The inuence of international scientic collaboration on the research performance of Brazilian academic institutions

Scientic collaboration, a practice that traces its roots back to the birth of modern science, has spread through the research community, expanding the ties between institutions and countries and becoming a strategy to improve research productivity. Collaboration with countries of renowned scientic leadership thus constitutes a clear opportunity for the scientic advancement of academics as well as institutions worldwide. This work focuses on the set of Brazilian papers indexed by InCites between 2010–2019 to analyze the advantages, measured in terms of the citation impact and percentage of publications in Q1 journals, as well as (just for the papers published between 2014 and 2018) the position in the ARWU Global Ranking of Academic Subjects, derived from the sustained scientic collaboration with institutions from Australia, Canada, United Kingdom, and the United States. Our results show that collaboration with these four countries presents clear advantages for Brazilian institutions in all areas of knowledge. In particular, our study shows that the percentage of publications in Q1 journals doubles, and the citation impact increases markedly for the set of papers in collaboration with the aforementioned countries. Our study also shows that, by and large, Brazilian academic institutions benet from these international collaborations to improve their positions in the current edition of the ARWU Global Ranking of Academic Subjects.

ensure the participation and integration of less developed countries into research networks. Collaboration with those four English-speaking countries presents impact advantages (Van Leeuwen 2009). Bordons et al. (2015) also identi ed differences in the impact related to the country of origin of the team leading the research effort. Studies led by scientists from countries of global scienti c leadership, such as Australia, Canada, the United Kingdom, and the United States, usually attract a high number of citations. Partnering with institutions from those countries becomes then very attractive for other researchers.
Together with the type of collaboration -national or international -and the scienti c performance of the collaborating country, the order of authorship is an important detail that has not been overlooked in the literature. Chinchilla Rodriguez et al. (2018) noted that Brazilian papers indexed on Clarivate Analytics from 2000 to 2016 follow a pattern that falls short of the world impact averages. Domestic papers (published among Brazilian authors) and those published in international collaboration in which Brazil takes the position of the corresponding author, reach a lower impact than the world average. Brazilian institutions lie above the world citation averages when they participate in international collaborations without occupying the position of the corresponding author.
University rankings gather indicators that should re ect several aspects of academic excellence (Hazelkorn, 2013). Bibliometric indicators of scienti c output and citation are present in every international university ranking and receive weights ranging from 20 to 100% of the nal score; despite employing different methodologies, most rankings produce similar results, suggesting that they measure a similar phenomenon (Robinson-Garcia et al., 2019) Kaycheng (2015) observed a strong correlation between the various criteria considered individually; some criteria contributed very little and would be masking, in fact, the relevant dimensions measured in the ranking. Robinson-Garcia et al. (2019) showed that, despite the differences in methodology as well as the weights attributed to publication and citation data, all rankings tend to measure a combination of the number of articles produced by the university and their relative citation impact. Therefore, despite the claims of precision and reliability that each ranking makes in its methodology, bibliometric indicators that combine the volume of publications and the volume of citations, as well as the average number of citations per publication, can offer an image of higher education that is very similar to the scenario shown by university rankings. It is in that context that coauthorship and the credits given to each author -and, as a result, to each university and country of a liation -gain some spotlight in current discussions.
The co-authorship phenomenon has raised the discussion about the methods for attributing credits or counting published papers and citations. The rst discussions on the subject date back to the early 1980s, when it was already noticeable that not all authors contribute equally (Solla Price 1981). In a paper signed in co-authorship, some authors may have contributed more; ideally, that should bear an impact on the attribution of credit. Authors' contributions in published papers fall into three categories: those who sign as the rst author, those who sign as the corresponding author, and those who contribute with no speci c role (Xiaojun, Rousseau, and Jin 2010). The corresponding author has gained status as the most important author among analytical approaches (Wouters et al. 2015). Corresponding authors are usually among the senior components of the group, the ones who most contribute to guarantee the research funding (Man et al., 2004); they are responsible for the research project, for bringing together the co-authors, and for the preparation of the paper. Several studies about scienti c collaboration (e.g. Bordons et al. 2015;Chinchilla Rodriguez et al. 2018) take the role of the corresponding author into account. Being a corresponding author has become a symbol of leadership and responsibility. In some countries, such as China, development agencies have started to register only the rst or the corresponding author. Clearly, the positions in the byline have become especially relevant in the scienti c arena (Xiaojun, Rousseau, and Jin, 2010). In this study, we will use the set of papers with at least a corresponding author belonging to an institution as one of the baselines to make comparisons with the papers derived from speci c international collaborations of the researchers from that institution.

