Temporal baselines alter cross-platform estimates of brand search attention during sporting events

1 Introduction

Major sporting events are high-salience communication environments in which brands compete for limited consumer attention. Sponsorship, athlete endorsement, broadcast visibility, and category relevance can increase awareness, contribute to customer-based brand equity, and shape consumer responses, but the realized effect depends on how event-linked information reaches audiences (Cornwell et al., 2005; Cornwell and Kwon, 2020; Keller, 1993; Meenaghan, 2001). In digital markets, this process is no longer confined to a single channel. Consumers may encounter a brand in a search engine, within a short-video ecosystem, through social media engagement, or across several environments during the same event. Treating digital attention as a platform-neutral outcome therefore risks concealing differences in how communication is encountered, interpreted, and subsequently searched.

Platform differentiation is well established in advertising and digital-consumer research. Digital environments shape what information is visible, how users interact with it, and which behavioral traces are recorded (Stephen, 2016). Social-media engagement and advertising response vary by platform type because platforms afford different user goals, content forms, and interaction patterns (Voorveld et al., 2018). Content characteristics also shape attention and engagement within broadcasts and social media environments (Herold and Breuer, 2023; Lee et al., 2018; Li and Xie, 2020). Sport marketing research has reached a similar conclusion: online technologies have transformed sponsorship into a dynamic content ecosystem, and social-media engagement returns depend on platform characteristics rather than exposure alone (Koronios et al., 2020; Naraine et al., 2022). Reviews of social media in sport likewise emphasize that platform-specific behavior should be treated as a substantive research object rather than a measurement nuisance (Rai et al., 2025).

Search indices are attractive in this setting because they record behavioral attention at high frequency and across geographic units. Search volume has been used as a direct proxy for investor attention (Da et al., 2011) and as a predictor of consumer and economic activity (Choi and Varian, 2012; Goel et al., 2010). Mobile search data have also been used to examine temporal and provincial variation in advertising response (Goh et al., 2015), while Baidu data have been used to measure provincial tourism demand in China (Ma et al., 2022). However, the use of multiple proprietary indices creates a measurement problem. Platform providers normalize their indices using platform-specific procedures, and the scale of one index is not commensurate with the scale of another. A change in the gap between two indices may reflect event-linked behavior, secular platform growth, changes in platform measurement, or a combination of these forces.

This problem is particularly important when an event occurs only once in the observed period. A full-sample fixed-effects model may compare an event window with months that are far away in time and located on a different part of a platform-specific trend. A local event window, by contrast, compares the same event with adjacent months but uses less information. The choice is not innocuous. If the relative level of two platform indices changes gradually over time, event coefficients can be sensitive to the temporal baseline even when cross-sectional sample size is large. In addition, event indicators vary at the month–platform level. Standard errors clustered only across brand–province units may therefore overstate precision by treating common platform-month shocks as if they were independent across 186 panels, a familiar concern in event-style panel designs with serially correlated outcomes (Bertrand et al., 2004; Cameron et al., 2011).

The present study reanalyzes a provincial panel of brand-search indices for six major sportswear brands on Baidu and Douyin from June 2022 through December 2025. Rather than assuming that Baidu records active attention while Douyin records passive exposure, we treat both outcomes as platform-specific search indices. Baidu represents search within a general-purpose search engine; Douyin represents search within a short-video content ecosystem. The analysis focuses on two high-salience events, the 2023 FIBA Basketball World Cup and the 2024 Paris Olympics. We ask three questions. First, do full-sample fixed-effects estimates indicate different cross-platform attention patterns during the two events when all observations, including zero-valued indices, are retained? Second, are those patterns robust to dependence-aware inference and alternative platform-specific temporal specifications? Third, are the full-sample patterns broadly shared across brands, and what do separate Baidu and Douyin analyses suggest about the source of the cross-platform gap?

The study contributes to advertising and marketing communication in three ways. First, it reframes cross-platform attention as a measurement and identification problem, not simply a comparison of raw platform levels. Second, it shows that transparent zero handling preserves the full sample without changing the substantive full-sample pattern, while clustering, temporal specification, and event-window choice can materially alter the interpretation of event-linked brand-search estimates. Third, it adds brand-level and platform-specific evidence showing that cross-sectional consistency across brands does not by itself establish a stable event effect when the underlying platform relationship evolves over time. The resulting contribution is deliberately methodological. The data reveal contrasting full-sample patterns, but they do not justify a stable causal claim about event type, platform mechanism, or downstream commercial outcomes.

