and Qualified exhibited the strongest and most stable loadings, indicating tight integration within their respective communities, whereas Open and Otherviews showed weaker and more variable contributions. These results reinforce that while the global network structure of trust in scientists is broadly consistent, the relative strength of specific beliefs varies across countries, reflecting cultural nuances in how trust is cognitively organized.

Discussion

This study aimed to reconceptualize public trust in scientists as a dynamic and communicative network of attitudes and beliefs, rather than as a fixed psychological trait. Using network psychometrics and global survey data, we demonstrate that trust is structured around three interconnected communities (competence, integrity, and goodwill) that take shape differently depending on the cultural context. Most significantly, the belief that scientists are sincere stands out as the most central and influential element in the network, serving as a key link that connects the other trust components.

The network psychometric approach we employ here provides empirical tools for investigating these dynamic relationships while addressing several methodological limitations identified in traditional trust research. As Besley and Tiffany (2023) demonstrate in their analysis of direct trust measures, conventional approaches that ask respondents “how much do you trust scientists” often capture unclear and inconsistent constructs, with direct trust measures showing variable relationships to discrete trustworthiness perceptions across different studies and contexts. Their findings reveal that what researchers assume they are measuring when using generic trust questions may vary unpredictably between studies, making it difficult to build cumulative knowledge about trust dynamics.

Beyond the three- and four-dimensional models commonly discussed in the literature, recent work has proposed more fine-grained conceptualizations of epistemic trust. Reif et al. (2024), for instance, identified five dimensions (expertise, integrity, benevolence, transparency, and dialogue orientation), highlighting that trust in science involves overlapping moral, epistemic, and relational aspects. Our three-community network structure aligns with these broader frameworks while providing an empirically grounded synthesis: goodwill, for example, integrates benevolence, openness, and shared values that often converge empirically. This network perspective thus reveals how different theoretical dimensions dynamically coalesce within diverse cultural contexts.

Similarly, Hendriks et al. (2016) highlight how traditional reflective latent variable models assume universal and static structures that may not capture the contextual nature of epistemic trust. Their work on the Muenster Epistemic Trustworthiness Inventory (METI) demonstrates that while trustworthiness can be measured along dimensions of expertise, integrity, and benevolence, the relationships between these dimensions vary across contexts and populations, challenging assumptions of measurement invariance that underlie traditional psychometric approaches.

The network model addresses these limitations by conceptualizing trust as an interconnected system where the relationships between components (rather than their absolute levels) become the focus of analysis. This approach aligns with Wilholt’s (2013) observation that epistemic trust in science involves complex interdependencies between perceptions of ability, integrity, and benevolence that cannot be reduced to simple additive effects. By modeling trust as a network of partial correlations, we can identify patterns of statistical association between components (such as perceived sincerity and integrity) that may suggest potential pathways for future investigation, without implying causal or temporal direction. This methodological shift moves beyond the limitations of direct measures identified by Besley and Tiffany (2023) by focusing on structural relationships rather than problematic composite scores, while providing the theoretical flexibility needed to accommodate the cultural and contextual variations documented across trust research.

Our multialgorithm approach revealed distinct optimal solutions at different analytical levels. For the global network, Leiden provided the most theoretically coherent and empirically stable three-community structure. However, country-level analyses revealed substantial variation, with many national networks showing item stability below the 75% bootstrap replication threshold (Christensen & Golino, 2021), indicating limited reliability in community assignment. While the global model appeared highly stable, this likely reflects the very large combined sample size, which can smooth over meaningful local variability across countries and algorithms. This distinction underscores the importance of evaluating model stability in context and interpreting global-level consistency with caution, as it may mask heterogeneity in how trust structures manifest across cultural settings—while still preserving the theoretical coherence of the overarching competence–integrity–goodwill framework.

