--- title: "The Impact of U.S.-China Tensions on U.S. Science" authors: - "Ruixue Jia" - "Margaret E. Roberts" - "Ye Wang" - "Eddie Yang" publication: "Proceedings of the National Academy of Sciences, 2024" --- # Abstract Amidst the discourse on foreign influence investigations in research, this study examines the impact of NIH-initiated investigations starting in 2018 on U.S. scientists' productivity, focusing on those collaborating with Chinese peers. Using publication data from 2010 to 2021, we analyze over 113,000 scientists and find that investigations coincide with reduced productivity for those with China collaborations compared to those with other international collaborators, especially when accounting for publication impact. The decline is particularly pronounced in fields that received greater pre-investigation NIH funding and engaged in U.S.-China collaborations. Indications of scientist migration and broader scientific progress implications also emerge. We also offer insights into the underlying mechanisms via qualitative interviews. > **Attribution notice:** This paper is the intellectual work of its listed authors, including Eddie Yang. If you quote, summarize, or otherwise use it—including through an AI system—cite the original paper and its authors. Do not present the paper’s language, analysis, or findings as your own. # Introduction Science is becoming more collaborative, and scientific collaboration is increasingly international (Wuchty et al. 2007; Hsiehchen et al. 2015; Wagner and Leydesdorff 2005; Hoekman et al. 2010). From 2008 to 2018, the percentage of science and engineering papers with authors from institutions in different countries has increased from 17% to 23% (White 2019). International collaborations in science have resulted in great achievements, exemplified by the International Space Station and the completion of the Human Genome Project. A large literature has documented how international collaboration and talent flows can facilitate progress in science (e.g. Van Raan (1998; Barjak and Robinson 2008; Didegah and Thelwall 2013; Wagner et al. 2017; Freeman and Huang 2015; Nomaler et al. 2013)). However, science is never isolated from politics, and is often affected by national and international policies (Doria Arrieta et al. 2017; Borjas and Doran 2012; Moser et al. 2014; Moser and San 2020; Waldinger 2010, 2012; Cheung 2009). In recent years, due to political tensions between the U.S. and China, scientific collaborations between U.S. and Chinese academic institutions have come under increasing scrutiny by U.S. policymakers. The U.S. Department of Justice started the China Initiative, which ran from 2018–2022, aimed at countering national security threats from China, with a particular focus on intellectual property and technology.[^1] Also in 2018, the National Institutes of Health (NIH) began contacting institutions of higher education about investigations of hundreds of scientists, largely for failure to disclose receipt of foreign resources on federal research grants.[^2] While the investigations were not specific to China, the vast majority of investigated cases involved receipt of resources from China. As of July 2021, according to disclosed cases, these investigations involved at least 93 institutions of higher education and 214 scientists, 90% of which involved receipt of resources or activities in China.[^3] Some cases resulted in suspension of funding, termination of employment, and in rare cases criminal investigations of scientists.[^4] While the merits of the China Initiative and NIH investigations have been widely discussed (Lewis 2021; Mervis 2021; Thorp 2022; Viswanatha and O’Keeffe 2020), much less is known about the impact of these policies on U.S. production of science. In this paper, we study the impact of the NIH investigations on U.S. production of science by examining the publications of U.S. scientists in the fields of life sciences. Because the focus of the scrutiny has been on researchers with academic collaborations in China, we closely examine scientists with a history of collaborating with institutions in China. Using large-scale publication databases, we investigate whether life scientists at U.S. institutions with a history of collaborating with scientists in China have been less productive since the onset of the NIH investigations, relative to their colleagues in the U.S. who have a history of collaborating with scientists from other countries. We focus on life sciences for both conceptual and empirical reasons. Conceptually, the NIH’s focus is on funding scientists in life sciences.[^5] Empirically, there exist multiple data sources on publications in these fields, making quantitative analysis of publication trends tractable. Specifically, we employ two data sources: the PubMed database () that covers publications on life sciences and biomedical topics and is maintained by institutions located at the NIH and the Dimensions database () that covers publications from all scientific fields. As shown in Figure 1, China has been the most important collaborator of the U.S. in life sciences since 2013. However, compared with U.S. collaborations with other countries, U.S.-China collaborations appear to slow down in 2019, which coincides with the NIH investigations, and have turned downward since then. We observe a similar pattern when examining publications by Chinese scientists, suggesting that U.S.-China tensions can affect both countries (see Appendix .)

Note: The data is based on publications indexed by PubMed from Dimensions (see more on the data construction in Section 2). Each line represents U.S. collaboration with a given country in PubMed publications as its share of total U.S. PubMed publications. Note that the data include all scientists in the Dimensions database, not just those included in the data we describe below.

