Congress Tweets

In April 2019, the China Data Lab started the “Congress Tweets” project. Since then, we have utilized Twitter’s API to gather 831,331 tweets, including 10,938 tweets related to China, authored by members of Congress.

Building off previous blogs in China Data Lab’s Congress Tweets series, we now will discuss the sentiment of tweets, the emotional content of Congress’s tweeting about China.

Part III: How does Congress Feel about China?

Harris Doshay, Lei Guang, Zeyu Li, Bailey Marsheck, Molly Roberts, and Young Yang

Up to now, our Congress Tweets series has focused on the who and what of Congressional tweeting on China. Starting here, we’ll be diving into sentiment, the emotional content expressed in the tweets, to see how the different parties and blocs within them relate to China online. You can find a discussion of how we coded sentiment in the notes section, and we’ll dedicate this post to discussing some of the basics of how it changed over time from 2017-2021 and whether it differs between the two political parties. Overall, we find interesting bipartisan similarity in this measure since around the start of the trade war in 2018, with negativity dominating the tweets in our database.

① Starting in 2018, Both Parties Went Equally Negative on China.

Figure 1. Congressional Sentiment toward China (Jan. 2017 – July 2021)

The charts above give us some indications about publicly tweeted Congressional attitudes towards China. What’s clear is that Republicans and Democrats both lean negative in terms of sentiment towards China, with each party’s average sentiment below the neutral rating of three at all times. Positive sentiment tweets were rare. Our findings are noteworthy for their consistency over time, as they do not show a sharp dropoff or uptick in any period, with the exception of Republican tweets in the first half of 2017 showing a bump in sentiment towards China before Trump’s summit with Xi at Mar-a-Lago in April 2017. Furthermore, there appears to be relative bipartisan convergence, with the mean sentiment similar in most periods. At first glance, “China bashing” appears to be a rare bipartisan consensus.

Figure 2. Congressional Sentiment toward China, Canada and Iran

Congressional sentiment toward China moved closer to its sentiment toward Iran after 2018.

How, then, does Congressional sentiment toward China compare to how MCs feel about other countries? We sampled 1,000 tweets by MCs discussing Iran and another 1,000 discussing Canada as baselines for discussion, using these two countries as the two poles representing “unfriendly” and “friendly” countries respectively.

Our data demonstrates that MCs’ sentiment towards China is much closer to their sentiment towards Iran, a country that is commonly positioned as an “enemy” of the United States, far below their sentiment towards a traditional “friendly” country like Canada.

There is, however, slightly more variation in tweets about China than those about Iran. It could be that discussions of China are more sensitive to domestic and international political events than Iran, where negative views are not yet as consistent in the case of China as they seem to be the case with countries like Iran and North Korea. As we will discuss in our next blog, trade with China has often had a positive dimension that congress likes to highlight to their constituents.

② Sentiment Doesn’t Change Much with Major Events


In the chart above, we zoom in on the initial divergence noted previously, with Republicans having slightly higher sentiment than Democrats early in the Trump administration. We can see that the initial convergence in negative sentiment happens around the first quarter of 2018, the time of the steel and aluminum tariff and the onset of Trump’s intensifying trade war. In our previous post, we discussed that there was a category of trade news that included relatively positive tweets by the MCs. It is possible that Congressional Republicans finally rallied around Trump’s negative views towards China after 2017, and their tweets started to converge to similar rhetoric on trade after the time of the tariff.

To see the change in sentiment during different periods, you can use the dropdown menu on the above chart. Perhaps surprisingly, neither COVID during its initial outbreak in China and subsequent spread in the States nor the outbreaks of protests and widespread repression in Hong Kong provoked a deterioration in sentiment similar to that provoked by the trade war. At this point, it may be that sentiment was already so negative there was little room for further downward spiraling. Once sentiment hit a “floor,” it was hard to go much farther down.

In summary, our initial analysis of sentiment leaves three takeaways. First, sentiment towards China is predominantly negative, close to the extremely low sentiment toward Iran. Second, the two parties are, for the most part, equally negative. The last period when there was divergence in Congressional sentiment towards China was before the trade war. Since then, negative sentiment seemed to have hardened inCongressional views of China, such that neither the outbreak of protests in Hong Kong nor a pandemic raging across the world was sufficient to cause a further fall in sentiment.

This leaves us with a few unanswered questions. Are there any areas of significant disagreement within each party when it comes to views of China? How do ideas of left-right politics map onto this sentiment-reading? Is this sentiment pattern reflective of deeper disputes between the two countries, or has China just been caught up in the crossfire of American domestic disputes? We’ll wrap up our “Congress tweets” series in our next installment by answering these questions.


Harris Doshay, Assistant Director of Research and Writing, 21st Century China Center, UC San Diego School of Global Policy and Strategy

Lei Guang, So Family Executive Director, 21st Century China Center, UC San Diego School of Global Policy and Strategy

Zeyu Li, Master of Chinese Economics and Political Affairs (MCEPA) candidate, UC San Diego School of Global Policy and Strategy

Bailey Marsheck, Chinese Language Fellow for the National Bureau of Asian Research, Tsinghua University's IUP Program; holds a bachelor's degree from UC San Diego and a master’s degree from Peking University

Molly Roberts, Associate Professor, Department of Political Science and the Halıcıoğlu Data Science Institute, UC San Diego; Co-Director, China Data Lab at the 21st Century China Center, UC San Diego School of Global Policy and Strategy

Young Yang, Research Data Analyst, China Data Lab at the 21st Century China Center, UC San Diego School of Global Policy and Strategy

Methodological note

Coding Sentiment

We had a two-step process for classifying the sentiment of tweets that mentioned a given country. First, we threw out all tweets that didn’t direct sentiment towards the country in question. These could be a tweet that had logistical information, like saying a given member of Congress met with a Canadian delegation, or could have offhand mention of the country with the main point of the tweet directing sentiment at another party. The second step involved digging further into those tweets that mentioned a country and directed sentiment towards it. This second group makes up the bulk of the tweets for our analysis.

We ranked each tweet with sentiment from 1-5, with one being the most negative and five being the most positive. Posts ranked one and five were reserved for explicitly negative and positive posts, respectively. Those ranked two and four had implicit or moderate negativity or positivity. Finally, those tweets ranked three either expressed a mix of negative and positive or neutral sentiment (but not purely factual information). This last group was relatively rare. Each tweet was manually ranked by two trained coders, and, when disagreement occurred, a .5 rating was given if they were only one apart. If there was greater disagreement a third party was brought in to have a discussion and/or provide a ruling. Tweets after 2021 had significantly less disagreement than prior tweets due to experience garnered by coders causing greater levels of inter-coder agreement.