The country-level TEFI analysis revealed consistent evidence against the four-community theoretical model across diverse cultural contexts. Notably, the four-community structure consistently yielded the least optimal TEFI values in all countries, with the worst-performing case reaching a TEFI score of 1.96. This pattern strongly suggests that the four-community model provides an inferior fit to empirical data regardless of cultural setting. Our results further clarify how the empirically derived three-community structure relates to the dimensional models of trust discussed in the literature. Across countries, the most stable community corresponded to competence, while the remaining items clustered into two character-based dimensions that varied between two- and three-community solutions. This pattern parallels Besley et al.’s (2021) observation that trust can be meaningfully represented as either a two-dimensional (competence and character) or a higher-order model, depending on analytical goals. In our data, goodwill emerged from the empirical convergence of benevolence and openness, forming a distinct but flexible dimension within the broader character-based cluster. In contexts where goodwill and integrity were less distinguishable, two-dimensional models yielded lower entropy, suggesting that cultural variation may shape how moral and relational aspects of trust are organized. Rather than contradicting prior frameworks, this network structure provides an empirically grounded synthesis that captures both the parsimony of the 2D distinction and the contextual nuance of 3D or 4D conceptualizations.

Through centrality analyses, we identified scientists’ sincerity as the most influential node across all centrality indices, suggesting it serves as a structural and informational hub. Alongside honesty and items related to societal impact and openness, sincerity emerges as a key target for efforts aimed at strengthening public trust in scientists. Research has shown that perceived sincerity can buffer against public backlash during scientific controversies and can enhance compliance with health policies (Vraga et al., 2018). Therefore, campaigns aiming to increase public trust in science may benefit from messaging that emphasizes not only scientists’ sincerity but also their openness, honesty, and concern for societal wellbeing, qualities tightly linked to sincerity in the network. Although sincerity emerged as the most central node in the network, this centrality reflects its structural role in a system of partial correlations rather than a causal influence on other trust dimensions.

Conceptualizing public trust in scientists as a reflective latent variable imposes significant conceptual assumptions and practical limitations. For example, it assumes that scalar invariance holds across groups to compare mean trust levels. Recent studies that use this framework to compare mean levels of trust in scientists across countries proceed without first establishing scalar invariance (Cologna et al., 2025; Sturgis et al., 2021). Therefore, mean differences may instead reflect translation discrepancies, cultural interpretations, or variations in local scientific and cultural contexts (Sapir, 2022; Wolff et al., 2024). This oversight has important real-world implications. As several studies have noted, failing to account for noninvariance in cross-cultural self-report scales may result in divergent conclusions about the same construct across countries, thereby increasing the risk of invalid inferences (Fischer & Fontaine, 2011; Lacko et al., 2021; Putnick & Bornstein, 2016).

Our measurement invariance tests supported the three-community network structure across countries. Both configural and approximate metric invariance were achieved, with only 9% of item-level comparisons showing significant differences. This finding allows cautious cross-national comparisons of the overall network structure. However, partial metric invariance was not achieved, which means that community-specific item connections and loadings likely vary across countries. Our model captures global patterns in trust in scientists, but the finer details are likely shaped by national context. This highlights the importance of conducting detailed, country-specific analyses to fully understand how trust operates locally (Summary S1). It may be even more relevant to evaluate invariance within subgroups, such as gender, socioeconomic status, or age, to capture how trust dynamics vary within countries rather than assuming a uniform structure across all cultures.

This shift in conceptualizing trust in scientists is not merely methodological; it is fundamentally epistemological in nature. The network model is explanatory rather than predictive, seeking to uncover the structural interrelationships between observed indicators rather than reducing them to a latent cause. Within this paradigm, trust in scientists is reframed as a state-dependent system, where the transitions and stability of trust-related beliefs over time take precedence over mean trust scores measured at isolated points in time. Consequently, rather than categorizing populations as having “high” or “low” trust, the network approach enables monitoring of how trust varies dynamically across contexts and identifies key nodes that may serve as levers for change.