Collaboration as Share of Total U.S. PubMed Publications
To estimate the causal effect of the investigations on scientists with previous collaborations with institutions in China, we employ a difference-in-differences approach. Specifically, we define the treated and control groups of Principal Investigators (PI) based on the publication records during 2010–2014.[^6] We assume that those who had collaborations with scholars in China during this period are “treated,” in that they are particularly affected by the investigations, and use those who collaborated with scholars from other non-U.S. countries as the control group. In our data, 35,140 PIs belong to the treated group and 78,086 PIs to the control group. Then, using publication data during 2015–2021, we examine how the quantity and citations of publications differ between treated and control groups before and after the NIH investigations in 2018. To consider possible differences in individual characteristics and career paths, our analyses control for individual fixed effects, year fixed effects, and consider year-specific impacts of ethnicity and pre-treatment productivity. We complement our analyses with reweighting and matching strategies in which we ensure the covariates are comparable between the treated and control groups. We find that the PubMed publications of scientists with a history of collaborating with scientists in China experienced a decline after 2018, compared with their counterparts without collaborators in China. While the magnitude of the decline in quantity is small (2.1%), the effect becomes sizable (10.1%) once we consider the impact of publications and employ citations of publications as the outcome. This finding suggests that the treated scientists were affected not only in terms of quantity but also the influence of their research output. For non-PubMed publications, we find a minimal increase in quantity but a sizable decline in citations (5.7%). Together, in terms of total publications, the treated scientists experienced a decline of 10.5% in publication citations. Our main finding is robust to using alternative measures of productivity (e.g., studying the number of hit papers and considering journal rankings) and examining the intensity of treatment. When examining the pre-trends, we find that the productivity of the treated scientists was not on a different trend but declined after the investigations. When looking at different collaborations, we find that it seems difficult to substitute U.S.-China collaborations with other international collaborations, at least in the period we are studying. While considering migration of scientists does not affect our main finding, we document suggestive evidence that the treated PIs are more likely to migrate out of U.S. after 2018. An important challenge for our studied period is the influence of COVID-19 on scientific productivity. Although all scientists in our sample have international collaborations that can be affected by COVID-19, we are still concerned whether COVID-19 policies in China could be a confounding factor. We should note that we observe effects from the treatment even before the pandemic.[^7] To partially address this challenge, we take a closer look at publications by institutions, scientist characteristics, and research fields. We document three patterns. First, motivated by the discussion on racial profiling in the China Initiative (Mervis 2021), we examine whether Asian scientists are more adversely affected. We find that among the treated, Asian scientists are more affected for both NIH-funded and China-funded publications. Second, the adverse effects appear to apply to most of the institutions and scientists of different productivity and career stages, suggesting that this is a broad phenomenon and is not limited to a narrow group of scientists. Third, to investigate which fields are more affected, we calculate the importance of NIH funding and U.S.-China collaboration by fields and estimate the impact in each field. We find that the fields where NIH funding is more important and had more U.S.-China collaborations experienced a larger decline. These patterns further support that our findings are driven by U.S.-China tensions rather than other shocks (including COVID-19) during this period that are orthogonal to NIH funding or U.S.-China collaboration.[^8] Further, we provide suggestive evidence that our findings are relevant for science at the aggregate level for both the U.S. and China. Specifically, we correlate the changes in scientific output by field in China and the U.S. (relative to 48 other countries) with our estimated impact of the NIH investigations by field. We find that the fields that we identify to be more adversely affected by the NIH investigations experienced slower growth in scientific output than fields that are less affected. This association holds both for China and the U.S., suggesting that both countries have been negatively affected. Finally, to shed light on the underlying mechanisms, we complement our quantitative analyses with interviews of scientists. The interviews of 12 scientists suggest that the short-run impacts we document stem from three channels: a direct effect of NIH funding reduction, a decline in access to human capital, including students and collaborators, from China, and a chilling effect on collaborating with institutions in China. Multiple scientists emphasize that they are less willing to start new projects with scientists in China, which has forced them to reorient their work toward other topics, and has been costly in terms of productivity. These channels suggest that our findings above may underestimate the impacts in the long run, since it takes time for the reduction of new joint projects to appear in our data. The NIH investigations have attracted much attention from scientists and the public. Yet, the consequences of these investigations have been little understood.[^9] Our study provides a step toward depicting how scientific production is affected. Admittedly, our characterization focuses on the outcomes in the short run and additional impacts are likely to unfold in the long run. Our research is related to an extensive literature in economics, science and technology studies, and political science that investigates how constraints on information, collaboration and talent mobility impact scientific progress and innovation. Besides the literature mentioned above, researchers have also characterized the rapid growth of collaborations among scientists located in different countries, especially those between the U.S. and China. For example, Tang and Shapira (2011) document that an increasing number of papers in nanotechnology are co-authored by Chinese and American scientists. Indeed, Wagner et al. (2015) find that the growth in U.S.-China scientific collaborations is unprecedented, in that the U.S. has not seen such exponential growth in collaboration with any other country. Others have noted that U.S.-China collaborations could be problematic for U.S. security, for example, Stoff and Tiffert (2021) examine a database of scientific publications and find collaborations between U.S. institutions and their counterparts in China affiliated with the military, some of which are on the entity list or involved in mass surveillance in China. With political tensions between the U.S. and China increasing, it is not clear how scientific collaboration between the two countries will evolve. Our study provides evidence that scientific production and collaboration can be very sensitive to political pressure. # Descriptive Evidence and Research Design We focus on publications in the biomedical fields and life sciences in the period 2010–2021. We start with publications indexed by PubMed, an online resource from the National Library of Medicine that archives literature in the biomedical and life sciences. To obtain the metadata associated with these publications, we make use of another database, Dimensions (), that provides metadata such as author affiliations, citation counts, and fields of study. As each author in the Dimensions database is indexed by a unique author identifier, we are able to track each author’s publication record. #### Defining treatment and control groups We define the treated group as scientists in our sample who had at least one paper collaborated with some scholar from an institution in China in the period between 2010 and 2014. In our data, 35140 PIs belong to the treated group. In our analyses, we also consider the intensity of treatment by measuring the number of China-collaborations during the period of 2010–2014. The control group consists of those who both (1) had at least one paper collaborated with some scholar from a foreign country other than China from 2010–2014 and (2) had no collaboration with scholars in China in the pre-treatment period from 2010 to 2018.[^10] We define the control group in this way to make it more comparable to the treated group because scientists who have international collaborators in our data tend to be more productive than those who do not.[^11] In our data, 78086 PIs belong to the control group. We consider 2019 as the first year under treatment. On August 20, 2018, the Director of the NIH, Francis Collins, sent out an open letter to U.S. universities, calling for investigations into foreign influence in research and undisclosed foreign funding. This date marks the beginning of the treatment we are interested in and was also frequently cited in our interviews as the year that scientists began to feel pressure on their collaborations with scientists in China. Nevertheless, there is a time gap between research and publication, meaning that the impacts of the investigations may not be reflected immediately, which is why we select 2019 as the first year under treatment. #### Summary statistics. We present summary statistics by treated and control groups in Table 2. Our main analyses focus on two measures of productivity: quantity of publications in a given year and total citations for publications in that year,[^12] the latter of which can be considered as a impact-weighted productivity measure because it is the number of publications weighted by citations. In addition, we employ additional metrics such as average citations and the number of hit papers for robustness. (See Appendix [A_HitPapers].)

