Abstract
Digital platforms provide increasingly accessible measures of brand attention, yet proprietary indices are normalized within platforms and may follow different temporal trajectories. This creates a methodological risk: the same sporting event may appear to favor different platforms depending on the temporal baseline and inferential specification. We examine monthly provincial brand-search indices for six sportswear brands across 31 provincial-level units in mainland China, 43 months, and two platforms, Baidu and Douyin, yielding 15,996 observations. The revised primary specification uses log(1 + index), retaining 29 zero-valued Baidu observations, with brand–province–platform and month fixed effects. In the full-sample model, the Paris Olympics × Douyin coefficient is 0.445 (two-way clustered
p
= .021), whereas the Basketball World Cu
p
× Douyin coefficient is −0.326 (
p
= .254). The positive-only logarithmic specification produces virtually identical point estimates. However, alternative platform-time specifications materially change coefficient magnitudes and, in some cases, signs; local event-centered contrasts are opposite to the full-sample pattern. A brand-level robustness analysis shows that all six brands share the full-sample sign pattern, but this cross-sectional consistency does not eliminate temporal-baseline sensitivity. The data therefore do not establish a stable event-type or platform mechanism. Instead, they show that the full-sample pattern is robust to zero handling but that conclusions drawn from proprietary cross-platform attention indices depend materially on temporal specification and common-shock inference. For advertising and marketing communication research, transparent outcome construction, platform-specific temporal evidence, and dependence-aware uncertainty should be treated as core elements of event evaluation.