Measuring the Effectiveness of Crop Improvement Research in Sub

Measuring the Effectiveness of Crop Improvement Research in Sub-Saharan Africa from the Perspectives of Varietal Output, Adoption, and Change: 20 Crops, 30 Countries, and 1150 Cultivars in Farmers’ Fields

July 2014

CGIAR INDEPENDENT SCIENCE AND PARTNERSHIP COUNCIL

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s Diffusion and Impact of Improved Varieties in Africa (DIIVA) Project

Tom Walker, Arega Alene, Jupiter Ndjeunga, Ricardo Labarta, Yigezu Yigezu, Aliou Diagne, Robert Andrade, Rachel Muthoni Andriatsitohaina, Hugo De Groote, Kai Mausch, Chilot Yirga, Franklin Simtowe, Enid Katungi, Wellington Jogo, Moti Jaleta and Sushil Pandey

Prepared for the Standing Panel on Impact Assessment (SPIA) CGIAR Independent Science & Partnership Council (ISPC)

July 2014 The CGIAR Independent Science and Partnership Council encourages fair use of this material provided proper citation is made.

Walker, T., Alene, A., Ndjeunga, J., Labarta, R., Yigezu, Y., Diagne, A., Andrade, R., Muthoni Andriatsitohaina, R., De Groote, H., Mausch, K., Yirga, C., Simtowe, F., Katungi, E., Jogo, W., Jaleta, M. & Pandey, S. 2014. Measuring the Effectiveness of Crop Improvement Research in Sub-Saharan Africa from the Perspectives of Varietal Output, Adoption, and Change: 20 Crops, 30 Countries, and 1150 Cultivars in Farmers’ Fields. Report of the Standing Panel on Impact Assessment (SPIA), CGIAR Independent Science and Partnership Council (ISPC) Secretariat: Rome, Italy. Contents

Tables and figures iv Acronyms and abbreviations vi Acknowledgements vii Foreword xi Executive summary 1 Introduction 6 1. Crop and country coverage 8 2. Scientific strength in crop improvement 10 Scientific staff strength by research program in 2010 10 The congruence rule: How many full-time equivalent (FTE) scientists are desirable? 13 Differences in scientific strength over time 14 Other aspects of scientific capacity: age, education, experience, and area of specialization 17 FTE scientists in the CGIAR 19 Agreement between the DIIVA and the Agricultural Sciences and Technology Indicators Initiative (ASTI) FTE estimates 21 Summary 23 3. Varietal output 25 Findings on varietal output in 1998 26 Updating varietal output in 2010 26 Varietal output over time 28 The historical record on CGIAR contributions to varietal output 30 The maturity of crop improvement programs over time 32 Summary 34 4. Varietal adoption 37 Defining modern varieties (MVs) and estimating adoption 37 Adoption levels of improved varieties by crop in 2010 39 Adoption rates by country 43 Named MVs in the adoption database 44 Spillovers in adoption 46 IARC-related adoption 47 Comparing adoption levels between 1998 and 2010 48 Summary 51 5. Varietal change 53 The importance of estimating varietal change 53 The velocity of varietal turnover in 2010 by crop 53 Varietal change and adoption by crop-by-country observations 55 The vintage of adopted varieties 56 Comparing levels of varietal change in 1998 and 2010 57 Summary 58 6. Validating adoption estimates generated by expert opinion from survey estimates 59 Validating expert opinion on adoption and the protocol used to elicit estimates from experts 59 Describing the validation surveys on the diffusion of MVs 61 Validating expert opinion with the survey estimates 62 Focusing on challenges in nationally representative adoption surveys 66 Summary 68 7. Conclusions 70 References 74 Annex: Summary data on FTE scientists, varietal release, and adoption of improved varieties by crop and country 79

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — iii Tables and figures

Table 1.1. Area coverage in SSA in 2010 by crop 9

Table 1.2. The number of crop-by-country observations in the DIIVA Project by type of data 9

Table 2.1. FTE scientists by crop improvement program in SSA in 2010 11

Table 2.2. Estimated research intensities by crop in SSA in 2010 from the perspectives of area, production, and value of production 12

Table 2.3. Comparing the actual allocation of FTE scientists to a normative allocation by crop 14

Table 2.4. Differences in estimated FTE scientists and research intensities between 1998 and 2010 by crop based on 65 paired comparisons 15

Figure 2.1. Change in scientific staff strength in food crop improvement programs between 1997/98 and 2009/10. 16

Table 2.5. Educational level of scientists in crop improvement programs by region in SSA 17

Table 2.6. Relative allocation of scientists by disciplinary specialization across roots and tubers, grain legumes, and cereals in SSA in 2010 in % shares 18

Table 2.7. Relative allocation of scientists by disciplinary specialization in SSA in 2010 in % shares 19

Figure 2.2. Number of PhD scientists working in the groundnut, pearl millet, sorghum and pulse improvement programs in ICRISAT by location for selected years in 1978–2010 20

Figure 2.3. Comparing ASTI and DIIVA estimates of scientific strength 22

Table 3.1. Counting the number of cultivars in the varietal release database by crop in SSA from before 1970 to 2011 27

Table 3.2. The frequency of cultivar release by decade by crop in SSA 28

Table 3.3. Performance in varietal release from 1999 to 2011 29

Figure 3.1. Crop-by-country observations with more releases in the 1980s than in the 2000s 31

Table 3.4. The contribution of IARCs of the CGIAR to varietal output output in SSA, 1980–2011 32

Figure 3.2. Number of improved maize varieties released by decade and by origin 33

Table 3.5. IARC-related percent share estimates over time with and without maize in ESA 33

Figure 3.3. Trend in the release of bean varieties by germplasm source, 1970–2010 34

Figure 3.4. Trend in cassava variety releases by IITA content, 1970–2010 35

iv — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Table 3.6. Trend in the number of sorghum, pearl millet and groundnut varieties released by germplasm origin, 1970–2010 35

Figure 3.5. Trend in cowpea varietal releases by germplasm content, 1970–2010 35

Table 4.1. Source of the national adoption estimates by number of observations 39

Table 4.2 Adoption of MVs of food crops in SSA in 2010 40

Table 4.3. Weighted area adoption levels by country in SSA in 2010 44

Table 4.4. Top-ranked varieties by commodity and country by area and value of production 45

Table 4.5. The contribution of the CG Centers to MV adoption in SSA in 2010 48

Table 4.6. Change in MV adoption between 1998 and 2010 in ten food crops in SSA 49

Figure 4.1. Change in the estimated level of adoption of improved maize and cassava varieties between 1998 and 2010 50

Figure 4.2. Change in the estimated level of adoption of improved bean, groundnut, pearl millet, potato, rice, sorghum and wheat varieties between 1998 and 2010 51

Table 5.1. The velocity of varietal turnover of improved varieties in farmers’ fields in SSA by crop 54

Figure 5.1. Distribution of crop-by-country improvement level of adoption and the mean 55

Figure 5.2. The distribution of country by crop improvement level of adoption and mean age of varieties in farmers’ fields 56

Table 5.2. Genetic improvement programs with higher adoption and more rapid varietal turnover by crop and country 57

Table 5.3. The vintage of varieties contributing to adoption in 2010 by criterion and by release period 57

Table 6.1. Description of the sampling features of the diffusion MV validation surveys conducted by participants in the DIIVA Project 62

Table 6.2. Validating adoption estimates from expert opinion with survey results by crop 63

Table 6.3. Agreement between expert and survey estimates of specific varieties by expert interval 65

Table 6.4. Distribution of rice by variety planted during major growing season in 2012 67

Table 6.5. Comparison of village-level and household-level interview data on varietal adoption using area grown under these varieties for rice in Nigeria 68

Annex Table 1: Summary data on FTE scientists, varietal release, and adoption of improved varieties by crop and country 79

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — v Acronyms and abbreviations

ASTI Agricultural Sciences and IFPRI International Food Policy Technology Indicators Research Institute BMGF Bill & Melinda Gates Foundation IITA International Institute of Tropical BSc Bachelor of Science Agriculture CATIE Centro Agronómico Tropical de INTSTORMIL International Sorghum and Millet Investigación y Enseñanza (Costa Collaborative Research Project of Rica) USAID CAR Central African Republic IRAT Institut de Recherches CIAT Centro Internacional de Agronomiques Tropicales Agricultura Tropical (International IRRI International Rice Research Center for Tropical Agriculture) Institute CIMMYT Centro Internacional de ISNAR International Service for National Mejoramiento de Maiz y Trigo Agricultural Research (International Center for the IVT Institute of Horticultural Plant Improvement of Maize and Breeding Wheat) JICA Japan International Cooperation CIP Centro Internacional de La Papa Agency (International Potato Center) MAPE Mean Absolute Percent Error CIRAD Centre de Coopération MAS Marker-assisted selection Internationale en Recherche MDV Millennium Development Goals Agronomique pour le MSc Master of Science Développement (Agricultural MV Modern variety Research for Development) NARO National Agricultural Research CRSP Collaborative Research Support Organization (Uganda) Program of USAID NARS National agricultural research CSIR Council for Scientific and system Industrial Research (Ghana) NERICA New Rice for Africa (AfricaRice) DIIVA Diffusion and Impact of Improved NGO Non-governmental organization Varieties in Africa NVRS National Vegetatable Research EAP Escuela Agricola Panamericana Station Zamorano (Zamorano Pan- OPV Open pollinated varieties American Agricultural School) PABRA Pan-African Bean Research ECABRN East and Central Africa Bean Alliance Research Network PhD Doctor of Philosophy EIAR Ethiopian Institute of Agricultural PPP Purchasing power parity Research PSC Project Steering Committee ESA East and Southern Africa PVS Participatory varietal selection FAO Food and Agriculture QPM Quality protein maize Organization of the United SABRN Southern Africa Bean Research Nations Network FHIA Fundación Hondureña de SPIA Standing Panel on Impact Investigación Agrícola (Honduras Assessment Foundation for Agricultural SSA Sub-Saharan Africa Research) SYs Scientist years FTE Full-time equivalent USAID United States Agency for GDP Gross Domestic Product International Development GxE Genotype by environment WCA West and Central Africa HYV High-Yielding Varieties WECABRN West Africa Bean Research IARC International Agricultural Network Research Center ICARDA International Center for Agricultural Research in Dry Areas ICRISAT International Crops Research Institute for the Semi-Arid Tropics

vi — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Acknowledgements

This Synthesis Report is the product of the effort of several hundred people, not all of whom are named here. The project it documents was first conceived in 2008 during a Standing Panel on Impact Assessment (SPIA) workshop organized by Flavio Avila in Brasilia, capital city of Brazil; where the need for a comprehensive assessment of the improved varieties of food crops in Sub-Saharan Africa (SSA) was strongly expressed. A project proposal workshop was subsequently held in May 2009 at the Bill & Melinda Gates Foundation (BMGF); a proposal was submitted by Bioversity International; and the grant was awarded later that year. The project ended on 30 June 2013.

Greg Traxler from the Foundation was the conceptual force behind the project that was fully supported by the BMGF in Prabhu Pingali’s Division of Agricultural Policy and Statistics. Greg served as an ex-officio member of the Project Steering Committee (PSC) that met 23 times during the life of the project. Derek Byerlee and Doug Gollin – the chairpersons of SPIA – headed the PSC, devoting considerable time and energy to its conduct and management. Tim Kelley guided all of the project’s activities at SPIA, interacting with the coordinator on a daily or at least weekly basis. Tim was ably assisted by James Stevenson who also sat on the PSC and was involved in several project-related activities, including the storage and public distribution of the data. We’re thankful for the time and inspiration of other PSC members too: Elisabetta Gotor, Mywish Maredia and Gerry O’ Donoghue.

We’re equally grateful to the following people who contributed to the success of the project in multiple ways – from timely consultancies, through full-fledged support at the CG Centers, to the organization of three project workshops: Jeff Alwang, Aden Aw-Hassan, Cynthia Bantilan, Giorgia Beltrame, Daniel Calendo, Deevi KumaraCharyulu, Maria Piedade Caruche, Cheryl Doss, Jessamyn Findlay, Keith Fuglie, Melanie Glover, Carolina Gonzalez, Ben Groom, Guy Hareau, Mariana Kim, Hilde Koper, Lakshmi Krishnan, Catherine Larochelle, Roberto LaRovere, C. Magorokosho, Bernadette Majebelle, George Norton, Bonny Ntare, Alastair Orr, David Raitzer, Byron Reyes, Bekele Schiferaw, Randy Shigetani, Cecily Stokes-Prindle, Gert-Jan Stads, Graham Thiele, Rob Tripp, Ludy Velasco, Stan Wood, and Di Zhang.

Others also warrant a special mention: Tim Dalton and Catherine Ragasa shared data with the DIIVA Project participants; Nienke Beintema hosted the DIIVA databases on the Agricultural Sciences and Technology Indicators (ASTI) initiative website; James Burgess contributed substantially to preparing the databases for distribution; and Tony Murray spent many hours tailoring software to their storage and distribution. In addition: Peter Gregory drafted a proposal for funding on DNA fingerprinting; Abdoulaye Adam oversaw the design of the sampling survey; Bea Amara managed all the travel-related requests from the project; Patti Sands organized the final workshop; and Brenda Zepeda was responsible for all the financial accounts and reporting of the project. Five reviewers – Derek Byerlee, Dana Dalrymple, Keith Fuglie, Doug Gollin and Tim Kelley – spent a lot of time vetting and making extensive comments on this Synthesis Report in draft; and Princess Ferguson worked on several project- related publications, including this one, which was edited and produced by Sarah Tempest and others in the Green Ink team (www.greenink.co.uk).

We acknowledge the following individuals from the CG Centers and their institutional partners for their invaluable collaboration in making this project a success:

AfricaRice – Patrice Y. Adegbola, Stella Everline Okello Adur, Daniel Andriambolanoro, Sekou Diawara, Lamin Dibba, Alioune Dieng, Sekou Doumbia, Chantal Ingabire, Dorothy Malaa Kenyi, Winfred A. Odenyo Kore, Nazir Mahamood; M. Hervé Nbedane, N. Robert Ngonde, Vivian T. Ojehomon, Mathieu Ouedraogo, Koku Domenyo Tsatsu, Alexander Nino Wiredu.

Bioversity – Edward Bbosa, Ronald Kakooza, Henry Kayondo, Ronald Mutebi, Betty Nakiyingi, Lydia Namayanja, Robert Ssebagala and Steven Ssengendo who assisted in data collection;

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — vii Dr Alex Barekye, Micheal Batte, Dr Deborah Karamura and Tendo Ssali scientists from the National BioResource Project, Bioversity and the International Institute of Tropical Agriculture (IITA); and the district agricultural and extension officers of various districts. Our thanks also go to the farmers interviewed in this study.

CIAT – Dr Rowland Chirwa, Prof. Paul Kimani and Dr Clare Mukankusi who are three regional breeders, along with bean program coordinators and national groups of experts from Democratic Republic of the Congo, Ethiopia, Malawi, Mozambique, Rwanda, Tanzania, Uganda, Zambia and Zimbabwe.

CIMMYT – Roberto Larovere for study start-up with Olaf Erenstein, Dan Makumbi, B. M. Prasanna and Bekele Shiferaw; and for country reports: Manuvanga Kiakanua (Angola), Lutta Muhammad (Kenya), Mahagayu Clerkson (Kenya), Alexander Phiri (Malawi), Zubeda Nduruma (Tanzania), Moses Mwesigwa (Uganda), Thomson Kalinda (Zambia) and Shamiso Chikobvu (Zimbabwe). Also Matthew Capelli at Virginia Tech, Stella Wambugo at Centro Internacional de La Papa (International Potato Center) along with Jean de Dieu Nsabimana, Jean-Claude Nshimiyimana, Josaphat Mugabo and Domitille Mukankubana at the Rwanda Agricultural Bureau.

CIP – scientists in potato and sweetpotato programs and officials in various non-governmental organization and extension services of Burundi, Ethiopia, Kenya, Malawi, Mozambique, Rwanda, Tanzania and Uganda, along with Maria Isabel Andrade, Merideth Bonierbale, Ted Carey, Paul Demo, Manuel Gastelo, Elmar Schulte-Geldermann, Marc Ghislain, Wolfgang Grüneberg, Stef de Haan, Diedonne Harahagazwe, Guy Hareau, Rogers Kakuhenzire, Imelda Kashaija, David Kipkoech, Juan Landeo, Jan Low, Everina Lukonge, Robert Mwanga, Obed Mwenye, Jean Ndirigwe, Jean Claude Nshimiyimana, Michel Ntimpirangeza, Jose Ricardo, Steffen Schulz, Ntizo Senkesha, Kirimi Sindi, Gorretie Ssemakula, Graham Thiele, Silver Tumwegamire and Gebremedhin Woldegiorgis.

ICARDA – The experts and their partners from the national research systems of Eritrea, Ethiopia and the Sudan whose contribution to this work deserves acknowledgment, include: the late Geletu Bejiga and his staff in the ICARDA-Ethiopia country office, Alemayehu Assefa, Adefris Chere, Gamal Elkheir, Tamer El-shater, Asnake Fikre, Stefania Grando, Abdelaziz Hashim, Muhammad Imtiaz, Musa Jarso, Gemechu Keneni, Shiv Kumar, Birhane Lakew, Fouad Maalouf, Mohammad Maatougui,Tamene Temesgen. We would also like to thank various teams at the Ethiopian Institute of Agricultural Research and the Holeta Agricultural Research Center for their logistical and administrative support, including the enumerators, data entry personnel and the cleaning crew, as well as the drivers and development agents without whom the project would not have been possible. Importantly, special thanks go to the Ethiopian farmers who volunteered their time to be interviewed.

ICRISAT – The experts and national partner institutions that participated in the interviews as well as those who provided useful support throughout the research, including: Dimphyna (Kenya), Albert Chamango and Geoffrey Kananji (Malawi), Stephen Lymo, Rose Ubwe and Mponda (Tanzania), Jasper Mwesigwa Batureine (Uganda) and Francisco Miti and Catherine Mungoma (Zambia). In West and Central Africa the main collaborators were: H. Ajeigbe, A. Amadou, O. Cisse, C.A. Echekwu, I. Faye, C.T. Hash, A. Ibro, A. Kassari, O. Kodio, A. Magagi, A. Minougou, A. Moutari, J. Naino, B.R. Ntare, M. Sanogo, A. Toure, M. Yeye and R. Zangre.

IITA – A number of people have contributed to the successful completion of the DIIVA study on modern variety release and adoption for IITA mandate crops in SSA. The authors would like to thank Jonas Mwalughali (late), Robertson Khataza and Makaiko Khonje for their outstanding research assistance. We also wish to acknowledge the contributions of the survey team leaders who coordinated data collection efforts and expert consultations in various countries – Adebayo Bamire, Tafireyi Chamboko, Hobayo Claude, Bisase Dennis, Mossi Illiassou, Eliya Kapalasa, Toussain Kendenga, Luke Olarinde and Asman Wesonga.

viii — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project We are particularly grateful to the national program and private sector leaders and researchers from over 20 countries in SSA who provided a great deal of information used in the DIIVA study through extensive consultations and discussion as well as review of institutional documents and records including:

Angola – Moniz Mutunda, Instituto de Investigação Agronómica.

Benin – Aly Djima, Romuald Dossou, Yallou Chabi Gouro, Abou Ogbon and Kodjo Siaka, Institut National des Recherches Agricoles du Bénin; and J. Detongnon, Société Nationale pour la Promotion Agricole.

Burkina Faso – Issa Drabo, Institut national d’Etudes et de Recherches Agricoles; and Jacob Sanou, Institut de l’Environnement et de Recherches Agricoles.

Burundi – Alexandre Congera, Institut des Sciences Agronomiques du Burundi.

Cameroon – Nguimgo Blaise, Zonkeng Celicard, Njualem Khumbah, Tandzi Liliane, Eddy Mangaptche, Ndioro Mbassa and Meppe Paco, Institute of Agricultural Research for Development.

Côte d’Ivoire – Louise Akanvou, Philippe Gnonhouri, Amani Michel Kouakou, Francois N’gbesso, Boni N’zue and Siaka Traoré, Centre National de Recherche Agronomique.

Democratic Republic of the Congo – Bambala Emmanuël, Mbuya Kankolongo and Lodi Lama, Institut National d’Etude et de Recherches Agronomiques.

Ghana – Joe Manu-Aduening, K. Obeng-Antwi , Henry Asumadu, Manfred Ewool and Emmanuel Otoo, Crops Research Institute; Cecil Osei, Root and Tuber Improvement and Marketing Program; and Kwabena Acheremu and Adjebeng Joseph, Savanna Agricultural Research Institute.

Guinea – Diallo Amadou, El Sanoussy Bah, Cheick Conde, Camara Fadjimba, Sekouba Korouma, Diallo Pathé and Camara Sékouna, Institut de Recherche Agronomique de Guinée.

Kenya – Joseph Kamau, Kenya Agricultural Research Institute.

Malawi – M.C. Dzinkambani and Vito Sandifolo, Department of Agricultural Research and Technical Services.

Mali – Ntji Coulibaly and Mamadou Touré, Institut d’Economie Rurale.

Mozambique – Anabela Zacarias, Agricultural Research Institute of Mozambique; Rogerio Cheulele, Universidade Eduardo Mondlane.

Niger – Moutari Adamou, Institut Nationale de Recherches Agronomiques du Niger.

Nigeria – Paul Ilona, CIAT-Nigeria; Hakeem Ajeigbe, ICRISAT-Nigeria; U.S. Abdullahi, Institute of Agricultural Research; S.R. Akande and S.A. Olakojo, Institute of Agricultural Research and Training; Shokalu Olumide, National Cereals Research Institute; S.O.S. Akinyemi, National Horticultural Research Institute; Njoku Damian, National Root Crops Research Institute; M.A.B. Fakorede, Obafemi Awolowo University; and Ibikunle Olumide, Premier Seed Nigeria Limited.

Senegal – Ndiaga Cisse and Abdou Ndiaye, Institut Sénégalais de Recherches Agricoles.

Tanzania – Edward Kanju, IITA-Tanzania.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — ix Togo – Akihila Didjeira, Etoudo N’kpenu, Komi Somana and Béré Tchabana, Institut Togolaise de Recherche Agronomique.

Uganda – Elvis James Mubiru, IITA-Uganda; and Godfrey Asea, Antony Bua and Nelson Wanyera, National Agricultural Research Organisation.

Zambia – Martin Chiona, B.L. Kaninga and Laston Milambo, Zambia Agriculture Research Institute.

Zimbabwe – Patrick Nyambo, Department of Research and Specialist Services; and Caleb M. Souta and Jacob Tichagwa, Seed Co Limited.

Finally, we wish to dedicate this report to Bonny Ongom who was a Global Information Systems (GIS) research assistant at CIAT in 2006–2011. Bonny was a key collector of in-country data for DIIVA objective 1, which he obtained by canvassing experts in project workshops. A master at facilitating participatory mapping exercises – a key method for eliciting information on bean production from expert panels – he also made an invaluable contribution to the second edition of the CIAT Bean Atlas. Bonny’s death was a blow to the DIIVA project.

x — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Foreword

Introduction leadership of Bob Evenson and drawing on the work of numerous collaborators, this For fifty years or so, development econo- study compiled data on the diffusion of mists have been concerned with tracking improved varieties of eleven food crops, the diffusion of improved agricultural tech- and it attempted to achieve global nologies in the developing world. This focus coverage. The project included three is not based on mere curiosity. One reason country case studies and several cross- for documenting diffusion is that it cutting analyses and modelling exercises. A provides a simple measure of the success of book (Evenson and Gollin, 2003) summa- agricultural research: when new crop varie- rized the main findings of the project and ties are taken up, or when new agronomic established a 1998 baseline for crop varietal practices are adopted by farmers, it adoption and diffusion data. provides information about the effectiveness of the research. A second reason for This report represents the first major documenting diffusion is that the resulting follow-up of the Evenson and Gollin data can be used to shed light on the multi- baseline. The following pages summarize dimensional impacts of the research – on the key findings of the DIIVA project (Dif- productivity, on farm income, even on fusion and Impact of Improved Varieties in poverty and inequality. A third reason for Africa), which was funded by the Bill and documenting diffusion is that the differen- Melinda Gates Foundation with the goal tial patterns observed across space and time of assessing incremental progress in Sub- can reveal underlying constraints or Saharan Africa in the years after 1998. The problems with technology diffusion: DIIVA project (and a companion project perhaps certain technologies fail to gain a focused on South Asia) have greatly foothold in particular agro-ecologies, or advanced our knowledge of varietal perhaps practices beloved by researchers adoption and diffusion, both by expanding have failed to spread widely. This informa- knowledge about areas where diffusion tion can feed back into the research process was previously not well documented and to inform scientists and shape further by improving the methodologies used for research. Indeed, information on diffusion measuring diffusion. can also inform the broader development community and can shape thinking about a The DIIVA project was organized around wide set of potential constraints to three distinct activities: documenting key adoption – resulting, perhaps, from failures performance indicators of crop genetic in financial markets, extension and infor- improvement, collecting nationally repre- mation, or simply reflecting high transport sentative survey data on varietal adoption, and transaction costs. and assessing the impact of varietal change. This synthesis paper reports on progress in Efforts to document the diffusion of the first two areas. A companion report in- improved crop varieties date back to the terprets and summarizes the results of groundbreaking work of Dana Dalrymple three related case studies on the impact of (1969, 1978, 1986a, 1986b). Dalrymple’s modern varieties in Sub-Saharan Africa. work drew on the cooperation of national research programs and international scien- The DIIVA report covers 20 crops and 30 tists, and it provided the data on which countries in Sub-Saharan Africa. Because were based many early analyses of the some crops are locally absent or unimpor- Green Revolution and its impacts. But for a tant, the report does not account for every variety of reasons, the important task of crop in every country; but coverage extends documenting diffusion was left to languish to 152 crop–country combinations that after Dalrymple’s last effort in 1986; the together account for over 70% of the gross next major effort to document diffusion value of agricultural production in Sub- came more than a decade later. Under the Saharan Africa.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — xi The report’s findings represent a major weighted average across 20 crops).1 Calcu- achievement in terms of both the scope lated another way, the annual growth rate and quality of data for Sub-Saharan in the adoption of MVs was 1.45% per Africa. In the Evenson-led study of 2003, annum over this period.2 This in itself is a the available data on varietal adoption remarkable achievement for agricultural and diffusion in Africa were very limited. research. Although one can still ask ques- Many of these data were based on a tions about the quality of the data, the combination of small-scale studies of DIIVA study provides important evidence adoption and rather vague regional esti- that agricultural research is continuing to mates; the specific crop-country estimates provide technologies of value to farmers. of varietal adoption were mostly the Technology adoption is, in some sense, a product of interpolation and triangula- logically sufficient measure of impact; tion. The current study has improved farmers would not use these technologies if enormously the quality of the evidence. they did not provide some advantage. In comparisons of adoption estimates between 1998 and 2010, it is important In the report that follows, Walker et al. to note that the new data is of substan- argue that the extent of diffusion of MVs is tially higher quality than the old data. one of the most important measures of the Thus, changes in the adoption estimates productivity, food security and poverty may simply reflect improvements in data benefits generated by investments in crop quality, as opposed to changes in the genetic research and development. We underlying patterns of varietal use. agree, in part, but we might add some important qualifiers. The extent of area under We note that the entire database for the MVs is only one of several changes that DIIVA study is publicly available, with full interact in complex ways. The pathway documentation, on the Agricultural from adoption to impact depends on a Sciences and Technology Indicators (ASTI) wide range of other factors such as fertilizer website (http://www.asti.cgiar.org/diiva). use, policy changes and road access. Docu- We encourage readers and researchers to menting impacts on poverty and food visit the website and to make use of the security is extremely challenging. data. In addition to the data on the adoption of modern varieties (MVs), the Nevertheless, the continued growth in area database includes observations on under MVs indicates that research is contin- varietal releases for each crop-country uing to provide farmers with useful combination and data related to the technologies – and that farmers are contin- number of full-time equivalent scientists uing to find ways to take up these new engaged in crop improvement research. technologies, in spite of the constraints that This will provide a benchmark at the they face. Of course, there are crop-country level of individual countries and crops so combinations where adoption of MVs is still that specific crop-country combinations quite low – 14 of the crops are character- can be tracked and analyzed over time. ized by a mean adoption rate below 35%. This, of course, assumes a comparable It will be important to analyze the factors effort will be sustained over time at regular intervals so that progress can be 1 If we look only at the paired comparison of 61 crop-by- assessed. country observations for the 10 continuing crops, area- weighted adoption was 27% in 1998 and 44% in 2010. 2 There are a number of qualifiers that must be kept Key findings and implications in mind when making comparisons here, given that the number and types of crops and crop-country combinations varied between the two periods and that Arguably the most significant finding of the methods used to elicit expert opinion were not this report is the impressive growth always consistent over the periods. Nevertheless, while achieved in terms of the share of cropped the confidence interval may be large – perhaps more so for the earlier survey result when less scrutiny was area now under MVs in Sub-Saharan Africa. applied to the method for eliciting expert opinion, there In 1998, about 20–25% of cropped area was is no reason to believe that there is a particular upward under MVs (based on a weighted average or downward bias in these different period estimates. All across 11 crops). By 2010, this figure had one can say is that the study is using the best available data, and the methods used to collect those data are grown to 35% (based on a similarly documented in the reports.

xii — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project that have limited adoption rates for these methods is closer to the truth. On the one crops. Conversely, there are crop-country hand, nationally representative household combinations that have already achieved a surveys might be presumed to be more relatively high (for Africa) level of MV reliable than expert opinion, since they are adoption (soybean, wheat, maize, cassava, based on data collected from individual rice) or where adoption has been quite farm households. On the other hand, there rapid – cassava, barley and maize doubled may be gaps in coverage (e.g. because of their share over this period. Here, too, the low probability of sampling from large there may be lessons to be learned. But an commercial farms). Moreover, the quality of important point to note is that, whether the data obtained from household respond- the 1998 base levels were relatively low or ents may not be higher – in many settings, high, over 90% of the crop-country obser- it is not clear that farmers can accurately vations experienced a rise in MV adoption identify the varieties, and the vernacular between the two studies. The notion that names that they assign to particular varie- African crop farming is stagnant is not sup- ties may make identification difficult.3 ported by the data from this study. Taken together, we conclude only that Over time, as the level of MVs approaches further research is needed to reconcile the full adoption, other measures of the success discrepancies between expert opinion data of crop improvement programs – in particu- and survey data on varietal adoption. It lar, the velocity of varietal change – will would be valuable to know whether there become more relevant. Even now, for many are consistent patterns that would allow us crops, this is an important measure of to predict which approach is more accurate success. The DIIVA study team looked at this for a particular crop-country combination. and found the area-weighted mean age of This is certainly an area worthy of further varieties in the field was 14 years across all analysis and research. SPIA is currently con- crops – not much change from the earlier ducting research to establish cost-effective period. More analysis is clearly needed here and reliable methods for measuring to understand the causes of this. Some adoption, using DNA fingerprinting as a older ‘modern’ varieties are proving to be benchmark to assess the accuracy of alter- remarkably robust in the face of many new native methods. varieties being released – or, alternatively, recent research has not always succeeded in Given that expert panel surveys are likely to producing genuinely useful technologies. remain a major source of data in the future when conducting large-scale adoption How reliable are the estimates of adoption surveys, there are valuable lessons to be emerging from this study? Is there any way learned from the report’s observations con- to measure their accuracy? These questions cerning how best to conduct expert surveys occupied the DIIVA project at every stage. (p. 37).4 These lessons should not be lost in By necessity, the DIIVA data largely draw on the vast array of data generated by this judgments made by expert panels. This remains the dominant method for estimating crop area under MVs at a large scale, 3 For instance, farmers may use the same name for distinct varieties, and they may use different names for due to the cost and complexity of collecting the same variety. data on varietal diffusion through other 4 In general, more effective elicitation was characterized means. Thus, the DIIVA study relied primar- by: ily on expert panel judgments (for 115 • close and intensive supervision of CG project-related crop-country combinations). In a number of staff cases, however, these expert data were • organization of, and attendance at, time-bound supplemented by estimates based on workshops with direct interaction with expert panel members household surveys (for 36 crop-country • greater spatial resolution in the elicitation of combinations). It was possible to compare estimates that were subsequently aggregated to these two methods for 18 observations. Of regional and national levels these, ten lined up reasonably well, but • including more members from the informal sector household survey estimates were lower for and from non-governmental organizations with eight observations. Unfortunately, there is geographic-specific expertise in technology transfer on the panels, and feedback from CG Center no easy way of knowing which of the breeders in the final stages of the process.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — xiii study. Indeed, a major effort to expand this high, have crop yields reached levels that database to Asia has already drawn on the might be viewed as satisfactory? If adop- lessons of the DIIVA study. tion in some crop-commodity combinations is nearly complete, and if yields are still low, what should we conclude? Is Issues emerging and future this evidence that crop genetic improve- directions for research ment is a weak tool in the Sub-Saharan context? Or should we expect that suc- cessive generations of improved varieties The synthesis report that follows is only a will increase yields where previous gen- descriptive summary of the data. We hope erations have failed? Or should we that many researchers will take advantage simply accept that high rates of adoption of the DIIVA data to construct estimates of provide sufficient evidence that productivity and impact, and we also hope improved varieties are useful, even if this for a lively conversation over the key is not manifested in crop yields? messages to be taken from the data. What can we learn from the patterns of diffusion that might inform the research The main results raise a number of issues process? What characteristics seem to be that deserve further exploration. Some are associated with high levels of take-up? easily answered. Others will require new How can we learn from the DIIVA study methods – or perhaps may be so challeng- to target future research more effectively? ing that they simply invite speculation. For instance: The need for continued data Is Africa finally experiencing a Green collection and analysis Revolution? If so, does Africa’s experience look like the Green Revolutions of Asia and Latin America and the The DIIVA study represents a major contri- Caribbean? Arguably, we are seeing dif- bution towards measuring and understand- fusion of MVs without seeing much ing the diffusion of modern crop varieties. intensification of accompanying inputs. The value of the study serves as a reminder In Asia, the spread of MVs was linked to of the importance of collecting similar data far greater use of fertilizer and mechani- on a regular basis – and of expanding the zation; but in Africa, the growth of these coverage across geographic areas. In the inputs has been much slower. long run, varietal adoption and diffusion Does yield growth in Sub-Saharan Africa data should ideally become a regular com- seem to match the diffusion of MVs? Do ponent of national agricultural statistics – we see substantial yield increases in the collected, for example, as part of national crops and countries where we see corre- agricultural censuses. In the short run, spondingly large increases in adoption? however, this task remains in the purview This seems like an important question to of research institutions such as the CGIAR ask, but perhaps a difficult one. A key and its partners. SPIA continues to support challenge is that, by many accounts, crop the collection of diffusion data and to yield data are very poor in quality. It is promote the institutionalization of data not clear whether many countries in collection. Africa conduct regular yield surveys based on crop cuts. Even theoretically, it Among the activities that SPIA is currently is possible that the diffusion of improved engaged in, as of mid-2014: varieties need not be accompanied by an increase in yields; for instance, a new With numerous partners, SPIA is current- trait (e.g. cold tolerance) might allow for ly working to pioneer and validate new crop area to expand along an extensive ways of measuring varieties in use, with margin where yields are lower. This the hope that these approaches can be could, in principle, result in a decline in incorporated routinely into micro-studies average yield. and household surveys. A related question: In the crops and SPIA is working to collect and report commodities where adoption levels are varietal adoption data from Asia.