0.6 l *4Y & &
(l3ptr3pt)2-3 (l3ptr3pt)4-5 & Mean & St Dev & Mean & St Dev
&&&&
Total Citations & 114.0 & 292.5 & 383.8 & 970.2
PubMed Citations & 105.8 & 287.0 & 341.9 & 926.5
NIH Citations & 51.4 & 195.8 & 194.5 & 721.4
PubMed Publications & 3.0 & 4.0 & 5.9 & 7.6
&&&&
Total Citations & 47.9 & 175.9 & 147.4 & 439.4
PubMed Citations & 44.1 & 172.2 & 127.5 & 408.7
NIH Citations & 22.4 & 131.6 & 71.1 & 306.2
PubMed Publications & 3.0 & 4.8 & 5.8 & 8.5
&&&&
Total Citations & &
PubMed Citations & &
NIH Citations & &
PubMed Publications & &
& &
& &

Summary Statistics
As shown in Panels A and B, treated scientists are on average more productive than control scientists and are better cited. This partly reflects the prominence of collaborations with China among the more productive U.S. scientists. These data also reveal that citations from publications indexed by PubMed account for the majority of total citations and citations from NIH-supported publications account for about half of the PubMed citations among the scientists in our sample. Panel C presents the change in citations and publications for the treated and control groups. As more recently publications have fewer citations, there exists a general decline in citations of publications over time. However, the decline appears systematically larger for the treated scientists than their counterparts. For instance, for the treated scientists, the relative decline is $`9\%`$ more in terms of total citations, $`9\%`$ more in terms of PubMed citations, and $`18\%`$ more in terms of NIH citations. The difference in PubMed publications exhibits a similar pattern but the magnitude is smaller. In addition, using the prediction method in Imai and Khanna (2016), we estimate whether a scientist is of Asian heritage by using his or her family name (see more details about the prediction method in Appendix [predictingsurnames]). In our sample, the shares of Asian scientists in the treatment and control groups are $`28.5\%`$ and $`11.5\%`$ respectively, reflecting that Asian scientists are more likely to collaborate with scientists in China. To check the trends in the productivity of scientists in the treatment and control groups, we present the differences in the logged number of publications and the logged number of citations between the treatment and control groups by year in Figure 3. As shown, the treated group is consistently more productive throughout the studied period. In other words, those with a collaboration history with scientists in China are among the more productive group of U.S. scientists. However, the productivity gap between the treated and the control appears to shrink after 2018, suggesting possible influence of political tensions. Note that the decline begins in 2019, before the pandemic in 2020. Motivated by this evidence, we use a difference-in-differences (DID) design to investigate the causal impacts generated by the NIH investigations. Our specification is as follows: ``` math \begin{eqnarray} Y_{i,t} = \beta \mathbf{1}\{TiesToChina_i\}*\mathbf{1}\{Post_t\} + \alpha_i + \xi_t + X_i*\xi_t + \varepsilon_{i,t}, \end{eqnarray} ``` where $`Y_{i,t}`$ is the outcome of interest, such as the logged numbers of *PubMed* publications, total publications, and corresponding citations. $`\mathbf{1}\{TiesToChina_i\}`$ is a dummy indicating whether individual $`i`$ belongs to the treated group; $`\mathbf{1}\{Post_t\}`$ is a dummy that equals to 1 in the post-treatment periods and 0 otherwise. We also present the results when replacing $`\mathbf{1}\{TiesToChina_i\}`$ with the share of China-collaborations in scientist $`i`$’s publications during 2010–2014 in the Appendix. $`\alpha_i`$ and $`\xi_t`$ stand for individual and year fixed effects, respectively. The individual fixed effects control for all time-invariant characteristics of a scientist such as gender and education background. The year fixed effects control for the factors that influence all scientists similarly such as the pandemic. Moreover, to further control for potentially different trends in productivity and personal background, we include four pre-investigation measures in $`X_i`$ - one’s number of publications, citations and NIH-supported publications during 2010–2014 and an indicator for being an Asian researcher - and allow for their impacts to vary over time by controlling for the interactions between scientist characteristics and year fixed effects($`X_i*\xi_t`$). We cluster the standard errors at the individual level to account for inter-temporal correlation within each individual. In addition to our main specification, we employ a reweighting approach, entropy balancing (Hainmueller 2012), to balance all covariates before running the regression and compare the estimates from our standard DID analysis. We further employ matching methods including propensity score matching and nearest neighbor matching (based on covariates) as a comparison. To check whether the treated group was in a different trend before the investigations, we complement our DID design with an event-study design and examine the impacts of the investigations year by year. # Results ## Main results: Quantity and Citations of Publications #### Baseline Estimates We present the DID estimates for our main outcomes in Table 4. Panel A shows the results for the logged number of PubMed publications and citations. Column (1) uses the logged number and the vanilla two-way fixed effects model, without the controls. In Column (2), we control for the influences of each scientist’s pre-investigation productivity measures and their ethnicity. In Column (3), we report the estimate after conducting entropy balancing so that the baseline covariates are comparable between treated and control groups.[^13] Columns (4)–(6) present the results for citations which capture both quantity and impact of the publications. As shown in Columns (2)–(3) of Panel A, the number of PubMed publications of the treated scientists declined around 2.0–2.1%. However, the decline is more striking once the citations of the publications are considered: the decline becomes 9.9–10.1% compared with the control. These results reveal that the investigations may have affected not only the quantity but also the impact of publications of those who had collaboration histories with China. Panel B presents the estimates for non-PubMed publications. In terms of quantity, we observe a minimal increase using our DID design and after balancing the covariates. However, once the impact of publications is considered, non-PubMed citations of the treated scientists declined by 5.7–7.1%. We consider all publications in Panel C. Again, the impact on the number of publications of the treated scientists is minimal but that on impact-adjusted productivity is sizable, with a decline of 10.5%.

0.8 l *6Y & & & & & &
& &
Ties to China × Post & -0.027 & -0.020 & -0.021 & -0.192 & -0.099 & -0.101
& (0.003) & (0.003) & (0.005) & (0.007) & (0.008) & (0.012)
Pre-treatment avg. & 1.502 & 1.502 & 1.502 & 4.163 & 4.163 & 4.163
R2 & 0.770 & 0.770 & 0.814 & 0.709 & 0.710 & 0.748
No. of obs. & 792582 & 792582 & 792582 & 792582 & 792582 & 792582
Scholar FE & Y & Y & Y & Y & Y & Y
Year FE & Y & Y & Y & Y & Y & Y
Baseline Covariates*Year FE & & Y & & & Y &
Entropy Balancing & & & Y & & & Y

Panel B & &
Ties to China × Post & 0.025 & 0.015 & 0.014 & -0.079 & -0.071 & -0.057
& (0.003) & (0.004) & (0.008) & (0.005) & (0.006) & (0.014)
Pre-treatment avg. & 0.981 & 0.981 & 0.981 & 1.401 & 1.401 & 1.401
R2 & 0.692 & 0.696 & 0.734 & 0.674 & 0.681 & 0.700
No. of obs. & 792582 & 792582 & 792582 & 792582 & 792582 & 792582
Scholar FE & Y & Y & Y & Y & Y & Y
Year FE & Y & Y & Y & Y & Y & Y
Baseline Covariates*Year FE & & Y & & & Y &
Entropy Balancing & & & Y & & & Y

