91016 A DECADE OF DEVELOPMENT IN SUB-SAHARAN AFRICAN SCIENCE, TECHNOLOGY, ENGINEERING & MATHEMATICS RESEARCH • A REPORT BY THE WORLD BANK AND ELSEVIER worldban k.org/africa/stemresearchreport 1 Contents Contents 1 Executive Summary & Policy Recommendations 2 CHAPTER 1 Methodology 9 Methodology 10 1.1 CHAPTER 2 Research Outputs & Citation Impact 13 Key Findings 14 2.1 Research Output 15 2.2 Citation Impact 19 2.3 Research Per Capita 24 2.4 2.5 Novel Measures of Research Impact 25 2.6 Interpretation and Discussion of Chapter Key Findings 29 CHAPTER 3 Research Collaboration 33 Key Findings 34 3.1 International Collaboration 35 3.2 Citation Impact of Collaboration 41 3.3 Cross-sector collaboration 42 3.4 Top collaborating institutions 44 3.5 3.6 Interpretation of Key Findings on Research Collaboration 48 CHAPTER 4 Researcher Mobility 49 Key Findings 50 4.1 Researcher Mobility Model 51 4.2 International Mobility 53 4.3 Cross-Region Comparisons 55 4.4 4.5 Interpretation of Key Findings on Researcher Mobility 56 APPENDICES A Author Credits, Advisory Groups, & Acknowledgements 58 B Glossary 59 C Data Sources and Methodology 61 Data Sources 61 Methodology and Rationale 62 Measuring International Researcher Mobility 63 Measuring Article Downloads 65 D Africa Region Classification 66 E Subject Classification 68 F International Researcher Mobility Maps 69 Southern Africa 69 South Africa 70 West & Central Africa 71 Notes 72 executive summary & policy recommendations 2 A DECADE OF DEVELOPMENT IN SUB-SAHARAN AFRICAN SCIENCE, TECHNOLOGY, ENGINEERING & MATHEMATICS RESEARCH A REPORT BY THE WORLD BANK AND ELSEVIER In March 2014, several African governments’ ministers agreed on a Joint Call for Action in Kigali to adopt a strategy “Higher education is now front and center of the that uses strategic investments in science and technology development debate – and with good reason. to accelerate Africa toward a developed knowledge-based More than 50 percent of the population of society within one generation. The represented governments sub-Saharan Africa is younger than 25 years are part of the Partnership for Applied Science, Engineering of age, and every year for the next decade, we and Technology (PASET), an initiative of the World Bank that expect 11 million youth to enter the job market. supports efforts by African governments and their partners This so-called demographic dividend offers a to strengthen the role of applied science, engineering, and tremendous opportunity for Africa to build a technology in the development agenda. The ministers unani- valuable base of human capital that will serve mously acknowledged the need for specific measures to as the engine for the economic transformation improve relevance, quality and excellence in learning, and re- of our continent. … To be more competitive, search in higher education. Which specific measures should expand trade, and remove barriers to enter be taken? Answering this question requires new analyses new markets, Africa must expand knowledge based on credible data and public debate on the findings. and expertise in science and technology. This report is part of a broader, on-going effort to provide From increased agricultural productivity to more evidence and analysis on the supply of and demand for higher energy production, from more efficient skills, education and research within Science, Technology, and broadly available ICT services to better Engineering and Mathematics (STEM) for Africa’s socio- employability around the extractive industries, economic transformation and poverty reduction under the building human capital in science and aegis of the PASET. technology is critical to empower Africa to take advantage of its strengths.” The World Bank and Elsevier are partnering on this report to examine the research enterprise over a decade from 2003 to Makhtar Diop 2012 of three different geographies in sub-Saharan Africa World Bank’s Vice President (SSA): West & Central Africa (WC), East Africa (EA), South- for the Africa Region ern Africa (SA). The research performance of these regions is compared to that of South Africa (ZA), Malaysia, and High-level Forum on Higher Vietnam; the latter two countries had a comparable research Education, Science, and base to the SSA regions at the beginning of the period of Technology in Africa analysis. The report analyzes all science disciplines, but with March 13, 2014 in Kigali a special emphasis on research in the Physical Sciences & Science, Technology, Engineering, and Mathematics (STEM). The report focuses on research output and citation impact, When reading the report, we encourage the reader to not important indicators of the strength of a region’s research only consider the findings on research performance from enterprise. These indicators are correlated with the region’s the narrow sense of academic knowledge generation, but long-term development and important drivers of economic also see research patterns as predictors of the sub-con- success. Moreover, research is a key ingredient for quality tinent’s future ability to train knowledge workers within higher education. Given the shortcomings of reliable statis- specific domains and sectors. As such, the patterns tics on education and research in Africa, we hope the infor- revealed through this report constitute a crystal ball to mation contained in a bibliometric database will shed light assess the future ability SSA’s scientific and educational on regional collaboration within Africa, academia-business ability to solve its development challenges through its collaboration, and STEM capacity. own capacity. 3 Methodology This report uses the Scopus abstract and citation database to evaluate trends in research growth in SSA. While the re- port recognizes that indicators on peer-reviewed research outputs do not fully capture all research activity in SSA, this is the most systematic and objective foundation for analysis currently available. Although previous studies have also analyzed research output trends in SSA, this is the first report that provides comprehensive policy analysis and recommendations at a regional level and builds an analyti- cal foundation for stakeholder dialogue in driving the STEM Figure E.0 — Map of sub-Saharan agenda. Africa regions analyzed in this report. Key Findings and Policy Recommendations despite the regions’ strong growth, countries with compara- This report presents four main developments over the past ble levels of research output in 2003 such as Malaysia and decade in research in SSA. Vietnam grew even faster over the same period. Further- more, SSA’s output growth has overwhelmingly been driven 1) Sub-Saharan Africa has greatly increased both the by advances in Health Sciences research (approximately quantity and quality of its research output. 4 percent annual growth), which now accounts for 45% of ► All three SSA regions more than doubled their yearly all SSA research. The progress in the Health Sciences is research output from 2003 to 2012. great and much welcome news for two reasons. First, due ► SSA’s share of global research has increased from to the tremendous health challenges the continent faces, 0.44% to 0.72% during the decade examined. improved Africa-relevant health research and well-trained ► Citations to SSA articles comprise a small but grow- health workers will have a great impact on health outcomes. ing share of global citations, increasing from 0.06%- Second, the impressive improvement in SSA’s research ca- 0.16% for each of the regions to 0.12%-0.28%. pacity in the Health Sciences demonstrates that persistent ► All regions improved the relative citation impact of their support and funding from development partners and gov- research, with East Africa and Southern Africa raising ernments pays off. There is clearly a large scientific talent their impact above the world average between 2003 base in Africa, but this needs to be trained and nurtured. and 2012. ► The percentages of each of SSA region’s total output The World Bank recommends that African governments and that are highly cited have grown steadily over time. development partners accelerate support to research and However, SSA still accounts for less than 1 percent of research-based education in Africa to build the necessary the world’s research output, which remains a far cry from human capital to further increase research on solving Afri- its share of global population at 12 percent. In addition, can problems by Africans for Africans. MALAYSIA (31.0% CAGR) 20,000 TOTAL NUMBER OF ARTICLES 15,000 SOUTH AFRICA (10.5% CAGR) 10,000 WEST & CENTRAL AFRICA (12.7% CAGR) EAST AFRICA (12.0% CAGR) 5,000 VIETNAM (18.8% CAGR) SOUTHERN AFRICA (8.5% CAGR) 0 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 YEAR OF PUBLICATION Figure E.1 — Overall number of articles and Compound Annual Growth Rate (CAGR) for SSA regions and comparator countries, 2003-2012. Source: Scopus. executive summary & policy recommendations 4 2) SSA research output in Science, Technology, Engi- 70% neering, and Mathematics (STEM) lags that of other 60% subject areas significantly. This is evident by the fol- 50% lowing indicators: 40% ► Research in the Physical Sciences & STEM makes up 30% only 29% of all research in SSA excluding South Africa, 20% as shown in Figure E.2. In contrast, STEM constitute the 10% largest share of Malaysia and Vietnam’s total output (an average of 68%), and that share continues to grow. 0% A IC N IC L A IA M ► The share of STEM research in SSA has marginally FR RA IC IC FR ER A S A A TN Y FR FR T TH LA N declined by 0.2% annually since 2002. In comparison, IE A A E U A V T TH C O A A M S & S A U the share of STEM research has declined 0.1% annually T E O S S E W in South Africa and grew 2% annually in Malaysia and Vietnam. ■ Physical Sciences & STEM ■ Health Sciences ► In 2012, the quality of STEM research in SSA, as meas- ured by relative citation impact, was 0.68 (32 percent Figure E.2 — Percentage of total article output in the below the global average). This is below that of all disci- Physical Sciences & STEM versus the Health Sciences for plines in SSA (0.92) and the global average (1.00), and sub-Saharan Africa regions and comparator countries, it has virtually stayed the same since 2003. In contrast, 2012. Source: Scopus. STEM research in Malaysia, Vietnam and South Africa in 2012 was slightly above the world average (1.02) and 3) SSA, especially East Africa and Southern Africa, re- has improved 15% since 2003. lies heavily on international collaboration and visiting These findings indicate that research in STEM in SSA is faculty for their research output. lagging in terms of research quantity and citation quality. Ca- ► A very large share of SSA research is a result of inter- pacity within other sciences, in particular health, is improving national collaboration. In 2012, 79%, 70% and 45% of substantially more than STEM. all research by Southern Africa, East Africa, and West & Central Africa, respectively, were produced through in- Building on the empirical basis outside of this report, the ternational collaborations. In contrast, 68%, 45%, and World Bank suggests that this large STEM gap could be 32% of Vietnam, South Africa and Malaysia’s research linked to several factors: the low quality of basic education output, respectively, were produced through interna- in science and math within SSA; a higher education system tional collaborations. skewed towards disciplines other than STEM such as the ► A large percentage of SSA researchers are non-local humanities and social sciences; international research fund- and transitory; that is, they spend less than 2 years at ing – which comprises the majority of SSA research funding institutions in SSA. In particular, 39% and 48% of all – prioritizing health and agricultural research. East and Southern African researchers, respectively, fall into this category. Analyses from parallel studies suggest that to undergo an The high level of international collaboration testifies to economic transformation, SSA needs more and better STEM the noteworthy effort and interest of academia outside skills and knowledge to boost value-added and productivity of Africa to support SSA’s research capacity. Moreover, within key sectors, such as extractive industries, energy, international collaboration is highly instrumental in raising transport, and light manufacturing. The World Bank recom- the citation impact of SSA’ publications. At the same time, mends the following policies: for the majority of SSA’s collaboration partners, the relative ► Accelerate and persistently pursue policies to improve citation impact of such collaborations is actually higher than the quality and quantity of teaching of STEM at all levels those partners’ overall average impact, suggesting that of the education system, including for research and the collaboration is a win-win situation for Africa and the research-based education. international collaborators. Furthermore, mobile research- ► Systematically scale up support to STEM disciplines at ers (those who move between institutions in the SSA and the higher education and research level through, for ex- the rest of the World) tend to be more productive in terms of ample: bilateral university collaborations, post-graduate publications and more highly cited than those researchers scholarships, and encouraging international firms to con- who primarily stay in SSA. tribute to the development of STEM capacity in Africa. ► Coordinate higher education strategies with development However, SSA’s high reliance on international collaboration needs and rigorously implement priorities through effec- for research is a concern for the World Bank; it signals a tive funding instruments. lack of internal research capacity and the critical mass to The box on the next page provides one example from Uganda. produce international quality research on its own; particu- 5 Supporting high quality and relevant research: Uganda Millennium Science Initiative The Uganda MSI project (2007 – 2013) is an example of an initiative that makes use of innovative funding mechanisms such as competitive grants to enhance research capacity through teams and collaboration. The project aimed to produce more and better qualified science and engineering graduates and higher quality and more relevant research. Component One ($ 16.7 million) of the project focused on developing research capacity through competitively awarded grants. Component Two ($16.7 million) aimed to improve public understanding and appreciation of science and strengthen the institutional capacity. Key policy innovations include: Building human capital by linking research with post-graduate education to ►  develop the country’s scientific future Building capacity of research teams for high quality scientific research ►  Encouraging statistical and policy analysis through scientific research ►  Project design was adopted to the Ugandan context and level of scientific ►  development Major achievements include: Increased human capital in STI: the number of researchers increased from 261 to ►  720 and the number of S&T students increased from 24 to 41 (Ph.D), from 245 to 633 (MSc), and from 3,241 to 4,892 (BSc) Established the fully functional competitive funding mechanism evaluated by ►  Ugandan and international scientists setting a high standard Ratio of applicants to fundable proposals was 11:1 (highly competitive), with ►  selection of high quality research proposals with strong leaderships Developed the capacity of the Uganda National Council for Science and ►  Technology for national statistics on STI and the Uganda Industrial Research Institute, where the number of services offered increased four-fold and revenue increased from nil to UGX 67 million to enhance efficiency and self-sustainability Source: Uganda Millennium Science Initiative Implementation Completion and Results Report, 2013 larly within STEM. Furthermore, the transitory nature of ► Continue international collaboration, and scale-up col- many researchers may prevent researchers from building laboration within STEM. relationships with African firms and governments, reducing ► Scale-up post-graduate education in Africa – possibly the economic impact and relevance of research. Analyzing through regional collaboration. the underlying reasons for lack of capacity goes beyond the ► Continue scholarship funding for studies in Africa, pos- scope of the current bibliometric analysis, but we speculate sibly through “sandwich-programs” to ensure interna- that the following are among the key reasons: shortcomings tional exposure and include funding support to raise the in the scale and quality of PhD programs; research funding; quality of the African post-graduate program. research equipment; and faculty time and incentives for research. To increase SSA’s research capacity, the World Bank encourages stakeholders to consider an initial set of policy recommendations below: executive summary & policy recommendations 6 100% 100% NON-LOCAL, TRANSITORY RESEARCHERS AS % 90% 90% INTERNATIONAL COLLABORATIONS AS 80% 80% PERCENTAGE OF TOTAL OUTPUT OF TOTAL RESEARCHER BASE 70% 70% 60% 60% 50% 50% 40% 40% 30% 30% 20% 20% 10% 10% 0% 0% SOUTH WEST & CENTRAL SOUTHERN EAST EAST SOUTHERN WEST & CENTRAL SOUTH AFRICA AFRICA AFRICA AFRICA AFRICA AFRICA AFRICA AFRICA Figure E.3 — Level of international collaboration for SSA regions (2012) and percentage of non-local, transitory researchers for SSA regions, 1996-2013. Source: Scopus. 4) Research collaboration in Africa features a number of ► Returning diaspora contribute significantly to rais- particular characteristics that are critical to under- ing the citation impact of SSA research, specifically stand for the design of successful policies in East and Southern Africa. The inflow of returnees ► SSA’s research capacity appears fragmentized across researchers (those who originally publish from an regions, with each of the regions collaborating very little African institution, left and published elsewhere, and with one another. Inter-SSA collaborations (collaborations then subsequently returned) make up a relatively small without any South-African or international collaborator) share of the region’s total researcher base (3.6% and comprise just 2%, 0.9%, and 2.9% of all East African, 2.1%, respectively), yet the relative citation impact of West & Central African, and Southern African total re- those returnees’ publications is quite high compared to search output. that of other SSA researchers. This empirical finding If this observation about fragmentation is confirmed through corroborates the widespread belief that the large and more detailed country level analyses, national governments well-trained scientific African diaspora in Europe, North and regional bodies should consider regrouping researchers America and elsewhere should be further tapped to into larger groups either through funding incentives for team raise the quantity and quality of SSA research. research or through institutional mergers of higher education and research institutions, which is already happening in many ► West & Central Africa displays somewhat different pat- countries. Increasing Africa-Africa collaboration in science terns of researcher mobility and collaboration than East can also generate gains. This could be done through scaling and Southern Africa. A higher share of West & Central up existing regional research and research-based education African researchers is sedentary – i.e. not migrating programs that stimulate regional collaboration, such the Af- to institutions outside of their region (44% for West & rican Institute for Mathematical Sciences, the Africa Centers Central Africa vs. 24% and 15% for East and Southern of Excellence, the Regional Initiative for Science Education, Africa, respectively). Moreover, the share of non-African the Pan-African University, the Nelson Mandela Institutes for transitory researchers – i.e. visiting scholars – as a per- Science and Technology, and RU-FORUM. centage of the total regional researcher base is smaller in West & Central Africa. Furthermore, there are smaller There appears to be little knowledge transfer and col- ►  differences in the relative research productivity and laboration between African academics and the corporate impact of sedentary researchers and mobile research- sector, as measured by corporate downloads of and pat- ers. International collaboration comprises a smaller ent citations to African academic research, especially for share (42%) of West & Central Africa’s total research STEM disciplines. To the extent to which such knowledge output, and there is less research collaboration between transfer occurs, it occurs within Health Sciences and academia and other partners (corporate, government, through collaborations with global pharmaceutical com- and medical). In contrast intra-regional collaboration is panies. Such trends suggest that corporations do not rely 24.7% in West and Central Africa compared to 13.6% much on African-generated knowledge and research for for East Africa and 5.67% for Southern Africa. West their competitiveness. and Central Africa is more integrated within the region 7 as a result of institutions and researchers collaborating within the region. This report speculates that these dif- ferences could be driven by several factors, such as: a higher degree of collaboration and mobility for historical or policy reasons; a measurement bias if Francophone research is not adequately published or indexed; less donor funding for research to this part of Africa; and/or a higher share of unstable political environments. INTER-REGIONAL COLLABORATION 0.9%-2.9% Inter-African collaboration (without any South-African or international collaborator) comprises 2% of all East African research, 0.9% of West & Central Africa, and 2.9% of Southern Africa. Figure E.4 — Inter-regional collaboration between SSA regions. Source: Scopus. Defining national policies Following this overview, the introductory chapter intro- The report discusses and provides a big picture of research duces the underlying database and the main methodologi- trends at a regional level. We emphasize that this is a report cal approaches and concepts used in the report. The next rich on data, and we have only described the main find- chapter provides a broad overview of the research enter- ings. We recommend further analyses in three directions: prise in the different regions and across different subject examination of specific indicators at the regional level, more groupings by using a variety of metrics to examine the nuanced analysis of the factors behind the above identified quantity, usage, and quality of research output. What types developments, and particularly additional country-level of knowledge and how much are being generated by SSA analysis. Any country-level policy discussion on science, researchers? By whom and how much is that knowledge technology and innovation policy should build upon country- being used? Chapter 3 focuses on key aspects of research level analyses of research performance and its link to collaboration for the Africa regions. How frequently do institutional factors and education, research, and economic researchers in the different regions co-author articles with policies. Moreover, given the lack of regionally and interna- international colleagues or colleagues in non-academic tionally comparable information on the latter topics, such institutions? How impactful are those co-authored articles, exercises would be best accompanied by additional data and with which institutions do researchers collaborate the collection on national research and research-based educa- most? The final chapter focuses on the mobility of re- tion sectors. searchers to and from the different regions. While the report calls for increased national and interna- tional funding to research and research-based education at the master and doctoral level in Africa with a strong focus on STEM, we must keep in mind the substantial opportunity costs of research funding. The estimated cost of one doc- toral degree (USD 50,000) can fund 5 classrooms benefit- ting around 400 pupils in primary education or 25,000 textbooks in math. Therefore, it makes sense to closely tie funding for research and research-based education to Afri- can development challenges and ensure research findings The report is available online at and knowledge is applied towards solving these challenges. worldbank.org/africa/stemresearchreport​ Nevertheless, with a larger share of SSA having attained or within reach of becoming middle-income countries, the regions’ development will increasingly require greater scien- tific and technological capacity. © 2014 Elsevier B.V. All rights reserved. 