Global online social response to a natural disaster and its influencing factors: a case study of Typhoon Haiyan

The study addresses how global online social media users respond to a catastrophic natural disaster and which socioeconomic factors drive these responses. Motivated by the importance of international relief and private-sector donations for disaster-stricken regions—especially in developing countries—the authors investigate whether online social responses overcome spatial barriers and which political, economic, social, cultural, natural risk, and demographic characteristics explain cross-country variation. The purpose is to characterize the spatiotemporal pattern of global attention to Typhoon Haiyan (2013) and to quantify the relative contributions of multiple factors to this online response. The work is significant for understanding public interest dynamics and informing humanitarian agencies on where and how to mobilize volunteers and donors globally.

Prior work on social media and disasters largely follows three strands: (1) Emotions on social media: studies examined emotional expression and retweet patterns across affected vs. non-affected areas (e.g., Chen et al., 2020), spatial clustering of negative emotions (Garske et al., 2021; Gruebner et al., 2018), and sentiment-based assessment of recovery (Yan et al., 2020; Contreras et al., 2022). (2) Thematic content of posts: analyses of posting themes during response and across stages of disasters (García-Ramírez et al., 2021; Brandt et al., 2019; Zhang and Cheng, 2021), including differences across demographic groups (Yuan et al., 2020; Zhu and Liu, 2021). (3) Communication mechanisms on Twitter: usage during Typhoon Haiyan within affected countries (Takahashi et al., 2015) and global attention modeled via Google search activity indicating Western dominance (Kam et al., 2021). Despite these advances, a gap remains in understanding global-scale online social responses to a specific disaster and systematically attributing cross-national socioeconomic drivers using comprehensive social media data.

Case: Super Typhoon Haiyan (2013), an extreme event with maximum sustained winds near the center of 315 km/h, causing severe losses in the Philippines and drawing global attention. Data collection and preprocessing: 234,042 Tweets related to Typhoon Haiyan were collected from Nov 4–20, 2013 using disaster-related hashtags and web crawling, following prior datasets/methods. Tweets with location attributes (self-reported locations or mobile location addresses) were geocoded to points in ArcGIS 10.7. Countries with fewer than 30 tweets or missing factor data were excluded. China was excluded due to very low Twitter penetration. The final attribution analysis covered 113 countries across all continents. Response metric and temporal segmentation: Daily counts of disaster-related tweets per country were computed (zonal statistics in ArcGIS) as the measure of online social response. Based on temporal clustering and the typhoon’s evolution, the period was divided into two stages: before landfall (Nov 4–8) and after landfall (Nov 9–20). Explanatory factors (2013 values) by dimension: - Political: Government Effectiveness (GE; World Bank) capturing service quality, policy implementation, and credibility. - Economic: Per capita GDP; Export value (goods/services) as a proxy for trade openness and international connectedness. - Social: Corruption Perceptions Index (CPI) as a proxy for social development and willingness to aid. - Cultural: Higher education enrollment rate; national proportions of major religions (Christianity, Islam, Buddhism, Hinduism) and nonreligion. - Natural risk: Expected mortality rate, GDP loss rate, and affected population rate from disasters (EM-DAT). - Demographic: Geographic distance (Euclidean from national capital to Manila); total population as proxy for social media user base. Analytical workflow: - Spatiotemporal pattern analysis of tweet counts; standard deviation ellipse to summarize spatial distribution. - Single-factor detection using Geodetector to quantify each factor’s explanatory power (q value) and identify leading factors, mitigating multicollinearity. - Multifactor modeling using Geographically Weighted Regression (GWR). One dominant factor per target layer was selected for GWR to reduce dimensionality. Export value represented the economic layer due to its connection to international linkages. Model fit: R²=0.94 (before), R²=0.90 (after). Coefficients were mapped to assess spatial heterogeneity and compared between stages to examine magnitude and sign changes. Statistical tests: Spatial autocorrelation checked via Moran’s I prior to GWR. Significance of Geodetector q values reported with p-levels.

