AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER Background report to Sri Lanka Poverty Assessment AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER Background report to Sri Lanka Poverty Assessment © 2021 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW Washington DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org This work is a product of the staff of The World Bank with external contributions. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy, completeness, or currency of the data included in this work and does not assume responsibility for any errors, omissions, or discrepancies in the information, or liability with respect to the use of or failure to use the information, methods, processes, or conclusions set forth. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Nothing herein shall constitute or be construed or considered to be a limitation upon or waiver of the privileges and immunities of The World Bank, all of which are specifically reserved. Rights and Permissions The material in this work is subject to copyright. Because The World Bank encourages dissemination of its knowledge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as full attribution to this work is given. Any queries on rights and licenses, including subsidiary rights, should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2625; e-mail: pubrights@worldbank.org. Cover design: Wojciech Wolocznik, Cambridge, United Kingdom Interior design and typesetting: Piotr Ruczyński, London, United Kingdom Contents Acknowledgements   6 Abbreviations   6 Executive summary   7 1. Introduction   8 2. Agricultural Production of Farm Households, 2006 – 16    12 3. Determinants of Higher Agricultural Productivity and Earnings     21 Diversification to Improve Productivity    22 Gender Gaps in Productivity and Earnings   29 4. Conclusion and Policy Implications    38 Appendix    41 References   44 Boxes Box 1 Are Self-employed Farmers and Wage Workers Different?   14 Box 2 Oaxaca-Blinder decomposition   32 Box 3 Measuring the gender gap in agricultural productivity: A literature review   36 Figures Figure 1 Share of farm households engaged in different types of products   13 Figure B1.1 Number of wage workers, by crop and gender (1,000)   14 Figure B1.2 Kernel density estimates of the log of wages, by crop   14 Figure 2 Trend in agricultural export value, 2011 – 2019 (Rs, billions)   15 Figure 3 Kernel density estimates of the log of gross output per acre, by crop   17 Figure 4 Crop productivity (Gross value of output/acre, Rs), 2006-2016   18 Figure 5 Correlation between the Export Orientation Index and productivity   23 Figure 6 Simpson Index of Diversification and Export Orientation Index, by productivity quintile   23 Figure 7 District level variation in diversification, export orientation, productivity, share of paddy farmers and share of paddy production in national production   25 Figure 8 Crop mix by farmer’s gender   30 Figure 9 Oaxaca-Blinder decomposition for productivity gender gap   33 Figure 10 Oaxaca-Blinder decomposition for earning gender gap   34 Figure 11 Differences in daily time use by gender   36 Figure 12 Agriculture workers’ work location, by gender   36 Tables Table 1 Shares of farm households engaged in selected crop activities by district   16 Table 2 Share of crop output kept for farm household’s own consumption (percent)   18 Table 3 Paddy output value as share of gross income and farm output value, by household income quintile (percent)   19 Table 4 Characteristics of farm households in Northern and Eastern Provinces (NE) vs. other provinces and by gender of household head   26 Table 5 Determinants of crop productivity and earnings   27 Table 6 Key summary statistics by crop and gender   31 Table 7 Productivity and earnings gap decompositions   34 Table A.1 Distribution of production systems for farm households, by district   41     Table A.2 Summary statistics of self-employed farmers 42 Table A.3 Summary statistics of agricultural wage workers   43 AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 6 Acknowledgements This report was prepared as a background paper to the "Sri Lanka Poverty Assessment. Accelerating Eco- nomic Transformation" (World Bank, 2021). It was written by Yeon Soo Kim (Senior Economist, World Bank), Emiko Fukase (Consultant, World Bank), Cristina Chiarella (Consultant, World Bank). The work was carried out under the overall guidance of Faris H. Hadad-Zervos (Country Director for Sri Lanka, Nepal and Maldives), Zoubida Allaoua (Regional Director, South Asia), Chiyo Kanda (Country Man- ager, Sri Lanka and Maldives), Tae Hyun Lee (Lead Country Economist), and Andrew Dabalen (Practice Manager, Poverty and Equity).The team is grateful for feedback and comments from Andrew Goodland (Lead Agriculture Specialist), Seenithamby Manoharan (Senior Rural Development Specialist), Athula Senaratne (Senior Agriculture Specialist), Kishan Abeygunawardana (Senior Economist) and Hafiz Zai- nudeen (Associate Operations Officer). The report also benefited from consultations with Wijaya Jaya- tilaka (Executive Director, Centre for Poverty Analysis). The team would like to thank the Government of Sri Lanka for its support and the Department of Census and Statistics (DCS) for sharing its data. Any remaining errors are the responsibility of the authors. Abbreviations ISIC International Standard Industrial Classification of All Economic Activities NE Northern and Eastern Provinces OLS ordinary least squares SID Simpson Index of Diversification AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 7 Executive summary Identifying opportunities to increase agricultural productivity and incomes is an important priority for rural development. Progress toward poverty reduction continued in recent years, but the contribution of the agriculture sector was weak, mainly because productivity improvements were relatively limited. Using detailed individual-level data on agricultural activities, this paper analyzes agricultural produc- tion patterns and associated productivity of farm households. Particular attention is paid to (i) diversifi- cation toward higher-value, export-oriented crops as a means to increase productivity and earnings; and (ii) gender differences in farming activities and outcomes. The role of structural factors such as access to land is also considered. There are three key findings in this paper. First, diversified farmers, especially those with a crop mix that is focused on export crops or other high-value crops have higher productivity and earnings. The productivity of paddy cultivation is significantly lower than that of other crops, leading to low earn- ings. Second, production patterns and productivity levels differ distinctively between men and women farmers. Female farmers have higher productivity, as measured by output value per acre, which is main- ly explained by their smaller plot size and a crop mix that consists of higher-value crops. However, despite higher productivity, overall farm incomes are lower among female farmers, mainly due to lower access to land. Third, once land size and crop mix are accounted for, unequal access to resources even- tually leads to a male productivity advantage — referred to as conditional advantage, after differential access to resources is controlled for via multivariate analysis. Policies to increase the crop mix toward higher-value, export-oriented crops and to equalize access to resources, including land and agricultural inputs, could help improve productivity and income, and reduce gender disparities. 1.  Introduction AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 1. Introduction 9 Sri Lanka has been steadily transitioning from a predominantly rural, agrarian economy toward a mod- ern economy structured around industries and services. The primary sector’s contribution to gross domestic product (GDP) was low, at 7.4 percent in 2019. Between 2002 and 2016, the share of employed engaged in agriculture dropped from around 30 percent to 25.8 percent. The majority shifted away from agriculture toward services, and the change was marked after 2012. While the production sector alone contributes a small share toward Sri Lanka’s total GDP, the broader agriculture and food sector is signif- icantly larger due to the sector’s strong backward and forward linkages. Food and beverage manufactur- ing alone accounted for about 6 percent of GDP and slightly over 5 percent of employment. 1 With a large share of the poor engaged in agriculture, identifying opportunities to increase produc- tivity and incomes in the sector is an important priority for poverty reduction and shared prosper- ity. While poverty decreased between 2012 and 2016, agricultural earnings showed weak improvements, in stark contrast to the preceding period (2009 – 2012), which saw strong progress in the sector primar- ily owing to favorable prices (World Bank 2021a). Sustained progress in the sector can be brought about by increases in productivity, particularly among household-based enterprises that comprise the major- ity of smallholder farmers. This paper analyzes agricultural production patterns and associated productivity of farm house- holds. It offers an overview of historical trends and patterns in agricultural production systems, crop choices, and productivity, and them more deeply analyzes the determinants of productivity and earn- ings. In addition to structural factors such as access to land, it focuses on diversification as a key means to increasing productivity and earnings, where diversification is understood to encompass (i) “a shift of resources from farm to nonfarm activities, (ii) use of resources in a larger mix of diverse and com- plementary activities within agriculture, and (iii) a movement of resources from low value agriculture to high value agriculture” (Joshi et al. 2004). The focus of this paper will be on the latter two types of diversification strategies. 2 Particular attention is paid to gender differentials in farming activities and outcomes. While many women engaged in the sector work as unpaid family workers (Hirimuthugodage et al 2014; World Bank 2021b)., an increasing number of women farmers have been successful at earning a living from crop- or livestock-related activities. Future poverty reduction hinges on strong and sustained improvements in rural livelihoods in the agricultural sector, especially among those groups and in areas where the produc- tivity and earnings gaps are the widest. As described in further detail below, there are large disparities between men and women farmers, and narrowing these disparities could help address these challenges. 1. Estimated share of GDP is from national accounts data for 2019; share of employment is based on HIES 2016. 2. The first strategy — a shift in resources from farm to nonfarm activities — is examined in a separate background paper titled “The Rural Nonfarm Sector and Livelihoods Strategies in Sri Lanka” (World Bank 2021). AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 1. Introduction 10 Gendered roles in agricultural activities have been documented for Sri Lanka in the past. For exam- ple, focus group discussions found that although both men and women participate in field production, men tend to dominate the later stages of the value chain, whereas women tend to participate in the most time-consuming activities at earlier stages. This arrangement could be due to the constraints on wom- en’s access to transportation; it allows women to work close to home while offering flexibility to also per- form household chores. It could also arise because men are more familiar with productive technologies, have more information about markets, and have more connections, influential networks and access to capital (FAO, 2018). These disadvantages may all be related to biases against women. Spice-based indus- tries (cinnamon and pepper) and tea and rubber crops in Sri Lanka have a high rate of female partici- pation, although the activities within these industries are also gendered (FAO 2018). Another important example of social norms in Sri Lanka agriculture is that although women participate in planting, most of them are considered unpaid family workers (Ratnayake 2009). The analysis relies on detailed individual-level data on agricultural activities that can also be disag- gregated by gender. The main data source is the Household Income and Expenditure Survey (HIES) 2016. The survey was conducted by the Department of Census and Statistics of Sri Lanka and the data are rep- resentative at the national and district levels. The survey includes questionnaires on demographic and socioeconomic characteristics of individuals and households as well as relatively detailed income mod- ules. As part of the income module, detailed information is gathered on wage employment and on agri- cultural and nonagricultural self-employment. Data on agricultural activities include information on the type of crops, cultivated area, value and quantity of output, agricultural inputs, own-consumption, and use of subsidies. While not all crops are individually listed, information is collected on paddy, other seasonal crops (chilies, onions, vegetables, cereals, yams, tobacco), and annual crops (tea, rubber, coco- nuts, coffee/pepper/betel, banana/fruits); information is also collected on livestock (meat, eggs, milk), fish, horticulture, and other crops and livestock. A key advantage of the data is that all this information is available for each individual farmer, allowing gender-disaggregated analysis. 