Policy narratives in the US on China have undergone a dramatic transformation in recent years. This column draws on various kinds of public records and employs text-as-data methods to examine the evolution of the US discourse on China over time and also across different sources of public narrative. Whereas the 19th century demonstrates oscillating trends, a clearer pattern emerges in the latter half of the 20th century, one driven less by fluctuating developments and more by ideological and political patterns. The overall trends in net sentiment suggest an interesting contrast between presidential discourse and other sources, with presidential speech appearing to have been a follower.
Policy narratives in the US on China have undergone a dramatic transformation in recent years. Increasingly, China is viewed as an economic rival that imperils America’s economy and a geopolitical adversary that undermines US national security. In the economics literature, the impact of import competition from China on local labour markets and political polarization have received considerable attention (Autor et al. 2016, 2020).
China’s turn towards the market after 1978 and the unprecedented economic growth that ensued were generally welcomed by the US foreign policy elite. The country came to be seen not only as a counterweight to the Soviet Union but as an economic engine for the world economy, an expanding market for US investment and exports, and a source of cheap consumer goods helping keep prices low in America’s supermarkets.
A stark contrast emerged over time as China’s political regime hardened and its economic success came to be seen by US elites less like an opportunity and more as a threat. Whatever role China’s exports might have played in America’s deindustrialisation, stagnant wages and growing joblessness in lagging regions, and decline in the middle class overall, much of the blame was placed on China’s feet.
Such a dramatic shift in policy attitudes towards another country in a comparatively short time raises many questions about its origins, spread across different policy actors, and scale in comparison with other similar changes in history. In a recent paper, we focus on the narrative discourse on China (Geyiktepe and Rodrik 2026). We place the change in the US discourse on China in historical perspective and compare its evolution across different sources of public narrative. When did official and public attitudes towards China begin to change? How does the current wave of negative discourse on China compare to previous historical episodes of realignment? How do presidential, congressional, media, and other sources differ in timing and scale of their shifts in attitudes? How do changes in narrative tone diffuse across actors?
To answer such questions, we use a large corpus of materials drawn from public records of various kinds (presidential and congressional records, newspaper articles, social media, think tank reports) and employ text-as-data methods that have become increasingly common in economics to quantify discourse (Gentzkow et al. 2019, Ash and Hansen 2023).
We develop and apply two measures of narrative discourse: salience and sentiment. The first is an indicator of the frequency with which China is covered in a particular text. The second is a proxy for how positively or negatively China is presented in the relevant passage.
We track the salience of China in a particular text through an exposure measure similar to that of Hassan et al. (2025): we calculate the number of sentences on China in a given document as a percentage of all sentences. This entails constructing a lexicon, or a dictionary, of keywords that is used to detect sentences discussing China in the text.
Our sentiment classification task requires a targeted approach, which is not based on identifying the sentiment of text as a whole but on specifically how China is discussed in that window. A single sentence might use ample positive language while portraying China in a negative light.
An effective way to address this is to carry out the classification task using a large language model (LLM). The use of LLMs as a tool for economics research has been growing with increased applications (Korinek 2023, Celebi and Penczynski 2024, Dell 2025). We use GPT-4o mini to label the tone of the passage as positive, negative, or neutral. This LLM approach allows us to craft a specific prompt tailored to our classification task. We further leverage few-shot prompting by providing example labels that serve as edge cases (Brown et al. 2020). Importantly, we use paragraphs instead of individual sentences, which can omit important context and prove misleading.
Figure 1 Sentiment classification comparisons of models
We compare our classifications of the China paragraphs from the presidential sample with results obtained using several established lexicons and transformer-based language models. We document that bag-of-words methods using dictionaries that categorise words into pre-determined sentiments do not do as well as our LLM approach. A DistilBERT based classifier, which evaluates the tone of the full passage by incorporating context, does better and tracks our results more closely. The closest match to our approach is obtained when we use an aspect-based sentiment analysis (ABSA) model that targets China as the concept of interest in the text.
Our presidential data source goes back to the early days of the republic, allowing us to take a very long-run view of salience and narrative sentiment on China. Furthermore, a long-term historical overview provides an informal test of the capacity of our exposure and net sentiment metrics to capture and quantify the phenomena we are interested in.
Figure 2 China in presidential speeches: Exposure and net sentiment
In the earlier relations between the two countries, sporadic engagements that spark discussion appear to be the pattern. The trend changes in the latter half of the 20th century. Exposure rises in the 1970s, following Richard Nixon’s official visit to China in 1972, after flatlining for over a decade. Notably, from this point on, we observe some oscillations, but their scope is much more limited compared to the patterns in the late 19th and early 20th century.
Similar to the patterns in our exposure measure, our sentiment metric tracks well with historical developments between the US and China and the framing of China in presidential discourse. Whereas the 19th century demonstrates oscillating trends, a clearer pattern emerges in the latter half of the 20th century, one driven less by fluctuating developments and more by ideological and political patterns. Net sentiment sharply declines around 1950 and turns negative. The escalation in tension resulting from a communist revolution in China was compounded by China’s entry into the Korean War and its direct armed confrontation with the US in 1950. At this point, presidential discourse dips to a historic negative and persists for nearly two decades.
The reversal of net sentiment back to positive levels follows once again a major shift in US national policy. Following Nixon’s first official visit in early 1970s, US policy towards China becomes drastically more conciliatory and amicable. Another major shift unfolds beginning in the 2000s as net sentiment declines and becomes ever more negative gradually. This parallels an increasing sense of competition that emerges with China’s growing economic and political power. The election of Donald Trump, alongside the tariffs targeting China, seem to have compounded the process. But it is also clear that the change in sentiment towards China predates Trump. The net sentiment metric reflects the broader shift in this recent period rather than a specific action or a particular president.
Since we use multiple sources (though these do not go as far back as the presidential records), we are able to delineate the ebbs and flows of the US relationship with China both at the elite and the grassroots level and across different types of political actors. To capture this diverse landscape of narrative discourse, we incorporate congressional speeches, newspaper articles, Reddit posts, and think tank publications into our corpus.
Figure 3 China exposure and net sentiment across sources
China exposure measures track similarly across the actors for decades, with think tanks as the standout. While exposure appears to be stable across actors for the most part, a rise beginning around 2010 is apparent for all of them. The uptick is more pronounced for think tanks, presidential coverage, and elite newspapers. This reflects how China came to become a focal national policy issue in the 21st century, with all parties involved in the policymaking process devoting more attention to the subject.
Notably, the overall trends in net sentiment suggest an interesting contrast between presidential and other sources. Beginning with the 1970s, presidential discourse around China assumed a positive tone and persisted for decades. A general downward trend only emerges in the 2000s and net sentiment turns negative only after 2008. For all other actors, net sentiment has been negative throughout, with a slight exception for non-elite media during the 1970s.
What these results reveal is that presidential sentiment on China has long been an outlier. Only recently has there been some convergence with the prevailing negative sentiment in all our other sources. The shift in the presidential discourse on China, notable as it is, seems to reflect a catching up to other sources of narrative discourse. Far from leading a change for the country as a whole, presidential speech appears to have been a follower.
Source : VOXeu
In Latin America and the Caribbean (LAC), one in five people report experiencing or knowing…
Commodity booms are easy to celebrate but hard to manage. For resource-rich economies, commodity windfalls—if…
Trade in critical metals essential for electrification has become centred on China over the last…
The voluntary carbon market was built on a single promise: that one credit cancels one…
Macroprudential policy is designed to make the financial system safer, but it can also reshape…
Anti-immigration policies often rest on the premise that immigrants worsen native-born employment and wages. This…