Can autocratic power influence the media in democracies? Evidence from China's expulsion of American journalists

Abstract

Can autocratic governments influence foreign media? This study finds that political pressures from autocrats can motivate the media in democracies to tone down their negativity. Exploiting China's sudden expulsion of American journalists in March 2020, I show that US news outlets targeted by the expulsion adopted a more positive tone toward China in their subsequent coverage, compared to outlets that were not targeted. Further analyses confirm that the observed pattern is not due to unexpelled outlets presenting more negative coverage of China, and that the expulsion has similar chilling effects on media outlets that could have been affected. The findings highlight the overt threats autocracies pose to media freedom, a fundamental pillar of democratic societies.

1. Introduction

Governments have both the incentives and means to control domestic media. In democracies where media freedom is constitutionally guaranteed, incumbents, driven by electoral concerns, often attempt to influence news outlets through monetary transfers (Di Tella and Franceschelli, 2011), legal regulations (Boas and Hidalgo, 2011; Stanig, 2015), pressure on editors and journalists (Qian and Yanagizawa-Drott, 2017), or selective information provision (Lai, 2025). In autocracies where media freedom is inherently restrained, ruling elites manipulate news with outright censorship, hoping to maintain political stability and prolong regime survival (Gehlbach and Sonin, 2014; Rozenas and Stukal, 2019; Gläßel and Paula, 2020; Guriev and Treisman, 2020; Carter and Carter, 2022).

Nonetheless, the influence of governments on news reporting can extend beyond their own borders. Policymakers have warned that autocratic leaders have been attempting to distort the media in democracies in order to shape foreign public opinion and enhance their global image. While China has invested heavily in news outlets abroad and invited journalists worldwide on all-expense-paid trips (Reporters without Borders, 2019), Russia has sought to influence the agenda of foreign media through disinformation campaigns (U.S. Department of State, 2022). Countries like Qatar have even cultivated networks of American lobbyists to improve their media image and, allegedly, have hacked journalists’ email accounts to deter criticism of the regime (Flanagan, 2019). This phenomenon has garnered increasing scholarly attention, with Chen and Han (2022) recently finding that US and UK news sites have compromised their reporting to maintain market access in China.

To further evaluate whether, and to what extent, news outlets in democracies are subject to political pressure from autocratic regimes abroad, I study the case of China's expulsion of American journalists. In March 2020, China's Ministry of Foreign Affairs announced that journalists from The New York Times, The Washington Post, and The Wall Street Journal would no longer be allowed to work in the country (including the Hong Kong and Macao Special Administrative Regions), and needed to surrender their press credentials within ten calendar days (Tracy et al., 2020). The expulsion came only a few weeks after the Trump administration's decision to designate five of China's leading news agencies, including Xinhua, China Daily, and The People's Daily, as foreign agents and to limit the number of Chinese nationals who can work in the United States for these organizations to 100 (Jakes and Myers, 2020). Although the expulsion was widely seen as retaliation by the Communist regime for these newspapers’ critical coverage of its mass detention of Muslims in Xinjiang, handling of the COVID-19 outbreak, among other issues (Tracy et al., 2020), the Ministry justified this policy as “a necessary countermeasure against the U.S. government's unreasonable oppression of Chinese news organizations” (MOFA, 2020). Editors from the three newspapers unanimously condemned China's decision (Tracy et al., 2020).

From a theoretical point of view, US news outlets with journalists expelled by the Chinese authorities could react in divergent ways. On the one hand, they may harshen their tone toward the regime. Aggrieved by the expulsion, which could adversely impact their work and careers as China correspondents, journalists may wish to unleash their grievances. The loss of on-the-ground-presence in China may also exacerbate journalists’ misunderstandings about certain policies that appear harsh from an external viewpoint but garner domestic support, such as the Social Credit System (Xu et al., 2022), leading to more negativity in the news. More importantly, being forced to leave China could reduce journalists’ concerns about offending the Chinese government with their critical coverage, essentially eliminating any incentive for self-censorship (Chen and Han, 2022). On the other hand, the expulsion may give news outlets greater incentives to self-censor. Generally, foreign media value their operational presence in China as it provides them with firsthand, exclusive insights into the world's largest autocracy, which holds substantial news value for their readership and can bring commercial benefits to the outlets. In an attempt to re-establish their journalistic presence in China, they may try to appease the Chinese government by intentionally softening their tone.