2 Materials and methods

2.1 Data and sample

The dataset contains monthly provincial brand-search indices for Anta, Li-Ning, Nike, Adidas, Xtep, and 361 Degrees. The sample covers all 31 provincial-level administrative units of mainland China from June 2022 through December 2025. For each brand–province–month combination, an index is observed for Baidu and Douyin. The balanced design contains 6 brands×31 provinces×43 months×2 platforms = 15,996 observations.

The original daily platform values were collected using the Chinese brand queries 安踏, 李宁, 耐克, 阿迪达斯, 特步, and 361度 and aggregated to monthly arithmetic means. The two source systems were Baidu Index and Ocean Engine Trend Insight/Douyin Index. The platform indices are proprietary metrics. Their internal scaling procedures are not fully disclosed, and each platform normalizes its index independently. Consequently, raw Baidu and Douyin levels are not interpreted as directly comparable quantities of attention.

Twenty-nine observations have a recorded Baidu index of zero; no Douyin observation is zero. To retain the complete balanced panel and avoid selection induced by dropping zero-valued observations, the revised primary outcome is log(1 + index), yielding all 15,996 observations. For positive index values, this transformation behaves similarly to the natural logarithm while remaining defined at zero. As a robustness check, we also reproduce the earlier positive-only specification using ln(index), which excludes the 29 zero-valued Baidu observations and contains 15,967 observations. We do not replace zeros with an arbitrary positive constant. The panel contains no duplicate brand–province–month–platform records and no missing raw index values.

2.2 Platform measures and event windows

The primary dependent variable is log(1 + platform-specific brand-search index). The transformation reduces skewness, retains zero values, and emphasizes within-series changes rather than raw cross-platform levels. Because Baidu and Douyin normalize their indices independently, neither the platform main effect nor the absolute magnitude of a cross-platform gap is interpreted as a common quantity of consumer attention. Event interactions are therefore interpreted only as changes in Douyin relative to Baidu within the transformed index framework.

The Paris Olympics indicator equals 1 in July and August 2024. The Games took place from 26 July to 11 August, so these two monthly observations cover the event and its immediate monthly aggregation. The Basketball World Cup indicator equals 1 in August and September 2023, covering the event held from 25 August to 10 September. Promotional-month controls equal 1 in January, February, June, and November, and a year-end control equals 1 in December. Because month fixed effects absorb all shocks common to both platforms, these controls enter through interactions with the Douyin indicator.

The monthly resolution prevents identification of day-level reactions. This is particularly consequential for sporting events that begin late in one month and end early in the next: monthly averaging can smooth short-lived peaks, combine pre-event and event days within the same observation, and obscure the precise timing of attention responses. Event indicators should therefore be interpreted as coarse period markers rather than precise treatment dates. The two events also differ in scale, media coverage, brand activation, team performance, and other unobserved features. No variable in the dataset directly measures emotional content, information intensity, advertising spending, algorithmic exposure, or user motivation. The analysis therefore does not label the events as causal representatives of broader event types.

2.3 Fixed-effects specification and inference

The primary stacked-panel model is specified in Equation (1):

log(1+Search_bpmt)=β_1(Olympics_m×Douyin_t)+β_2(WorldCup_m×Douyin_t)+β_3(Promo_m×Douyin_t)+β_4(YearEnd_m×Douyin_t)+α_bpt+δ_m+ε_bpmt.(1)

Here b indexes brand, p province, m month, and t platform. Brand–province–platform fixed effects α(b,p,t) absorb the stable level of each brand–province series on each platform. Month fixed effects δ(m) absorb macroeconomic conditions, national seasonality, and event main effects common to both platforms. The coefficients β₁ and β₂ therefore measure changes in the transformed Douyin–Baidu platform gap during the designated event months relative to the remaining sample, conditional on the controls. They do not identify an absolute increase on Douyin, a common-unit difference between platforms, or an effect on sales or brand performance.