xiv — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project We are looking to expand the set of the DIIVA project to fruition through their technologies for which adoption and dif- extraordinary efforts. fusion data are collected; specifically, we seek to extend the data to include obser- First and foremost, we were exceptionally vations on improved agronomic practices fortunate to have Tom Walker leading this (e.g. conservation agriculture); irrigation effort on behalf of SPIA. Tom was perhaps technologies; livestock technologies and uniquely qualified to lead this effort, on practices; and a range of other changes the basis of his long and distinguished that can potentially be linked to CGIAR record of research on agricultural technol- research. ogy adoption and its impacts. Not only did Tom effectively manage this large and In this sense, we think it is important that complex multi-partner undertaking, but he the DIIVA project be viewed as part of an also provided expertise at every stage of ongoing set of research activities designed the study. He provided crucial insights into to reveal the continuing diffusion of agri- methods of collecting varietal data – from cultural technologies, broadly defined. experts, from farmers, and from farm com- Much remains to be done, and SPIA munities. Tom’s careful probing and his welcomes partners and researchers who efforts to check and validate the data drew bring new approaches and ideas. on his deep and detailed knowledge of African agriculture. We are enormously grateful to Tom Walker for his leadership SPIA Chair’s acknowledgments and expertise; without him, the project could not possibly have achieved such a As will be apparent from this foreword and high-quality outcome. the document that follows, the DIIVA project involved a major undertaking. Any Perhaps no one was more important to the project of this size necessarily involves a conceptualization and completion of the team effort. In this case, the team was DIIVA study than Greg Traxler, Program large, including researchers at seven CGIAR Officer of the Gates Foundation. Along Centers and numerous national partner with Prabhu Pingali (who was based at the institutions. The acknowledgments section time at the Gates Foundation), Greg urged of this report lists the full cast of partici- the CGIAR to push ahead with a new effort pants, but I would like to take this opportu- to collect data on varietal diffusion – and nity to thank, on behalf of SPIA, all of those he then helped to mobilize the funding for who contributed time and effort. the project. Greg’s contributions went far beyond his role as a conduit for funding. The project depended in the final analysis Over the course of several years, Greg asked on the efforts and expertise of many persistently about the scope and quality of researchers based at CGIAR Centers and in a data and pushed to set a high standard for range of national research institutions the study. across Africa. We are grateful to the hundreds of scientists who contributed Another key figure in the history of the their time to this effort – whether through DIIVA project was my predecessor as SPIA participating in panels or filling out surveys Chair, Derek Byerlee, who has remained a or providing their field notes, based, in key participant throughout the duration of some cases, on years of data collection. The the project. Like Tom Walker, Derek brings detailed field knowledge of scientists was an encyclopaedic knowledge of African ultimately one of the main sources of data agriculture, based on years of fieldwork for the DIIVA project. We are grateful to all and personal experience in most of the these scientists for their generosity in countries covered by the DIIVA study. As a sharing time and for their desire to provide dedicated social scientist of the highest thoughtful and objective information caliber, Derek played a central role in the about patterns of adoption and diffusion. design and implementation of the study. My own term as SPIA Chair started as the Beyond this collective effort, however, I DIIVA project came to a close, so Derek was want to single out the outstanding contri- at the helm of SPIA for almost the entire butions of several individuals who brought duration of the project.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — xv Finally, two members of the SPIA Secretariat and substantively in the DIIVA study. As a staff – Tim Kelley and James Stevenson – member of the Project Steering Committee deserve special recognition for their contri- for DIIVA, James participated in every stage butions to the project. Tim Kelley’s role of the project; SPIA is fortunate to be able cannot be easily described. As the Head of to draw on his skills as a researcher and his the SPIA Secretariat, Tim played a key ad- thoughtful analysis. ministrative role in managing the study. But Tim’s first-hand knowledge of the CGIAR, In closing, I would like to honor the memory based on some thirty years as a researcher of Bob Evenson, who died in February 2013. and research manager, was ultimately of Bob’s career-long efforts to document the enormous importance in the quality of the diffusion and impact of agricultural technol- DIIVA project and its findings. I think it is ogies grew out of his passionate belief that no exaggeration to say that Tim read every science had the potential to improve the sentence produced by the DIIVA project; his lives of the poor and of rural people. His critical eye and high standards were illness prevented Bob from taking part in the matched by his constantly positive outlook. planning of the DIIVA project, but I have no Tim played a similar role in shepherding doubt that he would have been delighted and reviewing the earlier Evenson-led and impressed by the work that has been study, and this provided him with a done – and eager to see it continued valuable long-term perspective on the through the future. DIIVA study. In both cases, Tim’s contributions proved enormously valuable. Douglas Gollin Also at the SPIA Secretariat, James Stevenson Chairperson, SPIA has played a key role both administratively University of Oxford, UK

xvi — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Executive summary

By the 1970s, farmers began to benefit The raw material for the descriptive analysis from recently bred varieties in several has been drawn from three databases: primary and secondary food crops in Sub- (1) recent cross-sectional data on the Saharan Africa (SSA). In the late 1990s, a strength of human resources in national global impact assessment estimated that agricultural research systems (NARS) by modern varieties (MVs) accounted for only discipline; (2) historical data on varietal 22% of the growing area for most primary release; and (3) recent cross-sectional data food crops across SSA (Evenson and Gollin, on cultivar-specific levels of adoption 2003a; Maredia and Raitzer, 2006). estimates elicited mainly from expert panels and, to a lesser extent, from nationally The estimates reported by Evenson and representative surveys. The unit of Gollin (2003a) were based on partial results observation for the three datasets is a crop- and limited data for a number of crops and by-country combination. countries. As a result, the picture of MV adoption in SSA was unclear and fragment- In Section 7, inferences are examined ed, and over the last decade no comprehen- from the perspective of the Millennium sive study has updated or clarified these Development Goals (MDGs). In particular, estimates. the feasibility of measuring the adoption of MVs in the DIIVA Project provides an Here, the baseline established by Evenson opportunity to redress one of the deficien and Gollin (2003a) has been updated, cies of the MDGs, which have been criticized widened and deepened. We report on the for neglecting agriculture (Eicher, 2003). results of a CGIAR project: Diffusion and Impact of Improved Varieties in Africa The Project’s databases contain about 150 (DIIVA) Project – the first major study to crop-by-country observations, selected to focus on the diffusion of improved crop cover the most important food crops in the varieties in SSA. Supported by the Bill & main producing countries. Twenty crops Melinda Gates Foundation (BMGF), seven and two large maize-producing regions CGIAR Centers and their partners carried result in 21 crop categories. The area out adoption research and impact harvested within the 20 study crops in SSA assessments as part of DIIVA, which was totals around 140 million hectares. These directed and coordinated by CGIAR 20 primary and secondary food staples Independent Science & Partnership make up about three-quarters of the total Council’s (ISPC) Standing Panel on Impact crop area in SSA including annuals and Assessment (SPIA) and administrated perennials. Overall, 83% of the area of the through Bioversity International. included crops in SSA is covered in the 21 crop categories. For 62 observations, data This work has been driven by three are available for comparative analysis complementary activities that respond to between 1998 and 2010 on the scientific three project objectives: (1) documenting strength, varietal output, and adoption of the key performance indicators of crop MVs. genetic improvement; (2) collecting nationally representative survey data on The national scientific capacity of the varietal adoption; and (3) assessing the 150 crop improvement programs examined impact of varietal change. This synthesis in the DIIVA Project totals 1,289 full-time paper reports on progress in the first two equivalent (FTE) scientists. But the actual areas: documenting the performance of number of scientists who work in these crop improvement in SSA and validating programs is likely to be more than double estimates from expert panels with results this sum. In rice, for example, 125 FTE from nationally representative surveys on scientists equates to 289 researchers, the diffusion of MVs. because only about 25–30% of these

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 1 scientists commit 75–100% of their time to program. But in future an estimated lower rice research. More scientific resources are number of Bachelor of Science (BSc) holders allocated to maize than to any other crop in WCA is a cause for concern because in SSA. Cassava is a distant second to maize. fewer younger scientists are available for mentoring by older, experienced scientists. Of the 20 crops, cassava, yams, and pearl millet consistently rank at the bottom of Nevertheless, the overall number of scien- the charts on research intensity. Relative to tists with PhDs and Master of Science (MSc) their area, production, and value of qualifications is encouraging. Only 24 of production, all three of these semi- the 135 crop improvement programs do not subsistence food crops appear to be have a PhD presence. Only four programs extremely short on research resources. In have neither a PhD nor an MSc scientist terms of harvested area, groundnut and involved directly in their research. More sorghum are also characterized by very low than half of the programs have at least research intensities. 1.0 FTE PhD scientist working in research.

Results on the differences in scientific With regard to crop type and the pattern of strength over time are mixed. Between disciplinary research resource allocation, 1998 and 2010, more programs have gained the main distinction centers on roots and scientists than have lost researchers. tubers on one hand and cereals and grain However, because of rising levels of crop legumes on the other. Root and tuber production, mainly attributed to area programs invest considerably less in plant expansion, estimates of research intensity breeding per se but more in closely allied have not increased and have even declined disciplines such as tissue culture. Molecular for most of the 65 programs that have biology only accounts for 3.4% of the mean information available to carry out paired resources across the 150 programs in the comparisons. database. This 3.4% is equivalent to only 40 FTE scientists, 17 of whom are involved Comparing these findings to the 2010 in studies of banana in Uganda. results highlights several transparent differences. Nigeria, for example, has invested Varietal release, interpreted broadly to significantly in maize research, while its sci- include improved materials that are or are entific capacity in rice and cassava has also supposed to be available to farmers, is improved. But by far the largest increase in equated to output in this research. The his- scientific capacity has occurred in maize torical data on varietal release across the 20 across East and Southern Africa (ESA), crops approaches 3,600 entries. About 90% thanks largely to the dynamism of the of these have information on the year of private sector in this region. Notably, the release. Maize leads all crops with over comparisons also strongly suggest that 1,000 entries. Rice is a distant second. Both larger public sector crop improvement rice and maize in ESA have benefited from programs may be highly susceptible to multiple institutional sources of modern downsizing in times of financial crisis or genetic materials. By contrast, low research when donor support ends. intensities in pearl millet, sorghum, and cowpea in West Africa have translated into Concerns in scientific capacity in national low output intensities. programs in West Africa reflect not only a problem of relative numbers but also of sci- About 45% of 3,194 dated entries had been entist age. About 65% of the scientists released since 2000. The mid-point date for working on sorghum, pearl millet, and varietal release was 1998. Decade by groundnut in the five project countries in decade, the incidence of release has in- West Africa were older than 50 in 2010. creased steadily over time. Varietal output rose exponentially in maize in ESA between Scientists engaged in crop improvement the 1990s and the 2000s because of surging across West and Central Africa (WCA) private-sector releases. However, not all appear to be more highly educated than crops in all countries fit the pattern of a their ESA counterparts, with around 2.6 steady rise in varietal output over time. Doctor of Philosophy (PhD) holders per Between one-fifth to one-quarter of the

2 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project 146 crop-by-country observations were International Agricultural Research Centers characterized by more releases in the 1980s (IARC)-contributed genetic materials. The than in the 2000s. Some plausible explana- IARC-related share in adoption is about tions for declining varietal output centered 20% higher than its 45% contribution to on civil war, such as in Sierra Leone; the released varieties. strength of several crop improvement programs in Nigeria in the 1980s; and the The problem of lagging countries was also weak scientific capacity of the recent past evident in the cross-sectional adoption esti- in grain legume and coarse cereal improve- mates based on 152 crop-by-country obser- ment programs in West Africa. vations. Adoption of MVs was uniformly low in Angola, Mozambique, and Niger About 43% of the varieties released since across all crops. 1980 are related to the work of the CGIAR – a proportion that has remained relatively Spill-over varieties were prevalent, but, stable (40–45%) over the past three unlike in South Asia, so-called mega- decades. In total, the CGIAR contribution varieties that claimed millions of hectares was greater than 40% for 14 crops. Viable of arable land were not found. Of the over alternative international suppliers, a 1000 improved varieties listed in the DIIVA dynamic private sector, strong NARS, and adoption database, SOSAT C88, a short- failed breeding strategies figure prominent duration pearl millet variety, was the most ly as reasons why the CGIAR share is below extensively cultivated on just over one 40% for banana, beans, barley, field peas, million hectares in Nigeria, Mali, Niger, and maize, and sorghum. Burkina Faso. In maize, Obatanpa, derived from quality protein maize (QPM) The evidence was mixed for the develop- materials, and TZEE-Y, an International ment of crop improvement programs. A Institute of Tropical Agriculture (IITA)-bred few programs followed a transition over variety, fit the description of spillover time that reflected increasing sophistication varieties that have crossed over the borders in plant breeding research. These programs of several countries in WCA. The incidence initially relied on landrace materials for of spillover varieties appears to be higher in varietal release. Subsequently they engaged WCA than in ESA and in groundnut than in in multi-locational adaptation trials of in- other crops. troduced elite materials. Finally, they selected varieties from their own crosses or The paired comparison of 61 crop-by-coun- introduced progenies. But, for most, the try observations for the 10 continuing crops evidence was fuzzy or not transparent that showed that area-weighted adoption was programs had advanced in plant breeding 27% in 1998 and 44% in 2010. The 95% capabilities. More specifically, few released confidence interval for the 17.6% gain in varieties traced their origins to crosses adoption was 12.3% to 22.9%. Over 90% of made by national crop improvement the observations experienced a rise in MV programs. This negative finding stands in adoption, which increased at a rate equiva- contrast to the evidence that plant lent to a linear annual gain of 1.45 percent- breeding capabilities increased steadily over age points over the 13-year period. With time in Asia and Latin America (Evenson the exception of rice and potatoes, all crops and Gollin, 2003a). experienced an expansion in the use of MVs. Uptake was especially robust in barley, The area-weighted grand mean adoption cassava, and maize with adoption levels level of improved varieties across the 20 more than doubling during the period. Civil crops is 35%. The distribution of adoption war and changing methods in measuring of improved varieties is skewed as 14 of the adoption loom large as plausible explana- crops are characterized by a mean adoption tions for why improved varieties lost level that falls below 35%. Crops with an ground in a few countries and a few crops. estimated adoption performance superior Crop-by-country observations with a low to the overall average included soybean, level of MV adoption in 1998 were more wheat, maize, pigeonpea, cassava, and rice. likely to experience positive outcomes in About 23% of the 35% – i.e. a share of 65% adoption than those with moderate levels – of MV adopted area is related to of adoption in 1998.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 3 The velocity of varietal change is as becoming older. For rice, the average age important as the adoption level of MVs in of MVs was the highest of the four crops in assessing plant-breeding performance both 1998 and 2010. especially for countries approaching moderate to full adoption. Varietal About one-third of the resources of the turnover is measured by the age of DIIVA project were invested in nine nation- improved varieties weighted by their area ally representative adoption surveys that in production. were designed to validate the estimates from expert panels and provide raw The area-weighted mean age was 14 years, material for impact assessment. In carrying indicating that the average MV in farmers’ out the process of estimate elicitation from fields was released in 1996. The average a standardized protocol, participants also age-related results by crop were tightly generated considerable anecdotal evidence clustered in the range of 10–20 years. This on what worked. The protocol was adapted means that there may be few, if any, crops to regional and crop-specific circumstances where older adopted improved materials that featured a considerable amount of have substantially eroded the profitability ‘learning by doing’ by CG Center staff con- of plant breeding. But, by the same token, ducting the expert panels. In general, more there was also little evidence that rapid effective elicitation was characterized by: varietal change is taking place. Some crops and important producing countries are close and intensive supervision of CG characterized by older than expected project-related staff; improved varieties. For a new expanding organization of and attendance at time- crop, the youngest soybean varieties in bound workshops with direct interaction farmers’ fields in Nigeria are ‘old’ as they with expert panel members; were released in the early 1990s. greater spatial resolution in the elicitation of estimates that were subsequently Sixteen crop-by-country programs aggregated to regional and national scored well on both varietal adoption levels; and turnover. These better performing including more members from the infor- crop-by-country entries combined larger mal sector and from non-governmental area programs in maize, cassava, and organizations (NGOs) with geographic- cowpea with several smaller programs in specific expertise in technology transfer soybean and rice. on the panels; and feedback from CG Center breeders in the The largest area and value shares came final stages of the process. from varieties that were released in the late 1990s, suggesting that CGIAR Centers were Lessons on what did not work were able to supply materials for release by their transparent. The CG Center that relied NARS partners during a time of financial solely on NARS scientists as consultants to crisis. The 15% value share for varieties carry out expert elicitation was only able released in 2006–2011 is encouraging and to provide quality cultivar-specific adoption indicates that materials continue to find a estimates for two of their 14 assigned home in farmers’ fields. Materials released crop-by-country observations. Much more prior to 1980 in the early years of the intensive supervision was needed. CGIAR were comparatively limited and their impact has eroded over time. In contrast, Survey estimates and those from expert those produced in the 1980s account for opinion panels were within ten percentage more than 20% of the area and value points for ten of the 18 observations resulting from MV adoption. suitable for validation. Survey estimates were lower for the other eight observa- Comparing the 1998 estimates to those in tions, and for two of these they were 2010 suggests that improved varieties are markedly lower. Ignoring these two not getting any younger in farmers’ fields. outliers, survey estimates were about seven- For maize and wheat, age is roughly the eighths the size of expert elicitations. same as 12 years ago. For three of the four Therefore, the mean estimate of 35% for countries producing potatoes, varieties are MVs for SSA as a whole is likely to be over-

4 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project estimated because the majority of estimates for economic development in general and came from expert opinion. Applying the agricultural development in particular. seven-eighths finding from the validation Based on past performance, a target of exercise gives a more conservative estimate about 50% can be projected for MV of about 31% for MV adoption if surveys adoption by 2020. had replaced expert opinion panels. Realizing outcomes consistent with this Higher mean absolute percentage errors solid trend in performance will require between the two sources did not seem to several changes. Lagging crops and/or be associated with variations in the lagging countries or both were identified elicitation approach or specific to a crop as an area of concern, particularly for or country. They had more to do with the sorghum, pearl millet, and groundnut in extenuating circumstances of rapid change West Africa where scientists’ advancing age or disruption associated with rampant over- and low levels of research intensity are optimism about the prospects for large important issues that threaten enhanced technology transfer efforts and with civil varietal output and adoption. Efforts will war that also can be devastating for the have to be redoubled in West Africa if a applicability of prior knowledge in goal of 50% coverage in improved varieties circumstances where confirmation is for SSA is to be attained. Some thinking out difficult. of the box may be required to address the problem of stagnating and eroding Slightly over 70% of the mean adoption scientific capacity in several West African estimates in the national surveys were crop improvement programs on coarse composed of MVs for which the panel held cereals and grain legumes. Performance positive adoption beliefs; the other 30% also has to improve in Angola and came from unnamed or other named Mozambique, the only two countries in materials believed to be MVs. The size of southern Africa where adoption of MVs is the second component varies from survey low for all primary and secondary food to survey, but it is usually sizable as there crops. is always a leftover quantity of MV area that cannot be assigned to a specific The descriptive research in this paper clearly cultivar. For this reason, the summed area does not begin to exhaust the relevant of specific MVs will typically be less than an themes that can be tackled using the DIIVA aggregate adoption level. Surveys are likely 1998 and 2010 databases. The relationships to understate the importance of specific among scientific capacity, varietal output improved cultivars; detailed estimates from and varietal adoption need to be explored expert opinion that feature few, if any, in a more rigorous analytical manner. The varieties in a residual ‘other’ category are data can be integrated with other datasets likely to overemphasize the uptake of to shed light on recent tendencies in, and specific MVs. Accuracy in survey estimates determinants of, productivity growth in depends heavily on whether or not African agriculture. Indeed, documenting numerous regional- and location-specific productivity growth from adoption is a names can reliably be assigned to specific possibly more challenging proposition than varieties. documenting adoption.

This synthesis of the first two objectives of Most of the above research priorities are the DIIVA Project is laden with numbers. easy to visualize. Years from now they Some are more important than others. We should pale in comparison to output estimate that MVs of food crops in SSA in stimulated by making the database 2010 covered slightly more than 35% of the accessible to the public. Analytically, this total harvested area. In the period 1998– study does not break new ground. Its 2010 the uptake of MVs increased at novelty and value stems from its wide around 1.45% per year. scope in terms of crops and countries with intensive data collection, using the same Incorporating the adoption of improved protocols with an emphasis on validation. varieties into the MDGs would be desirable

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 5 Introduction

In 1978, Dana Dalrymple completed the global initiative on the impact assessment sixth edition of his life’s work: chronicling of varietal change estimated that MVs the development and spread of the High- accounted for only 22% of the growing Yielding Varieties (HYVs) of wheat and rice area of most primary food crops across SSA in developing countries. These semi-dwarf, (Evenson and Gollin, 2003; Maredia and short-duration varieties had entered Africa Raitzer, 2006). as early as the late 1960s. Dalrymple estimated that the diffusion of modern The estimates reported in Evenson and rice varieties had reached 4% by 1978. He Gollin (2003a) were based on partial results included 15 rice-growing countries in his with limited data available for a number of assessment that was based mainly on direct crops and countries. As a result, the picture communication with in-country scientists of MV adoption in SSA was somewhat fuzzy working on rice genetic improvement and fragmented even at that time and, in in Africa. the past decade, no comprehensive study has updated or clarified those estimates. Dalrymple could not have foreseen the Millennium Development Goals (MDGs) Here, the baseline established by Evenson that facilitate the assessment of progress and Gollin (2003a) has been updated, and guide development efforts in the early widened and deepened. We report on the 21st Century. But he realized the potential results of a CGIAR project – Diffusion and for improving poor people’s welfare from Impact of Improved Varieties in Africa the adoption of modern varieties (MVs). (DIIVA) Project – the first major study to The extent of the area planted with focus on the diffusion of improved crop improved varieties is often the most varieties in SSA. Supported by the Bill & important determinant of productivity, Melinda Gates Foundation (BMGF), seven food security, and poverty benefits CGIAR Centers and their partners carried generated by investments in crop genetic out adoption research and impact research and development (Walker and assessments as part of DIIVA, which was Crissman, 1996; Evenson and Gollin, 2003b; directed and coordinated by CGIAR Fuglie and Rada, 2011). Independent Science & Partnership Council’s (ISPC) Standing Panel on Impact He also knew how hard it was to measure Assessment (SPIA) and administrated varietal uptake and change. He typically through Bioversity International. The DIIVA began his edition-ending summary chapter Project began on 1 December 2009 and with a note of caution. ended on 30 June 2013.

The individual country data which are A budget of slightly under US$3 million summarized here, and the regional totals was allocated to three objectives designed themselves, are labeled estimates for good to: reason. They cannot be considered very precise because of problems in both attain a wider understanding of definition and in reporting (Dalrymple, key aspects of the performance of 1978, p. 125). food-crop genetic improvement in priority crop-by-country combinations He went on to describe in detail well- in SSA; documented cases where the spread of verify and gain a deeper understanding HYVs had been overestimated. of the adoption and diffusion of new varieties in selected priority countries By the 1970s, farmers began to benefit and food crops in SSA; and from recently bred varieties in several acquire more comprehensive insight primary and secondary food crops in Sub- about the impact of crop improvement Saharan Africa (SSA). In the late 1990s, a on poverty, nutrition, and food security.

6 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project The DIIVA project is viewed as a major (2) historical data on varietal release; building block in the construction of a and (3) recent cross-sectional data on routine system for monitoring varietal cultivar-specific levels of adoption estimates adoption and impact in SSA for the CGIAR taken mainly from expert panels and, to a Research Programs. This work has been lesser extent, nationally representative driven by three complementary activities surveys. The unit of observation for the that respond to three project objectives: three datasets is a crop-by-country (1) documenting the key performance combination. indicators of crop genetic improvement; (2) collecting nationally representative These datasets are comparable to, but survey data on varietal adoption; and more uniform than, those collected in (3) assessing the impact of varietal change. Evenson and Gollin (2003a) in the late 1990s. For the purposes of this report, the This Synthesis Report looks at progress work underlying the 13 crop chapters in in the first two areas, documenting the Evenson and Gollin (2003a) is referred to performance of crop improvement in SSA as the ‘1998 Initiative’. while validating estimates from expert panels with results from nationally The results of the validation of experts’ representative surveys on the diffusion adoption estimates are presented in of MVs. A companion report interprets Section 6. Inferences from Sections 2–6 and summarizes the results of three related are examined from the perspective of the case studies on the impact of MVs in SSA MDGs in Section 7. In particular, the (Larochelle et al., 2013; Ndjeunga et al., feasibility of measuring the adoption of 2011; Zheng et al., 2013). MVs in the DIIVA project provides an opportunity to redress one of the Results are presented in Sections 2–5 on deficiencies of the MDGs that have key performance indicators of crop genetic been criticized for neglecting agriculture improvement. The raw material for the (Eicher, 2003). descriptive analysis in these sections consists of three databases: (1) recent Before describing and discussing the results cross-sectional data on the strength of in the main body of the report, we briefly human resources in National Agricultural review crop, country, and data coverage in Research Systems (NARS) by discipline; the next section.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 7 1. Crop and country coverage

The Project’s databases contain about 150 crop-by-country observations selected to cover the most important food crops in the main producing countries (Map 1).5 The planned design of coverage was balanced in the proposal; but, for multiple reasons,6 the number of observations varies somewhat by type of data.

In Table 1.1 coverage is described for the national-level adoption data. Twenty crops and two large maize-producing regions result in 21 crop categories. About half of these were included in the ‘1998 Initiative’ and are described in Table 1.1 as ‘continuing’. The other half is ‘new’ indicating where a baseline on varietal diffusion has been constructed for the first time.

The area harvested within the 20 study Map 1: Frequency of crops in the 30 countries covered in crops in SSA totals about 140 million the DIIVA Project in SSA hectares. These 20 primary and secondary food staples make up about three-quarters of the total crop area in SSA including annuals and perennials. Overall, 83% of the crop study area in SSA was covered in the 21 cropping The number of observations varies from categories.7 Only three crops were sparsely one for lentil, wheat, banana, and field pea represented at a level below 60% area to 17 for cassava and 20 for maize in East coverage. Beans in Kenya, sorghum in and Southern Africa (ESA) and West and Ethiopia, and sweetpotatoes in Nigeria are Central Africa (WCA) combined. Maize is arguably the most important omissions in split regionally, not only because of its rel- the DIIVA Project at this time. evance as a food crop, but because the ESA and WCA are so distinct in their uptake of Breadth of coverage by database is hybrids in relation to improved open-polli- addressed in Table 1.2. The proposal nated varieties (OPVs). The private sector is envisaged coverage of 104 crop-by-country dynamic and now dominant as the source observations. Field pea, banana, and yam of MVs in several important maize-produc- were brought in during the course of the ing countries in ESA, but is only recently Project. Moreover, AfricaRice, the emerging in the production of hybrids in a International Center for Agricultural few West African countries. Research in the Dry Areas (ICARDA) and the International Institute of Tropical Agriculture (IITA) covered many more 5 The data are available online at: http://www.asti.cgiar.org/ countries than was initially planned. The diiva. International Sorghum and Millet collabora 6 Incomplete data collected on improved wheat varieties adopted on large irrigated farms in Kenya, Zambia, tive research project (INSTORMIL) of the and Zimbawbe and the lack of Food and Agriculture Organization (FAO) and/or national production data in very small producing countries for cowpea and 7 For banana, area coverage in 2010 refers to East soybean are two prominent considerations that led to Africa. For the purposes of the project and this paper, an unbalanced coverage across the three databases. production in South Africa is not included in SSA. South However, this effort is substantially more balanced than Africa was included in the ‘1998 Initiative’ for maize and the ‘1998 Initiative’. wheat.

8 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Table 1.1. Area coverage in SSA in 2010 by cropa Share (%) of total SSA Crop Description Number of countries for the DIIVA countries Faba bean Newb 2 100 Cowpea New 18 98 Maize–ESA Continuing 9 97 Yam New 8 95 Lentil New 1 95 Barley Continuing 2 91 Cassava Continuing 17 90 Soybean New 14 86 Maize–WCA Continuing 11 85 Wheat Continuing 1 84 Chickpea New 3 80 Pearl millet Continuing 5 80 Pigeonpea New 3 79 Rice Continuing 19 79 Sorghum Continuing 8 78 Banana New 1 71 Potato Continuing 5 65 Groundnut Continuing 10 63 Bean Continuing 9 59 Sweetpotato New 5 54 Field pea New 1 46 Total/ Weighted mean 152 83 a Refers to the national aggregate adoption data b Refers to crops that were not covered in the 1998 Initiative

United States Agency for International For 62 observations, data are available for Development (USAID) has partnered with comparative analysis between 1998 and the DIIVA Project to improve coverage in 2010 on scientific strength, varietal output, sorghum, which now includes the Sudan and MV adoption. This before and after (Zereyesus and Dalton, 2012). Hence, the analysis, along with a recent assessment of database contains about 50% more the 103 crop-by-country observations in the observations than was proposed. This bonus 1998 data, figure prominently in the next coverage drew on the efforts of CGIAR four sections in which the performance in- Centers and other partners to establish a dicators on crop genetic improvement in comprehensive baseline of varietal SSA are highlighted. diffusion in food crop production in SSA.

Expanded coverage by AfricaRice brought a few very small producers, such as the Table 1.2. The number of crop-by-country Central African Republic (CAR) and Guinea observations in the DIIVA Project by type Bissau, into the Project that now covers of data 30 countries. The median country in the Category Number national adoption database contributes five crop observations. Four countries – Proposed (intended) 104 CAR, Eritrea, Madagascar and Sierra Leone Scientist Years (SYs) 151 – have only one crop observation. At the Varietal release 149 other end of the range, Uganda supplies National adoption 152 11 of a possible 20 crop observations.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 9 2. Scientific strength in crop improvement

Information was compiled on different long-standing Agricultural Sciences and aspects of the scientific strength in crop Technology Indicators (ASTI) initiative that improvement programs in SSA. ‘Crop im- was started by International Service for provement’ is a broad definition that National Agricultural Research (ISNAR) includes plant breeding’s closely allied disci- and is currently led by the International plines, such as genetic resources, molecular Food Policy Research Institute (IFPRI). For biology, and tissue culture. It also covers matching crop-by-country observations, pathology, entomology, agronomy, and any the estimates of total FTE scientists are other discipline – such as social science and compared between the two data sources. post-harvest technology – that helps to identify priorities in the development of genetically improved materials. Natural resource management is excluded as is soil science, unless the research focuses on genotype by environment (GxE) interactions. Therefore, the definition of crop improvement used in this project and report focuses on genetic research – broadly defined and potentially fully supported.