9 CHAPTER 1 METHODOLOGY 1 methodology 10 Methodology 1.1  Approaches and Definition a recent report 6 only about 40% of the publications from Measuring scientific activity in low and middle-income the University of Dar es Salaam (UDSM) appear in serious, countries peer reviewed journals. Moreover, we acknowledge that a Past research studies have observed that the standards 1 lot of other peer-reviewed research is conducted in Africa used to measure and benchmark research performance that is not published in journals or proceedings covered by in advanced nations do not necessarily translate to less Scopus, often because these sources do not meet globally developed regions. First, the infrastructure for survey- accepted publication standards. Nevertheless, in focusing ing and collecting data on research and development on peer-reviewed research, the Scopus database captures (R&D) expenditures, number of researchers, and so forth one of the most common and globally commensurable is less developed.2 This report eschews such data collec- forms of research dissemination. tion issues by primarily focusing on research output data captured in Scopus. Scopus is an abstract and citation This report uses “article” as a shorthand to refer to the fol- database of peer-reviewed literature, covering over 50 lowing types of peer-reviewed document types indexed in million documents published in over 21,000 journals, book Scopus: articles, reviews, and conference proceedings. For series, and conference proceedings by over 5,000 publish- a more detailed explanation, see Appendix B: Glossary. ers. Moreover, one of the main advantages of this database is its multi-lingual and global coverage. Approximately 21% Defining subject areas of titles in Scopus are published in languages other than Properly and consistently defining subject areas is a key English, and the database contains over 400 peer-reviewed concern for quantitative approaches to research assess- titles from publishers based in the Middle East and Africa.3 ment. Based on discussions about the most relevant schema for categorizing sub-Saharan research, article and Second, the overall quantity of research inputs and outputs citation data were aggregated to 5 main subject group- of smaller, low-income countries are sometimes too small ings: Agriculture, the Physical Sciences & STEM (Science, and noisy to be reliably tracked and analyzed over time.4 Technology, Engineering, and Mathematics), the Health To avoid this issue, this report aggregates research output Sciences, the Social Sciences & Humanities, and the Life statistics from individual institutions and countries into four Sciences. We acknowledge that there could be alternate major regions. Moreover, the report draws on a range of groupings or classifications, such as combining Agricul- output metrics to better triangulate and verify broad sub- ture with the Life Sciences, and that the gains and impact Saharan Africa (SSA) trends in research performance. We of interdisciplinary sciences is not fully illustrated in the acknowledge, however, that the trade-off to this approach report. Nevertheless, these subject groupings are highly is that we cannot provide insights on country-level varia- instrumental for the analysis. tions in research performance that is important for national policymaking. Articles were classified in one or more of these groupings based on their underlying categorization according to the Third, as Siyanbola et al. (2014) note, the usual categories Scopus All Science Journal Classification (ASJC) codes. of science and technology indicators often do not capture This classification system does not and is not intended to or are not useful measures for "the local realities of STI map onto the department, program, or school divisions of systems. Agriculture, informal economy and indigenous any particular university or institution. For the calculation knowledge are three important aspects of African system of field-weighted citation impact, a more granular scheme that S&T indicators, to date, do not cover." 5 As the next encompassing more than 300 subject subfields (again, section details more extensively, this report defines subject consistent with the ASJC hierarchy) was used and then ag- groupings to more closely match the relevant dimensions gregated to the level of the main subject groupings. for SSA. More broadly, the analyses of research output data in this report are based upon recognized advanced Defining SSA regions and choosing comparator countries indicators, and our base assumption is that such indicators The choices to group SSA countries into the respective are useful and valid, though still imperfect and partial meas- regions detailed in Figure 1.1 were based on a preliminary ures. We acknowledge the limitations of drawing on publica- analysis of the respective similarities of various research in- tion data to capture even just research activity, let alone dicators across those countries. For example, due to funda- all scientific activity in SSA. Research activity has many mental differences in the state of research infrastructure, outlets for dissemination, from peer-reviewed research to the levels of research output, and the quality of research technical reports to policy briefs. For example, according to performance between South Africa and other Southern 1.1 methodology 11 Main Subject Grouping Scopus 27 Subject Classification Agriculture Agricultural and Biological Sciences Biochemistry, Genetics, and Molecular Biology Veterinary Physical Sciences & STEM Chemical Engineering Materials Science Chemistry Mathematics Computer Science Physics and Astronomy Earth and Planetary Science Energy Engineering Environmental Science Health Sciences Medicine Nursing Dentistry Health Professions Social Sciences & Humanities General (multidisciplinary journals such as Nature and Science) Arts and Humanities Business, Management, and Accounting Decision Sciences Economics, Econometrics, and Finance Psychology Social Sciences Life Sciences Immunology and Microbiology Neuroscience Pharmacology, Toxicology, and Pharmaceutics African countries, this report separates the former country measure of research citation impact to be explained later in from the latter region. In contrast, while Nigerian research this report, of Malaysian research published in 2003 was comprised more than 50% of the total output in West & 0.67 compared to that of West & Central Africa at 0.63. Central Africa between 2003 and 2012,7 the relative cita- Similarly, the FWCI of Vietnamese publications in 2003 tion impact of that country’s research, the distribution of was 1.02 compared to 0.88 for Southern Africa and 0.95 that country’s research across different subject areas, and for East Africa. However, we acknowledge that while the the relative rate of international collaboration were compa- research volume and citation impact of these countries rable to the larger region. As a result, although we consid- and regions have similar starting points, both Malaysia and ered treating Nigeria as a separate entity, its grouping with Vietnam have underlying economic differences that likely the larger West & Central Africa region does not distort the affected their capacity for scientific growth. The differ- larger trends. Throughout the report, numbers referring to ences in population size, income per capita and tertiary SSA as a whole exclude South Africa and refer specifically enrollment are all key to explaining the growth patterns that to East, West & Central and Southern Africa. are observed in the report. Analogously, Malaysia and Vietnam were selected as We also considered using the entirety of Southeast Asia comparators for the Africa regions due to the similarity as a comparator region, but we ultimately decided against in the quantity and impact of those countries’ research doing so for two reasons. First, as the somewhat divergent output with that of the Africa regions at the beginning trajectories undertaken by Malaysia and Vietnam attest, of this report’s analysis in 2003. For example, in 2003, there is considerably more variation in research perfor- Vietnam produced 587 research articles compared to 928 mance across countries in that region. Second, the level of by Southern Africa, and Malaysia produced 1815 research both research investment and the corresponding level of articles compared to 1900 by East Africa. Likewise, output for that region as a whole are much larger than all the field-weighted citation impact (FWCI), a normalized but South Africa. 1 methodology 12 West & Central Africa East Africa Benin Burundi Burkina Faso Comoros Cameroon Djibouti Cape Verde Eritrea Central African Republic Ethiopia Chad Kenya Congo, Democratic Republic of the Mauritius Congo, Republic of the Mayotte Cote d'Ivoire Rwanda Equatorial Guinea Seychelles Gabon Somalia Gambia, The South Sudan Ghana Tanzania Guinea Uganda Guinea-Bissau Southern Africa Liberia Angola Mali Botswana Mauritania South Africa Lesotho Niger Madagascar Nigeria Malawi Saint Helena, Ascension and Tristan da Cunha Mozambique Sao Tome and Principe Namibia Senegal Swaziland Sierra Leone Zambia Togo Zimbabwe Figure 1.1 — Map of sub-Saharan Africa regions analyzed in this report. 1 The OECD’s Frascati Manual is usually used as the gold standard. OECD. (2002). Proposed Standard Practice for Surveys on Research and  Experimental Development: Frascati Manual. Frascati, Italy. Retrieved from http://www.oecdbookshop.org/oecd/display.asp? LANG=EN&SF1=DI&ST1=5LMQCR2K61JJ 2 UNESCO. (2010). Measuring R&D: Challenges Faced by Developing Countries. Montreal.  Retrieved from http://www.uis.unesco.org/Library/Documents/tech 5-eng.pdf 3 For more information on Scopus, including its content coverage, please see Appendix C.  4 Gaillard, J. (2010). Measuring Research and Development in Developing Countries: Main Characteristics and Implications for the Frascati Manual.  Science Technology & Society, 15(1), 77–111. doi:10.1177/097172180901500104 5 Siyanbola, W. O., Adeyeye, A. D., Egbetokun, A. A., Sanni, M., & Oluwatope, O. B. (2014). From indicators to policy: issues from the Nigerian research  and experimental development survey. International Journal of Technology, Policy and Management, 14(1), 83. doi:10.1504/IJTPM.2014.058726 6 Thulstrup. E, Mlama .P, & Suntinen. E (2014). Study on Higher Education and Research in Tanzania. Report from Swedish Institute for Public  Administration. 7 To put things in perspective, if South Africa were treated as part of “Southern Africa,” South Africa’s research output would comprise ~85% of  “Southern Africa” total output. 13 CHAPTER 2 RESEARCH OUTPUTS & CITATION IMPACT This chapter provides a broad overview of how much research each SSA region produces and how impactful that research is. 2 research outputs & citation impact 14 Key Findings 2.1  “Forty or fifty years ago, many people thought that PUBLICATION OUTPUT GROWTH, 2003-2012 > 100% simply transferring technologies from industrial- ized to developing countries would close the tech- nology gap. Now we know that technologies devel- oped in industrialized countries may not be suitable for use in other environments. They may require a All SSA regions more than doubled their yearly particular type of infrastructure to operate. They research output. may need specialized parts or knowledge to mend when they break down. ... We now understand that innovative capacity must be built in different ways. Many developing countries can make important progress through simply adapting existing technol- SUBJECT AREA OUTPUT IN 2012 28.5% ogies. ... In a globalized world, technological devel- opment is a global venture. It requires a collective and coordinated effort by government, the private sector, scientists and civil society.” On average for the three SSA regions, research in un secretary-general ban ki-moon the Physical Sciences & STEM constituted 28.5% January 14, 2010 at Yale University  of their total output. In contrast, the average share of Health Sciences for the three regions was 45.2%. HIGHLY CITED ARTICLES IN 2012 FIELD-WEIGHTED CITATION IMPACT (FWCI) 7.5%-16% 0.92 Between 7.5% and 16% of the different SSA Research output across the three SSA regions regions’ total outputs were amongst the world’s achieved a FWCI of 0.92 in 2012, meaning it was top 10% most highly cited articles, but only 5.9% cited 8% less than the world average. However, the -10% of those same regions’ total output in the regions’ average FWCI in the Physical Sciences & Physical Sciences & STEM met that threshold. STEM was only 0.68 in 2012, and it has virtually stayed the same since 2003. 2.2 research output 15 Research Output 2.2  2.2.1. Total Research Output and Growth From 2003 to 2012, sub-Saharan Africa significantly in- creased the amount of peer-reviewed research it produced. As Figure 2.1 demonstrates, all three Africa regions more than doubled their total yearly article 8 output. For exam- ple, Southern Africa researchers produced 928 articles in 2003 and 1940 in 2012. West & Central Africa research- ers produced 3,069 articles in 2003 and 8,978 in 2012. The compound annual growth rates (CAGRs) 9 for research output exceeded 10% for both East and West & Central Africa (Southern Africa still grew at a respectable 8.5% annually). Despite the strong research output growth by the Africa regions, the comparator countries grew even faster over the same period. Malaysia, whose article output in 2003 was similar to that of East Africa, grew its output by 31% per year. Similarly, Vietnam, whose article output in 2003 was about two-thirds the level of Southern Africa, grew its output by 18.8% per year. MALAYSIA (31.0% CAGR) 20,000 15,000 TOTAL NUMBER OF ARTICLES SOUTH AFRICA (10.5% CAGR) 10,000 WEST & CENTRAL AFRICA (12.7% CAGR) 5,000 EAST AFRICA (12.0% CAGR) VIETNAM (18.8% CAGR) SOUTHERN AFRICA (8.5% CAGR) 0 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 YEAR OF PUBLICATION Figure 2.1 — Overall number of articles for SSA and comparator countries, 2003-2012. Source: Scopus. 8 This report uses “article” as a shorthand to refer to the following types of peer-reviewed document types indexed in Scopus: articles, reviews, and con-  ference proceedings. For a more detailed explanation, see Appendix B: Glossary. 9 Compound Annual Growth Rate (CAGR) is the year-on-year constant growth rate over a specified period of time. Starting with the earliest value in any  series and applying this rate for each time interval yields the amount in the final value of the series. The full formula for determining CAGR is provided in Appendix B: Glossary. 2 research outputs & citation impact 16 2.2.2. World Article Share Over the past decade, the total research output of the world has also risen, and world article share 10 provides a normal- ized measure of the regions’ growth. As Figure 2.2 shows, since every region’s world publication share increased from 2003 to 2012, their output growth rates outpaced the world’s overall growth. Collectively, the SSA’s share of glob- al research increased from 0.44% to 0.72%. The overall findings about sub-Saharan Africa’s world publication share suggest a reversal in the trends reported in Tijssen’s (2007) analysis of Africa’s research output from 1980-2004,11 which had found that “Africa's share in worldwide science has steadily declined.” However, certain regions grew more quickly than others. West & Central Africa increased its world article share from 0.23% in 2003 to 0.40% in 2012, achieving a CAGR of 6.3%. In contrast, Southern Africa barely increased its share from 0.07% to 0.09%. However, with a population of 0.9 billion, SSA accounts for 12.5% of the global population, a far cry from its less than 1% share of the world’s research output. This shows a large gap in Africa’s capacity to produce new knowledge in relation to its share of the world population and presents potential for rapid growth. 0.60% SOUTH AFRICA (4.4% CAGR) 0.50% 0.40% WEST & CENTRAL AFRICA (6.3% CAGR) WORLD PUBLICATION SHARE 0.30% EAST AFRICA (5.7% CAGR) 0.20% 0.10% SOUTHERN AFRICA (2.8% CAGR) 0% 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 YEAR OF PUBLICATION Figure 2.2 — World publication shares for SSA and comparator countries, 2003-2012. Source: Scopus. 10  The share of publications for a specific region expressed as a percentage of the total world output – see Appendix B: Glossary. 11  Tijssen, R. J. W. (2007). Africa’s contribution to the worldwide research literature: New analytical perspectives, trends, and performance indicators. Scientometrics, 71(2), 303–327. doi:10.1007/s11192-007-1658-3 2.2 research output 17 2.2.3. Output and Growth by Subject Grouping Although overall article outputs rose for all regions from 2003 - 2012, certain subject groupings grew faster than others. As Figure 2.3 shows, in every SSA region, research in the Health Sciences comprised the highest percentage of those regions’ total article output. At one extreme, research in the Health Sciences accounted for 47.8% of EA’s total output in 2012. On average, research in the Health Sciences comprised 45.2% of the SSA’s total research output. In contrast, the Physical Sciences & STEM has been the main area of focus for South Africa, constituting 44.7% of the country’s total output in 2012. However, for the other Africa regions, the Physical Sciences & STEM comprises between only 25% and 30% of their total research output in 2012. The Africa regions’ comparator countries provide a stark contrast. As Table 2.1 reveals, over 67% of Malaysia and Vietnam’s article output in 2012 was in the Physical Sciences & STEM. EAST AFRICA Agriculture SOUTHERN AFRICA Agriculture 50% 50% 40% 40% 30% 2003 30% 2003 Life Physical Life Physical 20% 20% Sciences Sciences / Sciences Sciences / 10% STEM 10% STEM 0% 0% 2012 2012 Social Sciences & Humanities Health Sciences Social Sciences & Humanities Health Sciences WEST & CENTRAL Agriculture SOUTH AFRICA Agriculture AFRICA 50% 50% 40% 40% 30% 2003 30% 2003 Life Physical Life Physical 20% 20% Sciences Sciences / Sciences Sciences / 10% STEM 10% STEM 0% 0% 2012 2012 Social Sciences & Humanities Health Sciences Social Sciences & Humanities Health Sciences Figure 2.3 — Percentage of total article output by subject grouping for SSA and South Africa, 2003 vs. 2012. Source: Scopus. 2 research outputs & citation impact 18 Moreover, as the individual radar charts reveal and Table 2.2 details more closely, while absolute output across all subject groupings increased over time, the share of STEM research in SSA has actually marginally declined by 0.2% annually since 2003. In contrast, despite Malaysia and Vietnam’s high relative output in the Physical Sciences & STEM, these comparator countries further increased their relative output in this area from 2003 to 2012, grow- ing 2% annually. On the other hand, relative output in the Health Sciences and the Social Sciences & Humanities increased in all SSA regions. Past research has also identified and expressed concern about the overall skew of African research toward the Health Sciences and Agriculture and away from the Physi- cal Sciences & STEM, a trend dating back to the 1990s.12 Pouris and Ho (2013) comment, “The continent’s research emphasizes medical and natural resources disciplines to the detriment of disciplines supporting knowledge based economies and societies.” 13 Table 2.1 — Percentage of total article output by subject groupings for Africa regions and comparator countries, 2012. For each subject area (row), the region with the highest percentage is encircled. Source: Scopus. Southern East West & Central South Africa Africa Africa Africa Malaysia Vietnam Physical Sciences & STEM 28.0% 25.3% 32.3% 44.7% 69.2% 67.9% Agriculture 33.4% 34.4% 28.2% 22.9% 15.3% 22.0% Health Sciences 44.8% 47.8% 43.1% 26.5% 13.1% 16.5% Social Sciences & Humanities 17.5% 15.4% 14.0% 21.8% 19.4% 8.4% Life Sciences 15.7% 15.0% 15.2% 8.7% 5.1% 8.6% Table 2.2 — CAGR for changes in percentage of total article output by subject groupings for Africa regions and comparator countries, 2003-2012. For each subject area (row), the region with the highest CAGR is encircled. Source: Scopus Southern East West & Central South Africa Africa Africa Africa Malaysia Vietnam Physical Sciences & STEM -1.7% -0.4% 1.4% -0.1% 2.1% 1.9% Agriculture 0.2% -2.6% -1.7% -3.7% -7.4% -1.9% Health Sciences 4.5% 4.1% 3.2% 2.8% -6.1% -2.9% Social Sciences & Humanities 3.6% 4.4% 5.1% 3.4% 9.1% 0.3% Life Sciences -2.6% -4.7% -3.7% -0.9% -3.3% -3.9% 12  Arvanitis, R., Waast, R., & Gaillard, J. (2000). Science in Africa: A bibliometric panorama using PASCAL database. Scientometrics, 47(3), 457–473. doi: 10.1023/A:1005615816165; Chuang, K.-Y., Chuang, Y.-C., Ho, M., & Ho, Y.-S. (2011). Bibliometric analysis of public health research in Africa: The overall trend and regional comparisons. South African Journal of Science, 107(5/6). doi:10.4102/sajs.v107i5/6.309 13  Pouris, A., & Ho, Y.