Framing the Crisis — Signature Work

My undergraduate Signature Work at Duke Kunshan University, submitted March 7, 2024 under Prof. Andrew Macdonald (Division of Social Science). Roommate Xingyu Shen helped with data collection and trimming.

Title: Framing the Crisis — A Study of Local vs. Central Media Propaganda During Shanghai COVID-19 Lockdown.

The question

When central and local state media cover the same crisis in the same language, how do their propaganda strategies differ — and is the relationship one of conflict, or of coordination?

The corpus

• 1,161 articles across the April 1 – June 1, 2022 lockdown window
• Central: People's Daily (207) + Global Times (179) → 386 articles
• Local: Jiefang Daily 解放日报 (316) + Xinmin Evening Post 新民晚报 (459) → 775 articles
• A Python crawler against CNKI (for People's Daily, Global Times, Jiefang Daily) plus the Xinmin website; Python translation pass for analysis

The method

Two passes layered on each other:

1. LDA topic modeling (Latent Dirichlet Allocation) on the full corpus. A coherence-score sweep landed on k=9 topics for central and k=10 for local; top-20-word vectors per topic, then I labeled them.
2. Content analysis on a stratified sample of 76 articles (9 central + 67 local) drawn from the start, midpoint, and end of the lockdown — coded along four dimensions: source (officials / professionals / ordinary), nature (info-sharing / persuasion / narrative), tone (authoritative / suggestive / warm / sympathetic), and framing (struggle / resilience / government response / local response / economy / health).

What I found

The dominant dynamic between central and local media wasn't conflict — it was a stratified division of labor. Two complementary halves of one apparatus:

• Central media → "hard propaganda." Authoritative tone (56% vs local's 24%). Government officials as the dominant source (67% vs 25%). Information-sharing as the dominant nature (78% vs 34%). Themes built around national cohesion, governmental capacity, and global leadership.
• Local media → "soft propaganda." Warm/sympathetic tone (61% warm vs central's 11%). Ordinary people as the dominant source (51% vs 11%). Narrative storytelling as the dominant form (55% vs 11%). Themes built around community resilience, neighborhood logistics, and human-centric stories.

The single most striking finding was temporal: in May 2022 — the peak of the lockdown — 91% of local-media coverage was framed around resilience. Not an accident. As public morale buckled, the local apparatus pivoted hard into a survival/endurance narrative.

What I learned

That an authoritarian propaganda system isn't monolithic — it's stratified, with central and local arms running different scripts at different audiences, in coordination. The interesting question isn't whether they agree (they do). It's how the labor of agreement gets divided.

Code and data: github.com/Waaangjl/local_central-level-propaganda.