Research Collaborations between Chinese and US Scientists and Engineers: A New Special Relationship? Abstract:

Transcription

1 1 Research Collaborations between Chinese and US Scientists and Engineers: A New Special Relationship? Abstract: This paper uses Web of Science and Pubmed data on scientific papers and the impact factor of the journal in which they are published and the citations received by the paper to assess research collaborations between Chinese-based researchers and Chinese-named authors on research publications produced in China and in the US in the 1990s and 2000s. It finds (1) a huge rise in the share of addresses from China in publications and citations (2) increasing partnership in publications between US and China. (3) many Chinese-named researchers conducting research in the US, with the share of Chinese names on publications associated with higher impact factors and citations (4) a growing number of Chinese researchers with US research experience doing research in China, producing publications with higher impact factors and citations. (5) Greater US-China collaborations from US cities with higher Chinese proportions of researchers (6) Higher research outcomes for the same Chinese researcher working in the US than working in China. These results give a positive answer to the title question. Yes, Chinese and US researchers have a special relations with benefits accruing to research in both localities. Richard Freeman and Wei Huang, Harvard and NBER American Economic Association Philadelphia, Pa. January 5, 2014 PRELIMINARY DRAFT FOR COMMENT ONLY

2 2 China has entered the world of modern science and engineering in a big way. The number of PhDs graduating in science and engineering in China exceeds the number in any other country the result of a massive expansion in higher education and R&D spending in the 1990s and 2000s. 1 Chinese students make up a huge proportion of international students worldwide, with many obtaining S&E PhDs from top universities in the US, Europe, and elsewhere. One striking aspect of China's advance to the forefront of science and engineering activity is the close ties that have developed between China and the US in both S&E education and research. Chinese students are found in virtually every top US graduate program. About 90% of Chinese nationals earning US S&E PhDs remain in the US for years after graduation. 2 They work in US universities and research centers, adding to the US's research output. Many other Chinese researchers come to the US as post-docs or on academic sabbaticals, in many cases supported by Chinese government grants that provide travel and living expenses to persons accepted as academic visitors to US and other advanced country institutions. Collaborations between Chinese and US scientists and engineers are substantial in almost every field. Chinese researchers who eventually return to China return with considerable research experience and connections with US researchers and research projects. This paper combines data on papers and citations from the Web of Science (WoS) for all physical sciences and engineering with data on papers and authors from PubMed for the biological or life science fields to examine China-US research collaborations from the 1990s through the 2000s when China moved rapidly from a bit player to a major contributor to the science and engineering literature. Per the title question, the number, pattern, and quality of research outcomes for US and China collaborations suggest that indeed the two countries have developed a special relationship in the globalization of knowledge creation. The paper is divided into four sections. Section 1 analyzes country addresses on papers in the WoS and PubMed to measure the magnitude and pattern of collaborations between researchers located in China and researchers located in the US. To deal with the fact that many scientific papers have co-authors in many different locations, 1 Haizheng Li Higher Education in China: Complement or Competition to US Universities? In Clotfelter, Charles American Universities in a Global Market University of Chicago Press Finn, Michael Stay Rates of Foreign Doctorate Recipients from U.S. Universities, 2007 January table 6

3 3 we use the number of addresses from US or China relative to the total number of addresses on a paper to construct a more subtle measure of geographic collaboration than the commonly used measure of whether a paper has an address from those or other countries. Our analysis shows a huge rise in the share of addresses from China in publications and citations and increasing partnership in publications between US and China. Section 2 analyzes the ethnicity of the names of authors in PubMed as an additional indicator of cross-country collaborations. Because Chinese names are distinguishable from other names and because most Chinese researchers are from mainland China or Taiwan, 3 we compare publications with Chinese names and with other names in the US and China, using disambiguated names in PubMed. This analysis shows that many Chinese-named researchers conduct research in the US and that a growing number of Chinese researchers in China have US research experience. Measuring China's research presence through names as well as addresses shows a more rapid rise of China in global science and engineering research and greater links between Chinese and US researchers than through either measure separately. Section 3 turns to the quality of the collaborative scientific work, as reflected by the impact factor of the journal in which the researchers published a paper and the citation performance of the paper. It shows that papers with US-only addresses do better in impact factors and citations the greater the proportion of Chinese authors on the paper; whereas papers written collaboratively with researchers in China reduce impact factors and citations. Papers written with Chinese addresses benefit from research done by Chinese researchers with prior US research experience and from collaboration with US-based researchers. Section 4 examines the relation between the growing Chinese presence in research in the US and US-China cross country collaborations and differential productivity of the same Chinese researchers when they work in the US and when they work in China. It finds greater US-China collaborations from US cities with higher Chinese proportions of researchers; and higher research outcomes for the same Chinese researcher when they work in the US than when they work in China. The conclusion summarizes the findings. 3 We have examined this by using first names and initials. A US-born Wang is likely to have an English first name like Robert or Harold whereas a China-born Wang is likely to have a first like Xu or Zhi or with initials at the end of the alphabet rather than the beginning. The vast majority of US-based Chinese authors of scientific papers have Chineseborn initials. See Freeman and Huang Collaborating with People Like Me: ethnic co-authorship within the US NBER WP to appear 2014.