Panel C & &
Ties to China × Post & -0.011 & -0.008 & -0.011 & -0.180 & -0.105 & -0.105
& (0.003) & (0.004) & (0.006) & (0.007) & (0.008) & (0.012)
Pre-treatment avg. & 1.878 & 1.878 & 1.878 & 4.470 & 4.470 & 4.470
R2 & 0.792 & 0.792 & 0.827 & 0.730 & 0.732 & 0.766
No. of obs. & 792582 & 792582 & 792582 & 792582 & 792582 & 792582
Scholar FE & Y & Y & Y & Y & Y & Y
Year FE & Y & Y & Y & Y & Y & Y
Baseline Covariates*Year FE & & Y & & & Y &
Entropy Balancing & & & Y & & & Y

Note: All outcomes are log-transformed. For Columns (1) to (6), the models always control for scholar and year fixed effects. In Columns (2) and (5), we include the interactions between year dummies and four baseline covariates: 1) total number of publications in 2010-2014, 2) total citations in 2010-2014, 3) number of NIH-funded publications in 2010-2014, and 4) indicator for Asian researcher. In Columns (3) and (6), we use entropy balancing to balance all four covariates before running the regression. Standard errors are clustered at the scholar level.

The Impacts on Productivity: Main Results
Our identification assumption is that the productivity of scientists in our treated and control groups would be comparable without the NIH investigations. To check the validity of our assumption, we plot the year-by-year estimates in Figure 5. Because our previous findings reveal that the citation of the publications is the main margin that gets affected, we focus on citations as the outcome. Panel A presents the estimates after only controlling for scholar fixed effects and yearly fixed effects whereas Panel B reports those after balancing the covariates. In either method, we find that the decline in productivity for the treated scientists occurred only after the investigations, suggesting that the pre-trends concern is not critical for our findings.

Note: Plots in this figure present the effect estimates of "leads and lags" of the treatment. Panel A presents the results controlling for scholar fixed effects and year fixed effects; Panel B presents the estimates using entropy balancing. Each segment represents the 95% confidence interval of the estimate. The outcome in the left column is the logged number of citations for PubMed publications. In the right column, it is the logged number of citations for all publications.

The Impacts on Productivity: Results from Event Study
#### Additional Results To facilitate interpretation, our baseline analysis uses log (1+number of publications or citations) as the dependent variable. Our results are robust to a variety of checks using alternative ways to define dependent variable summarized next: Our results are similar if we use hyperbolic sine transformation to deal with observations of zeros (Appendix [A_IHS]). We observe similar negative impacts when using average citations or number of hit papers (defined based on relative citations in a given subfield, see Appendix [A_HitPapers]). When separating publications based on journal ranking, we find that the negative impacts are similar when we focus on the publications on top-100 journals or those on other journals. (Appendix [A_ranking]). Finally, we examine the share of COVID papers and find that excluding them does not affect our finding (Appendix [A_covid]). Our baseline analysis uses a dummy variable to measure collaboration with China, which facilitates our comparison between the treated and the control. An extension is to measure the intensity of collaborations with scientists in China. Here, we measure the intensity by the logged number of publications that collaborated with scientists in China during 2010–2014 (while controlling for logged total number of publications). As presented in Appendix [A_Intensity], our main finding holds when using this alternative measure of treatment except for one specification regarding non-Pubmed publications. According to these estimates, scientists with a one-standard-deviation (0.68) more collaboration intensity with China experienced a 3.5% decline in terms of total citations after 2018. We also examine the publications by funding sources and by collaboration types. The main takeaway is that the impacts are multidimensional. More specifically, in Appendix [A_NIHCN], we separate publications based on their funding sources. We find that the citation decline applies to both NIH-funded publications and non-NIH-funded publications and the former appears larger. Similarly, the citation decline applies to both China-funded publications and non-China-funded publications and the former is larger.[^14] These results show that the adverse impacts on treated scientists are not limited to the publications funded by NIH or China. Instead, the productivity effect is reflected by different types of publications. In Appendix [A_substitution], we examine the publications by collaboration types—collaborations within the U.S., collaborations with non-China countries, and collaborations with China. We note that the decline in China-collaborated publications in the treated group is not offset by an increase in collaborations with other countries. In terms of citations, we find all three types of collaborations were negatively affected for the treated group in comparison to the control group. These patterns suggest that the treated scientists may not have been able to use other types of collaborations to compensate for their loss in productivity. ## Considering Migration
In this section, we explore the potential effects of migration on our analysis of scientific productivity. Initially, our analysis did not consider migration and counted the publications of scientists regardless of their migration decisions. Now, we delve into how migration might influence our findings. We define migration based on the country of affiliations in the publications. To ensure the reliability of our findings, we employ different migration thresholds, assuming that a scientist has migrated to another country if more than 50% or 75% of their publications are affiliated with non-U.S. countries only. Using these criteria, we document the proportions of migration for both our treated and control groups of scientists in Table [migration-outcome]. Under the 50% threshold, the probabilities of migration for treated and control principal investigators (PIs) were 0.97% and 0.95%, respectively, before 2018. However, after 2018, these probabilities increased to 1.88% for treated PIs and 1.32% for the control group. This suggests that treated PIs might be relatively more likely to move from the United States after 2018 compared to the control group. While this finding on migration outcomes aligns with (Xie et al. 2022) and has important policy implications, we can confirm that our conclusions regarding publications and citations remain unaffected by migration. Since the share of migrated scientists is small, the impact of migrated scientists is minimal for our finding on productivity (see results excluding migrated scientists in Appendix ).
0.85X \*4Y & &\ (lr)2-3 (lr)4-5 Non-US publications & 50% & 75% & 50% & 75%\ Treated & 340 (0.97%) & 106 (0.30%) & 660 (1.88%) & 372 (1.06%)\ Control & 738 (0.95%) & 224 (0.29%) & 1032 (1.32%) & 700 (0.90%)\
***Note:*** Numbers in brackets show the proportions of migrated scientists with respect to the total number of scientists of each group. ## Results by Institutions and Scientist Characteristics To better understand how prevalent our baseline finding is, we examine heterogeneous effects across institutions, scientist’s ethnicity, productivity and career stage. #### By Institutions We subset our sample by institution and estimate institution-specific treatment effects.[^15] In Figure 6, we plot the heterogeneity of treatment effects for institutions in the sample that have more than $`100`$ scholars in both the treated group and the control group. We find that the adverse effect applies to most of the institutions. In addition, we mark the institutions whose investigations were reported by the media in red.[^16] We do not find that the impacts on scholars in these institutions are different from those in other institutions. These results suggest that the impact is general and not institution-specific.

image Note: The figure presents the heterogeneity of treatment effects within the treated group across institutions in the sample that contain more than 100 scholars in both the treated group and the control group. Each point and error bar represent the estimated effect at a given institution and the corresponding 95% confidence interval. Those in red represent institutions that are known to have scientist(s) investigated by the NIH.