3 There are three key findings in this paper. First, diversification within agriculture plays a key role in improving productivity and income. Diversification includes diversifying the set of activities undertaken as well more broadly shifting from food crops toward high-value crops with higher export orientation. 3. While the survey aimed to collect individual data on production, it is possible that households manage farm plots jointly, where males may contribute to the work on a female-managed plot. The relevant question is understood as reporting the household member with the largest contribution to the plot, while contribution by others is difficult to know. Certain other data limitations should be acknowledged as well. As described above, information is available for most major crops but not all are listed separately (e.g., information on “vegetables” is collected in the aggregate). Agricultural inputs consist of seeds, fer- tilizers, chemicals, hired labor, agricultural equipment/rental, and others, but only the aggregated value is available in the data. This makes it difficult to know whether production practices changed and how they would have affected productivity. The absence of information on hours worked makes it difficult to construct detailed measures of labor input. There is also no information on technology use, cropping intensity, or actual use of extension services (the survey only asks about the availability of a community extension center). AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 1. Introduction 11 Paddy productivity is particularly low, constraining productivity growth among farmers who devote a large share of their land to paddy. Second, agricultural production patterns and productivity levels dif- fer distinctively between men and women farmers. Female farmers have higher productivity, as meas- ured by output value per acre, which is mainly explained by their smaller plot size and a crop mix that consists of higher-value crops. 4 However, despite higher productivity, overall farm incomes are low- er among female farmers, mainly due to lower access to land. Third, once land size and crop mix are accounted for, unequal access to resources eventually leads to a male productivity advantage — referred to as conditional advantage, after differential access to resources is controlled for via multivariate anal- ysis. Policies to increase the crop mix toward higher-value, export-oriented crops and to equalize access to resources, including land and agricultural inputs, could help improve productivity and income, and reduce gender disparities. 4. Farmers are defined as those who reported a positive gross crop output value in the survey. 2.  Agricultural Production of Farm Households, 2006 – 16 AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 2. Agricultural Production of Farm Households, 2006 – 16 13 Agricultural production decisions are influenced by a variety of factors, including agroecology and climate. Sri Lanka’s agricultural land is broadly divided along three climatic zones, i.e., dry, intermedi- ate and wet zones. Two thirds of the agricultural area is located in the dry zone, which covers the north- ern, eastern and south-eastern parts of the country. This is also where the bulk of the irrigation infra- structure is located. The variation in rainfall and altitude creates diverse agroecological regions that offer opportunities for diversified production. There are two distinct growing seasons that fall in line with seasonal rainfall patterns: these are the Yala (April – June) and Maha (October – June) seasons. The tea, rubber and coconut plantations have traditionally been the basis for export agriculture, with many of them located in the Southwest and highland areas. Extreme weather events have impacted produc- tivity in recent years: for example, the prolonged drought that started in 2016 and lasted through a good part of 2017 was arguably the worst drought in 40 years, affecting almost 4 million people, and leading to widespread crop failure and large income drops. Sri Lanka has been identified as a hot spot in future climate change scenarios (Mani et al. 2018). In Sri Lanka, the primary form of agriculture among farm households is crop production, with a large number of farmers engaged in rice cultivation. As of 2016, over 88 percent of farm households maintained a crop-only production scheme; about 7 percent produced livestock only and about 5 per- cent engaged in mixed production. Production systems have overall not changed much, apart from a slight decrease in mixed crop-livestock schemes. There is marked variation across districts, with FIGURE 1 Share of farm households engaged livestock activities concentrated in the Northern in different types of products and Eastern Provinces and crop production dom- Paddy inating in the Southern, Sabaragamuwa, and Uva Vegetables Cereals Provinces (annex table A.1). Food crops Yams Chilies Other annual The trend has been mixed for some higher-val- Onions ue crops, while participation in livestock and Tobacco other activities has increased. Nearly half of all Banana Cash crops Coffee farm households cultivate paddy, and this share Coconut has remained almost unchanged in the decade Tea/rubber Horticulture after 2006. The popularity of some higher-value Other livestock crops — such as coffee, pepper, betel, tea, and rub- Milk Other Egg ber — increased, while it decreased for banana/fruits Fish and coconut (figure 1), the latter a major ingredient Meat in Sri Lanka food and an increasingly important 0 5 10 15 20 25 30 35 40 45 50 export product. Vegetables are grown by 12 percent Percent 2006 2016 of farmers. The focus in this paper is on smallhold- er farmers since the analysis is based on household Source: World Bank staff calculation using HIES. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 2. Agricultural Production of Farm Households, 2006 – 16 14 surveys. Figure 1 may not fully reflect the production generated by large plantations, for example. Box 1 explores the different profiles and earning opportunities of smallholders and agricultural wage work- ers, including those in the estate sector. BOX 1 Are Self-employed Farmers and Wage Workers Different? According to the HIES 2016, there were 0.57 million agricultural wage workers involved in crop activities in 2016. Figure B1.1 shows the number of wage workers by crop type and wage worker’s gender. Tea production was the leading agricultural wage sector, employing 41 percent of wage workers, of whom 56 percent were women. Tea is the top agricultural export commodity in Sri Lanka: out of 300 million kg of tea produced, 293 million kg were exported in 2019 (Central Bank of Sri Lanka 2020). Smallholdings account for over 70 percent of the country’s tea production, and the remaining FIGURE B1.1 Number of wage workers, by crop share is produced in plantations (ILO 2018). The paddy sector and gender is the second largest employer of agricultural wage workers (87 percent of whom are male), and most paddy output is sold do- 250 Number of workers mestically. Spices (e.g., cinnamon) and coconuts are export-ori- 200 ented sectors and appear to hire more male wage workers. (thousand) 150 The summary statistics reported in appendix table A.2 reveal 100 somewhat different personal profiles for Sri Lanka agricultur- 50 al wage workers and self-employed farmers. Wage workers are younger (45 years old on average) than self-employed farmers 0 Coconut Other Paddy Spices Tea Vegetables (52 years old on average); they are less educated (6.1 years of ed- crops ucation, compared to 8.5 years of education for self-employed farmers) and less likely to be ethnic majority Sinhala (54 percent Sum of male Sum of female versus 91 percent). A disproportionately large share—33 per- Source: HIES 2016. cent—of wage workers work in the estate sector (43 percent for Note: Under ISIC (International Standard Industrial Classification of All women and 27 percent for men), while only 2 percent of self-em- Economic Activities) categorization, the crops that have more than 50 ployed farmers work in the estate sector. observations are reported as separate categories, including paddy (ISIC 112), vegetables (ISIC 113), coconut (ISIC 126), tea (ISIC 127), and spices Figure B1.2 and the accompanying table show average wages per (ISIC 128). The activities that are not crop specific and the crops that have crop and by gender. The figure clearly reveals different patterns fewer than 50 observations are aggregated as “other crop activities.” of earnings for self-employed and wage workers. In particular, tea is the sector that pays the lowest wages. This finding contrasts with the pattern among self-employed farmers, whose export-oriented crops such as tea generate higher productivity and income. Spices appear to be the highest paid crop, while paddy is in the middle. FIGURE B1.2 Kernel density estimates of the log of wages, by crop 1.0 Monthly wages (Rs) Log of monthly wages (Rs) 0.8 t-test 0.6 Total Male Female Total Male Female (p-value) Density 0.4 Paddy 14,575 15,414 8,794 9.39 9.48 8.81 0.000 *** Vegetables 13,183 15,052 10,203 9.30 9.45 9.06 0.000 *** 0.2 Tea 10,972 12,832 9,536 9.16 9.33 9.03 0.000 *** 0 Coconuts 15,801 17,204 12,422 9.57 9.65 9.35 0.0098 ** 6 8 10 12 14 Spices 24,695 25,459 20,597 9.92 10.00 9.49 0.001 *** Log of wages Other crop 13,541 15,428 11,435 9.32 9.45 9.19 0.000 *** Paddy Vegetables Tea activities Coconut Spices Total 13,330 15,460 10,319 9.3090 9.4718 9.0789 0.000 *** Source: HIES 2016 Note: Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 2. Agricultural Production of Farm Households, 2006 – 16 15 The results of Mincerian earnings regressions (not reported here) reveal that the spice and coconut sectors pay higher wages than the paddy sector, while wages in the tea and paddy sectors are not statistically different. These results differ from those for self-em- ployed farmers; their earnings from all the other crops (except other cereals) are higher than from paddy. The result of Oaxaca-Blinder decomposition suggests that wage workers in spices contribute to widening the wage gender gap in favor of men, as spice wage workers are relatively well paid and represented disproportionately by males. This suggests that farmers can be heterogenous even within the same sector. For instance, tea pluckers in the estate sector and small- holder entrepreneurs cultivating tea appear to face different constraints and opportunities and to show different gender gap patterns. The most important higher-value crops include tea, rubber, spices and coconut, and a large share of their production is aimed toward export markets. Key export crops have exhibited positive export per- formance in recent years (Figure 2). Sri Lanka is among the world’s leading tea producers as the climate in the central highlands and in some low-elevation areas is favorable for the production of high-qual- FIGURE 2 Trend in agricultural export value, ity tea. Today about 70 percent of tea production 2011 – 2019 (Rs, billions) is accounted for by smallholders, with plantations 500 450 accounting for the rest (ILO 2018). In the HIES data, 400 about 17 percent of farm households report growing Export value (Rs, billion) 350 tea and/or rubber in 2016. 5 Coconut production was 300 reported by about 20 percent of farm households in 250 2016, which is a notable decline from 31 percent in 200 2006. Given the recent boost in coconut production 150 and exports, it is possible that this change reflects 100 50 a consolidated shift toward production at scale. 0 2011 2012 2013 2014 2015 2016 2017 2018 2019 While paddy is cultivated across the country, the Coconut Rubber Tea production of tea/rubber tends to be concentrat- Other agricultural crops ed in certain provinces. Table 1 shows by district Source: Central Bank of Sri Lanka, Economic and Social Statistics (various years). the shares of farm households that produced paddy, vegetables, tea/rubber, coconut, and coffee/pepper/betel in 2016. Production patterns vary widely across the country. For instance, paddy is planted throughout Sri Lanka, possibly because of relatively high lev- els of self-consumption. The share of paddy-cultivating farmers was particularly high in districts such as Ampara in the Eastern Province (87 percent) and Polonnaruwa in the North Central Province (86 percent). Between 40 percent and 55 percent of farm households in the Kurunegala, Puttalam, and Gampaha dis- tricts (which form the so-called coconut triangle) engaged in coconut production, a much higher share 5. Prior to 2016, the HIES response option for tea and rubber was lumped together in one category as “tea/rubber.” It is thus dif- ficult to ascertain the trends for tea and rubber separately. The extent to which these trends can be generalized to the overall sub-sector depends on the sector’s farm size composition: that is, if the sub-sector is dominated by smaller farms, the trends captured in household surveys are more likely to reflect changes in the whole sector. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 2. Agricultural Production of Farm Households, 2006 – 16 16 than elsewhere in the country. In comparison, farm households in the Sabaragamuwa, Southern, and Western Provinces were more likely to be engaged in tea/rubber production, with the highest shares, at about 70 percent, recorded in Kalutara and Ratnapura districts. TABLE 1 Shares of farm households engaged in selected crop activities by district Coffee, Province District Paddy Vegetables Tea/rubber Coconut pepper, betel 0.30 0.14 0.17 0.23 0.10 Western Colombo (0.05) (0.04) (0.04) (0.05) (0.03) 0.23 0.04 0.03 0.55 0.07 Western Gampaha (0.03) (0.01) (0.01) (0.03) (0.02) 0.28 0.03 0.71 0.10 0.03 Western Kalutara (0.03) (0.01) (0.03) (0.02) (0.01) 0.27 0.18 0.17 0.15 0.40 Central Kandy (0.02) (0.02) (0.02) (0.02) (0.02) 0.63 0.14 0.01 0.13 0.28 Central Matale (0.03) (0.02) (0.01) (0.02) (0.03) 0.11 0.60 0.14 0.00 0.05 Central Nuwara Eliya (0.02) (0.03) (0.02) (0.00) (0.01) 0.13 0.03 0.64 0.22 0.12 Southern Galle (0.02) (0.01) (0.02) (0.02) (0.02) 0.26 0.02 0.62 0.18 0.17 Southern Matara (0.02) (0.01) (0.02) (0.02) (0.02) 0.52 0.18 0.01 0.20 0.12 Southern Hambantota (0.03) (0.02) (0.00) (0.02) (0.02) 0.29 0.09 0.00 0.06 0.02 Northern Jaffna (0.04) (0.03) (0.00) (0.02) (0.01) 0.62 0.06 0.01 0.09 0.00 Northern Mannar (0.06) (0.03) (0.01) (0.03) (0.00) 0.67 0.18 0.01 0.06 0.00 Northern Vavuniya (0.04) (0.03) (0.01) (0.02) (0.00) 0.73 0.01 0.01 0.03 0.00 Northern Mullaitivu (0.04) (0.01) (0.01) (0.01) (0.00) 0.51 0.07 0.00 0.08 0.06 Northern Kilinochchi (0.07) (0.04) (0.00) (0.04) (0.03) 0.43 0.12 0.00 0.01 0.00 Eastern Batticaloa (0.05) (0.03) (0.00) (0.01) (0.00) 0.87 0.03 0.01 0.02 0.02 Eastern Ampara (0.03) (0.01) (0.01) (0.01) (0.01) 0.70 0.08 0.01 0.02 0.01 Eastern Trincomalee (0.04) (0.02) (0.01) (0.01) (0.01) 0.67 0.04 0.02 0.42 0.09 North-Western Kurunegala (0.02) (0.01) (0.01) (0.02) (0.01) 0.32 0.10 0.00 0.52 0.00 North-Western Puttalam (0.03) (0.02) (0.00) (0.03) (0.00) 0.77 0.15 0.00 0.14 0.00 North Central Anuradhapura (0.02) (0.02) (0.00) (0.02) (0.00) AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 2. Agricultural Production of Farm Households, 2006 – 16 17 Coffee, Province District Paddy Vegetables Tea/rubber Coconut pepper, betel 0.86 0.04 0.00 0.09 0.00 North Central Polonnaruwa (0.02) (0.01) (0.00) (0.02) (0.00) 0.45 0.34 0.22 0.02 0.13 Uva Badulla (0.03) (0.02) (0.02) (0.01) (0.02) 0.61 0.07 0.04 0.11 0.25 Uva Monaragala (0.03) (0.01) (0.01) (0.02) (0.02) 0.17 0.05 0.70 0.10 0.12 Sabaragamuwa Ratnapura (0.02) (0.01) (0.02) (0.01) (0.02) 0.31 0.03 0.51 0.21 0.06 Sabaragamuwa Kegalle (0.03) (0.01) (0.03) (0.03) (0.01) 0.44 0.12 0.21 0.19 0.11 Total (0.01) (0.00) (0.01) (0.01) (0.00) Source: World Bank staff calculation using HIES. Notes: Standard errors are in parentheses. The level of productivity varies widely depending on the crop. Figure 3 shows kernel density estimates of the log of gross output per acre by crop, our measure of productivity. While paddy farmers devote a rel- atively large land area to the crop (1.9 acres on aver- FIGURE 3 Kernel density estimates of the log of gross age), both their productivity and earnings are the output per acre, by crop lowest among all crops. Export-oriented crops such 1.0 as tea appear to be significantly more productive. 0.8 The productivity of paddy production is not only 0.6 the lowest among food crops, it has also improved Density little in the last decade. A vast amount of land area 0.4 is devoted to paddy cultivation, but productivity and earnings are low. Figure 4 shows the average 0.2 productivity of different crops, measured as yield per acre (in annual terms). 6 Among the different 0 6 7 8 9 10 11 12 13 14 crops, paddy stands out for its low earnings per Log of gross output per acre acre of land as well as the low levels of productiv- Paddy Chilies ity growth experienced between 2006 and 2016. In Onions Yams fact, productivity remained almost stagnant during Vegetables Tea Rubber Coconut these years, in contrast to most other crops, which Banana / Fruits Coffee, Pepper, Betel experienced an annualized productivity growth Source: World Bank staff calculation using HIES 2016. rate of around 4 percent or more. 6. Productivity is measured as the real annual gross output value, divided by the total cultivated area. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 2. Agricultural Production of Farm Households, 2006 – 16 18 FIGURE 4 Crop productivity (Gross value of output/acre, Rs), 2006-2016 60 16 14 50 12 40 10 Rs. (Thousands) Growth rate (%) 8 30 6 20 4 2 10 0 0 -2 Paddy Chilies Onions Vegetables Cereals Yams Tobacco Other Tea and Coconut Coffee Banana annual rubber 2006 2009 2012 2016 Annualized growth (R) Source: World Bank staff calculation using HIES. Note: HIES combines tea and rubber as one category for 2006, 2009 and 2012. In 2016 tea and rubber are included as separate categories but are aggregated for this table. Gross value of output is temporally and spatially adjusted using the Colombo Consumer Price Index (CCPI). 1 percent and 99 percent tails are trimmed. While the majority of cultivation is marketed, paddy farmers keep a large share of output for their own consumption. The share of own-consumption among farm households has fallen across nearly all food crops over time, likely as a result of increased market access and commercialization. However, paddy is an exception to this trend, as paddy farmers still use more than 40 percent of paddy output to meet their own consumption needs. Somewhat surprisingly, the figure does not vary much across farm- ers in different income quintiles. The share of self-consumption of vegetables has decreased significantly, which likely implies a diversification strategy toward market crops (table 2). TABLE 2 Share of crop output kept for farm household’s own consumption (percent) Year Paddy Chilies Onions Vegetables Cereals Yams Tobacco Other annual 2006 49.6 25.5 4.8 19.6 20.1 16.9 0.3 10.0 2009 50.6 20.1 3.2 14.2 8.1 28.6 0.0 21.0 2012 40.7 6.9 2.2 6.8 1.5 3.3 0.0 5.0 2016 41.8 4.9 0.2 7.4 3.4 5.8 0.0 1.8 Q1 41.49 1.95 0.18 5.77 4.26 9.23 0.00 1.56 Q2 41.86 11.60 0.48 5.26 3.18 20.93 0.00 0.19 Q3 42.66 1.34 0.13 11.63 4.30 2.35 0.36 0.78 Q4 41.67 16.61 0.02 6.97 3.22 3.21 0.00 5.69 Q5 41.33 1.52 0.18 8.42 1.05 1.58 0.00 0.58 Source: Staff calculation using HIES. Note: Some large fluctuations in values are due to relatively small number of observations. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 2. Agricultural Production of Farm Households, 2006 – 16 19 Poor farmers rely much more on paddy cultivation for their income than better-off farmers. Table 3 below shows the share of paddy output out of total gross income (left) and out of total farm output value (right). Consistent with the low productivity of paddy, the poorest farmers — in the bottom 20 percent of the income distribution — are most reliant on paddy farming; over 40 percent of their farm output value comes from paddy, while the contribution of higher-value crops is low. In contrast, farmers in the top 20 percent receive only about 26 percent of their farm output value from paddy. The relative impor- tance of paddy income has not changed much over time. TABLE 3 Paddy output value as share of gross income and farm output value, by household income quintile (percent) Paddy output value as Paddy output value as share of gross income (percent) share of farm output value (percent) Quintile 2006 2009 2012 2016 2006 2009 2012 2016 Bottom 20% 9.4 9.5 9.0 8.4 39.7 44.4 43.7 41.3 Second quintile 4.0 5.5 3.8 3.5 32.5 36.8 35.2 32.2 Third quintile 2.9 4.1 2.9 2.9 30.6 34.2 32.5 31.2 Fourth quintile 2.3 3.5 2.3 2.2 27.4 34.0 29.9 29.4 Top 20% 1.4 2.8 1.8 1.4 19.0 26.0 23.2 25.7 Total 3.9 5.0 3.9 3.6 29.6 34.6 32.7 31.8 Source: Staff calculation using HIES. Note: Income quintiles are estimated based on per capita total household income. Most farmers who benefit from fertilizer subsidies apply them in paddy cultivation, and larger sub- sidies are associated with higher productivity. A larger share of male farmers receives crop subsidies than female farmers (38.9 percent versus 18 percent) 7, though this gap likely reflects the greater tendency of paddy farmers to be male. Most of the subsidy (90 percent) is accounted for by paddy, followed by tea (4 percent) and vegetables (2 percent). Regression results in section 3 show that the incidence of sub- sidy is negatively associated with productivity as well as earnings — i.e., poor farmers are more likely to be recipients of subsidies. However, among those that receive subsidies, larger subsidies are associated with higher productivity, which appears to support the long-held belief that fertilizer subsidies lead to increased land productivity, the goal of the subsidy support scheme (Weerahewa, Kodithuwakku, and Ariyawardana 2010). Fertilizer subsidies come at a fiscal and opportunity cost, and may need careful consideration in the broader context of spending to support the agriculture sector and rural development. Fertilizer sub- sidies constitute a major expenditure in the government’s agricultural budget and could result in a sub- optimal composition of agricultural spending when interventions to incentivize farmers to adopt cli- mate-smart technologies, help them access value chains or invest in better agro-logistics could have 7. Most of the subsidies consist of fertilizers. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 2. Agricultural Production of Farm Households, 2006 – 16 20 greater payoffs. More generally, there have been concerns that they might take away resources from oth- er spending on roads, health, and education, which could contribute to rural development more broadly. Fertilizer subsidies have been primarily used to meet production and therefore food security objectives, and were deemed less effective than direct income transfers from the perspective of providing support to low-income paddy farmers. Furthermore, these subsidies appear to have impacted market decisions by encouraging the cultivation of paddy and disincentivizing movements to other types of agriculture that have more potential for value addition (World Bank, 2013). 3.  Determinants of Higher Agricultural Productivity and Earnings AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 3. Determinants of Higher Agricultural Productivity 22 Diversification to Improve Productivity The analysis in this section focuses on key determinants of productivity and earnings. Since detailed input data to measure total factor productivity are not available, we use land productivity as a proxy for productivity, defined as gross income divided by cultivated area. A farmer’s real monthly income from crop production is proxied by the sum of the gross output value for each crop. 8 For the purpose of our analysis, “farmers” are defined as those who reported a positive gross crop output value in the HIES. By this definition, there were 1.7 million farmers in 2016 and about a quarter of them were female. 9 Detailed summary statistics are reported in appendix table A.2. On average, farmers were 51.7 years old, had 8.5 years of schooling, cultivated 1.7 acres of land and grew 1.3 types of crops. Their average crop income and productivity in 2016 were estimated respectively at Rs. 15,487 and Rs. 16,905 per acre. Diversification is generally considered a key channel to increase productivity and incomes. Joshi et al (2004) describe diversification as threefold, entailing “(i) a shift of resources from farm to nonfarm activities, (ii) use of resources in a larger mix of diverse and complementary activities within agricul- ture, and (iii) a movement of resources from low value agriculture to high value agriculture.” The first channel could be considered diversification at the external margin. Our findings below suggest that Sri Lanka farmers may be using nonfarm incomes to move out of agriculture. We focus on diversification within agriculture and allocation of resources between low-value and high-value activities and their roles for productivity and incomes. To operationalize these concepts, we construct an index to capture each. Following common practice in the literature, we measure a farm- er’s level of crop diversification using the Simpson Index of Diversification (SID): SIDi = – �k Pik , where SIDi is the Simpson Index of Diversification for farmer i, and Pik is the proportionate cultivated area dedicated to the kth crop. 10 A lower index would indicate high concentration (or low diversifica- tion). This measure considers crop farming only, excluding earnings from livestock activities. There are many reasons why farm households might benefit from increased diversification. These include strate- gies to reduce risks or to realize complementarities between different activities. 