Spatiotemporal response: - Total tweets: 234,042 (Nov 4–20, 2013). Global daily tweets rose sharply Nov 4–8 (peak on Nov 8 at Philippine landfall), then declined Nov 9–20 with a small bump Nov 10–11 (Haiyan’s impact on Hainan, China and active relief efforts). - The timing of online response closely followed typhoon intensity (Fig. 4). Spatial patterns: - Highest responses in the Philippines; strong attention in developed Western countries (Europe, North America) and neighboring countries (e.g., Australia, Vietnam, Indonesia). Africa and West Asia had the lowest responses. - Differences pre/post landfall: Most countries increased after landfall. Largest relative increases: Bhutan (×39), Botswana (×15), Luxembourg (×12.5), Mongolia (×12), Uruguay (×11.6), Nigeria (×11.4). Decreases: Andorra (-50%), Guyana (-50%), Seychelles (-50%), North Korea (-43%), Papua New Guinea (-10%); some dropped to zero (Liechtenstein, Chad, Antigua and Barbuda). Single-factor (Geodetector) results: - Before landfall: Economic factors dominated. Significant factors and q values: Export value q=0.2285 (p<0.05), GDP loss rate q=0.1882 (p<0.05), Population q=0.1788 (p<0.05), Per capita GDP q=0.1651, Proportion nonreligious q=0.1199, CPI q=0.0985. Political factor (GE) not significant; distance weak (q≈0.0437). - After landfall: Economic factors remained strongest. Factor ranking by q: Export value q=0.2392 (p<0.05), Per capita GDP q=0.1780, GDP loss rate q=0.1685 (p<0.1), Population q=0.1684 (p<0.05), Proportion nonreligious q=0.1389 (p<0.1), CPI q=0.1118 (p<0.1), GE q=0.0987 (p<0.1). Political effect became significant but weakest; distance remained weak (q≈0.0474). GWR (spatially varying effects): - Before landfall (R²=0.94): Export value mostly positive globally (higher in the Americas); CPI mixed with positive effects in Western Europe and negative in parts of the Americas; Proportion nonreligious mixed (positive in S. South America and Central Africa; negative in N. America, W. Europe, W. Africa); GDP loss rate positive in most countries (high in Ukraine 0.33, Russia 0.32, Lithuania 0.31; negative in Switzerland -0.17, Netherlands -0.15, France -0.15, U.K. -0.12); Population generally positive (high in Iceland 3.27, U.K. 1.47, France 1.47, Switzerland 1.23; negative in Chile -0.91, Peru -0.86, Brazil -0.84). - After landfall (R²=0.90): GE ranged roughly -0.155 to 0.100 with spatial heterogeneity; Export value remained largely positive with highest coefficients in South America (Chile 1.24, Peru 1.23, Argentina 1.22, Uruguay 1.21, Brazil 1.19); CPI positive in North America (Canada 0.19, U.S. 0.16, Mexico 0.14, Guatemala 0.13, El Salvador 0.13), negative in parts of Northern Europe; Proportion nonreligious mixed (Iceland 0.025; negatives in parts of Caribbean and Central America); GDP loss rate positive in many countries (Iceland 1.53, Ireland 0.47, Canada 0.35, U.S. 0.29, Cuba 0.15; negatives in parts of Europe); Population generally positive (Iceland 2.35, Canada 1.95, Ireland 1.52, Portugal 1.11, U.S. 1.09; negatives in parts of South America). - Directional stability across stages: Export value, GDP loss rate, and Population generally retained effect direction; CPI and Proportion nonreligious changed direction in some countries. Distance: - Geographical distance had little explanatory power overall, indicating online networks reduce spatial barriers to response.

The findings show that global online social responses to a catastrophic disaster closely track hazard development and are not strongly constrained by geographic distance, supporting the view that social media can overcome spatiotemporal barriers in information diffusion. Western countries form the primary foreign group exhibiting strong online responses, likely due to higher economic capacity and trade linkages with the Philippines, while neighboring countries also respond strongly owing to potential regional impacts of typhoons. The economy is the principal driver: trade openness (export value) and prosperity (per capita GDP) consistently explain higher attention, with population size further amplifying response capacity. Social (CPI), cultural (religiosity, education), and natural risk indicators contribute with marked spatial heterogeneity, and political effectiveness becomes weakly but significantly relevant in the post-disaster phase, aligning with shifts toward recovery and international assistance narratives. These insights address the research question by identifying where and why global public interest concentrates and by quantifying the relative importance and spatial variability of influencing factors, offering actionable intelligence for targeting communications and mobilizing international volunteers and donors.

Using 234,042 geolocated tweets about Typhoon Haiyan (2013), the study demonstrates that global online social responses align with disaster evolution and largely transcend geographic distance. Beyond the Philippines, Western and neighboring countries showed higher engagement than other regions. Economic factors (especially export value and per capita GDP) dominate the drivers of cross-country response both before and after landfall; social, cultural, and demographic factors are relatively weaker, and political factors matter modestly in the post-disaster stage. Identifying active online social groups in Western countries highlights opportunities for governments and humanitarian organizations to mobilize foreign volunteers and donors. The results advance understanding of online social behavior during disasters and can inform coordinated, efficient international relief outreach.

- Country coverage: Only 113 countries met tweet count and data availability thresholds; China excluded due to low Twitter penetration. This may affect generalizability and GWR stability. - Variable selection and measurement: Potential omitted variables and reliance on target-layer, first-level indicators only; religious composition sourced from Wikipedia introduces uncertainty; Government Effectiveness (definition/algorithm) is contested. - Proxies: Total population used as a proxy for social media user base due to lack of 2013 Twitter penetration data. - Event-specific nuance: Despite Haiyan’s impact on China’s Hainan, global online response was much weaker than for the Philippines; reasons remain for future investigation.