4 4 1.Collaborations over Space: US-China Links on Addresses of Papers We begin our analysis with papers in the Web of Science. Because of the destruction of China's higher education and research system in the Cultural Revolution, China had only a slight presence in science and engineering research until the 2000s. Table 1 records the total number of papers in the WOS and the numbers of papers and citations to papers with China and US addresses from 1998 through The WOS data goes back further but there is a sharp break in 1998 when WOS added more journals that creates a measurement problem but one largely irrelevant for the Chinese addressed papers which are minimal even after the jump in papers. We measure China and US international collaborations in a somewhat different way than is common in studies of international collaborations. Our measure is based on the proportion of Chinese or US addresses among all addresses on the paper and is thus a more refined measure than the 0-1 measure of whether a paper has a Chinese, US, other country address. To see the difference between our measure and the standard measure consider two papers one with two Chinese addresses and one with one Chinese address and one European address. We count the former paper as one China paper and the latter as 1/2 China paper. We have done calculations doing the cruder dichotomous measures and obtained results similar to those in our tables. Treating the country content of a paper by the proportion of addresses to the country can be further refined for the period from 2008 forward. In our data, the WOS lists author addresses on a paper without linking the addresses to each author. Thus it does not show whether any of the addresses had more authors (and presumptively more work)than others. Given this limitation we credit a paper with, say five authors and addresses in two countries, as being produced half in one country and half in the other. With the more refined data we might find that four of the authors were in one country or even the five were in one country and one had a second address in the other country, and thus compute more accurate measures. In 2008 the WOS linked each author to his or her address, but this does not allow us to improve measures for the period covered. Column 1 shows a 35% increase in total scientific papers in the ten years in our data. Of those papers column 2 shows that the share going to Chinese addresses increased from 2.49% to 9.02% a 360% increase. Papers with China addresses typically get lower numbers of citations than other papers in the WOS, presumably because Chinese researchers are relatively young with research programs just beginning. Column 3 shows that the citation share of papers from Chinese addresses falls markedly below the share of papers but that the citation share increased far more rapidly than the paper share. As

5 a result the citation rate of papers with Chinese addresses relative to all papers in column 4 trended upward steadily 1998 to 2007, suggests an upward trend in the quality of China-based research. Columns 5-7 give comparable figures for the US. The proportion of US addresses on papers drops from 28.8% to 24% the result of the expansion of scientific activity in the rest of the world, including China (column 5). With about 6% of the world's population, the US share of papers will almost keep trending downward from the extraordinary levels it held post world war II. The share of citations to US-based addresses in column 6 trends downward at about the same rate, thus maintaining relative share of citations to US addressed papers in column 7 at a fairly constant rate above the world average. The last two columns in table 1 record the average citations for China-based and US-based papers respectively. The averages are highest for the earlier years due to those papers having more time to cumulate citations. But the pattern of more citations for older papers is less consistent for the China-addressed papers than for US-addressed papers, possibly because the quality of Chinese papers has improved over time and/or because the increasing number of Chinese-addressed papers are more likely to cite Chinese papers through network effects. 4 To see how papers with China addresses and US addresses are connected, we examined the extent to which papers with at least one China or one US address were written with researchers in the other country or in some third country. Figures 1A-1C display the results of those calculations for papers with one or more addresses in China and one or more addresses in the US. Figure 1A shows the well-known trend in internationalization of papers with a US address. By contrast, the international proportion of Chinese declines moderately, which given the growth of international collaborations globally presents a bit of puzzle. The answer to the puzzle lies with the huge increase in research activity in China, which increased the potential for collaborations among researchers with a Chinese address. Figure 1B shows that the proportion of US addresses on papers with at least one address in China and the proportion of China addresses on papers with at least one address in the US have both increased, but at different rates. The proportion of US addresses on papers with a Chinese address was reasonably high at the outset and increases modestly over the period, indicating that US-China collaborative work has remained important even as China-based research has expanded greatly. The proportion of Chinese addresses on papers with a US address was quite small in 1998 at 0.66% but 4 We are currently examining the extent to which authors in US and China papers self-cite by their own papers and site papers from US or China addresses and from Chinese or non-chinese named authors. 5