8. Beginning in this section, the analysis moves to the individual farmer level except in table 4. 9. Following the definition used, there were an estimated 1.7 million farmers in 2016, of which 1.3 million were male and 0.4 million female. These numbers should not be interpreted as representing the gender breakdown of the workforce in the agricultural sector. Because the definition considers only those with a positive output as farmers, most unpaid family workers are not included, and it is well known that in Sri Lanka this group includes a large number of female workers. Another caveat is that the sample includes those individuals whose farm activities are secondary. In the 2016 HIES, 38.5 percent of farmers had some income from nonfarm activities such as nonfarm self-employment or wage work, while 47.0 percent of them had some non-labor income. 10. Calculating the index with the share of output value leads to similar results. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 3. Determinants of Higher Agricultural Productivity 23 Further, an export-orientation index is constructed to capture the relative weight of domestic-oriented and export-oriented crops in a farmer’s portfolio. Specifically, the index is calculated as follows: XINDi = �k ShX k * ShX ik , where XINDi is the Export Orientation Index for farmer i, and ShX k is the export intensity of crop k — i.e., the share of exports of the overall production quantity at the national level taken from FAOSTAT (2014 – 17 average). Export-related information is not available at the farm level. Paddy and most other food crops are geared toward serving the domestic market, whereas higher-value crops tend to have a higher share of exports. ShSik is the share of cultivated area for agricultural item k in individual i’s total cultivated area and is intended to compute a weight for the overall export intensity of the production mix.  1 1 XINDi intends to capture the extent to which the crop mix consists of export-oriented crops, which tend to be of high-value, e.g., tea, coffee, pepper, betel, and rubber. There is a positive relationship between productivity and the export-orientation of the crop mix (figure 5). It should be noted that this index captures export shares at the aggregate national level and there is likely sig- nificant heterogeneity across farmers. Unfortunately, the household survey data do not provide further infor- mation on the export share of farmers’ agricultural output. The relationship between productivity and the level of diversification in the portfolio appears to be less linear. Productivity increases with the level of diversi- fication, measured with the SID, but peaks in the middle quintile of the productivity distribution. In compar- ison, productivity increases monotonically with the weighted export intensity of the crop portfolio (figure 6). FIGURE 5 Correlation between the Export Orientation FIGURE 6 Simpson Index of Diversification and Export Index and productivity Orientation Index, by productivity quintile 13 0.5 12 0.4 11 Log of Productivity 0.3 10 Indexes 9 0.2 8 0.1 7 0 6 1 2 3 4 5 0 0.2 0.4 0.6 0.8 1.0 Productivity quintile Export Orientation Index XIND SID Source: World Bank staff calculation using HIES 2016. Source: World Bank staff calculation using HIES 2016. Note: XIND = Export Orientation Index; SID = Simpson Index of Diversification. The x axis displays the productivity quintile of the farmers (1=lowest 20%, 5=highest 20%).. 11. ShX k, or crop-level export shares, are computed as: paddy (0.0039), chilies (0.012), onions (0.0), vegetables (0.023), cereals (0.044), yams (0.0), tobacco (1.0), tea (0.87), rubber (0.18), coconuts (0.09), coffee, pepper, betel etc (0.399), banana/fruits (0.052) and horticulture (0.037) (average of vegetables and banana/fruits). AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 3. Determinants of Higher Agricultural Productivity 24 Crop diversification has also been adopted in Sri Lanka to increase the resilience of agricultural pro- duction. A case study in the dry zone of Sri Lanka found that although farmers in the area have been cultivating paddy for many decades, its cultivation has increasingly suffered pressure from high costs of production and the decrease in paddy market prices. To cope with such constraints, farmers have increas- ingly relied on crop diversification, changing cropping patterns and methods for cultivating, especial- ly during seasons with a deficit of water supply (Dharmasiri, 2008) and likely between seasons as well. The nature of agricultural livelihoods differs significantly across the country. The variation could be the result of a combination of factors, including differences in agroclimatic conditions and natural en- dowments — for example, smallholder tea farmers are concentrated in the Southern and Sabaragamuwa Provinces; vegetable production mainly takes place in the Central and Uva Provinces; many small coco- nut producers are located in the North-Western and Western Provinces; and finally, the North-Central Province accounts for the largest share of smallholder paddy production. Wide regional variation exists in the export-orientation and diversification of crops and productivity levels. Figure 7 shows the district-level variation in the average Simpson Index of Diversification, Export Orientation Index, productivity (Rs/acre), share of paddy farmers in total crop farmers in the district and the district’s share in total paddy production. Some clear patterns emerge: first, there are wide differ- ences in productivity across districts; and second, productivity is significantly higher in districts where export-oriented crops are cultivated and lower in districts where a large share of farmers cultivate pad- dy. However, while a high share of farmers in the NE produce at least some paddy, the majority of paddy output (as measured by output value) is accounted for by the districts of Anuradhapura (North-Central Province), Polonnaruwa (North-Central Province), Kurunegala (North-Western Province), Ampara (Eastern Province), and Hambantota (Southern Province). The spatial production patterns reflect the farmers’ choice as well as the local agro-ecological and climatic conditions. Meanwhile, farm households in the Northern and Eastern Provinces (NE) rely more on livestock activities and exhibit lower average crop productivity, measured by output value per acre of land. Table 4 compares household characteristics in the NE to others, disaggregated by the gender of the household head. Households in the NE rely more on livestock activities (table 4; see also annex Table A.1.) — for instance, 40 percent of farm households in Batticaloa were engaged in livestock-only activi- ties in 2016. The civil war has exerted lasting impacts on livelihoods in the NE. Post-conflict livelihood support, especially for female-headed households, tended to be focused on small livestock activities (Silva et al. 2018), which may explain the high share of chicken and goat ownership among this group. Crop productivity among NE farm households is lower than in the rest of the country, which could be due to several factors, including (i) larger agricultural land area, as the average land size owned by NE households is 1.7 acres, compared to 1.0 acre for non-NE households; there is a commonly found inverse relationship between land size and productivity (as further described below); and (ii) the crop AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 3. Determinants of Higher Agricultural Productivity 25 FIGURE 7 District level variation in diversification, export orientation, productivity, share of paddy farmers and share of paddy production in national production Simpson Index of Diversification (SID) Export Orientation Index (XIND) Productivity (Rs./acre) 0.16 – 0.20 0.4 – 0.61 20,000 – 35,000 0.12 – 0.16 0.3 – 0.4 15,000 – 20,000 0.08 – 0.12 0.2 – 0.3 10,000 – 15,000 0.04 – 0.08 0.1 – 0.2 5,000 – 10,000 0 – 0.04 0 – 0.1 0 – 5,000 Share of Paddy Farmers in Total Farmers (%) Share of Paddy Production (%) 80 – 92 12 – 18 60 – 80 9 – 12 40 – 60 6–9 20 – 40 3–6 0 – 20 0–3 Source: World Bank staff estimation using HIES 2016. choice, with NE farm households devoting a large area of land — a round 80 percent on average in most districts — to the cultivation of paddy, which is a low-productivity crop. It should be emphasized that available data on livestock activities are either limited or difficult to measure — for example, land pro- ductivity is only measured against crop production. Hence, the focus on the crop sector may not accu- rately reflect the overall productivity and incomes of farmers in the NE where livestock activities are much more widespread. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 3. Determinants of Higher Agricultural Productivity 26 TABLE 4 Characteristics of farm households in Northern and Eastern Provinces (NE) vs. other provinces and by gender of household head t-test to show if HHs Farm HHs in other in NE are different All Farm HHs Farm HHs in NE provinces from others headed headed headed headed headed headed headed headed Female Female Female Female Total Total Total Total Male Male Male Male Gender of HH head 16.8 0.097 0.176 0.000 (=1 if female) Education of HH head (years) 8.1 8.1 7.6 7.5 7.5 7.5 8.1 8.2 7.6 0.000 0.000 0.161 Monthly income (Rs) Total income 87,153 90,886 68,663 70,296 71,812 56,224 88,966 93,132 69,404 0.006 0.008 0.077 Per capita income 17,234 17,4 88 15,982 14,067 14,160 13,211 17,576 17,882 16,14 8 0.000 0.000 0.008 Crop productivity 16,376 16,332 16,600 9,911 9,782 11,301 16,981 17,009 16,844 0.000 0.000 0.003 at HH levela (Rs/acre) Land owned 1.1 1.1 0.8 1.7 1.8 1.4 1.0 1.1 0.8 0.000 0.000 0.008 by HH (acres) Paddy land owned 0.6 0.6 0.5 1.4 1.4 1.0 0.5 0.5 0.4 0.000 0.000 0.000 by HH (acres) Livestock ownership % HHs that own a cow 9.1 9.7 5.6 20.7 21.1 16.8 7.8 8.4 5 0.000 0.000 0.000 % HHs that own a goat 1.9 2.1 1.3 10.3 10.1 11.6 1 1.1 0.7 0.000 0.000 0.000 % HHs that own a chicken 10.2 10.2 10.3 32.5 31 46.5 7.8 7.8 8.2 0.000 0.000 0.000 Source: World Bank staff calculation using HIES 2016. Note: NE = Northern and Eastern Provinces; HH = household. a. Productivity is calculated as gross crop output divided by cultivated areas at the household level. There is a strong and positive association between productivity and the level of diversification, and between productivity and the export orientation of the crop mix; the association is confirmed after controlling for other factors that could determine productivity. Table 5 reports regression results relat- ing the log of productivity (regression (1)) or the log of earnings from crops (regression (2)) to a series of personal, household, location, and agricultural characteristics. The main results are presented in regres- sion (1) and (2), and suggest that higher export orientation of farmers’ production mix, captured by the coefficient for the Export Orientation Index,  1 2 and diversification of crop mix, reflected in the coeffi- cient for the Simpson Diversification Index, are positively associated with higher productivity and earn- ings (with statistical significance at the 1 percent level). 12. Alternatively, we include a series of the share of cultivated area for each crop in total cultivated area, with the share of paddy as the omitted category. The coefficients for the shares of all the other crops except “other cereals” turn out to be pos- itively significant, suggesting higher productivity of cash crops relative to staples. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 3. Determinants of Higher Agricultural Productivity 27 TABLE 5 Determinants of crop productivity and earnings (1) (2) (3) (4) (5) (6) (7) Log of pro- Log of crop Log of pro- Log of pro- Log of pro- Log of pro- Log of pro- Dependent variable ductivity income ductivity ductivity ductivity ductivity ductivity -0.072 *** -0.084 *** 0.254 *** 0.129 *** -0.065 * 0.013 -0.155 *** Gender (=1 if female) (0.025) (0.025) (0.036) (0.036) (0.035) (0.034) (0.033) 0.009 ** 0.008* 0.013 * 0.015 ** 0.014 ** 0.016 ** Age (0.005) (0.005) (0.007) (0.007) (0.006) (0.006) -0.000 ** -0.000 ** -0.000 ** -0.000 *** -0.000 *** -0.000 *** Age squared (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) 0.006 ** 0.007 ** 0.013 *** 0.026 *** 0.012 *** 0.024 *** Years of education (0.003) (0.003) (0.004) (0.004) (0.004) (0.003) 0.068 ** 0.071 ** 0.052 0.152 *** 0.052 0.143 *** Household head (0.027) (0.028) (0.041) (0.038) (0.038) (0.035) Share of nonfarm labor income in -0.509 *** -0.537 *** -0.607 *** -0.934 *** -0.500 *** -0.808 *** total income (%) (0.031) (0.032) (0.040) (0.041) (0.037) (0.039) Share of nonlabor income in total -0.579 *** -0.588 *** -0.723 *** -1.043 *** -0.544 *** -0.853 *** income (%) (0.041) (0.042) (0.054) (0.051) (0.052) (0.049) -0.036 -0.044 -0.137 ** -0.181 *** -0.065 -0.111 * Incidence of livestock activity (0.045) (0.045) (0.068) (0.064) (0.062) (0.058) 0.003 0.002 0.009 0.018 ** 0.012 0.020 ** Household size (0.006) (0.006) (0.009) (0.008) (0.009) (0.008) -0.004 -0.010 0.056 0.016 0.030 -0.005 Child dependency ratio (0.032) (0.033) (0.049) (0.044) (0.045) (0.041) -0.720 *** 0.202 *** -0.351 *** -0.322 *** Log of cultivated land (acres) (0.014) (0.015) (0.014) (0.014) 0.904 *** 0.867 *** 1.413 *** 1.292 *** Export Orientation Index (0.040) (0.042) (0.055) (0.051) 0.482 *** 0.519 *** Log of agricultural inputs (0.012) (0.012) 0.591 *** 0.630 *** Simpson Index of Diversification (0.048) (0.050) Number of unpaid agricultural family 0.046 ** 0.045 ** labor (0.018) (0.020) -0.718 *** -0.835 *** Incidence of crop subsidy (0.095) (0.097) 0.061 *** 0.077 *** Log of crop subsidy (0.011) (0.011) 0.088 *** 0.071 ** Tractor (0.025) (0.029) 0.127 *** 0.119 *** Access to finance (0.029) (0.027) 0.004 0.001 Access to IT (0.027) (0.027) Log of distance to nearest agricultur- 0.009 0.006 al service center (0.008) (0.008) AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 3. Determinants of Higher Agricultural Productivity 28 (1) (2) (3) (4) (5) (6) (7) Log of pro- Log of crop Log of pro- Log of pro- Log of pro- Log of pro- Log of pro- Dependent variable ductivity income ductivity ductivity ductivity ductivity ductivity -0.032 -0.028 Incidence of natural calamity (0.024) (0.025) Number of observations 5,613 5,683 5,613 5,613 5,613 5,613 5,613 R2 0.669 0.759 0.011 0.250 0.353 0.350 0.437 Source: World Bank staff estimation using HIES 2016. Note: The dependent variable for all regressions except (2) is the log of productivity. Productivity is measured by yield, i.e., gross real monthly output divided by the area cultivated (SL Rs/acre). The dependent variable for regression (2) is the log of earnings (SL Rs). Gross real monthly output is used as a proxy for crop earnings. The regressions are estimated using ordinary least squares (OLS). Except in regression (3), ethnicity dummies, sector dummies (urban, rural, estate), and district dummies are included as control variables. Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent. Robust standard errors are in parentheses. a. Agricultural inputs include seeds, fertilizer, chemicals, hired labor, transport cost, agricultural equipment/rental, and others. Cultivated area and productivity are negatively correlated, consistent with findings from other coun- tries in the literature. The coefficient for the log of land area in regression (1) is found to be negatively significant at the 1 percent level, confirming an inverse relationship between cultivated area and agri- cultural productivity that has been commonly found in the existing literature — e.g., Aguilar et al (2015) in Ethiopia; Ali et al (2016) and Carletto, Savastano, and Zezza (2013) in Uganda; Kilic, Palacios-Lopez, and Goldstein (2015) in Malawi; Oseni et al (2015) in Nigeria; and Slavchevska (2015) in Tanzania. In con- trast, the relationship between cultivated area and earnings is positive, suggesting that access to more land improves farmers’ earnings. This relationship can be seen from the positive and statistically sig- nificant coefficient for cultivated area in the earnings regression. Results also suggest that farmers tend to utilize their nonfarm income, including remittances, to move out of agriculture rather than to invest in agriculture. The coefficients for the share of nonfarm labor income and those for the share of nonlabor income are negatively significant. This suggests that the availability of other income sources is associated with lower productivity and lower crop income. This result is consistent with previous research in other countries, such as Kilic et al (2009) in Albania and Rozelle, Taylor, and DeBrauw (1999) in China. Other correlates of productivity such as agricultural inputs, family labor, mechanization, and access to finance turn out to be positively associated with productivity and income. The coefficients for the agricultural inputs, unpaid agricultural family labor,  1 3 the mechanization (proxied by the ownership of tractor(s)), and access to finance (proxied by access to a loan) are positively significant at least at the 5 percent level for both productivity and earning regressions. The coefficient for the proxy of access to 13. Agricultural unpaid “family labor” is defined as those household members whose main activity is in agriculture; whose employment status is either “own-account worker,” “contributing family worker,” or “employer”; and who report no farm income. Under this definition, 70 percent in this category were female; 58 percent in this category (2 percent of the males and 81 percent of the females) were spouses of household heads. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 3. Determinants of Higher Agricultural Productivity 29 “digital agriculture” 14 (Fabregas, Kremer and Schilbach, 2019) was not statistically significant. It is possi- ble that with nearly 90 percent of information technology (IT) ownership at the household level, varia- tions in access are not well proxied with this information. While the incidence of subsidy is associated with lower productivity and income, conditional upon receipt, subsidy appears to enhance productivity and income. More years of schooling and hence better skills also lead to higher productivity and income. Gender Gaps in Productivity and Earnings Delving further into the data on gender differences in productivity and earnings suggests that female farmers are on average more productive. The empirical measurement of gender gaps in agricultural productivity has been found to be challenging in the literature.  1 5 But the Sri Lanka HIES has a distinct advantage in providing farm activity information at the individual level, whereas many previous stud- ies have had to rely on certain assumptions. Interestingly, in Sri Lanka the productivity among female farmers (Rs 19,809 per acre) is found to be higher than that of male farmers (Rs 15,976 per acre). This is the difference in raw, unconditional pro- ductivity, not controlling for crop type or any other differences in inputs. Female productivity advantage is again confirmed through regression (3) in table 5: including only the gender dummy in the regression, the coefficient for the gender variable is positive and statistically significant. This is in stark contrast to other countries in the literature, which found male farmers to be more productive but which explained away the advantage by pointing to differential access to inputs. It was subsequently argued that gender gaps could be overcome if female farmers had the same level of access to inputs (see for example, Croppenstedt, Goldstein, and Rosas 2013; FAO 2011; Adeleke et al 2008; Udry 1996). Using high-quality plot-level data, some more recent papers find a persistent productivity gender gap in favor of males after controlling for inputs and other characteristics (e.g., Ali et al, 2016; Kilic et al. 2014; Oseni et al. 2015; Slavchevska 2015). The productivity advantage of female farmers appears surprising at first, given that they have less access to inputs and a less diversified crop mix. Appendix table A.2 shows that female farmers make less use of agricultural inputs (Rs 2,923/month) than male farmers (Rs 5,449/month). The share of female farmers who receive fertilizer subsidies (18.0 percent) is smaller than the share of male farmers (38.9 per- cent); and the amount received (Rs 658/month) is also smaller than for male farmers (Rs 935/month). Further disadvantages include lower access to finance (6.8 percent for female farmers versus 9.8 percent for male farmers); less access to unpaid agricultural family labor and a lower degree of mechanization 14. As proxy for access to information technology, a dummy variable is added to indicate whether any household member had an expenditure on a mobile phone, e-mail/internet, or computer. 15. See box 3 for a review of the literature. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 3. Determinants of Higher Agricultural Productivity 30 (proxied by tractor ownership). On average, male farmers cultivate a slightly higher number of crops which is also reflected in the corresponding Simpson Index of Diversification, at 11.5 for male farmers and 0.6 for female farmers. Female farmers also have fewer opportunities for nonfarm activities rela- tive to male farmers: while 44.9 percent of male farmers were engaged in nonfarm labor activities and drew 29.7 percent of income from these sources, only 19.3 percent of female farmers had access to non- farm labor opportunities, and these comprised only 10.3 percent of their income. While less diversified, the average crop mix of female farmers has a higher degree of export orienta- tion than that of male farmers. Figure 8 shows the composition of crop mix by farmer’s gender, as meas- ured by the average share of cultivated area for each crop in total cultivated area. The share of paddy for male farmers accounted for about 45 percent, which is more than twice that of female farmers (20 per- cent). In contrast, the crop mix of female farmers tends to skew toward export-oriented crops: for exam- ple, the share of cultivated area dedicated to tea was 31 percent for female farmers versus 14 percent for male farmers. As a result, the Export Orientation Index among female farmers was estimated at 33.9, nearly twice as large as the estimated index value of 17.7 among male farmers. FIGURE 8 Crop mix by farmer’s gender Bana Co Ban ffe Co na / ana e, P Chilies Chilies Others ffe Others ep e, Fruit / Fru Pe pe Ho pp r, B rti y s its er, dd cu ete 0.7% 0.5% 3.5% ltu 5.2% Pa Be 2.9% Ho l re 2.5% tel rtic 6.1 ultu .4% 0.3 7.3 re % % 20 % Coco 1.9 % nut s 12.2 % Onion addy 0.2% 45.4% P Coconut 16.1% 8.9% Vegetables Rubber 1.4% 3.9 0.7 % O .3% % % ther 14 1.8 Yam cerea Tea bb er s ls Ru .9% 30.6% .2% 0.6% Vegetables 8.4% r cer ms 0 3 eals Onion Ya Tea s Othe Source: World Bank staff estimations using HIES 2016. Notably, the unconditional productivity advantage enjoyed by female farmers relative to male farm- ers — despite numerous disadvantages described above — does not translate into higher incomes. The gap observed between male and female farmers in Sri Lanka reverses when it comes to crop earnings, with male farmers earning on average Rs 16,970 per month, significantly higher than the earnings of AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 3. Determinants of Higher Agricultural Productivity 31 female farmers at Rs 11,086. The male earnings advantage persists after adjusting for individual, house- hold, location, and agricultural production-related characteristics. This is seen from the coefficients on the gender dummy in regression (2) in table 5, which are negatively significant at the 1 percent level. The unconditional female productivity advantage appears to be explained by two factors: (i) lower access to land by female farmers, which results in higher productivity because of the inherent inverse relationship between land area and productivity; and (ii) the selection of a more profitable crop mix by female farmers. The productivity gap is further analyzed by conditioning it on different factors and attempting to identify what contributes to the female productivity advantage. Accounting for a num- ber of personal, household, and location characteristics, the magnitude of the coefficient for the gender dummy was reduced but remained positively significant (regression (4)). However, once land area is con- trolled for, the sign of the coefficient for the gender dummy is reversed and becomes significant at the 1 percent level, suggesting the inverse relationship between land area and productivity is an important factor behind the female productivity advantage (regression (5)). In fact, female farmers have access to about one acre of land on average, which is only about half that of male farmers. Moreover, female farmers tend to select a crop mix that consists of a higher share of higher-value crops, which increases their productivity relative to male farmers. Two sets of results support this observa- tion. When accounting for the export-orientation of the production mix while not controlling for land area, the gender difference in productivity becomes statistically insignificant (regression (6)). In addition, as shown in the last four columns of table 6, which compare the log of productivity between male and TABLE 6 Key summary statistics by crop and gender Gross crop income (Rs) Cultivated area (acre) Log of productivitya t-test Total Male Female Total Male Female Total Male Female (p-value) Paddy 9,951 10,435 6,513 1.9 1.99 1.31 8.40 8.41 8.34 0.010 *** Chilies 9,616 7,995 18,252 0.58 0.61 0.42 9.13 9.09 9.38 0.497 Onions 18,254 18,204 18,777 0.59 0.6 0.57 10.16 10.16 10.16 0.805 Yams 19,287 20,972 10,771 0.82 0.88 0.54 9.77 9.81 9.61 0.762 Vegetables 10,961 12,724 4,935 0.65 0.73 0.37 9.45 9.51 9.23 0.020 ** Tea 20,190 23,825 14,507 0.82 0.98 0.58 9.90 9.93 9.85 0.019 ** Rubber 20,702 20,757 20,539 1.36 1.37 1.36 9.43 9.47 9.28 0.119 Coconut 13,747 12,829 16,324 1.48 1.46 1.52 9.01 9.01 9.00 0.832 Coffee, pepper, betel 18,814 21,501 11,236 0.73 0.8 0.52 9.81 9.82 9.77 0.856 Banana/fruits 18,230 19,177 14,751 