To empirically discern these two possibilities, I start by collecting reports from US news outlets on China, spanning from September 2019 (six months before the expulsion) to September 2020 (six months after the expulsion). Next, I conduct a dictionary-based sentiment analysis to measure the tone of these articles, following established practices. My empirical approach mimics a difference-in-differences (DiD) design in the sense that it compares how news outlets that experienced journalists’ expulsion (The New York Times, The Washington Post, and The Wall Street Journal) changed their tone toward China after the expulsion, relative to those that maintained a notable presence in the country (Bloomberg News). The results show that the expulsion motivates news outlets to portray China in a relatively more positive light. Moreover, by comparing expelled news outlets with China Daily, a Chinese state-owned English-language newspaper that is unlikely to be affected by the expulsion, I confirm that the findings are not driven by a negative shift in Bloomberg's reports. In addition, I find that the expulsion has chilling effects on news outlets that could have potentially been affected, including the Associated Press, CNN, The Los Angeles Times, BBC, and Reuters.

This study is significant in a number of ways. Most importantly, it expands the existing research on information control of authoritarian regimes, which typically focuses on autocrats’ control over domestic citizens’ access to information (e.g., Roberts, 2018; Sanovich et al., 2018; Guriev and Treisman, 2020; Howells and Henry, 2021; Stukal et al., 2022). It demonstrates that autocratic powers, such as China, also possess the capacity to influence news produced outside their jurisdiction, thereby distorting the information accessible to foreign audiences. Additionally, this study supplements the findings of Chen and Han (2022) by providing further empirical evidence on the role of foreign autocratic governments in shaping the coverage of news outlets in democracies. In doing so, it extends the literature on government control of the media (Boas and Hidalgo, 2011; Di Tella and Franceschelli, 2011; Durante and Knight, 2012; Gehlbach and Sonin, 2014; Stanig, 2015; Qian and Yanagizawa-Drott, 2017; Lai, 2025) and contributes to the burgeoning body of work on the influence of autocracies abroad (DellaVigna et al., 2014; Peisakhin and Rozenas, 2018; Bail et al., 2020). Finally, this study joins a growing body of scholarship incorporating automated sentiment analysis into applied research (Crabtree et al., 2020; Lajevardi, 2021; Osmundsen et al., 2021).

2. Empirical design

2.1. Sample

To empirically test the effect of the expulsion, I start by constructing a sample of China-related articles published by news outlets that had journalists expelled—The New York Times (NYT), The Washington Post (WP), and The Wall Street Journal (WSJ), and news outlets that operate in China but did not face expulsion—Bloomberg News. The time frame for this sampling spans from September 2019, six months before the expulsion, to September 2020, six months after the expulsion. To find news outlets’ articles about China, I search the word “China” on the websites of NYT, WP, and WSJ, and scrape all articles returned by the search. In the case of Bloomberg News, since directly searching on its website only returns articles published within the last six months at the time of the search, I use Google to search the word “China” within Bloomberg's website,¹ and scrape the first 100 articles returned by the search each month.² Then, for all news outlets, China-related articles are defined those that mention “China,” “Chinese,” or “Xi Jinping” at least three times.³

Bloomberg's articles are included in the sample because, empirically, I seek to estimate the effect of the expulsion by comparing news outlets whose journalists were expelled with those whose journalists remained unaffected. I specifically choose Bloomberg News as the main comparison group because it was the only US news outlet that maintained a substantial presence in China after the expulsion. The “presence” of foreign media in China is quantified by the number of instances their journalists pose questions during the daily press conferences organized by China's Ministry of Foreign Affairs (MOFA). These press conferences are widely attended by foreign journalists as they provide rare opportunities to access information from, and interact with, the Chinese government. Beginning March 24, 2020, MOFA began to specify the affiliation of journalists posing questions during the press conferences in the released transcripts. Appendix Figure A1 shows that journalists working for the Associated Press and CNN, two other US news outlets operating in China, spoke up in fewer than ten press conferences over a six-month period,⁴ Bloomberg's reporters were regularly called upon, even posing more questions than journalists from China Daily, a state-owned English-language daily newspaper.