Materials And Methods
The data used in this paper derive from InCites, a database created by Clarivate Analytics. Data of papers (article type) published between 2010 and 2019 by the 100 most productive and most cited Brazilian institutions were collected in June 2020 and January 2021. The analyses were carried out for data in the six areas of the GIPP InCites' schema: Engineering & Technology (ENG), Clinical, Pre-Clinical & Health (MED), Life Sciences (LIFE), Physical Sciences (NAT), Social Sciences (SSCI), Arts & Humanities (HUM).
First, we carried out a descriptive analysis of the number of papers produced by the selected Brazilian universities, including the number and percentage of citations, Category Normalized Citation Impact, and the number and percentage of papers published in journals of the First Quartile. We will use the "ALL" label for this set of papers. Then, we identi ed all papers in which at least an author from a Brazilian university took the role of the corresponding author. In the Tables that follow, results associated with this set of papers will lie under the "CORR AUTHOR" label.
Considering the relevance of international collaboration with Australia, Canada, the United Kingdom, and the United States (González Alcaide et al. 2017;Van Leeuwen 2009;Bordons et al. 2015;Moya Anegón et al. 2013), our study then identi ed the papers co-authored by Brazilian researchers and at least one researcher from any of those four countries. Results associated with this set of papers will lie under the "ACKS COLL" label. As we will like to compare the impact of international collaboration when it involves or not those four countries, the "NO ACKS COLL" label will refer to the set of papers co-authored by Brazilian researchers not included in the "ACKS COLL" collection.
The Academic Ranking of World Universities (ARWU), a.k.a. the Shanghai Ranking, produces an annual classi cation by subjects -Shanghai Ranking's Global Ranking of Academic Subjects (ARWUGRAS)-that brings visibility to many Brazilian universities listed in subject rankings close to their elds of specialization. ARWUGRAS comprises 54 subjects split into large areas, namely, Natural Sciences, Engineering, Life Sciences, Medical Sciences, and Social Sciences. Those areas roughly correspond to ve of the Clarivate's GIPP areas. The 2020 rankings rely on bibliometric results from the period 2014-2018, aggregated through a weighted combination of the following indicators: AWARD: For some subjects ARWU makes use of an additional indicator that refers to the "total number of an institution's staff that has won a signi cant award in an Academic Subject since 1981. The awards were identi ed through the ShanghaiRanking's Academic Excellence Survey" (ARWUGRAS 2020).
In the 2020 edition of ARWUGRAS, all the bibliometric indicators relied on raw measures of research parameters for publications (article type only) between 2014 and 2018. As stated in the ARWUGRAS Methodology (ARWUGRAS 2020), for all the indicators except CNCI nal scores "are computed from raw measures as the square root of the percentage of the top-scored institution". As of CNCI, "The maximum value of the indicator in a subject is set as the lower value of: (1) the twice of the average CNCI for all institutions in this subject; (2) the maximum of the CNCI for all institutions in this subject". Institutions' scores on CNCI are then "calculated as the proportion of their CNCI to the maximum value. If an institution's CNCI is higher than the maximum value, its score on CNCI will be assigned as 100" (ARWUGRAS 2020).
For this study, we have chosen all the subjects that include at least ve Brazilian universities in the corresponding ARWUGRAS Ranking. We have also selected the three subjects in the Social Sciences showing the largest presence of Brazilian universities.