All participating CG Centers collected cross- sectional data on the number of Full-Time Equivalent (FTE) scientists which are Using genomic techniques in marker-assisted selection referred to as Scientist Years (SYs). Data to incorporate striga resistance in sorghum cultivars that farmers are already growing in the Sudan on the level of education and disciplinary orientation in the program were also collected. Like the other databases in the Scientific staff strength by research DIIVA Project, participants were given a program in 2010 set of guidelines on data collection and assembly in this narrowly focused but Table 2.1 on page 11 provides the estimat- important input that potentially affects ed numbers of FTE scientists involved in the crop improvement output (Walker, 2010). 151 crop improvement programs. Although Briefly, ‘scientists’ are defined as public the total number of FTE scientists approaches sector, private sector, and university staff 1,300, the actual number of scientists who who work in crop improvement research work in these programs is likely to be con- and who have an educational level siderably greater. For example, the 126 FTE equivalent to a Bachelor of Science (BSc) scientists in rice refers to the time allocated degree or above. Research technicians and to 289 researchers (Diagne et al., 2012). Only staff working in seed production and about 25–30% of these scientists commit related transfer and extension activities 75–100% of their time to rice research. are excluded, but scientists active in producing breeders’ seed are included. As expected, more scientific resources are allocated to maize than to any other crop Several CG Centers also collected individual in SSA (Table 2.1). Cassava is a distant information on gender, age, and experience second to the total for maize across its two of scientists as well as on program infra major regions of production. structure. Here, we focus mainly on the number of FTE scientists, but also cite other Maize in ESA, with a longstanding tradition aspects of scientific strength from the CG of national programs promoting hybrids, Center draft reports where appropriate. We has benefitted from a sharp and sustained conclude this section with a brief analysis increase in private sector maize breeding, on how these data fit into the broader and especially in Kenya, Malawi, Zambia, and

10 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Zimbabwe (De Groote et al., 2011). The of 40 scientists seems appropriate for the private sector has yet to make its presence importance of a crop in a country with 9.5 felt in maize production in much of WCA million tonnes of production (Kagezi et al., where national programs have emphasized 2012). OPVs (Alene and Mwalughali, 2012a). Relative to other crops, the public sector In contrast to other crops, the quantity has allocated substantial scientific resources of scientists engaged in maize improvement to maize research in several of the programs in ESA is not a cause for concern. 11 producing countries covered in WCA. The nine programs are all staffed by more than 10 FTE scientists with Angola and Maize in Nigeria has the largest scientific Mozambique (12 scientists each) tied for cadre of 77 scientists. Some of these are the smallest program. In other words, even university research staff who allocate part the smallest maize programs in ESA have of their time to maize research. more scientists than the median sized program for 16 of the 19 other crops (Table 2.1). The median program size is 8–9 FTE scientists, which should be sufficient to get the A median program size of 15 for wheat un- job done for all small and most medium- derscores the continuing commitment of sized producing countries unless the crop is governments to invest heavily in this import produced in highly diverse agro-ecologies substitute that is grown on large farms, or unless changes in basic knowledge lead often with access to irrigation, in Kenya, to a radical shift in the distribution of yield Zambia, and Zimbabwe. Ethiopia, where potential, as there are diminishing marginal wheat is grown by smallholders, is by far returns to sampling from the same distribu- the largest wheat producer in SSA. A value tion when knowledge is stagnant or only of 11 for barley reflects the emphasis that increasing incrementally (Kislev, 1977). Ethiopia places on agricultural research. Banana is the largest average-sized program by crop because only Uganda, a Pearl millet is at the other end of the large producing country is included. Its size human resource spectrum. Indeed, its

Table 2.1. FTE scientists by crop improvement program in SSA in 2010

Total FTE Crop Countries scientists Min Median Max Maize–ESA 9 243.2 12.0 17.0 62.0 Maize–WCA 11 139.5 3.0 5.8 77.5 Cassava 17 138.8 1.0 7.2 22.5 Rice 14 125.0 0.9 8.3 15.3 Bean 10 86.5 2.6 5.9 21.4 Potato 5 57.3 3.0 4.6 30.0 Cowpea 18 76.5 0.4 2.9 16.0 Wheat 4 70.1 12.0 15.0 28.0 Soybean 14 52.2 0.8 2.4 14.6 Sweetpotato 5 32.7 2.0 4.0 15.9 Yams 8 49.5 3.0 4.6 12.1 Sorghum 7 42.3 2.4 3.0 18.2 Groundnut 10 23.9 1.15 2.1 5.0 Banana 1 40.0 40.0 40.0 40.0 Chickpea 2 27.0 8.4 13.5 18.6 Pigeonpea 3 6.9 3.9 1.2 5.0 Barley 2 22.1 1.0 11.1 21.1 Pearl millet 5 20.4 1.5 4.5 6.8 Faba bean 2 15.5 6.9 7.8 8.7 Lentil 3 11.0 2.0 3.7 5.3 Field pea 1 6.9 6.9 6.9 6.9 Total/mean 151 1,289.0 na 8.2 na

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 11 largest country program only has about the perspectives of area, production, and seven FTE scientists. With the exception of value of production. As anticipated, crops the largest producing countries in West characterized by small area and value of Africa, pearl millet is almost always a production are associated with higher esti- shared program with other coarse cereals. mated research intensities than those with Groundnut suffers a similar outcome (Table very large area, production, and value of 2.1) and is often a member of a composite production. program made up of pulses and/or oilseeds. The ranking of the crops in terms of Saying something more conclusive about research intensity varies somewhat across the data in Table 2.1 requires adjusting the three criteria in Table 2.2. Potato ranks for the differences in the size of production high in research intensity on area but across different countries. Attaining a occupies a low position on production and critical mass of scientists is needed to value. Banana ranks high on area, low on make progress in large-producing countries production and high on value. However, and crossing a threshold size of production there are more aspects in common than is required before resources should are different across the three criteria. be committed to investing in crop improvement in very small-producing In general, several pulses rank high in countries (Maredia and Eicher, 1995). research intensity in all three criteria. The first five crops listed in the production In Table 2.2 (below), the size of production column of Table 2.2 are all pulse crops with has been normalized across crops and coun- relatively small areas of production. The tries by calculating research intensities exceptions are soybean in Nigeria and that express FTE scientists as ratios from pulses that are produced in Ethiopia, which

Table 2.2. Estimated research intensities by crop in SSA in 2010 from the perspectives of area, production, and value of productiona

Area Production Value of production FTE scientists per FTE scientists per million tonnes of FTE scientists per US$100 million of Crop production Crop million hectares Crop the crop Chickpea 112.4 Lentil 89.1 Banana 25.2 Pigeonpea 64.2 Chickpea 83.6 Soybean 21.4 Potato 61.3 Soybean 45.6 Chickpea 18.4 Lentil 55.6 Bean 43.3 Pigeonpea 17.5 Banana 45.9 Field pea 31.4 Lentil 16.2 Soybean 44.0 Wheat 20.5 Field pea 14.0 Wheat 42.9 Faba bean 20.5 Wheat 13.7 Beans 32.5 Pigeonpea 20.3 Barley 12.8 Field pea 29.7 Barley 15.1 Maize–ESA 8.5 Faba bean 25.3 Maize–ESA 12.3 Sweet potato 7.0 Rice 24.0 Cowpea 11.3 Faba bean 6.2 Barley 22.8 Rice 10.1 Beans 6.1 Sweetpotato 22.1 Maize–WCA 8.1 Maize–WCA 5.7 Maize–ESA 16.5 Potato 6.5 Cowpea 5.3 Maize–WCA 14.0 Groundnut 4.2 Rice 3.9 Cassava 12.6 Banana 4.2 Potato 3.4 Yam 10.6 Sweetpotato 3.6 Sorghum 2.2 Cowpea 6.6 Sorghum 2.9 Groundnut 1.4 Groundnut 5.3 Pearl millet 1.6 Cassava 1.2 Sorghum 2.5 Yam 1.0 Pearl millet 1.0 Pearl millet 1.4 Cassava 0.9 Yam 0.4

a All estimates are weighted averages.

12 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project has invested substantial scientific resources of production constant with binary vari- in its NARS – at least in terms of the ables for region and crop type reinforces number of scientists. Bean’s high ranking the evidence from the tabular results. speaks to the stability of the Pan-African Shifting to East Africa from West Africa is Bean Research Alliance (PABRA) – one of accompanied by an increase of five FTE the regional crop improvement associations scientists. In the same ‘holding everything that survived a shrinking budget for inter else equal’ format, shifting from roots and national crop improvement research in the tubers to the so-called ‘superior’ cereals of 1990s and early 2000s. Cowpea, which is maize and rice results in an even larger the lowest ranking pulse in Table 2.2, is increase, an addition of about nine FTE produced almost entirely in West Africa. scientists (Walker et al., 2011a).

Turning to the cereals in Table 2.2, barley does well because of its location in Ethiopia, The congruence rule: How many which has a large and regionally decentral FTE scientists are desirable? ized national program at the Ethiopian Institute of Agricultural Research (EIAR). Building on the estimated research intensi- Rice also displays a research intensity ties in Table 2.2, it is useful to compare the estimate above 10.0 from the perspective actual allocations of FTE scientists with of production. Potato has a leading normative allocations calculated from a position in roots and tubers because of congruence rule. This states that research its high market orientation and demand resources should be allocated in proportion in East Africa. to the value of production across commodities, if all other things are considered equal Cassava, yams, and pearl millet appear at (Alston et al., 1995). In priority setting, 2% the bottom of Table 2.2. Relative to their of value of production is a common as- area, production, and value of production, sumption because studies have shown that all three of these semi-subsistence food research investment proportional to agri- crops appear to be starved for research cultural gross domestic product (GDP) resources. In terms of area, groundnut often exceeds 2% in developed countries and sorghum are also characterized by (Walker et al., 2006). In developing coun- very low research intensities. tries in SSA, the 2% criterion is rarely obtained (Beintema and Stads, 2011). In Specific cases of resource deprivation can large countries, such as China and India, be identified by counting the incidence where economies of scale and size prevail, of falling below an arbitrary, but seemingly research investments of the order of 1% of reasonable, threshold of critical mass. agricultural GDP are commonplace. This lower bound threshold for large programs exceeding two million tonnes When comparing normative to actual of production is established at nine allocations, we have assumed that 1% of scientists (the median-size program as the value of production is desirable for the shown in Table 2.1). Ten large-producing size of research investment and that each crop-by-country combinations fall below scientist costs on average US$115,000 in this minimum threshold: cassava in Benin, purchasing-power parity (PPP) in 2010. The Côte d’Ivoire, Malawi, and Mozambique; latter assumption is well within the range cowpea in Côte d’Ivoire and Guinea; of comparable estimates in the ASTI groundnut in Nigeria; pearl millet in Niger country studies. We also cap the maximum and Nigeria; and sorghum in Nigeria. size of a crop-by-country program at 80 FTE From the perspective of production, the es- scientists, recognizing economies of size timated research intensity of these 10 crops and scale in agricultural research. This is in the range of 0.2–2.0 and averages 1.0. admittedly arbitrarily imposed limit is slightly above the size of the largest Conducting a simple regression analysis is program – maize in Nigeria. another way to shed light on the data on scientific strength in Tables 2.1 (page 11) In order to achieve congruence or parity and 2.2 (page 12). Regressing polynomial in research intensities across crops with a production terms on SYs to hold the effects fixed budget, resources would have to be

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 13 reassigned from the crops with positive Differences in scientific strength estimates in Table 2.3 (below) to the over time commodities with negative estimates. The sign and size of the estimates by crop are Results on differences in scientific strength sensitive to our assumptions on a desirable over time are mixed. Between 1998 and target for research intensity, the cost of 2010 more programs have gained scientists each FTE scientist, and the limit on the than have lost researchers but, because of size of the program. The relative position rising production, estimates of research of the crops in Table 2.3 will change intensity have not improved and have even somewhat as these assumptions vary, but declined for the majority of the 65 not as much as their numerical values. programs with information available to Assuming payoffs are the same, these carry out paired comparisons. Before more formal results reinforce the findings addressing changes over time, we briefly on research intensities in Table 2.2 (on examine the results of previous estimates page 12). Using the congruence rule to of scientific staff strength in 1998 for SSA set priorities shows that research into (Walker et al., 2011a). cowpea, groundnut, pearl millet, sorghum, and yams is underinvested 1. Nigeria stood out as a country with relative to other crops, from the consistently low researcher intensity. perspective of the value of production. Indeed, Nigerian farmers appeared to The deficit of FTE scientists encountered be afflicted by some of the lowest for these largely West African crops has readings on researcher intensity ever more to do with the reliability of FAOSTAT estimated anywhere in the world. Mean data (from the statistics division of the readings of the ratio of FTE scientists to Food and Agriculture Organization) on million tonnes of production were 0.1 the value of production than with for cassava, 0.5 for sorghum, 1.7 for economic assumptions on priority setting. rice, 1.8 for pearl millet and 2.6 for

Table 2.3. Comparing the actual allocation of FTE scientists to a normative allocation by crop

Simple average in FTE scientists

Crop Actual allocation Normative allocationa Difference Banana 42.0 11.1 30.9 Wheat 17.5 8.5 9.0 Chickpea 13.5 4.9 8.6 Maize–ESA 27.0 21.3 5.8 Barley 11.1 5.8 5.3 Pigeonpea 7.8 3.0 4.8 Soybean 3.9 1.2 2.7 Lentil 3.6 1.5 2.1 Field pea 6.9 5.2 1.7 Sweetpotato 6.5 6.3 0.3 Faba bean 7.8 8.4 -0.6 Beans 8.7 9.5 -0.9 Cowpea 4.5 5.6 -1.2 Maize–WCA 12.7 14.7 -2.0 Rice 9.6 16.6 -7.0 Potato 7.6 14.7 -7.1 Groundnut 3.4 16.4 -13.0 Sorghum 6.6 20.3 -13.7 Pearl millet 4.1 27.8 -23.8 Cassava 8.2 32.1 -24.0 Yams 7.0 47.3 -40.4

a Assumes a research intensity of 1% of value of crop production, a cost per FTE scientist of US$115,000 and a maximum program size of 80 FTE scientists.

14 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project maize, which benefited from some of the 65 programs had fewer FTE private sector participation in research. scientistis in 2010 than in 1998. Among Nigeria ranked among the lowest in the 35 programs that gained staff, two researcher intensity in each of the five observations were unduly influential in commodity groups to which it was a the results – maize programs in Nigeria major contributor. The country also and Zimbabwe both experienced increases figured prominently when the that were equivalent to over 40 FTE performance indicators for these scientists. same crops were aggregated. 2. Ethiopia, Kenya, South Africa, and Some of this change is undoubtedly real, Sudan were characterized by a higher but some may be attributable to an un- investment in scientific staff than other derestimation of scientific capacity in countries in the 1998 dataset. This 1998, e.g. maize in Nigeria included sub- behavior was reflected in positive and stantially more university researchers in statistically significant estimated 2010 than in 1998. Excluding maize in country coefficients in an additive Nigeria and Zimbabwe, the mean scientific effects model regressing total SYs on strength in 1998 was 8.4 FTE scientists production, crops and countries. compared to 9.7 in 2010, resulting in a 3. Researcher intensity was lower in positive but statistically insignificant cassava than in other crops even when change at the 0.05 level. The median the relatively inferior output value program also gained 1.3 FTE scientists as of cassava was factored into the the difference between the two time calculation. Rice and sorghum also periods was normally distributed. Overall, had lower than expected research these results suggest a marginal increase intensities, although not as extreme in scientific capacity. as cassava. 4. Estimates of researcher intensity Cassava appears in Table 2.4 as the largest declined exponentially as the size of loser of scientific capacity. Maize in ESA, production increased from under potato, rice, and wheat were the biggest 50,000 tonnes to more than 5 million gainers. tonnes. This is similar to the findings of other enquiries (Maredia and Eicher, These gains in staff were not sufficient to 1995). translate into increased research intensity in most crops. The net decline in research in- Data are available for a before-and-after tensity was about 1.7 scientists per million analysis of the changes in scientific tonnes of production, which suggests that capacity for 65 matching crop-by-country growth in production outstripped the observations that feature eight of the smaller positive changes in staffing. Maize continuing crops (Table 2.4, below). Thirty and wheat in ESA were the only crop

Table 2.4. Differences in estimated FTE scientists and research intensities between 2010 and 1998 by crop based on 65 paired comparisons

Mean FTE Median FTE Mean research Paired Crop scientists scientists intensity observations Beana -0.6 -0.8 1.3 8 Cassava -2.4 -2.3 -4.7 14 Maize–ESA 10.8 7.0 3.9 9 Maize–WCA 4.3 -3.3 -32.4 9 Pearl millet -1.1 -1.0 -0.9 5 Potato 6.9 3.6 -7.9 4 Rice 4.3 3.8 -4.3 6 Sorghum 1.9 1.4 -10.3 6 Wheat 7.3 8.5 110.9 4 a For Bean, the definition of scientists applies only to breeders.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 15 categories that accrued substantial gains in programs on the right-hand side of Figure researcher intensity (Table 2.4).8 2.1 could not sustain their staff strength. These were mainly concentrated in cassava- A first-difference comparison of the bulk of growing programs. For example, Benin, the overlapping crop-by-country observa- Guinea, and Tanzania downsized to only tions is presented in Figure 2.1 (below). For 2–3 scientists per program. reasons of scale, three high-end outliers are excluded: maize in Kenya that had very For a few maize programs in WCA, the large values in 1998 and 2010, and maize in numbers of scientific staff also declined Nigeria and Zimbawbe that had very high over time. But these declines were more values in 2010. than compensated for by Nigeria’s dramatic increase in scientific staff, discussed earlier A small majority of the 62 remaining obser- in this report (see Table 2.1). Overall, the vations increased their numbers of scientific data presented in Figure 2.1 convey the staff between the two periods. One of message that larger crop improvement these was cassava in Nigeria which added programs may be highly susceptible to about six scientists. Notably, we also see downsizing in times of financial crisis or that several of the largest commodity when donor support ends.

Potato 20 Kenya

Maize Tanzania

10 Cassava Nigeria

Maize Pearl millet Ethiopia Nigeria Cassava Difference in FTE scientists between 1998 and 2010 Difference -10 Cameroon Cassava Zambia Cassava Tanzania

0 5 10 15 20 25 FTE scientists in 1998

Figure 2.1. Change in scientific staff strength in food crop improvement programs between 1998 and 2010. (The size of the circles reflects the size of production value in 2010. Note that Nigeria’s observation for cassava is the largest circle in the bubble graph.) (Source: DIIVA SY Database)

8 Declining production in a crop can lead to rising research intensity. But among these observations, only wheat in Zimbabwe seems to demonstrate increasing research intensity attributed to steadily decreasing output.

16 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Other aspects of scientific capacity: crop improvement research (Eicher, 1995). age, education, experience, and Even with increasing participatory varietal area of specialization selection (PVS) and marker-assisted selection (MAS) it can take, on average, about 10 The problem of scientific capacity in NARS years from parental crossing to progeny in West Africa is not only a problem of release in the same country. PVS is relative numbers of scientific staff but also increasingly becoming a reality in rice and of age. About 65% of the scientists working beans among the food crops in the DIIVA on groundnut, pearl millet, and sorghum in Project. MAS is still rare and newsworthy in the five project countries in West Africa SSA. It has been applied to facilitate varietal were over 50 years in 2010 (Ndjeunga et al., development in only a few successful cases, 2012). Rice shows a more typical age profile such as sorghum in the Sudan (International – about half of the 289 rice scientists Crops Research Institute for the Semi-Arid documented in the Project were older than Tropics [ICRISAT], 2009). The DIIVA Project 50 years (Diagne et al., 2012). sought to collect information on the duration of varietal generation, selection, Scientists engaged in crop improvement and testing. However, reliable data over across WCA appear to be more highly time on this aspect of crop improvement educated than their ESA counterparts, with performance were not forthcoming, so we around 2.6 Doctor of Philosophy (PhD) cannot say whether the gestation period of holders per program. But in future an esti- new MVs is shortening or staying the same. mated lower number of BSc holders in WCA We can say, though, that instability in is a cause for concern because fewer scientific staffing levels within crop younger scientists will be available to be improvement programs can severely curtail mentored by, and capitalize on the experi- their potential. Full potential will only be ence of, older scientists (Table 2.5 below). reached if the routine work of varietal selection and testing takes place season The incidence of scientists with PhDs and after season and year after year. Master of Science (MSc) qualifications is encouraging (Table 2.5). Only 24 of the 135 Estimates on experience levels within the programs did not have a PhD presence. Only same area of research suggest that many four programs had neither a PhD nor an MSc scientific staff have been able to work on scientist involved directly in their research. the same crop for an extended period of More than half of the programs have at least time. For example, the 289 NARS rice 1.0 FTE PhD scientist working in research. For scientists had worked on rice improvement the most part, all crops and most countries for an average of 12.25 years as of 2010 have at least one program supported by (Diagne et al., 2012). Scientists with ten or several PhDs and MScs. Eritrea was the more years’ experience made up the exception among the 30 countries in the majority of staff in five of the ten bean DIIVA Project. Nonetheless, it was still programs in ESA (Muthoni and Andrade, possible to find programs, such as cassava in 2012). This level of experience is surprising Tanzania, that were severely understaffed because only about one scientist in six was from both a numerical and educational older than 50 years in 2010 across the ten perspective. programs.

Staff stability is a primary ingredient for a Estimates on the allocation of scientists recipe of sustained output from investing in across specialized areas of crop improvement

Table 2.5. Educational level of scientists in crop improvement programs by region in SSA

Mean number of FTE Scientists by educational level Number of observations PhD MSc BSc Total ESA 65 1.51 3.20 2.33 7.03 WCA 70 2.61 2.84 1.66 7.12 Total 135 2.08 3.01 1.98 7.07

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 17 are presented in Tables 2.6 (below) and 2.7 Three other findings in Table 2.6 warrant (overleaf) on two aspects: crop type and comment. First, molecular biology only strength of scientific resources. We expect accounts for 3.4% of the mean resources that relative allocations across areas of across the 150 programs in the database. specialization will vary substantially across This level of investment is not significantly cereals, grain legumes, and roots and different from tissue culture, which has been tubers. Root and tuber programs that are a staple area in root and tuber crop im- based on vegetatively propagated material provement since the 1970s. The 3.4% is and on clonal selection are hypothesized to equivalent to only 40 FTE scientists; 17 of be characterized by a more diverse area whom work on banana in Uganda. Secondly, allocation than cereals and grain legumes, the level of social science involvement in which typically are more heavily concen- crop improvement work is much higher than trated in classical plant breeding. It was expected. Thirdly, post-harvest work is con- expected that increasing human resources centrated on maize and cassava in Nigeria. would be accompanied by a lower concen- tration in plant breeding and agronomy, The differences between more sparsely which are conventionally viewed as the and densely staffed crop improvement core disciplinary areas of crop improvement programs were also less than anticipated. research. The largest programs in Quartile 4 in Table 2.7 display a more even disciplinary These expectations are largely confirmed in allocation pattern across disciplines than Tables 2.6 and 2.7, although the differences the smallest programs in Quartile 1, but the among programs based on generalized differences are milder than expected. On crop orientation as well as small versus average, even the smallest programs from large programs are not as obvious as antici the perspective of total scientists invest pated. With regard to crop type, the main about half of their resources in disciplines distinction focuses on roots and tubers on other than plant breeding. Nevertheless, one hand and cereals and grain legumes on the smallest programs invest relatively few the other. Root and tuber crop programs resources in molecular biology, entomology, invest considerably less in plant breeding social science, and post-harvest research, and more in the biotechnological areas of compared to programs in the quartiles with molecular biology and tissue culture than higher relative allocations. By contrast, the cereal and grain legume programs. With relative research allocations to tissue the exception of post-harvest research, the culture, pathology, agronomy, and seed other research areas are surprisingly similar production do not vary systematically by across the generalized crop types. The size of the program. This lack of response emphases on entomology, pathology, agro to program size suggests that these areas nomy, and social science are not markedly are viewed as essential investment areas for different across the three groups of crops. crop improvement.

Table 2.6. Relative allocation of scientists by disciplinary specialization across roots and tubers, grain legumes, and cereals in SSA in 20l0 in % sharesa

Broad areas of crop improvement Root and Grain work tuber crops (5) legumes (8) Cereals (7) All 20 crops Plant breeding including germplasm 21.8 45.8 44.39 39.6 conservation Plant pathology 8.3 10.9 7.80 9.2 Molecular biology and genetic engineering 11.4 0.5 1.22 3.4 Tissue culture 11.9 0.1 0.40 3.0 Entomology and nematology 5.4 6.1 7.38 6.3 Agronomy, weed science, and seed production 25.2 24.6 23.68 24.4 Social science 8.7 10.3 9.36 9.6 Post-harvest and food science 5.0 0.6 4.55 3.6 Other areas including soil science 1.2 0.2 0.20 0.6

a Numbers in parentheses refer to the number of observations in each crop category

18 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Table 2.7. Relative allocation of scientists by disciplinary specialization across program-size quartiles in SSA in 20l0 in % shares

Broad areas of crop improvement work Quartile 1 Quartile 2 Quartile 3 Quartile 4 Plant breeding including germplasm conservation 49.4 40.0 32.7 35.8 Plant pathology and virology 5.9 7.6 11.19 7.1 Molecular biology and genetic engineering 1.0 2.3 2.44 3.7 Tissue culture 3.2 3.9 2.74 3.3 Entomology and nematology 3.9 7.4 11.01 5.4 Agronomy, weed science, and seed production 24.3 20.4 15.51 20.5 Seed production 7.9 8.4 6.38 10.3 Social science 2.8 6.6 12.90 8.6 Post-harvest and food science 1.6 3.4 5.1 5.3 Total FTE scientists 63.1 137.9 292.0 796.1

The term ‘essential’ should not convey the Data from IITA address the issue of the notion that all programs are active in these relative importance of scientific capacity in areas. Fifty of the 150 programs do not the national programs as compared with have any representation in pathology, the CG Centers. Over its five mandated which historically has been one of the most crops, the ratio of the Centers’ FTE scientists productive areas in plant breeding in to the total number allocated to NARS screening for varietal resistance and toler- ranges from about 0.06 for cowpea and ance to economically important plant maize to 0.14 for yams, which have diseases. Investment in entomology in grain commanded little attention from national legumes was also lower than expected. programs. About 40 FTE Center scientists and 450 FTE national program scientists worked on the improvement of cassava, FTE scientists in the CGIAR cowpea, maize, soybean, and yams in 2009 (Alene and Mwalughali, 2012a). Hence, the Evidence on the levels of, and changes in, share of CG Center scientists in total investment in crop improvement in SSA scientific capacity was about 8%. from the perspective of scientific capacity in the CGIAR is not transparent. This is In four of the five IITA crop improvement because only two of the CG Centers, IITA programs, the disciplinary allocation of scien- and AfricaRice, operate exclusively in SSA. tists to CG Center programs differs markedly ICRISAT – which has several long-standing to their allocation in national programs. The regional programs and country research IITA allocations are based heavily on plant agreements – has allocated a large share breeding – just over half of IITA’s scientists of its budget to crop improvement research are plant breeders. Meanwhile, the in SSA. The International Center for Tropical allocations within national programs show Agriculture (CIAT), International Center for significantly more diversification across the Improvement of Maize and Wheat disciplines. The difference suggests that the (CIMMYT), International Potato Center CG Center programs are highly focused on (CIP) and ICARDA have also made sizeable genetic improvement. In contrast, the yam commitments. Only including staff posted program at IITA resembles national programs in SSA does not do justice to the size of in terms of its level of disciplinary diversifica- that investment, but partially assigning tion, although the disciplines vary between headquarters and even staff from other the two types of programs. Tissue culture regions to more fundamental research and social science are well represented in in SSA is a difficult exercise to assess. IITA’s yam improvement program, which Fortunately, there is sufficient evidence seems to be at the early stage of develop- to piece together a coherent story under ment when diagnoses of constraints and the assumption that financial trends in market opportunities are important. regional and global crop improvement research have not varied much from Center The cross-sectional evidence from IITA is to Center. indicative of the relative contribution of CG

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 19 Centers to the total scientific capacity within SSA was staffed by just 12 internationally SSA in the very recent past. Time-series infor- recruited scientists. More recently, scientific mation on scientific capacity at ICRISAT capacity has recovered with increased suggests that the past has been anything project funding from donors such as the but constant. Since its establishment in 1972, BMGF. the number of PhD scientists working in crop improvement programs on chickpea, ICRISAT – with its large cadre of national- groundnut, pearl millet, pigeonpea, and level PhD qualified scientific staff – was sorghum expanded rapidly in the 1970s at arguably more affected by the financial its headquarters in Patancheru, India crisis of the early to mid-1990s than any (Figure 2.2). During the 1980s and into the other CG Center. However, the same pattern early 1990s, that level stabilized at about 75 is visible over time when other definitions, FTE scientists. At the same time, the number such as the number of staff recruited of PhD scientists was steadily increasing in internationally, are used to describe the SSA until the early 1990s when capacity regional allocation of CG scientists over peaked at about 30 scientists. In the 1970s time. Moreover, the loss of scientific and 1980s, most of these scientists worked capacity in the 1990s and early 2000s is on pearl millet and sorghum improvement not unique to ICRISAT. For instance, the in West Africa. In the late 1980s and into the investment made by CIAT in bean early to mid-1990s, genetic research improvement grew from about US$350,000 expanded to include groundnut and two and two principal scientists in 1970 to about pulse crops (chickpea and pigeon pea) as US$8 million in 1985 with 20 principal well as millet and sorghum improvement in scientists. Expenditures peaked around southern Africa. US$13.8 million in 1990 with 26 principal scientists and seven internationally recruited The mid-1990s ushered in a period of breeders. By 1997/98, investment in bean budget tightening that threatened institu- improvement by CIAT had declined to about tional collapse. CG Centers were required to US$7.7 million with 18.5 internationally make an investment in newer initiatives recruited scientists (Johnston et al., 2003). such as resource management research. PhD scientists in crop improvement plum- Rating budget expenditures by meted to 30 (Figure 2.2). By the late 1990s internationally recruited scientists over time and early 2000s, genetic improvement in shows that CIAT’s resources allocated to

90 80 70 60 50 Headquarters, India 40 SSA Other regions

PhD scientists 30 20 10 0 1978 1982 1987 1992 1998 2002 2010 Selected years

Figure 2.2. Number of PhD scientists working in the groundnut, pearl millet, sorghum, and pulse improvement programs in ICRISAT by location for selected years in 1978–2010. (Source: ICRISAT Annual Reports for 1978, 1982, 1987, 1992, 1998, 2002, and 2010)

20 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project bean improvement further declined from the late 1980s, economists at ISNAR and about US$7 million in 2002 to just under now at IFPRI, working under ASTI, have US$3 million in 2007. A substantial decrease collected comprehensive information on in core funding and major restructuring agricultural research in SSA. processes in the CGIAR and CIAT led to a reduction in major operations in the bean Although DIIVA focuses on specific crop improvement program. The most affected improvement programs and ASTI addresses bean research activities were in CIAT country-level sectoral agricultural research headquarters in Cali, Colombia. as a whole, the substantive findings in this section resonate well with those from a Recently, funding has increased and stabi- recent analysis of the latest round of ASTI lized at about US$5.5 million annually. The inquiries (Beintema and Stads, 2011). Results entry of new donors who were interested in on the lack of investment in agricultural targeting genetic improvement as a means research in West Africa and concerns about of increasing food security led to a recovery scientists’ ageing profiles are common to in funding. Recent growth has been both DIIVA and ASTI. The absence of BSc targeted at small-scale bean production in entry-level scientists in agricultural research Africa and is mediated through the PABRA in the Francophone countries as an network. important component of the demographic problem has also been identified in both The story of budgetary woe for crop studies. In addition, this has called attention improvement also applies to CIP in the to the empirical fact that variation in mid-1990s to the mid-2000s. By the late investment in agricultural research is high 1980s, more than 70 PhD scientists were in SSA, with sizable gainers and losers in a working for CIP in a decentralized research generalized picture of stagnation. The ASTI organization that featured outreach in eight reports also found plausible explanations for regional programs. In 1988, the genetic the results on the changing effectiveness of improvement mandate also expanded to NARS (Alene and Mwalughali, 2012a). sweetpotato. By the early 2000s, the number of scientists fell to 30 whereas the regional In general, ASTI researchers collect data research programs reduced to two, one in on all institutional agencies engaged in ESA and the other in Southeast Asia. The agricultural research and aggregate the in- distribution of plant materials became a formation to the national level, while trickle of the flow of genetic resources and relevant budgetary information is docu- elite varieties exchanged in the 1980s and mented annually. Data collection for the early 1990s (Thiele et al. 2008). DIIVA Project was at a lower, more disag- gregate level – its sources of information Following the establishment of the com- were the scientists in, and leaders of, commodity Centers and their rapid growth in modity improvement programs. Many of the 1970s and 1980s, there is perhaps some- these contacts were long-standing partners thing to be said for a rationalization of the participating CG Centers. process to augment the efficacy of genetic improvement. But the cuts in the 1990s and The DIIVA approach to information early 2000s were too severe to be classified gathering was not cost effective for all as necessary adjustments. They jeopardized data. With hindsight, data gathered on the the stability of crop improvement and age, educational level and gender of exacted a high opportunity cost in terms of scientists could have been more reliably and varieties that were never released and easily obtained at the institutional level. impact that was never realized. What is more, DIIVA Project data – sourced, collected, and analysed for different thematic areas – was limited compared with Agreement between the DIIVA and ASTI’s information for the same country. the ASTI FTE estimates But there is no reason to hypothesize that one or more crop improvement programs The DIIVA Project was not the first to would seriously depart from the general gather information on the health of, and tendencies of the research institute or trends in, agricultural research in SSA. Since country in these cases.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 21 On the other hand, ASTI has not gathered baseline of scientific capacity estimated in information systematically on the scientific the DIIVA Project. Although about 25% of capacity in crop improvement programs at the ASTI estimates – such as maize in the level of discipline, such as plant Nigeria and banana in Uganda – are lower breeding, entomology, pathology, etc. than the DIIVA figures, the majority of ASTI However, ASTI researchers have collected estimates exceed the DIIVA estimates. data on the total number of FTE scientists per crop, commodity, or thematic research The matched observations are for the larger area in more than 20 countries in SSA. research programs as only a few of these Those data are presented as percentage are listed numerically for each country on shares at the national level and are avail- the ASTI website. The smaller research areas able on the ASTI website. are relegated to a residual ‘other‘ category. This means that the base DIIVA estimate for Nonetheless, complementarity between the these programs is also high with a mean of two sets of estimates could be high. Docu- 13.5 FTE scientists per program. At the mid- menting and understanding the differences point of the 48 observations, the ASTI between the two sets of data could also be estimate exceeds the DIIVA estimate by used to highlight future research priorities 4.15 FTE scientists, implying a 30% increase. in assessing the scientific capacity of crop improvement programs. Three other explanations are also plausible. Firstly, researchers in higher education are Forty-eight paired observations are readily well-represented in each country in the available from the two sources in 21 coun- ASTI estimate. Few universities release vari- tries. The difference in FTE scientists is eties directly in SSA, but most have staff charted in Figure 2.3 (below) from a who engage in research in addition to their

150

Wheat Ethiopia 100

Cassava Nigeria

Maize 50 Sorghum Ethiopia Sudan

Maize Maize Tanzania Kenya 0 Banana Uganda Maize Nigeria Difference in FTE scientists: ASTI and DIIVA estimates ASTI and DIIVA in FTE scientists: Difference

-50

0 20 40 60 80 FTE scientists in 2010: DIIVA estimates

Figure 2.3. Comparing ASTI and DIIVA estimates of scientific strength. (Source: ASTI estimates from the ASTI website; DIIVA estimates from DIIVA SY database)

22 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project teaching responsibilities. On average, The easiest way to do this would be to higher education contributes about 25% of compare rosters of scientists working on the total FTE estimate across the 21 coun- the crop in large countries characterized by tries in the ASTI database. About 1% of FTE large differences in the estimates. Highly scientists are located in the non-profit focused, comparative checking should lead sector – in this specific ASTI database, the to converging estimates during the next contribution of the private sector was neg- round of inquiry. ligible. In contrast, DIIVA Project participants focused their attention on public sector institutions that had a well-known Summary history and reputation in carrying out crop improvement research. The national scientific capacity of the 150 crop improvement programs examined in Supporting this explanation is the finding the DIIVA Project approaches 1,300 FTE sci- that the deviations in the two estimates entists. But the actual number of scientists were greatest in the largest and strongest who work in these programs is likely to be NARS in terms of numbers of FTE scientists. more than double this sum. In rice, for Marked disparities between estimates were example, 125 FTE scientists equates to 289 found in Ethiopia, Kenya, Sudan, and researchers, because only about 25–30% of Tanzania where the ASTI estimates were these scientists commit 75–100% of their considerably higher than the DIIVA time to rice research. More scientific re- estimates. In these larger countries, tech- sources are allocated to maize than to any nology transfer and other programs not other crop in SSA. Cassava is a distant strictly related to research may find their second to maize. way into the ASTI estimates. For example, in Kenya, a large potato seed program Of the 20 crops, cassava, yams, and pearl related primarily to technology transfer was millet consistently rank at the bottom of initially included but subsequently excluded the charts on research intensity. Relative to from the DIIVA FTE estimate for potato their area, production, and value of produc- scientific capacity in Kenya. That program tion, all three of these semi-subsistence absorbs most of the discrepancy between food crops appear to be research resources. the two estimates; therefore, in practice, it In terms of harvested area, groundnut and is likely that DIIVA’s definition of crop im- sorghum are also characterized by very low provement is narrower than ASTI’s. research intensities.