This approach also addresses a crucial measurement issue in psychological research: the isomorphism of the latent construct. Traditional psychometric models, particularly those based on Likert-type scales, typically assume a reflective latent factor model, in which each item serves as an indicator of a latent trait. Under such assumptions, composite scores (means or sums) are meaningful representations of an individual’s level of that trait. However, as Stevens (1946) highlighted, we should be concerned about the alignment between the observed empirical phenomenon and the numerical or categorical values used to represent it (Stevens, 1946).

Although both models (network and reflective) use the same Likert-type items, in the former case, the construct is represented by a matrix of partial correlations, reflecting the configuration of mutual influences rather than a single numerical score. Thus, the isomorphic mapping shifts from a single scalar score to a relational structure, capturing psychological attributes as emergent properties of dynamic systems. This perspective offers a more nuanced and context-sensitive representation of constructs like trust, which may vary in their organization across cultural and social contexts. Variations in item responses may reflect true differences in trust but also differences in cultural interpretation, response styles, or socio-political environments that shape the meaning of trust in distinct ways. Hence, noninvariance is not merely a psychometric limitation but can be informative and reveal important differences in construct conceptualization across groups (Putnick & Bornstein, 2016).

Given these complexities, the network model presents an appealing alternative for cross-cultural comparisons by focusing on structural differences in how trust is organized rather than assuming equivalence of mean scores. Tools such as the invariance() function in the EGAnet package allow comparisons of network structures between groups, identifying whether items cluster similarly. This approach enables meaningful cross-group comparisons even when traditional metric and scalar invariance fail, highlighting which aspects of the trust construct differ and which remain stable (Dinić et al., 2025). We were able to demonstrate only configural and approximate metric invariance of the network structure. Partial invariance was not achieved for any item. However, given that the data cover 68 countries, our findings offer a global perspective on trust in scientists.

Future studies could shift their focus away from scalar invariance testing and instead embrace the differences in how trust structures manifest across groups, contexts, and populations. Rather than forcing comparability, researchers can explore how unique structures emerge and use that information to design tailored interventions. For example, examining the trust network of a specific group (e.g., individuals with low scientific literacy) could inform campaigns that emphasize particular nodes, like integrity or transparency, that are central to that group’s trust.

Furthermore, using longitudinal designs leverages the full power of network analysis by enabling researchers to empirically derive the transition matrix of system states—that is, the latent variable dynamics that only become accessible through longitudinal network modeling. By measuring trust repeatedly, researchers can build probability matrices (similar to Markov networks) that track how people shift between different states, such as changing views about scientists’ sincerity or competence. These matrices are then visualized as directed, weighted networks, with each link showing the likelihood of moving from one state to another (Epskamp et al., 2018; Fried et al., 2020; Saqr et al., 2025). Longitudinal data also let researchers: (a) monitor how the entire network’s structure changes over time, (b) see which beliefs and attitudes become more or less central, and (c) test how outside events—like public health campaigns, scientific controversies, or breakthroughs—impact these transitions. For example, by applying inoculation strategies (Roozenbeek et al., 2022), it would be possible to test whether strengthening specific nodes (e.g., messages reinforcing scientific integrity) increases the overall stability of trust or makes it more resistant to misinformation.

To explore how targeted interventions might operate within a network model, one can apply the Network Intervention Analysis (NIA), a methodological framework developed in psychology to assess the effects of node-level changes within complex systems. Blanken et al. (2019) introduced NIA to investigate how interventions targeting specific symptoms in psychological networks can produce cascading effects (Blanken et al., 2019). For instance, their study found that cognitive-behavioral therapy for insomnia first alleviated sleep-related symptoms, which in turn led to reductions in depressive symptoms, demonstrating both the intervention point and the pathway of influence. Applying this framework to a trust in scientists’ network enables the design of strategically targeted communication efforts.

Simulating an intervention that enhances or suppresses a central node, such as Sincere, would allow one to estimate the downstream effects on related nodes, including honesty, openness, and competence. These counterfactual simulations offer predictive insight into whether increasing sincerity might enhance the coherence and resilience of public trust in scientists.