The brand–province–platform fixed effects are more restrictive than brand–province fixed effects because they allow every brand–province unit to have a different baseline on each platform. The model is estimated by within transformation. We report two inferential approaches. The first clusters standard errors by brand–province, matching the cross-sectional panel structure. The second uses two-way clustering by brand–province and month (Cameron et al., 2011). Month clustering is important because the event interactions vary at the platform–month level and platform-specific shocks are shared across brands and provinces. Statistical conclusions emphasize the two-way clustered results.

2.4 Outcome sensitivity and alternative temporal specifications

We use several complementary analyses to assess whether the event coefficients depend on outcome construction, scale, or temporal specification. First, the earlier positive-only ln(index) model is retained as a zero-handling robustness check. Second, log(1 + index) is standardized within each brand–province–platform series to a mean of zero and standard deviation of one, expressing coefficients in within-series standard-deviation units.

Third, we estimate alternative platform-time specifications rather than treating trend controls as mere diagnostics. These include a Douyin-specific linear trend, a Douyin-specific quadratic trend, a linear trend combined with month-of-year interactions, and a flexible model with platform-specific year and month-of-year effects. These specifications represent plausible alternative descriptions of the strong secular movement in the cross-platform gap. Material changes in sign or magnitude are therefore interpreted as evidence that a stable event effect is not separately identified from the evolving platform relationship.

Fourth, we construct a descriptive paired gap for each brand–province–month, G(b,p,m) = log[1 + Douyin(b,p,m)]−log[1 + Baidu(b,p,m)], and compare the average gap in each two-month event window with the immediately preceding two months and the immediately following two months. These local contrasts are descriptive because only two monthly observations define each event period, but they show whether the full-sample sign is also visible near the event itself.

2.5 Brand-level robustness and platform-specific decomposition

To assess whether the aggregate pattern is driven by particular brands, we follow the reviewer-suggested brand-level robustness design. For each brand b and month m, we average log(1 + index) across provinces separately by platform and construct Y(b,m) = mean across provinces[log(1 + Douyin)]−mean across provinces[log(1 + Baidu)]. For each event, a simple descriptive regression is estimated separately for each brand, as shown in Equation (2):Here D(m) identifies the corresponding event period. Because each brand series contains only 43 monthly observations, these coefficients are used as descriptive robustness evidence rather than as a replacement for the stacked panel. We additionally examine Baidu and Douyin separately in auxiliary regressions and local event-window contrasts. Exact month fixed effects cannot be included in a single-platform event regression because the event dummy is defined at the month level and would be absorbed. The separate-platform results are therefore interpreted only as decompositions of the observed cross-platform pattern, not as causal estimates.

Y_bm=α_b+β_bD_m+u_bm.(2)

2.6 Use of generative AI-assisted tools

OpenAI ChatGPT (GPT-5.6 Thinking and GPT-5.6 Sol models; OpenAI, San Francisco, CA, United States; accessed July–August 2026) was used to assist with language editing, structural reorganization, statistical code checking, document formatting, and preparation of the revision. The authors reviewed and verified all analyses, factual statements, citations, and conclusions and take full responsibility for the manuscript.

3 Results

3.1 Descriptive platform dynamics

Table 1 summarizes the raw platform indices. The Baidu index has a mean of 215.684 and a median of 178.270, whereas the Douyin index has a mean of 10,059.463 and a median of 5,340.870. These level differences are not substantively interpretable because each platform uses its own normalization. They do, however, illustrate why the analysis must focus on within-platform changes rather than raw cross-platform levels.

Table 1. Descriptive statistics of the platform indices.

Indices are normalized within platforms and are not comparable in level. Twenty-nine Baidu observations equal zero.

Figure 1 plots the monthly mean paired gap, log(1 + Douyin)−log(1 + Baidu). The gap rises from approximately 1.57 in June 2022 to 4.01 in December 2025, a secular movement that is large relative to the event coefficients. Around the Basketball World Cup, the gap is approximately 2.50 in August 2023, 3.09 in September, and 3.63 in October. Around the Paris Olympics, it is 3.76 in June 2024, 3.59 in July, 3.55 in August, and 3.68 in September. The descriptive trajectory therefore suggests a pronounced evolving mean rather than a visually stable cross-platform difference around which isolated event deviations occur. We do not infer stationarity, cointegration, or structural breaks from the graph alone. Given the 43 monthly observations, we do not treat formal cointegration or structural-break analysis as necessary for the present revision; instead, Figure 1 and the brand-level evidence below provide descriptive evidence on the evolving relationship.