2.2. Measuring news tone

I measure the tone (i.e., sentiment) of news articles using the Lexicoder Sentiment Dictionary (LSD) developed by Young and Soroka (2012). The dictionary consists of 4576 positive and negative words and phrases drawn from the three largest lexical resources: Roget's Thesaurus, the General Inquirer, and the Regression and Imagery Dictionary. It is primarily tailored for political texts and has been shown to yield assessments that align better with human coders than other commonly used dictionaries across a range of topics, including foreign affairs, crime, the environment, and the economy (Young and Soroka, 2012). Scholars have employed the LSD to analyze the tone of news on the economy (Soroka et al., 2015a; Wlezien et al., 2017; Soroka et al., 2018), elections (Giasson, 2012; Fournier et al., 2013), and political leaders (Balmas, 2017).

Following Soroka et al. (2018, 2015b), I calculate the sentiment of a given article by subtracting the count of negative words from the count of positive words, then dividing the difference by the total number of words in the article.⁵ This method measures the percentage-point difference between positive and negative words in an article, thereby capturing both the direction and magnitude of tone while controlling for article size (Soroka et al., 2015a). Figure 1 shows how the news tone fluctuates over time for NYT, WP, WSJ, Bloomberg, and China Daily. As expected, China Daily adopts the most positive tone toward China. More importantly, the difference in tone between NYT, WP, WSJ, and Bloomberg visibly narrows after the expulsion. Appendix Figure A2 shows the distribution of news tone before and after the expulsion, confirming this pattern.⁶

Figure 1. Changes in news tone.

I also use the sentimentR R package, which includes an algorithm for calculating text polarity sentiment, to analyze the tone of news articles, following Osmundsen et al. (2021). The algorithm computes sentiment at the sentence level while accounting for valence shifters, such as not, really, hardly, and but, which can significantly alter the sentiment (or the intensity of the sentiment) conveyed in a text. To determine the sentiment of a given article, I average the sentiment scores of all its sentences. The output is used both to cross-validate the LSD-based sentiment score and to check the robustness of the main results. Reassuringly, as depicted in Appendix Figure A3, the two new tone measures are highly correlated, with a correlation coefficient of 0.78. Summary statistics are presented in Appendix Tables A2 and A3.⁷

2.3. Model

I specify the main estimation model as follows:

$$\textit{Tone}_{i, j, t} = \beta ( T_{i, j} \times \textit{Post}) + {\bf X}_{i}\theta + \mu_{\,j} + \nu_t + \epsilon_{\,j}$$

where i, j, and t index article, news outlet, and year-month respectively. Tone _i,j,t is the sentiment score of article i published by news outlet j in month t. T _i,j is a dummy variable that takes 1 if article i is published by NYT, WP, or WSJ and 0 if published by Bloomberg. Post takes 1 if the article is published in or after March 2020, and 0 otherwise. X is the vector of article-level control variables, which include the length of the article and the number of times the article mentions China-related keywords (“China,” “Chinese,” and “Xi Jinping”). I add fixed effects at the news outlet and year-month levels, which render T _i,j and Post redundant in the model.⁸ Standard errors are calculated based on 1000 bootstrap resamples.⁹

Although the model mimics a DiD design, I do not claim that DiD is used for causal identification in this study. The reason is straightforward: the use of DiD rests on the parallel-trend assumption, which is difficult to justify in this case. The Chinese government's decision to expel American journalists was a response to the Trump administration's decision to limit the number of Chinese citizens who could work for the Chinese media operating in the United States. It is highly probable that journalists from NYT, WP, and WSJ were specifically targeted due to their increasingly negative reporting on China. In other words, their reporting tone could be on a different trend than Bloomberg's, which may continue post-expulsion. However, this implies a downward bias in my estimates. That is to say, if β is found to be positive, its size is very likely to be underestimated, and if β is found to be negative, its size is very likely to be overestimated.