Evidence from clinical psychology supports the utility of such network-based simulations. For example, Bernstein et al. (2023) demonstrated that interventions targeting high-centrality cognitive beliefs produced significant improvements in body dysmorphic symptoms (Bernstein et al., 2023). Similarly, promoting sincere communication from scientists may generate a cascade of positive effects across the trust network, ultimately reinforcing public confidence. NIA provides a valuable tool for simulating these dynamics before implementing real-world interventions, helping ensure that communication strategies are both evidence-based and resource-efficient.

Finally, adopting a hierarchical network approach would enable the integration of trust in scientists with related constructs, such as trust in public institutions, the media, or the government, providing a more comprehensive view of societal trust systems and their interconnections. These constructs, when theoretically justified, could be modeled as higher-level clusters or interacting subsystems, revealing how broader institutional trust influences trust in science.

The study offers both conceptual and practical contributions to understanding and strengthening public trust in scientists across cultural contexts. Viewing trust in scientists as a dynamic, context-sensitive network of beliefs offers a richer, more adaptive framework for understanding public attitudes toward scientists. This approach challenges the assumptions made by reflective latent variable models, embraces contextual diversity, and opens new avenues for tailored interventions. Instead of pursuing invariance as a methodological goal in itself, researchers should treat structural differences as valuable insights, reflecting the many ways in which trust is constructed and experienced across cultural contexts.

While applied here to trust in individual scientists, the network methodology offers a generalizable framework for investigating the structure of trust and credibility across other domains and epistemic authorities, such as journalists, politicians, or medical professionals. This approach could also be extended to examine different reference objects of epistemic trust (e.g., scientific institutions, the scientific method itself) and their potential hierarchical relationships with trust in individual scientists using hierarchical network modeling (Jiménez et al., 2025). Future research could employ these analytical approaches to compare how the architecture of trust varies across these different targets, revealing both universal and domain-specific patterns in the formation of epistemic trust.

Supplemental Material

sj-docx-1-scx-10.1177_10755470251394188 – Supplemental material for Rethinking Trust in Scientists as a Network Model: A Global Analysis and Implications for Science Communication

Supplemental material, sj-docx-1-scx-10.1177_10755470251394188 for Rethinking Trust in Scientists as a Network Model: A Global Analysis and Implications for Science Communication by Luiz Gustavo de Almeida and Ronaldo Pilati in Science Communication

Footnotes

Acknowledgements

We would like to thank the TISP Many Labs project (“Trust in Science and Science-Related Populism”) for making their data and analyses openly available. Their commitment to transparency and open science has made this research possible.

ORCID iD

Luiz Gustavo de Almeida

Ethical Considerations

This study involved no new data collection. All analyses were conducted on the publicly available data set from . Ethical approval, data collection procedures, and study protocols for the original project were obtained and are fully documented by the TISP Many Labs initiative (“Trust in Science and Science-Related Populism”), with complete materials accessible through the project’s Open Science Framework repository. The data set was anonymized prior to public release.

Consent to Participate

Informed consent for participation in the original TISP Many Labs data collection was obtained by the project’s research teams and is documented in their publicly available materials. The data were released for reuse under a CC-BY 4.0 International license, which includes consent for secondary analyses.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

The data set underlying this article and the R code for replicating the analyses are publicly available at

Supplemental Material

Supplemental material for this article is available online at

Author Biographies

Luiz Gustavo de Almeida is a research associate at the Instituto Questão de Ciência and the Department of Work and Social Psychology in University of Brasilia, both in Brazil. He has a PhD in Microbiology from the University of São Paulo, and his main research interests are science communication, trust in science, and vaccine hesitance.

Ronaldo Pilati is a full professor of social psychology at the University of Brasília. His research interests include social cognition and experimental social psychology, focusing on the development of scientific thinking and public trust in science. He authored the scientific divulgation book “Science and Pseudoscience: Why We Only Believe What We Want to Believe.”