6 6 increases over three-fold to 9.8% in Figure 1C shows that the differing trends in the share of international papers and in the share of papers with US or China as partner country produce similar trends of increased collaboration with persons in the other country. US international collaborations involve more China-based researchers. Chinese international collaborations involve more US-based researchers. From the WOS addressbased metric, the US and China are becoming greater partners in global research activities. Table 2 focuses more closely on the pattern and relation between papers with at least one China address or with at least one US address in the PubMed data base. Column 1 shows that the rising number of China-based papers in PubMed began accelerating around 1997 when 1.04% of addresses on papers were from China to 12.02% of addresses in 2007 (column 2). 5 The increase in the China share of addresses is moderately larger in the PubMed than in the WOS data in table 1. The patterns in the columns for papers with 1 or more China addresses in table 2 are comparable to those in table 1 for the period beginning in The proportion of international addresses on China-address papers declines beginning in the late 1990s through 2007 as the number of China-based papers increases. The proportion of US addresses on China-address papers declines modestly from 1998 to 2007 at a much smaller rate than the decline in the proportion of international addresses. The result is that the US-China proportion of international collaborations with a China address increases substantially in the 2000s in the PubMed data as in the WOS data. The pattern for China from 1990 through 1998 differs markedly from that after 1998 because of the small number of China-based researchers/papers in the earlier period. With few Chinese groups doing research, 40% to 50% of the addresses on papers with a Chinese address were international outside the US while 20% to 25% were US based. The proportion of addresses on US-based Pub Med papers declines throughout the period, though the decline is more rapid in than in the earlier period. The decline exceeds the increase in the China share of addresses, reflecting the increase in international addresses beyond China in the PubMed data base. The proportion of China addresses on US-based papers (column 8) increases more rapidly than the proportion of international addresses on US-based papers (column 7) so that the China share of US international collaborations increases substantially as it did in the WOS data. In short in both data sets, the share of China addresses on papers increases substantially while 5 It would be fruitful to compare the logistic growth curve for China-based addresses with those for European countries, Japan, or Korea when they began increasing their shares of scientific papers to see whether there is a general pattern or whether the China pattern is distinct.

7 the share of US addresses declines; and the two country's shares of the other country's international collaborations increases. 7 2.Collaborations by Ethnicity of Names The presence of Chinese names on research papers with US co-authors provides another indicator of the collaboration between Chinese and American researchers in global science and engineering. China born researchers working outside of China with researchers of other ethnicities or with other Chinese researchers through like associating with like homophily collaborations beyond that captured with country addresses. Finally, China-born researchers who gain research experience in the US (and other countries) and use that experience upon returning to their native land represent another channel of collaboration. To measure these linkages between China and US researchers, we link PubMed data on the names of the authors of scientific papers to the likely ethnicity of the researchers. 6 We examine PubMed data fro three reasons. First, because the program that determines the ethnicity of names associates the names in PubMed to an ethnic group at a higher rate (90%) than it does for the names in WOS (80%), in part because PubMed reports relatively more first names as opposed to initials than does the WOS. Second, because Torvik and Smalheiser's (2009) disambiguation of names in PubMed 7 allows us to follow the publications of individuals over their careers and thus to estimate the extent to which Chinese-named authors have research experience in the US prior to publishing a given paper. Third, because the PubMed data does not show a sudden sharp break in the numbers of publications in The cost of using PubMed is that it restricts analysis to the biological or life sciences, which make up about one-fourth of WoS papers. To get the impact factor and citation of papers we link the PubMed papers to the same paper in the WOS. Column 1 of table 3 shows a fourfold increase in the proportion of Chinese names among authors in Pubmed papers from 1990 to In part, this simply reflects the increased share of papers with a Chinese address shown in table 2 (on which most authors naturally have Chinese names). But it also reflects an increase in Chinese names on papers with addresses in the US and other countries. Column 2 shows that among papers with only US addresses, the proportion of Chinese-named authors 6 For this we use the latest version of the names-ethnicity identification program of William Kerr, "Ethnic Scientific Communities and International Technology Diffusion," The Review of Economics and Statistics, MIT Press, vol. 90(3), pages , August. 7 Vetle I. Torvik and Neil R. Smalheiser Author name disambiguation in MEDLINE ACM Transactions on Knowledge Discovery from Data Volume 3 Issue 3, July 2009 Article No. 11