-S. (2013). Research emphasis and collaboration in Africa. Scientometrics, 98(3), 2169–2184. doi:10.1007/s11192-013-1156-8 2.3 citation impact 19 Citation Impact 2.3  2.3.1. World Citation Share often than those in mathematics. Second, different types of The number of citations received by an article from subse- articles are cited with varying baseline frequencies. Review quently published articles is widely recognized as a proxy articles receive on average more citations than regular jour- for the quality or importance of that article’s research.14 As nal articles. A more sophisticated way of analyzing citation Figure 2.4 shows, citations to articles by the SSA regions impact is to use field-weighted citation impact (FWCI). FWCI and their comparator countries comprise a small but grow- normalizes for differences in citation activity by subject ing share of global citations. For example, Southern Africa’s field, article type, and publication year. This enables the share of global citations more than doubled from 0.06% in comparison of citation impact across subject areas with dif- 2003 to 0.12% in 2012, a CAGR of 8%. The other regions ferent publication velocities and or publication type norms. experienced similarly strong growth rates in their world ci- tation share, though they are modest in comparison to that The world is indexed to a value of 1.00. A FWCI of more of the comparator countries in Asia. For instance, Malay- than 1.00 indicates that the entity’s publications have been sia’s global citation share increased more than six-fold from cited more than would be expected based on the global 0.09% to 0.56%, which is less surprising given Malaysia’s average for similar publications. For example, Southern corresponding increase in research output. Africa’s FWCI in 2012 of 1.39 indicates that the average article from that region in that year has been cited 39% 2.3.2. Field-Weighted Citation Impact more than the world average. In contrast, Southern Africa’s Although citations provide an intuitive proxy for research FWCI in 2003 of 0.88 indicates that articles from that impact, they can be problematic for two reasons. First, region in that year were cited 12% less than the world aver- citations are usually not comparable across fields. For age. Collectively, the SSA regions achieved of FWCI a 0.92 instance, articles in the Life Sciences tend to be cited more in 2012. For more details, please see Appendix B: Glossary. 0.80% SOUTH AFRICA 0.70% 0.60% MALAYSIA 0.50% WORLD CITATION SHARE 0.40% 0.30% EAST AFRICA WEST & CENTRAL AFRICA 0.20% VIETNAM 0.10% SOUTHERN AFRICA 0% 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 YEAR OF PUBLICATION Figure 2.4 — World citation share across all subject groupings for SSA regions and comparator countries, 2003-2012. Source: Scopus. 14  Davis, P. M. (2009). Reward or persuasion? The battle to define the meaning of a citation. Association of Learned and Professional Society Publishers. doi: 10.1087/095315108X378712 2 research outputs & citation impact 20 Figure 2.5, which graphs the impact of research produced 2.3.3. Impact by Subject Groupings by the Africa regions and their comparator countries Just as the relative quantity of outputs produced by the dif- against their respective world article share over time, ferent regions varied across subject groupings, the relative provides a visual contrast of the different paths that the quality of said outputs also differed. Figure 2.6 to Figure regions took over the past decade. All three SSA regions 2.8 display the trends in the FWCIs in the Physical Sciences improved the relative citation impact of their research, but & STEM, Agriculture and Health Science versus the respec- there are significant variations across the regions in the tive world article shares in those subject groupings from baseline FWCI level and the trends in FWCI growth or stag- 2003-2012. nancy from 2003 to 2012. The regions’ impact in the Physical Sciences & STEM is Southern Africa has improved the impact of its research much lower than their overall average. For instance, South- output the most, growing its FWCI from 0.88 in 2003 to ern Africa’s overall FWCI in 2012 was 1.39, but its FWCI 1.39 in 2012. However, Southern Africa did not increase in the Physical Sciences & STEM was 0.94, just below the its world article share much. In contrast, West & Central world average. More importantly, the impact of the regions’ Africa increased the quantity of its output over time, output in the Physical Sciences & STEM has improved little outpacing the world’s average growth to improve its world over time. All three SSA regions still have subject grouping article share, but it made little gains in the overall quality FWCIs below the world average. Although West & Central of its research. Likewise, Vietnam modestly increased its Africa’s research impact in the Physical Sciences & STEM world article share, but it did not significantly change its improved from 0.56 in 2003 to 0.63 in 2008, it regressed aggregate citation impact. to 0.56 in 2012. In contrast, the impact of Malaysia and Vietnam’s research output in the Physical Sciences & STEM East Africa and South Africa developed in a hybrid manner, have both improved significantly over the past decade. initially increasing their overall FWCI and then shifting to- ward increasing their world article share. South Africa and Similarly, while the Africa regions grew the impact of their East Africa have also increased the impact of their research research output in Agriculture at roughly the same rate as output from below the world average to above the world their overall impact, the baseline impact for Agriculture average. Similarly, Malaysia has increased both its world was much lower. However, in contrast to the Physical Sci- article share and its research impact, though as of 2012, it ences & STEM, the impacts of those regions’ outputs have is still below the world average (0.92). increased over time. 1.4 2003 2012 SOUTHERN AFRICA 1.3 FIELD-WEIGHTED CITATION IMPACT 1.2 EAST AFRICA SOUTH AFRICA 1.1 VIETNAM WORLD AVERAGE 1.0 0.9 MALAYSIA 0.8 0.7 WEST & CENTRAL AFRICA 0.6 0% 0.1% 0.2% 0.3% 0.4% 0.5% 0.6% 0.7% 0.8% 0.9% 1.0% WORLD ARTICLE SHARE FOR ALL SUBJECT GROUPINGS Figure 2.5 — FWCI versus world article share for all subject groupings for SSA regions and comparator institutions, 2003-2012. Source: Scopus 2.3 citation impact 21 In contrast to the other subject groupings, the regions’ out- The contrast between the trends in Figure 2.6 and Figure put in the Health Sciences achieved a much higher impact 2.8 provide another perspective on the regions’ divergent than those regions’ overall output. For instance, articles by subject grouping trajectories. While all three SSA regions Southern Africa in this subject grouping in 2012 attained increased both the quantity and quality of their output in a FWCI of 1.85, or nearly 85% above the world average. the Health Sciences, progress in the Physical Sciences & Similarly, East Africa and South Africa’s output in 2012 STEM has been more limited. Health Sciences has driven attained impact levels far above their aggregate regional the regions’ overall research growth. average. Even West & Central Africa, whose FWCI in the Health Sciences was still below the world average (0.77), outperformed its overall FWCI (0.66). 1.2 1.1 SOUTH AFRICA WORLD AVERAGE 1.0 FIELD-WEIGHTED CITATION IMPACT SOUTHERN AFRICA VIETNAM 0.9 0.8 EAST AFRICA MALAYSIA NB: Malaysia's world article share reaches 1.16% 0.7 and its FWCI reaches 1.00 in FWCI. 0.6 WEST & CENTRAL AFRICA 0.5 0% 0.2% 0.4% 0.6% 0.8% 1.0% WORLD ARTICLE SHARE FOR THE PHYSICAL SCIENCES & STEM Figure 2.6 — Field-weighted citation impact versus world article share for the Physical Sciences & STEM for SSA regions and comparator countries, 2003-2012. Source: Scopus 1.2 1.1 SOUTHERN AFRICA WORLD AVERAGE 1.0 FIELD-WEIGHTED CITATION IMPACT VIETNAM SOUTH AFRICA 0.9 EAST AFRICA 0.8 MALAYSIA 0.7 0.6 WEST & CENTRAL AFRICA 0.5 0% 0.2% 0.4% 0.6% 0.8% 1.0% WORLD ARTICLE SHARE FOR AGRICULTURE Figure 2.7 — Field-weighted citation impact versus world article share for Agriculture for SSA regions and comparator countries, 2003-2012. Source: Scopus 2 research outputs & citation impact 22 2.0 VIETNAM 1.8 EAST AFRICA 1.6 FIELD-WEIGHTED CITATION IMPACT 1.4 SOUTH AFRICA 1.2 SOUTHERN AFRICA WORLD AVERAGE 1.0 MALAYSIA 0.8 0.6 WEST & CENTRAL AFRICA 0.4 0% 0.2% 0.4% 0.6% 0.8% 1.0% WORLD ARTICLE SHARE FOR THE HEALTH SCIENCES Figure 2.8 — Field-weighted citation impact versus world article share for the Health Sciences for SSA regions and comparator countries, 2003-2012. Source: Scopus 2.3.4. Research Excellence outputs in 2012 achieved that mark, reflecting a CAGR of Citations are not evenly distributed across articles. There is 7.0% over the decade. usually a strongly skewed distribution, with a small propor- tion of all published articles receiving the majority of the However, similar to the trends in FWCI, West & Central citations, a “long tail” of articles receiving the remainder, Africa lags behind the other regions in terms of its relative and a significant proportion of all articles never receiving a production of highly-cited articles. It grew its percentage single citation.15 Recent research suggests that not only is of 90th percentile articles from 5.5% in 2003 to 7.5% in an examination of the small proportion of the most highly- 2012, levels below what one would expect if the region’s cited articles a robust approach to research assessment,16 output matched the world average distribution. it may yield insights hidden from aggregate measures. Figure 2.10 to Figure 2.12 provide a more in-depth ex- Similar to the methodology behind FWCI, this report defines amination of the regions’ highly cited output at the subject highly cited articles as those in the top X% worldwide in grouping level. South Africa consistently increased its citation counts relative to all articles published in the same highly cited article output in the Physical Sciences & STEM, year and subject area. As Figure 2.9 shows, the percent- but the trends for the other regions are less even. From age of each of the regions’ total output that are highly cited 2003 to 2012, South Africa grew the percentage of its articles – that is, articles that meet the threshold for being Physical Sciences & STEM output in the world's top 10% considered amongst the world’s top 10% (e.g., those in the from 10.5% to 14.5%. For the other regions, the level of 90th percentile) in terms of citation count, has grown stead- highly cited articles in this subject grouping increased from ily over time. For instance, for East and Southern Africa, 2003 to 2008 but declined from 2008 to 2012. For exam- highly cited articles comprised at least 14.6% of their total ple, the percentage of East Africa’s output in the world's top output in 2012. While 8.7% of Southern Africa’s outputs in 10% grew from 11.8% in 2003 to 13.4% in 2008 before 2003 were in the world's top 10%, 16.0% of that region’s falling to 9.8% in 2012. 15  De Solla Price, D.J. (1965). “Networks of Scientific Papers”. Science 149 (3683): pp. 510-515. doi: 10.1126/science.149.3683.510 16  Bornmann, L., Leydesdorff, L., Walch-Solimena, C., & Ettl, C. (2011). Mapping excellence in the geography of science: An approach based on Scopus data. Journal of Informetrics 5(4): pp. 537–546. doi: 10.1016/j.joi.2011.05.005; Bornmann, L., & Marx, W. (2013). How good is research really? Measuring the citation impact of publications with percentiles increases correct assessments and fair comparisons. EMBO reports 14(3): pp. 226–30. doi: 10.1038/embor.2013.9 2.3 citation impact 23 Across all three SSA regions, although the percentage of highly cited article output in Agriculture remained well below the regions’ overall percentages, it increased significantly from 2003 to 2012. For example, in 2003, only 3.3% of Southern Africa’s outputs in Agriculture were in the world's top 10% in terms of citation counts, but in 2012, 7.9% were. The regions’ relative output of highly cited articles in the Health Sciences has consistently increased over the past decade, with Southern Africa achieving the highest absolute percentage growth. From 2003 to 2012, Southern Africa grew its percentage of output in the world's top 10% in the Health Sciences from 10.0% to 17.3%. Southern Africa Southern Africa 20% 10% 8% 15% 2012 Vietnam South Africa Vietnam 6% South Africa 2003 10% 4% 5% 2% 2012 0% 2003 0% 2008 2008 Malaysia East Africa Malaysia East Africa West & Central Africa West & Central Africa Figure 2.9 — Percentage of total publications with citation Figure 2.10 — Comparing percentage of publications on counts in the 90 percentile worldwide for SSA regions and th the Physical Sciences & STEM with citation counts in the comparator countries, 2003-2012. Source: Scopus. 90th percentile worldwide for SSA regions and comparator countries, 2003-2012. Source: Scopus. Southern Africa Southern Africa 15% 20% 15% 2012 10% Vietnam South Africa Vietnam 2003 South Africa 10% 5% 2012 5% 2008 0% 0% 2008 2003 Malaysia East Africa Malaysia East Africa West & Central Africa West & Central Africa Figure 2.11 — Comparing percentage of publications on Figure 2.12 — Comparing percentage of publications on Agriculture with citation counts in the 90 percentile world- th the Health Sciences with citation counts in the 90th percen- wide for SSA regions and comparator countries, 2003- tile worldwide for SSA regions and comparator countries, 2012. Source: Scopus. 2003-2012. Source: Scopus. 2 research outputs & citation impact 24 Research Per Capita 2.4  Research productivity at a national level refers to the & Central Africa and East Africa are slightly more produc- capability of converting research inputs, such as R&D tive in terms of articles per million USD$ GDP. In 2011, expenditures and human capital, into research outputs, West & Central Africa produced 0.048 articles per million such as articles and citations. Due to limitations in the data USD$, while East Africa produced 0.034 articles per million availability of more precise research inputs  17 for the Africa USD$. regions, this report draws on basic population and GDP data from the World Bank Africa Development Indicators. When normalizing for population size, however, South In contrast to previous indicators, data is available only for Africa is the most productive, producing 242.6 articles per 2006-2011. million people in 2011, an increase from 160.5 articles per million people in 2006. In contrast, the closest SSA region As Figure 2.13 shows, although South Africa’s GDP (and is West & Central Africa, which generated 47.8 articles per hence capacity to invest in R&D, training human capital, and million people in 2011, an increase from 30.2 articles per so forth) is much larger than that of the SSA regions, West million people in 2006. 0.050 WEST & CENTRAL AFRICA 0.040 ARTICLES PER MILLION USD$ GDP EAST AFRICA 0.030 SOUTH AFRICA 0.020 SOUTHERN AFRICA 0.010 Figure 2.13 — Articles per million 0 GDP (PPP, current US$) for Africa 2006 2007 2008 2009 2010 2011 regions, 2006-2011. Source: Scopus YEAR OF PUBLICATION and Africa Development Indicators. 50 NB: Values for South Africa too large to be displayed WEST & CENTRAL AFRICA 45 40 ARTICLES PER MILLION PEOPLE 35 30 25 20 EAST AFRICA 15 SOUTHERN AFRICA 10 5 Figure 2.14 — Articles per million 0 people (population) for Africa regions, 2006 2007 2008 2009 2010 2011 2012 2006-2011. Source: Scopus and YEAR OF PUBLICATION Africa Development Indicators. 17  According to the UNESCO Institute of Statistics, data on gross expenditures on R&D (GERD) is available for only 11 of the 52 countries comprising the three Africa regions in 2008 and only 5 countries in 2012. Likewise, data on the number of researchers is available for only 7 of the 52 countries in 2008 and only 4 in 2012. Trend analyses are not possible but the boxes at the end of this chapter provide insights on the GERD and researcher numbers across fields. http://www.uis.unesco.org/ScienceTechnology/Pages/research-and-development-statistics.aspx 2.5 novel measures of research impact 25  ovel Measures of 2.5 N Research Impact  Citations represent one path through which academic re- to accrue. Research on publication download measure- search is utilized, but it is neither meant to nor does a good ments and their implications is an emerging topic within the job of capturing the impact of academic research outside bibliometric community.20 academia. There is increasing interest in creating more and better indicators of the use and commercialization of Since full-text journal articles reside on a variety of research. Download usage and patent citations may provide publisher and aggregator websites, there is no central new, alternative ways of understanding usage of academic database of download statistics available for comparative research and linking academic research to larger societal analysis. Despite this, downloads are nonetheless a useful impact.18 indicator of early interest in, or the emerging importance of, research. This report uses full-text article download data 2.5.1. Article Downloads as Potential from Elsevier’s ScienceDirect database, which provides Predictor of Future Impact approximately 20% of the world’s published peer-reviewed Article downloads from online platforms are an alterna- journal articles, to offer an alternate perspective on how an tive metric used as a predictor of future research impact. institution’s research is being used around the world. Measuring impact through citations is particularly difficult for recently published articles. Citation impact is by defini- For this report, a download is defined as either download- tion a lagging indicator. The accumulation of citations takes ing a PDF of an article on ScienceDirect or looking at the time. After publication, articles need to first be discovered full text online on ScienceDirect, without downloading the and read by the relevant researchers; then, those articles actual PDF. Views of paper abstracts are not counted. might influence the next wave of studies conducted and Multiple views or downloads of the same article in the same procedures implemented. For a subset of those studies, the format during a user session are filtered out, in accord- results are written up, peer-reviewed, and published. Only ance with the COUNTER Code of Practice. 21 Moreover, as then can a citation be counted toward that initial article. a proxy for the influence and impact of Africa’s research Moreover, citations do not necessarily capture the full on industry, this report separately analyzes the download extent to which an article is being used and may systemati- trends of ScienceDirect users in the corporate institutions cally understate the impact of certain types of research versus non-corporate ones. (clinical versus basic). 19 Since the pipeline from initial publication to receiving a citation is long and leaky, data on article downloads are an appealing alternative. When measuring downloads, one can start tracking usage immediately after the publication of an article, instead of waiting months or even years for citations 18  Bornmann, L. (2013). What is societal impact of research and how can it be assessed? a literature survey. Journal of the American Society for Information Science and Technology, 64(2), 217–233. doi:10.1002/asi.22803; Tijssen, R. J. . (2001). Global and domestic utilization of industrial relevant science: patent citation analysis of science–technology interactions and knowledge flows. Research Policy, 30(1), 35–54. doi:10.1016/ S0048-7333(99)00080-3 19  Van Eck, N. J., Waltman, L., van Raan, A. F. J., Klautz, R. J. M., & Peul, W. C. (2013). Citation analysis may severely underestimate the impact of clinical research as compared to basic research. PloS One, 8(4), e62395. doi:10.1371/journal.pone.0062395 20  Kurtz, M.J., & Bollen, J. (2012). Usage Bibliometrics. Annual Review of Information Science and Technology Volume 44, Issue 1. doi: 10.1002/aris.2010.1440440108; Moed, H. F. (2005). Statistical relationships between downloads and citations at the level of individual documents within a single journal. Journal of the American Society for Information Science and Technology, 56(10), 1088–1097. doi:10.1002/asi.20200; Schloegl, C., & Gorraiz, J. (2010). Comparison of citation and usage indicators: the case of oncology journals. Scientometrics, 82(3), 567–580. doi:10.1007/s11192-010-0172-1; Schloegl, C., & Gorraiz, J. (2011). Global usage versus global citation metrics: The case of pharmacology journals. Journal of the American Society for Information Science and Technology, 62(1), 161–170. doi:10.1002/asi.21420; Wang, X., Wang, Z., & Xu, S. (2012). Tracing scientist’s research trends realtimely. Scientometrics, 95(2), 717–729. doi:10.1007/s11192-012-0884-5 21  http://usagereports.elsevier.com/asp/main.aspx; http://www.projectcounter.org/code_practice.html 2 research outputs & citation impact 26 Table 2.3 presents the average number of downloads region. For example, Southern Africa’s output in Agriculture that articles published between 2008 and 2012 by the is downloaded on average 17% more frequently than its respective regions have thus far received. The first column overall output, and Southern Africa’s output in the Health provides the overall average, and the next five columns Sciences is downloaded on average 9% less frequently provide the number of downloads per article for each of than its overall output. the five subject groupings. For example, East Africa has 4,231 articles on ScienceDirect, and those articles have Output in Agriculture is downloaded more frequently for all been downloaded on average 928 times, the most of any three Africa regions and South Africa, and it is downloaded region in this report. Moreover, across all the Africa regions’ at an even higher relative rate for the two comparator coun- outputs in different subject groupings, East Africa’s 1,376 tries (41% and 32% for Malaysia and Vietnam, respective- articles in the Physical Sciences & STEM have received the ly). Likewise, for all SSA regions, research in the Physical most average downloads per paper at 1,086. In general, Sciences & STEM is downloaded at a rate higher than the sub-Saharan research articles published between 2008 overall regional average. In contrast, for all the regions, out- and 2012 have been downloaded on average at least 650 put in the Health Sciences is downloaded on average less times. frequently than those respective regions’ overall output. To better benchmark and compare the relative number of downloads across subject groupings, Table 2.4 divides the downloads per article measure for each subject grouping by the overall downloads per article measure for a given Table 2.3 — Downloads per article by subject grouping for SSA regions and comparator countries, 2008-2012. Source: ScienceDirect. Social Physical Health Sciences & Life All Sciences Agriculture Sciences Humanities Sciences East Africa 928 1086 991 757 1022 807 Southern Africa 884 949 1033 801 813 820 West & Central Africa 676 781 752 511 671 860 South Africa 875 816 968 956 791 1103 Malaysia 898 843 1265 803 1252 1172 Vietnam 832 763 1100 838 984 1035 Table 2.4 — Downloads per article by subject grouping relative to regional averages for SSA regions and comparator countries, 2008-2012. Source: ScienceDirect. Social Physical Health Sciences & Life All Sciences Agriculture Sciences Humanities Sciences East Africa 1.00 1.17 1.07 0.82 1.10 0.87 Southern Africa 1.00 1.07 1.17 0.91 0.92 0.93 West & Central Africa 1.00 1.16 1.11 0.76 0.99 1.27 South Africa 1.00 0.93 1.11 1.09 0.90 1.26 Malaysia 1.00 0.94 1.41 0.89 1.39 1.30 Vietnam 1.00 0.92 1.32 1.01 1.18 1.24 2.5 novel measures of research impact 27 One particularly interesting audience of sub-Saharan research is international corporations. They provide both an early indicator of what types of research could attract further corporate R&D funding and a test for whether such research is more broadly applicable. Corporations, however, often have differing tastes in and uses for research than academics. As Table 2.5 exemplifies, downloads from cor- porate users comprises only a fraction of the total amount of usage data. For example, while