Agricultural Statistics & Data Science · Foundations · also Agricultural basics (Unit I), Bridge course in agriculture
The agriculture a statistician must know to plan a trial, frame a crop survey or read a breeder's question: how crops, livestock and fisheries are produced, what pests and diseases do to yield, how agricultural research and statistics are organised, and the rural-development programmes surveys evaluate. It is Unit I of AIEEA-PG Group 07.
In the exams: AIEEA-PG Group 07 Unit I (heavy: the agriculture unit); ASRB ARS/NET interviews (present)
A statistician who works in agriculture needs to know what is being measured and why. Planning a fertiliser trial requires knowing that responses level off and that plots of different crops need different sizes; framing a crop survey requires knowing when crops are sown and harvested and how yield is defined; analysing a breeder's data requires knowing what heritability and heterosis mean; and evaluating a rural programme requires knowing what the programme was supposed to change. This manual is the agriculture that such a statistician must carry, written for graduates of mathematics, statistics and computing who enter agricultural statistics, and as the agriculture unit of postgraduate entrance examinations in statistical sciences.
It covers the place of agriculture, forestry, livestock and fisheries in national economies; how crops are grown, in which seasons and cropping systems, and how yield is built from its components; soils, nutrients and the response of crops to fertiliser; the main pests and diseases and how their incidence, severity and damage are measured; the basics of livestock and fish production and their performance measures; what plant and animal breeders select for; elementary agricultural economics and extension; how agricultural research, education and extension are organised nationally and internationally; the rural development programmes that statisticians are asked to measure; and the units, conversion factors and crop calendars that turn field records into comparable statistics.
The treatment is global. Seasons, crops, institutions and programmes differ by region, and the manual names examples from several continents rather than one country's system. Where a number varies by crop, breed, soil or country (a conversion factor, a fertiliser recommendation, a typical yield), the manual gives the principle and says that the value must be taken from local sources. Examination questions on this unit test definitions, simple calculations (cropping intensity, harvest index, land equivalent ratio, disease index, heterosis, feed conversion, unit conversions) and the reasoning that links agricultural facts to statistical design.
Agriculture in the broad sense used by national accounts includes crop production, livestock, forestry and fishing. In the System of National Accounts these form one sector, whose gross value added is the value of output minus the value of intermediate inputs such as seed, feed, fertiliser and fuel. Its share of gross domestic product (GDP) varies enormously. For the world as a whole agriculture, forestry and fishing contribute only a few per cent of GDP, but in many low-income countries the share is a quarter or more, and in some it is the largest single sector. Its share of employment is much larger than its share of output: roughly a quarter of the world's employed people work in agriculture, mostly as smallholders and family workers in Africa and Asia.
As economies grow, the share of agriculture in GDP and employment falls, even while agricultural output rises. This structural transformation follows from Engel's law, the observation that the share of household income spent on food falls as income rises, and from faster productivity growth in agriculture, which releases labour to industry and services. Countries such as the Republic of Korea, Brazil and China show the pattern sharply over recent decades; high-income countries in Europe and North America completed it earlier and now employ a few per cent of their workforce on farms while producing large surpluses.
The low share in GDP understates agriculture's importance. It supplies food security and raw materials for agro-industries (textiles, sugar, oils, leather, paper); it earns foreign exchange in countries that export coffee, cocoa, soybeans, beef, palm oil, rice or cut flowers; it holds most of the world's poor, so agricultural growth reduces poverty more than growth in other sectors in low-income economies; and it manages most of the land, water and biodiversity on which environmental outcomes depend. Agriculture is also a large source of greenhouse-gas emissions and the largest user of fresh water, which brings new statistical demands for environmental accounts.
Livestock (cattle, buffalo, sheep, goats, pigs, poultry, camels and others) provide meat, milk, eggs, fibre, hides, draught power and manure, and act as savings and insurance for poor households. Livestock products account for a large share of the value of agricultural output worldwide, commonly put at around 40%, and the share is higher in pastoral and high-income economies. Pastoral systems of the Sahel, the Horn of Africa, Central Asia and the Andes depend almost entirely on livestock.