8 8 more than doubled from 4.5% to 10.9%. Column 3 shows an even larger percentage point increase in the proportion of Chinese-named researchers on papers with only China addresses. This presumably also reflects the increased number of Chinese-named researchers. Finally, columns 4 and 5 show large increases in the proportion of Chinese named authors on China-US collaborations and on China-non US international collaborations. Consistent with the development of a special relation between China and the US researchers the proportion of Chinese names on China-US papers exceeds the the proportion of Chinese names on China-non US international collaborations. One possible reason for this is the large number of China-born researchers who work in the US and who, we shall soon see, have a special proclivity for collaborating with researchers in China. Table 4 examines the effects of the changing number of Chinese authors in a different way. It shows of the distribution by address of papers written exclusively by persons with Chinese names. It is stunning to see that from 1990 through 1998 the majority of papers with all Chinese names had all US addresses! This reflects homophily in the US market, where Chinese (and other ethnic groups) are far more likely to write with persons of the same ethnic background than they would by chance 8 and possibly the effects of the Tiananmen Square protests and suppression on the decisions of many Chinese graduate students and academics to remain in the US rather than to return home. As China increased its production of PhDs and scientific activity the share of all Chinese written papers shifts massively toward China, though there still remains a notable minority of all Chinese papers with only US addresses, US and China collaborations, and even with US and other non-china addresses. As noted, many China-born doctorate students obtain degrees in the US, many Chinese postdocs work in the US, and many Chinese scholars come to the US as visiting researchers. The result is that many Chinese scientists and engineers have cumulated substantial US research experience. Without searching CVs or data sources that contain information on the experiences of particular researchers such as ProQuest for PhD dissertations or LinkedIn, it is difficult to measure the USexperience of Chinese scientist and engineers. With the disambiguated names on the PubMed data set, however, we can determine if a Chinese-named researcher wrote a paper while in the US. If the researchers name is on a paper with all US addresses, the researcher must have done their research for the paper at a US location. This measure of US research experience is conservative because it ignores the potential that a Chinese researcher on a US-China (or other country) paper conducted their research 8 Freeman and Huang Collaborating with People Like Me: ethnic co-authorship within the US NBER WP to appear 2014.

9 9 in the US and because it gives no US research experience for past collaborations with Americans. Figure 2 displays our estimate of the proportion of Chinese-named authors of papers published in a given year who had a US-only address paper at some earlier time. The figure distinguishes among papers with only Chinese addresses, papers with China-US addresses, and papers with China-non-US international addresses. The estimated proportions increase substantially in the late 1990s and 2000s for all three address groups. By 2007, 10.2% of Chinese researchers on China-based papers had previous US experience; 15.3% of Chinese researchers on China-non US international collaborations had such experience; and 25.5% of Chinese researchers on China-US collaborations had such experience. That one quarter of Chinese researchers on China-US collaborations had previously written a paper with a US address suggests a strong role for the effect of research visits to the US in creating the links and ideas that develop into cross-country collaborations. 3.Relation of Chinese-US Address and Name Collaborations with Impact factor and citations Bibliometric data provide two widely used measures of the quality of papers: the impact factor of journal in which paper appeared and the number of citations to the paper from year of publication to the most recent period. Almost all more sophisticated indexes of the contribution of papers to scientific knowledge derive from these measures. While we recognize that both impact factors and citations are imperfect measures of quality, we use them in this section to assess the productivity of different forms of US and China collaborations. One problem with the impact factor of the journal of publication is that high impact journals contain mundane as well as outstanding articles. An article may be accepted in a high impact journal because it fits with prevailing scientific orthodoxy rather than challenging it; or because the researchers are connected to editors and reviewers through networks or are from highly prestigious universities or research centers. The number of citations to articles in high impact journals vary greatly (as does the number of citations to articles in low impact journals). Just because an article appears in the same journal with an important article does not mean it is a greater contribution to science than if it had appeared in a lower impact journal. Indeed, it is arguable that an article with 20 citations in a lower impact journal may be better than an article with 20 citations in a high impact journal because researchers had to work harder to find it. Still, the impact factor has virtues as a measure of quality. It immediately identifies a paper as one that passed a critical review process that is presumptively more difficult than those in less