3.2 Full-sample fixed-effects estimates

Table 2 presents the revised primary fixed-effects estimates using log(1 + index) and all 15,996 observations. The Paris Olympics×Douyin coefficient is 0.445, while the Basketball World Cup×Douyin coefficient is −0.326. These coefficients describe changes in the transformed Douyin–Baidu gap relative to the full-sample baseline; they are not interpreted as common-unit cross-platform percentage effects. When errors are clustered only by brand–province, both coefficients are highly significant (p < .001).

Table 2. Primary fixed-effects estimates using log(1 + index).

Dependent variable: log(1 + platform search index). All 15,996 observations are retained. All models include brand–province–platform and month fixed effects. Two-way clustering is by brand–province and month; two-way p-values use a t reference with 42 degrees of freedom.

The apparent precision is much lower under two-way clustering. The two-way standard error for the Olympic estimate is 0.186, producing p = .021, whereas the standard error for the Basketball World Cup estimate is 0.282, producing p = .254. The promotional and year-end interactions also lose statistical significance under two-way clustering. The result reinforces that nearly 16,000 panel observations do not imply nearly 16,000 independent event shocks when the identifying variation is concentrated in a small number of common months.

3.3 Zero handling and scale robustness

Table 3 compares the revised primary outcome with the earlier positive-only logarithmic specification. Dropping the 29 zero-valued Baidu observations and using ln(index) produces an Olympic coefficient of 0.444 and a Basketball World Cup coefficient of −0.322, compared with 0.445 and −0.326 under log(1 + index). The corresponding two-way p-values are likewise similar. The zero-valued observations therefore do not drive the full-sample point estimates, but retaining them avoids an unnecessary selection rule and is methodologically preferable.

Table 3. Outcome construction and zero-handling robustness.

All standard errors are two-way clustered by brand–province and month. The positive-only model excludes 29 zero-valued Baidu observations; the revised primary model retains them through log(1 + index).

Within-series standardization of log(1 + index) preserves the point-estimate directions but not conventional significance: the Olympic coefficient is 0.486 standard deviations (two-way p = .261) and the Basketball World Cup coefficient is −0.473 (p = .225). This result again shows that cross-sectional consistency does not resolve uncertainty created by a small number of common time shocks.

3.4 Sensitivity to alternative temporal specifications

Table 4 and Figure 2 report the alternative temporal specifications. Adding a platform-specific linear trend reduces the Olympic estimate from 0.445 to 0.140 and changes the Basketball World Cup estimate from −0.326 to 0.011. A quadratic platform trend produces estimates of −0.163 and −0.251. A model with platform-specific seasonality and a linear trend produces estimates of 0.433 and 0.114, while the flexible year-and-seasonality model yields 0.054 and −0.106. The Olympic estimate changes sign in the quadratic specification, and the Basketball World Cup estimate changes sign in two specifications. These are substantively different model implications, not minor diagnostic fluctuations around a stable event effect.

Table 4. Sensitivity of event coefficients to alternative temporal specifications.

Dependent variable: log(1 + index). All standard errors are two-way clustered by brand–province and month. Specifications differ only in the platform-specific temporal controls described in the Methods. These are alternative model specifications rather than diagnostic checks.

The event-centered local contrasts in Table 5 reinforce this conclusion. For the Basketball World Cup, the average transformed platform gap is 2.496 in June–July 2023 and 2.797 in August–September, an event-minus-pre-event contrast of +0.301. For the Paris Olympics, the average gap is 3.705 in May–June 2024 and 3.568 in July–August, a contrast of −0.138. Both local signs are opposite to the corresponding full-sample coefficients. The post-event gaps are 3.663 after the Basketball World Cup and 3.740 after the Olympics, suggesting that the observed movements cannot be reduced to a stable contemporaneous event effect.

Table 5. Event-centered descriptive platform-gap contrasts.

Gap = mean[log(1 + Douyin index)−log(1 + Baidu index)]. Pre-event, event, and post-event periods each contain two consecutive months. These are descriptive contrasts, not causal estimates.