East African articles from 2008-2012 were downloaded on average over 900 times, each paper was downloaded only 15.5 times on average from corporate users. More importantly, as Table 2.6 shows, the distribution of corporate interest in the different regions’ subject outputs is very different from that of the academic sector. In particular, while the output in the Health Sciences received fewer downloads on average relative to that from all sectors, such output received between 27% and 87% more downloads from the corporate sector. In contrast, research in the Physical Sciences & STEM received between 9% and 30% less downloads on average. Table 2.5 — Corporate downloads per article by subject grouping for SSA regions and comparator institutions, 2008-2012. Source: ScienceDirect. Social Physical Health Sciences & Life All Sciences Agriculture Sciences Humanities Sciences East Africa 15.5 10.9 10.5 24.4 4.5 26.0 Southern Africa 18.5 14.8 19.0 26.1 4.0 32.0 West & Central Africa 13.3 12.1 12.3 17.0 4.7 24.8 South Africa 22.3 18.7 21.7 41.6 5.5 48.6 Malaysia 16.6 14.7 27.2 32.5 6.1 37.2 Vietnam 16.2 11.1 24.1 40.6 6.9 43.4 Table 2.6 — Corporate downloads per article by subject grouping relative to regional averages for SSA regions and comparator institutions, 2008-2012. Source: ScienceDirect. Social Physical Health Sciences & Life All Sciences Agriculture Sciences Humanities Sciences East Africa 1.00 0.70 0.68 1.58 0.29 1.68 Southern Africa 1.00 0.80 1.02 1.41 0.21 1.73 West & Central Africa 1.00 0.91 0.92 1.27 0.35 1.86 South Africa 1.00 0.84 0.97 1.87 0.25 2.18 Malaysia 1.00 0.89 1.64 1.95 0.37 2.24 Vietnam 1.00 0.69 1.49 2.52 0.43 2.69 2 research outputs & citation impact 28 2.5.2. Patent Citations as an Alternative In terms of raw numbers, given the size and maturity of Measure of Impact South Africa’s research enterprise, it is unsurprising that Past studies suggest that academic researchers and indus- South Africa has attained more than twice as many patent try interact in a multitude of channels,22 and patent citations citations overall than any SSA region (804 compared to the is one of the more public lenses for understanding the link- next closest, West & Central Africa, at 351). More surpris- age between academic research and intellectual property. ing, however, is the disparity in the relative distribution of patent citations across subject groupings. Research in the Intellectual property (IP) describes intangible assets, such Physical Sciences & STEM by East Africa has only been as discoveries and inventions, for which exclusive rights cited 32 times compared to 90 and 87 times for research in may be claimed. Common types of IP include that which is the Health Sciences and Agriculture, respectively. South- codified in copyright, trademarks, patents, and designs. ern Africa and West & Central Africa show similar trends. Typically, a patent application must include one or more In contrast, for Malaysia, research in the Physical Sciences claims that define the invention, and these claims should be & STEM has garnered more patent citations (256) over the novel and non-obvious from the prior art (i.e., from existing, past decade than research in any other subject grouping. publicly-available documentary sources). As such, many patent applications cite journal articles which either provide When patent citations are normalized by the regions’ total information that supports or are related to the claims but publication outputs, the disparities between the regions that do not constitute prior art. get smaller. For example, the ratio of patent citations to all publications was 0.60% for East Africa and 0.50% for Drawing on indexed patent citation data from Lexis-Nexis Malaysia. However, even when patent citations are normal- TotalPatent and Scopus, this section examines the percent- ized by the regions’ publication outputs per subject, there is age of each Africa region’s output that is referenced by glob- still a noticeable focus amongst the SSA regions on Agricul- al patent applications from the World Intellectual Property ture and Health Sciences instead of the Physical Sciences Organization (WIPO). The numbers in Table 2.7 correspond & STEM. The ratio of patent citations to all publications for to the total number of citations in patent applications from West & Central Africa was 0.33% in the Physical Sciences & 2003-2012 to journal articles published by the respective STEM versus 0.82% in Agriculture and 0.61% in the Health regions (and when applicable, the respective subject group- Sciences. For Malaysia and Vietnam, the ratio of patent cita- ings) between 2003 and 2012. To normalize for differences tions to all publications in the Physical Sciences & STEM is in the underlying number of publications produced by each quite low (0.42% and 0.02%) relative to that of other subject region (and hence the number of publications that could be areas because of those comparator countries’ high levels cited in patents), Table 2.8 presents the number of patent ci- of output in the Physical Sciences & STEM, not necessarily tations divided by the total number of publications produced because the research conducted by the countries in those by a region in a subject area. subject areas is not particularly helpful to inventors. Table 2.7 — Patent citations to academic output in dif- Table 2.8 — Patent citations to academic output as percent- ferent subject groupings for SSA regions and comparator age of total publication output in different subject groupings institutions, 2003-2012. Source: LexisNexis TotalPatent for SSA regions and comparator institutions, 2003-2012. and Scopus. Source: LexisNexis TotalPatent and Scopus. es es nc nc M cie M cie re re in t in t TE l S TE l S up c up c gs gs tu tu ro b j e ro b j e es es & s ica & s ica ul ul c c S lt h S lt h u u en en ic ic S S ea ea y y gr gr S S ci ci ll ll Ph Ph A A A A H H G G East Africa 205 32 87 90 0.60% 0.38% 0.68% 0.64% Southern Africa 63 8 26 26 0.46% 0.19% 0.61% 0.48% West & Central Africa 351 60 167 151 0.56% 0.33% 0.82% 0.61% South Africa 804 315 338 211 0.90% 0.79% 1.43% 0.99% Malaysia 450 256 203 98 0.50% 0.42% 1.33% 0.77% Vietnam 88 17 39 26 0.65% 0.20% 1.21% 1.05% 22  D’Este, P., & Patel, P. (2007). University–industry linkages in the UK: What are the factors underlying the variety of interactions with industry? Research Policy, 36(9), 1295–1313. doi:10.1016/j.respol.2007.05.002; Schartinger, D., Rammer, C., Fischer, M. M., & Fröhlich, J. (2002). Knowledge interactions between universities and industry in Austria: sectoral patterns and determinants. Research Policy, 31, 303–328. doi: 10.1016/S0048-7333(01)00111-1 2.6 interpretation and discussion of chapter key findings 29 Interpretation and Discussion 2.6  of Chapter Key Findings The following section interprets the key findings on re- declined by 0.2% annually since 2003. In comparison, search output and impact in SSA and provides insights into the share of STEM research has declined 0.1% annually the expected drivers of the key findings. in South Africa and grew 2% annually in Malaysia and Vietnam. The key findings point in our view to three main interpreta- c. In 2012, the quality of STEM research in SSA, as meas- tions: ured by relative citation impact, was 0.68 (32 percent 1. Africa is rising in research. Both the quantity and qual- below the global average). This is below that of all disci- ity of research performance is improving. Capacity in plines in SSA (0.92) and the global average (1.00), and the African higher education and research sector has it has not significantly changed since 2003. In contrast, clearly progressed in the decade from 2003-2012. STEM research in Malaysia, Vietnam and South Africa The improvements are primarily driven by increased in 2012 was slightly above the world average and has research capacity in the Health Sciences. This interpre- improved significantly since 2003. tation is supported by the following key findings: a. Research production has increased by more than 100% Below is a short discussion of some of the key factors that in SSA since 2003. may drive the key findings of this chapter. Since the main b. SSA’s share of global research has increased from scope of this report is research output, the following is 0.44% in 2003 to 0.72% in 2012. based on factors observed in other regions and findings c. Between 7.5% and 16% of the different SSA’s total from other relevant, country-wise studies explaining re- publications were amongst the world’s top 10% most search output in SSA. Subsequent research should further highly cited articles, but only 5.9% -10% of those same examine these explanatory factors. region’s total output in the Physical Sciences & STEM ► Volume of Funding: Research outputs are greatly de- met that threshold. termined by international and national funding for R&D d. On average for the three SSA regions, research in the which finances necessary salaries, equipment and other Health Sciences constituted 45.2% of their total out- research costs. As an example, Box 1 summarizes how put. increased R&D expenditures in South Africa were an essential driver behind this country’s growth in research 2. A large gap in research capacity still exists between outputs. Box 3 presents the latest available R&D ex- SSA and the rest of the world. penditure data for SSA. a. SSA’s research output remains less than 1% of the ► Sectoral R&D Funding: Similarly, disciplinary alloca- world, while its share of the population is 12%. tion of R&D funding may heavily influence disciplinary b. Research output by comparator countries grew even research output. Box 2 presents anecdotal data for 3 faster than that of Sub-Sahara Africa. Malaysia, whose countries. Although data is scarce, the research funding article output in 2003 was similar to that of East Africa, provided by international development partners, such grew its output by 31% per year. Similarly, Vietnam, as the US Government and SIDA, to health research whose article output in 2003 was about two-thirds the in Africa is expected to be a major factor behind the level of Southern Africa, grew its output by 19% per improved research output in SSA. The increases in year. health R&D spending and output is encouraging and im- portant. First, due to the tremendous health challenges 3. SSA research capacity within Science, Technology, the continent faces, improved Africa-relevant health re- Engineering and Mathematics is underdeveloped and search and well-trained health workers will have a great lags significantly. This is evidenced by absolute and impact on health outcomes. As recent research shows, comparative shortcomings in the quantity and quality of although SSA assumes the heaviest burden of major STEM research. diseases such as HIV/AIDS, malaria, and tuberculosis, a. STEM research makes up only 29% of all research it is primarily Western countries that have the highest in SSA. In contrast, STEM research constitutes the research intensities in said subjects, with the exception largest share of each of the comparator countries’ total of South Africa.23 Second, the impressive improve- outputs (45% for South Africa and an average of 68% ment in SSA’s research capacity in the Health Sciences for Vietnam and Malaysia). demonstrates that persistent support and funding b. The share of STEM research in SSA has marginally from development partners and governments pays off. 2 research outputs & citation impact 30 On the other hand, Pouris and Ho (2013) 24 argue that Further, it is critical that institutional incentives are Africa’s heavy dependency on international scientific transferred within each institution to its faculty. collaboration may be stifling research individualism and ► Research infrastructure: Research in most STEM, Ag- affecting the continent’s research evolution and priori- ricultural, Health, and Life Sciences require substantial ties. Researchers argue that Africa's dependence on equipment as well as access to international databases international research funding implies that some of its and science literature. Research infrastructure is built research priorities are underfunded, STEM being a criti- and depleted over time. Lack of research infrastructure cal one. Governments and development partners could in Africa is a frequent explanation espoused by re- use lessons learnt from the rapid growth in health R&D searchers working in Africa. Unfortunately, no system- to boost growth in other sciences, specifically STEM. atic data is collected on this topic. ► Funding mechanisms: How research funding is allocated ► Number of researchers: The number of PhD hold- and the accountability for results equally matters for ers, faculty, and post-docs and PhD students is a key research output. Box 1 describes one example on how determinant of research output. Similar to research a change in research funding to South Africa universi- infrastructure, research human capital is built and de- ties fostered a marked increase in research output. The pleted over time. Box 4 provides a snapshot of available gold standard for research funding is open, transpar- information on the sectoral composition of the number ent, competitive, and peer-reviewed research funding. of researchers in SSA. Box 1. R&D Funding and Funding Mechanisms Matter: The Case of South Africa The following figures provide data on the growth of R&D in South Africa as a result of increased funding and better managed funding mechanisms. As shown in Figure A, starting from 2000, R&D funding in South Africa rose with GERD reaching $4.3 billion (in 2005 dollars) by 2008. This increase in funding volume has led to a sharp rise in research output in the past decade. The line in Figure A represents the introduction of a new funding formula for the provision of incentives by the Department of Education to universities. It is clear that this led to a sharp rise in the number of publications. Pouris (2011) 25 concludes that R&D funding and funding mechanisms matter for research output. 5,000 10,000 4,500 9,000 GROSS EXPENDITURES R&D (GERD) 4,000 8,000 NUMBER OF PUBLICATIONS 3,500 7,000 3,000 6,000 Number of 2,500 Publications 5,000 2,000 4,000 1,500 GERD 3,000 1,000 2,000 500 1,000 0 0 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Figure A — Trends in GERD and overall number of articles over time for South Africa, 1996-2008, with GERD in millions 2005 dollars - constant PPP. The line at 2001 notes when the new funding formula was introduced (see Pouris 2011) 26 Source: OECD Main Science and Technology Indicators and Scopus. 23  Huggett, S. (2009). Research supports UN millennium development goals. Research Trends (14). Retrieved from http://www.researchtrends.com/issue14-december-2009/behind-the-data/ 24  Pouris, A., & Ho, Y.-S. (2013). Research emphasis and collaboration in Africa. Scientometrics, 98(3), 2169–2184. doi:10.1007/s11192-013-1156-8 25  Pouris, A. (2012). Science in South Africa: The dawn of a renaissance? South African Journal of Science, 108(7/8). doi:10.4102/sajs.v108i7/8.1018 26  Ibid. 2.6 interpretation and discussion of chapter key findings 31 Growth Mirrors Allocation of Resources: Box 2.  Learning from Health in SSA GERD by field of science Over the years both Uganda and Mozambique have increased their funding in S&T but it remains lower than that of health. In Malaysia, the GERD in STEM is 28% while that in Health is 4%. In contrast, in 2010 the spending on STEM in Mozambique and Uganda was 15% and 12%, respectively of the total countries’ expenditures on research. In Africa, Health has seen great improvements given the national priorities and presents an example that can be followed in STEM. 30 GROSS DOMESTIC EXPENDITURE ON R&D (%) ■ 2008 ■ 2010 25 20 15 10 5 0 MOZAMBIQUE MOZAMBIQUE UGANDA UGANDA MALAYSIA MALAYSIA (STEM) (Health) (STEM) (Health) (STEM) (Health) GROSS DOMESTIC EXPENDITURE ON R&D BY FIELD (%) Figure B — Gross domestic expenditure on R&D by field (%), 2008 versus 2010. Source: UNESCO Institute of Statistics. Box 3. Low Country Investments in R&D Most West African countries are placing less than 0.25% of the GDP in R&D investments, while East Africa remains largely below 0.5% of the GDP. ■ 0.00% − 0.25% ■ 0.26% − 0.50% ■ 0.51% − 1.00% Figure C — Gross domestic expenditure on R&D (GERD) as a percentage of GDP, 2011 or latest available year for sub-Saharan Africa. Source: UNESCO Institute of Statistics. 2 research outputs & citation impact 32 Box 4. Researchers are concentrated in the field of medical and health sciences The number of researchers mirrors the flow of resources. As shown in the figure, the share of researchers in medical science and health far exceed the share of researchers in engineering and technology, e.g. In Burkina Faso, 46% of the researchers focus on Medical & Health Sciences while in Ethiopia and Kenya, it is 21% and 25% respectively. In contrast, the percentage of researchers that focus on Engineering & Technology in those countries are 16%, 6%, and 14%, respectively. 100% 90% 80% 70% 60% 9 15 46 % % 50% % 13 19 % % 0% 40% 14 % 25 19 30% 21 % 7% % % 20% 13 % 10% 0% O A A A * I L O A ^ * LI R W E E A PI N Y D S G A U A W LA G N A N FA TO IO C IQ M E B E H A S A N A TH K B G G A A M B E M U IN G E M S ZA A K ZI D R O A U M M B ■ Natural Sciences ■ Agricultural Sciences ■ Engineering & Technology ■ Social Sciences ■ Medical & Health Sciences ■ Humanities ■ Not Specified Figure D — Percentage of researchers in different fields for select SSA countries. Note: Data in this graph are based on FTE from 2010 counts unless otherwise noted (* = data from 2011, ^ data from 2012). Source: UNESCO Institute of Statistics. 33 CHAPTER 3 RESEARCH COLLABORATION This chapter focuses on how various types of collaboration affect citation impact. It examines the levels of extra-regional (i.e. international) and intra- regional collaboration, the corresponding impact of research resulting from such collaborations, and the top institutional collaborators with each region. 3 research collaboration 34 Key Findings 3.1  EXTRA-REGIONAL COLLABORATION INTER-REGIONAL COLLABORATION 42%-79% 0.9%-2.9% In 2012, the dominant share of SSA research is a Inter-African collaboration (without any South- result of international collaboration (42%, 68%, and African or international collaborator) comprises 79% of total research for West & Central, East, and 2% of all East African research, 0.9% of West & Southern Africa, respectively. Central Africa, and 2.9% of Southern Africa. CROSS-SECTOR COLLABORATION TOP ACADEMIC COLLABORATOR 1%-2.4% Harvard Academic-corporate collaborations comprise Harvard University ranked amongst the top 10 between 1% and 2.4% of SSA’s total research academic collaborators for the three SSA regions. output from 2003-2012. COLLABORATION CITATION IMPACT CROSS-SECTOR COLLABORATION 3.23-3.82 CITATION IMPACT Extra-regional (i.e., international) collaborations 2.81-6.09 In 2012, West & Central Africa’s academic- for SSA regions were between 3.23 and 3.82 corporate collaborations received more than six times as impactful as those respective regions’ times as many relative citations as the average institutional collaborations. article. Southern and East Africa’s academic- corporate collaborations also achieved high multipliers of 3.71 and 2.81, respectively. TOP CORPORATE COLLABORATORS GlaxoSmithKline, Novartis From 2003-2012, GlaxoSmithKline and Novartis were amongst the top 3 corporate collaborators for the three SSA regions. 3.2 international collaboration 35 International Collaboration 3.2  3.2.1. Methodology As technological advances facilitate long-distance commu- nication and low-cost travel, researchers are increasingly collaborating with international partners.27 Moreover, past research suggests that such collaborations are quite produc- tive. Internationally co-authored articles are associated with higher field-weighted citation impact.28 For this report, publications are classified as single-author (self-explanatory) or into one of three, mutually-exclusive types of geographic collaboration based on the nature of co-authorship: 29 extra- regional (i.e., international), intra-regional, and institutional. Table 3.1 — Typology of Different Types of Geographic Collaboration. Type of Collaboration Definition Extra-regional (i.e., international) Multi-authored research outputs with authors affiliated Collaboration with institutions in at least two different regions (e.g., East Africa and non-Africa, or West & Central Africa and Southern Africa) Intra-regional Collaboration Multi-authored research outputs with authors affiliated with more than one institution but both institutions within the same Africa region (e.g., University of Nairobi and National University of Rwanda, both in East Africa region) NB: for country comparators, regional collaboration is synonymous with national collaboration Institutional Collaboration Multi-authored research outputs with all authors affiliated with the same institution Single Author Single-authored research outputs 27  Pan, R. K., Kaski, K., & Fortunato, S. (2012). World citation and collaboration networks: uncovering the role of geography in science. Scientific reports, 2, 902. doi: 10.1038/srep00902 28  Science Europe & Elsevier. (2013). Comparative Benchmarking of European and US Research Collaboration and Researcher Mobility. Retrieved from http://www.scienceeurope.org/uploads/Public documents and speeches/SE and Elsevier Report Final.pdf; The Royal Society. (2011). Knowledge, networks and nations: Global scientific collaboration in the 21st century. (J. Wilson, L. Clarke, N. Day, T. Elliot, H. Harden-Davies, T. McBride, … R. Zaman, Eds.) (p. 113). London: The Royal Society. Retrieved from http://royalsociety.org/policy/projects/knowledge-networks-nations/report/ 29  Melin, G., & Persson, O. (1996). Studying research collaboration using co-authorships. Scientometrics, 36(3), 363–377. doi:10.1007/BF02129600; Glänzel, W., & Schubert, A. (2004). Analyzing Scientific Networks through Co-Authorship. In H. F. Moed (Ed.), Handbook of Quantitative Science and Technology Research (pp. 257–276). Amsterdam: Kluwer Academic Publishers. 