Forests cover roughly a third of the world's land area. They supply timber, fuelwood, which remains the main cooking fuel for many rural households, and non-timber forest products such as fruits, resins, bamboo and medicinal plants, and they provide watershed protection, carbon storage and habitat. Forestry statistics come mainly from national forest inventories and from FAO's periodic Global Forest Resources Assessment.
Fisheries and aquaculture provide animal protein, livelihoods for millions of fishers and fish farmers, and major exports. Capture fisheries take wild fish and other aquatic animals from oceans, lakes and rivers; aquaculture farms them in ponds, cages, raceways and tanks. Capture production has been roughly stable for decades, while aquaculture has grown rapidly and now supplies about half of the aquatic animal food produced, with Asia producing most of it. Fishery statistics depend on landings records, catch-and-effort surveys and stock assessment, which are covered in the statistical ecology subject.
For a statistician, each sub-sector has its own units of observation (plot, holding, herd, animal, forest stand, vessel, pond), its own seasonality and its own measurement problems: livestock numbers change through births, deaths, sales and slaughter during a year; forest growth is slow and measured over decades; fish catches are mobile and often unrecorded. Agricultural censuses and surveys must be designed around these differences.
Crop seasons are set by temperature in temperate and high-altitude regions and by rainfall in the tropics. In temperate zones a spring-sown crop grows through summer, while winter cereals (winter wheat, barley, rye) are sown in autumn, survive winter in a vegetative state and are harvested the next summer; a cold period (vernalisation) is needed for some varieties to flower. In the monsoon tropics, a wet-season crop is sown with the onset of rains and a dry-season crop grows on residual moisture or irrigation; in South Asia these are called kharif and rabi. In East Africa, regions with two rainy seasons grow crops in the long rains and the short rains. In central Brazil, a second maize crop (safrinha) follows soybean within the same rainy season. Day length (photoperiod) also controls flowering: short-day crops such as many rice and sorghum varieties flower as days shorten.
A cropping system is the pattern of crops on a piece of land over time and the way they are managed. Monocropping grows the same crop each year; crop rotation alternates crops (for example a cereal after a legume) to break pest cycles and use nutrients better. Multiple cropping grows two or more crops on the same land in a year, either in sequence (double or triple cropping) or together. Intercropping grows two or more crops simultaneously: mixed intercropping without row arrangement, row intercropping in distinct rows (maize with beans in alternate rows), strip intercropping in strips wide enough for separate management, and relay intercropping, in which the second crop is sown into the standing first crop before its harvest.
Two statistics summarise land use intensity. Cropping intensity is gross cropped area divided by net sown area, times 100: a farm of 10 ha that grows two crops on 6 ha and one crop on 4 ha has a gross cropped area of 16 ha and a cropping intensity of 160%. The land equivalent ratio (LER) measures the advantage of an intercrop: LER = Σ (yield of crop i in intercrop / yield of crop i as sole crop). An LER of 1.3 means that 1.3 ha of sole crops would be needed to produce what 1 ha of intercrop produced.
Yield is the quantity of economic product per unit area: grain, tubers, fibre, fruit or forage, usually expressed in tonnes or kilograms per hectare at a stated moisture content. Biological yield is the total dry matter produced; economic yield is the harvested part. Their ratio is the harvest index, HI = economic yield/biological yield. Modern semi-dwarf wheat and rice varieties raised the harvest index from about 0.3 in traditional tall varieties to around 0.5, which explains much of the yield gain of the Green Revolution.
