Reading Shanghai's lockdown through four newspapers

The number I remember best from my undergraduate thesis is 91%. In the hand-coded local-media sample from May 2022—the hardest month of Shanghai's COVID lockdown—roughly ninety-one articles in every hundred spoke through a frame of resilience.

This essay is a plain-language companion to my Signature Work at Duke Kunshan, Framing the Crisis (2024), supervised by Prof. Andrew Macdonald. The thesis itself contains 1,161 articles, LDA topic models, and hand-coded content analysis. The point here can be understood without first carrying all of that machinery.

What I actually read

From April into June 2022, Shanghai lived under a citywide lockdown. Twenty-six million people remained at home while food delivery and testing systems strained under the scale of the city. I collected coverage from four newspapers. The full corpus ran from April 1 through June 1; the hand-coded time sample also included June 2 so that I could compare the beginning, middle, and end:

• People's Daily (人民日报) and Global Times (环球时报), the two central outlets.
• Jiefang Daily (解放日报) and Xinmin Evening Post (新民晚报), Shanghai's two big local papers.

The event and language were the same; the readers were not. The corpus held 1,161 articles, 386 central and 775 local. Local coverage was larger because it had to meet the city at the scale of daily life: which compound was sealed, where the testing line had broken down, and which residents' committee was arranging food. Central outlets could pass over those details. Local papers could not.

What the papers were doing differently

I used two passes. First came an LDA topic model: words that repeatedly appeared together formed clusters, which I then named. Second came a stratified sample of seventy-six articles, each hand-coded for whom it cited, what kind of article it was, how it sounded, and what frame it placed around events.

Translated out of the bar charts, the pattern looked like this:

• Sources. Central media cited government officials in 67% of pieces; local media in 25%. Local media cited ordinary people in 51% of pieces; central media in 11%.
• Article type. Among central articles, 78% shared information or carried government announcements, while only 11% were narrative. Among local articles, 34% shared information and 55% were narrative.
• Tone. Central media leaned authoritative (56%); local media leaned warm and sympathetic (61% warm, 37% sympathetic).
• Framing. Both used resilience heavily, but local papers also concentrated on local community response (46% against central's 11%), while central concentrated on the national economy (67% against local's 22%).

Together, the numbers sound like two voices inside one apparatus. Central media set the line: an authoritative register, official sources, and national cohesion. Local media carried that line into daily life: a warmer register, ordinary people, and community resilience. The two were not arguing. They were doing different parts of the same work.

The one number I keep coming back to

The local papers changed most clearly over time. In April, as the lockdown began, their frames were still dispersed: struggle appeared in 45% of the sample, resilience in 45%, health in 50%, and local community response in 55%. Frames could overlap within one article. The pages read like a city trying to tell itself what was happening.

By May, when conditions were at their worst and public morale was visibly fraying, that dispersion had vanished. In the hand-coded local-media sample, 91% of articles used a resilience frame; other explanations were pushed to the margins. The coverage became one long instruction: endure, persist, hold on.

That turn can be read as cynicism: when reality ceased to cooperate, the apparatus concentrated on managing morale. It can also be read as a local obligation: readers were living inside the crisis, and a city newspaper could not speak only from above it. Both were probably true. The 91% comes from the hand-coded sample, and its direction echoes the community-resilience themes in the topic model. By June, as the city began to reopen, resilience fell to 43% of the coded sample while local response—mutual help, neighborhood organization, recovery—rose to 48%. The story turned once more.

What 91% changed for me

Two things. First, "Chinese state media" is too blunt a unit of analysis. The system has levels: central outlets signal and control information; local outlets perform empathy and manage morale. The coordination between them explains more than either voice alone.

Second, what propaganda omits—and what it makes heavier over time—often reveals more than a direct claim. A story about resilience in May 2022 need not have been false. The coded sample was simply talking about almost nothing else. When ninety-one percent of a sample looks through one frame, the frame is no longer only describing reality. It has begun to work on reality.

The thesis page explains the methodology and findings. The underlying newspaper corpus is not redistributed here because the source articles remain copyrighted.