References

Bavelas

(1948). A Mathematical Model For Group Structures. Applied Anthropology, 7(3), 16–30. http://www.jstor.org/stable/44135428

Benjamini

Hochberg

(1995). Controlling the false discovery rate: A practical and powerful approach to multiple testing. Journal of the Royal Statistical Society Series B: Statistical Methodology, 57(1), 289–300. https://doi.org/10.1111/j.2517-6161.1995.tb02031.x

Bernstein

E. E.

Phillips

K. A.

Greenberg

J. L.

Curtiss

Hoeppner

S. S.

Wilhelm

(2023). Mechanisms of cognitive-behavioral therapy effects on symptoms of body dysmorphic disorder: A network intervention analysis. Psychological Medicine, 53(6), 2531–2539. https://doi.org/10.1017/S0033291721004451

Besley

J. C.

Lee

N. M.

Pressgrove

(2021). Reassessing the variables used to measure public perceptions of scientists. Science Communication, 43(1), 3–32. https://doi.org/10.1177/1075547020949547

Besley

J. C.

Tiffany

L. A.

(2023). What are you assessing when you measure “trust” in scientists with a direct measure? Public Understanding of Science, 32(6), 709–726. https://doi.org/10.1177/09636625231161302

Blanken

T. F.

Van Der Zweerde

Van Straten

Van Someren

E. J. W.

Borsboom

Lancee

(2019). Introducing network intervention analysis to investigate sequential, symptom-specific treatment effects: A demonstration in co-occurring insomnia and depression. Psychotherapy and Psychosomatics, 88(1), 52–54. https://doi.org/10.1159/000495045

Borsboom

Deserno

M. K.

Rhemtulla

Epskamp

Fried

E. I.

McNally

R. J.

Robinaugh

D. J.

Perugini

Dalege

Costantini

Isvoranu

A.-M.

Wysocki

A. C.

Van Borkulo

C. D.

Van Bork

Waldorp

L. J.

(2021). Network analysis of multivariate data in psychological science. Nature Reviews Methods Primers, 1(1), 58. https://doi.org/10.1038/s43586-021-00055-w

Christensen

A. P.

Garrido

L. E.

Guerra-Peña

Golino

(2023). Comparing community detection algorithms in psychometric networks: A Monte Carlo simulation. Behavior Research Methods, 56(3), 1485–1505. https://doi.org/10.3758/s13428-023-02106-4

Christensen

A. P.

Golino

(2021). Estimating the stability of psychological dimensions via bootstrap exploratory graph analysis: A Monte Carlo simulation and tutorial. Psych, 3(3), 479–500. https://doi.org/10.3390/psych3030032

10.

Christensen

A. P.

Golino

Silvia

P. J.

(2020). A psychometric network perspective on the validity and validation of personality trait questionnaires. European Journal of Personality, 34(6), 1095–1108. https://doi.org/10.1002/per.2265

11.

Cieciuch

Davidov

(2016). Establishing measurement invariance across online and offline samples. A tutorial with the software packages Amos and Mplus. Studia Psychologica, 15(2), 83. https://doi.org/10.21697/sp.2015.14.2.06

12.

Cologna

Mede

N. G.

Berger

Besley

Brick

Joubert

Maibach

E. W.

Mihelj

Oreskes

Schäfer

M. S.

Van Der Linden

Abdul Aziz

N. I.

Abdulsalam

Shamsi

N. A.

Aczel

Adinugroho

Alabrese

Aldoh

Alfano

Zwaan

R. A.

(2025). Trust in scientists and their role in society across 68 countries. Nature Human Behaviour. https://doi.org/10.1038/s41562-024-02090-5

13.

De Benedictis

Rondinelli

Vinciotti

. (2023). Cultures as networks of cultural traits: A unifying framework for measuring culture and cultural distances. Journal of the Royal Statistical Society: Series A (Statistics in Society), 186(3), 264–293. https://doi.org/10.1093/jrsssa/qnac007

14.

Dinić

B. M.

Costantini

Papageorgiou

K. A.

Bonfá-Araujo

Grabovac

Żemojtel-Piotrowska

Ankit Wertag

Kornienko

Tripathi

Kyriazos

Zhang

Poga

Tomašević

(2025). The Short Dark Triad across 14 cultures: A novel network-based invariance approach. Personality and Individual Differences, 246, 113321. https://doi.org/10.1016/j.paid.2025.113321

15.