Yield is the product of components that develop at successive stages. For cereals: grain yield per unit area = number of panicles (or ears) per unit area × number of filled grains per panicle × weight of one grain. With 400 panicles per m², 80 filled grains per panicle and 25 mg per grain (a thousand-grain weight of 25 g), yield is 400 × 80 × 0.025 g = 800 g/m², that is 8 t/ha. For pulses, the components are plants per unit area, pods per plant, seeds per pod and seed weight; for tubers, plants, tubers per plant and mean tuber weight.
Components are determined at different times (panicle number around tillering, grain number around flowering, grain weight during grain filling) and compensate for one another: a thin stand produces more tillers or heavier grains. This compensation produces negative correlations among components even when each contributes positively to yield, which is why breeders use path analysis rather than simple correlations to judge their effects.
Plant population is set by seed rate and spacing. With rows 30 cm apart and plants 10 cm apart in the row, each plant occupies 0.03 m², giving 333,333 plants per hectare. Seed rate (kg/ha) is the desired plant population times seed weight, adjusted for germination and field emergence. Yield is limited by the environment and management: potential yield is set by radiation, temperature and variety with water and nutrients unlimited; attainable yield by water and nutrient limits; actual yield further reduced by pests, weeds and management. The gap between potential and actual yield, the yield gap, is a central concern of agricultural development and a frequent subject of statistical analysis.
Soil is the medium that holds water and nutrients for roots. Its texture is the proportion of sand, silt and clay particles; in the widely used USDA system sand is 2 to 0.05 mm, silt 0.05 to 0.002 mm and clay below 0.002 mm, and the textural triangle assigns names such as sandy loam or clay loam. Texture controls water-holding capacity, drainage and nutrient retention: clays hold more water and nutrients, sands drain fast. Soil structure, organic matter, depth, pH and salinity complete the picture. Soil pH affects nutrient availability; most crops grow best in slightly acid to neutral soils, acid soils are limed, and saline or sodic soils are reclaimed by leaching and amendments such as gypsum. Cation exchange capacity (CEC), the soil's capacity to hold exchangeable cations, rises with clay and organic matter. National soil classifications differ, and the FAO-based World Reference Base and USDA Soil Taxonomy are the two main international systems.
Seventeen elements are generally regarded as essential for higher plants. Carbon, hydrogen and oxygen come from air and water. The primary macronutrients are nitrogen (N), phosphorus (P) and potassium (K); the secondary macronutrients are calcium, magnesium and sulphur; the micronutrients are iron, manganese, zinc, copper, boron, molybdenum, chlorine and nickel. Deficiency symptoms are characteristic: nitrogen deficiency yellows older leaves first, because nitrogen is mobile in the plant.
Fertiliser contents are traditionally expressed as N, P₂O₅ and K₂O. Urea contains 46% N; diammonium phosphate is 18% N and 46% P₂O₅; muriate of potash (potassium chloride) is 60% K₂O. To convert, P = 0.436 × P₂O₅ and K = 0.830 × K₂O. To supply 60 kg N/ha from urea needs 60/0.46 = 130 kg urea. Organic manures supply nutrients and organic matter, and legumes fix atmospheric nitrogen through symbiosis with rhizobia, which is why rotations with legumes reduce nitrogen needs.
Crop yield responds to an increasing nutrient dose with diminishing returns: each additional kilogram adds less than the one before, until yield reaches a maximum and may then fall. Liebig's law of the minimum states that growth is controlled by the scarcest resource, so adding a nutrient that is not limiting does nothing until the limiting one is supplied. Mitscherlich formalised diminishing returns as y = A(1 − e^(−c(x + b))), where A is the maximum yield, x the dose, b the soil supply in dose units and c an efficiency constant.
In practice response is often fitted with a quadratic, y = a + bx − cx², from a trial with several doses. The dose giving maximum yield is x_max = b/(2c). The economic optimum dose is lower: it is where the extra yield from one more unit of nutrient just pays for it, that is where the marginal product equals the price ratio, b − 2cx = p_x/p_y, giving x_opt = (b − p_x/p_y)/(2c), where p_x is the price of the nutrient and p_y the price of the crop. With y = 2,000 + 30x − 0.1x² (kg/ha, x in kg N/ha), the maximum yield dose is 150 kg N/ha; with a nutrient price six times the grain price, the optimum is (30 − 6)/0.2 = 120 kg N/ha. Other forms (linear-plateau, quadratic-plateau, square-root) fit some data better and can give very different optima, so model choice matters.