3.5 Brand-level robustness and platform-specific evidence

Figure 3 implements the reviewer-suggested brand-level gap analysis. In separate full-sample event-period regressions, all six brands have a positive Olympic coefficient, ranging from 0.147 for 361 Degrees to 0.723 for Adidas, and all six have a negative Basketball World Cup coefficient, ranging from −0.089 for Xtep to −0.953 for Adidas. The aggregate full-sample sign pattern is therefore not generated by a single brand or a narrow subset of brands.

However, the same brand-level data also reproduce the temporal-baseline reversal. In the two-month local comparison, all six brands have a positive Basketball World Cup event-minus-pre-event gap, whereas five of six brands have a negative Paris Olympics event-minus-pre-event gap. Thus, cross-sectional sign consistency across brands coexists with temporal instability within brands. Supplementary Table S1 reports the brand-level coefficients and local contrasts.

Auxiliary platform-specific analyses provide additional decomposition rather than causal identification. In simple full-sample event-period regressions, the aggregated Baidu series has full-sample event-period coefficients of +0.028 for the Olympics and +0.105 for the Basketball World Cup, whereas the corresponding Douyin coefficients are +0.423 and −0.308. Brand-level Douyin coefficients share the same full-sample sign across all six brands for both events, while Baidu coefficients are more heterogeneous. Models with brand–province fixed effects plus year and month-of-year controls yield the same qualitative message that platform-specific patterns differ, but exact month fixed effects cannot be retained in these separate-platform regressions. We therefore report them as descriptive decomposition in Supplementary Tables S2, S3 rather than as evidence of a platform mechanism.

4 Discussion

4.1 Main findings

The full-sample fixed-effects model reproduces a visually compelling cross-platform reversal: the Paris Olympics is associated with a higher Douyin–Baidu transformed gap, whereas the Basketball World Cup is associated with a lower gap. That pattern is virtually unchanged by retaining zero observations with log(1 + index), and the full-sample sign is shared by all six brands. These findings rule out two narrow explanations: the reversal is not created by the 29 zero-valued Baidu observations, and it is not driven by a single brand.

Neither result, however, establishes a stable event effect. Two-way clustering substantially widens uncertainty; alternative temporal specifications change magnitudes and signs; and local event-centered comparisons reverse the full-sample pattern for every brand during the Basketball World Cup and for five of six brands during the Olympics. The most defensible conclusion is therefore methodological: cross-platform event estimates based on proprietary search indices can be simultaneously cross-sectionally consistent and temporally unstable. A large panel can document the former while still providing weak identification of the latter because the event treatment varies in only a few common months.

4.2 Implications for advertising and marketing communication theory

The findings reinforce the view that digital platforms are not interchangeable communication environments, but the present data do not identify why. Platform heterogeneity has at least two meanings. Substantive heterogeneity refers to genuine differences in user behavior, exposure, and communication processes across platforms. Measurement heterogeneity refers to differences in proprietary index scaling, normalization, updating, or coverage. The observed Baidu–Douyin divergence is compatible with either explanation, and the current design cannot empirically separate them.

This distinction places a hard boundary on mechanism claims. Baidu search occurs in a general-purpose query environment, whereas Douyin search occurs inside a short-video content ecosystem. Douyin's recommendation architecture may stimulate subsequent search, but the available variable is still a search index, not a direct measure of recommendation exposure, passive attention, emotional arousal, algorithmic delivery, or advertising contact. The auxiliary platform-specific regressions show different patterns across the two indices, but they cannot determine whether those differences are behavioral or measurement-based. Future research should combine search indices with content-level exposure, advertising spend, video engagement, user-level clickstream measures, or a common external outcome.

4.3 Methodological implications

Five practices follow from the analysis. First, researchers should avoid interpreting raw cross-platform index levels, because logarithmic transformation does not make independently normalized scales commensurate. Second, zero handling should be explicit and should preserve observations when a well-defined transformation such as log(1 + index) is available. Third, platform-specific temporal trajectories should be visualized before event models are estimated; Figure 1 reveals a long-run movement in the gap that is too large to treat as a minor nuisance.

Fourth, standard errors should reflect the level at which event variation occurs. Clustering by thousands of geographic or brand observations is insufficient when the identifying variation is shared by a small number of months. Time clustering, two-way clustering, randomization inference over event dates, or other dependence-aware procedures should be considered. Fifth, full-sample estimates should be accompanied by local event windows, brand-level checks, and alternative platform-time specifications. Agreement across these approaches is evidence of stability; disagreement is itself an empirical result that should narrow the claim rather than be described as a routine diagnostic.