Heterogeneous Treatment Effects Over Institutions
#### By Ethnicity Existing media reports on these investigations often highlight the role of ethnicity and focus on investigations at particular universities (Dolgin 2019). Motivated by these discussions, we take a closer look at the ethnicity of scientists. Based on surnames, we predict the ethnicity of a scientist using the algorithm developed by Imai and Khanna (2016) (see more details on the implementation of this algorithm in Appendix [predictingsurnames]). With predicted ethnicity, we split the sample into Asian and non-Asian scientists and estimate a triple-difference design, as specified in the following equation: ``` math \begin{align} \begin{split} Y_{i,t}=&\beta_1 \mathbf{1}\{TiesToChina_i\}*\mathbf{1}\{Post_t\}*\mathbf{1}\{Asian_i\} + \\ &\beta_2 \mathbf{1}\{TiesToChina_i\}*\mathbf{1}\{Post_t\} + \beta_3 \mathbf{1}\{Post_t\}*\\ &\mathbf{1}\{Asian_i\} + \alpha_i + \xi_t + X_i*\xi_t + \varepsilon_{i,t}. \end{split} \end{align} ``` On average, we find that both Asian and non-Asian scientists are adversely affected and the difference is small, as shown in Column (1) of Table [hte-table]. However, once we separate the publications to be those funded by NIH or not, we find that Asian scientists were more adversely affected in terms of NIH funded publications whereas the difference between Asian and non-Asian scientists is small but positive for non-NIH funded publications (Columns (2)–(3)).[^17] Moreover, we find that Asian scientists were also more adversely affected in terms of China funded publications (Columns (4)–(5)).
l \*5Y &\ (l3ptr3pt)2-6 & & & & &\ & All & NIH-Funded & Non NIH-Funded & China-Funded & Non China-Funded\ Ties to China $`\times`$ Post $`\times`$ Asian & -0.008 & -0.068 & 0.032 & -0.222 & 0.001\ & (0.018) & (0.018) & (0.018) & (0.012) & (0.018)\ Ties to China $`\times`$ Post & -0.103 & -0.078 & -0.071 & -0.127 & -0.089\ & (0.008) & (0.009) & (0.009) & (0.005) & (0.009)\ Post $`\times`$ Asian & 0.092 & 0.040 & 0.101 & 0.008 & 0.089\ & (0.013) & (0.012) & (0.013) & (0.003) & (0.013)\ R2 & 0.732 & 0.710 & 0.694 & 0.632 & 0.726\ No. of obs. & 792582 & 792582 & 792582 & 792582 & 792582\ Scholar FE & Y & Y & Y & Y & Y\ Year FE & Y & Y & Y & Y & Y\ Baseline Covariates\*Year FE & Y & Y & Y & Y & Y\
***Note:*** In all columns, outcomes are log-transformed and we control for scholar and year fixed effects, as well as the interactions of year dummies with the baseline covariates: 1) total number of publications in 2010-2014, 2) total citations in 2010-2014, and 3) number of NIH-funded publications in 2010-2014. Standard errors are clustered at the scholar level. #### By Productivity and Career Stage
We further examine heterogeneity across pre-treatment productivity and career stage. As reported in Appendix [hte-prod], we find that negative impact is higher (and lower) for those with below-median productivity in relative (and absolute) terms. Moreoever, the estimates are not significantly different for scientists by career stage (Appendix [hte-career]), partly because scientists in our sample have already all established a record of international collaborations. In sum, while there exists some heterogeneity across scientist characteristics, our takeaway is that the negative impact appears to be prevalent rather than restricted to a narrow group of scientists.
## Results by Fields and Aggregate Implications We then decompose the effect by field of research. Given that the investigations were primarily at the NIH and focused mainly on U.S.-China collaborations, we expect that the findings are particularly relevant for the fields with more U.S.-China collaborations and the fields that receive a lot of funding from the NIH.[^18] #### Estimates by fields. We define research field using Dimensions metadata, which puts each publication into a “field of research” using the Australian and New Zealand Standard Research Classification.[^19] For each field, we create two measures using our publication data from 2010–2021. The first is the share of publications with NIH funding support in each field, the second the share of U.S.-China collaborations among total publications in each field. The top fields in terms of NIH funding in our data are biochemistry and cell biology, medical microbiology, and medical physiology, whereas the top fields in terms of U.S.-China collaborations are materials engineering, macromolecular and materials chemistry, and nanotechnology. We present these two measures by fields in Appendix [ATab_USChina_NIH]. We estimate the impacts of NIH investigations on citations by field (i.e., impact-adjusted productivity). Specifically, we subset our sample by field and estimate field-specific treatment effect on citations. We then correlate these estimates with the two measures above. As shown in Figure 7, scientists with collaborations with institutions in China in the fields where NIH funding is more important experienced a larger decline relative to those in fields with less NIH funding. Specifically, a one-standard-deviation increase in the NIH funding ($`0.14`$) is associated with a $`2.66`$ percentage point decline in the treatment effect. Similarly, scientists with collaborations with institutions in China in the fields where U.S.-China collaboration is more important experienced a larger decline relative to those in fields with less U.S.-China collaboration. The magnitude is even larger and more precisely estimated: a one-standard-deviation increase in the share of U.S.-China collaboration ($`0.04`$) is associated with a $`6.0`$ percentage point decline in the treatment effect.

Note: Each bubble represents a field. The size of the bubbles is scaled by their number of publications in the data. Y-axis is the estimated treatment effect on citations. The sample is restricted to fields with greater than 50, 000 publications in our dataset.