Analysis of the smaller programs in the ASTI Results on the differences in scientific ‘other’ category could partially reverse the strength over time are mixed. Between finding of lower DIIVA estimates of scientif- 1998 and 2010, more programs have gained ic strength. In terms of size distribution, the scientists than have lost researchers. DIIVA estimates could even be higher than However, because of rising levels of crop ASTI’s as smaller programs tend to be production, mainly attributed to area ex- shared across the same crop types, whose pansion, estimates of research intensity components are sensitive to assumptions on have not increased and have even declined the allocation of scientists’ time across crops for most of the 65 programs that have in- in the same program. formation available to carry out paired comparisons. The findings on cassava and maize in Nigeria are one of the puzzling features of Comparing these findings to the 2010 this exercise. The same researchers estimat- results highlights several transparent differ- ed substantially more FTE scientists in maize ences. Nigeria, for example, has invested and substantially fewer in cassava significantly in maize research, while its sci- compared to the ASTI estimate for each entific capacity in rice and cassava has also crop in the same country. Indeed, this com- improved. But by far the largest increase in parison suggests that rationalizing the dif- scientific capacity has occurred in maize ferences between the estimates for cassava across ESA, thanks largely to the dynamism in several of the larger growing countries, of the private sector in this region. Notably, such as Nigeria and Tanzania, is a priority. the comparisons also strongly suggest that

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 23 larger public sector crop improvement Nevertheless, the overall number of programs may be highly susceptible to scientists with PhDs and MSc qualifications downsizing in times of financial crisis or is encouraging. Only 24 of the 135 crop when donor support ends. improvement programs do not have a PhD presence. Only four programs have neither Concerns over scientific capacity in national a PhD nor an MSc scientist involved directly programs in West Africa reflect not only in their research. More than half of the a problem of relative numbers but also of programs have at least 1.0 FTE PhD scientist scientist age. About 65% of the scientists working in research. working on sorghum, pearl millet, and groundnut in the five project countries With regard to crop type and the pattern of in West Africa were older than 50 years disciplinary research resource allocation, in 2010. the main distinction centers on roots and tubers on one hand and cereals and grain Scientists engaged in crop improvement legumes on the other. Root and tuber across WCA appear to be more highly programs invest considerably less in plant educated than their ESA counterparts, breeding per se but more in closely allied with around 2.6 PhD holders per program. disciplines such as tissue culture. Molecular But in future an estimated lower number biology only accounts for 3.4% of the mean of BSc holders in WCA is a cause for concern resources across the 150 programs in the because fewer younger scientists are database. This 3.4% is equivalent to only available for mentoring by older, experi 40 FTE scientists, 17 of whom are involved enced scientists. in studies of banana in Uganda.

24 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project 3. Varietal output

‘Output’ refers to the expansion that varieties may be available for adoption can be attributed to genetic improvement but may not appear in release registries. in the potential availability of valuable Escapes from breeding programs may be genotypes for cultivation. Ideally, attri widely adopted and not well identified. bution is measured from a with and without perspective, i.e. the difference Almost all countries have well described between what is potentially available with procedures for varietal release, but genetic improvement and what is few – like Ethiopia and Kenya – have available without an investment in plant compiled comprehensive release registries breeding. for downloading on the Internet. An exhaustive review of varietal registration By its nature, crossing and selection is a in 24 rice-growing countries shows that winnowing process that is characterized nine do not have an established release by a search for a smallish number of gen- and registry system in place (Sanni et al., otypes that are perceived to be valuable. 2011). In some countries with established Elements of perceived value are encoded systems, release committees do not meet in government registry and release prac- periodically and are financially constrained tices that place an imprimatur on because of pressures on government breeders’ elite selections. Official release operating budgets. is tantamount to saying that ‘liberated’ varieties are potentially valuable for culti- Moreover, changes in the release practices vation in the sense that they have satis- over time may give the illusion of increased fied rigorous criteria, such as threshold varietal output when, in fact, its true trajec- yield advantages, compared to check vari- tory has not changed. Comparing release eties in multi-locational testing on lists over two points in time also suggests research stations over time. In well-func- that older improved varieties can reappear tioning systems of varietal release and at a later date in the registry, giving the im- registry, information on the quantity and pression of recent output when in fact the location of breeders’ seed is published. In cultivar was generated much earlier. this report, output is synonymous with varietal release – the most immediate and One can also cite cases such as Guinea observable indicator of progress in crop (with a rare institutional setup of multiple improvement. institutions releasing varieties of the same crop) where more than 100 rice varietal Varietal release is not a perfect indicator releases in the 1980s and 1990s has resulted and, in specific cases, may not even be a in limited discernible adoption. However, good measure of varietal output in agri- rice in Guinea is an outlier in the joint culture within developing countries. Both varietal release and adoption database. private sector and public sector improved The estimated simple correlation between total historical releases and the percentage of adoption for improved varieties in 2010 is a statistically significant but modest 0.17. However, the ‘weighted by area’ association is markedly higher at 0.47.

The relationship between varietal release, adoption and subsequent impact is asymmetric. Large numbers of releases can result in substantial or no adoption, but Cowpea seeds show wide varietal diversity. Varietal release, zero or negligible releases rarely result in for all of its imperfections, is still an important benchmark appreciable adoption of improved varieties. for assessing progress in varietal output Absence of release activity is synonymous

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 25 with negligible output from plant breeding. in the 1990s. The crop improvement Performance in crop improvement needs programs of the CGIAR were most likely to be measured and varietal release, for a force that contributed to offsetting all its imperfections, is still an important differences in initial advantage in benchmark for assessing progress in varietal research endowments because most output. CG Centers reached their full potential to generate varietal output in the 1990s. Findings on varietal output in 1998 4. Across the eight food crops in the study, the higher and more stable release rate In the ‘1998 Initiative’, most CGIAR in wheat was anticipated. In contrast, participants were successful in assembling the very low release intensity for cassava valid release data for almost all countries, was unanticipated. Cassava ranked last supplemented by information on so-called in the average varietal output by a wide informal releases of suspected improved margin on any criterion of release varieties. For maize in ESA, release was intensity. For cassava, the size of country equated to varietal availability in the production was not positively correlated market in the late 1990s because of heavy with the number of releases. Cassava did private sector participation in seed have a colonial legacy of genetic production and distribution. In spite of the research in the 1960s to draw from, but inherent difficulties in inferring varietal governments were slower to invest in output from varietal release, such data this important staple than in grain crops, present an historical benchmark that, once where technological change was consolidated carefully, can provide a firm perceived to be more of a reality foundation for updates over time. (Nweke, 2009). Other crops, especially rice, have had a substantially richer In the pooled analysis of varietal release institutional milieu in the form of covering the period 1965–1998 (Walker et national, regional, and international al., 2011a), relevant findings included: organizations that have been actively involved in promoting crop improve 1. Across all crops, annual releases ment over the past 50 years in SSA. increased at an accelerating rate from 5. Release profiles were often punctuated the 1960s to the late 1990s. This positive by bursts of activity sandwiched trend in the rate of release over time is between long periods of inactivity. For one of the shared findings across the example, Sierra Leone released one commodity chapters in Evenson and variety of rice in 1964, five in 1978, and Gollin (2003b). However, beans, cassava 18 in 1988. Most, but not all, extreme and maize in ESA were the only cases in release behavior could be commodity groupings that truly fit the explained. positive-trend stereotype. Varietal output for the other crops peaked in the 1980s and was maintained at Updating varietal output in 2010 roughly the same level in the 1990s. 2. Political instability adversely affected Updating the database for the continuing varietal output in some crops in key crops and assembling fresh historical data countries in the 1990s. on varietal output for the new crops in 3. Some crops were characterized by high Table 3.1 (overleaf) broadly confirms the numbers of releases prior to 1975. A five findings cited above from the 1998 few countries could identify stable lines analysis. The historical data on varietal of research that generated early varietal output across the 20 crops contains output, which existed prior to and 3594 entries. continued immediately after the country declared independence. These About 90% of these have information on early positive performers also released the year of release. The undated entries in substantially more varieties from the the database are associated with modern mid-1970s to the late 1980s; however, materials that were judged to be available the advantage of an early start vanished to farmers or are located in countries that

26 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project do not maintain a formal release registry. the results to show that less extensively Many of these come from the IITA report grown crops are characterized by higher (Alene and Mwalughali, 2012a) and are levels of output intensity. Indeed, this listed as ‘informal’ releases. Participants expectation was confirmed for lentil, were encouraged to add escapes and other soybean, potato, and wheat, all of which adopted materials perceived as modern to were associated with strong market the release database so that information on demand. Additionally, during the mid-20th their identity and characteristics was avail Century, both wheat and potato benefited able (Walker, 2010). Most, but not all, the from a strong program of genetic dated entries in Table 3.1 imply official improvement thanks to the Rockefeller release. Program in Mexico. The genetic base for many released varieties in SSA came from Maize leads all crops with over 1000 entries that early work. in the cultivar-release database. Rice is a distant second. Both rice and maize in ESA At the other end of the spectrum, five have had access to multiple institutional crops fell under the low threshold of less sources of modern genetic materials. than 20 cultivars released per million ha of harvested area in 2009. Low research A simple index of output intensity can also intensities in pearl millet and sorghum have be constructed for comparative analysis translated into low output intensities. The across crops. In Table 3.1 below, output same finding applies to countries producing intensity is expressed in terms of total cowpea. Relatively few varieties have been releases per million hectares (ha) in 2009. released recently (Alene and Mwalughali, Similar to research intensity, we expected 2012a). A low estimated research intensity

Table 3.1. Counting the number of cultivars in the varietal release database by crop in SSA from before 1970 to 2011a

Number of released Number of cultivars cultivars with Output intensity Number of in the varietal year of release (total releases/ Crop countries release data information million ha) Banana 1 13 6 14 Barley 2 41 41 42 Bean 9 250 232 100 Cassava 17 355 207 32 Chickpea 2 27 26 108 Cowpea 17 200 157 17 Faba bean 2 28 28 46 Field pea 1 26 26 113 Groundnut 10 140 137 22 Lentil 3 15 14 158 Maize–ESA 8 692 664 47 Maize–WCA 11 330 271 33 Pearl millet 5 121 120 9 Pigeonpea 3 17 17 46 Potato 5 117 117 190 Rice 11 436 428 64 Sorghum 8 174 180 11 Soybean 15 201 156 170 Sweetpotato 5 89 89 60 Wheat 5 244 243 146 Yam 8 78 35 17 Total/average 148 3594 3194 68 a This count also includes the same cultivar released in different countries under a different name.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 27 for banana is derived from the observation (Table 3.2, below). The mid-point for data that hybridization is still difficult. More release was 1998. Decade by decade, the than all other crops in Table 3.1 (page 27), incidence of release has steadily increased low output intensity in yams is attributed to over time. historically negligible levels of research investment. However, not all crops fit the pattern of a steady rise in varietal output over time. In The parity between the output intensity of ESA, varietal output rose exponentially in cassava and that of maize in WCA is maize between the 1990s and the 2000s perhaps the most interesting finding in because of surging private-sector releases. Table 3.1. The total number of releases and On the other hand, groundnut displays a their total harvested area are almost identi- flat trajectory in output for more than four cal for the two crops. The example of decades and then output rises abruptly cassava suggests that low research intensity from 2000. Unfortunately, this increase in does not preordain mediocre performance releases is confined mainly to smaller in output. producing countries in ESA. Meanwhile, WCA is still associated with stagnation in the incidence of released varieties, e.g., Varietal output over time varietal output in cowpea has declined sharply from its peak in the 1990s. Tracking cultivar release over five time periods that mostly correspond to decades Three cereals have also not been able supports the anticipated finding that to maintain an increase in varietal varietal output has been increasing over production. Varietal output in pearl millet time. About 45% of the 3194 dated entries peaked in the 1980s. Meanwhile, varietal in Table 3.1 were released since 2000 performance in sorghum tapered off in

Table 3.2. The frequency of cultivar release by decade by crop in SSA

Released varieties and hybrids by decade

Crop Pre-1970 1970s 1980s 1990s 2000sa Banana 0 0 0 0 6 Barley 0 3 3 4 31 Bean 1 6 22 73 130 Cassava 0 2 31 61 113 Chickpea 0 3 2 9 12 Cowpea 3 8 49 65 32 Faba bean 0 3 2 8 15 Field pea 0 2 2 10 12 Groundnut 20 23 25 21 48 Lentil 0 0 4 5 5 Maize–ESA 7 10 34 159 455 Maize–WCA 12 25 75 76 82 Pearl millet 1 7 46 28 38 Pigeonpea 0 0 3 2 12 Potato 3 18 29 24 43 Rice 27 53 133 138 77 Sorghum 2 25 36 63 54 Soybean 2 13 32 52 57 Sweetpotato 0 0 9 20 60 Wheat 20 43 43 40 97 Yam 0 0 0 5 30 Total 98 244 580 863 1409

a The end year for the period is either 2009, 2010, or 2011 depending on the crop.

28 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project the 2000s. In spite of the widespread maize-growing countries released more introduction of the New Rice for Africa than 29 varieties during this recent period. (NERICA) varieties starting in the mid-1990s in most rice-growing countries in SSA, But, in general, releases were unevenly varietal release also slowed in rice in the distributed across all countries within each 2000s. Political instability and civil war in crop. Thirty country programs reported no Côte d’Ivoire and Sierra Leone severely releases, and 45% of the 148 crop-country curtailed releases caused by the closure programs released fewer than five varieties of several rice research stations. With the during the 12-year period. The country with exception of Senegal, West Africa shows the most releases often accounted for more a downturn in releases in the 2000s than one-third of the total releases and, in compared to the 1980s and 1990s. Even the case of yams in Côte d’Ivoire, the vast in Guinea, where varietal output exceeded majority of total releases. In contrast with 100 varieties in the 1980s and 1990s, cowpea, none of the 17 countries in the rice releases are becoming increasingly rare. dataset released more than ten varieties in the ten-year period. Releases in the post-1998 period are described in Table 3.3 (below). Five crops Returning to the top of Table 3.3, wheat’s have been able to maintain a simple position in weighted annual release rate average annual release rate of at least one was anticipated. Ethiopia is by far the variety released per program. Fueled by largest producer and recently has been Kenya’s and Zambia’s high production – prolific in varietal release, which explains with over 100 varieties released since 1998, why the weighted annual rate is substan- mostly by the private sector – maize in ESA tially higher than the simple annual rate. easily tops the list at five varieties released The release performance of the smaller per annum per program. Seven of the eight wheat-growing countries of Kenya,

Table 3.3. Performance in varietal release from 1999 to 2011 by country program

Annual release rate Total releases

Weighted by Crop Total releases Simple area Maximum Minimum Maize–ESA 485 5.1 5.1 143 0 Wheat 106 1.8 4.0 53 5 Barley 31 1.3 2.2 28 3 Bean 148 1.4 1.4 27 8 Maize–WCA 91 0.6 1.4 37 0 Yam 30 0.3 1.3 23 0 Cassava 128 0.6 1.2 20 0 Sweetpotato 66 1.1 1.1 28 1 Faba bean 15 0.6 1.0 14 1 Field pea 12 1.0 1.0 12 12 Chickpea 12 0.5 1.0 12 0 Potato 47 0.8 0.8 24 1 Sorghum 58 0.6 0.6 30 0 Banana 6 0.5 0.5 6 6 Rice 77 0.6 0.5 23 0 Cowpea 34 0.2 0.5 8 0 Soybean 61 0.3 0.4 16 0 Pearl Millet 39 0.7 0.4 17 1 Pigeonpea 12 0.3 0.4 6 2 Groundnut 46 0.4 0.4 9 0 Lentil 5 0.1 0.3 4 0

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 29 Tanzania, Zambia, and Zimbabwe has Leone, potato in Rwanda, and rice in slowed somewhat recently. Côte d’Ivoire. For other observations, the explanation appears to be country- or Ethiopia’s sustained efforts in varietal region-specific. Most of the observations release also explain barley’s ranking come from West Africa. As the balloons near the top of Table 3.3. Moreover, a in Figure 3.1 show, Nigeria accounts for decentralized regional research emphasis a large share of total area of all the obser- has reinforced release activities in Ethiopia. vations. Cowpea, groundnut, pearl millet, The buoyancy and productivity of the rice, and sorghum are well represented in aforementioned PABRA network – the Figure 3.1. With the exception of rice, these umbrella organization that oversees crops finished at the bottom of Table 2.2 three regional genetic networks in SSA (page 12) describing estimated research – contributed heavily to the release per intensities in 2010. formance of beans in the recent period. Sweetpotato programs also released varieties at a rate of more than 1% per The historical record on CGIAR annum. The fruition of a longstanding contributions to varietal output CIP-supported breeding program in Mozambique made a substantial The commodity centers of the CGIAR can contribution to this output. leverage varietal output through the direct distribution of elite material and their The lower end of Table 3.3 shows the same finished varieties, progenies for selection, crops that displayed lagging levels of and parents for direct crossing by NARS. human resources investment in genetic About 43% of the varieties released since improvement programs. The estimated 1980 in Table 3.1 (above) are related to the release rate for cowpea, groundnut, pearl work of the CGIAR. millet, and sorghum indicate one release per program every three to five years. The CGIAR contribution is greater than 40% for the majority of crops in Table 3.4 (page The low position of soybean for the recent 32). In several cases, two or more CG period in Table 3.3 is a surprise for an Centers contribute to varietal releases of expanding commercial crop from a very the same crop. Notable examples of joint small production base in most countries contributions include ICRISAT and ICARDA except Nigeria. Such countries are most for chickpea; IITA and CIMMYT for maize in likely following a cost-effective strategy of WCA; and AfricaRice, the International Rice capitalizing on finished materials from Research Institute (IRRI) and IITA for rice other tropical and semi-tropical countries, (before IITA closed its rice program). especially Brazil and Argentina. Neverthe less, those varieties should still appear in The six crops below the 40% contribution the varietal registries maintained by level in Table 3.4 are suitable candidates for countries in SSA. discussion about why their estimates are lower than those of other crops. Barley and Between one-fifth to one-quarter of the field pea are primarily grown in Ethiopia 146 crop-by-countries observations were and are researched in a strong NARS setting characterized by more releases in the 1980s where the crops have considerable genetic than in the 2000s. These observations are diversity as a locus of domestication. identified in Figure 3.1 (overleaf) by the number of releases in the 1980s and the Other institutional suppliers play a change in releases between the two large part in the reported estimates for periods. The observations included in this banana and maize in ESA. The Honduras dropline graph imply declining productivity Foundation for Agricultural Research (FHIA) in crop improvement over time. Some of has contributed significantly to the these observations were casualties of civil improvement of banana in SSA, especially war during the 1990s and early 2000s. in finding cultivars resistant to Fusarium Wilt – a soil-borne fungal disease – in the Civil war, as a major explanation for falling brewing, cooking, and dessert types of varietal output, applies to rice in Sierra banana.

30 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project 0 Rice Madagascar

Sorghum Groundnut Niger Niger Soybean Cowpea Nigeria -50 Cowpea Burkina Faso Nigeria Cassava Benin Sorghum Senegal Cowpea Benin Groundnut -10 Nigeria Sorghum Nigeria Rice Côte d'Ivoire Maize Benin Rice Nigeria

-15 Potato Rwanda

Rice Sierra Leone Change in releases between the 2000s and 1980s Change in releases Pearl millet Niger -20

0 5 10 15 20 Number of releases in the 1980s

Figure 3.1. Crop-by-country observations with more releases in the 1980s than in the 2000s.

Between 1958 and 2010, the private The 39% estimate for beans approaches the sector – without direct participation from average level of CGIAR contribution in other institutions – was responsible for Table 3.4. Multiple smaller institutional 56% of maize releases in ESA (De Groote providers have added a global perspective to et al., 2011). In Figure 3.2 (page 33), the CIAT’s primary role as a source of genetic CGIAR is credited with a 23% share of materials for the generation of bean varietal improved maize variety releases, output in ESA. These include the Bean and together with NARS and the private Cowpea Collaborative Research Support sector. This estimate is substantially Program in the USA, Institute of Horticultural higher than currently shown in the DIIVA Plant Breeding (IVT) in the Netherlands, database, but even a 23% contribution Zamorano Pan-American Agricultural School to varietal output is low compared with (EAP) in Honduras, Tropical Agricultural estimates for other crops in Table 3.4.9 Research and Higher Education Center Historically, the public sector’s (CATIE) in Costa Rica, National Vegetable contribution to varietal research declines Research Station (NVRS) – Wellsbourne when the private sector becomes Project in the UK, and the Tokachi established in cross-pollinated crops that Agricultural Experiment Station in Japan. can be readily hybridized (Fuglie and Genetic materials from the gene bank in Walker, 2001). The private sector is well Beltsville, Maryland, USA have also figured established in Kenya, Zambia, and prominently in several varietal releases. Zimbabwe where hybrids dominate the market. The Institut de Recherches Agronomiques Tropicales (IRAT) now Agricultural Research for Development (CIRAD) has played a large role in generating 9 Recently, CIMMYT has released more than 100 varieties in SSA as part of its Drought Tolerance for Maize in materials that have resulted in varietal Africa Initiative. change in several food crops in West

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 31 Table 3.4. The contribution of IARCs of the confirmed here. Between the 1980s and CGIAR to varietal output in SSA, 1980–2011 1990s the CGIAR share in varietal output Number of Share of rose from 42 to 46% (Table 3.5, page 33). dated released CGIAR-related But, contrary to our expectation, the role of varieties varieties to the CGIAR declined in the 2000s compared related to total dated with the 1990s.10 This decline could be Crop CGIAR activity releases in % attributed to the funding crisis in the mid- Chickpea 23 95.8 to late-1990s and early 2000s when the Lentil 13 86.7 exchange of germplasm and genetic Pigeonpea 14 82.4 materials became more constricted. The Potato 72 75.0 increasing rate of private sector releases in Yam 26 74.3 maize in ESA – especially in Kenya and Maize–WCA 173 74.2 Zambia with more than 100 releases since Cassava 143 68.1 2000 – has directly had a dampening effect Sweetpotato 59 66.3 on the CGIAR share. When maize in ESA is Cowpea 88 57.5 omitted, the revised estimate in the second Rice 179 51.4 row of Table 3.5 shows a plateauing of the Soybean 69 48.9 CGIAR contribution at about 56% in the Wheata 81 45.0 1990s and 2000s. Groundnut 41 43.6 Pearl millet 45 40.2 Faba bean 10 40.0 The maturity of crop improvement Bean 88 39.1 programs over time Sorghum 38 24.8 One of the relevant findings in the 1998 Maize–ESA 171 22.8 global Initiative (Evenson and Gollin, 2003b) Barley 8 21.1 concerned the sequential nature of the Banana 1 16.7 breeding process at NARS and International Field pea 4 16.7 Agricultural Research Centers (IARCs) over a The share estimate for wheat is understated because data collected time. The commodity centers in the CGIAR in the smaller producing countries did not contain information on began by sending out elite material and the institutional source of genetic material since 2000. finished products for testing. This stage was quickly followed by distribution of progenies for selection by NARS who later matured to the stage of crossing parental materials Africa. CIRAD also works on non-staple often provided by the commodity centers. crops and has historically placed less NARS also started the process with the emphasis on genetic enhancement than introduction of local landraces that they the CGIAR. But the relatively low level of ‘purified’ for subsequent distribution. CGIAR contribution to sorghum releases The transition from the introduction of in West Africa is not related to strong finished materials to progeny selection NARs in centers of diversity or to from introduced crosses to selection from alternative suppliers of material. The national crosses was not linear, but it was overly aggressive pursuit of a breeding well documented for several crops at the strategy focusing on shorter statured, global level in Evenson and Gollin (2003a). photoperiod-insensitive material is a This was especially true for stronger NARS plausible explanation for why ICRISAT’s with larger volumes of production, where it contribution is not greater, especially in made economic sense to enter the mature West Africa (Ndjeunga et al., 2012). phase. Direct crossing in the target country Farmers strongly prefer tall, photoperiod- was hypothesized to enhance prospects for sensitive Guinean types of sorghum. making progress when confronted with strong GxE interactions. The commodity centers in the CGIAR mostly date from the late 1960s and the early 1970s. We would expect to see a rising 10 In one sense, this could be viewed as progress: NARS and the private sector are taking on additional contribution from CG-related materials responsibilities, freeing the CGIAR to focus on basic over time from 1980. That expectation is research.

32 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project 300

263 250 <=1980 1981–1990 1991–2000 200 2001–2010

150 139 122 100

54 Number of maize varieties released 50 47 26 22 18 21 21 12 9 9 4 8 6 7 2 0 Private sector NARS and CGIAR NARS NARS and CGIAR and private sector private sector

Figure 3.2. Number of improved maize varieties released by decade and by origin (Source: DeGroote et al., 2011)

Table 3.5. IARC-related percent share estimates over time with and without maize in ESA

Basis for the estimation The 1980s The 1990s The 2000s Average share All crops and regions in the database 41.5 45.8 41.0 42.8 Without maize in ESA in the estimation 43.6 55.9 56.2 51.9

There is weak evidence for this transition in finished products from ICRISAT in the 1970s the DIIVA database, but it is not usually and 1980s,to progeny selection in the 1990s transparent as each crop seemingly has a and finally to direct crossing of ICRISAT different story to tell. For beans, the parental material in the 2000s. transition from the selection of local landraces to the selection from introduced Cowpea seems to be a counterfactual to progenies is marked in Figure 3.3 (overleaf). the transition hypothesis as the relative Since 2000, the majority of released shares between bred and parental material varieties have been derived by NARS from have stayed relatively constant over time introduced progenies. Direct crossing by (Figure 3.5, page 35). Cowpea is also a NARs and subsequent selection is still rare. counterfactual to the confirmed secular rise in the incidence of released varieties. The cassava story fits the transition stereotype quite well (Figure 3.4, overleaf). The In general, the incidence of direct crossing incidence of IITA-bred materials plateaued was less than expected in most crops in in the 1990s and the importance of NARS- the DIIVA Project release database. Even bred material from IARC parents rose from large NARS programs, such as rice in a small base in the 2000s. Nigeria, still rely heavily on introduced finished varieties, although they generated A weaker story applies to groundnut, and released varieties from direct crosses sorghum and pearl millet in West Africa. in-country as early as the mid-1980s. The data in Table 3.6 (page 35) are Releases from landraces continue to figure consistent with a gradual shift in the prominently in a sizeable minority of relative importance of basing varieties on programs in the 2000s.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 33 90 80 70 60 50 Crosses 40 Purification of local landraces 30 Number of varieties 20 10 0 1970–1979 1980–1989 1990–1999 2000–2010 Selected years

Figure 3.3. Trend in the release of bean varieties by germplasm source, 1970–2010 (Source: Muthoni and Andrade 2012)

140

120

100

80 IITA bred IITA parent 60 Non IITA 40 Total

Number of varieties released 20

0 1970–1979 1980–1989 1990–1999 2000–2010 Selected years

Figure 3.4. Trend in cassava variety releases by IITA content, 1970–2010 (Source: Alene and Mwalughali 2012a)

Table 3.6. Trend in the number of sorghum, pearl millet, and groundnut varieties released by germplasm origin, 1970–2010 (Source: Ndjeunga et al. 2012)

Year range

1970–80 1980–90 1990–2000 2000–2010 Total Parent-ICRISAT/cross NARS 0 0 0 5 5 Cross ICRISAT/selection NARS 0 5 11 2 18 Cross ICRISAT/Selection ICRISAT 4 17 14 5 40 Total 38 92 86 50 266

34 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project 70

40 IITA bred IITA parent 30 Non IITA 20 Total

Number of varieties released 10

0 1970–1979 1980–1989 1990–1999 2000–2010 Selected years

Figure 3.5. Trend in cowpea varietal releases by germplasm content, 1970–2010 (Source: Alene and Mwalughali 2012a)

Summary on civil war, such as Sierra Leone; the strength of several crop improvement Varietal release, interpreted broadly to programs in Nigeria in the 1980s; and the include improved materials that are or are weak scientific capacity in the recent past in supposed to be available to farmers, is West Africa of grain legume and coarse equated to output in this research. The cereal improvement programs. historical data on varietal release across the 20 crops approaches 3,600 entries. About About 43% of the varieties released since 90% of these have information on the year 1980 are related to the work of the CGIAR of release. Maize leads all crops with over – a proportion that has remained relatively 1,000 entries. Rice is a distant second. Both stable (40–45%) over the past three rice and maize in ESA have benefited from decades. In total, the CGIAR contribution multiple institutional sources of modern was greater than 40% for 14 crops. Viable genetic materials. By contrast, low research alternative international suppliers, a intensities in pearl millet, sorghum, and dynamic private sector, strong NARS, and cowpea in West Africa have translated into failed breeding strategies figure promi- low output intensities. nently as reasons why the CGIAR share is below 40% for banana, beans, barley, field About 45% of 3,194 dated entries had been peas, maize, and sorghum. released after 2000. The mid-point date for varietal release was 1998. Decade by The evidence was mixed for the develop decade, the incidence of release has ment of crop improvement programs. A increased steadily over time. Varietal few programs followed a transition over output rose exponentially in maize in ESA time that reflected increasing sophistication between the 1990s and the 2000s because in plant breeding research. These programs of surging private-sector releases. However, initially relied on landrace materials for not all crops in all countries fit the pattern varietal release. Subsequently they engaged of a steady rise in varietal output over time. in multi-location adaptation trials of Between one-fifth to one-quarter of the introduced elite materials. Finally, they 146 crop-by-country observations were selected varieties from their own crosses or characterized by more releases in the 1980s introduced progenies. But, for most, the than in the 2000s. Some plausible explana- evidence was fuzzy or not transparent that tions for declining varietal output centered programs had advanced in plant breeding

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 35 capabilities. More specifically, few released in contrast to the evidence that plant varieties traced their origins to crosses breeding capabilities increased steadily made by national crop improvement over time in Asia and Latin America programs. This negative finding stands (Evenson and Gollin, 2003a).