Finally, we have included in the list of 21 subjects four additional topics that cover a substantial share of the scienti c production of Brazilian universities, namely Biology, Chemistry, Human Biology, and Materials Science & Engineering. Table 1 shows the list of subjects that we will analyze in the paper, along with the number of Brazilian universities in the corresponding ARWUGRAS classi cation (N BRAZIL), as well as the total number of universities listed (N TOTAL). To perform the simulations needed for the study we extracted from InCites the raw measures of all Brazilian universities ranked in the chosen subjects. The data for these simulations correspond to the period 2014-2018, as stated by the 2020 edition of ARWUGRAS. It constitutes, therefore, a subset of the complete set of data used in the other part of the study. To cancel the effect of international collaboration we recomputed the scores as follows: for every institution and subject, we assigned the average raw values of the indicators (Q1, CNCI, and TOP) obtained for the CORR AUTHOR collection of papers to the set of papers ACKS COLL.
For better readability, we have used acronyms for all the universities included in the tables that follow. Table 2 shows the correspondence of the chosen acronyms to the universities' names.  Tables 3 to 18,

Computation of the modi ed scores
To compute nal scores, we follow ARWUGRAS's methodology.
Let X be a Brazilian institution with a raw value of for a particular indicator. Let be the top raw score for that particular indicator among all the institutions ranked.
Hence, applying ARWU's methodology, the score of the institution X, , in that indicator is computed as follows: . Now, after discounting the effect of the papers in ACKS COLL, both the raw score of X and the maximum raw score in the table may change. Let then be the new raw score of X, and the new maximum score. The new nal score on the indicator, , will then be computed in the same fashion.
We use UNESP as an example to calculate the score in the indicator Q1 for the Agricultural Sciences. According to InCites, UNESP shows a raw value on Q1, , of 768. The top raw value among the institutions in Agricultural Science corresponds to Wageningen University & Research, 2531. UNESP score, , would then be ARWU rounds the value to the rst decimal digit, 55.1, as the ARWUGRAS webpage for the Agricultural Sciences shows.
Let us now compute the new scores after canceling the effect of the international collaboration.
Researchers from UNESP published 3354 papers in the Agricultural Sciences between 2014 and 2018, 768 of them in Q1 journals. The percentage of Q1 papers in that period (taking into account only papers with at least a corresponding author a liated with UNESP) was 22.2 for that subject. Out of the 768 Q1 contributions from UNESP in the period, 212 papers were included in the ACKS COLL set. Now, the total number of papers of UNESP in ACKS COLL was 451. Hence, assuming that those papers were from the "CORR AUTHOR" set, the expected number of Q1 papers would have been 100 (22% of 451). Therefore, the new raw score for UNESP in Q1 would be 768 − 212 + 100 = 656.
Carrying out the same computation for Wageningen University & Research produces a new Q1 raw score for that institution, 2487, which again reaches the maximum value of all the modi ed scores of universities in the ARWUGRAS list.
The new score in the indicator would then be In the next section, we present and discuss the results of the complete simulations.

Results And Discussion
To investigate the extent to which scienti c collaboration bene ts the research performance of Brazilian universities in terms of bibliometric impact, we have identi ed the total number of papers indexed in InCites for each university and the number of papers in which an author a liated with a Brazilian university assumed the position of the corresponding author.   On average, researchers from the 25 universities in Table 3 occupy the corresponding author position in 49.3% of their publications. Those Brazilian universities show, except in a few cases, a share of papers signed as the corresponding author larger than the average share of all Brazilian institutions (43.6%). Table 3 shows several disparities, starting with the great difference (an order of magnitude) between the institutions with the highest and lowest scienti c production in the period analyzed (column PUB). Although less noticeable, differences in institutional pro les can also be seen in CIT, PQ1, and CNCI.
The percentage of papers with a corresponding author from the institution varies from 38% at UERJ to 59.7% at UFSM and shows no signi cant correlation with the total number of papers (Pearson Correlation -0.125, p=.55). The percentage of citations to papers with a corresponding author from the institution (PCITC in Table 3) varies from 19.3% at UERJ to 59.7% at UFSM and shows no signi cant correlation with the total number of papers either (Pearson Correlation -0.061, p=.77).