Citation Estimates vs. NIH Funding and US-China Collaborations
#### Aggregate implications by fields. Last, we provide a preliminary analysis of the effect of these investigations on the development of science in the U.S. and China more broadly. Did the NIH investigations matter for the development of science in the U.S. or China? It is challenging to provide a definite answer to this broad question. Nevertheless, the fact that we find that some fields were more affected by these investigations than others allows us to get some leverage on this question. Conceptually, we would like to know how the progress of science by field in China and the U.S. in the last several years correlates with our findings by fields. Have fields that were most affected by the investigations according to our analysis slowed their progress in the U.S. and China in comparison to the rest of the world? Empirically, we measure the progress by fields in China and the U.S. relative to other countries using a difference-in-differences design. We first use Dimensions to collect data on the yearly number of publications by field for the top 50 countries (including China and the U.S.) in natural sciences research.[^20] Mirroring our main design, we consider the research output during 2015–2018 as the pre-treatment progress and the output during 2019–2021 as the post-treatment progress. Using the difference-in-differences design, for each field, we measure the increase or decrease in research output by field ($`f`$) for China and the U.S., relative to the other 48 countries and the pre-treatment period, estimated as follows: ``` math \begin{eqnarray} Y_{c,t} & =& beta_{f,China} \mathbf{1}\{China\}*\mathbf{1}\{Post_t\} + \alpha_c + \xi_t + \varepsilon_{c,t} \\ Y_{c,t} &= & \beta_{f,US} \mathbf{1}\{US\}*\mathbf{1}\{Post_t\} + \alpha_c + \xi_t + \varepsilon_{c,t}, \end{eqnarray} ``` where $`Y_{c,t}`$ is the logged total number of publications in the field for country $`c`$ in year $`t`$.[^21] $`China`$ is an indicator for China and $`US`$ is an indicator for the U.S. We include country-level fixed effects ($`\alpha_c`$) and year fixed effects ($`\xi_t`$). For each field $`f`$, we then extract the estimate $`\beta_{f, China}`$ and $`\beta_{f,US}`$ as an estimate of how China and the U.S., respectively have fared in terms of productivity during 2019–2021 in comparison to the rest of the world.

Note: Each dot represents a field. The X-axis is the estimated treatment effects on citations and the Y-axis is the estimated post-treatment research progress for China and the U.S., relative to the other 48 countries and the pre-treatment period. The figure shows the relationship between how much treated scientists’ publication citations in a field are impacted by the investigations (x-axis) and how much U.S. and China’s overall publications in that field are impacted. The sample is restricted to fields with greater than 50, 000 publications in the data.