10 prestigious journals and as one that editors felt made a broad contribution to science. Citations are presumably a better measure than impact factors of the quality of a paper as they reflect the paper itself rather than where it appeared. But they also have problems. There are differences in the level of citations across specialties (which we deal with by including a large number of subfield dummy variables in ensuing regressions). The life cycle of citations (which we treat by including publication year dummies) creates difficulties in assessing new papers. Citation networks affect the number of citations an article gets. Papers written in countries with larger numbers of researcher such as the US are likely to get more citations than comparable papers written in countries with smaller research communities. To the extent that US papers disproportionately cite US research, the high US share of papers will itself add to the US share of citations. As noted in section one, the rising share of citations to Chinese authored papers may be partially driven by the rising share of Chinese-authored papers. We use impact factors and citations to assess the benefits of US-Chinese collaboration on papers with US addresses and papers with Chinese addresses. We use the proportion of Chinese names among authors to measure the Chinese contribution to a US addressed papers We use the US experience of Chinese-named authors to measure the US contribution to a Chinese addressed paper. In both we cases we include variables measuring collaborations across country lines as well, differentiating between papers with both US and China addresses and those with US and non-china country addresses. Table 5 records the regression coefficients and standard errors for the impact of having Chinese co-authors and cross-country collaborations for US addressed papers. Columns 1 and 2 show that the proportion of Chinese-named authors on papers with only a US address is associated with higher impact factors and citations, conditional on a host of covariates listed at the bottom of the table. Columns 3 and 4 extend the sample to all papers with at least one US address and adds dummy variables for whether paper was a collaboration with China-based researchers or a collaboration with non-china based researchers. Contrary to the widely held view that international collaborations are associated with higher quality papers, the regressions in columns 3 and 4 show that the impact factor and citations for US addressed papers written collaboratively with researchers in China or in other countries are negatively related to the two outcome measures while the Chinese author proportion continues to have substantial positive effect. 9 To see the net effect of collaborations, we estimated the 9 The belief that international papers are more productive than others is based on comparisons of mean levels of outcomes without any consideration of covariates. Papers with more co-authors tend to have higher impact factors and citations and papers with international collaborations tend to have more co-authors, but conditional on numbers of authors and numbers of addresses, international collaborations do no appear to be associated with better papers. See 10

11 11 equations underlying columns 3 and 4 leaving out the Chinese proportion of authors and still obtained negative albeit smaller estimated effects of collaborations outside the US. Table 6 turns to impact of Chinese authors having US research experience on the impact factor and citations of China-addressed papers. Columns 1 and 2 show that the proportion of Chinese-named authors with US experience substantially increases both impact factors and citations on papers with only a Chinese address, conditional on a host of covariates listed at the bottom of the table. Columns 3 and 4 show a similar pattern for the proportion of Chinese authors with US experience in the wider set of papers with at least one China address. The implication is that US research experience improves the quality of researchers in China and thus that the government program of sending researchers to spend some time in the US payoffs, at least in part, in improved Chinese research activities. The estimated coefficients on the dummy variables for collaborations in table 6 differ markedly from the estimated coefficients for the dummy variables found for US addressed papers in table 5. In table 6 collaborations between US and China based researchers have a substantial positive relation to impact factors and citations the opposite of their negative link in table 5. The difference in signs relates to the reference groups. The reference group in table 6 are papers written only in China. The reference group in table 5 are papers written only in the US. Thus China-US collaborations produce impact factors and numbers of citations that exceed those on papers produced in China, where citations per paper fall below the global average, but that fall short of impact factors and citations in the US, where citations per paper exceed the global average. We expect that as the number of Chinese researchers continues to increase and as Chinese researchers gain more experience, these differences will diminish. 4 Chinese researchers in US: links to US-China collaborations and research productivity This section examines two further aspects of the special relationship between US and China research: the link between the growing number of Chinese-named researchers on papers with US addresses and US-China collaborations; and the productivity of Chinese researchers working in US compared to their working in China. Both analyses make use of the PubMed disambiguated names of researchers together with WoS addresses, impact factors and citations to obtain appropriate independent and dependent variables. We begin with the concordant increase in Chinese-named researchers in the US and increased Freeman,Richard and Ina Ganguli, and Raviv Murciano-Goroff The Why and Wherefore of Increased Scientific Collaboration (2013)