The percentage of Q1 contributions decreases in the set of papers CORR AUTHOR but the gap is barely noticeable on average (8.5%), ranging from 0.9% for UFV to 20.8 for UERJ. This means that the association with other institutions (in or outside Brazil) in which the other institution takes the lead of the investigation helps but does not radically improve the ability of Brazilian universities to publish in the rst quartile. We will check later whether reducing the collaboration to the four English-speaking countries selected for our analysis does have a substantial impact on %Q1 gures.
When comparing the Category Normalized Citation Impact of all papers published by Brazilian institutions included in Table 3 ( Only four universities presented a CNCI equal to or greater than 1 when all papers are analyzed: UFPEL, UERJ, UFSC, and USP.
The indicator ranges from 0.63 for UFLA to 1.10 for UFPEL. The normalized impact of papers whose corresponding author is a liated with a Brazilian university, CNCIC, has a smaller range, reaching from 0.76, for FIOCRUZ and UFPEL, to 0.52 for UFMS.
FIOCRUZ and UFPEL, despite not being among the largest institutions in Table 2, are noteworthy due to their research impact, considering that they present the highest CNCI for the total number of papers and corresponding author. FIOCRUZ is an institution dedicated to the research in Biomedical Sciences, and UFPEL is a federal university with recognized leadership in that area, despite its multidisciplinary work. Both institutions are recognized for their research in Epidemiology (Guimarães, Lourenço and Cosac 2001) and are protagonists in the research on Covid 19 (Hallal 2021, FIOCRUZ 2021. FIOCRUZ is concentrated on three research areas (Parasitology, Infectious Diseases, and Tropical Medicine) that account for more than 30% of its scienti c production. It is no wonder than González Alcaide et al. (2017) have reported that the area of Tropical Medicine reveals Brazil's scienti c leadership above the United States, the United Kingdom, and other European countries, based on the high number of papers signed as the corresponding author. Table 4 presents results about the collaboration pro le of the 25 Brazilian institutions with the largest scienti c production between 2010 and 2019.  However, the difference in mean between CNCIA and CNCIC is about 4.5 times larger than the one between CNCIN and CNCIC.
It is worth pointing out that the dispersion of CNCIC is very low, with a short range (0.52, 0.76) and a standard deviation of 0.05.
On the other hand, CNCIA shows a far larger range (1.13, 3.04) and standard deviation (0.50). This indicates that a sizable share of the differences in impact among Brazilian institutions may be attributed to the collaboration with the four English-speaking countries. Indeed CNCI (from Table 3) and CNCIA (from Table 4) are positively correlated, and the association is statistically very signi cant (Pearson Correlation 0.89, p < .001).
To check whether this effect was associated with the size of the institutions we carried out the same analysis for the sample of 230 institutions in Brazil with at least 200 papers between 2010 and 2019. We found CNCI and CNCIA positively correlated and the association statistically very signi cant (Pearson Correlation 0.786, p < .001). We also investigated (for the sample of 230 institutions) the effects of the variables related to PUB in the relationship between CNCI and CNCIA through partial correlation analyses. We found that the association holds in all the cases, indicating that those covariates (PUB, PPUBA, PPUBC, and PPUBN) had little in uence in controlling for the relationship between CNCI and CNCIA. Table 5 shows the partial correlation results.
On average, the share of Q1 papers by Brazilian institutions in ACKS COLL is larger than the share of Q1 papers in CORR PAPER.
We  Table 5 shows, the covariates had little in uence in the relationship between PQ1 and PQ1A, except PPUBA (the percentage of papers in ACKS COLL) that explains part of the effect of PQ1A in PQ1. All the correlations statistically signi cant at p < .001 The advantage of collaborating with researchers from these four countries is also clear in the analysis of speci c areas. Table 6 shows the results of the collaboration with the four English-speaking countries in the six areas of the GIPP schema. We will contrast the results of the ACSK COLL set of papers with two baselines: the CORR AUTHOR and the NO ACSK COLL sets of papers. The percentage of papers in ACKS COLL uctuates around 20%, except in the case of Engineering & Technology and, most dramatically, the Arts & Humanities. The gains from the collaboration in terms of normalized impact are apparent for both baselines, particularly in the Health Sciences. In the case of the Arts &Humanities, in spite of the short number of papers in ACKS COLL and NO ACKS COLL there are substantial gains associated with international collaboration. However, one of the areas, Engineering and Technology, does not appear to improve from the collaboration to the same extent.