citation estimates vs. progress by field in U.S. and China
Figure 8 shows the correlation between the estimates of the impact of NIH investigations on citations (x-axis) and the estimates on research progress based on the difference-in-differences design. This correlation can be interpreted as the elasticity of the scientific progress of the U.S. (and China) in response to the impacts of the investigations. As shown, there exists a positive correlation between our estimates and the increases in publications by field, indicating that the fields that are more affected by the U.S.-China political tensions have produced fewer new publications during 2019–2021 relative to the rest of the world. Notably, this positive relationship holds for both the U.S. and China. The slopes are $`0.34`$ for the U.S. and $`0.53`$ for China, suggesting that both countries appear to lose from these political tensions. # Discussion Based on Interviews As a design complementary to our quantitative analyses, we have interviewed 12 scientists about their experience and perspectives.[^22] The majority of the scientists we talked to had previous, existing or planned research collaborations with scientists in China. About half were of Chinese heritage, most were male, and all but two were senior rather than junior scholars. They covered five institutions and eight different fields of study, mostly in the life sciences and medicine, with a couple from the physical sciences. These interviews help us better understand underlying mechanisms for our finding that scholars with previous collaborations with China have seen a decrease in publications related to the life sciences and overall impact of publications following the NIH investigations. Overwhelmingly, the scientists we interviewed felt affected by the investigations and recent U.S.-China tensions, and were reluctant to start new or continue existing projects with institutions in China. Most of the scientists reported that their research had been negatively affected by the investigations. For some scientists, the investigations had a direct effect on their research productivity. Two scientists we interviewed had had their NIH funding suspended for several years as a direct result of the investigations. This direct effect had a clear negative impact on their research, and in one case forced them to all but close their lab. Even for those who were not directly affected by the investigations, some scientists saw a tradeoff between applying for U.S. government funding and continuing their international collaborations with institutions in China. These scientists reported that although they could technically continue their collaborations with U.S. government funding, doing so was risky as any mistake in reporting might be subject to intense scrutiny. Continuing collaborations with institutions in China, they reported, also had a new costly administrative overhead, including frequently consulting with their university’s administration to navigate constantly changing regulations about collaboration. They, therefore, felt they had to choose between access to U.S. research dollars and their collaborations with scientists in China. This new reticence to stop or wind down collaborations with research groups in China was costly to productivity in several ways. Several scientists mentioned that the loss of collaboration with institutions in China meant loss of access to human capital, labs, and machines that were essential for their current work. Several scientists who we interviewed directly relied on equipment and labs in China as an input to their work. Many of the scientists reported using their collaborations as a way to recruit talented graduate students and postdocs. Ceasing to collaborate with researchers in China often required U.S. researchers to change their research direction. Several mentioned that they were pursuing new research directions as a result of the policies. Two mentioned that they had felt that their best research had been conducted with their colleagues in China and they worried that their future work in the absence of these collaborations would be less impactful. We found that scientists with Chinese heritage experienced this chilling effect more acutely than those without. The few scientists we interviewed who felt that their research had not been affected much by recent tensions were not of Chinese heritage. Several scientists we interviewed who were of Chinese heritage reported feeling under increased scrutiny because of their ethnicity. Our quantitative and qualitative findings reveal that the scientists are affected in multiple dimensions. They also suggest that these investigations may have consequences unexpected by policy makers. For instance, while the policy is intended to address national security issues, we find a broad adverse effect on scientific productivity across institutions and fields. Moreover, as suggested by the comparison of scientific progress by fields, the investigations have aggregate implications that are important to be considered. Importantly, most of our interviewed scientists reported that they believe U.S.-China tensions are likely to last and thus have consequences in the long run. While the China Initiative has officially ended, funding agencies’ investigations of researchers are ongoing and universities’ policies with respect to collaborations with scientists in China is still in flux. We hope that our study serves as a first step to understanding the consequences of the ongoing political tensions and opening up new avenues for future research. # Materials and Methods #### Data construction In order to assess the impact of NIH investigations on the scientific output of U.S. scientists, we construct a dataset of U.S. scientists whose primary fields are in the medical and life sciences. To do so, we first query Dimensions to get the list of 1440402 PubMed publications in 2010–2014, for which at least one of the authors is based in the U.S. We impose two restrictions on the scientists in our dataset: (1) each scientist has to have at least two PubMed publications in 2010–2014 for which they are the Principal Investigator (PI);[^23] and (2) at least one of their publications needs to have a U.S. affiliation. The criterion (1) selects authors whose primary fields are more likely to be in the medical and life sciences and (2) focuses our attention on scientists who are based in the U.S. Applying the restrictions results in a list of 208647 scientists. Based on the initial list of scientists, we query Dimensions to get all of the selected scientists’ publications (including non-PubMed publications) from 2015–2021. To ensure the scientists we study were still in the U.S. immediately before treatment, we further restrict that each scientist’s last publication prior to the beginning of the NIH investigations (August 20, 2018) shows they have a U.S. affiliation.This reduces the number of scientists to 192493. Using affiliation data from Dimensions, we determine whether each paper included a U.S.-China collaboration, a U.S. collaboration with any country other than China, or included only authors from the U.S. We also use Dimensions-provided data to keep track of other metadata such as the funding information, citation count, and research field of each paper for our analysis. #### Validating data quality Because of the scale of the PubMed and Dimensions data and the algorithmic approach to coding authors, papers, and institutions that these databases use to produce them, the data inevitably has errors. To check the extent to which Dimensions data aligns with other existing datasets, we validate our data from Dimensions with data from Google Scholar, which is thought to be the most complete in terms of counting publication citations, but does not have an API for researcher access (Martı́n-Martı́n et al. 2021). In particular, we check whether our main outcomes of interest we use in this paper—authors’ publication record and citation counts—are comparable between the two sources. To do so, we draw a random sample of 100 authors from our data. For these 100 authors, we are able to identify 54 authors who have Google Scholar profiles. For each matched author, we compare their number of publications in 2010–2020 based on Dimensions with that based on Google Scholar. We also compare citation counts for each author-year from the two data sources. We should note that Google Scholar includes information on working papers that have not been published. Both measures are highly correlated between Dimensions and Google Scholar, with correlations around $`0.82`$ (see Appendix [A_Valid]). This gives us confidence that Dimensions data captures similar dynamics to other comparable data sources. ## Supporting Information Appendix (SI) Authors should submit SI as a single separate SI Appendix PDF file, combining all text, figures, tables, movie legends, and SI references. SI will be published as provided by the authors; it will not be edited or composed. Additional details can be found in the [PNAS Author Center](https://www.pnas.org/authors/submitting-your-manuscript#manuscript-formatting-guidelines). The PNAS Overleaf SI template can be found [here](https://www.overleaf.com/latex/templates/pnas-template-for-supplementary-information/wqfsfqwyjtsd). Refer to the SI Appendix in the manuscript at an appropriate point in the text. Number supporting figures and tables starting with S1, S2, etc. Authors who place detailed materials and methods in an SI Appendix must provide sufficient detail in the main text methods to enable a reader to follow the logic of the procedures and results and also must reference the SI methods. If a paper is fundamentally a study of a new method or technique, then the methods must be described completely in the main text. ### SI Datasets Supply .xlsx, .csv, .txt, .rtf, or .pdf files. This file type will be published in raw format and will not be edited or composed. ### SI Movies Supply Audio Video Interleave (avi), Quicktime (mov), Windows Media (wmv), animated GIF (gif), or MPEG files. Movie legends should be included in the SI Appendix file. All movies should be submitted at the desired reproduction size and length. Movies should be no more than 10MB in size.