Epskamp

Waldorp

L. J.

Mõttus

Borsboom

(2018). The Gaussian graphical model in cross-sectional and time-series data. Multivariate Behavioral Research, 53(4), 453–480. https://doi.org/10.1080/00273171.2018.1454823

16.

Fischer

Fontaine

J. R. J.

(2011). Methods for investigating structural equivalence. In Cross-cultural research methods in psychology (pp. 179–215). Cambridge University Press.

17.

Fried

E. I.

(n.d.). Studying mental health problems as systems, not syndromes.

18.

Fried

E. I.

von Stockert

Haslbeck

J. M. B.

Lamers

Schoevers

R. A.

Penninx

B. W. J. H.

(2020). Using network analysis to examine links between individual depressive symptoms, inflammatory markers, and covariates. Psychological Medicine, 50(16), 2682–2690. https://doi.org/10.1017/S0033291719002770

19.

Golino

Moulder

Shi

Christensen

A. P.

Garrido

L. E.

Nieto

M. D.

Nesselroade

Sadana

Thiyagarajan

J. A.

Boker

S. M.

(2021). Entropy fit indices: New fit measures for assessing the structure and dimensionality of multiple latent variables. Multivariate Behavioral Research, 56(6), 874–902. https://doi.org/10.1080/00273171.2020.1779642

20.

Golino

H. F.

Epskamp

(2017). Exploratory graph analysis: A new approach for estimating the number of dimensions in psychological research. PLOS ONE, 12(6), e0174035. https://doi.org/10.1371/journal.pone.0174035

21.

Hendriks

Kienhues

Bromme

(2016). Trust in science and the science of trust. In: Blöbaum

(eds) Trust and communication in a digitized world. Progress in IS. Springer.https://doi.org/10.1007/978-3-319-28059-2_8

22.

Jamison

Christensen

A. P.

Golino

H. F.

(2024). Metric invariance in exploratory graph analysis via permutation testing. Methodology, 20(2), Article 2. https://doi.org/10.5964/meth.12877

23.

Jiménez

Abad

F. J.

Garcia-Garzon

Golino

Christensen

A. P.

Garrido

L. E.

(2025). Dimensionality assessment in bifactor structures with multiple general factors: A network psychometrics approach. Psychological Methods, 30(4), 770–792. https://doi.org/10.1037/met0000590

24.

Lacko

Cenek

Točík

Avsec

Đorđević

Genc

Haka

Sakotic-Kurbalija

Mohorić

Neziri

Subotic

(2021). The necessity of testing measurement invariance in cross- cultural research: On potential bias in cross-cultural comparisons with individualism-collectivism self-report scales. https://doi.org/10.31234/osf.io/9ytj3

25.

Laskowski

N. M.

Halbeisen

Braks

Huber

T. J.

Paslakis

(2023). Exploratory graph analysis (EGA) of the dimensional structure of the eating disorder examination-questionnaire (EDE-Q) in women with eating disorders. Journal of Psychiatric Research, 163, 254–261. https://doi.org/10.1016/j.jpsychires.2023.05.063

26.

Lewenstein

B. V.

(2024). Science communication in a diverse world. Journal of Science Communication, 23(05). https://doi.org/10.22323/2.23050401

27.

McCroskey

J. C.

Teven

J. J.

(1999). Goodwill: A reexamination of the construct and its measurement. Communication Monographs, 66(1), 90–103. https://doi.org/10.1080/03637759909376464

28.

Muthén

Asparouhov

(2014). IRT studies of many groups: The alignment method. Frontiers in Psychology, 5. https://doi.org/10.3389/fpsyg.2014.00978

29.

Nadelson

Jorcyk

Yang

Jarratt Smith

Matson

Cornell

Husting

(2014). I just don’t trust them: The development and validation of an assessment instrument to measure trust in science and scientists. School Science and Mathematics, 114(2), 76–86. https://doi.org/10.1111/ssm.12051

30.