In Tables 7 to 18  We found again that the association (both for CNCI and PQ1) holds when controlling for relevant covariates, indicating that all the covariates had little in uence in the relationship between the two variables under study, as Table 8 shows. All the correlations statistically signi cant at p < .001 Table 9 presents the results for institutions with at least 1,800 papers in the Health Sciences between 2010 and 2019. Papers in the Health Sciences also show higher numbers in citation impact when written in collaboration with researchers from the four English-speaking countries. On average, institutions in Table 9 multiply the citation impact by 3.5, and all of them except UNESP and UFSM see their CNCI increase at least two-fold. The Health Sciences present the highest volume of citations and CNCIs in Brazilian institutions when compared with the other GIPP areas.
We As of the papers within NO ACKS COLL, the association also holds: CNCIN and CNCIC differ in the scores, CNCIN (M = 0.83, SD = 0.33) and CNCIC (M = 0.58, SD = 0.21), with t (180) = 10.73, p < .001. However, the difference in mean between CNCIA and CNCIC is about 6.5 times larger than the one between CNCIN and CNCIC.
Brazilian institutions show PQ1 data substantially higher for ACKS COLL papers (PQ1A) than for CORR AUTHOR papers (PQ1C).
We found again that the association (for both indicators, CNCI and PQ1) holds when controlling for relevant covariates, indicating that those covariates had little in uence in the relationship between the two variables under study, as Table 10 shows. All the correlations statistically signi cant at p < .001 Table 11 presents the results for institutions with at least 1,800 papers in the Physical Sciences between 2010 and 2019. We found again that the association (both for CNCI and PQ1) holds when controlling for relevant covariates, indicating that all the covariates had little in uence in the relationship between the two variables under study, as Table 12 shows. All the correlations statistically signi cant at p < .001 Table 13 presents the results for institutions with at least 300 papers in the Social Sciences between 2010 and 2019. Papers in the Social Sciences show higher numbers in citation impact when written in collaboration with researchers from English-speaking countries. On average, institutions in Table 13 multiply the citation impact by 3.5, and all of them except UNIFESP, UFPE, FGV, and PUCRS see their CNCI increase two-fold. PPUBI is negatively correlated with PPUBC, but the association is not statistically signi cant at p=.01.
We Brazilian institutions show %Q1 data substantially higher for ACKS COLL papers (PQ1A) than for CORR AUTHOR papers (PQ1C). There was a signi cant difference in the scores for PQ1A (M=45.75, SD=17.1) and PQ1C (M=27.50, SD=10.5), t(76) = 11.9, p <.001. PQ1 and PQ1A positively correlated, and the association is statistically very signi cant (Pearson Correlation 0.62, p<.001). We found again that the association (both for CNCI and PQ1) holds when controlling for relevant covariates, indicating that all the covariates had little in uence in the relationship between the two variables under study, as Table 14 shows. All the correlations statistically signi cant at p < .001 Table 15 presents the results for institutions with at least 100 papers in the Arts & Humanities between 2010 and 2019, and at least 10 papers in ACKS COLL. This is an area not well represented in the Web of Science; however, the analysis shows that papers in the Arts & Humanities do bene t from the ACKS collaboration in terms of citation impact, although we can compute the results only in a limited number of universities. On average institutions in Table 15, except FIOCRUZ, see their CNCI increase three-fold. PPUBI is negatively correlated with PPUBC, but the association is not statistically signi cant at p=.01.
We extended the analysis to the sample of 36 institutions in Brazil with at least 40 papers in the Arts & Humanities between 2010 and 2019. Brazilian institutions show CNCI data substantially higher for ACKS COLL papers (CNCIA) than for CORR AUTHOR papers (CNCIN). There was a signi cant difference in the scores for CNCIA (M=1.13, SD=1.5) and CNCIC (M=0.21, SD=0.12), t(35) = 3.7, p =.001. CNCI and CNCIA are positively correlated and the association is statistically very signi cant (Pearson Correlation 0.735, p<.001). As of the collaboration with other countries, the association also holds: CNCIN and CNCIC differ in the scores, CNCIN (M=0.74, SD=1.1) and CNCIC (M=0.21, SD=0.12), t(35) = 3.1, p =.004. However, the difference in mean between CNCIA and CNCIC is about 2 times larger than the one between CNCIN and CNCIC.
