Research study 03 · Qin, Strömberg & Wu (2018)
How often do newspapers mention Xinhua?
Share of collected articles containing 新华社, the Chinese name of China’s state news agency, Xinhua.
1 September 2026–30 September 2026 · 22 newspapers · 30,624 collected articles · texts as of 12 October 2026
Shares range from 9.8% in People’s Daily to 31.5% in Tibet Daily.
| Newspaper | Share |
|---|---|
| Tibet Daily西藏日报 | 31.5%265 / 842 |
| Gansu Daily甘肃日报 | 29.3%325 / 1,111 |
| Inner Mongolia Daily内蒙古日报 | 26.6%295 / 1,108 |
| Jilin Daily吉林日报 | 26.1%305 / 1,170 |
| Zhejiang Daily浙江日报 | 24.9%339 / 1,361 |
| Jiangxi Daily江西日报 | 23.2%322 / 1,385 |
| Jiefang Daily解放日报 | 22.4%320 / 1,429 |
| Sichuan Daily四川日报 | 21.1%236 / 1,116 |
| Nanfang Daily南方日报 | 20.8%279 / 1,339 |
| Guangming Daily光明日报 | 20.7%374 / 1,805 |
| Chongqing Daily重庆日报 | 20.5%228 / 1,111 |
| Fujian Daily福建日报 | 17.4%257 / 1,478 |
| Heilongjiang Daily黑龙江日报 | 16.8%210 / 1,252 |
| Shaanxi Daily陕西日报 | 16.8%268 / 1,598 |
| Xinjiang Daily新疆日报 | 16.2%162 / 1,002 |
| Hubei Daily湖北日报 | 14.5%148 / 1,022 |
| Henan Daily河南日报 | 13.1%161 / 1,232 |
| Hunan Daily湖南日报 | 12.9%172 / 1,333 |
| Guizhou Daily贵州日报 | 12.5%218 / 1,749 |
| Xinhua Daily新华日报 | 11.8%223 / 1,885 |
| Dazhong Daily大众日报 | 11.7%203 / 1,739 |
| People's Daily人民日报 | 9.8%250 / 2,557 |
Ordered by literal mention rate, not political bias or loyalty.
A mention may be a news credit, a photo credit or an institution’s name. It does not by itself establish that Xinhua supplied the article.
What a mention can mean
A dispatch credit
The article opens with a Xinhua dateline. Here the string marks a news credit.
新华社北京9月8日电李干杰会见蒙古民主党代表团
People's Daily · 人民日报 ·
Read the original — 李干杰会见蒙古民主党代表团A venue name
This report names the Xinhua Art Museum as an exhibition venue. Its byline credits a Shaanxi Daily reporter; the match does not establish Xinhua authorship.
参加了在北京新华社美术馆举办的画乡里的诗意中秋
Shaanxi Daily · 陕西日报 ·
Read the original — 画乡里的诗意中秋The study and our method
Qin, Strömberg and Wu (2018) ask how market competition affects political bias in Chinese newspapers. Their study includes a measure based on the string 新华社. This card tests that component; it does not reproduce their nine-part bias index or their findings about competition.
We count an article once if its title or text contains 新华社, in the version the database held on 12 October 2026: with the papers’ own site text removed, and before the articles cut at 5,000 characters were collected again in full. The denominator is every usable collected article in that paper from 1–30 September 2026. Cross-paper copies remain separate observations. The table is ordered by this literal mention rate. A stored record may contain several briefs or a digest; records do not always correspond to standalone print stories.
All 22 papers have some readable text on each of the same 30 dates. This does not establish complete editions. Tianjin and Hainan are excluded because this snapshot chiefly contains their front pages; Ningxia and Yunnan have no coverage in this month; Hebei Economic Daily is a substitute paper with only 22 covered days.
The 30,624 records include 420 empty bodies and 745 flagged truncated bodies. Title-only records stay in the denominator and can match on their title. Missing text may conceal a mention. Recalculations excluding flagged cuts, excluding both cuts and empty bodies, and weighting days equally are saved with the research results. The range is 9.8%–31.2% after excluding both empty and flagged cut bodies; daily equal weighting gives 10.1%–34.6%. These checks retain differences but do not establish a definitive ranking of newspapers.
The members’ tables hold the texts collected again since then, so they find slightly more: on 14 October 2026, 5,629 matching articles instead of 5,560 across the 22 papers, for example People’s Daily 257 of 2,557 instead of 250. Each difference is an article whose full text, collected again, mentions 新华社 after the old 5,000-character cut. Their figures will move until the cut articles have all been collected again.
An independent SQL recount checks every paper’s numerator, denominator and text-quality counts. A fixed sample selects two matching articles per paper, plus 12 diagnostic selections without typical credit phrases; one overlaps, so there are 55 distinct texts. AI-assisted close reading checks what the matches mean. This is not a human-validated estimate of authorship accuracy.
The original study uses annual WiseNews searches from 1999–2010 and an approximate article denominator. Its exact search fields are not specified. Our monthly, cleaned-text count is an adaptation. Commercial newspapers, market information and the historical reform setting would be needed to test the paper’s competition question.
Qin, Strömberg & Wu (2018). Media Bias in China.
Research preview · Data as of 12 October 2026.