The effect of official intervention on reducing the use of potentially discriminatory language during the COVID-19 pandemic in China

The paper examines how official intervention by the Chinese government influenced the public’s use of potentially discriminatory language (e.g., “Wuhan pneumonia”) in online searches during the initial stages of COVID-19. Motivated by the widespread problem of social stigma linked to disease naming and origins, the study asks whether and how government guidance and official naming can reduce discriminatory search behavior, how such behavior evolves over time, and how regional socioeconomic factors shape these patterns. By leveraging large-scale behavioral data from Baidu searches, the study seeks to provide evidence on the effectiveness of official communications in mitigating stigma and to inform strategies for public health messaging during epidemics.

The study situates disease-related stigma within broader social stigma theory (e.g., Goffman, Sontag) and public health practice. Historically, names such as “Spanish flu,” “Ebola,” and “Mexican influenza (H1N1)” contributed to stigma, hostility, and barriers to disease control. WHO and UN agencies emphasize that naming should avoid references to places, people, or animals (WHO 2015), and have advocated for government, media, and community roles in preventing stigma during COVID-19. Prior scholarship and policy actions (e.g., UK and Australia) indicate that official intervention can reduce stigma. In early 2020 China, fear and uncertainty created conditions conducive to stigmatizing language; official guidance and later formal naming sought to counteract this. The paper thus hypothesizes that official intervention would reduce discriminatory search behavior and potentially diminish the influence of socioeconomic drivers of stigma.

Data source and measure: The study uses Baidu Index, combining PC and mobile search data, to quantify daily search volumes for two keywords: the potentially discriminatory “Wuhan pneumonia” and the neutral “novel coronavirus pneumonia.” The discriminatory search index is computed as the daily proportion: searches for “Wuhan pneumonia” divided by the sum of searches for “Wuhan pneumonia” and “novel coronavirus pneumonia.” Time frame and periods: A three-week window (January 19–February 8, 2020) is analyzed, partitioned by key events: (1) initial stage around the Wuhan lockdown on January 23; (2) before official guidance on discriminatory language (up to January 29, when CCTV and the National Health Commission issued guidance discouraging stigmatizing terms); (3) after official guidance (from January 29); and (4) following official naming on February 8, when the Joint Prevention and Control Mechanism temporarily named the disease “novel coronavirus pneumonia (NCP)” and media were required to use this terminology. Macro covariates: Provincial-level variables (mostly from 2018 sources) include land area, population density, GDP, proportion of fiscal expenditure, CPI, registered unemployment rate, years of education per capita, internet penetration rate, and average number of official social media (Weibo) posts. Analytic strategy: The authors conduct cross-sectional regressions for 30 provinces at each of the four time points to assess determinants of discriminatory search behavior. They then form panel data capturing declines in discriminatory searches between adjacent periods and estimate models (fixed/random/mixed; tests include Hausman, LM, and F-tests) incorporating a dummy for official intervention (starting January 29) and the role of official social media posting. Model diagnostics include R-squared, F-tests, and heteroscedasticity checks.

- Temporal shifts: The discriminatory search index peaked at 0.933 and fell to 0.179 over 21 days. On January 23, “Wuhan pneumonia” searches peaked at 613,238 versus 162,438 for “novel coronavirus pneumonia,” with a discrimination index of 0.7906. Between January 29 and 30, the index dropped sharply from 0.7422 to 0.2574 as official guidance took effect, and it continued to decline after official naming on February 8. - Descriptive statistics (provincial means across periods): initial stage 0.7312 (SD 0.0615; min 0.5770; max 0.8473); before guidance 0.6677 (SD 0.0885; min 0.4427; max 0.8257); after guidance 0.2337 (SD 0.0382; min 0.1602; max 0.3353); following official naming 0.1715 (SD 0.0274; min 0.1316; max 0.2448); overall 3-week average 0.4360 (SD 0.0448; min 0.3411; max 0.5489). - Regional patterns: Shanghai consistently ranked high in discriminatory search usage; Guangdong and Beijing were high initially but dropped markedly after guidance and fell out of the top ten following official naming. In contrast, Qinghai and Tibet showed increases after guidance, consistent with lower internet penetration and less exposure to official media. - Media ecosystem: Portal sites (Sina, NetEase, Sohu, Tencent) accounted for 53% of headlines using “Wuhan pneumonia,” while central/party/government media were more cautious. After February 8, party/government outlets ceased using “Wuhan pneumonia” in headlines. Of 18 headlines using “novel coronavirus pneumonia,” 55.6% were from party/government media (e.g., five from CCTV News), indicating official media leadership in terminology normalization. - Cross-sectional regressions: In early periods, higher population density, GDP (lnGDP), and years of education per capita were associated with higher discriminatory search usage; livelihood variables (higher CPI and registered unemployment), internet penetration, and more official social media posts were associated with lower discriminatory search usage. Land area was negatively associated. Models showed strong fit early (adjusted R² ≈ 0.93 initial; 0.92 before guidance; heteroscedasticity not detected initially), but explanatory power weakened after guidance (R² ≈ 0.68) and especially after naming (R² ≈ 0.36). - Panel analysis (decline between periods): Official intervention significantly accelerated the reduction of discriminatory searches (official guidance coefficient positive and highly significant in the reduction model). The interaction/role of the number of official social media posts was positively associated with larger declines. The panel model explained about 50% of the variance in reductions (R² ≈ 0.51; F-test p=0.0000).

Findings indicate that official intervention—both guidance to avoid stigmatizing language and formal disease naming—substantially reduced the use of potentially discriminatory language in internet searches. The most pronounced behavioral shift occurred immediately after initial guidance (January 29–30), with sustained reductions following official naming. Media channels mattered: official/party media adopted and propagated neutral terminology, while portals were initially less cautious; after naming, stigmatizing terms disappeared from central outlets. Regional heterogeneity reflected socioeconomic and media-access differences: provinces with higher economic development and education were initially more prone to discriminatory searches yet exhibited faster adjustment post-guidance; areas with lower internet penetration (e.g., Qinghai, Tibet) showed weaker responsiveness, highlighting the dependence of intervention effectiveness on information access. Overall, results support the hypothesis that timely, coordinated official communication can rapidly mitigate stigmatizing behaviors online and can offset socioeconomic drivers of stigma under conditions of broad media reach and engagement.

The study demonstrates that timely official guidance and formal naming conventions significantly reduced the prevalence of potentially discriminatory language in online searches during early COVID-19 in China. The abrupt drop in discriminatory search behavior after January 29 underscores the power of targeted public communication. Regional analyses reveal initially higher discriminatory tendencies in more developed, denser, and more educated provinces, but these differences diminished after guidance. Policy implications include: (1) implement early, clear official media guidance to prevent stigmatizing language; (2) adhere to WHO best practices for naming infectious diseases to avoid geographic or group associations; and (3) ensure transparent, accessible dissemination of health information to alleviate fear and misinformation. Future work could extend to longer timelines, additional platforms, and comparative contexts to further evaluate generalizability and mechanisms.

The study relies on search behavior as a proxy for discriminatory attitudes and explicitly notes that causality between searches and discrimination cannot be directly inferred. Model explanatory power declined in later periods, suggesting structural changes after official interventions that early-period regressors no longer captured. Effectiveness varied by region, with limited impact in provinces with low internet penetration and less exposure to official media (e.g., Qinghai, Tibet). The analysis focuses on a short, early-pandemic window and aggregated provincial data, which may mask within-province heterogeneity and dynamic nuances.