Efficiency indices summarise response. Agronomic efficiency is the extra yield per unit of nutrient applied, (Y_f − Y₀)/F, in kg grain per kg nutrient. Apparent recovery efficiency is the extra nutrient uptake per unit applied. Partial factor productivity is yield divided by the nutrient applied. Fertiliser recommendations are based on soil tests calibrated against response trials: a low soil-test value predicts a large response. Recommended doses vary widely by crop, soil, climate and prices, and must come from local trials; a statistician's role is to design the response trials and fit and compare the response models.
Crop pests in the broad sense include insects and mites, pathogens (fungi, oomycetes, bacteria, viruses and phytoplasmas), nematodes, weeds, and vertebrates such as rodents and birds. FAO and others estimate that pests and diseases reduce global crop production by a large fraction, commonly put at 20 to 40%, with weeds causing the largest potential loss in many crops. Examples that recur in statistics and surveys include rust and blast diseases of wheat and rice, late blight of potato, fall armyworm (which spread from the Americas to Africa and Asia from 2016), desert locust outbreaks across Africa and South-West Asia, stem borers of cereals, whitefly-transmitted viruses of cassava and vegetables, and parasitic weeds such as striga in sub-Saharan Africa.
Disease development depends on the disease triangle: a susceptible host, a virulent pathogen and a favourable environment, all three needed. Epidemics are monocyclic (one infection cycle per season, as in many soil-borne diseases) or polycyclic (many cycles, as in rusts and late blight), and polycyclic epidemics grow roughly logistically, which is the basis of the epidemic models used in forecasting.
Integrated pest management (IPM) combines resistant varieties, cultural practices, biological control and need-based pesticide use. Its central decision tool is the economic threshold. The economic injury level (EIL) is the pest density at which the cost of control equals the value of the damage it prevents; the economic threshold (action threshold) is a lower density at which control should start, to prevent the population reaching the EIL. A widely used formula is EIL = C/(V × I × D × K), where C is the cost of control per unit area, V the value per unit of produce, I the injury per pest, D the damage (yield loss) per unit of injury, and K the proportional reduction of injury achieved by control. Thresholds vary by crop, pest, price and region and are set from local data.
Disease incidence is the proportion (or percentage) of plants, leaves or other units that are diseased, regardless of severity. Disease severity is the proportion of tissue affected (the percentage of leaf area covered by lesions), assessed by eye with the help of standard area diagrams or by image analysis. Severity is often scored on an ordinal scale, such as 0 to 5 or 1 to 9, and summarised by a percent disease index: PDI = (sum of all numerical ratings)/(number of units assessed × maximum rating) × 100. For 50 plants scored on a 0-4 scale with 10, 15, 12, 8 and 5 plants in classes 0 to 4, the sum of ratings is 0 + 15 + 24 + 24 + 20 = 83, and PDI = 83/(50 × 4) × 100 = 41.5%.
Epidemics are summarised over time by the area under the disease progress curve (AUDPC), computed by the trapezoidal rule from repeated assessments: AUDPC = Σ [(yᵢ + yᵢ₊₁)/2] × (tᵢ₊₁ − tᵢ), where yᵢ is severity at time tᵢ. It integrates both the timing and the intensity of disease and is widely used to compare varieties and fungicides.
Insect populations are measured by absolute methods (counts per plant, per unit area or per unit volume of soil) and relative methods (catches in light traps, pheromone traps, sticky traps, sweep nets, or counts per unit time), which index population changes without giving densities. Damage is measured as the percentage of damaged plants, tillers, fruits or pods: dead hearts and white ears for stem borers, bored fruits for fruit borers, defoliation for leaf feeders.