3 research collaboration 36 3.2.2. International Collaboration with different types of geographic partners. Across all the “International” collaboration has been an especially popular regions, there is a common trend in the decline of single topic in past studies of Africa’s research performance. Since authorship and to a lesser extent, institutional collabora- many studies have analyzed this variable at the country in- tions. stead of the intra-regional level, 30 this report cannot provide a direct, apples-to-apples comparison of research measures. In addition, with the exception of West & Central Africa, Instead, this report’s definition of regional collaboration sub- international collaborations as a percentage of total output sumes both co-authored publications between two institu- rose for all the Africa regions. As Figure 3.2 and Figure 3.4 tions in the same country (e.g., University of Nairobi and Moi show, international collaboration consistently comprised University in Kenya) as well as co-authored publications be- over 60% of East Africa’s and Southern Africa’s total tween institutions in different countries but the same region research outputs, with no other type of collaboration con- (University of Swaziland and Catholic University of Angola, stituting more than 20% from 2003 to 2012. However, both in the SA region). Likewise, this report’s definition of for East Africa, intra-regional collaboration has increased international collaborations refers to collaborations between over time from 9.8% of its total output in 2003 to 13.6% researchers inside a particular Africa region and researchers in 2012. outside that region (i.e., extra-regional collaboration). Thus, the terms international and extra-regional collaboration are As Figure 3.3 shows, international collaborations as a used interchangeably in this chapter. percentage of West & Central Africa’s total output actually fell between 2003 to 2010 from 44.1% to 35.1% before Figure 3.1 presents the amount of international collabora- rebounding to 42.2% in 2012. Nevertheless, during those tions as the relative percentage of a region’s total output. The years, intra-regional collaboration rose from 14.3% in international collaboration rate is quite high especially for 2003 to 24.7% in 2012. Southern Africa and East Africa. Between 2003 and 2012, international collaborations as a percentage of Southern Afri- Malaysia provides interesting contrasts to the Africa ca’s total article output increased from 60.7% to 79.1%. For regions. International collaborations as a percentage of Eastern Africa, international collaborations consistently com- Malaysia’s total output has actually fallen over time, and prised between 65% and 71% of the region’s total output. institutional collaborations now constitute the largest share of all Malaysian research. In contrast, Vietnam’s The figures on the next few pages provide another perspec- heavy emphasis on international collaboration mirrors that tive on the degree to which the Africa regions collaborate of East Africa and Southern Africa. 90% 80% SOUTHERN AFRICA EAST AFRICA 70% VIETNAM PERCENTAGE OF TOTAL OUTPUT 60% 50% SOUTH AFRICA WEST & CENTRAL AFRICA 40% MALAYSIA 30% 20% 10% Figure 3.1 — Level of international 0% collaboration for SSA regions and 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 comparator countries, 2003-2012. YEAR OF PUBLICATION Source: Scopus. 30  Mêgnigbêto, E. (2012). Scientific publishing in Benin as seen from Scopus. Scientometrics, 94(3), 911–928. doi:10.1007/s11192-012-0843-1; Mêgnigbêto, E. (2013). Scientific publishing in West Africa: comparing Benin with Ghana and Senegal. Scientometrics, 95(3), 1113–1139. doi:10.1007/s11192-012-0948-6 3.2 international collaboration 37 80% EAST AFRICA INTERNATIONAL 70% 60% PERCENTAGE OF TOTAL OUTPUT 50% 40% 30% INTRA-REGIONAL INSTITUTIONAL 20% SINGLE AUTHOR 10% Figure 3.2 — Different types of 0% collaborations as percentage of 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 East Africa’s total output, YEAR OF PUBLICATION 2003-2012. Source: Scopus. 50% WEST & CENTRAL AFRICA 45% INTERNATIONAL 40% 35% PERCENTAGE OF TOTAL OUTPUT 30% 25% INTRA-REGIONAL 20% INSTITUTIONAL 15% 10% SINGLE AUTHOR 5% Figure 3.3 — Different types of 0% collaborations as percentage of 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 West & Central Africa’s total output, YEAR OF PUBLICATION 2003-2012. Source: Scopus. 90% SOUTHERN AFRICA 80% INTERNATIONAL 70% PERCENTAGE OF TOTAL OUTPUT 60% 50% 40% 30% INSTITUTIONAL SINGLE AUTHOR 20% INTRA-REGIONAL 10% Figure 3.4 — Different types of 0% collaborations as percentage of 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Southern Africa’s total output, YEAR OF PUBLICATION 2003-2012. Source: Scopus. 3 research collaboration 38 50% SOUTH AFRICA 45% INTERNATIONAL 40% 35% PERCENTAGE OF TOTAL OUTPUT 30% 25% INSTITUTIONAL 20% SINGLE AUTHOR 15% INTRA-REGIONAL 10% 5% Figure 3.5 — Different types of 0% collaborations as percentage of 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 South Africa’s total output, YEAR OF PUBLICATION 2003-2012. Source: Scopus. 50% MALAYSIA 45% INSTITUTIONAL 40% 35% PERCENTAGE OF TOTAL OUTPUT INTERNATIONAL 30% 25% 20% INTRA-REGIONAL 15% 10% SINGLE AUTHOR 5% Figure 3.6 — Different types of 0% collaborations as percentage of 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Malaysia’s total output, YEAR OF PUBLICATION 2003-2012. Source: Scopus. 90% VIETNAM 80% 70% INTERNATIONAL PERCENTAGE OF TOTAL OUTPUT 60% 50% 40% 30% INSTITUTIONAL INTRA-REGIONAL 20% SINGLE AUTHOR 10% Figure 3.7 — Different types of 0% collaborations as percentage of 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Vietnam’s total output, YEAR OF PUBLICATION 2003-2012. Source: Scopus. 3.2 international collaboration 39 3.2.3. Inter-Regional Collaboration provides a measure of un-brokered collaborations between In addition to “international” collaboration, researchers and co-authors at institutions in two or more different Africa policymakers are particularly interested in better under- regions.35 standing the degree to which the different Africa regions collaborate with one another. Are there indications of the Relative to East Africa’s overall rates of international col- rise of a sub-Saharan research network independent from laboration (which comprise over 60% of East Africa’s total ties to European and American foci? Past studies have output), its level of inter-regional collaboration with other found low rates of both intra-regional and inter-regional SSA regions is low, at about 2%. Yet, East Africa’s collabo- collaboration. 31 For example, Boshoff’s 2009 study of the rations with South Africa have increased considerably over Southern African Development Community (SADC) found time, from 3.9% in 2003 to 7.9% in 2012. This growth that only 5% of all SADC papers from 2005-2008 were co- has been driven mostly through collaborations involving authored between a researcher in the SADC and another partners at institutions in developed countries. The annual African researcher. 32 From their network analysis of Africa’s growth rate of East Africa-South Africa collaborations research output that similarly demarcated the continent without an OECD partner has only been 3.3%, compared to into three large regions (Southern-Eastern, West, and 8.2% with an OECD partner. Northern), Toivanen and Ponomariov similarly found low lev- els of inter-regional collaboration. They argue, “So great is the heterogeneity between these three regions and so weak are the inter-regional linkages, that it raises the broader question of optimal organization of African research. Considering that African research effort and capacity are increasing rapidly, Africa as a whole stands the risk to miss synergies inherent in well integrated innovation systems and which are foundational for knowledge economy.” 33 To calculate the number of collaborations between East Africa and West & Central Africa, for example, this report counted all publications in which at least one author holds an affiliation to an East African institution and another author holds an affiliation to a West & Central African insti- tution. Thus, the counts of inter-regional collaborations are subsets of the counts of international (i.e., extra-regional) collaborations from the previous section. Figure 3.8 dis- plays the trends of inter-regional collaboration for East Af- rica vis-à-vis the other regions. The first and top three trend lines correspond to all collaborations between East Africa and West & Central Africa, Southern Africa, and South Africa respectively. The bottom three trend lines corre- spond specifically to collaborations in which no co-authors were affiliated with institutions in OECD countries.34 This 31  Adams, J., Gurney, K., Hook, D., & Leydesdorff, L. (2013). International collaboration clusters in Africa. Scientometrics, 98(1), 547–556. doi:10.1007/ s11192-013-1060-2; Onyancha, O. B., & Maluleka, J. R. (2011). Knowledge production through collaborative research in sub-Saharan Africa: how much do countries contribute to each other’s knowledge output and citation impact? Scientometrics, 87(2), 315–336. doi:10.1007/s11192-010- 0330-5 32  Boshoff, N. (2009). South–South research collaboration of countries in the Southern African Development Community (SADC). Scientometrics, 84(2), 481–503. doi:10.1007/s11192-009-0120-0 33  Toivanen, H., & Ponomariov, B. (2011). African regional innovation systems: bibliometric analysis of research collaboration patterns 2005–2009. Scientometrics, 88(2), 471–493. doi:10.1007/s11192-011-0390-1 34  OECD member countries include: Australia, Austria, Belgium, Canada, Chile, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Israel, Italy, Japan, Korea, Luxembourg, Mexico, Netherlands, New Zealand, Norway, Poland, Portugal, Slovak Republic, Slovenia, Spain, Sweden, Switzerland, Turkey, United Kingdom, and the United States. 35  These counts may still reflect collaborations amongst two Africa regions and non-OECD countries, so they are not necessarily pure, un-brokered research collaborations. 3 research collaboration 40 8% with SOUTH AFRICA EAST AFRICA 7% PERCENTAGE OF TOTAL EAST AFRICA OUTPUT 6% with WEST & CENTRAL AFRICA 5% with SOUTHERN AFRICA 4% with SOUTH AFRICA with WEST & CENTRAL AFRICA 3% with SOUTHERN AFRICA Figure 3.8 — Different types of 2% inter-regional collaborations as 1% percentage of East Africa’s total output, 2003-2012. Dashed lines 0% refer to rates of inter-regional 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 collaboration excluding additional YEAR OF PUBLICATION OECD partners. Source: Scopus. 5% with SOUTH AFRICA PERCENTAGE OF TOTAL WEST & CENTRAL AFRICA OUTPUT WEST & CENTRAL AFRICA 4% with EAST AFRICA 3% with SOUTH AFRICA with SOUTHERN AFRICA with EAST AFRICA 2% with SOUTHERN AFRICA Figure 3.9 — Different types of inter- 1% regional collaborations as percentage of West & Central Africa’s total output, 2003-2012. Dashed lines refer to 0% rates of inter-regional collaboration 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 excluding additional OECD partners. YEAR OF PUBLICATION Source: Scopus. 25% SOUTHERN AFRICA PERCENTAGE OF TOTAL SOUTHERN AFRICA OUTPUT with SOUTH AFRICA 20% with EAST AFRICA 15% with SOUTH AFRICA with WEST & CENTRAL AFRICA with EAST AFRICA 10% with WEST & CENTRAL AFRICA Figure 3.10 — Different types of inter- 5% regional collaborations as percentage of Southern Africa’s total output, 2003-2012. Dashed lines refer to 0% rates of inter-regional collaboration 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 excluding additional OECD partners. YEAR OF PUBLICATION Source: Scopus. Source: Scopus. 3.3 citation impact of collaboration 41 Citation Impact of Collaboration 3.3  Previous studies suggest there exists a strong positive How, if at all, have the citation impacts of the regions’ correlation between international collaboration and citation international collaborations changed over time? As Figure impact.36 Table 3.2 shows adjusted FWCI with different 3.11 shows, the FWCI of Southern Africa’s international types of collaboration normalized against the FWCI of collaborations increased from 1.16 in 2003 to 1.66 in institutional collaborations. For all SSA regions, the FWCI 2012, reflecting a 4% CAGR. Since international collabora- associated with international collaborations is at least 3.23 tions comprised no less than 60% of Southern Africa’s total times higher than that attained by institutional collabora- output over this period, the rise in the impact of its overall tions. Moreover, while comparator countries Malaysia and research output can be primarily traced to the increases in Vietnam also have multipliers above 1, they are much lower the impact of its international collaborations.39 than those values for the SSA regions. Paralleling the growth in the impacts of the Africa regions’ Corroborating the results of past studies,37 the citation collaborative outputs, Malaysia also saw the impact of its impacts of intra-regional collaborations were higher than international collaborations grow from 0.89 (below world that of single-institution collaborations in East and South- average) to 1.14 (above the world average). However, ern Africa. However, in contrast to past research, which Vietnam saw little change over time in the FWCI of its suggest that single-authored publications achieve lower international collaborations. levels of impact than all types of collaboration, 38 in all three SSA regions, single-authored publications were actually more impactful than collaborations between researchers at the same institution. Table 3.2 — Adjusted FWCI associated with different types of collaboration (e.g., FWCI for single-authored, intra-regional, and international collaboration normalized against FWCI of institutional collaboration) for SSA regions and comparator coun- tries, 2012. Source: Scopus. Single Author Institutional Intra-regional International East Africa 1.08 1.00 1.03 3.23 Southern Africa 1.07 1.00 1.24 3.82 West & Central Africa 1.13 1.00 0.92 3.64 South Africa 0.95 1.00 1.12 2.67 Malaysia 0.62 1.00 0.93 1.34 Vietnam 1.18 1.00 1.02 1.92 36  Adams, J. (2013). Collaborations: The fourth age of research. Nature, 497(7451), 557–60. doi:10.1038/497557a; Franceschet, M., & Costantini, A. (2010). The effect of scholar collaboration on impact and quality of academic papers. Journal of Informetrics, 4(4), 540–553. doi:10.1016/j. joi.2010.06.003; Guerrero Bote, V. P., Olmeda-Gómez, C., & de Moya-Anegón, F. (2013). Quantifying the benefits of international scientific collaboration. Journal of the American Society for Information Science and Technology, 64(2), 392–404. doi:10.1002/asi.22754; The Royal Society, & Society, T. R. (2011). Knowledge, networks and nations: Global scientific collaboration in the 21st century (p. 113). Retrieved from http://royalsociety.org/uploadedFiles/Royal_Society_Content/policy/publications/2011/4294976134.pdf; Sooryamoorthy, R. (2009). Do types of collaboration change citation? Collaboration and citation patterns of South African science publications. Scientometrics, 81(1), 177–193. doi:10.1007/s11192-009-2126-z 37  Apolloni, A., Rouquier, J.-B., & Jensen, P. (2013). Collaboration range: Effects of geographical proximity on article impact. The European Physical Journal Special Topics, 222(6), 1467–1478. doi:10.1140/epjst/e2013-01937-5 38  Gazni, A., & Didegah, F. (2011). Investigating different types of research collaboration and citation impact: a case study of Harvard University’s publications. Scientometrics, 87(2), 251–265. doi:10.1007/s11192-011-0343-8; Hsu, J., & Huang, D. (2010). Correlation between impact and collaboration. Scientometrics, 86(2), 317–324. doi:10.1007/s11192-010-0265-x 39  To confirm this hypothesis, this report further analyzed the impact trends associated with Southern Africa’s single author, institutional, and intra- regional collaborations. In 2012, such collaborations (or lack thereof) was relatively low and stable with FWCI from 0.45 to 0.54 and CAGRs from -1.0% to 0.9%. 3 research collaboration 42 SOUTH AFRICA 1.80 SOUTHERN AFRICA 1.60 EAST AFRICA FIELD-WEIGHTED CITATION IMPACT 1.40 VIETNAM WEST & CENTRAL AFRICA 1.20 MALAYSIA 1.00 Figure 3.11 — FWCI of international 0.80 collaboration for SSA regions and 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 comparator countries, 2003-2012. YEAR OF PUBLICATION Source: Scopus. Cross-sector collaboration 3.4  Cross-sector collaboration potentially provides another the highest level of all types of cross-sector collabora- lens into understanding the improvement of Africa regions’ tions. For example, 17.4% of Southern Africa’s total output research citation impact over the past decade. Much recent over the past decade belonged to this category, a growth research focuses on the benefits of and complementarity from 13.7% in 2003 to 19.6% in 2012 or a 4.1% annual between academic and commercially oriented research.40 increase. Academic-government collaborations constituted Measuring co-authored publications across sectors is one a similarly large minority of Vietnam’s total output (14%) proxy for cross-sector collaboration. For this report, the af- over the same period, but they made up only a small portion filiation of every (co-)author in an article has been assigned (3.3%) of Malaysia’s total output. to one of four sectors: academic, corporate, government, or medical.41 When an article is co-authored by authors with Academic-corporate collaborations account for only a small affiliations in different sectors, that article is added to the percentage of each region’s total output, but it has grown count of cross-sector collaboration between those sectors. over time. For instance, Southern Africa published only 16 This section investigates the rates at which authors col- articles co-authored between an academic and a corporate laborate across sectors within the different regions.42 institution in 2003 but 74 in 2012. Such collaborations are particularly interesting for two reasons. First, they are a Table 3.3 presents the amount of cross-sector collabo- signal of and proxy for deeper connections between the two ration as the relative percentage of each region’s total sectors, which suggests a greater transfer of knowledge. output between 2003 and 2012. Across all the SSA Second, the academic-corporate collaborations act as a regions, academic-government collaborations comprised harbinger of future, alternative funding channels. 40  Larsen, Maria Theresa (2011). The implications of academic enterprise for public science: An overview of the empirical evidence. Research Policy 40(1): pp. 6-19. doi: 10.1016/j.respol.2010.09.013 41  The overwhelming majority of corporate research is conducted by mostly large, multinational corporations with significant R&D workforces, such as Microsoft, Merck, Boeing, General Electric, and so forth. We acknowledge that our current measures of research output and performance do not provide a good proxy for the level of collaboration and exchange between smaller African companies and their university counterparts. Please see Appendix B: Glossary on Sectors for more details on how institutions are specifically assigned to these sectors. 42  These cross-sector counts do not distinguish between whether both institutions are located in a particular AR or, if only one of the co-authors is from the AR, to which sector that author’s institution belongs. In practice, the great majority of AR institutions that engage in cross-sector collaborations are academic institutions. 3.4 cross-sector collaboration 43 For each region, Table 3.4 displays the citation impact as- The impact associated with different types of cross- sociated with different types of cross-sector collaborations sector collaborations has grown significantly over the past relative to the impact of all articles produced by that region decade. Figure 3.12 is most relevant for understanding in 2012. For example, the 112 articles from West & Central the influence of academic-government collaborations on Africa in 2012 that were academic-corporate collabora- the citation impact of the Africa regions’ total output since tions received more than six times as many citations on av- academic-government collaborations comprise such a erage as articles from the region overall. More importantly, sizeable minority of those regions’ output. For example, be- academic-government collaborations received two or more tween 2003 and 2012, the impact of such collaborations times as many relative citations on average as articles from for Southern Africa grew at nearly 6% per year, from 1.66 the regions overall. in 2003 to 2.80 in 2012. Table 3.3 — Cross-sector collaboration as percentage of total publications for SSA regions and comparator institutions, 2003-2012. Source: Scopus Academic  Corporate Academic  Government Academic  Medical East Africa 2.4% 17.2% 6.0% Southern Africa 2.4% 17.4% 7.5% West & Central Africa 1.0% 10.5% 4.2% South Africa 2.8% 12.6% 3.0% Malaysia 1.3% 3.3% 1.7% Vietnam 2.1% 14.0% 3.8% Table 3.4 — Adjusted FWCI of different types of cross-sector collaboration (e.g., FWCI for cross-sector collaboration normal- ized against FWCI of all articles) for SSA regions and comparator countries, 2012. Source: Scopus.43 Overall Academic  Corporate Academic  Government Academic  Medical East Africa 1.00 2.81 2.00 2.69 Southern Africa 1.00 3.71 2.01 2.43 West & Central Africa 1.00 6.09 2.67 2.48 South Africa 1.00 2.88 2.07 3.71 Malaysia 1.00 1.90 1.64 2.03 Vietnam 1.00 3.32 1.95 2.76 2.90 SOUTHERN AFRICA 2.70 SOUTH AFRICA 2.50 FIELD-WEIGHTED CITATION IMPACT EAST AFRICA 2.30 2.10 VIETNAM 1.90 1.70 WEST & CENTRAL AFRICA MALAYSIA 1.50 1.30 Figure 3.12 — FWCI of academic- 1.10 government collaboration for SSA 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 regions and comparator countries, YEAR OF PUBLICATION 2003-2012. Source: Scopus. 43  NB: There were at least 50 articles published in 2012 for each category of cross-sector collaboration in every country. This ensures that there were enough observations to draw meaningful conclusions. 3 research collaboration 44 Top collaborating institutions 3.5  To further investigate the trends in international and cross- institutions from that respective region from 2008-2012. sector collaboration, this section analyzes those institu- Certain institutions appear on the list of top collaborators tions with which the different Africa regions collaborate the for multiple regions and are represented by concentric most and the frequency and impact of those collaborations. circles of the respective regional colors. Notably, Harvard Jones et al.’s research (2008) suggests that the returns to University and the London School of Hygiene and Tropical collaboration in terms of citation impact depend not just on Medicine rank amongst the top ten academic collaborators whether one collaborates but with whom one collaborates. for all three Africa regions, while the University of Copen- The returns are predictably stratified by the rank or pres- hagen, the University of Liverpool, and the University of tige of the collaborating institution. 44 Oxford are amongst the top ten academic collaborators for two of the three. Past studies of Africa’s research output from the 1990s suggest that the institutions with whom African institu- Further corroborating past studies, the top collaborating tions collaborate the most are from the US and Europe. 45 institutions for South Africa tend to be based in the UK. Giv- Moreover, the exact list of top countries with which African en the French colonial history associated with many West institutions collaborate depends on those institutions’ colo- & Central African countries, it is unsurprising that four of nial ties – for example, South Africa and other former British its top ten overall collaborators are French organizations colonies tended to collaborate more with the UK, 46 while (CIRAD, Institut Pasteur, and IRD). Francophone countries collaborated more with France, and so forth. 