As of the percentage of Q1 papers, Brazilian institutions show PQ1 data slightly higher for ACKS COLL papers (PQ1A) than for CORR AUTHOR papers (PQ1C), but the differences in the mean are not statistically signi cant. PQ1 and PQ1A are positively correlated, and the association is statistically very signi cant (Pearson Correlation 0.582, p<.001). We found that the association (for CNCI and PQ1) holds when controlling for relevant covariates, indicating that the covariates had little in uence in the relationship between the two variables under study, as Table 16 shows.   CNCIN (M=0.95,SD=0.32) and CNCIC (M=0.80,SD=0.22), t(121) = 7.1, p <.001. However, the difference in mean between CNCIA and CNCIC is about 3 times larger than the one between CNCIN and CNCIC.
Brazilian institutions show PQ1 data substantially higher for ACKS COLL papers (PQ1A) than for CORR AUTHOR papers (PQ1C).
PQ1 and PQ1A positively correlated, and the association is statistically very signi cant (Pearson Correlation 0.62, p<.001). We found again that the association (both for CNCI and PQ1) holds when controlling for relevant covariates, indicating that all the covariates had little in uence the relationship between the two variables under study, as Table 18 shows. All the correlations statistically signi cant at p < .001 As we have seen, the analysis of the number of citations, CNCI, and percentage of rst quartile publications supports the validity of hypotheses a) and b), formulated in the Introduction, about the gains for Brazilian researchers from the scienti c collaboration with authors from Australia, Canada, the United Kingdom, or the United States.
Since major academic classi cations rely on bibliometric data, it is clear that the kind of international collaborations analyzed in the present study are bound to have a positive impact on the ranking positions of Brazilian universities. In particular, the analysis of the Shanghai Ranking by Subjects will enable us to identify the areas in which the positions in the ranking of Brazilian universities most bene t from the collaboration with the four English-speaking countries.
As stated in the section devoted to discussing the methodology, we recalculated the score of all the universities in the ARWUGRAS classi cations when bibliometric gures for papers in collaboration with those four countries were replaced by averages over papers in which the corresponding author belongs to the institution under analysis. The fact that we used the same procedure for all the institutions on the ARWUGRAS list, not only to Brazilian universities, helps in ensuring a fair comparison.
For each Brazilian university we computed the variation in the position in the ranking, as well as the ratio using the position in ARWUGRAS 2020 as baseline. We include the complete analysis of one of the subjects in which the presence of Brazilian universities is noticeable and then summarize the results of all the subjects analyzed in the paper. Table 19 shows the results in   Ecology, whereas Table 20 presents a general overview of all the subjects, including the average of all Brazilian universities listed in the corresponding ARWUGRAS subjects, along with the results of the universities listed in at least ve subjects.  As shown in Table 20, the areas that concentrate the largest share of the scienti c throughput of Brazilian universities clearly highlight the impact of the collaboration with Australia, Canada, The United Kingdom, and the USA. In particular, two subjects in the Health Sciences (Clinical Medicine and Public Health) account for more than 75% of the scienti c production of Brazilian institutions in that area: the large negative ratios point to the great impact in the respective AWUGRAS rankings of the international cooperation with researchers from major English-speaking countries. The two largest areas in the Life Sciences, Biology, and Human Biology, also show large negative ratios. The results in those four subjects are consistent with the large impact of the international ACKS collaboration in CNCI and the proportion of Q1 publications analyzed before in the paper.
In the Natural Sciences, Physics is the subject in which the bene ts of the cooperation are more noticeable. In the Social Sciences, the bene ts are also clear in Economics and Political Sciences.