Insecticide efficacy trials correct for natural mortality in the control. Abbott's formula gives corrected mortality = (T − C)/(100 − C) × 100, where T and C are percentage mortality in the treated and control groups. With 80% mortality in the treated group and 20% in the control, corrected mortality is 60/80 × 100 = 75%. Henderson and Tilton's formula extends the correction to populations counted before and after treatment. Because pest counts are clumped and disease scores are ordinal or proportions, these data usually need transformation or count and ordinal models in analysis.
Livestock systems range from extensive pastoral herding on rangelands, through mixed crop-livestock systems in which animals eat crop residues and provide manure and draught, to intensive confined systems for poultry, pigs and dairy. Production depends on genetics, nutrition, health, reproduction and management, and each has standard performance measures that appear in surveys and experiments.
For dairy animals the key records are milk yield per day and per lactation, lactation length, peak yield, fat and protein content, age at first calving, dry period and calving interval. Lactation records are often standardised to a 305-day lactation so that cows can be compared; a calving interval of about 12 to 13 months is commonly regarded as a target for dairy cattle, and longer intervals reduce lifetime production. Fertility is measured by conception rate, services per conception and calving rate.
For meat animals the records are birth weight, weaning weight, average daily gain, slaughter weight and dressing percentage (carcass weight as a percentage of live weight). Feed conversion ratio (FCR) is feed consumed per unit of live-weight gain (or per unit of eggs or milk); a lower FCR is better. A broiler flock that eats 3,400 kg of feed and gains 2,000 kg has an FCR of 1.7. Values differ widely between species, breeds and systems.
Poultry egg production is reported as hen-day production (eggs laid divided by the number of hens alive each day, summed over the period) or hen-housed production (eggs laid divided by the number of hens originally housed). Hen-housed production penalises mortality and is always lower than or equal to hen-day production. Mortality, livability and flock uniformity complete the record.
Livestock numbers of different species are added using livestock units: a common international standard, the tropical livestock unit, equals an animal of 250 kg live weight, with conversion coefficients for each species that differ by region. Livestock surveys must deal with herd dynamics (births, deaths, purchases, sales and slaughter) and with mobile herds, which complicate area-based sampling.
Capture fisheries statistics record catches (landings) by species, gear, vessel and area, and fishing effort (number of trips, days at sea, hooks, net hauls). Catch per unit effort (CPUE), catch divided by effort, is used as an index of stock abundance, on the assumption that catch is proportional to abundance times effort. That assumption fails when fishing technology improves, when fishers concentrate on remaining dense patches, or when fish school, so CPUE trends must be interpreted with care. Stock assessment models, maximum sustainable yield and yield-per-recruit analysis are treated in the statistical ecology subject.
Aquaculture is practised in fresh water (carps, tilapia, catfish), brackish water (shrimp, milkfish) and the sea (salmon, seaweeds, mussels and oysters), at scales from household ponds to industrial cage farms. Systems are extensive (natural food, low stocking density), semi-intensive (fertilisation and supplementary feed) and intensive (complete feed, aeration, high density). Key production records are stocking density (number or weight of fish per unit area or volume), survival rate (harvested number as a percentage of stocked number), growth, yield per unit area and FCR.
Growth of fish over a culture period is often summarised by the specific growth rate, SGR = 100 × (ln W₂ − ln W₁)/t, the percentage increase in weight per day, where W₁ and W₂ are mean weights at the start and end and t is the number of days. Fish growing from 10 g to 250 g in 150 days have SGR = 100 × ln(25)/150 = 2.15% per day. Water quality (dissolved oxygen, temperature, ammonia, pH) governs survival and growth and is routinely monitored. For statisticians, aquaculture experiments raise the problem that the experimental unit is the pond, tank or cage, not the individual fish, a frequent source of pseudo-replication in published trials.