36 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project 4. Varietal adoption

Defining MVs and estimating For the few crop-by-country observations adoption where FAOSTAT was not used, the alter native area estimates may be higher or Arriving at reliable estimates of varietal lower than the FAOSTAT national figures. adoption of improved varieties is an For beans, the greatest variations in the important step towards assessing impact data sources were observed in Burundi, and providing a baseline for commodity- Tanzania, and Uganda. In Tanzania, for oriented collaborative research projects. example, 393,716 ha under beans were The use of different definitions of ‘adoption’ identified from agriculture production can lead to misleading results across com- census data for 2008/2009; this compares to modities and countries, as well as over time. 1,245,623 ha in FAOSTAT, when averaged Because establishing a reliable baseline was over three years from 2008 to 2010. In a major aim of the DIIVA project, ‘adoption’ Burundi, bean area was estimated at needed to be well defined. 405,715 ha by expert sources, whereas the FAOSTAT estimate was 203,367 ha. Overall, Harvested area is the basis for the adoption the expert panel estimated 19% of the estimates that follow, and has been taken total agricultural land in Burundi to be from the FAO production database with under beans, whereas FAO results imply some notable commodity and country that only 9% of area cultivated is put to exceptions. These are where national and beans. A large discrepancy was also regional surveys are believed by experi- observed in the data from Uganda; the enced crop scientists and technology Uganda Bureau of Statistics identified transfer practitioners to be more reliable. 532,883 ha under beans, whereas FAOSTAT Exceptions pertained to a few countries for reported 917,000 ha (Muthoni and potatoes and sweetpotatoes and to banana Andrade, 2012). in Uganda. For beans, area data from the CIAT Bean Atlas are used. Similar discrepancies between FAOSTAT alternative estimates have been noted in potato (Labarta, 2012). Compared to FAOSTAT, alternative estimates are characterized by substantially higher area estimates for potato in Ethiopia and substantially lower estimates in Malawi, where it is suspected that ‘sweetpotato’ historically has been classified as ‘potato’ in FAOSTAT.

Harvested area is a desirable measure because it is relatively easy to interpret in terms of production impacts. However, expert panels and focus group respondents in community surveys are much more comfortable giving adoption estimates in terms of the percentage of farmers using improved varieties. This is measured easily in surveys of farm households and does not require estimates of the area planted. It also says something about the access of individual farmers to technology. But Pearl millet OPV SOSAT C88 is the most extensively grown using the percentage of farmers leads to improved variety documented in the DIIVA Project. Arriving an overestimate of the use of specific at reliable adoption estimates of improved varieties is an improved varieties and cannot easily be important step in assessing impact. Photo: T. Hash/ICRISAT used to interpret economic impact.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 37 Furthermore, an area measure imposes the patory varietal selection from improved added discipline that area shares between materials, and breeding outputs in traditional and improved varieties must add countries that do not have a functioning to 100, and area shares among specific formal release and registry system. Focusing improved varieties must add to their only on released varieties would understate aggregate total. the performance of investments in crop improvement. What constitutes an improved variety is perhaps the most important aspect of esti- The issue of the frequency of seed renewal mating adoption levels. Reaching a robust for improved OPVs and hybrids in open- definition begs several questions: pollinated crops is one that needs to be addressed in the future. This aspect acquires Would the variety be available to farmers heightened importance in comparing esti- without research in crop improvement? mated adoption levels in maize between Are breeding and selection embodied in regions in SSA. Where survey data were readily identified materials that farmers available, i.e. maize in Ethiopia, an OPV was are growing but that have not achieved considered to be an MV if the age of the formal release? seed was three years or less. In the Ethiopi- Is the seed of improved OPVs renewed an nationally representative survey of periodically? varietal adoption, “maize seed was consid- Is the seed of hybrids renewed annually? ered an improved variety if the farmer used Should very old released varieties be freshly purchased hybrid seeds, freshly pur- included? chased Open-Pollinated Varieties, and recycled OPVs for not more than three pro- Responding to these questions often duction seasons” (Jaleta et al., 2013). requires more information on the seed sector as well as careful inspection of the Unfortunately, survey data on seed vintage varietal release database in the country in OPVs were only available in this one case. studied. Ideally, landrace materials should Older OPVs released in the 1970s and 1980s not qualify as improved varieties in their in countries where seed renewal from gov- country of origin, even though their seed is ernment or private sector sources is limited purified and they are formally released. were also deleted from the list of improved Farmers would most likely be producing varieties. Matuba, which was released in these materials with or without a crop im- 1984 in Mozambique, where civil war pre- provement program; although their identi- vailed until 1992, and where seed programs fication does require some effort in selec- are still not institutionally well developed, tion. Moreover, variety-specific comparisons is an example of an OPV that is now consid- show that productivity gains from released ered to be a local variety. In general, the in-country landraces are substantially problem of outcrossing in defining an lighter than those estimated for more improved variety is most pronounced in modern materials with greater breeding maize, pearl millet, and pigeonpea among content (Dalton and Guei, 2003). the 20 crops in this study.

For most crops, some released landraces Likewise, we have not confronted the issue have been included within the category of of varietal age with great analytical rigor improved varieties in our adoption esti- in defining MVs. Arbitrarily, we have used mates, but have been excluded for bean 1970, the year IR-8, the first Green Revolu- and sweetpotato where landraces are more tion semi-dwarf HYV, was introduced into important. Released landraces from other SSA (Dalrymple, 1986), as the cutoff point countries are also included on the assump- to define MVs. The use of 1970 as a cutoff tion that such materials would not be avail- point means that improved materials bred able to farmers without the intervention of in SSA during the colonial era are not adaptation trials by the national crop im- included. Using an earlier cutoff date, such provement program. as 1950 or 1960, would have resulted in markedly greater estimated adoption Our definition of an ‘improved variety’ is levels in groundnut and in rice in several inclusive of escapes, products of partici countries in West Africa. The adoption

38 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project results for most crops and countries are validation are described in detail in Section not very sensitive to the use of an earlier 6 of this report. cutoff date. Similar to the ‘1998 Initiative’, adoption es- In groundnut, the estimates based on a timates were mainly generated by elicting large survey in northern Nigeria and expert expert opinion. This may still be the only opinion in Mali suggests that two varieties cost-effective option for arriving at reason- bred in the colonial period are still able adoption estimates for SSA, which is extensively grown. Variety 55-437 is characterized by multiple food crops and a estimated to cover about 40% of diversity of countries. One hundred and groundnut-growing area in Nigeria; 47-10 eleven crop-by-country combinations in the is believed to be cultivated on about the DIIVA adoption dataset of 152 observations same percentage of groundnut area in Mali are based on expert opinion (Table 4.1, (Ndjeunga, et al., 2012). Groundnut below). Highly focused, nationally represen- varieties seem to have remarkable staying tative surveys account for 36 observations power in farmers’ fields. Groundnut is self- – 16 of these were financed and canvassed pollinated, has a low multiplication ratio by the DIIVA Project; 20 drew on comple- that results in slow diffusion, does not mentary research by other CG Centers and degenerate appreciably from seed-borne donors, especially AfricaRice’s Japan Project; viruses, and is not characterized by a high several others were inferred from recent lit- rate of somatic mutations. Variety 55-437 erature; and one observation, maize in still appears to be expanding in area in Tanzania, relied on variety-specific seed Nigeria (Ndjeunga et al., 2012). production information.

In rice, using a cutoff date of 1960 or 1965 would bring several popular introduced Adoption levels of improved purified landraces into play. Inclusion of varieties by crop in 2010 these materials in the set of modern cultivars would result in a sharp rise in the The area-weighted grand mean adoption adoption level of some large rice-growing level of improved varieties across the 20 crops agroecologies in West Africa (Dalton and in the project is 35% (Table 4.2, overleaf). Guei, 2003). They were listed, but not included, as improved varieties in the ‘1998 Two-thirds of the crop entries in Table 4.2 Initiative’; therefore, we opted for the same fall below the mean weighted level of course of action for consistency in compar- adoption of 35%. Starting at the bottom of ing estimates over time. the table, the low estimate for field pea, which is produced primarily in Ethiopia, is The source of the adoption information po- not surprising. Internationally and nation- tentially affects estimates in terms of their ally, field pea is arguably the crop species in variance and bias. With 20 crops in multiple Table 4.2 that has had the smallest amount countries, a uniform application of a of resources allocated to its improvement. protocol is needed to elicit information on In contrast, both chickpea and lentil have adoption (Walker, 2010). The same protocol benefited from international agricultural also needs to be used in the future to research in the CGIAR since the early- to generate valid time-series estimates. The mid-1970s. Although progress has been protocol used in the ‘1998 Initiative’ was adhered to as strictly as possible in this study and featured the elicitation of Table 4.1. Source of the national adoption adoption estimates based on expert estimates by number of observations. opinion. That protocol was administered by Source Number seven different institutional partners resulting in some variation. But, in general, Expert opinion 111 usable estimates were obtained. In the DIIVA adoption survey 15 small minority of cases where such informa- Non-DIIVA adoption survey 20 tion was incomplete, survey estimates were Inferred from the literature 5 relied on if they were nationally represen- Seed production and trade 1 tative. The expert opinion protocol and its Total 152

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 39 made, adoption of improved cultivars of anywhere in the world. A focus on disease both crops is concentrated in small pockets resistance is necessary, but entrenched con- of production regions in Ethiopia where ex- sumption preferences are potentially major tension programs have been active (Yigezu constraints to adoption which may be only et al., 2012a). This apparent location speci- partial in the best of circumstances (Kagezi ficity is typical of pulse crops, but it is sur- et al., 2012). prising in light of improved lentil varieties that have reportedly significantly heavier The National Banana Research Program yields than their local counterparts. of Uganda's National Agricultural Research Organization (NARO) also faces the chal- Adoption levels of faba bean and chickpea lenge that elite clones for evaluation were are buoyed by a reportedly higher only introduced on farms from 1991. NARO penetration of improved varieties in the has made a considerable commitment to Sudan. Indeed, chickpea in the Sudan is the biotechnology in order to exploit to the only crop-by-country observation to have fullest the opportunity for varietal change been at full adoption level in 2010, albeit and has mobilized several international on a very small area of 21,000 ha (Yigezu et partners in the supply of elite clonal material., 2012a). Meanwhile, Ethiopia harvests als. The potential for harnessing biotech- more than 0.5 million ha of faba bean, yet nology in Uganda for regional varietal the perceived adoption of improved change is a recurring theme that has been cultivars is very low at 3.5%. reported in the DIIVA Project for other clonally propagated crops such as cassava Cooking, dessert, and beer bananas in (Alene and Mwalughali, 2012a). Uganda are also characterized by a low rate of adoption. This finding is not surprising. Groundnut, sorghum, and pearl millet also Stimulating varietal change in a clonally fall below the adoption average of 35% in propagated crop – and one that is not an Table 4.2. They are produced extensively in annual – is a challenging proposition the Sahelian, Sudanian, and Guinean zones

Table 4.2. Adoption of MVs of food crops in SSA in 2010

Country Crop observations Total area (ha) Adopted area (ha) % MVs Soybean 14 1,185,306 1,041,923 89.7 Maize–WCA 11 9,972,479 6,556,762 65.7 Wheat 1 1,453,820 850,121 62.5 Pigeonpea 3 365,901 182,452 49.9 Maize–ESA 9 14,695,862 6,470,405 44.0 Cassava 17 11,035,995 4,376,237 39.7 Rice 19 6,787,043 2,582,317 38.0 Potatoes 5 615,737 211,772 34.4 Barley 2 970,720 317,597 32.7 Yams 8 4,673,300 1,409,309 30.2 Groundnut 10 6,356,963 1,854,543 29.2 Bean 9 2,497,209 723,544 29.0 Sorghum 8 17,965,926 4,927,345 27.4 Cowpeas 18 11,471,533 3,117,621 27.2 Pearl millet 5 14,089,940 2,552,121 18.1 Chickpea 3 249,632 37,438 15.0 Faba bean 2 614,606 85,806 14.0 Lentils 1 94,946 9,874 10.4 Sweetpotato 5 1,478,086 102,143 6.9 Banana 1 915,877 56,784 6.2 Field peas 1 230,749 3,461 1.5 Total/weighted average 152 107,721,630 37,469,577 34.78

40 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project of West Africa. All three crops share the of improved barley varieties in Ethiopia same poor country-specific outcomes in has slowly but steadily increased over time. terms of adoption – negligible diffusion Both improved food and malting barleys of improved varieties in Burkina Faso and have contributed substantially to MV no recorded adoption in Senegal where adoption (Yigezu et al., 2012b). varietal output has paled in comparison to the robust performance in Mali (Ndjeunga Cowpea adoption outcomes are dominat- et al., 2012). ed by the performance of crop improvement research in Niger and Nigeria which, Scientists in West Africa have also gone when combined, have a harvested area of down some blind alleys. For example, over 8 million ha. Niger is characterized by sorghum breeding overemphasized a harsh production environment and Caudatum types that could not compete unstable crop research featuring a high with the dominant Guinean materials level of donor instability. These conditions prevalent in the region (Ndjeunga et al., have resulted in an adoption estimate of 2012). Photoperiod-insensitive, short- 9% that has kept cowpea from entering duration Caudatum materials were high the top half of Table 4.2. yielding, but they were susceptible to pests, disease, and bird damage and did not According to FAO production data, yams measure up to the consumption expect have the highest calculated value of ations of semi-subsistence producers who production of any crop, including cassava also consume a sizeable share of their and maize, in SSA. This fact seems output. incredible because maize and cassava are usually considered the staple food crops in Additionally, groundnut crop improvement SSA, but an absence of crop improvement scientists in the Francophone countries have research targeted on a species as spatially to compete with old improved cultivars concentrated as yams does not seem grown prior to Independence. The afore- surprising. The 30% adoption estimate for mentioned groundnut variety 55-437, yams in Table 4.2 is attributed to a 75% released some 40 years ago, is still the outcome for improved varieties in Côte dominant variety in Senegal and even in d’Ivoire, the second largest producer in Anglophone Nigeria (Ndjeunga et al., West Africa. C18 is the prevalent variety. 2012). In Mali, groundnut varieties 47-10 Following its introduction in Côte d’Ivoire and 28-206 released in the 1950s are the in 1992, C18 expanded rapidly, covering most popular cultivars. large areas of yam cultivation where it sometimes represents 100% of the area In spite of the dearth of investment in the cultivated to Dioscorea Alata – otherwise improvement of these crops in West Africa known as ‘yellow’ or ‘water’ yam – one of as well as scientists’ ageing profiles, some six economically important yam species. C18 progress has occurred which has been is known for making tasty yellow porridge. below the radar for some time. SOSAT C88 – an improved, ICRISAT-related short- Both beans and sweetpotato partially owe duration pearl millet variety released in their position in the lower half of Table 4.2 1988 in Mali and Niger, and in 2000 in (page 40) to this study’s stance on excluding Nigeria – lays claim to an area slightly ex- released local landraces from the definition ceeding 1 million ha. This variety is grown of MVs. The adoption level for beans would in a larger area than any of the over 1,000 rise to 50% with a broadening of this improved adopted cultivars listed in the definition, while the adoption level of DIIVA database. Varietal change in ground- improved varieties of sweetpotato would nut in East Africa, especially in Uganda, is triple to 24%. another success story that was stimulated by an impressive partnership between Among grain legumes in Table 4.2, NARO, ICRISAT, and USAID’s Peanut CRSP. improved varieties of beans rank third in the adoption outcomes. Bean MVs are char- Barley, cowpea, and yams also appear in acterized by a substantially higher uptake the lower half of Table 4.2. Starting from a in Ethiopia than MVs for any other grain very low base of 11% in 1998, the uptake legume in the DIIVA Project; presumably

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 41 because Ethiopia has developed a vibrant superior performance. Following a longer- export industry for haricot beans. term CIP presence, Malawi has only recently released improved varieties that are now in In 1984, a regional breeding program was the very early phase of adoption. The established in the Great Lakes region of greater uptake of improved clones in SSA. It focused on breeding for resistance Ethiopia and Kenya has not compensated to bean pests and diseases in conditions of for the sharp downturn in the use of low and declining soil fertility typical of improved materials in Rwanda since the small rural household production. To meet 1994 genocide which destroyed not only this challenge, PABRA was launched as a the potato improvement program in CIAT project in 1996. It now consists of Ruhengeri – the hub of CIP activities in the three regional genetic improvement Great Lakes region (Rueda et al., 1996) – networks – the Eastern and Central Africa but also devastated an effective seed Bean Research Network (ECABREN), the program. Although potato is a priority food Southern Africa Bean Research Network crop, recovery in Rwanda has been slow for (SABRN) and West and Central Africa Bean improved clones, which were believed to be Research Network (WECABREN) – and close to full adoption in the early 1990s, encompasses 29 countries in SSA. PABRA prior to civil war. has a record of sustainability and growth that is matched only by a few other Cassava is perhaps the most surprising regional IARC-related crop improvement member of the set of seven crops with networks (Lynam, 2010). above-average adoption in Table 4.2. In spite of low levels of research intensity The sustainability of the PABRA umbrella documented in Section 2 of this paper, the network has strongly influenced these performance of cassava crop improvement positive outcomes for adoption in a crop has been solid and steady with regard to that is often characterized by niche adoption outcomes. The majority of the specificity in terms of production conditions countries included in this study have and market preferences. Identification of substantially higher levels of uptake of improved bean varieties in farmers’ fields is improved varieties now compared to the an onerous undertaking. With a few late 1990s (Alene and Mwalughali, 2012a). notable exceptions, improved bean A strategy that has emphasized high yield varieties are believed to account for only combined with disease resistance in a small areas in most countries, thereby mostly sweet, rather than bitter, background making the validation of such spatially seems to have yielded good dividends in fragmented expert opinion a difficult task. many countries. Additionally, donors have actively supported programs to propagate In the 1970s and 1980s, little research and widely distribute improved planting was conducted on sweetpotato in SSA. materials. Sweetpotato owes its rather modest position in Table 4.2 to a stable and sustained The location of pigeonpea in the top half of breeding effort in Uganda and Mozambique Table 4.2 was also expected. All three study (Labarta, 2012). Interest in orange-fleshed countries in East Africa have a commer cial sweetpotato for its high beta-carotene demand for high-yielding medium-duration content has also helped to stimulate and types that are well adapted to bi-modal marshal investment in what was once a seasonal rainfall in Kenya, Malawi, and relatively neglected secondary food crop in Tanzania (Simtowe and Mausch, 2012). SSA. The adoption of improved varieties in Table 4.2 is split about equally between Maize in ESA benefited from the large white- and orange-fleshed varieties. number of released varieties stimulated by liberalization policies and private sector Adoption of potato MVs are at the mean investment in maize breeding. As discussed level in Table 4.2. Given the crop’s market in the previous section, varietal output orientation and rapidly increasing growth borders on prodigious in some countries, rate in SSA over the past two decades, an such as Zambia, which has enacted policies adoption level that approaches the mean strongly favoring maize production. value across all crops could not be termed However, excellent performance in Zambia

42 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project and Malawi has not compensated for the the largest producer in SSA. Wheat would lack of tangible progress in Angola and be likely to occupy a higher position in Mozambique. In Angola, the dominant Table 4.2 if reliable data on adoption had released cultivars only account for about been collected for Kenya, Tanzania, 5% of the area planted and date from the Zambia, and Zimbabwe. These countries mid- to late-1960s prior to Independence. were at the level of full adoption of wheat MVs in 1998. Assuming full adoption in Adoption outcomes seem to be at a moder- 2010 is eminently plausible, as wheat in ately high level for rice, which is grown in these four countries is mainly produced in well-defined agro-ecological settings large farms with irrigation. The inclusion of throughout SSA. Aggregate adoption levels these four countries results in a rise in the still depend heavily on what happens in adoption estimate to 70%, which is still Nigeria and Madagascar, countries that substantially lower than that for soybean in together account for more than half of the Table 4.2. The limited penetration of rice-growing area in the 14 countries improved Durum varieties into farmers’ studied that had data available on this fields in Ethiopia is a major constraint to aspect. Aggregate adoption levels also full adoption of wheat HYVs in SSA. hinge on adoption outcomes in the rainfed lowlands and the uplands. The aggregate Soybean ranks first in our crop adoption level is also sensitive to adoption outcomes table. Soybean is a new crop characterized in Guinea, which arguably has released by strong market demand. Genetic more varieties with less ensuing adoption materials are mostly imported from abroad; than any other of the 152 crop-by-country sufficient time has not elapsed to allow national adoption observations. Recent many local landrace materials to develop. gains in adoption in several countries Although improved soybean adoption appear to have been driven by a positive levels are not surprising, their varietal age response from farmers to the NERICA vari- is – as discussed in Section 5. Given eties (Diagne et al., 2012). More than any soybeans’ scope for global expansion, the other crop, rice was negatively affected by crop seems to be taking its time in finding a the decision to define MVs from 1970 – an home in farmers’ fields in SSA. Nigeria still earlier starting date in 1960 would have led harvests more soybean area than the other to higher adoption levels, but this points to 12 countries in Table 4.2 combined. the continued use of very old varieties.

Maize in WCA secures the second spot Adoption rates by country in adoption performance in Table 4.2. Improved maize varieties in WCA gained Aside from the Central African Republic’s more ground in adoption than any other second place ranking – attributed to the crop in SSA between 1998 and 2010. These adoption of rice MVs – there are relatively gains were accomplished without significant few counterintuitive findings in the private sector input (Alene and Mwalughali, adoption estimate by country rankings 2012a). Most of these gains were recorded (Table 4.3, overleaf). One is the relatively via the adoption of OPVs. Some of these high placing of the Democratic Republic of are getting older and undoubtedly not all the Congo in achieving an above-average farmers renew seed in a timely fashion, adoption outcome across all crops in spite raising questions about the sensitivity of our of stagnating institutional and economic definition of improved varieties. Factoring development. in seed renewal rates would lead to a lower adoption estimate, but the uptake of The five countries at the bottom of Table 4.3 improved maize varieties would still be all have a weighted adoption estimate very impressive (Alene et al., 2009). below 15%. Burkina Faso is the outlier with a high adoption performance in maize and Wheat topped the crop adoption table in rice. Burkina Faso is also the first adopter of 1998. The increasing transition in area from Bacillus thuringiensis (Bt) cotton varieties durum to spring bread wheat was one of aside from South Africa. Burkina Faso’s the factors leading to the higher adoption position is attributed to negligible adoption of improved varieties in Ethiopia – by far of groundnut, sorghum, and pearl millet

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 43 Table 4.3. Weighted area adoption levels by country in SSA in 2010

Country MV adoption % Number of crop observations Zimbabwe 92 4 Central African Republic 72 1 Cameroon 68 6 Zambia 67 6 Kenya 63 8 The Gambia 56 1 Côte d’Ivoire 55 6 Ghana 53 6 Benin 52 6 Malawi 47 8 Senegal 45 6 Sudan 41 4 Nigeria 41 9 DR Congo 36 6 Madagascar 35 1 Mali 35 6 Ethiopia 33 9 Uganda 33 11 Tanzania 32 10 Guinea 29 5 Togo 22 6 Rwanda 21 4 Angola 17 2 Sierra Leone 16 1 Burundi 14 4 Niger 14 4 Eritrea 13 2 Burkina Faso 13 6 Mozambique 13 5

MVs. Other countries, like Angola, Mozam- regional and cultivar-specific information. bique, and Niger, have uniformly low rates The other 13% refers to aggregate of adoption of improved cultivars across all adoption only at the national level. crops. The regional and cultivar-specific database Optimism is warranted about the prospects accounts for slightly over 33 million ha. for enhancing adoption in such countries Adopted area is attributed to named as Ethiopia, Mali, and Uganda that are now and unnamed varieties where they are characterized by average levels for SSA as available. Unnamed varieties are aggregat- a whole. However, attaining a moderately ed into a category called ‘other’. Every high adoption rate of 50% as a hypotheti- effort was made to minimize the number cal development goal by 2020 will be a of varieties in the ‘other’ category. Most of daunting challenge, unless adoption pros- the specific entries come from survey data pects improve markedly for countries in and refer to names that are believed to be the bottom half of the table. MVs, but that could not be linked to a specific released variety. A few of the observations based on expert opinion also Named MVs in the adoption database have a small residual ‘other’ category.

About 87% of the MV adopted area is There are 1173 named releases in the associated with detailed data containing cultivar-specific database. They account for

44 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project 98% of the 33 million ha described above. some crops. Value of production is an The size distribution of area planted with important criterion because varietal these varieties is heavily skewed, consistent change in crops with more attractive with previous findings in the "1998 prices and/or higher base yields has the Initiative" for maize in ESA, potato, rice, potential to generate greater net benefits and wheat. Most of the varieties are grown per ha of adopted area. By either criterion, on small areas, whereas the median-sized the top 100 varieties account for about variety distribution is cultivated on about 60–65% of the total adopted area and 7000 ha. In total, 250 entries were adopted value of production of all adopted on less than 1000 ha. The 75th percentile of varieties. the cumulative distribution occurs at about 22,000 ha. Only 76 varieties exceed 100,000 Based on a value criterion, the share of ha of adopted area. cereals in the top 100 falls and the share of vegetatively propagated crops rises Few, if any, of these varieties could be dramatically. According to FAO production called mega-varieties with potential to data, one ha of cooking banana, yams, or cover tens of millions of ha, such as the potato can be worth the equivalent of rice variety Swarna (also Sona Masoori 25–30 ha of sorghum and pearl millet in and Samba Masuri, among other names) value. Therefore, it is not surprising to see that is grown in South Asia. The most relatively small areas of improved clones extensively grown variety is SOSAT C88 – of these crops claim a larger share in the the leading pearl millet cultivar in Nigeria top 100, when value of production is and the second-ranking improved variety the criterion. Indeed, a small majority in Mali. It is one of the subjects of impact of the varieties in the top-value 100 are assessment in the DIIVA Project (Ndjeunga vegetatively propagated. et al., 2011). The top ten ranking varieties are listed in Most of the more extensively grown or Table 4.4 (below). Cereals dominate the more economically valuable improved area classification. Only SOSAT C88 makes varieties are concentrated in a small it into the top 10 when the categorization subset of crops and countries. The value is based on value. Under either criterion, of production estimates complement Nigeria contributes more varieties than all harvested area in describing the economic other countries combined. Aspects of importance of adopted varieties. Value several of these economically important of production adjusts for heavier yields varieties are described in the next section leading to the more attractive prices of on spillovers.

Table 4.4. Top-ranked varieties by commodity and country by area and value of production

Area Value

Rank Name Crop Country Name Crop Country TMS 30572 1 SOSAT C88 Pearl millet Nigeria Cassava Nigeria (Nicass 1) 2 Wad Ahmed Sorghum Sudan C18 Yams Cote d’Ivoire 3 Oba 98 Maize Nigeria TDr 89/02660 Yams Nigeria TMS 30572 TMS 4(2)1425 4 Cassava Nigeria Cassava Nigeria (Nicass 1) (Nicass 2) NR 8082 5 ICSV 111 Sorghum Nigeria Cassava Nigeria (Nicass 14) 6 Kubsa Bread wheat Ethiopia TDr 89/02602 Yams Nigeria 7 ICSV 400 Sorghum Nigeria TDr 89/02665 Yams Nigeria 8 Suwan 1-SR Maize Nigeria SOSAT C88 Pearl millet Nigeria 9 Tabat Sorghum Sudan Sadisa (91/203) Cassava DR Congo Afisiafi 10 C18 Yams Côte d’Ivoire Cassava Ghana (TMS 30572)

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 45 Spillovers in adoption Confirming the potential for spillovers, the products of older regional crop Although the history of crop improvement improvement programs are still visible in research is marked by spillovers in adoption their respective geographical sphere of in SSA, spillovers are not the first thing that influence. The Armani Regional Station, comes to mind when thinking of adoption now in Tanzania but at one time covering outcomes in the harsh rainfed production all East Africa, has been the location for environments of Africa. Adaptability research that has led to long-term spillovers appears to be restricted by low fertility in since the 1950s and 1960s in cassava and environments characterized by seemingly sweetpotato materials as progenitors and high levels of location specificity. in a few cases as finished elite clones. Researchers at Armani developed the Positive evidence for spillover outcomes sweetpotato variety known as ‘Tanzania’ in was well documented in the colonial era in Uganda and Rwanda, as ‘Sinama’ in SSA. For example, in collaboration with Tanzania, as ‘Enaironi’ in Kenya, as ‘Kenya’ the British, scientists in Sierra Leone had in Malawi, as ‘ADMARC’ in central been working to increase regional rice Mozambique, and ‘Chingovwa’ in Zambia production in the difficult mangrove agro- (Labarta, 2012). In the five countries ecology since 1934. The locus of their included in the CIP study, this variety is activities – curtailed in the 1990s because estimated to be cultivated on an area of the civil war – was the Rokupr Rice approaching 200,000 ha, equivalent to 13% Research Station. Before Independence of the total sweetpotato area. (Because of this was known as the West African Rice its age, ‘Tanzania’ is not considered in the Research Institute and its mandate was to set of improved varieties.) It combines high promote spillovers. Several of the released dry matter and a marked preference in East ROK rice varieties became popular, not Africa with a strong background of virus- only in Sierra Leone but also in Guinea and resistance in the Great Lakes Region. Guinea Bissau. They have also been the subject of adoption studies and impact In many of the study crops within the DIIVA assessments (Adesina and Zinnah, 1993; Project, researchers have been able to Edwin and Masters, 1998). identify more recent examples of spillovers, where investing in varietal improvement in The case of the high-yielding, late-maturing one country has benefited neighboring maize hybrid SR 52 – the world’s first triple- countries or other countries in SSA. cross hybrid grown commercially – released Spillovers in adoption are not as common in the early 1960s in Rhodesia is a well- as spillovers in releases, but they are very known example of varietal output that visible when they occur. generated benefits to neighboring countries in southern Africa (Eicher, 1995). A lesser IITA researchers were able to describe in known example after Independence focused detail the occurrences of spillovers in on late blight-resistant potato cultivars in adoption for all five of their mandated the Great Lakes region of East Africa. In the crops in the DIIVA Project (Alene and early 1970s, three late blight-resistant Mwalughali, 2012a). In cassava, TMS 30573 varieties – at the time, recently released occupies 17.8% of the total cassava area from Mexico – were imported into Uganda in Nigeria, 17.5% in Uganda, 7% in Benin, and Kenya via the Rockefeller Foundation. and 3.2% in Guinea. Though not officially Although these varieties never laid claim to released, the same clone is also being a significant area in Mexico, they quickly grown extensively in Kenya, where it covers became popular in several smaller countries 24% of the cassava area and, to a much in East Africa. Before the 1994 genocide in lesser extent, is produced in Côte d’Ivoire. In Rwanda, Sangema was the dominant variety cowpea, popular multi-country varieties in Rwanda and was arguably the most include: IT82E-32, covering 23% of the total economically important in the ESA region in cowpea area in Ghana, 11% in Benin, and the 1970s and early 1980s. Even today, 2% in Cameroon. This is followed by Rosita, a synonym for Sangema, is the VITA-7, accounting for 22% of total cowpea prevailing potato variety in Malawi and area in Guinea, and 13% in DR Congo Mozambique. (Alene and Mwalughali, 2012a). The

46 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project adoption level for variety IT81D-1137 is zones of West Africa cut across broad estimated at 17% in DR Congo and 14% in swathes of several countries. This makes for Benin. These varieties are attractive to more homogeneous agro-ecological condi- farmers because they feature high yield tions going from west to east across coun- potential, good disease tolerance and short tries than from north to south within the duration. same country. The incidence and size of spillovers also varies by crop and is lower in In maize, Obatanpa – derived from quality beans and higher in potatoes in East Africa. protein maize (QPM) materials – and In ESA, spillover events in maize were not TZEE-Y fit the description of spillover as large; although they were probably un- varieties that have crossed over the borders derestimated because of incomplete and of several countries in WCA (Alene and low-quality data. SC 627 is a hybrid that Mwalughali, 2012a). Two improved soybean scores well on wider adaptation and is varieties are also widely cultivated in the grown extensively in Tanzania and Malawi region. Firstly, TGx 1448-2E – a shattering (De Groote et al., 2011). and frog-eye, leaf-spot resistant IITA-bred variety – is sown on more than 60% of In West Africa, the incidence of spillovers soybean area in Nigeria and on more than also varies from crop to crop. Spillover vari- 20% of harvested area in Cameroon and eties are readily visible in pearl millet and Ghana. Secondly, TGx 1835-10E – another groundnut but less so in sorghum. The IITA-developed variety that is desired pearl millet variety SOSAT C88 mentioned for its early maturity and resistance to previously has been adopted in four West soybean rust, pod shattering and lodging – African countries. Similarly, the groundnut dominates soybean areas in Uganda (50%) variety Fleur 11 is also spreading in West and covers 26% of soybean area in Kenya Africa from Senegal to Mali and Niger as well as 6% in Nigeria. (Ndjuenga et al., 2012).