3. Results

3.1. Main results

Table 1 presents the main results. The positive and significant coefficients of the interaction between T _i,j and Post indicate that NYT, WP, and WSJ toned down their negativity toward China (i.e., reported on China more positively) after the expulsion, compared to Bloomberg (columns 1 and 2). To put the effect size into perspective, it is about 7 percent of the average difference in news tone between NYT (−0.013) and China Daily (0.042) before the expulsion. In columns 3–5, I estimate the effect for the three newspapers separately, without adding any fixed effects for parsimony. The negative coefficients of T indicate that, throughout the study period, NYT, WP, and WSJ portrayed China more negatively than Bloomberg, and the negative coefficients of Post suggest that all news outlets adopted a more negative tone toward China following the expulsion. However, the positive coefficients of T × Post again imply that the reporting tone of NYT, WP, and WSJ improved after the expulsion relative to Bloomberg's.¹⁰ This change, though, was not enough to completely offset the pre-existing differences in the news outlets’ tone toward China and the overall decline in their reporting tone post-expulsion. Finally, the varying size and statistical significance of the coefficients for T × Post suggest that NYT and WSJ responded more strongly than WP to the expulsion. A probable explanation for this phenomenon is that NYT and WSJ, which operate Chinese-language websites, require a higher volume of news stories about China and thus have more incentives to reestablish their presence in the country compared to WP, which only translates some of its articles into Chinese on a regular basis.

Table 1. Effect of the expulsion on news tone

Notes. g = Articles sampled by Google Search. Standard errors based on 1000 bootstrap resamples are reported in parentheses.

*p < 0.10, **p < 0.05, ***p < 0.01.

As robustness checks, I first re-run the above models using the sentimentR measure of news tone (Appendix Table B2). Reassuringly, the interaction term remains positive and statistically significant. The effect size, approximately 6 percent of the average pre-expulsion news tone difference between NYT (−0.017) and China Daily (0.179), remains consistent. Next, in the left panel of Appendix Table B3, I use NYT, WP, and WSJ articles sampled by Google Search.¹¹ Recall that, in constructing the main sample, I search the word “China” on the websites of NYT, WP, and WSJ directly, but use Google to search “China” within the website of Bloomberg News. While Appendix Figure A5 illustrates that NYT, WP, and WSJ articles returned by Google Search are not systematically different in tone toward China compared to those returned by website searches, Appendix Table B3 further confirms that the method of article collection (using Google to search within news outlets’ websites versus searching directly on news outlets’ websites) does not affect the results. In the right panel of Appendix Table B3, I increase the threshold of what counts as “China-related” articles,¹² and find almost identical results. Moreover, I show that the finding is not driven solely by opinion articles (Appendix Table B4), articles discussing the Trade War (Appendix Table B5), or Sino–US trade relations (Appendix Table B6). I also check whether re-coding the Post variable¹³ (Table B7) has any impact on the estimate. Finally, in Table B8, I allow “treated” and “control” news outlets to follow different time trends, using a news outlet-specific linear trend instead of the simpler news-outlet and year-month fixed effects specification outlined above (Foos and Bischof 2021).¹⁴ The results are unchanged.

3.2. Alternative explanations

One alternative explanation for the finding is that it is driven by the negative change in Bloomberg's reporting tone in response to the incident. In theory, Bloomberg's reporters might react to the expulsion of their colleagues at NYT, WP, and WSJ by adopting a more critical stance toward China, even in the absence of personal grievances. To empirically address this concern, I use China Daily as the comparison group instead of Bloomberg News. There is no reason to expect that Chinese state media would respond to the incident with a more negative depiction of China.¹⁵ The results presented in Table 2 affirm the previous conclusion that the expulsion of journalists motivates news outlets to adopt a more positive tone in their China coverage than they otherwise would have. Furthermore, column 6 indicates that Bloomberg's reporting remained largely unaltered—the coefficient of T × Post is small and statistically indistinguishable from zero. Appendix Table B9, utilizing the sentimentR news tone measure, echoes similar trends.

Table 2. Effect of the expulsion on news tone (compared to China Daily)

Notes. g = Articles sampled by Google Search. Standard errors based on 1000 bootstrap resamples are reported in parentheses.

*p < 0.10, **p < 0.05, ***p < 0.01.