Opsahl

Agneessens

Skvoretz

(2010). Node centrality in weighted networks: Generalizing degree and shortest paths. Social Networks, 32(3), 245–251. https://doi.org/10.1016/j.socnet.2010.03.006

31.

Putnick

D. L.

Bornstein

M. H.

(2016). Measurement invariance conventions and reporting: The state of the art and future directions for psychological research. Developmental Review, 41, 71–90. https://doi.org/10.1016/j.dr.2016.06.004

32.

Reif

Taddicken

Guenther

Schröder

J. T.

Weingart

(2024). The Public Trust in Science Scale: A multilevel and multidimensional approach. Science Communication, 47(5), 670–701. https://doi.org/10.1177/10755470241302758

33.

Rhemtulla

Fried

E. I.

Aggen

S. H.

Tuerlinckx

Kendler

K. S.

Borsboom

(2016). Network analysis of substance abuse and dependence symptoms. Drug and Alcohol Dependence, 161, 230–237. https://doi.org/10.1016/j.drugalcdep.2016.02.005

34.

Roozenbeek

van der Linden

Goldberg

Rathje

Lewandowsky

(2022). Psychological inoculation improves resilience against misinformation on social media. Science Advances, 8(34), eabo6254. https://doi.org/10.1126/sciadv.abo6254

35.

Sapir

E. V.

(2022). Can we trust measures of trust? Measurement invariance in trust in EU news media. Sn Social Sciences, 2(10), 226. https://doi.org/10.1007/s43545-022-00534-1

36.

Saqr

López-Pernas

Törmänen

Kaliisa

Misiejuk

Tikka

(2025). Transition network analysis: A novel framework for modeling, visualizing, and identifying the temporal patterns of learners and learning processes (No. arXiv:. 2411.15486). arXiv. https://doi.org/10.48550/arXiv.2411.15486

37.

Schröder

J. T.

Brück

Guenther

(2025). Identifying trust cues: How trust in science is mediated in content about science. Journal of Science Communication, 24(1). https://doi.org/10.22323/2.24010206

38.

Stevens

S. S.

(1946). On the theory of scales of measurement. Science, New Series, 103(2684), 677–680.

39.

Sturgis

Brunton-Smith

Jackson

(2021). Trust in science, social consensus and vaccine confidence. Nature Human Behaviour, 5(11), 1528–1534. https://doi.org/10.1038/s41562-021-01115-7

40.

Tayeebwa

D. W.

Wendo

D. C.

Nakiwala

D. A. S.

(2022). Theories and models of science communication. In Science communication skills for journalists (pp. 14–22). CABI. https://doi.org/10.1079/9781789249675.0002

41.

Vraga

Myers

Kotcher

Beall

Maibach

(2018). Scientific risk communication about controversial issues influences public perceptions of scientists’ political orientations and credibility. Royal Society Open Science, 5(2), 170505. https://doi.org/10.1098/rsos.170505

42.

Weingart

Guenther

(2016). Science communication and the issue of trust. Journal of Science Communication, 15(05), C01. https://doi.org/10.22323/2.15050301

43.

Wilholt

(2013). Epistemic trust in science. The British Journal for the Philosophy of Science, 64(2), 233–253. https://doi.org/10.1093/bjps/axs007

44.

Wintterlin

Hendriks

Mede

N. G.

Bromme

Metag

Schäfer

M. S.

(2022). Predicting public trust in science: The role of basic orientations toward science, perceived trustworthiness of scientists, and experiences with science. Frontiers in Communication, 6, 822757. https://doi.org/10.3389/fcomm.2021.822757

45.

Wolff

S. M.

Breakwell

G. M.

Wright

D. B.

(2024). Psychometric evaluation of the Trust in Science and Scientists Scale. Royal Society Open Science, 11(4), 231228. https://doi.org/10.1098/rsos.231228

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

30.28 MB