In yams, examples of large spillover effects The emphasis on spillover varieties in this are harder to find, but a few improved subsection does not detract from the em- cultivars are found in two countries. Florido pirical fact that the varieties selected and is planted in Benin and Togo, and used solely within a country are likely to TDr 89/02665 is propagated in Ghana and contribute far more to total adopted area Nigeria in 5–10% of the total planted area. in SSA than multi-country varieties. Moreover, as pointed out earlier in this Groundnut seems to be the exception to section, none of the identified spillover the finding that the prevalence of wide varieties can yet be called mega-varieties, adaptability and spillover varieties is such as the rice variety Swarna Masuri sown uncommon in ESA. Similarities in the results on millions of hectares in South Asia. were less numerous than contrasts between the two regions. In four of the five ground- The identification of definite spillover vari- nut study countries in the ESA region, ro- eties serves mainly as a reminder that small sette-resistant ICGV-SM 90704 and drought- NARS can still reap some benefits from tolerant ICGV 83708 ranked first or second national and international research. A in the adoption of improved varieties. stable crop improvement presence in the region can generate returns that far exceed Finally, in rice, NERICA 1 is presently grown national benefits for the investing country. in five of the 12 producing countries with cultivar-specific information in the DIIVA adoption database. Earlier, BG 90-2 from IARC-related adoption Sri Lanka was a commonly introduced cultivar that was released by the majority of Most IARCs have been heavy contributors rice-producing countries in West Africa and to the varietal change that has taken later became popular in several countries. place in their mandated crops in SSA (Table 4.5, overleaf) – about 22% of the The incidence of spillover varieties appears area harvested is in IARC-related genetic to be higher in West Africa than in East materials. The relative importance of those Africa. The Sahelian, Sudanian and Guinean materials approaches two-thirds of total

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 47 area in improved varieties. non-CG institution and in any crop in the DIIVA Project, CIRAD (IRAT) has had a The crops in Table 4.5 are ordered by the marked impact on the adoption of rice MVs difference between their estimated share in several countries of Francophone Africa, in varietal output and adoption. It is including Madagascar. This important refreshing to see sorghum, pearl millet, institutional connection is a plausible and groundnut at the head of this table. explanation for why rice does not rank Released varieties of these crops may have higher in Table 4.5. Likewise, the small had somewhat limited acceptance by negative value of maize in ESA could be farmers (see Table 4.2 on page 40), but attributed to the late start by CIMMYT in IARC-related cultivars have had better the region, and to alternative suppliers in adoption outcomes than most in a difficult the burgeoning private sector. rainfed production environment.

The mean weighted difference between Comparing adoption levels between the adoption and release shares is 20%, 1998 and 2010 which is higher than expected. The crops towards to bottom of Table 4.5 are The 1998 benchmark provides a basis for relatively new to crop improvement carrying out a before-and-after comparison research in the CGIAR, so we did not of the level of varietal adoption for the ten anticipate having high shares of IARC- continuing crops in the DIIVA Project partnered adoption. (Table 4.6, overleaf).

Perhaps more than any other international On average, the 61 observations represent

Table 4.5. The contribution of the CG Centers to MV adoption in SSA in 2010

Adoption Release Difference between adoption and Estimated IARC-related release shares Crop adoption (%) (%) Share IARC (%) Share IARC (%) (%) Sorghum 27.4 20.6 75.0 24.8 50.2 Pearl millet 18.1 15.7 86.6 40.2 46.4 Groundnut 29.2 25.0 85.8 43.6 42.2 Bean 29.0 23.5 81.0 39.1 41.9 Wheat 58.5 37.7 64.5 45.0 19.5 Banana 6.2 2.2 34.9 16.7 18.2 Potato 34.4 31.2 90.8 75.0 15.8 Sweetpotato 6.9 5.6 81.3 66.3 15.0 Cassava 39.7 32.7 82.5 68.1 14.4 Soybean 87.9 55.6 63.2 48.9 14.3 Lentil 10.4 10.4 100.0 86.7 13.3 Cowpea 27.2 18.1 66.7 57.5 9.2 Maize–WCA 65.7 53.0 80.6 74.2 6.4 Chickpea 15.0 15.0 100.0 95.8 4.2 Barley 32.7 7.5 23.0 21.1 1.9 Pigeonpea 49.9 41.8 83.9 82.4 1.5 Rice 38.0 19.2 50.6 51.4 -0.8 Maize–ESA 44.0 12.9 29.4 30.3 -0.9 Field pea 1.5 0.0 0 16.7 -16.7 Yams 30.2 15.1 50.0 74.3 -24.3 Faba bean 14.0 0.5 3.7 40.0 -36.3 Weighted averagea 35.25 23 65.6 45.5 20.0

a Weighted by total area, except the share in adoption estimates that are weighted by total adopted area in each crop

48 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Table 4.6. Change in MV adoption between 1998 and 2010 in ten food crops in SSA

1998 2010 Relative importance in 2010 (% area Number coverage of paired MV adoption MV adoption of paired Crop observations Area (ha) (%) Area (ha) (%) observations) Barley 1 897,360 11.0 91,3863 33.8 86 Bean 6 1,738,000 14.6 1,903,964 35.1 45 Cassava 15 8,777,800 21.0 10,033,995 42.0 81 Groundnut 3 496,517 12.6 724,019 56.7 7 Maize 19 18,566,300 25.6 24,366,088 52.8 91 Pearl Millet 1 1,285,540 22.0 1,520,440 31.1 9 Potatoes 4 353,852 49.2 569,921 37.1 60 Rice 7 3,639,110 48.4 3,787,146 36.5 44 Sorghum 4 12,711,129 19.3 13,354,489 32.4 58 Wheat 1 1,330,000 56.0 1,453,820 63.5 84 Total/weighted 61 49,795,608 25.0 58,627,745 43.9 55 average

about 55% of the area of the crops grown and in Tanzania and Uganda were almost in SSA. Coverage is adequate in eight of the entirely fueled by the spread of improved ten crops to draw inferences about varietal OPVs. In general, the cassava-growing change between 1998 and 2010. However, countries were characterized by lower coverage is too scanty to make inferences adoption levels in 1998 than the maize- about progress in varietal uptake in producing countries; but, aside from groundnut and pearl millet. Tanzania, every cassava-producing country displayed a propensity for the greater Two important empirical facts emerge from uptake of improved clones in 2010 than Table 4.6 (above). First, the level of varietal in 1998. adoption was 25% in 1998. Secondly, and more importantly, MV adoption increased A potentially relevant aspect of this before- at a rate equivalent to a linear annual gain and-after comparison centers on the accrual of 1.45 percentage points over the 13-year of adoption gains by level of adoption in period. 1998. The difference in adoption between the two periods is negatively associated with With the exception of rice and potatoes, all the magnitude of adoption in 1998. crops experienced an expansion in the use Countries that commenced with levels of of MVs. Uptake was especially robust in adoption equal to, or below, 40% tended to barley, beans, cassava, and maize, with accrue more increments in adoption. Those adoption levels doubling during the period. that started with moderately high rates of adoption of improved varieties were hard The before-and-after data points for the pressed to achieve even more positive primary staples, maize and cassava, are outcomes in adoption. We expect this type arrayed in Figure 4.1 (overleaf) where the of behavior when a country approaches full balloons in the droplines are weighted by adoption, but not when it is at such a area in 2010. Maize in DR Congo was the moderate to high level of MV acceptance as only crop-by-country observation to improved maize cultivars were in Burkina experience a steep decline in the estimated Faso, Ghana, and Kenya in 1998. adoption rate between 1998 and 2010. Gains in the uptake of maize hybrids were Lack of progress in countries with already significant in Zambia and Malawi. Hybrids moderately high rates of adoption indicates also played an important role in Ethiopia. the existence of marginal production Increases in the West African countries regions where MVs do not compete favor-

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 49 60 Benin Zambia Nigeria Malawi Cameroon Maize Guinea Mali Uganda Zambia Cassava Zimbabwe 40

Tanzania Benin Malawi Kenya Togo

Nigeria DR Congo

20 Ethiopia

Angola Zimbabwe Ghana Senegal Uganda Ghana Togo Côte d'Ivoire Guinea Cameroon Burkina Faso 0 Tanzania Mozambique Angola Kenya Change in adoption between 1998 and 2010

DR Congo -20

0 20 40 60 80 100 Adoption (% area) in 1998

Figure 4.1. Change in the estimated level of adoption of improved maize and cassava varieties between 1998 and 2010

ably with traditional varieties on a few im- devastation in Rwanda caused by the 1994 portant characteristics. It will be interesting genocide, which did not predate the ‘1998 to see if the new entrants in moderate-to- Initiative’ because the adoption estimates high adoption group in Figure 4.1 (above) for Rwanda referred to 1993. will be able to consolidate and expand on their gains in the next version of the DIIVA In contrast, the estimated deteriorating Project. position of MVs in rice could be attributed to a change in methods. Expert opinion Comparable before-and-after data on the panels were used to generate all the esti- remaining crop in Table 4.6 (on page 49) mates for rice MVs in 1998. Surveys funded are presented in Figure 4.2 (overleaf). Many by the Japan International Cooperation relatively small producing countries made Agency (JICA) were deployed by researchers relatively large gains in the adoption of in AfricaRice to arrive at nationally repre- beans and groundnut. Sorghum in the sentative estimates of MV adoption in Sudan was the largest crop-by-country 20 African countries from 2008 to 2011. combination to register appreciable gains in adoption. If progress in MV adoption was slow, switching methods could be sufficient to Unlike cassava and maize in Figure 4.1, the change a small positive outcome to a relative importance of MVs declined in meager negative consequence. several countries between the 1990s and 2010. In particular, the adoption estimate Similar to the evidence presented in for improved clones of potato decreased Figure 4.1, countries characterized by sharply from 97% of the harvested area in moderately high levels of adoption in 1998 1993 to 35% in 2010. As discussed, potato had a hard time maintaining these levels, MVs became less important because of the let alone achieving gains in adoption. Rice

50 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project BE Malawi

50 G Malawi BE Tanzania G Uganda BA = Barley BE Ethiopia G Zambia BE = Bean S Tanzania R Côte d'Ivoire G = Groundnut PM = Pearl Millet 25 BA Ethiopia R Senegal P = Potato BE Uganda R = Rice S Sudan S = Sorghum P Ethiopia P Kenya W = Wheat PM Mali W Ethiopia BE Rwanda S Nigeria S Mali 0

R Sierra Leone R Nigeria P Uganda R Guinea R Ghana -25

BE DR Congo

R Mali Change in adoption between 1998 and 2010 -50

0 20 40 60 80 100 Adoption (% area) in 1998

Figure 4.2. Change in the estimated level of adoption of improved bean, groundnut, pearl millet, potato, rice, sorghum, and wheat varieties between 1998 and 2010

in Senegal and, to a lesser extent, wheat in Summary Ethiopia are the only two crop-by-country observations that exhibited substantial The area-weighted grand mean adoption gains in adoption from ‘moderate’ to level of improved varieties across the 20 ‘high’. Gains in adoption were concentrat- crops is 35%. The distribution of adoption ed at the lower end of the x axis in of improved varieties is skewed as 14 of the Figure 4.2 in much the same manner as crops are characterized by a mean adoption very positive outcomes were clustered in level that falls below 35%. Crops with an the same region of Figure 4.1. estimated adoption performance superior to the overall average included soybean, About 90% of the paired observations in wheat, maize, pigeonpea, cassava, and rice. Table 4.6 on page 49 showed an increase About 23% of the 35%, i.e., a share of in the uptake of improved varieties 65% of MV adopted area is related to (Figures 4.1 and 4.2). Again, disadoption IARC-contributed genetic materials. The and/or overestimation of MV adoption IARC-related share in adoption is about levels in 1998 occurred mainly in potatoes 20% higher than its 45% contribution to and rice. The finding of a few cases of released varieties. disadoption is unexpected because the ending of fertilizer subsidies is frequently The problem of lagging countries was also mentioned as a motivation for reversion to evident in the cross-sectional adoption local varieties. The evidence for disadoption estimates based on 152 crop-by-country is sparse, particularly for maize, which is a observations. Adoption of modern varieties relatively intensive user of fertilizer among was uniformly low in Angola, Mozambique, the food crops in the DIIVA Project. and Niger across all crops.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 51 Spillover varieties were prevalent, but, 95% confidence interval for the 17.6% unlike in South Asia, so-called mega- gain in adoption was 12.3%–22.9%. varieties that claimed millions of hectares Over 90% of the observations experienced of arable land were not found. Of the over a rise in MV adoption, which increased 1000 improved varieties listed in the DIIVA at a rate equivalent to a linear annual adoption database, SOSAT C88, a short- gain of 1.45 percentage points over the duration pearl millet variety, was the most 13-year period. With the exception of extensively cultivated on just over one rice and potatoes, all crops experienced million ha in Nigeria, Mali, Niger, and an expansion in the use of modern Burkina Faso. In maize, Obatanpa, derived varieties. Uptake was especially robust in from QPM materials, and TZEE-Y, an IITA- barley, cassava, and maize with adoption bred variety, fit the description of spillover levels more than doubling during the varieties that have crossed over the borders period. Civil war and changing methods of several countries in WCA. The incidence in measuring adoption loom large as of spillover varieties appears to be higher plausible explanations for why improved in WCA than in ESA and in groundnut than varieties lost ground in a few countries in other crops. and a few crops. Crop-by-country observations with a low level of MV The paired comparison of 61 crop-by- adoption in 1998 were more likely to country observations for the 10 continuing experience positive outcomes in adoption crops showed that area-weighted adoption than those with moderate levels of was 27% in 1998 and 44% in 2010. The adoption in 1998.

52 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project 5. Varietal change

The importance of estimating be factored into the evaluation. Having varietal change rapid varietal turnover with less than 10% adoption does not imply significant The level of adoption of improved cultivars economic progress in plant breeding. only tells part of the story about the results of investment in crop improvement. The Past studies have documented large velocity of varietal change is an equally im- disparities in varietal turnover rates in portant aspect, especially for countries different agricultural settings. Farmers where levels of adoption are already high. growing irrigated wheat in the Yaqui Valley The rate of change or the replacement of of Mexico replace their varieties every three old varieties by new ones says a lot about to four years. The breaking down of disease the effectiveness of genetic improvement resistance and the steady increase in yield programs. Past research suggests that if gains are positive incentives for rapid newer materials are not replacing their varietal change. In the corn belt of the USA, older counterparts, returns to genetic im- farmers switch to newer hybrids every two provement stagnate (Brennan and Byerlee, to three years. In contrast, potato growers 1991). Simply put, the permanency of first- in specialized compact regions of outstand- generation improved varieties in farmers’ ing production potential in the USA have fields points to a problem of declining pro- limited incentives to replace the Russet ductivity in the search for, and release of, Burbank potato with newer varieties. new varieties in crop improvement. Russet Burbank is difficult to grow, but it is highly productive and has strong market Varietal turnover is measured by the age of demand. For potato growers in Canada and varieties against their area in production. the USA, estimated varietal age has fluctu- The date of release usually initiates a vari- ated between 40 and 50 years since the ety’s age calculation – i.e. when it became 1990s indicating a low rate of return to available to the public for adoption. There- most state and national potato improve- fore, age is measured from the year of ment programs in North America (Walker, release to the current year, unless farmers 1994, Walker et al., 2011b). had access to the variety many years prior to the date of release. Only improved varieties are calculated irrespective of their The velocity of varietal turnover in adoption level. The appropriate measure is 2010 by crop the area of variety, x or y, in the total area of improved varieties. The results show that The estimated average weighted age is varietal age will fall irrespective of whether presented in Table 5.1 (overleaf) by crop for younger varieties replace older improved 117 country observations where crop-specific varieties or traditional varieties, because the share of younger varieties will increase in the group of improved varieties.

Area-weighted age estimates under 10 years indicate rapid varietal change and steady progress in plant breeding from an economic perspective. Estimates of varietal turnover that exceed 20 years indicate that more recent materials are having a hard time competing with earlier materials. Rising varietal age is associated with declining marginal returns to plant breeding. Farmers in Benin celebrate a bumper crop of NERICA rice. However, the adoption level also needs to Velocity of varietal turnover is as important as adoption level in assessing plant-breeding importance

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 53 adoption and varietal release information and expanding cultivation in SSA. However, were available. The average results by crop the youngest soybean varieties in farmers’ are tightly clustered in the range of 10–20 fields in Nigeria are ‘old’ as they were years. This means that there may be few, if released in the early 1990s. any, crops where older adopted improved materials have substantially eroded the The lack of difference in varietal age profitability of plant breeding. But, by the between maize in WCA and ESA is also un- same token, there was also little evidence expected. Improved cultivars in WCA are that rapid varietal change is taking place. OPVs; hybrids dominate maize production in The area-weighted grand mean is 14 years ESA. Historically and, especially in the last indicating that the average MV in farmers' decade, many more hybrids have been fields in 2010 dated from 1996. released in ESA than OPVs in WCA. Yet, the genetic and seed market-related differences The results in Table 5.1 (below) are between these two contrasting types of somewhat counterintuitive because crops material have not translated into substantial such as sweetpotato and banana, with high differences in varietal turnover. H-614 is the multiplication ratios, are characterized by a dominant hybrid in Kenya. It was released in younger portfolio of varieties compared 1986. HB-660 is less dominant, but it is the with several propagated crops with stronger leading improved cultivar in Ethiopia. Both market demand. However, this is not surpris- hybrids are closely related with the same ing because of the dearth of earlier research parental materials. They trace their roots to on these clonal crops that translated into the Kitale Station in Kenya from crosses few, if any, releases in the 1980s and 1990s. between Kitale Synthetic and Ecuador 573, a landrace from the Andean Highlands collect- Table 5.1 does contain a few surprises. For ed by the Rockefeller Foundation in 1953 example, soybeans should have performed (De Groote, personal communication, 2013). better on this criterion given its emerging In Kenya, the mean varietal age of hybrids and improved OPVs across the six maize pro- Table 5.1. The velocity of varietal turnover ducing agro-ecologies was 24 years in a na- of improved varieties in farmers’ fields in tionally representative adoption survey in SSA by crop 2010 (Swanckaert et al., 2012).

Varietal age One way to shed more light on the perfor- Crop (yrs) Number mance of crop improvement across crops is Banana 10.2 1 to combine the information on varietal Sweetpotato 10.3 5 adoption and age. Mean levels of both Groundnut 11.7 5 variables are drawn in Figure 5.1 (overleaf) Chickpea 11.9 2 to identify different combinations of Cowpea 11.9 16 outcomes for varietal adoption and change. Lentil 12.5 1 This delineation results in four quadrants Maize–WCA 12.8 11 that are roughly indicative of the compara- Wheat 12.8 1 tive profitability of investments in crop Maize–ESA 13.0 8 improvement. Programs in the upper left- Beans 13.8 9 hand quadrant are characterized by poor Cassava 14.1 17 outcomes in both varietal adoption and Soybean 14.2 11 change; commodities in the lower left-hand Pearl millet 14.8 3 quadrant are problematic in terms of adoption levels; and those in the upper Rice 15.8 4 right-hand quadrant are problematic in Sorghum 17.4 6 terms of the velocity of varietal change. Pigeonpea 17.9 2 Only maize and wheat are clearly placed in Yams 18.4 5 the positive outcome space of both criteria Barley 18.5 2 in the lower right-hand quadrant. Cassava, Field pea 18.9 1 beans, and soybean border on the positive Potato 19.4 5 outcome space, suggesting that they are Faba bean 20.7 2 superior to most crops on one criterion and Weighted mean/Total 14.0 117 are average on the other.

54 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Fababean 20 Potato Field pea Barley Yams 18 Pigeonpea Sorghum

16 Rice

Age Pearl millet Soybean 14 Cassava Beans Maize WCA Maize ESA Wheat Lentil 12 Chickpea Cowpea Groundnut

Banana Sweetpotato 10

0 20 40 60 80 100 Adoption

Figure 5.1. Distribution of crops by adoption level and percent and varietal age in years. Note: reference lines index the mean level of varietal adoption and age

Varietal change and adoption by Figure 5.2. Several of the programs in crop-by-country observations Table 5.1 were described in Section 4 as success stories in the discussion on adoption The same analytical format can be used levels by crop. to array adoption and age information for all of the 117 observations for which The composition of country and crop there is complete information on cultivar- programs in Table 5.2 also begs some specific adoption and on the date of questions, such as: varietal release. These unlabeled results are presented in Figure 5.2 (overleaf). How has DR Congo managed to Sixteen programs are located in a attain better than average estimates favorable position in the lower right-hand in varietal adoption and varietal change quadrant, showing better than average in cassava in the presence of severe adoption of improved varieties with institutional and economic constraints? younger varietal age (less than or roughly This remains a question that requires equal to 10 years). further research. Why do Burkina Faso and Senegal A quick inspection of Table 5.2 (page 57) have desirable outcomes for diffusion shows that these better-performing and varietal age in maize but also have crop-by-country entries are a blend of negligible results in varietal releases larger-area programs in maize, cassava, and and adoption in sorghum, pearl millet, cowpea with several very small programs in and groundnut? Apparently, maize soybean and rice. But, on average, these is a new crop in Burkina Faso and 16 programs are not significantly different Senegal and is expanding rapidly in from the mean size of the crop-by-country terms of area. observations in Table 5.1 (page 54) and

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 55 40

20 Age in years

0 20 40 60 80 100 Adoption %

Figure 5.2. Distribution of crop-by-country improvement level of adoption and the mean age of varieties in farmers’ fields. Note: benchmark lines are set at the mean adoption level and 10 years of adoption age

Returning to Figure 5.2, programs located 5.3 (overleaf). The largest area and value in the top-left quadrant with unfavorable share come from the cohort of varieties readings on both varietal adoption and that were released in the late 1990s. This varietal change are larger than average. finding is not necessarily surprising, but it Their mean area is about 1 million ha does show that CG Centers were able to greater than the mean of all 117 crop-by- supply materials for release by their NARS country observations. These lagging partners during a time of financial crisis programs in both varietal adoption and in the late 1990s and early 2000s. From turnover are critical to attaining this, it is possible to infer that financial dynamism in varietal diffusion as well as constraints did not entirely stop the flow development in the medium and longer of materials in the pipeline. A 15% value term in SSA. They risk being bypassed by share for varieties released between 2006 their crop and country neighbors that are and 2011 is encouraging and indicates located in the other three quadrants of that materials in the pipeline are finding a Figure 5.2. home in farmers’ fields. A sizeable chunk of the recent difference between the area and value share has been attributed to the The vintage of adopted varieties release of two promising improved yam clones in Nigeria. A small majority of the 1145 cultivars in the adopted variety database carry The share estimates in Table 5.3 also hint information on the date of release. These at the longer term impact of varietal varieties account for about 80% of the change. Improved varieties in the early adopted area and value of production. 1980s are still making a substantive Their age distribution is presented in Table contribution that cannot be ignored.

56 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Table 5.2. Genetic improvement programs with higher adoption and more rapid varietal turnover by crop and countrya

Adoption of improved varieties Crop Country (%) Varietal age (yrs) Total area (‘000 ha) Cassava Kenya 44 8 60 Cassava DR Congo 49 8 1850 Cowpea Nigeria 39 10 3768 Cowpea DR Congo 87 8 96 Groundnut Uganda 56 9 239 Maize Senegal 97 6 191 Maize Burkina Faso 49 8 555 Maize Cameroon 82 5 686 Maize Malawi 68 10 1655 Maize Zambia 73 9 1081 Rice Rwanda 69 7 13 Rice Uganda 81 10 120 Soybean Zambia 100 1 45 Soybean Zimbabwe 100 4 67 Soybean Uganda 97 3 148 Soybean Cameroon 75 8 13 a Observations in Table 5.1 with better than average adoption and age <= 10 years

A case in point is IITA’s release of its Comparing levels of varietal change important cassava variety TMS 30572 in 1998 and 2010 in 1984. In contrast, materials released prior to 1980 in the early years of the Information is sparse for comparing aspects CGIAR were relatively limited in of varietal change between 1998 and 2010. number and their impact has eroded In 1998, cultivar-specific profiles at the over time. national level could only be constructed for maize in ESA, wheat, rice, and potato. Vari- eties were younger in wheat and potato and older in maize and rice. The average Table 5.3. The vintage of varieties age ranged from about 10 years in potato contributing to adoption in 2010 by to 18 in rice. criterion and by release period Heisey and Lantican (2000) estimated Criterion weighted average varietal age for improved wheat varieties from the 1998 Area share Value share Release period (%) (%) dataset for seven countries in SSA. Estimat- ed age varied from a low of 2.5 in 1970–1975 1.7 1.1 Zimbabwe to 15.9 in Ethiopia. Estimated 1976–1980 2.7 2.9 varietal age for six of the seven countries 1981–1985 8.3 10.6 ranged from 10–16 years. Those estimates 1986–1990 12.7 12.8 were roughly the same for comparable data 1991–1995 19.4 15.6 in 1990 indicating stagnant progress in 1996–2000 27.1 23.9 varietal turnover. In Zimbabwe, varietal 2001–2005 17.7 17.4 turnover was very rapid because several 2006–2011 10.3 15.2 large homogeneous wheat producers were Total Area 27,477.4 effective in realizing their demands for (‘000 ha) varietal change. Total Value in US$ (million) 12,095.20

Note: Based on 590 varieties that account for 80–85% of total area Comparing the 1998 estimates to those in and value of production of all varieties in the database Table 5.1 (page 54) for 2010 suggests that

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 57 improved varieties are not getting any and important producing countries are younger in farmers’ fields. For maize and characterized by older than expected wheat, age is roughly the same as 14 years improved varieties. For a new expanding ago. For three of the four countries produc- crop, the youngest soybean varieties in ing potatoes, varieties are becoming older. farmers’ fields in Nigeria are ‘old’ as they For rice, the average age of MVs was the were released in the early 1990s. highest of the four crops in both 1998 and 2010. Returning to maize in Kenya, varietal Sixteen crop-by-country programs scored age has increased steadily from 17 years in well on both varietal adoption and turnover. 1992 to 22 years in 2001 to 24 years in 2010 These better performing crop-by-country (Swanckaert et al., 2012). Although age has entries combined larger area programs in fallen markedly in the dry transitional zone maize, cassava, and cowpea with several in response to rapid varietal adoption and smaller programs in soybean and rice. change, new private sector seed suppliers have not been able to penetrate into other The largest area and value shares came zones where adoption levels are stagnating. from varieties that were released in the late 1990s, suggesting that CGIAR Centers were able to supply materials for release by their Summary NARS partners during a time of financial crisis. The 15% value share for varieties The velocity of varietal change is as impor- released in 2006–2011 is encouraging and tant as the adoption level of MVs in assess- indicates that materials continue to find a ing plant-breeding performance especially home in farmers’ fields. Materials released for countries approaching moderate to full prior to 1980 in the early years of the adoption. Varietal turnover is measured by CGIAR were comparatively limited and their the age of improved varieties weighted by impact has eroded over time. In contrast, their area in production. those produced in the 1980s account for more than 20% of the area and value re- The area-weighted mean age was 14 years, sulting from MV adoption. indicating that the average MV in farmers’ fields was released in 1996. The average Comparing the 1998 estimates to those in age-related results by crop were tightly 2010 suggests that improved varieties are clustered in the range of 10–20 years. This not getting any younger in farmers’ fields. means that there may be few, if any, crops For maize and wheat, age is roughly the where older adopted improved materials same as 12 years ago. For three of the four have substantially eroded the profitability countries producing potatoes, varieties are of plant breeding. But, by the same token, becoming older. For rice, the average age there was also little evidence that rapid of modern varieties was the highest of the varietal change is taking place. Some crops four crops in both 1998 and 2010.

58 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project 6. Validating adoption estimates generated by expert opinion from survey estimates

What did we learn from the DIIVA Project that will improve the accuracy, precision and cost-effectiveness of future estimation?

Shedding light on the major issues and arriving at orders of magnitude is the intent of this section that is based on three further questions:

1. Were there systematic differences between adoption estimates from the expert panel and survey sources? Using genomic techniques in marker-assisted selection 2. Do we need to revise the estimate of to incorporate striga resistance in sorghum cultivars that 35% of MV adoption for 2010 in light farmers are already growing in the Sudan of the validation results? 3. What were the main weaknesses in using expert panels and surveys in estimating MV adoption? In the DIIVA Project Implementation Workshop held in Addis Ababa, Ethiopia in Prior to describing the validation surveys February 2010, a 13-step protocol was de- and analyzing systematic differences, we scribed for eliciting expert opinion on the begin with a brief review of the literature adoption of improved cultivars (Walker and the protocol for eliciting expert 2010): opinion. 1. Ensure that the historical information on varietal release has been updated Validating expert opinion on and is available. In other words, the adoption and the protocol used to varietal release database precedes and elicit estimates from experts lays the foundation for the assessment of adoption perceptions. The DIIVA Project was not the first to 2. Canvass background evidence on recent compare subjective estimates on adoption adoption studies and variety-specific from expert panels with more objective seed distribution and sales. data. In the spirit of the ‘1998 Initiative’, 3. Convene an expert panel (usually NARS CIMMYT economists assessed the congru- crop improvement scientists of the ence between expert opinion from NARS commodity of interest and other scientists, mainly plant breeders, and ag- experts with extensive field-level gregate adoption estimates from data on knowledge of varietal adoption). seed sales of hybrids and OPVs for maize- 4. Divide the country into sub-regions or growing countries in southern and East recommendation domains that the Africa (Hassan et al., 2001). Their assess- experts are most comfortable with (as ment showed that expert opinion on few as two or three or as many as 10 or adoption in countries where hybrids were more). These sub-regions should be as popular and approaching full adoption fully described as possible in the form were very consistent with estimates of a map or a listing of sub-national derived from seed production data. In administrative units. contrast, estimates from the two sources 5. Assign relative areas to each sub-region diverged as the importance of OPVs in- from the sub-national Harvest Choice creased. Expert opinion in Uganda and database from the most recent year or Tanzania resulted in markedly higher esti- a three-year average of recent years. mates than those inferred from annual 6. Assess the correspondence between the seed-related data. Harvest Choice agro-ecological,

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 59 socioeconomic classification and the Harvest Choice database in Steps 5–6 was experts’ description of sub-regions. optional, depending on circumstances. 7. Elicit perceptions on the rank of specific Through trial and error, the Project’s improved varieties and local varieties as Coordinator and its Steering Committee a group in descending order of Members expected that some CG Center popularity in each sub-region. The participants would arrive at an assessment reference point for the ranking is 100% process that was superior to this one in of the crop’s area in the sub-region. terms of cost-effectiveness and precision. 8. Check the congruence between varieties in the expert adoption It is too early to tell if this hope of schedule and the release list. superiority in methods-related adaptation 9. Elicit descriptive information on non- will become a reality, but a review of local varieties that are sub-regionally methods deployed by the CG Centers shows important (they are on the expert several concrete examples of adaptation perception adoption schedule) but are (J. Stevenson and J. Burgess, 2013, personal not on the release list. Such information communication): relates to the date of first use, institutionally specific classification in CIAT used a very inclusive approach to the release database, distinguishing conduct its 13 expert panels that were characteristics, etc. widely attended by knowledgeable 10. Translate the cultivar-specific people from over 150 institutions, perceptions on ranks into perceptions including the private sector and NGOs of a percent (%) of area for each (Muthoni and Andrade, 2012). ranked category. Do this for each Several CG Centers, including CIP and sub-region and for the most recent ICRISAT, adapted the process when they cropping year, say 2009–2010. The saw that progress depended on easiest way to do this may be to start increasing the level of hands-on with the aggregate category group of management by conducting more local varieties for a percent (%) area in-country visits, complemented by a estimate and then estimate percent (%) GIS-orientation to build estimates from area for the dominant MV, the second the ground up (Ndjeunga et al., 2013). most dominate MV, the third ranked IITA created teams in Francophone and MV, etc. Anglophone SSA to hold workshops, 11. Highlight issues of greatest uncertainty assemble data and canvass information. in the perceptions of percent (%) area; Wisely, IITA staff discouraged the note ranges where uncertainty is review of adoption studies because greatest. such research could unduly influence 12. Discuss areas of discrepancies between the thinking of participants and keep the background information and the them from reflecting on their personal elicited perception and revise the experience. perceptions if the discrepancies are large and if revisions are warranted. It was also apparent what did not work: 13. Draft a brief one- to two-page report mailing out questionnaires to key documenting the substance and the collaborators and hoping for responses and process (composition of the expert delegating the lion’s share of responsibility panel; a description of the sub-regions; to in-country consultants. These approaches background information on adoption; resulted in considerable e-mail fatigue but details on how perceptions were little in the way of reliable information. assessed; a description of the varieties Because US$6,000–7,000 was allocated for included in the adoption perception each crop-by-country observation, a more schedule that were not on the release aggressive supervisory approach could be list; and magnitude and reasons for any pursued. That approach usually bore fruit. revisions to expert opinion) for each priority crop-by-country combination. Participants now agree that a more geo- graphically decentralized process, featuring Some of these steps were viewed as less wider participation by different institution- essential than others. For example, the al actors in society, is needed to arrive at