12 US-China research collaborations. While these two developments could reflect independent trends resulting from the massive expansion of Chinese research activities, we expect that they are related at a more fine-grained level. Evidence that Chinese (and other ethnic group) researchers tend to write more papers with members of their own ethnicity than if collaborations occurred independently of ethnicity 10 suggests that China-named researchers in the US would be more likely than other researchers to seek out and accept collaborations with researchers in China; and conversely that Chinese researchers in China would be more likely to seek out and work with Chinese researchers in the US. We chose a geographic-based analysis of the location of individual authors in cities in the US to test this expectation. We calculated for each of some 300 US cities the number of authors on papers written in the city and the number of Chinese names on those papers. For each city-year cell we formed the proportion of researchers who were Chinese in the city. We next calculated the proportion of papers from the city that had China-addresses, indicating US-Chinese collaborations. If Chinese named researchers are more likely to coauthor with persons of Chinese ethnicity in China, then cities with a larger proportion of Chinese US-based researchers should see a higher number of US-China collaborations. In the period under study, when US-China collaborations increased greatly, the increases ought to be related to increases in the Chinese proportion of US-based researchers by city. Table 7 records the results of regressions that yield just such results. Column 1 is a pooled cross-section regression of the proportion of papers in a city-publication cell that are US-China collaborations on the proportion of Chinese named authors in that city-publication cell. The coefficient on the Chinese proportion is positive significant. Column 2 adds city dummy variables to the regression, which turns the computation into a panel/ longitudinal regression that links deviations of the proportion of papers that are US-China collaborations around their city means on deviations in the China share of authors in the city around their city means. The estimated coefficient on the proportion of authors with Chinese names is reduced to less than a third its value but remains highly significant positive. The increased proportion of Chinese-named researchers in the US is closely linked to the increase in US-China collaborations. The way in which the productivity of a given researcher varies in different research settings is more difficult to determine, due in part to the endogeneity of decisions of where to conduct one's research. Kahn and MacGarvie's (2012) analysis of the effects of Fulbright awards requirement that recipients return to their home country on the scientific career of persons found that Fulbright scholars 10 Freeman and Huang Collaborating with People Like Me: ethnic co-authorship within the US NBER WP to appear

13 13 from low income countries had fewer publications and less impact on global knowledge relative to a control group of scientists who did not face the Fulbright requirement to return home. 11 The implication is that the US research environment encourages more and better academic publications than the those researchers find in their own country. Our analysis in table 8 suggests that this may also be the case for Chinese researchers. In this table we regress impact factors and citations of papers for Chinese scientists with some US research experience (ie paper written with US address only) on papers written in the US, papers written with only Chinese addresses, China-US, and China-other country collaborations. By including author fixed effects in the regressions, the analysis compares papers written by the same person in different settings/collaborations. The estimated coefficients show that a researcher writing a paper with a China address has a significantly lower impact factor and fewer citations than the same researcher writing a paper in the US (the reference group). In addition, collaborative papers with persons with a US address also have lower impact factors and citations than the papers written by the person in the US. Absent an analysis of the factors that lead Chinese researchers to locate in the US or China or elsewhere, it would be a stretch to conclude that the US research environment encourages higher quality research from Chinese researchers than in their own country, though the regressions clearly point in that direction. 5. Conclusion This paper has found a huge rise in the share of addresses from China in publications and citations and increased participation of Chinese-named researchers on publications with US only addresses as well with on publications with other addresses. It has shown that the two trends are related as US cites with relatively greater increases in the share of Chinese-named authors also have more rapid increases in collaborations with China. The analysis also found the US-based papers have higher impact factors and citations the higher the proportion of Chinese authors while Chinese research papers have higher impact factors and citations the greate the US research experience of China-based authors. Finally, a given Chinese researcher gains higher impact factor and citations for papers published based on working in the US than from China-addressed papers. All told, the answer to the title question is Yes, the US and China have developed a special relationship in research collaborations that benefits research in both countries in different ways. 11 Shulamit Khan and Megan MacGarvie The Effects of the Foreign Fulbright Program on Knowledge Creation in Science and Engineering in Josh Lerner & Scott Stern, "The Rate and Direction of Inventive Activity Revisited," NBER Books, National Bureau of Economic Research, Inc, number lern11-1, November.