These recommendations extend beyond sport. Product launches, celebrity endorsements, viral campaigns, sales festivals, and public controversies are frequently evaluated with Google Trends, Baidu Index, social-media search, and proprietary engagement metrics. In each case, a platform-specific growth trend can be mistaken for an event effect if the baseline is remote or poorly modeled.

4.4 Practical implications

For practitioners, the results caution against using raw dashboard differences or a single full-period regression to allocate marketing budgets. A platform whose proprietary index is trending upward may appear to outperform another during an event even when the local event-period anomaly is weak or points in the opposite direction. Brand teams should construct platform-specific baselines, monitor deviations from recent trends, and compare several reference windows before attributing attention changes to an event.

The results also suggest that cross-platform reporting should distinguish descriptive monitoring from causal evaluation. Descriptive monitoring asks where search activity moved. Causal evaluation asks whether the event produced that movement. The first may be answered with well-designed dashboards; the second requires stronger counterfactual information, such as untreated brands, campaign-spend variation, staggered exposure, or event content measures.

4.5 Limitations and future research

The study has several limitations. First, the platform indices are proprietary, and their exact normalization rules, historical recalibration procedures, and potential methodological changes are not observable. This is not merely a routine data limitation: it is a fundamental obstacle to interpreting cross-platform levels or assigning a common substantive meaning to a coefficient such as 0.445. Transformations can improve within-series modeling but cannot make two undisclosed scales commensurate. Second, monthly aggregation is coarse for events that begin late in one month and end early in the next. It can smooth short-lived peaks and mix pre-event and event days, so daily data would permit more credible event-time analysis and local counterfactuals. Third, only two events and one product category are examined, and the 43-month time span provides limited power for a definitive analysis of cointegration, structural breaks, or a stable long-run relationship between the two platform indices.

Fourth, the design is observational and does not include advertising spending, creative content, athlete performance, sponsorship activation, media volume, algorithmic exposure, or user-level behavior. Consequently, substantive platform heterogeneity cannot be separated from measurement heterogeneity, and event coefficients remain associations rather than causal effects. Fifth, the two-way clustered analysis has only 43 month clusters, and each event occupies two months; the resulting uncertainty is more appropriate than cross-sectional clustering alone but remains imprecise. Sixth, the separate-platform regressions necessarily relax exact month fixed effects and are therefore descriptive decompositions, not stronger causal tests. Future studies should use more events, longer time spans, pre-registered event definitions, daily or higher-frequency observations, direct measures of paid and organic communication, and validation against a common external outcome such as website visits, sales, or survey-based recall.

5 Conclusion

This study examined cross-platform brand-search attention during the 2023 FIBA Basketball World Cup and the 2024 Paris Olympics using a 15,996-observation provincial panel. A full-sample fixed-effects model using log(1 + index) produces opposite-signed event coefficients, and those signs are shared across all six brands. Yet the interpretation changes under dependence-aware inference, alternative temporal specifications, and local event baselines. The evidence therefore does not establish a stable event-type asymmetry or a platform mechanism. It establishes a methodological point with direct relevance to advertising and marketing communication: conclusions drawn from proprietary cross-platform attention indices can be cross-sectionally consistent while remaining highly sensitive to temporal baseline, measurement construction, and common-shock inference. Researchers and practitioners should preserve zero observations transparently, visualize platform-specific dynamics, report brand-level and local-window evidence, and interpret cross-platform coefficients as relative associations rather than common-unit or causal effects.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. OpenAI ChatGPT (GPT-5.6 Thinking and GPT-5.6 Sol models; OpenAI, San Francisco, CA, United States; accessed July–August 2026) was used to assist with language editing, structural reorganization, statistical code checking, document formatting, and preparation of the revision. The authors reviewed and verified all analyses, factual statements, citations, and conclusions and take full responsibility for the manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcomm.2026.1938976/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Tereza Semerádová, Technical University of Liberec, Czechia

Reviewed by: Francisco Corona, National Institute of Statistics and Geography, Mexico

Isabel A. Vázquez-Sacristán, Universidad Francisco de Vitoria, Spain