60 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project more precise and accurate expert opinion three crops. CIP and CIAT jointly undertook estimates of MV adoption. Balancing surveys with their NARS partners in Rwanda knowledgeable experts with representation on beans, potatoes, and sweetpotatoes and from a wider sectorial audience is a chal- in Uganda on beans and sweetpotatoes. lenge when seeking expert opinion on ICRISAT also carried out a multi-crop survey progress in varietal research. in Tanzania on groundnut, pigeonpea, and sorghum. In some contexts, other methods would be eminently more suitable than expert The guidelines for the survey opinion. But what is frequently overlooked recommended a stratified cluster sampling is the fact that, to arrive at a significantly (Walker and Adam, 2011). Most of the better outcome, alternative methods participants followed this recommended require relatively sophisticated skills in framework. Sample size varied from 841 application and energetic, persistent households in the cassava survey in five interviewers. states of southwestern Nigeria, to 5445 households in the rice survey also in Nigeria where all 36 states were covered. Describing the validation surveys on Households interviewed per village ranged the diffusion of MVs from 10–18. Community interviews based on focus groups preceded the household Nine large-scale adoption surveys were interviews in most of the surveys. funded and undertaken by the DIIVA Project. Their coverage and sampling Oral responses on seed usage and on area features are described in Table 6.1 planted to specific varieties provided the (overleaf). Although multi-purpose in raw material for the subsequent nature, their primary intent was to validate calculation of adoption estimates. The the adoption estimates generated by the cassava survey team complemented their national expert panels. Eight of the nine household interviews with field surveys were nationally representative; measurements that featured varietal cassava’s inquiry was regional for photographs using mobile phones (Alene Southwest Nigeria. and Mwalughali, 2012b). These were analyzed by research scientists who were Previous adoption surveys, if they existed, able to assess varietal identity from the were largely restricted to small project pictures displaying morphological plant areas in the other crop and country characteristics. Without high resolution settings. Both NARS and IARC participants photographs from mobile phones, requested a national survey to complement identification of specific varieties would their project-specific inquiries that often have been impossible. addressed only the initial uptake and very early adoption of well-defined introduced The 15 crop observations in the surveys materials. described in Table 6.1 were complemented by a more limited survey that canvassed The average cost of the nine surveys was four regions in Uganda to assess adoption about US$100,000. During the Project Im- of recently released clonal material in plementation Workshop, project partici- banana (Kagezi et al., 2012). We also used pants were encouraged to pool their re- output from a recent IFPRI–Council for sources and canvass joint surveys. They Scientific and Industrial Research, Ghana were reluctant to do so initially. But the (CSIR) survey on adoption of maize and reality of a fixed budget for survey work, rice MVs in Ghana (Ragasa et al., 2013a combined with the desire for greater and 2013b). country coverage in their crops of interest, subsequently spawned a more collaborative Because of the lack of close supervision and approach. ICARDA and CIP worked the existence of recent survey estimates on together with the Ethiopian national adoption, cultivar-specific estimates could program to carry out a survey on MVs of not be elicited from expert panels for rice barley, faba bean, and potatoes in Ethiopia in Nigeria; therefore, the validation exercise in mostly shared agro-ecologies across the for this survey focuses on the aggregate

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 61 Table 6.1. Description of the sampling features of the diffusion MV validation surveys conducted by participants in the DIIVA Project

Sample size

Primary House- Geographic basis Sampling Unit holds per Number of Community Crop Country for sampling (PSU) PSU householdsa survey The 3 major regions where Barley Ethiopia 123 Kebeles 12 1,469 (1,280) Yes barley is grown 10 major agro-ecological Bean Rwanda 80 communities 18 1,440 Yes regions 19 districts, Bean Uganda 4 major geographic regions 18 1,908 Yes 108 communities All 5 States in 80 Enumeration Cassava Nigeria 10–12 841 Yes Southwest Nigeria Areas 10 major groundnut- Groundnut Nigeria 243 villages 10 2,739 Yes producing States 77 Wards, Groundnut Tanzania 7 main producing regions 15–16 1,622 (1,046) Yes 104 villages Production potential from Maize Ethiopia 156 Kebeles 15–16 2,455 No 118 maize-growing districts 77 Wards, Pigeonpea Tanzania 7 main producing regions 15–16 1,622 (816) Yes 104 villages The 3 major regions where Potato Ethiopia 123 Kebeles 12 1,469 Yes potato is grown 10 major agro-ecological Potato Rwanda 80 communities 18 1,440 Yes regions 589 Enumeration Rice Nigeria All 36 States in Nigeria 10 5,445 Yes Areas 77 Wards, Sorghum Tanzania 7 main producing regions 15–16 1,622 (902) Yes 104 villages 10 major agro-ecological Sweetpotato Rwanda 80 communities 18 1,440 Yes regions 4 major geographic regions in 19 districts, Sweetpotato Uganda 18 1,908 Yes Uganda 108 communities 8 wheat-growing agro- Wheat Ethiopia 125 Kebeles 15–18 2,096 (1,839) No ecologies

a First number denotes total sample size; numbers in ( ) are households growing the crop.

adoption of MVs as a group in relation to according to Mean Absolute Percentage local varieties. AfricaRice also undertook a Error (MAPE). In appraising congruence, similar national survey in 2009. If the expert both the percentage differences in column panel had had access to those results, its 5 and MAPE in column 6 provide responses could have been contaminated complementary information. by that information. The estimated adoption levels in Table 6.2 are based on what experts stated were Validating expert opinion with improved varieties during the elicitation the survey estimates exercise. For beans and sweetpotato in Rwanda and Uganda, released local Three small national, regional, and landraces were included in the set of cultivar-specific databases were available improved varieties. for matching adoption estimates from different sources. Congruence between For groundnuts in Nigeria, 55-437, national estimates is described in Table 6.2 described in Section 4, was included as (overleaf) where the 18 matching obser an improved variety. The definition of vations in the database are ordered improved varieties needs to be constant

62 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project across all sources in comparing estimates. In sweetpotato vines were transferred to most contexts, reviewing the national the northern region in response to relief release list was the basis for defining MVs. programs. Interest in orange-fleshed sweetpotatoes has also sparked a massive Adoption estimates elicited by IITA from transfer of planting material in selected NARS participants in Ghana closely matched districts. Historically, drought tolerance the results of the IFPRI national survey on of planting material is a known weakness maize adoption. The fit is also reasonably of improved varieties and improving good for the next 10 observations in tolerance is a primary breeding objective. Table 6.2. MAPEs are less than 30%, and It appears that transfer of large quantities differences are under 10% with the of planting material fueled exuberance borderline exception of pigeonpea in and optimism about the prospects for Tanzania. In contrast, a lack of agreement adoption that departed sharply from the is apparent in the last seven observations. reality of propagating sweetpotato in a drought-prone environment where a few Arguably, the most egregious mismatch well-defined local varieties reign. between expert opinion and survey estimates centers on sweetpotato in The second explanation focuses on varietal Uganda. Estimates were elicited for the invisibility in the sweetpotato crop, which four main geographic regions of the seldom exceeds 0.5 ha, is often planted in country and were aggregated to generate association with other crops, and usually a national estimate. The discrepancy is harvested piecemeal. It is a crop between sources of estimates was most characterized by poor road visibility that marked in the eastern and northern leads to blurred perceptions in identifying regions with differences exceeding 75%. varieties that farmers are growing. As a Labarta et al. (2012) give two plausible result, varietal identity and diversity is not reasons for the wide divergence between apparent without taking the time and the expert opinion and the survey effort to make field visits, especially at estimates. Large quantities of improved flowering.

Table 6.2. Validating adoption estimates from expert opinion with survey results by crop

Estimate of MV Adoption (%)

Mean Absolute Country/ Percent Error Crop region Expert opinion National survey Difference (%) (MAPE) Maize Ghana 57.0 59.6 -2.6 4 Maize Ethiopia 26.5 27.9 -1.4 5 Sorghum Tanzania 42.3 38.7 3.6 9 Rice Nigeria 50.4 56.2 -5.8 10 Bread wheat Ethiopia 87.7 77.8 9.9 13 Groundnut Tanzania 32.2 28.4 3.8 13 Beans Rwanda 68.2 60.1 8.1 13 Potato Ethiopia 25.2 22.2 3.0 14 Barley Ethiopia 29.2 33.8 -4.7 14 Pigeonpea Tanzania 39.5 49.7 -10.2 21 Banana Uganda 8.0 6.2 1.8 29 Cassava SW Nigeria 68.0 52.0 16.0 31 Sweetpotato Rwanda 41.6 27.9 13.7 49 Beans Uganda 60.0 40.0 20.0 50 Groundnut Nigeria 51.2 31.0 20.2 65 Potato Rwanda 84.9 35.6 49.3 138 Sweetpotato Uganda 78.8 17.9 60.9 340 Durum wheat Ethiopia 13.0 0.5 12.5 2,500

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 63 Inspection of the regional and cultivar- the same group of wheat improvement specific databases also provides clues about scientists. the likely reasons for the poor congruence between the estimate sources for the other The simple mean MV adoption level was six observations. Although problematic 48% for expert perceptions and 36.5% regions and cultivars can be identified, for survey estimations. The 11% mean explanations for what led to these large difference in a paired t-test is statistically order of magnitude disparities are mostly significant at the 5% level. If we designate speculative. Nevertheless, based on these sweetpotato in Uganda and potato in seven cases, we can say that over-optimism Rwanda as outliers because of their chang about technology transfer programs can ing contextual situations and re-estimate result in substantial overestimates of the means for the remaining 16 observa- technology adoption. The case of potato in tions (see Table 6.2), the mean difference Rwanda shows that civil war may lead to between the two estimation methods the collapse of MV varietal adoption. These narrows to 5.5% and is, again, significant are two contextual situations that analysts statistically at 0.05. Proportionally, the need to be aware of in measuring the long- survey estimate is seven-eighths the size of term uptake of improved varieties. the expert opinion estimate. Reducing our 35% estimate for aggregate adoption of Knowing what is going on in farmers’ MVs in SSA in Table 4.1 on page 39 by the fields is desirable when scientists give same proportion yields a revised estimate expert opinion on adoption. But such that approaches 31%. This revised estimate knowledge is not always recorded at the incorporates a correction from the finding main research station. The elicitation that expert opinion tends to generate process of expert opinion also does not somewhat higher levels of adoption than seem to play a significant role in under properly conducted survey estimation. standing variations in congruence. Nor does more prestigious science make for Redoing the above calculation on the regional more congruent estimates. dataset of 34 observations (excluding potato in Rwanda and sweetpotato in Uganda) As mentioned earlier, NARO’s sweetpotato gives identical results. The simple-mean, breeding program in Uganda is highly expert-opinion estimate of 36.4% is 4% respected and is the hub of sweetpotato higher than the survey estimate of 32.4%. improvement in the Great Lakes region. IITA’s Center is located in Southwest Nigeria The mean adoption estimate of 36.5% in where the cassava survey was conducted. the national surveys was made up of 26.5% The same people and the same process gen- from MVs named by the panel and 10% erated the congruent estimates for maize from unnamed or other named materials in Ghana and the rather ‘disagreeable’ believed to be MVs. The size of the second estimates for cassava in southwestern component varies from survey to survey, Nigeria. With the same elicitation process, but it is usually sizeable as there is always CIP was responsible for congruent estimates a leftover quantity of MV area that cannot for potato in Ethiopia and divergent esti- be assigned to a specific cultivar. For this mates for potato in Rwanda. reason, the area of specific MVs will typically be proportionally less than total Likewise, differences in crops and countries adoption levels. Because the ability to des- do not seem to feature as explanations of ignate specific areas to MVs is imperfect, the variation in types of estimates. Ethiopia survey-specific cultivar estimates will often was associated with convergent estimates be substantially lower than comparable es- for barley, maize, and potato, and divergent timates from expert panels. estimates for durum wheat (Yigezu et al., 2012b; Jaleta et al., 2013; Labarta et al., Thus far, we have presented comparative 2012). Adoption estimates for bread wheat results from the 18-observation national in Ethiopia did not vary much by source but, database. Similar differences in MV in relative terms, the estimates for durum adoption between expert opinion and wheat were substantially different (Yirba et household surveys were also found in the al., 2012). These estimates were elicited from 34-observation regional database. Findings

64 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project for the 274-cultivar specific database are The message conveyed in Table 6.3 is that presented in Table 6.3 (below), which probably neither surveys nor expert panels divides the varieties into four categories de- can do a good job in delivering accurate pending on the level of perceived adoption estimates of cultivar-specific adoption. by experts. For example, of the 279 varieties Expert panels will most likely overestimate from the national comparisons in Table 6.3, the importance of specific varieties; surveys experts perceived that 44 had a level of will understate their relevance. Although adoption that exceeded 10% of cultivated skillful use of both methods may suffice for area of the crop. Experts believed that, on our purposes, we should be aware of the average, these were sown on 24% of the sources of bias when the focus is on MV- area available; the mean survey estimate specific adoption. Accuracy in survey for the same 44 varieties was about 13%, estimates depends heavily on whether or resulting in a difference between the two not a plethora of names can be identified sources of about 11%. From the previous reliably with specific varieties. discussion of the national and regional data, it was likely that expert estimates Of the DIIVA-funded surveys, bean were higher than the survey estimates at researchers in Rwanda worked hardest in all levels, except the lowest estimates. tracing the identities of farmers and their crop varieties in many locations. They This tendency for systematic differences assigned successfully (with an 88% certainty) to emerge between the two sources of the area available to local, selected, and estimates applies to all levels of estimates, improved varieties of bean (Katungi and but figures most prominently for expert Larochelle, 2012). CIAT researchers and their estimates in the range of 5–10% (Table 6.3). partners had considerable experience in the For the lowest level of adoption in the MV counting of bean varieties. Their work in the cultivar database, the survey estimates are DIIVA Project built on interviews with village higher than expert opinion, which to some focus groups carried out in 2000 and 2005 extent neglected these varieties. Restricting when respondents ranked the importance of the analysis to only positive observations for different varieties. With the addition of data expert opinion in this lowest interval does not from 2010, patterns emerging over time reverse the finding that the survey estimates could be seen. are higher than those for expert panels. On the other hand, expert opinion tends Soon after values greater than one are to focus on a subset of varieties while reached, the difference between the two ignoring the relative importance of other sources of estimates widens as expert MVs. The otherwise excellent survey work in estimates become significantly higher than southwest Nigeria was an apt example survey estimates (Table 6.3). Proportionally, of not being inclusive enough in eliciting the gap is widest for the interval 5–10%. estimates from experts – the elicitation did The mean expert estimate of 8% is matched not mention the leading MV found in the by a survey estimate of only 2.8%. Con household survey, apparently because it did gruence is a little better for the leading not appear on the release list. varieties that were expected to account for more than 10% of the area. For these Being too inclusive can also prove to be dominant cultivars, the simple mean survey a risky strategy. Returning to beans in estimate rises to 50% of the expert estimates. Rwanda, experts allocated very small areas

Table 6.3. Agreement between expert and survey estimates of specific varieties by expert interval

Estimate (%) 0–1 1.01–5 5.01–10 >10 Expert 0.39 2.97 8.02 24.12 Survey 0.70 1.15 2.84 12.86 Difference -0.30 1.82 5.18 11.27 Number of observations 105 100 30 44

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 65 to 22 improved varieties. Sixteen of these Mandii seems to be expanding – its area in had negligible adoption outcomes in the 2009 was estimated at 7%. Given the uncer- household survey results. An additional tainty about its origins, the ‘What to do 25 MVs accounted for about 10% of the about Mandii?’ question can most likely area. These did not receive an area alloca- only be addressed by DNA analysis. tion by the expert panel. The DIIVA Project has also reconfirmed the need for field measurement in cases where Focusing on challenges in nationally varieties are difficult to distinguish morpho- representative adoption surveys logically. The survey of cassava in southwestern Nigeria epitomizes this case (Alene and Two challenges have been highlighted Mwalughali, 2012b). Farmers knew improved in arriving at cost-effective MV adoption varieties by a group name but could not dis- levels from survey data. The first is tinguish relatively small morphological and responding to the need to save resources phenotypic differences that allowed for the by covering multiple crops in shared agro- elicitation of reliable data on specific MV cul- ecologies. There are usually tradeoffs tivars. In this case, there was no substitute between the potential for cost saving and for field measurement, which is more doable the reliability of estimates (C. Ragosa, 2013, in cassava because it is in the field for a personal communication). The other is the longer time in a mature state. aforementioned problem of identifying a specific MV from a multiplicity of names Survey performance could be improved that exhibit widespread spatial variation. if focus groups generated reliable Judging whether a cultivar is or is not an information on varietal adoption. The use MV is a corollary to the identification of focus group interviews in a community problem. The recycling of seed in cross- questionnaire was one of the features of pollinated crops is another difficult issue the surveys supported by the DIIVA Project that calls for standardization. (Table 6.1, page 62). In their validation reports, project participants formally Levels of MV adoption can vary widely even compared responses from focus groups and in well-conducted surveys. Based on a household questionnaires. Although these national-level survey of rice in Ghana in results have yet to be analysed, reading the 2012, researchers estimated aggregate MV reports suggests that focus groups can adoption approaching 60% (Table 6.4, provide useful information about the overleaf). However, researchers from an relative importance of the variety in the earlier national survey carried out in 2009 village and the adoption levels of individual arrived at a comparable estimate exceeding farmers; but household data are strongly 80% (Diagne et al., 2012). The difference is preferred if cultivar-specific area estimation not attributable to the differing survey years is the goal (Mausch and Simtowe, 2012). – the disparity in estimates emanates from decisions researchers had taken in classifying Researchers from AfricaRice were more varieties as ‘improved’ or ‘traditional’. optimistic about the use of community- based instruments than most researchers In both surveys, the leading variety was from other CG Centers (Diagne et al., 2013). Jasmine 85, an IRRI variety bred in Thailand They still opted for household schedules in the 1960s (Table 6.4). (Jasmine 85 was over focus group interviews, where it was officially released in Ghana in 2009 after necessary to collect cultivar-specific infor- it had already been adopted widely by mation. They also found ways to collect farmers.) But the key question is: What to do area data at the community level which about Mandii, the second leading variety compared favorably with information laying claim to 19% of area? Researchers in gathered from household interviews the 2012 survey classified it as a ‘local (Table 6.5, page 68). variety’ in Table 6.4, while researchers in the 2009 survey designated it as ‘improved’. Although there is an 11% gap between the Their list of improved varieties contains 104 two estimates on the level of adoption of names with only 7 dated released varieties MVs as a whole, the matched ranking with adopted area. between community and household surveys

66 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Table 6.4. Distribution of rice by variety planted during the major growing season in 2012

Varieties Area (%) CSIR-released varieties 34 JASMINE 85/Gbewaa/Lapez (2009) 27 Digang (also called Abirikukuo or Aberikukugu) (2002) 3 GR 18 (Afife) (1986) 2 GR 21 (1986) 1 Sikamu/TOX 3108 (1997) 1 NERICA 1 (2009) 1 FARO 15 (1970s) 0 Bodia (2010) 0 Other promising varieties being evaluated (already with certified seed production) 19 Togo Marshall 11 Jet 3 4 Aromatic Short 2 IR20 1 TOX 3107/Bumbaz 1 NERICA 14 0 NERICA 9 0 WITA 7 1 Other varieties that came from MOFA but cannot be named 5 Indigenous/traditional/local varieties 40 Mandii (originally from Sierra Leone, introduced by MOFA in the 1970s) 19 Farmers did not know variety 2 Total 100

MOFA = Ministry of Food and Agriculture, Ghana. Source: Ragasa et al., 2013b, CRI/SARI/IFPRI survey (November 2012–February 2013).

is characterized by widespread agreement in the household survey could not find an on the relative importance of specific identical partner in the community focus cultivars. The simple correlation between group inquiries which also embraced a the community and the household rankings relatively large set of varieties. Most of approaches 0.80 in Table 6.5. these unmatched household varieties were planted in very small areas. Common bean cultivars show even stronger agreement in Rwanda and Uganda The high estimated correlation coeffici- between a community estimate in the ents for bean cultivars in Rwanda and percentage of villages where a cultivar was Uganda suggests that well-constructed mentioned in a focus group and the per- community focus groups can provide centage area of the same cultivar from the valuable information on the relative household survey. For 67 common cultivars importance of leading varieties in a in Rwanda, the estimated correlation well-defined regional setting. Therefore, coefficient exceeds 0.80; for the leading community focus groups could provide a 19 common varieties in Uganda, the esti- valuable means to ground truth expert mated correlation coefficient exceeds 0.95. opinion in a rapid rural appraisal format. In both of these meticulously carried out This finding raises several important issues, surveys, the community survey focused on such as the number of communities that is the top three cultivars at three points in sufficient for the generation of significant time. All the cited focus group varieties and positive associations under different could be paired up with household respons- variance-related conditions. Given that es, but not all the household responses travel to the community is usually the could be matched to the top varieties largest cost component of any rural perceived by focus group respondents. survey in SSA, the issue of relative costs More than 100 varietal names generated is relevant.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 67 Table 6.5. Comparison of village-level and household-level interview data on varietal adoption using area grown under these varieties for rice in Nigeria

Village interview Household interview (2009)

Variety Percentage Rank Percentage Rank Gap Traditional 54.75 1 43.73 1 11.02 Modern 45.25 2 56.27 2 -11.02 FARO 44 / SIPI 4 8.04 1 12.35 1 -4.31 CHINA 7.03 2 8.76 2 -1.73 IMPROVED 3.48 3 4.83 3 -1.35 NERICA (Others) 2.94 4 4.36 4 -1.42 FARO 15 2.80 5 4.03 5 -1.23 FARO 46 / WITA 1 2.39 7 2.77 6 -0.38 FARO 52 / WITA 4 2.68 6 2.45 7 0.23 FARO 55 / NERICA 1.77 8 2.15 8 -0.38 FARO 37 / WITA 3 1.59 9 2.01 9 -0.42 FARO 29 / BG 90- 1.48 10 1.58 10 -0.10 FARO 54 / WAB 18 1.38 11 1.32 11 0.06 BUTUKA 0.17 31 1.20 12 -1.03 FARO 51 / CISADA 0.95 12 1.09 13 -0.14 TURN 2 0.53 14 1.07 14 -0.54 ECWA 0.28 21 1.03 15 -0.75 IR 8 0.57 13 0.71 16 -0.14 CAROLINA 0.50 16 0.66 17 -0.16 WILLY RICE 0.33 20 0.59 18 -0.26 FARO 21 0.34 17 0.55 19 -0.21 FARO 35 / WITA 2 0.52 15 0.38 20 0.14 YARJOHN 0.27 22 0.28 21 -0.01 FADAMA2 0.06 52 0.27 22 -0.21 Other Improved 5.15 1.83

Source: Diagne et al., 2013

Nonetheless, in the next phase of DIIVA, we on what worked. The protocol was adapted need to find a cost-effective alternative to to regional and crop-specific circumstances the 500–700 representative household that featured a considerable amount of surveys in order to validate expert opinion ‘learning by doing’ by CG Center staff con- from the more qualitative perspective of ducting the expert panels. In general, more ‘Do the elicited estimates roughly reflect effective elicitation was characterized by: reality or not?’ Well-structured, community focus group discussions combined with field close and intensive supervision of CG visits could be one such alternative. project-related staff; organization of and attendance at time- bound workshops with direct interaction Summary with expert panel members; greater spatial resolution in the elicitation About one-third of the resources of the of estimates that were subsequently DIIVA project were invested in nine nation- aggregated to regional and national ally representative adoption surveys that levels; were designed to validate the estimates including more members from the infor- from expert panels and provide raw mal sector and from NGOs with geo- material for impact assessment. In carrying graphic-specific expertise in technology out the process of estimate elicitation from transfer on the panels; and a standardized protocol, participants also feedback from CG Center breeders in the generated considerable anecdotal evidence final stages of the process.

68 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Lessons on what did not work were extenuating circumstances of rapid transparent. The CG Center that relied change or disruption associated with solely on NARS scientists as consultants to rampant over-optimism about the carry out expert elicitation was only able to prospects for large technology transfer provide quality cultivar-specific adoption efforts and with civil war that also can be estimates for two of their 14 assigned crop- devastating for the applicability of prior by-country observations. Much more knowledge in circumstances where intensive supervision was needed. confirmation is difficult.

Survey estimates and those from expert Slightly over 70% of the mean adoption opinion panels were within ten percentage estimates in the national surveys were points for ten of the 18 observations composed of MVs for which the panel held suitable for validation. Survey estimates positive adoption beliefs; the other 30% were lower for the other eight observa- came from unnamed or other named tions, and for two of these they were materials believed to be MVs. The size of markedly lower. Ignoring these two the second component varies from survey outliers, survey estimates were about seven- to survey, but it is usually sizeable as there eighths the size of expert elicitations. is always a leftover quantity of MV area Therefore, the mean value of 35% for MVs that cannot be assigned to a specific for SSA as a whole is likely to be overesti- cultivar. For this reason, the summed area mated because the majority of estimates of specific MVs will typically be less than came from expert opinion. Applying the an aggregate adoption level. Surveys are seven-eighths finding from the validation likely to understate the importance of exercise gives a more conservative estimate specific improved cultivars; detailed of about 31% for MV adoption if surveys estimates from expert opinion that feature had replaced expert opinion panels. few, if any, varieties in a residual ‘other’ category are likely to overemphasize the Higher mean absolute percentage errors uptake of specific MVs. Accuracy in survey between the two sources did not seem to estimates depends heavily on whether or be associated with variations in the not numerous regional- and location- elicitation approach or specific to a crop specific names can reliably be assigned to or country. They had more to do with the specific varieties.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 69 7. Conclusions

This Synthesis Report on the first two objec- For some crops, leveraging adoption in one tives of the DIIVA Project is laden with country is the key to significant progress. numbers. Some are more important than For cowpea, that country is Niger. If an others. We estimate that MVs of food crops improved cowpea variety could reach a in SSA in 2010 covered slightly more than 20% adoption level in Niger, it would 35% of the total harvested area. Between become one of the very few food-crop 1998 and 2010, uptake of MVs increased at varieties to cover 1 million ha in Africa. a pace equivalent to a linear annual gain Presently, cowpea varieties planted on only of 1.45%. 3–4% of the total area make the crop eligible for membership in the group of the Incorporating the adoption of improved 100 most extensively grown food-crop varieties as a MDG would be desirable for varieties in SSA. economic development in general and agricultural development in particular. For other crops, enhancing the adoption Although there are still several methods- prospects in a single agro-ecology of just related and semantic issues to resolve, the one country would make a difference. For results of the DIIVA Project show that it is rice, that agro-ecology is the rainfed possible to make a timely and concerted lowlands of Madagascar. effort to measure the diffusion of MVs for a relatively small quantity of resources. Meeting this proposed MDG also means Extrapolating the past performance to the that varieties should turn over faster in future would establish a target of about farmers’ fields. We documented that the 50% for MV adoption by 2020. Continuation weighted average age of improved of a trend may seem like a pedestrian target, varieties in farmers’ fields is 14 years since but this level of adoption is ambitious release. This rather mediocre estimate for because scientists working on several crops varietal turnover could have been in the DIIVA initiative have little research or considerably worse. However, a slender basic knowledge on which to found body of evidence suggests that it also has improvements. not improved over time. Improved varieties in farmers’ fields today are not getting any Realizing outcomes consistent with this solid younger for most crops and countries. The trend in performance will require several potential of biotechnology to increase changes. Lagging crops and/or lagging coun- varietal turnover through techniques such tries or both were identified as an area of as MAS is still largely a promise that needs concern, particularly for sorghum, pearl to come to fruition. millet, and groundnut in West Africa where scientists’ advancing age and low levels of The analysis also underscores the research intensity are important issues that importance of the IARCs of the CGIAR in threaten enhanced varietal output and supplying material to, and sharing of adoption. Efforts will have to be redoubled material with the NARS. About 65% of the in West Africa if a goal of 50% coverage in varietal adoption originated in IARC-related improved varieties for SSA is to be attained materials. The CG Centers contribution for by 2020. Some 'out-of-the-box' thinking may 10 of the 20 crops was above 80%. Their be required to address the problem of stag- share in adoption was also substantially nating and eroding scientific capacity in higher than their share in varietal output several West African crop improvement for most crops. programs on coarse cereals and grain legumes. Performance also has to improve in In the proposal for the DIIVA Project, over Angola and Mozambique, the only two 30 hypotheses were specified for testing on countries in Southern Africa where adoption the strength of NARS, varietal output, of MVs is low for all primary and secondary adoption and change, and exchange of food crops. materials between CG Centers and NARS.

70 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Some of these hypotheses, such as equal national crop improvement programs in the shares in varietal releases and adoption form of crossing and selection was increas- from the contributions of the CGIAR to ingly evident. That hypothesis was broadly varietal change, were rejected; others were confirmed in the global ‘1998 Initiative’. broadly confirmed; and others still could However, this scenario on the development not be tested as data were insufficient or of national plant breeding in SSA does not incomplete mainly in the area of the even apply to rice, which was the crop that germplasm exchange of materials. was its ‘poster child’ in Asia and Latin America (Evenson and Gollin, 2003a). Findings from the DIIVA Project also show that output in the form of released The descriptive research in this Synthesis varieties is increasing for most crops, but Report clearly does not begin to exhaust is still characterized by a high level of the relevant themes that can be explored instability from year to year. Part of this via the DIIVA 1998 and 2010 databases. The instability is attributed to donor programs relationships among scientific capacity, that periodically reinforce public sector varietal output, and varietal adoption research because governments do not have need to be explored in a more rigorous the resources or political will to invest in analytical manner. The data can also be in poor people’s crops. Estimated research tegrated with other datasets to shed light intensities, i.e. the relationship between on recent tendencies in, and determinants size of country production and investment of, productivity growth in African agriculture. in research scientists, has not changed over time for most crops. The input of university Indeed, documenting productivity growth research in varietal release is still negligible from adoption is an equally challenging in most crops and countries. proposition. The genetic contribution to yield growth is low in marginal production During the discussion of substantive results environments characterized by erratic in the sections on varietal output, adoption, rainfall and declining soil fertility. In the and change, the DIIVA Project has also had companion Green Cover Report that its share of surprises. Prominent among presents the impact assessment research in these unexpected findings were the increas- the DIIVA Project, large and statistically sig- ing demand for maize OPVs in West Africa, nificant yield effects from improved varietal the steady productivity record of cassava in adoption were found in maize in regions of the face of well-documented resource relatively high production potential. scarcity, and the advanced age of cultivars However, comparable productivity impacts in the expanding soybean crop in Nigeria. for improved varieties of bean in Rwanda and Uganda and improved sorghum and The prolificacy of the private sector in pearl millet cultivars in Northern Nigeria producing new maize hybrids in ESA is not were considerably lower. Measuring the surprising, but the observation that varietal genetic contribution from improvement in age has not yet started a downward trend traits that do not reflect higher yield poten- is. The slow pace of private sector develop tial – such as short duration for drought, ment in maize research in West Africa was escape or varietal disease-resistance – is also unanticipated. The moderately high likely to understate the size of these effects, adoption performance in DR Congo across unless highly focused and intensive data col- several food crops was an unexpected lection is pursued. achievement that warrants more careful monitoring when varietal adoption is next A 35% share in MVs for food crops in SSA in updated. 2010 is roughly the level Asia was at in the early 1970s, that Latin America attained in The continuing emphasis on the release of the mid-1980s, and that the Middle East purified landrace materials in the very and North Africa reached in the early recent past for some countries in some 1990s (FARA, 2006, as cited in Lynam, crops and the central role of adaptive 2010). But this comparison that relies on breeding in most national crop improve- aggregate old data is flawed, because it ment programs point to rejection of the is not adjusted for the reality of rainfed hypothesis that applied plant breeding by agriculture in SSA.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 71 The findings on adoption in SSA need to be Regional commodity networks are often compared to complementary studies that seen as critical institutions for generating are now nearing completion by IRRI and varietal output and subsequent adoption ICRISAT on the adoption of rice, sorghum, and change in developing countries, espe- pearl millet, and groundnut improved cially in SSA. Largely anecdotal evidence varieties in South Asia; and by CIAT on the suggests that few of these organizations diffusion of rice, bean, and cassava MVs in survived intact from the budgetary restric- Latin America. These investigations were tions of the 1990s and early 2000s. These inspired by, and follow the pattern of, the multinational networks have been heavily DIIVA Project. dependent on funding from donors and technical leadership from the CGIAR. Quan- From the perspective of SSA, their findings titative information and analysis is needed on rainfed agriculture are especially relevant, to test the early evidence that these genetic so that progress can be assessed and realistic improvement networks were seriously targets established on performance. Unless damaged by the budgetary crisis of the basic knowledge changes the yield distri 1990s and early 2000s, and that the bution of basic food crops in rainfed exchange of germplasm was severely agriculture, the evidence from South Asia reduced because of funding shortfalls suggests that program results (even well- caused by donor withdrawal. developed crop improvement initiatives) can fall considerably short of full adoption and Most of the above research priorities are rapid varietal turnover for some crops in easy to visualize. Years from now they difficult agro-ecologies. should pale in comparison to output stimulated by making the database accessible to For example, the dominant cultivar in the public. Providing access is past due and post-rainy season sorghum production in is critical for generating timely and wider peninsular India is still M35-1, a variety research results for, and by, a larger released in the late 1930s about 10 years audience who are aware of the project and prior to Independence. Both groundnut in have expressed a firm demand for its use. peninsular India and rice in East India are Analytically, this study does not break new characterized by slow rates of varietal ground. Its novelty and value stems from its turnover because of the permanency of wide scope in terms of crops and countries first-generation MVs in farmers’ fields. with intensive data collection via the same Nevertheless, adoption of improved culti- protocols with an emphasis on validation. vars is substantially lower in SSA than in roughly comparable agro-ecologies in The achievements of the DIIVA Project are South Asia and Latin America. Questions of more substantive than methodological. We ‘How much lower?’ and ‘What are reason- did confirm the differences between the able targets?’ are researchable issues that estimates obtained from expert opinion need to be addressed using comparative and from surveys. The results of the valida- evidence on varietal uptake and turnover tion comparisons suggest that our 35% in South Asia and Latin America. adoption estimate could decline to about 31% if we had the resources and the During the discussion of results, several capacity to substitute national representa- priorities for research were implicitly tive surveys for expert opinion panels. identified (see Sections 2–6). For example, documented budgetary shortfalls at the We also conducted a thorough assessment CG Centers implied a significant contraction of the availability of experimental data for in the flow of germplasm to the NARS via undertaking an ex-post rate of returns donor-funded plant breeding networks. analysis. However, these data were not Had tightening budgets for germplasm available and their quality appears to have enhancement and development not been deteriorated since the 1980s when Farming less in evidence in the 1990s and into Systems Research figured prominently in the early 2000s for the CGIAR crop several CG Centers. On the other hand, CG improvement programs, the pace of Center scientists made unforeseen project- varietal change could have accelerated related contributions in the conduct of faster than 1.45% per annum. multi-crop surveys in shared agro-ecologies

72 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project and in field measurement that permitted varietal name registries over an extensive, the used of highly trained technical but well-defined, geographic area could assistants in a cost-effective manner. be a productive undertaking. Learning from pilot studies with DNA fingerprinting We also acquired more awareness about remains a priority for the next phase of this the magnitude of some of the problems project. If Dana Dalrymple was charged with associated with measuring varietal adoption. evaluating the DIIVA Project, he would be Foremost among these is the issue of varietal impressed with the comprehensive coverage identification, which could be welcomed of the endeavor, but would probably say by anthropologists. Investing in the that not much has changed in the way we development and maintenance of pilot chronicle varietal adoption.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 73 References

Adesina, A. & Zinnah, M.M. 1993. Technol- Dalrymple, D. 1969. Technological change ogy characteristics, farmer’s perceptions in agriculture: effects and implications and adoption decisions: A Tobit model for developing nations. United States application in Sierra Leone. Agricultural Department of Agriculture (USDA): Economics 9: 297–311. Washington DC, USA.