14 14 Table 1: China and US addresses in Web of Science Paper, (1) (2) (3) (4) (5) (6) (7) (8) (9) Publication Total Prop. Of Citation Citation Ave. citations for Ave. citations Prop. Of USbased papers year number of China-based share from (3)/(2) share from (6)/(5) China-based for US-based papers papers China US papers papers , % 1.01% % 42.1% , % 1.32% % 40.7% , % 1.85% % 39.9% , % 2.26% % 39.5% , % 2.73% % 38.7% , % 3.39% % 38.3% , % 4.06% % 37.3% , % 4.79% % 36.5% , % 5.60% % 35.5% , % 6.17% % 34.7% Note: Date are from WOS. Papers published between 1998 and 2007 are kept. China-based = Sum[China address prop.]/sum of papers US-based = Sum[US address prop.]/sum of papers Citation Share from China = Sum [Citations received * (China address prop.)]/sum of citations Citation Share from US = Sum [Citations received * US address prop.)]/sum of citations Ave. citations for China-based papers = Sum [Citations received * (China address prop.)]/sum[china address prop.] Ave. citations for China-based papers = Sum [Citations received * (China address prop.)]/sum[china address prop.]

15 Figure 1A. Prop. of Int'l addresses in papers w/ China address, and in papers w/ US address 15 Fig 1B. Prop. of US addresses in papers w/china address, and Prop. of China addresses in papers w/ US address Fig 1C. Prop. of US addresses and Prop. of China addresses among Int'l papers for each country

16 16 Table 2: China and US addressed papers in PUBMED, (1) (2) (3) (4) (5) (6) (7) (8) (9) Publication year Num. of Papers containing China addresses China-based papers prop. Int'l address prop. Papers with 1+ China Address US address prop. Ratio (4)/(3) Prop. of US-based Papers Papers with 1+ US Address Int'l address prop. China address prop. Ratio (8)/ (7) % 25.2% 12.8% 50.9% 94.4% 10.2% 0.22% 2.15% % 26.1% 11.5% 44.2% 94.0% 11.1% 0.19% 1.71% % 25.5% 11.6% 45.7% 93.5% 11.9% 0.23% 1.90% % 27.6% 11.8% 42.7% 92.8% 13.2% 0.24% 1.81% % 28.0% 10.9% 38.9% 92.0% 14.7% 0.23% 1.57% % 27.7% 11.1% 40.1% 91.7% 15.4% 0.23% 1.47% % 24.5% 9.7% 39.7% 91.2% 16.0% 0.23% 1.45% % 24.8% 8.3% 33.6% 90.4% 17.0% 0.23% 1.36% % 22.5% 8.1% 36.1% 89.4% 18.0% 0.32% 1.80% % 20.0% 7.2% 36.0% 88.2% 19.1% 0.40% 2.09% % 17.9% 6.2% 34.9% 86.7% 20.1% 0.52% 2.61% % 17.5% 6.7% 38.1% 85.2% 21.5% 0.72% 3.35% % 17.1% 6.4% 37.7% 83.9% 22.4% 0.84% 3.76% % 16.4% 6.5% 39.6% 82.5% 23.2% 1.03% 4.45% % 17.1% 6.7% 38.8% 81.3% 24.1% 1.23% 5.11% % 16.6% 6.7% 40.7% 79.6% 25.0% 1.56% 6.23% % 15.5% 6.5% 42.0% 77.4% 25.8% 1.91% 7.43% % 15.2% 6.4% 42.1% 75.3% 27.2% 2.21% 8.14% Note: The sample is composed of papers in PUBMED with any China or any US address, determined from Web of Science. Column 3 is the mean value of share of international addresses (Address outside China) in papers with 1+ China address. Column 4 is the mean value of share of US addresses in papers with 1+ China address. Column 7 is the mean value of share of international addresses (Address outside US) in papers with 1+ US address. Column 8 is the mean value of share of China addresses in papers with 1+ US address.

17 17 Table 3: Proportion of Chinese Names in PUBMED, by addresses of papers, (1) (2) (3) (4) (5) Proportion of Chinese Name authors Publication year only China-based Full sample only US-based papers China-US papers papers China-Int'l other than US papers % 4.5% 84.8% 46.0% 46.3% % 5.0% 85.9% 48.5% 42.2% % 5.4% 82.3% 53.0% 38.4% % 6.3% 84.0% 46.1% 39.5% % 7.1% 85.1% 53.1% 41.4% % 7.5% 85.1% 49.2% 42.3% % 7.9% 88.3% 50.5% 41.8% % 8.1% 89.7% 50.2% 41.7% % 8.0% 90.0% 55.2% 46.4% % 8.2% 89.8% 55.4% 50.2% % 8.4% 88.3% 55.9% 46.2% % 9.0% 89.1% 57.3% 49.0% % 9.3% 90.9% 59.8% 50.3% % 9.8% 92.0% 62.4% 53.3% % 10.2% 92.4% 64.5% 54.0% % 10.6% 93.3% 66.8% 55.1% % 11.0% 93.8% 67.9% 57.7% % 10.9% 94.2% 68.7% 58.9% Note: The whole sample is composed of PUBMED papers with any US or any China address. "Chinese" in this table is defined as the likelihood of surname ethnicity being Chinese. The statistics in Columns 1-5 are conditional probability. For example, column 2 shows the Chinese author proportion in only US-based papers, and column 3 shows the Chinese author proportion in only-china based papers.