The results in the Engineering subjects show that Brazilian universities, when all the institutions see the effect of the ACKS collaboration removed, keep up with their positions in the ARWUGRAS rankings. Three cases are worth highlighting: Mathematics, Pharmaceutical Sciences, and Management. In those three areas, the quality of the scienti c production having a corresponding author a liated with the Brazilian institutions helps in advancing through the ranking.

Conclusions
Results from this paper show that when a Brazilian university takes on the role of the corresponding author, the impact on the research decreases in most elds. Between 2010 and 2019, on average Brazilian institutions had a CNCI for papers in which a corresponding author was associated with them of 0.66, in contrast with the total CNCI of 0.86. These results are in line with the ones obtained by Moya Anegón et al. (2013). Those authors observed that there is a tendency for the impact to decrease for all countries with respect to the papers in which the corresponding author belongs to the country (except for the USA) when that impact is calculated in relation to the all the papers published by researchers from that country. The results presented in Tables 3 and 4 show that Brazil is not an exception to that rule.
The impact gains of papers in which the corresponding author is not a liated with a Brazilian institution was also observed by Grácio et al. (2019; for papers indexed by SCOPUS between 2003 and 2015. The authors showed that the impact increased by 68.1% when the Brazilian author was not the corresponding author. For the institutions in Table 3, the increased impact corresponding to the period 2010-2019 was 57.7%, very much in agreement with Grácio et al. ndings. Whether these ndings indicate that researchers led by the Brazilian university teams have a lower impact because they usually deal with topics of local interest is a question that we leave for further work, as it needs more research to be adequately addressed.
The quartile analysis of the journals in which the papers were published also reveals the positive in uence of collaborating with major English-speaking countries. On average, 51.5% of the papers in collaboration with these countries were published in journals of the rst quartile, almost twice the percentage found for the papers whose corresponding author is a liated with a Brazilian institution -which represent the 27.5% of the total number of papers.
When analyzing the results of the ARWUGRAS ranking, our ndings point to a general advantage gained by Brazilian universities in collaboration with Australia, Canada, the United Kingdom, or the United States. The results help in shedding light over one of the leading questions of this study posed in Sect. 1: "Is there an area of Brazilian science that bene ts the most from collaboration?" For a number of subjects, there would be a loss of positions if the collaborations with English-speaking countries were canceled; that would happen, notoriously, for Clinical Medicine, Public Health, Physics, Biology, and Human Biology, subjects that account for more than 40% of the scienti c production of Brazilian universities. To a lesser extent, the effect is also noticeable in the areas of Materials Science & Engineering, and Chemistry, subjects that account for more than 10% of the scienti c production of Brazilian universities. In the Social Sciences, the bene ts are apparent in Economics and Political Sciences.
The second question posed in Sect. 1, "Is there an area in which the Brazilian authors who take on the position of the corresponding author get more recognition?" can also be partially answered using the results from Table 20. Mathematics, Pharmacy, and Management emerge as the areas in which the extra help from international collaboration is not needed to advance in the respective rankings.
The study's initial hypotheses -namely, international collaboration with Australia, Canada, the United Kingdom and the USA were proven right based on the presented data.
Corroborating Bordons et al. (2015) and Wagner (2008), our results for some of the analyzed subjects show that researchers and institutions with unique knowledge can be found in Brazil. Those scientists can become strategic partners not only for their geographical, biological, or social conditions but also for their expertise in speci c disciplines. According to the report on scienti c collaboration published by the Royal Society (2011), science is becoming increasingly global, and the participation of all countries is necessary to face problems that affect all countries.
Prompted by an analysis of highly cited scientists in Latin America, Martinez and Sá (2020) stated that while yet a regional leader, as far as scienti c production is concerned, Brazil is still relatively peripheral to global science. Mc Manus et al. (2020) argue that their ndings point in a different direction, one in which "Brazilian researchers are seen to be effectively collaborating to world prominent themes of high impact and to advance the innovative science". Our results corroborate Mc. Manus et al. (2020) ndings in what concerns the advancement of innovative science in Brazil through international scienti c collaboration.
However, we have also shown that the gap between the impact of the science produced in Brazil with or without international collaboration with major English-speaking countries is still noticeable, particularly in Physics and the Health & Life Sciences.