Alene, A.D., Menkir, A., Ajala, S.O., Badu- Dalrymple, D. 1978. The development and Aprak, B., Olanrewaju, A.S., Manyong, spread of the high-yielding varieties of V.M. & Ndiaye, A. 2009. The economic wheat and rice among less-developed and poverty impacts of maize research nations. 6th Edition. Foreign Agricultural in West and Central Africa. Agricultural Economic Report No. 95. United States De- Economics 40(5): 535–550. partment of Agriculture (USDA) and United States Agency for International De- Alene, A.D. & Mwalughali, J. 2012a. The velopment (USAID): Washington DC, USA. effectiveness of crop improvement programs in Sub-Saharan Africa from Dalrymple, D. 1986a. Development and the perspectives of varietal output and spread of high-yielding rice varieties in adoption: The case of cassava, cowpea, developing countries. United States maize, soybean and yam. Diffusion and Agency for International Development Impact of Improved Varieties In Africa (USAID): Washington DC, USA. (DIIVA), Objective 1 Report. Internation al Institute of Tropical Agriculture (IITA): Dalrymple, D. 1986b. Development and Lilongwe, Malawi. spread of high-yielding wheat varieties in developing countries. United States Alene, A.D. & Mwalughali, J. 2012b. Agency for International Development Adoption of improved cassava varieties (USAID): Washington DC, USA. in southwestern Nigeria, Objective 2 Technical Report. International Institute Dalton, T.J. & Guei, R.G. 2003. Productivity of Tropical Agriculture (IITA): Lilongwe, gains from rice genetic enhancements in Malawi. West Africa: Countries and ecologies. World Development 31(2): 359–374. Alston, J.M., Norton, G.W. & Pardey, P.G. 1995. Science under scarcity: Principles De Groote, H., Gitonga, Z. & Mugo, S. 2011. and practice for agricultural research Diffusion and impact of improved maize evaluation and priority setting. Cornell and wheat varieties in East and Southern University Press: Ithaca, New York, USA Africa. Diffusion and Impact of Improved (republished in soft cover by CAB Varieties in Africa (DIIVA), Objective 1 International: Wallingford, UK, 1998). Report. International Maize and Wheat Improvement Center (CIMMYT): Nairobi, Beintema, N.M. & Stads, G.J. 2011. African Kenya. agricultural R&D in the new millennium: Progress for some, challenges for many. Diagne, A., Kinkingninhoun-Medagbe, F.M., IFPRI Food Policy Report. International Amovin-Assagba, E., Sanni, K. & Toure A. Food Policy Research Institute (IFPRI): 2012. To attain a wider understanding of Washington DC, USA. key aspects of the performance of food- crop genetic improvement in priority Brennan, J.P. & Byerlee, D. 1991. The rate of country-by-commodity combinations in crop varietal replacement on farms: sub-Saharan Africa: The case of rice crop. Measures and empirical results for Diffusion and Impact of Improved wheat. Plant Varieties and Seeds 4: Varieties in Africa (DIIVA), Objective 1 99–106. Report. AfricaRice Center: Cotonou, Benin.

74 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Diagne, A., Kinkingninhoun-Medagbe, F.M., Journal of Agricultural and Applied Ojehomon, V.T., Abedayo O.O., Amovin- Economics 33(3): 459–475. Assagba, E. & Nakelse, T. 2013. Assessing the diffusion and adoption of improved Hassan, R.M., Mekuria, M. & Mwangi, W. rice varieties in Nigeria. Diffusion and 2001. Maize breeding research in Impact of Improved Varieties in Africa Eastern and Southern Africa: Current (DIIVA), Objective 2 Report. AfricaRice status and impacts of past investments Center: Cotonou, Benin. made by the public and private sectors 1966–1997. International Maize and Edwin, J. & Masters, W.A. 1998. Returns to Wheat Improvement Center (CIMMYT): rice technology development in Sierra Mexico City, Mexico. Leone. Sahelian Studies and Research I:67–74. Heisey, P. & Lantican, M.A. 2000. Interna- tional wheat breeding research in Eicher, C.K. 1995. Zimbabwe’s maize-based Eastern and Southern Africa 1966–97. Green Revolution: Preconditions for CIMMYT Economics Working replication. World Development 23(5): Paper 00–01. International Maize and 805–818. Wheat Improvement Center (CIMMYT): Mexico City, Mexico. Eicher, C.K. 2003. Flashback: Fifty years of donor aid to African agriculture. ICRISAT. 1978, 1982, 1987, 1992, 1998, 2002, Conference paper No. 16. Paper and 2010. Annual Reports. International prepared for the Conference “Successes Crops Research Institute for the Semi- in African Agriculture” December 1–3, Arid Tropics (ICRISAT): Patancheru, India. 2003. Pretoria, South Africa. ICRISAT. 2009. First marker-assisted sorghum Evenson, R.E. & Gollin, D. (Eds.) 2003a. Crop varieties released in Sudan. ICRISAT variety improvement and its effect on Happenings. International Crops productivity: the impact of international Research Institute for the Semi-Arid agricultural research. CABI Publishing: Tropics (ICRISAT): Patancheru, India. Wallingford, UK. Jaleta, M., Yirga, C., Kassie, M., de Groot, H. Evenson, R.E. & Gollin, D. 2003b. Crop & Shiferaw, B. 2013. Knowledge, genetic improvement in developing adoption and use intensity of improved countries: Overview and summary. Pps. maize technologies in Ethiopia. 7–38 in Evenson, R.E. & Gollin, D. (Eds.) CIMMYT’s Report on Objective 2. Part I. (2003). Crop variety improvement and its International Maize and Wheat Improve- effect on productivity: the impact of ment Center (CIMMYT): Nairobi, Kenya. international agricultural research. CABI Publishing: Wallingford, UK. Johnston, N., Pachico, D. & Wortmann, C.S. 2003. The impact of CIAT’s genetic im- FARA. 2006. Framework for African agri provement research on beans. Pps. cultural productivity / Cadre pour la 257–274 in Evenson, R.E. & Gollin, D. productivité agricole en Afrique. Forum (Eds.) (2003). Crop variety improvement for Agricultural Research in Africa and its effect on productivity: the impact (FARA): Accra, Ghana. of international agricultural research. CABI Publishing: Wallingford, UK. Fuglie, K.O. & Rada, N. 2011. Policies and productivity growth in African agricul- Kagezi, G., Mpiira, S., Jogo, W., Karamura, ture. Conference paper No. 19, prepared E., Gotor, E., Garming, H. & Johnson V. for the ASTI–FARA Workshop, 2012. Adoption of new banana varieties 5–7 December 2011, Accra, Ghana. in Uganda: Report for the DIIVA Project. Diffusion and Impact of Improved Fuglie, K.O. & Walker, T.S. 2001. Economic Varieties in Africa (DIIVA), Objective 1 incentives and resource allocation in U.S. Report. Bioversity International: Rome, public and private plant breeding. Italy.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 75 Katungi, E. & Larochelle, C. 2012. Validating Maredia, M.K. & Raitzer, D.A. 2006. CGIAR adoption of modern bean varieties in and NARS partner research in sub- Rwanda and Uganda using household Saharan Africa: evidence of impact to and community surveys. Diffusion and date. CGIAR Science Council Secretariat: Impact of Improved Varieties in Africa Rome, Italy. (DIIVA), Objective 2 Report. International Center for Tropical Agriculture (CIAT): Maredia, M.K. & Eicher, C.K. 1995. The eco- Kampala, Uganda. nomics of wheat research in developing countries: The hundred million dollar Kislev, Y. 1977. A model of agricultural puzzle. World Development 23(3): research. Pps 265–280 in Arndt, T.M., 401–412. Dalrymple, D.G. & V. Ruttan (Eds.) (1977). Resource allocation and productivity in Mausch, K. & Simtowe, F. 2012. Adoption national and international agricultural rates and their reliability evidence research. University of Minnesota: Min- based on expert panel discussions, neapolis, Minnesota, USA. community focus groups and household survey in Tanzania – groundnuts, Labarta, R. 2012. The genetic improvement pigeonpea and sorghum. Diffusion and of potato and sweetpotato in Sub- impact of improved varieties in Africa Saharan Africa. Diffusion of Improved (DIIVA), Objective 2 Report. Varieties in Africa (DIIVA) Project Report International Crops Research Institute on Objective 1. International Potato for the Semi-Arid Tropics (ICRISAT): Center (CIP): Nairobi, Kenya. Nairobi, Kenya.

Labarta, R., Wambugu, S., Yirga, C., Muthoni, R. & Andrade, R. 2012. The impact Mugabo, J., Dieu Nsabimana, J., Mulwa, of bean genetic improvement research. C., Sulz, S., Larochelle, C., Alwang, J., Diffusion and impact of improved Andrade, R. & Yigezu, Y. 2012. The varieties in Africa (DIIVA), Objective 1 adoption of improved potato and Report. International Center for Tropical sweetpotato varieties in Ethiopia, Agriculture (CIAT): Kampala, Uganda. Rwanda and Uganda: Comparing expert opinion estimates and the results of Ndjeunga, J., Hash, C.T., Faye, I., Sanogo, nationally representative surveys. M., Echekwu, C.A., Yeye, M., Moutari, Diffusion and Impact of Improved A., Kodio, O., Toure, A., Cisse, O., Varieties in Africa (DIIVA), Objective 2 Naino, J., Magagi, A., Zangre, A., Report. International Potato Center Minougou, A., Ibro, A. & Kassari, A. (CIP): Nairobi, Kenya. 2012. Assessing the effectiveness of agricultural R&D in West Africa: Cases Larochelle, C., Alwang, J., Norton, G.W., of pearl millet, sorghum and Katungi, E. & Labarta, R.A. 2013. Ex-post groundnut crop improvement impact of adopting improved bean programs. Diffusion and Impact of varieties on poverty and food security in Improved Varieties in Africa (DIIVA) Uganda and Rwanda. Diffusion and Project, Objective 1 Report. impact of improved varieties in Africa International Crops Research Institute (DIIVA), Objective 3 Report. Virginia for the Semi-Arid Tropics (ICRISAT): Tech: Blacksburg, Virginia, USA, and Niamey, Niger. International Center for Tropical Agriculture (CIAT): Cali, Colombia. Ndjeunga, J., Ntare, B.R., Ajeigbe, H., Echekwu, C.A., Ibro, A. & Amadou, A. Lynam, J. 2010. Evolving a plant breeding 2013. Adoption and impacts of modern and seed system in Sub-Saharan Africa in groundnut varieties in Nigeria. an era of donor dependence. Prepared Diffusion and Impact of Improved for the Global Initiative on Plant Varieties in Africa (DIIVA), Objective 2 Breeding. Food and Agriculture Organi- Report. International Crops Research zation of the United Nations (FAO): Institute for the Semi-Arid Tropics Rome, Italy. (ICRISAT): Niamey, Niger.

76 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Ndjeunga, J., Umar, J., Ahmed, B., Aba Aba, liberalisation policies on the maize seed Ibro, A., Abdoulaye, A. & Gwadi, K. systems in Kenya based on micro- 2011. Adoption and impacts of modern evidence from 1992 to 2010. Interna- sorghum and pearl millet varieties in tional Maize and Wheat Improvement Northern Nigeria. International Crops Center (CIMMYT): Nairobi, Kenya. Research Institute for the Semi-Arid Tropics (ICRISAT): Niamey, Niger. Thiele, G., Hareau, G., Suarez, V., Chujoy, E., Bonierbale, M. & Maldonado, L. 2008. Nweke, F. 2009. Controlling cassava mosaic Varietal change in potatoes in develop- virus and cassava mealybug in Sub- ing countries and the contribution of Saharan Africa. IFPRI Discussion Paper the International Potato Center 1972– 912. International Food Policy Research 2007. Working Paper 2008–6. Interna- Institute (IFPRI): Washington DC, USA. tional Potato Center (CIP): Lima, Peru.

Ragasa, C., Dankyi, A., Acheampong, P., Walker, T.S. 1994. Patterns and implications Nimo Wiredu, A., Tripp, R., Chapoto, A. of varietal change in potatoes. Social & Asamoah, M. 2013a. Patterns of Science Department Working Papers adoption of improved maize Series No. 1994–3. International Potato technologies in Ghana. Draft Report. Center (CIP): Lima, Peru. International Food Policy Research Institute (IFPRI): Washington DC, USA. Walker, T., Pitoro, R., Tomo, A., Sitoe, I., Salência, C., Mahanzule, R., Donovan, C. Ragasa, C., Dankyi, A., Acheampong, P., & Mazuze, F. 2006. Priority-setting for Nimo Wiredu, A., Tripp, R., Chapoto, A. public-sector agricultural research in & Asamoah, M. 2013b. Patterns of Mozambique with the national agricul- adoption of improved rice technologies tural survey data. IIAM–DFDTT Research in Ghana. Draft Report. International Report No. 3E. Instituto de Investigacao Food Policy Research Institute (IFPRI): de Agricultura Mozambicana: Maputo, Washington DC, USA. Mozambique.

Rueda, J.L., Ewell, P.T., Walker, T.S., Soto, Walker, T.S. 2010. Guidelines for data M., Bicamumpaka, M. & Berrios, D. collection for Objective 1 of the DIIVA 1996. Economic impact of high-yielding, Project. Fletcher, North Carolina, USA. late-blight-resistant varieties in the Eastern and Central African Highlands. Walker, T.S. & Adam, A. 2011. Guidelines for In: Walker, T. & Crissman, C. (Eds.) data collection for Objective 2 of the Case studies of the economic impact of DIIVA Project. Fletcher, North Carolina, CIP-related technologies. International USA. Potato Center (CIP): Lima, Peru. Walker, T.S., Alene, A., Diagne, A., Labarta, Sanni, K., Toure, A. & Diagne, A. 2011. R., LaRovere, R. & Andrade, R. 2011a. Variety release and registration practices Assessment of the late 1990s IARC for rice in African countries. Study commodity by country data on varietal Report. AfricaRice: Cotonou, Benin. release, cultivar-specific adoption, and strength of NARS in crop improvement Simtowe, F. & Mausch, K. 2012. Assessing the in Sub-Saharan Africa. Fletcher, North effectiveness of agricultural R&D in Carolina, USA. eastern and southern Africa: Cases of groundnuts, pigeonpea and sorghum crop Walker, T., Thiele, G., Suarez, V. & Crissman, improvement programs. Diffusion and C. 2011b. Hindsight and foresight about impact of improved varieties in Africa potato production and consumption. (DIIVA), Objective 1 Report. International Social Sciences Working Paper 2011-5. Crops Research Institute for the Semi-Arid International Potato Center (CIP), Lima, Tropics (ICRISAT): Nairobi, Kenya. Peru.

Swanckaert, J., De Groote, H. & D’Hawese, Walker, T. & Crissman, C. 1996. Case studies M. 2012. Mapping the effect of market of the economic impact of CIP-Related

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 77 technologies. International Potato of improved maize and wheat varieties Center (CIP): Lima, Peru. in east and southern Africa. CIMMYT’s Report on Objective 2. Part II – Wheat Yigezu, Y.A., Yirga, C. & Aw-Hassan, A. in Ethiopia. International Maize and 2012a. Varietal output and adoption in Wheat Improvement Center (CIMMYT): barley, lentils, chickpeas, faba beans, Nairobi, Kenya. and field peas in Ethiopia, Eritrea, and Sudan. Diffusion and impact of Zheng, D., Alwang, J., Norton, G.W., improved varieties in Africa (DIIVA), Ob- Shiferaw, B. & Jaleta, M. 2013. Ex-post jective 1 Report. International Centre for impacts of improved maize varieties on Agricultural Research in the Dry Areas poverty in rural Ethiopia. Diffusion and (ICARDA): Aleppo, Syria. impact of improved varieties in Africa (DIIVA), Objective 3 Report. VPI Blacks- Yigezu, Y.A., Yirga, C., Aw-Hassan, A. & burg, Virginia, USA, and International Labarta, R. 2012b. Barley in Ethiopia: Maize and Wheat Improvement Center Verifying the adoption estimates from a (CIMMYT): Nairobi, Kenya. panel of experts with secondary data and household and community focus Zereyesus, Y.A. & Dalton, T.J. 2012 (draft). group surveys. Diffusion and impact of The Impact of sorghum varietal improved varieties in Africa (DIIVA), improvement research in Sudan. Objective 2 Report. International Centre Diffusion and impact of improved for Agricultural Research in the Dry varieties in Africa (DIIVA), Objective 3 Areas (ICARDA): Aleppo, Syria. Report. Kansas State University: Manhattan, Kansas, USA. Yirba, C., Jaleta, M., De Groote, H. & Gitonga Z. 2012. Diffusion and impact

78 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project d – EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO Survey Survey Survey Survey Source – 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2010 2009 2010 2009 2009 2009 2009 2009 2010 2009 2010 2009 Year a 2 6 8 9 – 31 66 29 36 20 48 36 21 44 61 19 46 31 39 35 44 52 13 15 34 16 44 55 13 19 46 31 100 (% area) Adoption Adoption of improved varieties by characteristic Adoption of improved – 9,525 6,667 (ha) 61,000 60,000 49,000 56,857 75,145 839,000 242,000 205,000 349,000 858,840 133,000 183,820 941,000 921,475 121,000 398,000 194,930 233,440 915,877 945,559 410,100 224,584 207,494 260,287 108,000 285,000 393,716 532,883 1,900,000 3,600,000 Total crop area area crop Total b 0 0 0 5 2 0 5 4 1 0 4 0 0 0 7 0 0 5 7 0 0 1 4 1 0 0 – 28 17 20 16 30 10 entries Undated 0 4 0 0 0 9 9 8 2 0 7 6 3 9 – 14 20 19 17 20 16 25 12 15 32 25 26 29 15 12 45 25 17 1990–2010 0 5 0 0 0 4 0 0 0 4 2 0 5 0 0 0 0 4 1 0 0 6 5 2 3 0 2 3 0 1 – 18 12 1970–1989 Released varieties by time period if dated c 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 – Before 1970 Before FTE 9.4 2.4 1 5.1 3 3.6 8.9 2.2 4 7.2 1.5 0 8.4 1 2.6 5.1 6.4 5.1 8.7 5.2 5.3 11.5 14.2 22.5 10.8 11 20.5 18.6 42 21.1 21.4 14.5 12.2 Scientists in 2009 Angola Benin Burundi Cameroon Cote d’Ivoire DR Congo Ghana Guinea Kenya Malawi Mozambique Nigeria Tanzania Togo Uganda Zambia Zimbabwe Eritrea Ethiopia Sudan Country Uganda Eritrea Ethiopia Burundi DR Congo Ethiopia Malawi Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe References for and caveats about these data are given at http://www.asti.cgiar.org/diiva where the detailed dataset is stored References for and caveats about these data are given at http://www.asti.cgiar.org/diiva that are not formally released Undated entries refer to improved varieties Indicates missing data EO = Expert opinion

Cassava Cassava Cassava Cassava Cassava Cassava Cassava Cassava Cassava Cassava Cassava Cassava Cassava Cassava Cassava Cassava Cassava Chickpea Chickpea Chickpea Crops Banana Barley Barley Bean Bean Bean Bean Bean Bean Bean Bean Bean Bean a. b. c. d. Annex 1. Summary data on FTE scientists, varietal release, and adoption of improved varieties by crop and country varieties by crop and adoption of improved Annex 1. Summary data on FTE scientists, varietal release,

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 79 S S – EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO Source – 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2010 2010 2010 2010 2010 2010 2009 2009 2009 2009 2009 Year a

– 4 2 51 10 70 27 44 81 63 10 53 11 17 39 27 31 40 16 17 45 87 25 47 58 20 12 19 47 32 55 57 10 (% area) Adoption Adoption of improved varieties by characteristic Adoption of improved – 4000 3,000 3,900 (ha) 68,000 83,076 74,330 74,000 77,000 20,640 94,946 622,830 110,650 128,000 163,612 263,786 352,400 218,904 147,721 178,707 537,606 230,749 458,222 266,946 335,031 588,651 535,000 253,000 204,073 5,200,000 3,800,000 2,600,000 1,100,000 Total crop area area crop Total b – 9 1 0 0 0 2 0 0 0 0 5 0 0 0 0 0 0 0 0 2 0 0 0 0 0 1 0 0 0 0 0 11 entries Undated – 1 4 5 1 4 2 9 3 3 0 3 1 6 3 0 6 5 6 7 7 6 1 7 10 12 14 10 17 17 22 17 12 1990–2010 – 9 4 5 7 4 3 0 0 5 0 5 2 3 3 2 0 4 1 4 8 0 0 6 8 6 3 0 1 0 4 18 16 1970–1989 Released varieties by time period if dated – 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 0 0 0 3 5 0 2 0 2 4 0 0 4 0 0 Before 1970 Before FTE 8.5 1.1 4 0.8 4.5 2 1.6 3.4 0.4 6.4 4.8 2.4 5.2 1.8 1.9 1 6.85 8.68 6.85 2.5 3 1 2 2.25 1.55 1.6 3.83 5 1.15 2 5.28 10.4 16 Scientists in 2009 Country Benin Burkina Faso Cameroon Cote d’Ivoire DR Congo Ghana Guinea Malawi Mali Mozambique Niger Nigeria Senegal Tanzania Togo Uganda Zambia Zimbabwe Ethiopia Sudan Ethiopia Burkina Faso Kenya Malawi Mali Niger Nigeria Senegal Tanzania Uganda Zambia Eritrea Ethiopia References for and caveats about these data are given at (website) where the detailed dataset is stored that are not formally released Undated entries refer to improved varieties Indicates missing data EO = Expert opinion

Crops Cowpea Cowpea Cowpea Cowpea Cowpea Cowpea Cowpea Cowpea Cowpea Cowpea Cowpea Cowpea Cowpea Cowpea Cowpea Cowpea Cowpea Cowpea bean Faba bean Faba pea Field Groundnut Groundnut Groundnut Groundnut Groundnut Groundnut Groundnut Groundnut Groundnut Groundnut Lentil Lentil a. b. c. d. (cont.) Annex 1. Summary data on FTE scientists, varietal release, and adoption of improved varieties by crop and country varieties by crop and adoption of improved (cont.) Annex 1. Summary data on FTE scientists, varietal release,

80 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO Survey Survey Survey Survey Survey Survey Survey Survey Source Seed Sales Seed Sales Seed Sales Seed Sales Seed Sales Seed Sales 2006 2009 2009 2008 2006 2010 2006 2006 2006 2009 2009 2009 2009 2009 2012 2009 2009 2009 2009 2009 2010 2010 2010 2009 2010 2009 2009 2010 2009 2009 2009 2009 2009 Year a

5 3 1 10 28 69 43 10 35 54 81 93 54 49 82 56 15 60 73 95 72 97 31 11 25 35 50 50 50 23 29 36 74 (% area) Adoption Adoption of improved varieties by characteristic Adoption of improved (ha) 72,000 45,816 887,000 911,942 830,000 555,175 693,027 302,253 863,577 440,000 408,606 190,624 500,000 118,167 175,734 164,146 152,998 150,777 102,000 1,600,000 1,800,000 1,900,000 1,600,000 1,600,000 3,000,000 1,500,000 1,500,000 3,700,000 1,300,000 6,500,000 1,500,000 3,700,000 1,100,000 Total crop area area crop Total b – 0 0 0 0 2 0 0 0 5 7 8 2 1 3 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 24 11 21 entries Undated – 0 7 4 1 8 8 3 6 5 6 3 8 7 36 64 82 32 73 19 31 15 10 44 11 15 15 27 30 10 12 200 194 1990–2010 – 0 7 4 9 0 3 9 0 2 9 2 8 2 1 8 7 5 2 1 0 2 10 11 16 13 38 11 22 10 10 15 10 1970–1989 Released varieties by time period if dated – 0 0 4 2 0 0 0 1 2 0 0 0 0 1 1 0 0 8 0 0 0 0 1 0 0 0 0 0 3 0 0 0 Before 1970 Before FTE 4.35 3.7 6.4 5 7.1 2.93 3.4 5.85 5.8 4.46 6.43 5 1.5 3.2 5 0.67 1.2 3 4.6 3.7 12 21.45 62 16 12 30.75 17 13 59 17.5 77.5 16 30 Scientists in 2009 Country Angola Ethiopia Kenya Malawi Mozambique Tanzania Uganda Zambia Zimbabwe Benin Burkina Faso Cameroon Cote d’Ivoire DR Congo Ghana Guinea Mali Senegal Nigeria Togo Burkina Faso Mali Niger Nigeria Senegal Kenya Malawi Tanzania Ethiopia Kenya Malawi Rwanda Uganda References for and caveats about these data are given at (website) where the detailed dataset is stored that are not formally released Undated entries refer to improved varieties Indicates missing data EO = Expert opinion

Crops Maize–ESA Maize–ESA Maize–ESA Maize–ESA Maize–ESA Maize–ESA Maize–ESA Maize–ESA Maize–ESA Maize–WCA Maize–WCA Maize–WCA Maize–WCA Maize–WCA Maize–WCA Maize–WCA Maize–WCA Maize–WCA Maize–WCA Maize–WCA Pearl millet Pearl millet Pearl millet Pearl millet Pearl millet Pearl Pigeonpea Pigeonpea Pigeonpea Potato Potato Potato Potato Potato (cont.) Annex 1. Summary data on FTE scientists, varietal release, and adoption of improved varieties by crop and country varieties by crop and adoption of improved (cont.) Annex 1. Summary data on FTE scientists, varietal release, a. b. c. d.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 81 EO EO EO EO EO EO EO EO EO EO EO EO Survey Survey Survey Survey Survey Survey Survey Survey Survey Survey Survey Survey Survey Survey Survey Survey Survey Survey Survey Experts Source Survey & EO 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2010 2009 2009 2009 2009 2009 2009 2009 2010 2009 2010 2010 2009 2010 2009 2010 2009 2009 2009 2009 2009 2009 Year a

3 4 83 71 35 52 33 39 58 22 85 35 25 50 69 43 16 13 24 59 46 40 33 15 20 41 40 38 50 75 94 100 100 (% area) Adoption Adoption of improved varieties by characteristic Adoption of improved 1065 2556 1000 (ha) 38,700 92,243 15,969 99,653 12,775 73,000 36,492 38,000 13,000 35,000 70,000 569,000 482,400 122,700 712,800 646,100 101,000 434,200 627,600 120,000 173,172 240,425 874,219 1,400,000 1,200,000 1,700,000 1,100,000 2,500,000 4,700,000 6,700,000 Total crop area area crop Total b – – – – – – – – 0 4 0 0 0 0 0 0 0 4 0 2 0 7 4 0 0 0 0 2 7 2 0 0 0 entries Undated – – – – – – – – 9 5 7 2 8 3 8 0 5 7 3 7 2 7 11 26 82 16 14 35 25 10 50 16 15 1990–2010 – – – – – – – – 9 3 2 4 3 6 6 1 3 1 1 3 3 0 14 16 20 10 40 10 32 10 22 10 31 1970–1989 Released varieties by time period if dated – – – – – – – – 1 0 2 0 2 4 4 0 1 1 0 0 0 0 0 0 1 1 0 0 0 0 0 0 12 Before 1970 Before FTE – – – – – – – 7 6.25 8 4.5 8.25 6.75 7 2.96 3 7.75 5.45 2.5 2.4 1.55 3.2 2.6 4.5 0.8 1.9 12 15.25 15.25 11.75 11 12 18.2 Scientists in 2009 Country Benin Burkina Faso CAR Cameroon Cote d’Ivoire DR Congo Ghana Guinea Kenya Madagascar Mali Nigeria Rwanda Senegal Sierra Leone Tanzania Gambia The Togo Uganda Burkina Faso Kenya Mali Niger Nigeria Senegal Sudan Tanzania Benin Burundi Cameroon Cote d’Ivoire DR Congo Ghana References for and caveats about these data are given at (website) where the detailed dataset is stored that are not formally released Undated entries refer to improved varieties Indicates missing data EO = Expert opinion

Crops Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Sorghum Sorghum Sorghum Sorghum Sorghum Sorghum Sorghum Sorghum Soybean Soybean Soybean Soybean Soybean Soybean a. b. c. d. (cont.) Annex 1. Summary data on FTE scientists, varietal release and adoption of improved varieties by crop and country varieties by crop and adoption of improved (cont.) Annex 1. Summary data on FTE scientists, varietal release

82 — Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project

– EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO EO Survey Survey Survey Source Assumed Assumed Assumed Assumed (1998 Results) (1998 Results) (1998 Results) (1998 Results) – 2009 2009 2009 2009 2009 2009 2009 2009 2010 2010 2010 2010 2010 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 2009 Year a

– 9 0 0 9 4 9 5 74 92 79 39 97 28 63 75 10 25 15 14 100 100 100 100 100 100 100 (% area) Adoption Adoption of improved varieties by characteristic Adoption of improved – 3000 9000 2000 (ha) 80,000 60,450 45,000 67,300 34,296 13,000 33,000 60,000 613,000 148,000 125,000 130,000 123,086 480,000 620,000 131,594 149,200 171,000 758,000 367,500 254,000 1,490,764 3,000,000 Total crop area area crop Total b 3 0 0 0 6 5 4 1 0 0 0 0 0 0 0 1 0 0 0 1 0 0 3 14 15 10 14 entries Undated 7 5 1 1 7 7 2 5 0 0 5 3 4 0 0 10 11 19 33 15 10 20 66 12 29 25 23 1990–2010 0 5 0 8 2 1 0 9 1 0 8 0 0 7 0 0 0 0 0 0 0 0 10 34 16 16 13 1970–1989 Released varieties by time period if dated 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 19 Before 1970 Before FTE – – 1 2.1 1 3.2 2.8 2 2 8.7 2.1 4 3.3 7.1 3 3.1 4.5 4.6 14.6 10.9 15.9 18.05 28 12 12 12.1 11.8 Scientists in 2009 Country Kenya Malawi Mozambique Nigeria Tanzania Togo Uganda Zambia Zimbabwe Burundi Mozambique Rwanda Tanzania Uganda Ethiopia Kenya Tanzania Zambia Zimbabwe Benin Cameroon Cote d’Ivoire Ghana Guinea Nigeria Togo Uganda References for and caveats about these data are given at (website) where the detailed dataset is stored that are not formally released Undated entries refer to improved varieties Indicates missing data EO = Expert opinion

Crops Soybean Soybean Soybean Soybean Soybean Soybean Soybean Soybean Soybean Sweetpotato Sweetpotato Sweetpotato Sweetpotato Sweetpotato Wheat Wheat Wheat Wheat Wheat Yams Yams Yams Yams Yams Yams Yams Yams (cont.) Annex 1. Summary data on FTE scientists, varietal release, and adoption of improved varieties by crop and country varieties by crop and adoption of improved (cont.) Annex 1. Summary data on FTE scientists, varietal release, a. b. c. d.

Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project — 83

Independent Science and Partnership Council Secretariat c/o FAO Viale delle Terme di Caracalla snc 00153 Rome, Italy www.sciencecouncil.cgiar.org t 39 06 57056782 f 39 06 57053298 [email protected] Synthesis Report for Objectives 1 and 2 of Bill & Melinda Gates Foundation’s DIIVA Project Independent Science and Partnership Council July 2014