Another explanation to consider is whether the observed pattern results from broader changes in these outlets’ attitudes toward authoritarian regimes. To investigate this, I collect their articles on Russia and Iran using the method outlined above and explore how they adjust their reporting tone on these two countries (Table 3). The coefficient of the interaction term is not statistically significant. If anything, compared to Bloomberg, the coverage of Russia and Iran by both the NYT and WP becomes more negative in the post-expulsion period (columns 3 and 4), while the WSJ's coverage shifts toward a more positive tone (column 6).

Table 3. Effect of the expulsion on news tone on Russia and Iran

Notes. g = Articles sampled by Google Search. Standard errors based on 1000 bootstrap resamples are reported in parentheses.

*p < 0.10, **p < 0.05, ***p < 0.01.

I also combine the articles on China, Russia, and Iran and estimate a triple-DiD model in which I use Russia- and Iran-related articles as an additional comparison group, following the approach of Chen and Han (2022).¹⁶ The results are presented in Appendix Table B10. While the coefficient of the second-degree interaction T × Post is negative, the triple interaction T × Post × China is positive and statistically significant in the preferred model presented in column 2. Taken together, we can conclude that “treated” outlets adopt a somewhat more negative tone toward Russia and Iran, but a more positive tone toward China after the expulsion, compared to “control” outlets.

3.3. Placebo test

I conduct a placebo test by examining the effect of the expulsion on news outlets that have near-zero on-the-ground presence in China. In theory, the effect should be null, given that they don't have any pre-expulsion presence to reclaim and therefore lack the motivation to depict China more favorably post-expulsion. I identify two major US news outlets that frequently report on China but do not have stationed journalists there—The Boston Globe (BG) and The Chicago Tribune (CT). I repeat the exercise and compare their China-related articles with those from Bloomberg News and China Daily. The results are detailed in Table 4. In line with the theoretical expectations, the expulsion does not have a discernible effect on them (p = 0.593 and 0.870 in columns 1 and 2, respectively). If expelling journalists of NYT, WP, and WSJ systematically affected outlets that do not have any meaningful presence in China, the coefficients of the interaction term should be more consistent in sign, larger in magnitude, and stronger in statistical significance. Such findings also alleviate the concern that Bloomberg News became more critical of China in response to the expulsion—if that were true, we should find T × Post to be positive and statistically significant.

Table 4. Effect of the expulsion on news tone of outlets that could not face expulsion

Notes. g = Articles sampled by Google Search. Standard errors based on 1000 bootstrap resamples are reported in parentheses.

*p < 0.10, **p < 0.05, ***p < 0.01.

3.4. Chilling effects of expulsion

Finally, I assess the potential chilling effects of the expulsion. That is, news outlets that were not directly affected by the expulsion also reported China less negatively in the aftermath of the incident, fearing that they could be the next target of the Chinese government. To investigate this, I gather China-related articles from three other US news outlets—the Associated Press (AP), CNN, and The Los Angeles Times, and two UK news outlets—BBC News and Reuters. Unlike BG and CT, these outlets maintained some presence in China both before and after the expulsion, as evidenced by their occasional participation in MOFA press conferences.¹⁷ Table 5 shows the results, which support the idea that the expulsion has chilling effects on foreign media operating in China. The coefficients of T × Post are consistently positive and statistically significant, implying that AP, CNN, LAT, BBC, and Reuters adopted a less hostile tone in their China coverage following the expulsion compared to China Daily, even though their journalists were not expelled.

Table 5. Effect of the expulsion on news tone of outlets that could face expulsion

Notes. g = Articles sampled by Google Search. Standard errors based on 1000 bootstrap resamples are reported in parentheses.

*p < 0.10, **p < 0.05, ***p < 0.01.

A natural question arising from these findings is why the expulsion hasn't had any chilling effect on Bloomberg News. One probable cause is that Bloomberg's coverage of China was already sufficiently positive. Therefore, there was minimal room for further positive shifts without appearing overtly biased. Notably, the average tone of Bloomberg's articles is more favorable than that of any other foreign news outlet included in my sample (AP, BBC, BG, CNN, CT, LAT, NYT, Reuters, WP, WSJ). This corresponds with anecdotal evidence suggesting that Bloomberg has been practicing self-censorship for a long time to avoid displeasing the Chinese government. In 2013, for example, it was reported that Bloomberg's editor-in-chief withheld a story detailing the concealed financial ties between a wealthy individual in China and the families of top Chinese officials (Wong, 2013). The outlet's favorable attitudes toward China are also reflected by the high frequency with which its correspondents participate in the Q&A sessions of the MOFA press conferences. From Bloomberg's perspective, it might believe that it was shielded from the expulsion due to its long-standing amicable relationship with the regime.