44 Jones, B. F., Wuchty, S., & Uzzi, B. (2008). Multi-University Research  Teams: Shifting Impact, Geography, and Stratification in Science. Sci- Figure 3.13 presents a global overview of those institutions ence, 322(5905), 1259–1262. doi:DOI 10.1126/science.1158357 with whom the different regions collaborate with the most, 45 Narváez-Berthelemot, N., Russell, J. M., Arvanitis, R., Waast, R., &  with Figure 3.14 and Figure 3.15 providing insets of the Gaillard, J. (2002). Science in Africa: An overview of mainstream United States and Europe. The colors of the circles corre- scientific output. Scientometrics. doi:10.1023/A:1016033528117 spond to the specific SSA regions with whom those institu- 46 Sooryamoorthy, R. (2009). Collaboration and publication: How  tions collaborate highly, and the size denotes the number collaborative are scientists in South Africa? Scientometrics, 80(2), of publications that that institution has co-authored with 419–439. doi:10.1007/s11192-008-2074-z WORLD MAP OF SSA REGIONS’ TOP COLLABORATING INSTITUTIONS Figure 3.13 — World map depicting top institu- Region Co-publications tions collaborating with different SSA regions ● West & Central Africa and South Africa, 2003-2012. Source: Scopus. ● East Africa Plotted using R/ggplot & rgdal, and free vector ● Southern Africa and raster map data @ naturalearthdata.com. ● South Africa 250 500 750 1000 1250 3.5 top collaborating institutions 45 Figure 3.14 — Inset of world map, focusing on the United States, depicting top institutions collaborating with different SSA regions and South Africa, 2003-2012. Source: Scopus. Plotted using R/ggplot & rgdal, and free vector and raster map data @ naturalearthdata.com. Figure 3.15 — Inset of world map, focusing Region Co-publications on Europe, depicting top institutions collabo- ● West & Central Africa rating with different SSA regions and South ● East Africa Africa, 2003-2012. Source: Scopus. Plotted ● Southern Africa using R/ggplot & rgdal, and free vector and ● South Africa 250 500 750 1000 1250 raster map data @ naturalearthdata.com. 3 research collaboration 46 Past research on Africa’s international research collabora- with the previous figures, bubble size denotes the number tions has been especially sensitive about the asymmetry of of collaborations between that institution and a particular North-South partnerships. Collaborations between African Africa region. The FWCI of co-authored articles between institutions and those in more developed countries tend to the regions and the great majority of their top collaborators rely on the funding of and hence be driven by the needs and are above the relative baselines (y=1, x=1), indicating that research interests of the latter. The distribution of work as those collaborations were beneficial to both parties. well as credit tends to be unequal. Moreover, rather than a mutually beneficial partnership, scholars suggest that col- However, the relative impact of these top collaborations laboration partners in Africa receive a boost in their citation varies by region. In particular, all of South Africa’s and most impact, while those in more developed countries experience of Southern Africa’s top collaborating institutions can be a relative decline.47 found in the top-right quadrant. About half of East Africa’s top collaborating institutions are above the relative base- For the top 10 collaborators (from any sector) for each re- line, while most of West & Central Africa’s top collaborating gion, Figure 3.16 graphs the relative FWCI associated with institutions are located below of the relative baseline. Thus, articles co-authored between that institution and an Africa in contrast to previous research, these results show that region, relative to the FWCI of all internationally co-au- institutions in more developed countries do benefit from thored articles from those institutions (on the vertical axis) collaborations with institutions in Africa regions, though or from that particular region (on the horizontal axis). As this varies across the different regions. TOP 10 COLLABORATORS IN EACH REGION 2.5 FWCI OF CO-AUTHORED ARTICLES RELATIVE TO FWCI OF ALL OF INSTITUTION'S INTERNATIONAL CO-AUTHORED ARTICLES +− ++ 2.0 Collaborations below Collaborations par for region, beneficial to 1.5 above par for both parties institution 1.0 0.5 −− −+ Collaborations below Collaborations below par for both parties par for institution, above par for region 0 0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 FWCI OF CO-AUTHORED ARTICLES RELATIVE TO FWCI OF ALL OF REGION'S INTERNATIONALLY CO-AUTHORED ARTICLES ● West & Central Africa ● East Africa ● Southern Africa ● South Africa Figure 3.16 — Top 10 collaborators with each SSA region and South Africa in terms of total co-authored publications, 2003- 2012. Size of circle indicates the volume of co-authored publications between the collaborating institutions. Source: Scopus. 47  Boshoff, N. (2009). Neo-colonialism and research collaboration in Central Africa. Scientometrics, 81(2), 413–434. doi:10.1007/s11192-008-2211- 8; Gaillard, J. F. (1994). North-South research partnership: Is collaboration possible between unequal partners? Knowledge and Policy, 7(2), 31–63. doi:10.1007/BF02692761; Jentsch, B., & Pilley, C. (2003). Research relationships between the South and the North: Cinderella and the ugly sisters? Social Science & Medicine, 57(10), 1957–1967. doi:10.1016/S0277-9536(03)00060-1 3.5 top collaborating institutions 47 3.5.1. Top Collaborators in Southern Africa Figure 3.17 provides a more granular view of Southern Africa’s top ten collaborators. Given the skew of the region’s output and impact in the Health Sciences, nearly all of its top collaborators are institutions that specialize in medicine and health-related research. This skew towards high levels of (and impacts associated with) collaboration in the Health Sciences is also found in the lists of top institutions for the other Africa regions, including South Africa. Further reinforcing the dominance of Health Sciences in the regions’ collaborations, past research has found that in terms of absolute publication output, public-private research collaborations are most common in the fields of medicine.48 Unsurprisingly, the two companies that ap- peared in all three SSA regions’ lists of top 10 corporate collaborations were GlaxoSmithKline and Novartis. Moreo- ver, other pharmaceutical companies comprise the majority of each region’s top corporate collaborator lists. TOP 10 COLLABORATORS IN SOUTHERN AFRICA +− ++ University of the FWCI OF CO-AUTHORED ARTICLES RELATIVE TO FWCI OF ALL OF Collaborations below Collaborations Witwatersrand 2.0 par for region, beneficial to INSTITUTION'S INTERNATIONAL CO-AUTHORED ARTICLES above par for both parties Johns Hopkins University institution Centers for Disease Control and Prevention University of Cape Town 1.5 University of Liverpool Harvard University University of KwaZulu-Natal London School of Hygiene 1.0 and Tropical Medicine Institut Pasteur World Health Organization 0.5 −− −+ Collaborations below Collaborations below par for institution, par for both parties above par for region 0 0 0.5 1.0 1.5 2.0 2.5 3.0 FWCI OF CO-AUTHORED ARTICLES RELATIVE TO FWCI OF ALL OF REGION'S INTERNATIONALLY CO-AUTHORED ARTICLES Figure 3.17 — Top 10 Collaborators with Southern Africa in terms of total co-authored publications, 2003-2012. Source: Scopus. 48  Abramo, G., D’Angelo, C. A., Costa, F. Di, & Solazzi, M. (2009). University–industry collaboration in Italy: A bibliometric examination. Technovation, 29(6-7), 498–507. doi:10.1016/j.technovation.2008.11.003 3 research collaboration 48 Interpretation of Key Findings 3.6  on Research Collaboration The following section summarizes the findings on research not rely much on African-generated knowledge and collaborations in sub-Saharan Africa and suggests interpre- research for their competitiveness. tations and background factors for the key findings. 4. SSA’s research capacity appears fragmented across 1. A very large share of SSA research is a result of inter- regions, with each the regions collaborating very little national collaboration. In 2012, 79%, 70% and 45% of with one another. Inter-African collaboration (without all research by Southern Africa, East Africa, and West & any South-African or international collaborator) com- Central Africa, respectively, were through international prises 2% of all of East Africa’s research, 0.9% of West collaborations. In contrast, 68%, 45%, and 32% of & Central Africa’s, and 2.9% of Southern Africa’s. Vietnam, South Africa, and Malaysia’s research output, respectively, were international collaborations. 5. West & Central Africa displays somewhat different pat- terns of collaboration than East and Southern Africa. 2. International collaboration is highly instrumental in International collaboration comprises a smaller share raising the citation impact of SSA publications. Such (42%) of West & Central Africa’s total research output, collaborations were between 3.23 and 3.82 times as and there is less research collaboration between aca- impactful as those respective regions’ institutional demia and other partners (corporate, government, and collaborations. In contrast, the multiplying factors for medical). South Africa, Malaysia, and Vietnam were 2.7, 1.3, and 1.9, respectively. Although international collaboration is the major driver of African research, raising the citation impact of research in Africa, Africa today still lacks sufficient capacity and critical mass to produce international quality research on its own, in particular within STEM. While the success of SSA’s diaspora demonstrates that talent abounds on the continent, that scientific talent may be insufficiently nurtured due to shortcomings in the quality of science and math basic education, the availability of high quality post-graduate training, research infrastructure, faculty time, research funding, and incentives to pursue an academic career. In most public research institutes, the governments only cover operational costs and salaries, and the research itself is financed through collaborations. In ad- dition, research funding often comes through international collaboration (often salaries are covered, but not opera- tional, travel, and equipment costs). As a result, research would, independent of student training capacity, tend to be associated with international collaboration. 3. There appears to be little knowledge transfer and col- laboration between African academics and the corpo- rate sector, as measured by corporate downloads of and patent citations to African academic research, espe- cially for STEM disciplines. To the extent such knowl- edge transfer occurs, it occurs within Health Sciences and through collaborations with global pharmaceutical companies. Such trends suggest that corporations do 49 CHAPTER 4 RESEARCHER MOBILITY This chapter examines the geographic mobility of different types of African researchers as they move to and from the larger African diaspora. 4 researcher mobility 50 Key Findings 4.1  "Africa has reached a stage of development where HIGHLY MOBILE RESEARCHER BASE 85.3% it has become a destination for doing world-class science - a place that has individuals, facilities and institutions that attract scientists from around the world to work on the continent. … As an example, the SKA project has resulted in a net brain gain to 85.3% of Southern Africa’s researcher base has the region, with leading astronomers, ranging from published an article while outside of Southern Africa. post-doc[toral]s to research professors, choosing to work in Africa.“ Professor Justin Jonas  A ssociate Director of South Africa’s Square Kilometre Array VISITING SCHOLARS 57%-65% (SKA) and Professor of Physics and Electronics at Rhodes University Source: http://www.bbc.com/news/science-environment-21851042 Transitory researchers – those who spend less than 2 years in or outside the region – comprise 57% and 65% of East Africa’s and Southern Africa’s total researcher base. HIGH IMPACT RESEARCHERS Returnee Inflow Returning diaspora contribute significantly to raising the citation impact of SSA research, specifically in East and Southern Africa. While they make up a relatively small share of the region’s total researcher base (3.6% and 2.1%, respectively), the relative citation impact of those returnees’ publication is quite high compared to that of other SSA researchers. 4.2 researcher mobility model 51 Researcher Mobility Model 4.2  Brain circulation has been a key area of interest for Africa. which draws on the methodology detailed in Moed et al. Although the concepts of brain drain and brain gain have (2013).53 It shares many characteristics with the approach traditionally been discussed in terms of losers and winners, used in previous studies conducted to analyze the mobil- new research and theoretical frameworks suggest that ity of UK researchers 54 and compare European and US talent mobility results in win-win situations where all parties researchers.55 accrue benefits both in the short-term and the long-term. 49 Measuring international researcher mobility In the context of academic mobility, although a country or This report’s approach uses Scopus author profile data to institution may lose some of its best scientific talent to else- derive a history of active researchers 56 affiliated with insti- where (especially for graduate training), many researchers tutions in the respective Africa regions, as recorded in their come back with stronger skills, strengthening collaboration published articles. These are then used to assign research- ties between the countries and institutions and improving ers to mobility classes defined by the type and duration of the quality of their research. 50 Moreover, those that remain observed moves. abroad still maintain strong ties to their place of origin, ena- bling the flow of ideas and providing trainee opportunities for other researchers from that country.51 In the context of especially medical training in Africa, researchers emphasize the benefits of these positive network externalities over potential declines in the stock of local human capital.52 The availability of comprehensive publication databases containing articles with complete author affiliation data, such as Scopus, has enabled the development of a system- atic approach to researcher mobility analysis through the use of authors’ addresses listed in their published articles as a proxy for their location. The following section describes the individual components of that brain circulation model, 49  Teferra, D. (2005). Brain Circulation: Unparalleled Opportunities, Underlying Challenges, and Outmoded Presumptions. Journal of Studies in International Education, 9(3), 229–250. doi:10.1177/1028315305277619; Tung, R. L. (2008). Brain circulation, diaspora, and international competitiveness. European Management Journal, 26(5), 298–304. doi:10.1016/j.emj.2008.03.005; Ciumasu, I. M. (2010). Turning brain drain into brain networking. Science and Public Policy, 37(2), 135–146. 50  Easterly, W., & Nyarko, Y. (2009). Is the Brain Drain Good for Africa. In J. Bhagwati & G. H. Hanson (Eds.), Skilled Immigration Today: Prospects, Problems, and Policies (pp. 316–60). Oxford ; New York: Oxford University Press.; Scellato, G., Franzoni, C., & Stephan, P. (2012). Mobile Scientists and International Networks. Retrieved from http://www.nber.org/papers/w18613; Murakami, Y. (2013). Influences of return migration on international collaborative research networks: cases of Japanese scientists returning from the US. The Journal of Technology Transfer. doi: 10.1007/s10961-013-9316-9 51  In praise of the “brain drain”. (2007). Nature, 446(7133), 231. doi:10.1038/446231a 52  Weinberg, B. A., Hanson, G., & Rapoport, H. (2011). Developing science: Scientific performance and brain drains in the developing world. Journal of Development Economics, 95(1), 95–104. doi:10.1016/j.jdeveco.2010.05.009; Docquier, F., & Rapoport, H. (2012). Globalization, Brain Drain, and Development. Journal of Economic Literature, 50(3), 681–730. doi:10.1257/jel.50.3.681 53  Moed, H. F., Aisati, M., & Plume, A. (2012). Studying scientific migration in Scopus. Scientometrics, 94(3), 929–942. doi:10.1007/s11192-012-0783-9 54  UK Department of Business Innovation and Skills. (2011). International Comparative Performance of the UK Research Base - 2011. (A. Plume, M. El Aisati, M. Amin, N. Gracy, N. Weertman, & N. Fowler, Eds.) (p. 88). London: Elsevier. Retrieved from http://www.bis.gov.uk/assets/BISCore/science/docs/I/11-p123-international-comparative-performance-uk-research-base-2011.pdf 55  Science Europe & Elsevier. (2013). Comparative Benchmarking of European and US Research Collaboration and Researcher Mobility. Retrieved from http://www.scienceeurope.org/uploads/Public documents and speeches/SE and Elsevier Report Final.pdf 56  See Appendix C for more details on what constitutes an active researcher. 4 researcher mobility 52 Mobility Classes The model generates several main categories of researchers. Category Description Sedentary Researchers who have only published with affiliations to institutions within a particular region. This includes researchers who move between institutions within the same region. Inflow Researchers who come to the region. Outflow Researchers who leave the region. Returnees (Inflow) Researchers who first publish while at an institution in the region, leave and publish with an affiliation to an institution outside of the region for two or more years, and ultimately return to back to the region. The institutional affiliation of their return destination need not be the same as their “original institution”. Returnees (Outflow) Researchers who first publish elsewhere, come and stay in the region for two or more two years, and then leave to publish elsewhere. The institutional affiliation of their post-region destination need not be the same as their “original institution”. Transitory Researchers that spend less than two years at an institution in the region or an institution out- side the region at any given time; within this group, this report separately analyzes those that publish the majority of their work with region-affiliations versus non-region affiliations, denoting the former as Local Transitory researchers and the latter as Non-local Transitory researchers. Indicators For each of the mobility classes, the analysis provides several indicators to characterize the publication profile of the sets of researchers: Indicator Description Relative Productivity The number of papers published per year (PPY) since the first appearance of each researcher as an author in the database during the period 1996-present, relative to all researchers in that re- gion for the same period. The analysis calculates the relative productivity for an author’s entire output of articles, not just those articles with a particular regional affiliation. Relative productivity somewhat normalizes for career length, enabling comparisons of produc- tivity across different groups (e.g. those comprising mostly early career researchers versus those comprising mostly more senior academics). For instance, a group that has a relative productivity of 1.28 produces 28% more PPY than that institution’s overall average PPY. Relative Age The number of years since the first appearance of each researcher as an author in the database relative to all researchers in the region in the same period. The analysis calculates relative age for the author’s entire output in articles (e.g., not just those with a particular regional affiliation). Since the dataset goes as far back as 1996, reporting on relative age is right-censored (e.g., the time since a researcher’s first appearance as an author has an upper limit of 17 years). FWCI The FWCI (see Appendix B for full definition) of all articles associated with a researcher, regard- less of whether that researcher lists the given region as an affiliation on said articles. 4.3 international mobility 53 4.3 International Mobility 4.3.1. East Africa For conciseness, this section presents the brain circulation model of East Africa. The brain circulation models for the other regions can be found in Appendix F. The brain circula- tion model in Figure 4.1 is based on the movement of 8750 active East African researcher profiles. These profiles account for 87% of all articles published with an affiliation to an institution in the East African between 1996-2013. As a comparison, the FWCI of articles associated with all East African researchers over this period is 1.65, while that of articles associated with active East African researchers is 1.74. The Outflow groups of East African researchers tend to be more senior, productive, and impactful. Returnees Outflow, or those researchers that spend more than 2 years in East African institutions before leaving to publish elsewhere, are amongst the most productive and impactful of all mobil- ity classes - they produce 35% more papers per year on average than the typical East African researcher, and the average FWCI of their papers at 2.14 is well above the 1.74 average associated with all active East African researchers. Transitory researchers comprise the great bulk of East African researchers at 57.2%. Within this group, there is a big difference between non-local transitory researchers (visiting scholars) and local transitory researchers. The for- mer are much more productive (relative productivity of 1.38 versus 0.57), senior (relative age of 1.16 versus 0.84), and impactful (FWCI of 1.81 versus 1.42). Relative to past studies of other regions, East Africa has a low number of sedentary researchers (24%); in contrast, 31.7% and 56.8% of US and European active research- ers remain in their respective regions throughout their careers.57 Such researchers tend to be less productive (relative productivity of 0.47) but also younger (relative age of 0.70). 57  Science Europe, & Elsevier. (2013). Comparative Benchmarking of European and US Research Collaboration and Researcher Mobility. Retrieved from http://www.scienceeurope.org/uploads/Public documents and speeches/SE and Elsevier Report Final.pdf 4 researcher mobility 54 BRAIN CIRCULATION MODEL EAST AFRICA BRAIN INFLOW TRANSITORY BRAIN MOBILITY BRAIN OUTFLOW Researchers: 9.1% Researchers: 57.2% Researchers: 9.7% Relative Productivity: 0.75 Relative Productivity: 1.17 Relative Productivity: 1.09 Relative Age: 1.18 Relative Age: 1.06 Relative Age: 1.22 FWCI: 1.74 FWCI: 1.76 FWCI: 1.99 Inflow Transitory (mainly East Africa) Outflow Researchers: 5.4% Researchers: 18.4% Researchers: 5.0% Relative Productivity: 0.66 Relative Productivity: 0.57 Relative Productivity: 0.81 Relative Age: 1.17 Relative Age: 0.84 Relative Age: 1.14 FWCI: 1.50 FWCI: 1.42 FWCI: 1.73 Returnees Inflow Transitory (mainly non-East Africa) Returnees Outflow Researchers: 3.6% Researchers: 38.8% Researchers: 4.7% Relative Productivity: 0.89 Relative Productivity: 1.38 Relative Productivity: 1.35 Relative Seniority: 1.20 Relative Age: 1.16 Relative Age: 1.31 FWCI: 2.00 FWCI: 1.81 FWCI: 2.14 SEDENTARY Researchers: 24.0% Relative Productivity: 0.47 Relative Age: 0.70 FWCI: 1.13 Figure 4.1 — International mobility of East African researchers, 1996-2013. 