18 Table 4: Distribution of Papers with All Chinese named authors, by distribution of addresses 18 Note: The percentages in the five columns under the distribution by addresses sums to 100%

19 19

20 20 Table 5: Estimated Relation between Percentage of Chinese Co-authors and US-China collaborations and Impact Factors and Citations, For US Addressed Papers (1) Papers with only US address Papers with 1+ US address VARIABLES Impact factor Citations Impact factor Citations Chinese author prop *** 1.363*** 0.299*** 1.567*** (0.0139) (0.295) (0.0133) (0.275) Paper addresses composition: Reference group is only USA US-China (Yes = 1) *** *** (0.0245) (0.507) US-Other (Yes = 1) *** *** ( ) (0.121) Observations 1,541,702 1,541,702 1,923,510 1,923,510 R-squared Publication year dummies Yes Yes Yes Yes Subfield dummies Yes Yes Yes Yes Interactions of above three Yes Yes Yes Yes Number of authors dummies Yes Yes Yes Yes References number categories Yes Yes Yes Yes Reference number linear trend Yes Yes Yes Yes Note: Standard errors in parentheses are clustered in paper level. *** p<0.01, ** p<0.05, * p<0.1. Sample is restricted to papers with US addresses.

21 21 Table 6: Estimated Relation Between Chinese Authors with US experience and types of collaboration and impact factor and citations, for China-addressed papers (1) (2) (3) (4) Papers with all addresses in China and written by Chinese Papers with 1+ addresses in China VARIABLES Impact Factor Citations Impact Factor Citations US experience prop *** 2.782*** 0.714*** 4.327*** (0.0399) (0.367) (0.0346) (0.363) Type of Collaboration by Paper addresses: Reference Addresses, ref group is only China China-US 0.526*** 2.807*** (0.0223) (0.234) China-other * (0.0195) (0.205) Chinese author prop *** *** (0.0307) (0.322) Chinese language *** ** *** (0.0522) (0.481) (0.0654) (0.684) Observations 56,663 56, , ,858 R-squared Publication year dummies Yes Yes Yes Yes Subfield dummies Yes Yes Yes Yes Interactions of above two Yes Yes Yes Yes Number of authors dummies Yes Yes Yes Yes References number categories Yes Yes Yes Yes Reference number linear trend Yes Yes Yes Yes Note: Standard errors in parentheses are clustered in paper level. Data are are PUBMED and WOS. Regressions are at paper level. Sample is restricted to papers with at least one China address in this table. A paper written by Chinese is defined as the mean author ethnicity being over 0.9 likely to be Chinese.

22 22 Table 7: Estimated Relationship Between Chinese Proportion of Authors in city-year cell and proportion of papers in cell with China address (1) (2) VARIABLES US-China Based Proportion (0-1) Chinese Prop. in City (0-1) *** *** ( ) ( ) Author number *** ( ) ( ) Observations 36,571 36,571 R-squared City fixed effect No Yes Publication year dummies Yes Yes Note: Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Regression are done in city-publication year level. Sample is restricted to US, China or US-China papers and papers with at least one US address are kept in this table. Chinese Proportion in city level are calculated among only US-based papers. Dependent variable is mean of US-China based papers proportion in corresponding city-publication cell.

23 23 Table 8: Estimated Relationship Between type of collaboration and impact factor and citations for Chinese-named authors with US research experience (1) (2) Impact factor Citations Reference group: Only US-based Only China *** *** (0.0631) (0.827) China-US *** *** (0.0711) (0.882) China-Other *** *** (0.0787) (1.040) Observations 87,148 87,148 R-squared Publication year dummies Yes Yes Authors fixed effect Yes Yes Chinese author prop. Categories Yes Yes Number of authors dummies Yes Yes References number categories Yes Yes Reference number linear trend Yes Yes Note: Standard errors in parentheses are clustered in author level. *** p<0.01, ** p<0.05, * p<0.1. Regression are conducted in paper level. The sample is composed of papers written by those Chinese once worked in US and would work in China afterwards.