4. Discussion and conclusion

This study leverages China's sudden expulsion of American journalists in March 2020 to inform our understanding of how foreign autocratic regimes can influence the media in democracies. Contrary to the conjecture that the expulsion would result in a more negative portrayal of China by the affected news outlets, the exact opposite is found—the tone of China-related articles from outlets that experienced journalist expulsion, relative to that from outlets not affected by the expulsion, turned more positive thereafter. My analyses also demonstrate that the findings cannot be explained by unaffected news outlets portraying China less positively post-expulsion, or by broader changes in news outlets’ attitudes toward authoritarian regimes. Moreover, a placebo test comparing news outlets unaffected by the expulsion to those with virtually no on-the-ground presence in China further confirms that the expulsion did not have a systematic impact on the latter group. Finally, I show that other news outlets with journalists stationed in China, although not directly affected by the expulsion, also modified their reporting tone on China in response to the event. Taken together, these findings suggest that autocratic regimes can utilize available policy tools to shape the reportage from free and independent media abroad, thereby potentially influencing public opinion in democratic societies.

The findings presented here appear to contradict the results of Chen and Han (2022), which show a negative effect of being blocked from the Chinese Internet on news outlets’ tone toward China. The authors interpret this as evidence that these outlets self-censor to maintain market access to China before the blockage. Why do measures taken by the Chinese government to restrict foreign media's access to China yield such contrasting outcomes? The answer likely lies in the distinction between losing information access and losing market access, with the former being less permanent than the latter, which leads to different behavioral responses from the news outlets. Websites blocked by the Great Fire Wall are rarely unblocked—NYT, WP, and WSJ have been permanently blocked since 2012, 2019, and 2014, respectively.¹⁸ After the blockage, these outlets have little incentive to portray China more positively, as the chance of having the blockage reversed is slim to none.¹⁹ However, once expelled, journalists can potentially be granted re-entry into the country in a much shorter time frame. Some degree of self-censorship may have already been in place before the expulsion—it is just that journalists self-censor more heavily thereafter, in the hope of appeasing the Chinese government and accelerating the process of re-entry. Indeed, after a high-level diplomatic talk between the two countries’ leaders in November 2021, the three news outlets were permitted to send their journalists back to China. The expulsion thus motivates journalists to tone down their negativity because of its temporary and remediable nature.

Of course, China is not the only authoritarian country that seeks to affect how foreign reporters write about them and distort the information accessible to foreign citizens, and expelling journalists or blocking websites is not these countries’ only means to achieve their goals—as mentioned before, their toolkit ranges from financial incentives to intimidation and violence. To what extent are the study's findings generalizable to other contexts, given that the estimates are specific to China's expulsion of American journalists? A key scope condition for autocrats’ success is that foreign news outlets need to have sufficient incentives to comply. This means the outlets must have a significant on-the-ground presence in the authoritarian country, this presence must be vital for their information gathering, and the information must be of substantial news value to generate commercial benefits. When these conditions are not met—such as when journalists can operate from nearby countries using secondary sources, or when stories from the authoritarian country do not attract significant readership—coercive actions like expulsion are more likely to provoke backlash rather than compliance from news outlets with nothing to lose.

Whether autocrats’ other attempts to influence the media abroad have succeeded or failed is ultimately an empirical question that this study leaves unanswered. The implications of these efforts for domestic and foreign public opinion also remain to be explored. Moreover, how would the media in autocratic countries respond when their own government targets foreign news outlets, or when they themselves become targets of democratic governments?²⁰ Addressing these questions in future research will help us better understand governments’ control of information and its cross-border ramifications, and enrich our understanding of how information flows and narratives are constructed in the globalized world (Norris and Inglehart, 2009).