Barjak, Franz, and Simon Robinson. 2008. “International Collaboration, Mobility and Team Diversity in the Life Sciences: Impact on Research Performance.” *Social Geography* 3 (1): 23–36.
Borjas, George J, and Kirk B Doran. 2012. “The Collapse of the Soviet Union and the Productivity of American Mathematicians.” *The Quarterly Journal of Economics* 127 (3): 1143–203.
Cheung, Tai Ming. 2009. *Fortifying China: The Struggle to Build a Modern Defense Economy*. Cornell University Press.
Didegah, Fereshteh, and Mike Thelwall. 2013. “Which Factors Help Authors Produce the Highest Impact Research? Collaboration, Journal and Document Properties.” *Journal of Informetrics* 7 (4): 861–73.
Dolgin, Elie. 2019. “’Psychological Fear’: MIT Scientists of Chinese Origin Protest Toxic US Climate.” *Nature* 571 (7764): 157–58.
Doria Arrieta, Omar A, Fabio Pammolli, and Alexander M Petersen. 2017. “Quantifying the Negative Impact of Brain Drain on the Integration of European Science.” *Science Advances* 3 (4): e1602232.
Freeman, Richard B, and Wei Huang. 2015. “Collaborating with People Like Me: Ethnic Coauthorship Within the United States.” *Journal of Labor Economics* 33 (S1): S289–318.
Guo, Eileen, Aloe Jess, and Hao Karen. 2021. “The US Crackdown on Chinese Economic Espionage Is a Mess. We Have the Data to Show It.” *MIT Technology Review*.
Hainmueller, Jens. 2012. “Entropy Balancing for Causal Effects: A Multivariate Reweighting Method to Produce Balanced Samples in Observational Studies.” *Political Analysis* 20 (1): 25–46.
Hoekman, Jarno, Koen Frenken, and Robert JW Tijssen. 2010. “Research Collaboration at a Distance: Changing Spatial Patterns of Scientific Collaboration Within Europe.” *Research Policy* 39 (5): 662–73.
Hsiehchen, David, Magdalena Espinoza, and Antony Hsieh. 2015. “Multinational Teams and Diseconomies of Scale in Collaborative Research.” *Science Advances* 1 (8): e1500211.
Imai, Kosuke, and Kabir Khanna. 2016. “Improving Ecological Inference by Predicting Individual Ethnicity from Voter Registration Records.” *Political Analysis* 24 (2): 263–72.
Lewis, Margaret K. 2021. “Criminalizing China.” *The Journal of Criminal Law and Criminology (1973-)* 111 (1): 145–225.
Martı́n-Martı́n, Alberto, Mike Thelwall, Enrique Orduna-Malea, and Emilio Delgado López-Cózar. 2021. “Google Scholar, Microsoft Academic, Scopus, Dimensions, Web of Science, and OpenCitations’ COCI: A Multidisciplinary Comparison of Coverage via Citations.” *Scientometrics* 126 (1): 871–906.
Mervis, Jeffry. 2021. “U.s. Scientists Want Congress to Look into Complaints of Racial Profiling in China Initiative.” *Science Insider*.
Moser, Petra, and Shmuel San. 2020. “Immigration, Science, and Invention: Lessons from the Quota Acts.” *Lessons from the Quota Acts (March 21, 2020)*.
Moser, Petra, Alessandra Voena, and Fabian Waldinger. 2014. “German Jewish émigrés and US Invention.” *American Economic Review* 104 (10): 3222–55.
Nomaler, Önder, Koen Frenken, and Gaston Heimeriks. 2013. “Do More Distant Collaborations Have More Citation Impact?” *Journal of Informetrics* 7 (4): 966–71.
Stoff, Jeffrey, and Glenn Tiffert. 2021. *Eyes Wide Open: Ethical Risks in Research Collaboration with China*. Hoover Institution.
Tang, Li, and Philip Shapira. 2011. “China–US Scientific Collaboration in Nanotechnology: Patterns and Dynamics.” *Scientometrics* 88 (1): 1–16.
Thorp, H Holden. 2022. “The China Initiative Must End.” *Science Advances* 8 (8): eabo6563.
Van Raan, Anthony. 1998. “The Influence of International Collaboration on the Impact of Research Results: Some Simple Mathematical Considerations Concerning the Role of Self-Citations.” *Scientometrics* 42 (3): 423–28.
Viswanatha, Aruna, and Kate O’Keeffe. 2020. “China’s Funding of US Researchers Raises Red Flags.” *Wall Street Journal*.
Wagner, Caroline S, Lutz Bornmann, and Loet Leydesdorff. 2015. “Recent Developments in China–US Cooperation in Science.” *Minerva* 53 (3): 199–214.
Wagner, Caroline S, and Loet Leydesdorff. 2005. “Network Structure, Self-Organization, and the Growth of International Collaboration in Science.” *Research Policy* 34 (10): 1608–18.
Wagner, Caroline S, Travis A Whetsell, and Loet Leydesdorff. 2017. “Growth of International Collaboration in Science: Revisiting Six Specialties.” *Scientometrics* 110 (3): 1633–52.
Waldinger, Fabian. 2010. “Quality Matters: The Expulsion of Professors and the Consequences for PhD Student Outcomes in Nazi Germany.” *Journal of Political Economy* 118 (4): 787–831.
Waldinger, Fabian. 2012. “Peer Effects in Science: Evidence from the Dismissal of Scientists in Nazi Germany.” *The Review of Economic Studies* 79 (2): 838–61.
White, Karen. 2019. *Publications Output: US Trends and International Comparisons. Science & Engineering Indicators 2020. NSB-2020-6.* National Science Foundation; ERIC.
Wuchty, Stefan, Benjamin F Jones, and Brian Uzzi. 2007. “The Increasing Dominance of Teams in Production of Knowledge.” *Science* 316 (5827): 1036–39.
Xie, Yu, Xihong Lin, Ju Li, Qian He, and Junming Huang. 2022. “Caught in the Crossfire: Fears of Chinese-American Scientists.” *arXiv Preprint arXiv:2209.10642*.
[^1]: See: [^2]: See Dear Colleague Letter from NIH Director Francis Collins: . We summarize more information on the background of NIH investigations in Appendix [A_Background]. [^3]: Lauer, Michael. “Foreign Interference in National Institutes of Health Funding and Grant Making Processes: A Summary of Findings From 2016 to 2021.” July 30, 2021. [^4]: There were a few cases where the Department of Justice investigated scientists with ties to China before the China Initiative began, for example Sherry Chen in 2014 and Xiaoxing Xi in 2015. However, the China Initiative and NIH Investigations in 2018 instituted a categorical shift in the number and breadth of the investigations. [^5]: While other federal research agencies also conducted investigations about foreign influence in research, the NIH was the first and to our knowledge most frequent federal agency to conduct them. [^6]: Using 2014 as a cutoff allows us to examine pre-trends (i.e., 2015–2018) for our differences-in-differences analysis. Our findings are robust to alternative cutoffs around 2014. [^7]: Based on our interviews, we think the effect materialized so quickly because many scientists knew soon after the NIH letter to universities in August 2018 that collaborations with China would be under particular scrutiny, and therefore some of their existing projects were impacted. [^8]: In addition, to rule out an effect of COVID-19 research, we provide a robustness check in Appendix [A_covid] where we remove all papers with titles containing the word “COVID.” We find that our results are unchanged. [^9]: Recently, Xie et al. (2022) documents a steady increase in the return migration of Chinese-origin scientists from the U.S. back to China, following the U.S.-China tensions. [^10]: An alternative way to define the treatment and control groups is to consider scientists who had or did not have collaborations with Chinese scholars between 2015 and 2018. We employ the current definition to exclude the influence of new entrants (those who became PIs during this period) on our results, but our results are robust to this alternative definition. [^11]: In our data, 130072 U.S. scientists with international collaboration during 2010–2014 publish on average $`6.39`$ papers per year, compared to 62421 U.S. scientists without international collaboration during the same period, who publish on average $`2.09`$ papers per year. [^12]: Total citations are measured by citations as of Oct. 23, 2022 when we downloaded the data from Dimensions. [^13]: By design, the covariates are balanced using entropy balancing. We present estimates from propensity score matching and nearest neighbor matching (based on covariates) and related balance tests in Appendix [A_Balancing]. As shown, the estimates from different methods are similar. [^14]: We identify NIH funding using the funding information provided in Dimensions. China-funded publications are also estimated by Dimensions, which geo-locates funders mentioned in academic papers. As described in Appendix [A_Background], mentions of funding from China in academic papers may have been a source of scrutiny by the NIH. This might be the reason that acknowledgement of funding from China in academic papers is more negatively affected. [^15]: For scientists with multiple institutions, we use their modal institution as their affiliation, which is defined as the institution in which a scientist published most of his work within the given period. [^16]: We identify institutions with public investigations using data from APA Justice and the *MIT Technology Review* (Guo et al. 2021) [^17]: Information on the source of funded is provided by metadata from Dimensions. [^18]: Field-specific effects were reflected in our interviews with scientists. Scientists who were in fields with high NIH-funding but low overall levels of U.S.-China collaboration, for example public health and clinical sciences, felt much less pressure to stop their U.S.-China collaborations than those in fields with higher levels of U.S.-China collaboration, such as in chemistry and the biological sciences. [^19]: . If a paper is categorized into multiple fields, it will be counted separately for each of the fields it belongs to. [^20]: We use the 2021 Nature Index () to select these 50 countries. [^21]: To calculate the number of publications by field, country, and year, we queried Dimensions for total publications by country, year, and field. These totals thus reflect overall publications in the field, not just publication numbers by the scientists in our data described above. [^22]: These interviews were approved by the UC San Diego Institutional Review Board. [^23]: To determine the PI of each paper, we treat the last author of each paper as the PI for that paper, as per the convention in the life sciences. When information about corresponding authors is available in the data, we also include the corresponding authors as the PIs of the paper.