4.4 cross-region comparisons 55 4.4 Cross-Region Comparisons Although transitory researchers account for the larg- Moreover, relative to the each region's overall average est part of each region’s total researcher base, there is researcher FWCIs, West & Central Africa's Brain Outflow significant variation in the relative distribution of other researchers have the highest adjusted FWCI (1.25). This researcher classes. West & Central Africa has the highest suggests that while East Africa loses the most impactful percentage of sedentary researchers at 41.8% while researchers amongst all the regions, the relative effect of Southern Africa has the lowest percentage at 14.7%. In West & Central Africa’s Brain Outflow is more acute. other words, 85.3% of all Southern African researchers have published an article while outside of Southern Africa. Table 4.3 provides a more granular breakdown of the dif- Taking together the total Outflow (5.7%) and total Inflow ferent mobility classes. Returnees outflow researchers (8.5%), West & Central Africa has a net inflow of research- – those who initially publish in abroad, move to an Africa ers (2.8%), while Southern Africa has a substantial net region for more than two years, and then go abroad again – outflow (-5.6%). have the highest adjusted FWCI amongst all categories of African researchers. Table 4.2 shows the adjusted FWCI associated with the different mobility classes, normalized against each Southern African researchers categorized as Returnees region’s overall researcher FWCI. West & Central Africa’s Inflow have the highest FWCI (2.02) associated with any Sedentary researchers have the lowest adjusted FWCI regions’ Returnees Inflow. However, East Africa’s Re- (0.43 compared to the next lowest region’s researchers, turnee Inflow researchers have high adjusted FWCI and South Africa at 0.60). In other words, while moving abroad comprise the largest (though still small) percentage of is positively associated with the impact of researchers' the Africa region’s total researcher pools at 3.6%. This outputs across all regions, the relative benefit of doing suggests that, amongst all the regions, East Africa has so is largest for West & Central African researchers. benefited the most from academic returning migrants. Table 4.1 — Researcher mobility classes as percentage of total active research base for SSA regions and South Africa based on brain circulation models, 1996-2013. Sedentary Brain Outflow Transitory Brain Inflow East Africa 24.0% 9.7% 57.2% 9.1% South Africa 34.0% 8.0% 49.1% 8.9% Southern Africa 14.7% 13.1% 64.7% 7.5% West & Central Africa 41.8% 5.7% 44.1% 8.5% Table 4.2 — Adjusted FWCI associated with researcher mobility classes (e.g., FWCI for individual mobility classes normal- ized against each region’s overall researcher FWCI) for SSA regions and South Africa based on brain circulation models, 1996-2013. Overall Sedentary Brain Outflow Transitory Brain Inflow East Africa 1.00 0.65 1.14 1.01 1.00 South Africa 1.00 0.60 0.94 1.10 0.92 Southern Africa 1.00 0.67 0.98 1.03 0.96 West & Central Africa 1.00 0.43 1.25 1.14 0.98 Table 4.3 — Adjusted FWCI associated with detailed researcher mobility classes (e.g., FWCI for individual mobility classes normalized against each region’s overall researcher FWCI) for SSA regions and South Africa based on brain circulation models, 1996-2013. Returnees Non-Local Local Returnees Overall Outflow Outflow Transitory Transitory Inflow Inflow East Africa 1.00 0.76 1.13 1.05 0.82 1.06 0.91 South Africa 1.00 0.90 0.98 1.16 0.74 0.98 0.89 Southern Africa 1.00 0.99 1.23 1.04 0.82 1.15 0.86 West & Central Africa 1.00 1.07 1.41 1.27 0.64 0.98 0.98 4 researcher mobility 56 Interpretation of Key Findings 4.5  on Researcher Mobility The following section interprets the main findings on and Southern Africa. The inflow of returnees research- research mobility in SSA and make five overall suggestions ers - those who originally publish from an African for interpretation: institution, left and published elsewhere, and then sub- sequently returned – make up a relatively small share 1. African researcher are highly mobile, particularly those of the region’s total researcher base (3.6% and 2.1%, from East and Southern Africa. Transitory research- respectively), yet the relative citation impact of those ers – those who spend less than 2 years in or outside returnees’ publication is quite high compared to that of the region – comprise 57.2% and 65% of East Africa’s other SSA researchers. and Southern Africa’s total researcher base. In contrast, 44% and 49% of West & Central Africa and South 4. Visiting faculty (transitory mainly publishing at institu- Africa’s research base, respectively, are transitory tions outside of Africa), which also can be diaspora, researchers). Moreover, a large percentage of SSA contribute even more to raising the volume and impact researchers are non-local and transitory; that is, they (citations) of research. Such researchers produce spend less than 2 years at institutions in SSA. 39% research that is between 4% and 27% more impactful and 48% of all East and Southern African researchers, than the average researcher in the region. respectively, fall into this category. 5. West & Central Africa displays a different pattern of The high percentage of non-local transitory researchers is researcher mobility. A higher share of West & Central concerning. The transitory nature of many researchers may African researchers is sedentary – i.e. not migrating prevent researchers from building relationships with Afri- to institutions outside of their region (44% for West & can firms and governments, reducing the economic impact Central Africa vs. 15% and 24% for Southern and East and relevance of research. Africa, respectively). Several key drivers could explain the high level of re- Several particularities of West & Central Africa could searcher mobility: inadequate research infrastructure, explain these differences: (i) a large part of West & Central low African production of PhDs/researchers, shortages in Africa is Francophone. This could reduce international funding, a high degree of international funding for interna- scientific collaboration which is in many cases conducted in tional researchers, lower dynamism, incentives, and scale English. (ii) Another potential contributing factor is meas- of research environments within the region. The interesting urement bias if Francophone research is not adequately and unique research topics, including within health and ag- published or indexed; (ii) a higher share of unstable political ricultural, that Africa offers could be highly attractive to re- environments could lower the willingness of researchers to searchers from other regions and the African diaspora. This travel to this part of Africa. genuine commitment to support the development of African science from a large number of international academics, including diaspora, should not be underestimated. 2. The research productivity and impact of the mobile African researcher is markedly higher than those of sedentary African researchers. For the SSA regions, the latter type of researcher produces research that is between 33% and 57% less impactful than sedentary researchers. This is likely to be the results of several factors: prior, unobserved differences between the types of researchers and collaboration with interna- tional researchers, exposure to new ideas, and access to better resources internationally. 3. Returning diaspora contribute significantly to raising the citation impact of SSA research, specifically in East 57 APPENDICES 58 APPENDIX A AUTHOR CREDITS, ADVISORY GROUPS, & ACKNOWLEDGEMENTS This study was commissioned by the World Bank. It was jointly conducted and written by George Lan, Dr. Judith Kamalski, Georgin Lau and Jeroen Baas at Elsevier and Andreas Blom and Mariam Adil at the World Bank. Special thanks to Dr. Peter Materu, Dr. Sajitha Bashir, Michael Crawford, Casey Torgusson and Kofi Anani at the World Bank, Dr. Nkem Khumbah at the University of Michi- gan, Dr. Rudiger Klein at the Max Planck Institute of Neu- robiology, Sudi Jessurun, Steven Scheerooren, Sarah Hug- gett, Matthew Richardson, Mohamed Kamel, Olga Barham, Josine Stallinga, and Hanna Sohn at Elsevier, Emilio Bunge and Molly Haragan at Development Finance International, Inc. for providing helpful reviews of and feedback on drafts of this report. Thank you to the Norwegian Government for funding for World Bank staff time through its Africa Post- Basic trust fund. The study is amongst of a series of technical outputs being produced under the World Bank Partnership for Applied Sciences, Engineering and Technology (PASET) initiative. Preliminary findings from this report were presented and reviewed at a high-level forum on Higher Education for Sci- ence, Technology, and Innovation in Kigali, Rwanda in March 2014 that was co-hosted by the Government of Rwanda and the World Bank and attended by representatives from the governments of Ethiopia, Mozambique, Rwanda, Sen- egal, and Uganda as well as private sector participants and development partners. The findings were presented and further refined at an internal World Bank seminar in March 2014, the University of Michigan STEM-Africa Initiative Third Biennial Conference in April 2014, and a World Bank review panel in August 2014. We thank the participants at all of these events for their helpful insights and feedback. This report was designed for online and print by CLEVER°FRANKE, www.cleverfranke.com The report is available online at www.worldbank.org/africa/stemresearchreport 59 APPENDIX B GLOSSARY Article (unless otherwise indicated) denotes the main claims of the work citing it. The number of citations received types of peer reviewed documents published in journals: by an article from subsequently-published articles is a articles, reviews, and conference papers. proxy of the quality or importance of the reported research. Article output for an institution or region is the count Downloads are defined as either downloading a PDF of of articles with at least one author from that institution an article on ScienceDirect, Elsevier’s full-text platform, (according to the affiliation listed in the authorship byline). or looking at the full-text online on ScienceDirect without All analyses make use of ‘whole’ rather than ‘fractional’ downloading the actual PDF. Views of abstracts are not counting: an article representing international collaboration included in the definition. Multiple views or downloads of (with at least two different countries listed in the authorship the same article in the same format during a user session byline) is counted once each for every institution listed. will be filtered out, in accordance with the COUNTER Code of Practice Release 4.58 ScienceDirect provides download Article share (world) is the share of publications for a data for approximately 16% of the articles indexed in Sco- specific region expressed as a percentage of the total world pus. It is assumed that user downloading behavior across output. Using article share in addition to absolute numbers countries does not systematically differ between online of article provides insight by normalizing for increases in platforms. Field-weighted download impact is calculated overall growth of the world’s research enterprise. from these data according to the same principles applied to the calculation of field-weighted citation impact. CAGR (Compound Annual Growth Rate) is defined as the year-over-year constant growth rate over a specified period FWCI (Field-Weighted Citation Impact) is an indicator of of time. Starting with the first value in any series and apply- mean citation impact, and compares the actual number of ing this rate for each of the time intervals yields the amount citations received by an article with the expected number in the final value of the series. of citations for articles of the same document type (article, 1 review or conference proceeding paper), publication year  CAGR (t0, tn ) = (V (tn ) / V (t0)) tn  t0  1 and subject field. Where the article is classified in two or more subject fields, the harmonic mean of the actual and V (t0): start value expected citation rates is used. The indicator is therefore V (tn ): finish value always defined with reference to a global baseline of 1.00 tn  t0: number of years. and intrinsically accounts for differences in citation accrual over time, differences in citation rates for different docu- ment types (reviews typically attract more citations than Citation is a formal reference to earlier work made in an research articles, for example) as well as subject-specific article or patent, frequently to other journal articles. A cita- differences in citation frequencies overall and over time and tion is used to credit the originator of an idea or finding and document types. It is one of the most sophisticated indica- is usually used to indicate that the earlier work supports the tors in the modern bibliometric toolkit. Field-Weighted Citation Impact (FWCI) Publication Article Subject Actual # of FWCIx: Ca/Ce year type Area(s) citations: Ca Publication X Calculate average # of citations to that set of publication. Collect set of all publications with same Expected # of citations: Ce publication year, subject area, and article type appendix b 60 When field-weighted citation impact is used as a snap- Journal is a peer-reviewed periodical in which scholarship shot, an un-weighted variable window is applied. The relating to a particular research field is published, and is field-weighted citation impact value for ‘2008’, for the primary mode of dissemination of knowledge in many example, is comprised of articles published in 2008 and fields. Research findings may also be published in confer- their field-weighted citation impact in the period 2008- ence proceedings, reports, monographs and books and the 12, while for ‘2012,’ it is comprised of articles published significance of these as an output channel varies between in 2012 and their field-weighted citation impact in 2012 fields. alone. When field-weighted citation impact is used in trend analysis, a weighted moving window is applied. R&D (Research and Development) is any creative sys- The field-weighted citation impact value for ‘2010’, for tematic activity undertaken in order to increase the stock example, is comprised of the weighted average of the of knowledge, including knowledge of man, culture and unweighted variable field-weighted citation impact values society, and the use of this knowledge to devise new ap- for 2008 and 2012 (weighted 13.3% each), 2009 and plications. R&D includes fundamental research, applied 2011 (weighted 20% each) and for 2010 (weighted research in such fields as agriculture, medicine, industrial 33.3%). The weighting applies in the same ratios for chemistry, and experimental development work leading to previous years also. However, for 2011 and 2012 it is new devices, products or processes. not possible to extend the weighted average by 2 years on either side, so weightings are readjusted across the Research collaboration is indicated by articles with remaining available values. at least two different institutions listed in the authorship byline. Highly cited articles (unless otherwise indicated) are those in the top-cited X% of all articles published and Sectors in this report refer to the different organization cited in a given period. types used to categorize institutional affiliations. The main sectors are: Hypercollaboration – while no definition exists on the number of co-authors required to constitute ‘hyper- Academic – universities, colleges, medical schools, and collaborative’ co-authorship, numbers in the hundreds research institutes or thousands seem worthy of the term. The most multi- authored research paper of all time was published in April Corporate – companies and law firms 2010 and has 3,222 authors from 37 countries 53. As an indication of the frequency of such hypercollaborative Government – government and military organizations articles, 74 articles published in 2012 had more than 3,000 authors; like the record holder, all of them reported Medical – hospitals results from the ATLAS experiment at CERN’s Large Hadron Collider in Switzerland. Indeed, hypercollabora- Other – non-governmental organizations, policy institutes, tive co-authorship may be a consequence of the rise of foundations, and other non-profit organizations so-called ‘Big Science’ – a term used to describe research that requires major capital investment and is often, but not always, international in nature.59 While such hypercollaborative articles may represent ex- treme outliers in co-authorship data, they are included in all the analyses since they remain proportionally few and because they are counted only as a single internationally co-authored article for each country represented in the article, and for each country pairing. Intellectual property (IP) are intangible assets such as discoveries and inventions for which exclusive rights may be claimed, including that which is codified in copyright, trademarks, patents, and designs. International Collaboration (i.e., research collabo- 58 http://www.projectcounter.org/r4/COPR4.pdf ration) in this report is indicated by articles with at least 59 Hand, E. (2010). “Big science” spurs collaborative trend.  two different countries listed in the authorship byline. Nature, 463(7279), 282. doi:10.1038/463282a 61 APPENDIX C DATA SOURCES & METHODOLOGY Data Sources For this report, a static version of the Scopus database cov- The key findings and insights discussed in this report are ering the period January 1, 1996-December 1, 2013 was based on a bibliometric analysis of the relevant publica- aggregated by country, region, and subject. Subjects were tion data from 2003-2012, which comes from Elsevier’s defined by ASJC subject areas (see elsewhere Appendix search and discovery research abstract database, Sco- C for more details). When aggregating article and citation pus.60 To augment the view of knowledge exchange, this counts, an integer counting method was employed where, report also draws on usage data from ScienceDirect,61 for example, a paper with two authors from a Rwanda (in Elsevier’s full-text journal article platform, and citation- East Africa) address and one from a South Africa address indexed patent data from LexisNexis TotalPatent and the would be counted as one article for each region (i.e. 1 East United State Patent and Trademark Office (USPTO). Africa and 1 South Africa). This method was favored over fractional counting, in which the above paper would count Scopus is Elsevier’s abstract and citation database as 0.67 for East Africa and 0.33 for South Africa, to main- of peer-reviewed literature, covering 50 million docu- tain consistency with other reports (both public and private) ments published in over 21,000 journals, book series we have conducted on the topic. and conference proceedings by some 5,000 publishers. Reference lists are captured for 29 million records pub- A body of literature is available on the limitations and ca- lished from 1996 onwards, and the additional 21 million veats in the use of such ‘bibliometric’ data, such as the ac- pre-1996 records reach as far back as the publication cumulation of citations over time, the skewed distribution of year 1823. citations across articles, and differences in publication and citation practices between fields of research, different lan- Scopus coverage is multi-lingual and global: approxi- guages, and applicability to social sciences and humanities mately 21% of titles in Scopus are published in lan- research. In social sciences and humanities, the bibliometric guages other than English (or published in both English indicators presented in this report for these fields must be and another language). In addition, more than half of interpreted with caution because a reasonable proportion Scopus content originates from outside North America, of research outputs in such fields take the form of books, representing many countries in Europe, Latin America, monographs and non-textual media. As such, analyses of Africa and the Asia Pacific region. In particular, Scopus journal articles, their usage and citation, provides a less comprises over 400 titles from publishers based in the comprehensive view than in other fields, where journal arti- Middle East and Africa. For more information, see http:// cle comprise the vast majority of research outputs. www.elsevier.com/__data/assets/pdf_file/0019/148402/ SC_Content-Coverage-Guide_July-2014.PDF 60 Scopus is the largest abstract and citation database of peer-reviewed  literature, covering 50 million records published in over 21,000 Scopus coverage is also inclusive across all major titles including over 400 titles from publishers in the Middle East research fields, with 6,900 titles in the Physical Sci- and Africa. This assuages concerns raised by researchers such ences, 6,400 in the Health Sciences, 4,150 in the Life as Tijssen (2007) that past bibliometric analyses have excluded a Sciences, and 6,800 in the Social Sciences (the latter significant portion of Africa’s research output placed in African- including some 4,000 Arts & Humanities related titles). published journals. Please see Appendix C: Data Sources for Titles which are covered are predominantly serial publica- more details. Tijssen, R. J. W. (2007). Africa’s contribution to the tions (journals, trade journals, book series and confer- worldwide research literature: New analytical perspectives, trends, ence material), but considerable numbers of conference and performance indicators. Scientometrics, 71(2), 303–327. papers are also covered from stand-alone proceedings doi:10.1007/s11192-007-1658-3 volumes (a major dissemination mechanism, particularly 61 Usage is defined as full-text article downloads or full-text article  in the computer sciences). Acknowledging that a great views online from Elsevier’s ScienceDirect database, which provides deal of important literature in all fields (but especially in approximately 20% of the world’s published journal articles. For more the Social Sciences and Arts & Humanities) is published information on the coverage and distribution of scientific content in in books, Scopus has begun to increase book coverage in ScienceDirect, please see Appendix C: Measuring Article Downloads 2013, aiming to cover some 75,000 books by 2015. for more details. appendix c 62 ScienceDirect is Elsevier’s full-text journal articles fields, and they may even change in magnitude over time. platform. With an invaluable and incomparable customer Given the complexities of determining and accounting for base, the use of scientific research on ScienceDirect.com the time lags between input and output, this report does provides a different look at performance measurement. not attempt to directly link the two. Readers are welcome to ScienceDirect.com is used by more than 12,000 institutes further interpret this report’s findings from a productivity worldwide, with more than 11 million active users and over perspective, such as normalizing article output and citation 700 million full-text article downloads in 2012. The aver- counts by a region’s population, per-unit R&D expenditure, age click through to full-text per month is nearly 60 million. or researcher headcount. However, such measures are More info can be found on http://www.elsevier.com/online- more meaningful in a comparative rather than absolute tools/sciencedirect sense. LexisNexis is a leader in comprehensive and authoritative Methodology and Rationale legal, news and business information and tailored applica- Our methodology is based on the theoretical principles and tions. LexisNexis® is a member of Reed Elsevier Group plc. best practices developed in the field of quantitative science