0.98 1.09 0.5 9.42 9.47 9.21 0.194 Totala 15,487 16,970 11,086 1.7238 1.9452 1.0438 9.14 9.09 9.30 0.000 Source: Staff calculation using HIES 2016. Note: Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent. a. Productivity is measured by yield, i.e., gross real monthly output divided by the area cultivated (Rs/acre). AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 3. Determinants of Higher Agricultural Productivity 32 female farmers by crop, there is no female productivity advantage within the same crop. In fact, there appears to be a male productivity advantage for paddy, tea and vegetables, as seen from the t-test results in the last column, which indicate whether differences are statistically significant. This result suggests that productivity differences between crops rather than within crops are the driving force behind the higher unconditional productivity among female farmers. Once both the log of cultivated area and the Export Orientation Index are controlled for, the coefficient for the gender dummy becomes negatively significant, revealing a conditional male productivity advantage (regression (7) in table 5). Since regres- sion (7) does not adjust for the variables concerning agricultural activities such as agricultural inputs, the magnitude of gender gap in favor of men is about 10 percentage points larger than that in our base regression (regression (1)) which controls for other production-related variables. A decomposition analysis further helps identify the relative weight or importance of different factors that drive the productivity and earnings differentials. The decomposition is based on the same regres- sions but allows us to estimate the size of an endowment and structural effect for each variable, which can be used to assign a relative weight to the different characteristics. The original methodology was conceived to understand gender wage gaps and to break down the gap into an endowment effect (i.e., the difference in the gap that is due to differences in characteristics themselves) and a structural effect (i.e., the difference in the gap that is due to differential returns to characteristics, such as returns to education). Methodological details of the decomposition are briefly summarized in box 2. Previous studies have applied this decom- position method to analyze gender gaps in agriculture; see box 3 for a detailed review of the literature. BOX 2 Oaxaca-Blinder decomposition Using an Oaxaca-Blinder decomposition, we decompose productivity and earnings gaps into (i) a component driven by gender dif- ferences in levels of observable attributes (endowment effect) and (ii) a component coming from gender differences in returns to the same set of observables (structural effect) (Ali et al [2016]; Jann [2008]). Specifically, YMK – YFK = �k= K X MK – X FK βK* (endowment effect) + �k= K X MK βMK – βK * + �k= K X FK – βK * + βM – βF (structural effect) where Y MK – YFK is the mean gender difference in log of productivity or crop income, and X MK and X FK are the average value of covariate K for men and women respectively. βM , βF , βMK and βFK are the returns to covariate K obtained from the ordinary least squares (OLS) regressions run separately by the gender of the farmer, and βK are the returns to covariate K estimated by the * pooled OLS model. The decomposition confirms that the productivity gap in favor of women is explained by a smaller cultivated area and a higher-value crop mix among female farmers. The female productivity advan- tage can be explained by the endowment effect in favor of women, which outweighs the structural AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 3. Determinants of Higher Agricultural Productivity 33 effect in favor of men. The results from a produc- FIGURE 9 Oaxaca-Blinder decomposition for tivity decomposition are illustrated in figure 9 and productivity gender gap table 7. The green and blue bars in the figure rep- 0.4 resent the size and direction of each characteristic 0.3 in explaining the gap. A negative (positive) bar in 0.2 figure 9 and negative (positive) sign in table 7 indi- 0.1 cate that the characteristic works in favor of female (male) farmers. For example, the “Total” bar on 0 Log difference the far right in figure 9 represents the gender gap -0.1 owing to either endowment or structural effects. -0.2 The negative endowment part is greater in value -0.3 than the positive structural part, indicating female -0.4 farmers are more productive. Specifically, panel B of table 7 reports that the female productivity advan- -0.5 tage of -25.5 percent is decomposed into an endow- -0.6 Fin Fla HHh Inp Lan Mec NFl NFn SID Sub XIN Total ment component of -32.5 percent and a structur- Structural Endowment Combined al component of 7.2 percent. The detailed decom- Source: World Bank staff estimations using HIES 2016. position in panel C by individual variables shows Note: Fin = access to finance; Inp = agricultural inputs; Fla = number of unpaid that the endowment effects are statistically signif- agricultural family labor; Lan = log of cultivated area; Mec = mechanization; HHh = household head; NFI = share of nonfarm labor income; NFN = share icant for all the variables while none of the struc- of nonlabor income; SID = Simpson Index of Diversification; Sub = subsidy (sum of the coefficients for the incidence of subsidy and the log of the value tural effects are statistically significant. of subsidy); XIN = Export Orientation Index. Access to land is the largest contributor to female endowment and ultimately to productivity advan- tage. The second largest contributor is the relatively large presence of export-oriented crops in the pro- duction mix among female farmers. This result contrasts with previous literature which found that the cultivation of cash crops contributes to male endowment advantage since male farmers are more likely to grow high-value cash crops; see for example Kilic, Palacios-Lopez, and Goldstein (2015) on Malawi; Ali et al (2016) on Uganda). Access to nonfarm labor income contributes to female productivity advan- tage. The latter could be because female farmers are less likely to have a nonfarm labor activity, and the availability of nonfarm labor income tends to be associated with lower agricultural productivity. On the other hand, factors such as greater access to agricultural inputs, crop diversification, and small- er receipt of nonlabor income contribute to male productivity advantage. This can be seen by compar- ing the specifications in regressions (7) and (1) in table 5, and the narrowing of the coefficient on the gen- der dummy once additional production-related factors are accounted for. Male farmers are able to utilize far more agricultural inputs such as fertilizer, seeds, chemicals, hired labor, and means of transportation, which contribute to a male endowment advantage over female farmers (29.1 percent). Greater access to fertilizers may in part be due to male farmers receiving more subsidies that are mostly linked to paddy. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 3. Determinants of Higher Agricultural Productivity 34 Female farmers are also less diversified, and their greater access to nonlabor income reduces the female productivity advantage because the availability of income from other sources tends to decrease crop pro- ductivity. Finally, access to finance is positively asso- ciated with both productivity and earnings (table FIGURE 10 Oaxaca-Blinder decomposition for earning 5) but does not seem to make an appreciable differ- gender gap ence between men and women farmers (figure 10), 0.5 possibly because commercial financing options are 0.4 limited for most farmers. This also appears to be the case for mechanization (proxied with tractor 0.3 ownership) and unpaid agricultural family labor. Log difference 0.2 Male farmers outearn female farmers significant- ly, mainly owing to greater use of inputs. Both 0.1 the endowment effect (34.8 percent) and to a less- 0 er extent the structural effect (7.8 percent) contrib- ute to male earnings advantage. Male farmers out- -0.1 earn female farmers by a significant share — 42.6 percent (figure 10, table 7). Greater access to agri- -0.2 Fin Fla HHh Inp Lan Mec NFl NFn SID Sub XIN Total cultural inputs among male farmers is the leading Structural Endowment Combined contributor, as suggested by the large endowment Source: Staff estimations using HIES 2016. effect of 29.1 percent. Contrary to the result from Note: Fin = access to finance; Inp = agricultural inputs; Fla = number of unpaid the productivity gap decomposition, having access agricultural family labor; Lan = log of cultivated area; Mec = mechanization; HHh = household head; NFI = share of nonfarm labor income; NFN = share to more land confers a large earnings advantage to of nonlabor income; SID = Simpson Index of Diversification; Sub = subsidy (sum of the coefficients for the incidence of subsidy and the log of the value male farmers (18.5 percent). of subsidy); XIN = Export Orientation Index. TABLE 7 Productivity and earnings gap decompositions   Decomposition of productivity gap Decomposition of earnings gap A. Mean gender differential (log)   Mean productivity of male farmers 9.034 *** Mean productivity of female farmers 9.288 *** Mean gender differential in productivity -0.254 *** Mean earnings of male farmers 9.190 *** Mean earnings of female farmers 8.764 *** Mean gender differential in earnings 0.426 *** Endowment Structural Endowment Structural B. Aggregate decomposition (%) Total -0.325 *** 0.072 *** 0.348 *** 0.078 *** AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 3. Determinants of Higher Agricultural Productivity 35 Endowment Structural Endowment Structural C. Detailed decomposition (%) Fin Access to finance 0.004 *** 0.005 0.003 *** 0.004 Inp Agricultural inputsa 0.291 *** -0.067 0.292 *** -0.027 Fla Number of unpaid agricultural family labor 0.008 ** -0.007 0.008 ** -0.007 Lan Log of cultivated area -0.487 *** -0.008 0.185 *** -0.011 Mec Mechanization 0.007 *** -0.005 0.006 *** 0.001 HHh Household head 0.022 ** 0.02 0.022 ** 0.013 NFI Share of nonfarm labor income -0.099 *** -0.012 -0.098 *** -0.011 NFn Share of nonlabor income 0.066 *** -0.005 0.065 *** -0.004 SID Simpson Index of Diversification 0.032 *** 0.004 0.032 *** 0 Sub Sum of subsidy -0.039 *** -0.08 -0.037 *** -0.072 Incidence of crop subsidy -0.154 *** -0.063 -0.159 *** -0.054 Log of crop subsidy 0.115 *** -0.017 0.122 *** -0.018* XIN Export Orientation Index -0.134 *** 0.03 -0.135 *** 0.029 Source: World Bank staff estimation using HIES 2016. Note: Negative sign indicates female productivity or earning advantage, while positive sign shows male productivity or earning advantage. Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent. a. Agricultural inputs include seeds, fertilizer, chemicals, hired labor, transport cost, and agricultural equipment/rental. Social norms could also be driving a wedge between male and female productivity, besides productiv- ity-related factors discussed and analyzed above. It has been widely documented that across countries, women spend more time in household work and child care, while men spend more time in market work (Ferrant, Pesando, and Nowacka 2014; Suárez Robles 2010). Even more important, if both market work and home activities are accounted for, women spend more time working in total than men (Rubiano- Matulevich and Viollaz 2019; Ilahi 1999). Such unequal patterns are typically driven by gender and social norms. Unequal time allocation (and unequal distribution of inputs of production) is also found to con- tribute to inefficiencies in agricultural productivity (Udry et al 1995). 16 Overall, our findings above are consistent not only with input constraints, but also with a context in which gender-specific norms influence agricultural production decisions and productivity. The litera- ture frequently describes women’s limitations in the activities that can be taken up and notes their pref- erence for agricultural work, including small livestock farming, that can be conducted from home and allows them to tend to household chores and childcare at the same time. Our regression result shows that the productivity gender gap in favor of men remains after controlling for personal, household, loca- tion, and agricultural variables (regression (1) in table 5). Gender-specific norms are a plausible factor that accounts for the persistent gender gap. 16. Analysis is based on the Sri Lanka Time Use Survey 2017, which collects information on how individuals spend their time in 15-minute intervals throughout the day, with detailed records of the activity conducted. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 3. Determinants of Higher Agricultural Productivity 36 Indeed, time use data suggest that women working in agriculture spend significantly less time on paid employment and more time on unpaid domestic work, whether compared to male workers in agriculture or their counterparts working outside of agriculture. Figure 11 reports the differences in a day’s time use by gender, with hours spent on employment, unpaid domestic work, other unpaid work, and nonproductive activities (such as leisure, self-care, learning, etc.). The number of hours spent on paid work by women engaged in the agriculture sector is 4.8 hours per day on average. This is significantly lower than the number of hours spent on paid work by male workers in the same sector (7.4 hours) or women employed outside of the agriculture sector (7.7 hours). Moreover, women are about twice as like- ly to be working from home, which might reflect the need to tend to household chores and care respon- sibilities (figure 12). FIGURE 11 Differences in daily time use by gender FIGURE 12 Agriculture workers’ work location, by gender 24 100 20 90 80 16 Hours per day 70 60 Percent 12 50 8 40 30 4 20 10 0 0 Male, Female, Male, Female, Male workers, Female workers, agriculture agriculture non-agriculture non-agriculture agriculture agriculture Employment Unpaid domestic work Home Non-productive activities (e.g., leisure, self-care) Workplace Other productive activities Other Source: World Bank staff estimation using Sri Lanka Time Use Survey 2017. Source: World Bank staff estimation using Sri Lanka Time Use Survey 2017. BOX 3 Measuring the gender gap in agricultural productivity: A literature review The measurement of gender gaps in agricultural productivity is empirically challenging. Studies on the gender agricultural productivity gap typically rely on one of two strategies: (i) interhousehold differences, based on agricultural production of male- or female-headed households, or (ii) intrahousehold differences, based on agricultural production of male- or female-managed plots. The problem with comparing agricultural productivity based on the head of household is that there are possible omitted varia- bles, since most women who head households tend to be widowed, have a migrant husband, or in general display systematic differ- ences from male-headed households. Moreover, both strategies rely on linear estimations, which could result in biased coefficients if the endogeneity of input choices is not addressed (see Quisumbing [1996] for a review). Despite this issue, comparing male- and female-managed plots allows for analyses of the returns to productive inputs. In this regard, the Sri Lanka HIES provides a unique opportunity to study the determinants of the gender gap, as it has detailed information for male- and female-managed plots. In previous studies, the gender gap in agricultural productivity is driven by the structural effect. A widely used strategy to evalu- ate the agricultural productivity gender gap is the Oaxaca-Blinder decomposition (Oaxaca 1973; Blinder 1973). The Oaxaca-Blinder strategy decomposes the simple difference in productivity between genders in a component explained by differences in farmer and plot characteristics (endowment effect) and an unexplained component or differences in the returns to such factors (structural AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 3. Determinants of Higher Agricultural Productivity 37 effect). Some studies using this approach have found that the gender gap in agricultural productivity is largely driven by the struc- tural component. For example, Aguilar et al (2015) estimate a gap of 23.4 percent in favor of men in Ethiopia, with 57 percent of it still unexplained after accounting for land and manager characteristics. Ali et al (2016) estimate a gap of 17.5 percent in Uganda, with 30.4 percent attributed to unexplained returns to endowments. Oseni et al (2015) find a gap of 28 percent in favor of men in northern Nigeria, again mainly driven by differentiated returns. Backiny-Yetna and McGee (2015) find the gender gap in Niger to be 18.3 per- cent in favor of men, with 148 percent of it explained by the structural effect. But differences across countries are important in deter- mining the relative contribution of these effects. Contrary to these findings, Kilic et al (2015) estimate a gender gap of 25 percent in favor of men in Malawi but find that 82 percent of it is driven by differentials in observables, rather than the structural component. The child dependency ratio, male labor, crop diversification, productive inputs, and land size are among the main drivers of the gender gap in productivity. An advantage of the Oaxaca-Blinder strategy is that it not only allows decomposition of the gender gap in the endowment and structural effects, but also allows the evaluation of each factor’s marginal contribution. This strategy has allowed previous studies to highlight the most important determinants to the structural component of the agricultural gender gap. For example, differences in the returns to extension services, land certification, land extension, and product diversification have been found to be the main drivers of the structural gap in Ethiopia (Aguilar et al. 2015). In Uganda, the main drivers of the structural gap were identified as coming from the child dependency ratio, uptake of and return to improved seeds and pesticides, and male- owned assets (Ali et al. 2016). For Niger, the main drivers of the structural effect were family and hired labor, the child dependency ratio, distance of the household to the nearest major road, and elevation of the plot (Backiny-Yetna and McGee 2015). In Tanzania, the main drivers of the structural effect were plot area and family labor (Slavchevska 2015). In Malawi, where the endowment effect was the main driver of the gender gap, the most important factors were high-value crop cultivation and household male labor inputs (Kilic, Palacios-Lopez, and Goldstein 2015). To summarize, existing literature has found the most important factors contributing to the gender gap in agricultural productivity may be the child dependency ratio, male labor, crop diversification, and productive inputs. The inverse relationship between land size and agricultural productivity is a critical factor to explain differences in agricul- tural output. The inverse relationship between land size and agricultural productivity has been long studied. a Most studies have found this relationship to hold in a variety of contexts (Berry and Cline 1979; Carter 1984; Eswaran and Kotwal 1986; Benjamin 1995; Barrett 1996), although a few did not find evidence of such a relationship (Kevane 1996; Zaibet and Dunn 1998). It is not clear yet what explains the inverse relationship, though there is some consensus that family labor could be an important driver (Carter 1984; Barrett 1996). Smaller farms use more family labor than larger farms (Carter 1984), and there is a U-shaped relationship as larger commercial farms, investing in capital and commercial crops, have monotonic increases to land size (Carter and Wiebe 1990). Other factors identified as important drivers are price risk (Barrett 1996) and land quality (Benjamin 1995). Mismeasurement of plot sizes was also put forward as a possible explanation (Lamb 2003), but recent studies using more sophisticated measures of land size con- firmed the inverse relationship (Carletto, Savastano, and Zezza 2013; Larson et al. 2014). The literature has not yet reached an agreement on the association between the inverse relationship and gender. That asso- ciation has been less explored because of the methodological issues outlined above (Quisumbing 1996). Some studies found evi- dence of returns to land size contributing in favor of women (Aguilar et al. 2015; Kilic, Palacios-Lopez, and Goldstein 2015; Ali et al. 2016; Backiny-Yetna and McGee 2015; Slavchevska 2015). The most widely accepted explanation for this result is that the inverse relationship plays in favor of women, as women control smaller plots. As discussed, it is widely agreed that the inverse relationship holds in the context of family labor allocation. Udry et al (1995) and Udry (1996) document that plots controlled by women in the African context are farmed less intensely than similar plots controlled by men, although this could be context dependent. For exam- ple, Barrett, Bellemare, and Hou (2010) argue that activities are gendered in Madagascar, but control over plots does not vary signif- icantly within households, suggesting that the inverse relationship found in their study is not driven by intrahousehold asymmetries. a. The origin of the study of the inverse relationship has been attributed for historical studies to Chayanov (1966), on the Soviet Union in the early 20 th century; for contemporary studies, the seminal work is Sen (1962) in the context of India. 4.  Conclusion and Policy Implications AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 4. Conclusion and Policy Implications 39 Using detailed information on agricultural activities among farm households in Sri Lanka, this note describes and further disaggregates by gender patterns in production, productivity, and incomes. The analysis describes varying spatial patterns in production across the country, in which high-productivity areas coincide with those having a high share of high-value crops that are more export-oriented, and low-productivity areas are mainly located in places with a high level of paddy production. Subsistence farming has declined significantly over time but remains at high levels for paddy. Female farmers enjoy an unconditional productivity advantage which is primarily attributed to lower access to land. Women cultivate a smaller area of land, and the underlying inverse relationship between cultivated area and productivity leads to higher average productivity. Despite this productivity advantage, the small size of cultivated land among female farmers leads to lower crop income relative to male farmers. The second source of female productivity advantage is that the crop mix of female farmers contains a relatively large portion of export-oriented/higher-value crops. Female farmers tend to grow more high-value crops, such as tea while a relatively large share of male farmers engages in paddy farming, which is domestically oriented and yields low returns. Thus promoting export-oriented/high-value crops could contribute to higher agricultural productivity and incomes as well as enhance gender equality. Once controlling for land area and crop mix, male farmers have a conditional productivity advantage over female farmers. This is largely explained by their greater access to resources, including agricultural inputs, transportation, subsidy, unpaid family labor, and mechanization. The analysis in this paper points to several areas where policies could focus to improve rural liveli- hoods. First, diversification accompanied by a shift to high-value agriculture could help raise agricul- tural productivity. Nearly half of Sri Lanka smallholders are engaged in paddy production and their productivity is low. In contrast, the cultivation of export-oriented crops — which tend to be high-value crops — is associated with higher productivity and crop income. For this, the local agroecological context will matter for the type of diversification as not all areas are suitable for the cultivation of high-value plantations crops. Increasing the productivity of paddy farmers is another important route to improv- ing overall productivity. Data suggest that improvements in the last decade have been slow. It will be important to understand which factors might have played a role. Policies to equalize access to resources like land and agricultural inputs, as well as interventions that could free up women’s household responsibilities, are likely to help increase women’s access to land and generate more crop income. With regard to land ownership, there are some long-standing issues that may warrant attention. For example, the land law can discriminate against women who opt to be governed by personal laws. Women who marry under the Thesawalami law cannot gain control of prop- erty without their husband’s consent (Zainudeen 2016). Under the Land Development Ordinance of 1935 and its subsequent amendments, the grant of state land in agricultural settlement schemes continues to AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 4. Conclusion and Policy Implications 40 favor men over women because grants are generally made to the male head of the household (Ranaraja 2020). It would also be important to understand the constraints that lead to unequal access to agricul- tural resources. The analysis shows that male productivity advantage persists even after adjusting for agricultural resources and other characteristics, possibly reflecting remaining unequal patterns driven by gender and social norms. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 41 Appendix TABLE A.1 Distribution of production systems for farm households, by district Province District Crops only Mixed crop-livestock Livestock only Western Colombo 79.95 2.8 17.25 Western Gampaha 81.25 3.88 14.87 Western Kalutara 97.57 1.27 1.16 Central Kandy 89.83 4.61 5.56 Central Matale 92.28 3.21 4.51 Central Nuwara Eliya 74.78 9.86 15.36 Southern Galle 97.33 1.3 1.37 Southern Matara 95.32 2.32 2.36 Southern Hambantota 89.33 3.67 7 Northern Jaffna 60.9 11.8 27.3 Northern Mannar 65.23 8.26 26.5 Northern Vavuniya 63.9 22.05 14.05 Northern Mulaitivu 66.93 21.59 11.4 8 Northern Kikinochchi 70.74 4.17 25.09 Eastern Batticaloa 50.04 10.12 39.84 Eastern Ampara 88.67 6.27 5.06 Eastern Trincomalee 75.57 7.85 16.58 North-Western Kurunegala 90.25 6.44 3.31 North-Western Puttalam 76.42 7.79 15.79 North Central Anuradhapura 86.96 8.87 4.17 North Central Polonnaruwa 88.07 5.95 5.98 Uva Badulla 91.4 5.01 3.59 Uva Monaragala 97.52 0.87 1.61 Sabaragamuwa Ratnapura 97.87 0.85 1.28 Sabaragamuwa Kegalle 95.37 2.05 2.58 Total 88.43 5.07 6.49 Source: World Bank staff calculation using HIES 2016. Note: The table shows the share of farm households engaged in different production systems. Livestock activities are identified if the household reports output value from meat, fish, eggs, milk, and other livestock. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER Appendix 42 TABLE A.2 Summary statistics of self-employed farmers a Mean t-test Total Male Female (p-value) Monthly real gross crop income (Rs) 15,487 16,970 11,086 0.000 *** Cultivated area (acres) 1.724 1.945 1.044 0.000 *** b Productivity (Rs/acre) 16,905 15976 19,809 0.000 *** Household head 0.793 0.876 0.546 0.000 *** Age 51.7 51.9 51.1 0.000 *** Years of education 8.524 8.461 8.711 0.007 *** Access to nonfarm labor income 0.385 0.449 0.193 0.000 *** Access to non-labor income 0.470 0.451 0.529 0.000 *** Share of nonfarm labor income (%) 0.248 0.297 0.103 0.000 *** Share of nonlabor income (%) 0.174 0.14 4 0.262 0.000 *** Number of unpaid agricultural family labor 0.179 0.220 0.059 0.000 *** Agricultural inputs (Rs/month) 4812 5449 2923 0.000 *** Tractor (=1 if tractor(s), =0 otherwise) 0.104 0.122 0.048 0.000 *** Incidence of natural calamity 0.114 0.109 0.132 0.071 * Access to finance 0.090 0.098 0.066 0.000 *** Access to IT 0.880 0.882 0.874 0.267 Incidence of crop subsidy 0.337 0.389 0.180 0.000 *** Value of crop subsidy (Rs/month) 898 935 658 0.150 Export Orientation Index 0.216 0.177 0.339 0.000 *** Simpson Index of Diversification 0.102 0.115 0.060 0.000 *** = 1 if urban, = 0 otherwise 0.0271 0.0266 0.0287 0.9931 = 1 if rural, = 0 otherwise 0.9500 0.9490 0.9529 0.2583 = 1 if estate, = 0 otherwise 0.0229 0.0244 0.0184 0.104 8 Source: World Bank staff calculation using HIES 2016. Note: Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent. a. Self-employed farmers are defined as those who had income from at least one crop activity. Those farmers having only livestock activity are not included due to the lack of land information. b. Productivity is measured as gross income divided by cultivated area in acres. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER Appendix 43 TABLE A.3 Summary statistics of agricultural wage workersa Mean t-test Total Male Female (p-value) Monthly real crop wage (Rs) 13,330 15,460 10,319 0.000 *** Household head 0.544 0.739 0.269 0.000 *** Age 45.4 44.9 4 6.1 0.019 ** Years of education 6.088 6.361 5.704 0.000 *** Share of nonfarm labor income (%) 0.002 0.002 0.001 0.523 Share of farm self-employment (%) 0.085 0.121 0.035 0.000 *** Share of nonfarm non labor income (%) 0.091 0.087 0.097 0.164 = 1 if urban, = 0 otherwise 0.011 0.013 0.008 0.256 = 1 if rural, = 0 otherwise 0.657 0.720 0.568 0.000 *** = 1 if estate, = 0 otherwise 0.332 0.267 0.425 0.000 *** Source: World Bank staff calculation using HIES 2016. Note: Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent. a. Wage workers are defined as those who responded that their main activities were crop activities at the ISIC four-digit level and who had a wage. Self-employed farmers and agricultural wage workers are not mutually exclusive since some individuals have both self-employed and wage jobs. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER 44 References Adeleke, O., O. Adesiyan, O. Olaniyi, K. Adelalu, and H. Department of Census and Statistics. 2018. Household Matanmi. 2008. “Gender Differentials in the Productivity Income and Expenditure Survey 2016. HIES Final of Cereal Crop Farmers: A Case Study of Maize Farmers Report. Department of Census and Statistics. Ministry in Oluyole Local Government Area of Oyo State.” of National Policies and Economic Affairs. Agricultural Journal 3: 193–98. Department of Census and Statistics. 2020. Sri Lanka Time Use Aguilar, A., E. Carranza, M. Goldstein, T. Kilic, and G. Survey Final Report 2017. Department of Census and Statis- Oseni. 2015. “Decomposition of Gender Differentials tics. Ministry of Finance, Economy and Policy Development. in Agricultural Productivity in Ethiopia.” Agricultural Dharmasiri, L. M. 2008. “Crop Diversification for Economics 46 (3): 311–34. Sustainable Agriculture: A Case Study from the Mahaweli Ali, D., D. Bowen, K. Deininger, and M. Duponchel. 2016. Development Programme in Sri Lanka.” Sri Lanka Journal “Investigating the Gender Gap in Agricultural Productivity: of Agrarian Studies 12 (1). Evidence from Uganda.” World Development 87: 152–70. Eswaran, M., and A. Kotwal. 1986. “Access to Capital Backiny-Yetna, P., and K. McGee. 2015. “Gender Differentials as a Determinant of the Organization of Production and and Agricultural Productivity in Niger.” Policy Research Resource Allocation in an Agrarian Economy.” Economic Working Paper 7199, World Bank, Washington, DC. Journal 96: 482–98. Barrett, Christopher B. 1996. “On Price Risk and the Fabregas, R., M. Kremer, and F. Schilbach. 2019. “Realizing Inverse Farm Size Productivity Relationship.” Journal the Potential of Digital Development: The Case of Development Economics 51 (2): 193–215. of Agricultural Advice. Science 366 (6471). Barrett, C. B., M. F. Bellemare, and J. Y. Hou. 2010. “Reconsid- FAO (Food and Agriculture Organization). 2011. The State ering Conventional Explanations of the Inverse Produc- of Food and Agriculture 2010–11: Women in Agriculture— tivity–Size Relationship.” World Development 38 (1): 88–97. Closing the Gender Gap for Development. Rome: Food and Benjamin, D. 1995. “Can Unobserved Land Quality Agriculture Organization. Explain the Inverse Productivity Relationship?” Journal FAO (Food and Agriculture Organization). 2018. Country of Development Economics 46 (1): 51–84. Gender Assessment of Agriculture and the Rural Sector in Sri Berry, R. A., and W. Cline. 1979. Agrarian Structure and Lanka. Rome: Food and Agriculture Organization. Productivity in Developing Countries. Baltimore: Johns Ferrant, G., L. M. Pesando, and K. Nowacka. 2014. “Unpaid Hopkins University Press. Care Work: The Missing Link in the Analysis of Gender Blinder, A. 1973. “Wage Discrimination: Reduced Form and Gaps in Labour Outcomes.” OECD Development Centre. Structural Estimates.” Journal of Human Resources 8 (4): Hirimuthugodage, D., Madurawala, S., Senaratne, A., 436–55. Wijesinha, A. and Edirisinghe, C. 2014. Female entrepre- Carletto, C., S. Savastano, and A. Zezza. 2013. “Fact neurship and the role of business development services in pro- or Artifact: The Impact of Measurement Errors moting small and medium women entrepreneurship in Sri on the Farm Size–Productivity Relationship.” Journal Lanka. Institute of Policy Studies of Sri Lanka and Oxfam of Development Economics 103: 254–61. International, Sri Lanka. Carter, M. 1984. “Identification of the Inverse Relationship Ilahi, N. 1999. “Gender and the Allocation of Adult Time: Evi- between Farm Size and Productivity: An Empirical dence from the Peru LSMS Panel Data.” Policy Research Analysis of Peasant Agricultural Production.” Oxford Working Paper 2744, World Bank, Washington, DC. Economic Papers 36: 131–45. ILO (International Labour Organization). 2018. Future Carter, M., and K. Wiebe. “Access to Capital and Its Impact of Work for Tea Smallholders in Sri Lanka. Colombo: ILO. on Agrarian Structure and Productivity in Kenya.” Jann, B. 2008. “The Blinder–Oaxaca Decomposition for American Journal of Agricultural Economics 72 (5): 1146–50. Linear Regression Models.” Stata Journal 8 (4): 453–79. Central Bank of Sri Lanka. 2020. Economic and Social Joshi, P. K., A. Gulati, P. S. Birthal, and L. Tewari. 2004. Statistics of Sri Lanka 2020. Colombo: Central Bank of Sri “Agriculture Diversification in South Asia: Patterns, Lanka Statistics Department. Determinants and Policy Implications.” Economic and Chayanov, A., 1966. The Theory of Peasant Economy. Political Weekly, 2457–67. Published for: Irwin, R. The American Economic Kevane, M. 1996. “Agrarian Structure and Agricultural Association. Homewood, IL. Practice: Typology and Application to Western Sudan.” Croppenstedt, A., M. Goldstein, and N. Rosas. 2013. “Gender American Journal of Agricultural Economics 78 (1): 236–45. and Agriculture: Inefficiencies, Segregation, and Low Kilic, T., C. Carletto, J. Miluka, and S. Savastano. 2009. “Rural Productivity Traps.” World Bank Research Observer 28 (1): Nonfarm Income and Its impact on Agriculture: Evidence 79–109. from Albania. Agricultural Economics 40 (2): 139–60. AGRICULTURAL PRODUCTIVITY, DIVERSIFICATION AND GENDER References 45 Kilic, T., A. Palacios-Lopez, and M. Goldstein. 2015. “Caught Slavchevska, V. 2015. “Gender Differences in Agricultural in a Productivity Trap: A Distributional Perspective Productivity: The Case of Tanzania.” Agricultural on Gender Differences in Malawian Agriculture.” World Economics 46 (3): 335–55. Development 70: 416–63. Suárez Robles, P. 2010. “Gender Disparities in Time Lamb, R. L. 2003. “Inverse Productivity: Land Quality, Labor Allocation, Time Poverty, and Labor Allocation across Markets, and Measurement Error.” Journal of Development Employment Sectors in Ethiopia.” In Gender Disparities Economics 71 (1): 71–95. in Africa’s Labor Market, edited by Jorge Saba Arbache, Larson, D., K. Otsuka, T. Matsumoto, and T. Kilic. 2014. Alexandre Kolev, and Ewa Filipiak, 299–332. Washington, “Should African Rural Development Strategies Depend DC: Agence Française de Développement and World on Smallholder Farms? An Exploration of the Inverse Bank. Productivity Hypothesis.” Policy Research Working Paper Udry, C. 1996. “Gender, Agricultural Production, and the 6190, World Bank, Washington, DC. Theory of the Household.” Journal of Political Economy Mani, M., S. Bandyopadhyay, S. Chonabayashi, A. 104 (5): 1010–46. Markandya, and T. Mosier. 2018. South Asia’s Hotspots: Udry, C., J. Hoddinott, H. Alderman, and L. Haddad. 1995. The Impact of Temperature and Precipitation Changes “Gender Differentials in Farm Productivity: Implications on Living Standards. South Asia Development Matters. for Household Efficiency and Agricultural Policy.” Food Washington, DC: World Bank. Policy 20 (5): 407–23. Oaxaca, R. 1973. “Male-Female Wage Differentials in Urban Weerahewa, Jeevika, Sarath Kodithuwakku, and Anoma Labor Markets.” International Economic Review 14: 693–709. Ariyawardana. 2010. “The Fertilizer Subsidy Program Oseni, G., P. Corral, M. Goldstein, and M. Winters. 2015. in Sri Lanka.” In Food Policy for Developing Countries: “Explaining Gender Differentials in Agricultural Produc- Case Studies, edited by Per Pinstrup-Andersen and Fuzhi tion in Nigeria.” Agricultural Economics 46 (3) : 285–310. Cheng. Ithaca, NY: Cornell University Press. Quisumbing, A. 1996. “Male-Female Differences World Bank. 2013. “What Is the Cost of a Bowl of Rice? The in Agricultural Productivity: Methodological Issues and Impact of Sri Lanka’s Current Trade and Price Policies Empirical Evidence.” World Development 24 (10): 1579–95. on the Incentive Framework for Agriculture.” Agriculture Ranaraja, Shyamali. 2020. “Assessment of Gender Equality and Environmental Services Discussion Paper 2, World and the Legal Framework of Sri Lanka.” Unpublished Bank, Washington, DC. background report for Sri Lanka Systematic Country World Bank. 2015. Sri Lanka: A Systematic Country Diagnostic. Diagnostic Update. Washington, DC: World Bank. Ratnayake, R. 2009. “Gender Issues in Agriculture—Sri World Bank. 2021a. “Sri Lanka Poverty Update.” Background Lanka.” Working paper, Sri Lanka Water Partnership. paper to Sri Lanka Poverty Assessment. World Bank, Rozelle, S., J. E. Taylor, and A. DeBrauw. 1999. “Migration, Washington, DC. Remittances, and Agricultural Productivity World Bank. 2021b. “The Rural Nonfarm Sector and in China.” American Economic Review 89 (2): 287–91. Livelihood Strategies in Sri Lanka.” World Bank, Rubiano-Matulevich, E., and M. Viollaz. 2019. “Gender Washington, DC. Differences in Time Use: Allocating Time between the Zaibet, L. T., and E. G. Dunn. 1998. “Land Tenure, Farm Size, Market and the Household.” Policy Research Working and Rural Market Participation in Developing Countries: Paper 8981, World Bank, Washington, DC. The Case of the Tunisian Olive Sector.” Economic Sen, A., 1962. An Aspect of Indian Agriculture. The Economic Development and Cultural Change 46 (4): 831–48. Weekly, Annual Number February. Zainudeen, Mohamed Hafiz. 2016. “Land Administration Silva, Kalinga Tudor, M. G. M. Razaak, Dhammika Her- in Sri Lanka. Issues and Challenges.” Unpublished manu- ath, Ramila Usoof-Thowfeek, S. Sivakanthan, and Vikne- script. World Bank. swaran Kunanayaham. 2018. “Post-war Livelihood Trends in Northern and Eastern Sri Lanka.” International Centre for Ethnic Studies and World Bank Group, Colombo.