Patents are obtained via a partnership with LexisNexis and and technology studies, particularly in science and technol- include those from the United States Patent and Trade- ogy (S&T) indicators research. The Handbook of Quantita- mark Office (USPTO), the European Patent Office (EPO), tive Science and Technology Research: The Use of Publica- the Japanese Patent Office (JPO), the Patent Cooperation tion and Patent Statistics in Studies of S&T Systems (Moed, Treaty (PCT) of the World Intellectual Property Organization Glänzel and Schmoch, 2004) 64 gives a good overview of (WIPO) and the UK Intellectual Property Office (UKIPO). this field and is based on the pioneering work of Derek de Solla Price (1978) 65, Eugene Garfield (1979) 66 and Francis World Bank Africa Development Indicators is a collection Narin (1976) 67 in the USA, and Christopher Freeman, Ben of development indicators compiled from officially recog- Martin and John Irvine in the UK (1981, 1987) 68, and in nized international sources, presenting the most current several European institutions including the Centre for and accurate global development data available. This study Science and Technology Studies at Leiden University, the particularly draws on data about SSA GDP and population Netherlands, and the Library of the Academy of Sciences in size to calculate research output per capita. More info can Budapest, Hungary. be found on http://data.worldbank.org/data-catalog/africa- development-indicators The analyses of research output data in this report are based upon recognized advanced indicators (e.g., the Changes in measures over time concept of relative citation impact rates). Our base assump- The main data sources used in this report (Scopus, Science- tion is that such indicators are useful and valid, though Direct usage data, LexisNexis patent citations index based imperfect and partial measures, in the sense that their on USPTO data) represent dynamic databases that are numerical values are determined by research performance regularly updated throughout the year. The indicators are and related concepts, but also by other, influencing factors therefore a snapshot taken from the data at a point in time. that may cause systematic biases. In the past decade, the For instance, the citation counts associated with South field of indicators research has developed a best practices Africa’s publications will increase over time. In some cases, which state how indicator results should be interpreted and the most recent values may be provisional as earlier data which influencing factors should be taken into account. Our may be revised as a result of initiatives to expand data com- methodology builds on these practices. pleteness. For example, in Scopus, a significant expansion of journal coverage in the Arts & Humanities beginning in Article Types 2009 has resulted in a more robust view of journal articles For all research output analyses, only the following, peer- and related output indicators in that area. This report used reviewed document types are considered: data from a December 1, 2013 snapshot of the aforemen- tioned data sources. ► Article (ar) ► Review (re) Time lags between inputs and outputs ► Conference Proceeding (cp) In the input-output model of research & development (R&D) evaluation 62, inputs such as R&D expenditure or human Article Counting and Deduplication capital must precede outputs such as journal articles and All analyses make use of whole counting rather than frac- citations. The results of a research grant awarded in 2010 tional counting. For example, if a paper has been co-au- may not be published in the peer-reviewed literature for thored by one author from East Africa and one author from several years, and a patent application may follow after an Southern Africa, then that paper counts towards both the even longer delay. 63 Such lags vary by indicator and subject publication count of East Africa as well as the publication appendix c 63 count of Southern Africa. Total counts for each region are surname by differentiating on data elements associated with the unique counts of publications. the article (such as affiliation, subject area, co-authors, and so on). An article can be counted in more than one subject group- ing. However, it is calculated only once toward the count of The Scopus algorithm favors accuracy and only groups a region’s total publications. For example, a West & Central together publications when the confidence level that they Africa publication on the impact of increased corn produc- belong together – the precision of matching – is at least tion on pricing may be counted once each toward the totals 99% (that is, in a group of 100 papers, 99 will be correctly of that region’s research output in Agricultural & Biological assigned). This level of accuracy results in a recall of 95% Sciences and Economics, Econometrics, & Finance. Howev- across the database: if an author has published 100 papers, er, this publication counts only once toward the aggregate on average, 95 of them will be grouped together by Scopus. entity of all West & Central Africa’s publications. These precision and recall figures are accurate across the entire Scopus database. There are situations where the Deduplication in the calculation of measures concentration of similar names increases the fragmenta- All analyses make use of whole counting rather than frac- tion of publications between Author Profiles, such as in the tional counting of articles. For example, if an article has well-known example of Chinese authors. Equally there are been co-authored by one author from East Africa and one instances where a high level of distinction in names results in author from Southern Africa, then that article is added to- a lower level of fragmentation, such as in Western countries. wards both the output of East Africa, as well as the output of Southern Africa. Total counts for each region are the The matching algorithm can never be 100% correct because unique count of articles. the data it is using to make the assignments are not 100% complete or consistent. The algorithm is therefore enriched The same article may be part of multiple smaller component with manual, author-supplied feedback, both directly through entities, such as the calculation of article counts in subject Scopus and also via Scopus’ direct links with ORCID (Open groupings. However, this report deduplicates all articles Researcher & Contributor ID 69). within an aggregate entity. For example, an article from Southern Africa on the impact of increased corn production What determines whether an author is an “East African on pricing may be counted once each toward the totals of researcher” or an analogous researcher from the other that region’s output in Agriculture and the Social Sciences sub-Saharan regions? & Humanities. However, the article is counted only once To define the initial population for study, East African au- toward the aggregate total of all articles from that region. thors were defined as those that had listed an affiliation with an East African institution on at least one publication (arti- Citation Counting and Self-Citations cles, reviews and conference papers) published across the Self-citations are those by which an entity refers to its sources included in Scopus during the period 1996–2013. previous work in new publications. Self-citing is normal and expected academic behavior, and it is an author’s respon- sibility to make sure their readers are aware of related, 62 Godin, B. (2007). Science, accounting and statistics: The  relevant work. For this report, self-citations are included in input–output framework. Research Policy, 36(9), 1388–1403. citation counts and the calculation of FWCI. doi:10.1016/j.respol.2007.06.002 63 Shelton, R. D., & Leydesdorff, L. (2012). Publish or patent:  Measuring International Researcher Mobility Bibliometric evidence for empirical trade-offs in national funding The approach presented here uses Scopus author profile strategies. Journal of the American Society for Information Science data to derive a history of active author affiliations record- and Technology, 63(3), 498–511. doi:10.1002/asi.21677 ed in their published articles and to assign them to mobil- 64 Moed H., Glänzel W., & Schmoch U. (2004). Handbook of Quantitative  ity classes defined by the type and duration of observed Science and Technology Research, Kluwer: Dordrecht. moves. 65 de Solla Price, D.J. (1977–1978). “Foreword,” Essays of an  Information Scientist, Vol. 3, v–ix. How are individual researchers unambiguously identi- 66 Garfield, E. (1979). Is citation analysis a legitimate evaluation tool?  fied in Scopus? Scientometrics 1 (4): pp. 359-375. Scopus uses a sophisticated author-matching algorithm to 67 Pinski, G., & Narin, F. (1976). Citation influence for journal aggregates  precisely identify articles by the same author. The Scopus of scientific publications: Theory with application to literature Author Identifier gives each author a unique ID and groups 68 Irvine, J., Martin, B. R., Abraham, J. & Peacock, T. (1987). Assessing  together all the documents published by that author, match- basic research: Reappraisal and update of an evaluation of four radio ing alternate spellings and variations of the author’s last astronomy observatories. Research Policy 16(2-4): pp. 213-227. name and distinguishing between authors with the same 69 http://orcid.org/ appendix c 64 What is an ‘active researcher’? ► Total Outflow: the sum of Outflow and Returnee Outflow The total authors identified for this reports’ analysis include groups. a large proportion with relatively few articles over the entire Inflow: active researchers whose Scopus author data ►  ten-year period of analysis. As such, it was assumed that for the period 1996-2013 indicate that they have they are not likely to represent career researchers, but migrated from institution(s) outside of the region to individuals who have left the research system. A productiv- institution(s) in the region for at least 2 years without ity filter was therefore implemented to restrict the analysis leaving that region. to those authors with at least 1 article in the most recent Returnees Inflow: active researchers whose Scopus ►  5-year period 2009–2013 and at least 10 articles in the author data for the period 1996-2013 indicates that entire period 1996-present, or those with fewer than 10 they have migrated from institution(s) in the region to articles in 1996-present, and at least 4 articles in 2009- institution(s) outside the region for at least 2 years, and 2013. For instance, after applying the productivity filter then subsequently migrated back to institution(s) in the on the initial set of 58,293 researchers identified as being region for at least 2 years. affiliated with institutions in West & Central Africa, a set ► Total Inflow: the sum of Inflow and Returnee Inflow of 15,019 active researchers was defined and formed the groups. basis of the study. Transitory How are mobility classes defined? Transitory (mainly non-Africa region): active Africa ►  The measurement of international researcher mobility by region researchers whose Scopus author data for the co-authorship in the published literature is complicated by period 1996-2013 indicates that they were based the difficulties involved in teasing out long-term mobility in institution(s) in the Africa region for less than 2 from short-term mobility (such as doctoral research visits, years at a time and have been predominantly based in sabbaticals, secondments, etc.), which might be deemed institution(s) outside the Africa region. instead to reflect a form of collaboration. In this study, stays Transitory (mainly Africa region): active Africa re- ►  overseas of 2 years or more were considered migratory and gion researchers whose Scopus author data for the were further subdivided into those where the researcher period 1996-2013 indicates that they are based in remained abroad or where they subsequently returned to institution(s) outside the Africa region for less than 2 their original institution. Stays of less than 2 years were years at a time and have been predominantly based in deemed transitory, and were also further subdivided into institution(s) in the Africa region. those who mostly published under an ego-region or a Total Transitory: the sum of Transitory (mainly non-Afri- ►  non—ego-region affiliation. Since author nationality is not ca region) and Transitory (mainly Africa region) groups. captured in article or author data, authors are assumed to be from the institution where they first published (for migra- Sedentary tory mobility) or from the institution where they published Sedentary: active Africa region researchers whose Sco- ►  the majority of their articles (for transitory mobility). In pus author data for the period 1996-2013 indicates individual cases, these criteria may result in authors being that they have not published outside institution(s) in the assigned migratory patterns that may not accurately reflect Africa region. the real situation, but such errors may be assumed to be evenly distributed across the groups and so the overall What indicators are used to characterize each mobility pattern remains valid. Researchers without any apparent group? mobility based on their published affiliations were consid- To better understand the composition of each group de- ered sedentary. fined on the map, three aggregate indicators were calcu- lated for each to represent the productivity and seniority Migratory of the researchers they contain, and the field-weighted Outflow: active researchers whose Scopus author data ►  citation impact of their articles. for the period 1996-2013 indicates that they have migrated from institution(s) in the region to institution(s) Relative Productivity represents a measure of the ►  outside of the region for at least 2 years without return- articles per year since the first appearance of each ing to the respective region. researcher as an author during the period 1996-2013, Returnees Outflow: active researchers whose Scopus ►  relative to all Africa region researchers in the same author profile data for the period 1996-2013 indicates period. that they have migrated from institution(s) outside the Relative Seniority represents years since the first ap- ►  region to institution(s) in the region for at least 2 years, pearance of each researcher as an author during the and then subsequently migrated back to institution(s) period 1996-2013, relative to all Africa region re- outside the Africa region. searchers in the same period. appendix c 65 Field-weighted citation impact is calculated for all ►  articles in each mobility class. All three indicators are calculated for each author’s entire output in the period (i.e., not just those articles listing a corresponding ad- dress for that author). Measuring Article Downloads Citation impact is by definition a lagging indicator: newly- published articles need to be read, after which they might influence studies that will be carried out, which are then written up in manuscript form, peer-reviewed, published and finally included in a citation index such as Scopus. Only after these steps are completed can citations to the earlier article be systematically counted. For this reason, inves- tigating downloads has become an appealing alternative, since it is possible to start counting downloads of full text articles immediately upon online publication and to derive robust indicators over windows of months rather than years. While there is a considerable body of literature on the meaning of citations and indicators derived from them,70 the relatively recent advent of download derived indicators means that there is no clear consensus on the nature of the phenomenon that is measured by download counts.71 A small body of research has concluded however that down- load counts may be a weak predictor of subsequent citation counts at the article level.72 In this report, a download is defined as the event where a user views the full-text HTML of an article or downloads the full-text PDF of an article from ScienceDirect, Elsevier’s full-text journal article platform; views of an article abstract alone, and multiple full-text HTML views or PDF downloads of the same article during the same user session, are not included in accordance with the COUNTER Code of Prac- tice 73. ScienceDirect provides download data for approxi- mately 20% of the articles indexed in Scopus; it is assumed that user downloading behavior across countries does not systematically differ between online platforms. Field- weighted download impact is calculated from these data according to the same principles applied to the calculation of field-weighted citation impact. 70 Cronin, B. (2005). "A hundred million acts of whimsy?" Current Science 89 (9) pp. 1505-1509; Bornmann, L., & Daniel, H. (2008). "What do citation  counts measure? A review of studies on citing behaviour." Journal of Documentation 64 (1) pp. 45-80. 71 Kurtz, M.J., & Bollen, J. (2010). “Usage Bibliometrics” Annual Review of Information Science and Technology 44 (1) pp. 3-64.  72 Moed, H.F. (2005). “Statistical relationships between downloads and citations at the level of individual documents within a single journal” Journal of  the American Society for Information Science and Technology 56 (10) pp. 1088-1097; Schloegl, C. & Gorraiz, J. (2010). “Comparison of citation and usage indicators: The case of oncology journals” Scientometrics 82 (3) pp. 567-580; Schloegl, C. & Gorraiz, J. (2011). “Global usage versus global citation metrics: The case of pharmacology journals” Journal of the American Society for Information Science and Technology 62 (1) pp. 161-170. 73  ttp://usagereports.elsevier.com/asp/main.aspx; http://www.projectcounter.org/code_practice.html h 66 APPENDIX D AFRICA REGION CLASSIFICATION e a od ic C fr A er a r al ic ct r y to tr fr a ra n t ara ic en A a ha ic fr n C ou p fr ry er -C A C om t& tA th th nt 3 C es ou ou O ou as IS W S S E C Angola AGO ⚫ Benin BEN ⚫ Botswana BWA ⚫ Burkina Faso BFA ⚫ Burundi BDI ⚫ Cameroon CMR ⚫ Cape Verde CPV ⚫ Central African Republic CAF ⚫ Chad TCD ⚫ Comoros COM ⚫ Congo, Dem. Rep. ZAR ⚫ Congo, Rep. COG ⚫ Cote d'Ivoire CIV ⚫ Djibouti DJI ⚫ Equatorial Guinea GNQ ⚫ Eritrea ERI ⚫ Ethiopia ETH ⚫ Gabon GAB ⚫ Gambia, The GMB ⚫ Ghana GHA ⚫ Guinea GIN ⚫ Guinea-Bissau GNB ⚫ Kenya KEN ⚫ Lesotho LSO ⚫ Liberia LBR ⚫ Madagascar MDG ⚫ Malawi MWI ⚫ Malaysia MYS ⚫ Mali MLI ⚫ Mauritania MRT ⚫ Mauritius MUS ⚫ Mayotte MYT ⚫ Mozambique MOZ ⚫ Namibia NAM ⚫ Niger NER ⚫ Nigeria NGA ⚫ Rwanda RWA ⚫ Saint Helena, Ascension and Tristan da Cunha SHN ⚫ Sao Tome and Principe STP ⚫ appendix d 67 e a od ic C fr A er a r al ic to ct tr fr a ra ra ic en A a pa ha ic fr n C ry fr ry er om -C A t& tA nt th th nt 3 C ou es ou ou O ou as IS W C S S E C Senegal SEN ⚫ Seychelles SYC ⚫ Sierra Leone SLE ⚫ Somalia SOM ⚫ South Africa ZAF ⚫ South Sudan SSD ⚫ Swaziland SWZ ⚫ Tanzania TZA ⚫ Togo TGO ⚫ Uganda UGA ⚫ Vietnam VNM ⚫ Zambia ZMB ⚫ Zimbabwe ZWE ⚫ 68 APPENDIX E SUBJECT CLASSIFICATION Background on Scopus All Science Classification System (ASJC) Titles in Scopus are classified under four broad subject clusters (life sciences, physical sciences, health sciences and social sciences & humanities) which are further divided into 27 major subject areas and 300+ minor subject areas. Titles may belong to more than one subject area. For a complete list of titles associated with these subject areas, please see http://www.elsevier.com/online-tools/scopus/ content-overview For this report, these 27 subject areas are then aggregated into five major subject groupings: Agriculture; Physical Sciences & STEM, Health Sciences, Social Sciences & Humanities, and the Life Sciences. The main foci of the report are Agriculture, the Physical Sciences & STEM, and the Health Sciences. 69 APPENDIX F INTERNATIONAL RESEARCHER MOBILITY MAPS BRAIN CIRCULATION MODEL SOUTHERN AFRICA BRAIN INFLOW TRANSITORY BRAIN MOBILITY BRAIN OUTFLOW Researchers: 7.5% Researchers: 64.7% Researchers: 13.1% Relative Productivity: 0.77 Relative Productivity: 1.13 Relative Productivity: 0.93 Relative Age: 1.06 Relative Age: 1.03 Relative Age: 1.21 FWCI: 1.82 FWCI: 1.96 FWCI: 1.86 Inflow Transitory (mainly Southern Africa) Outflow Researchers: 5.4% Researchers: 16.3% Researchers: 7.5% Relative Productivity: 0.73 Relative Productivity: 0.48 Relative Productivity: 0.71 Relative Age: 1.06 Relative Age: 0.77 Relative Age: 1.17 FWCI: 1.73 FWCI: 1.56 FWCI: 1.45 Returnees Inflow Transitory (mainly non-Southern Africa) Returnees Outflow Researchers: 2.1% Researchers: 48.3% Researchers: 5.6% Relative Productivity: 0.88 Relative Productivity: 1.28 Relative Productivity: 1.21 Relative Seniority: 1.05 Relative Age: 1.11 Relative Age: 1.27 FWCI: 2.02 FWCI: 1.99 FWCI: 2.15 SEDENTARY Researchers: 14.7% Relative Productivity: 0.43 Relative Age: 0.67 FWCI: 1.28 Figure F.1 — International mobility of Southern African researchers, 1996-2013. appendix f 70 BRAIN CIRCULATION MODEL SOUTH AFRICA BRAIN INFLOW TRANSITORY BRAIN MOBILITY BRAIN OUTFLOW Researchers: 8.9% Researchers: 49.1% Researchers: 8.0% Relative Productivity: 0.96 Relative Productivity: 1.27 Relative Productivity: 0.88 Relative Age: 1.14 Relative Age: 1.10 Relative Age: 1.18 FWCI: 1.60 FWCI: 1.91 FWCI: 1.63 Inflow Transitory (mainly South Africa) Outflow Researchers: 6.5% Researchers: 13.6% Researchers: 4.9% Relative Productivity: 0.89 Relative Productivity: 0.57 Relative Productivity: 0.77 Relative Age: 1.11 Relative Age: 0.98 Relative Age: 1.16 FWCI: 1.54 FWCI: 1.28 FWCI: 1.57 Returnees Inflow Transitory (mainly non-South Africa) Returnees Outflow Researchers: 2.4% Researchers: 35.5% Researchers: 3.1% Relative Productivity: 1.13 Relative Productivity: 1.44 Relative Productivity: 1.05 Relative Seniority: 1.23 Relative Age: 1.14 Relative Age: 1.31 FWCI: 1.71 FWCI: 2.02 FWCI: 1.70 SEDENTARY Researchers: 34.0% Relative Productivity: 0.50 Relative Age: 0.78 FWCI: 1.04 Figure F.2 — International mobility of South African researchers, 1996-2013. appendix f 71 BRAIN CIRCULATION MODEL WEST & CENTRAL AFRICA BRAIN INFLOW TRANSITORY BRAIN MOBILITY BRAIN OUTFLOW Researchers: 8.5% Researchers: 44.1% Researchers: 5.7% Relative Productivity: 0.94 Relative Productivity: 1.27 Relative Productivity: 1.19 Relative Age: 1.26 Relative Age: 1.12 Relative Age: 1.33 FWCI: 1.21 FWCI: 1.40 FWCI: 1.54 Inflow Transitory (mainly West & Central Africa) Outflow Researchers: 5.0% Researchers: 19.7% Researchers: 3.5% Relative Productivity: 0.85 Relative Productivity: 0.65 Relative Productivity: 0.97 Relative Age: 1.24 Relative Age: 0.98 Relative Age: 1.28 FWCI: 1.21 FWCI: 0.79 FWCI: 1.32 Returnees Inflow Transitory (mainly non-West & Central Africa) Returnees Outflow Researchers: 3.5% Researchers: 24.3% Researchers: 2.2% Relative Productivity: 1.07 Relative Productivity: 1.67 Relative Productivity: 1.50 Relative Seniority: 1.30 Relative Age: 1.23 Relative Age: 1.42 FWCI: 1.20 FWCI: 1.56 FWCI: 1.74 SEDENTARY Researchers: 41.8% Relative Productivity: 0.57 Relative Age: 0.78 FWCI: 0.53 Figure F.3 — International mobility of West & Central African researchers, 1996-2013. 72 Notes @2014 Elsevier B.V. All rights reserved. Elsevier Research Intelligence® is a registered trademark of Elsevier Properties S.A. used under license. ELSEVIER