Breeding improves the genetic merit of crops and livestock. Breeders measure traits (yield, quality, maturity, plant height, disease resistance, milk yield, growth rate, fertility) on many individuals or lines, separate genetic from environmental variation, and select the best as parents or as varieties.
The phenotypic value of an individual is P = G + E, its genotypic value plus an environmental deviation. Phenotypic variance is partitioned as V_P = V_G + V_E (plus genotype-by-environment interaction when several environments are involved), and V_G further into additive variance V_A, due to the average effects of genes and transmitted from parent to offspring, dominance variance and epistatic variance. Broad-sense heritability is H² = V_G/V_P; narrow-sense heritability is h² = V_A/V_P, the proportion of phenotypic variance that responds to selection. Heritability is a property of a trait in a population and environment, not a constant of the trait: the same trait can have different heritabilities in different populations or environments.
The response to selection is predicted by the breeder's equation, R = h²S, where S, the selection differential, is the difference between the mean of the selected parents and the population mean. Equivalently, R = i h² σ_P, where i is the intensity of selection (the selection differential in standard-deviation units, which depends on the proportion selected) and σ_P the phenotypic standard deviation. If h² = 0.3 and the selected parents average 12 units above the population mean, the expected response is 3.6 units per generation. Response per year is the response per generation divided by the generation interval, which is why shortening the generation interval (rapid generation advance in crops, genomic selection of young bulls in dairy cattle) speeds genetic gain.
Genotype-by-environment interaction (G × E) means that varieties rank differently in different environments. It is the reason variety trials are conducted at many locations over several years and analysed for stability, which the multi-environment trials subject covers.
Breeding methods follow the mode of reproduction. Self-pollinated crops (wheat, rice, soybean, many pulses) are bred as pure lines by pedigree, bulk or single-seed descent methods after crossing. Cross-pollinated crops (maize, rye, many forage crops) are improved by population improvement (recurrent selection) and by hybrids. Vegetatively propagated crops (potato, cassava, sugarcane, banana) are bred as clones, which fix any genotype once found.
Heterosis (hybrid vigour) is the superiority of a hybrid over its parents. Relative heterosis is measured against the mid-parent value: (F₁ − MP)/MP × 100, where MP is the mean of the two parents. Heterobeltiosis is measured against the better parent, and standard or economic heterosis against the best commercial check variety, which is what matters for release. If the parents yield 4.0 and 6.0 t/ha and the F₁ yields 6.6 t/ha, relative heterosis is (6.6 − 5.0)/5.0 = 32% and heterobeltiosis (6.6 − 6.0)/6.0 = 10%. Heterosis is greatest in the F₁ and declines in later generations as heterozygosity is lost, which is why farmers buy fresh hybrid seed each season.
Animal breeders estimate an animal's breeding value, its genetic merit as a parent, from its own records and those of its relatives, and rank animals by estimated breeding values (EBVs). Progeny testing evaluates a sire by the performance of its daughters, which is accurate but slow; best linear unbiased prediction (BLUP) combines all relatives' records and is now standard; genomic selection adds DNA-marker information, covered in the statistical genomics subject. Crossbreeding of breeds exploits heterosis and complementarity, as in crossbred dairy cattle in many tropical countries.
Before release, new crop varieties go through official trials. Distinctness, uniformity and stability (DUS) testing establishes that a variety is a new, distinct and stable entity, the basis of plant variety protection under the UPOV system; value for cultivation and use (VCU) trials, run over several locations and years, establish that it performs better than existing varieties. Statistical design and analysis of these trials is a major professional role in agricultural statistics.
Agricultural economics applies economic principles to farming, markets and policy. The production function relates output to inputs. Its three regions, defined by the law of variable proportions, are: stage I, where average product is still rising and more input should be used; stage II, between the maximum of average product and the point where marginal product reaches zero, where a rational farmer operates; and stage III, where marginal product is negative and more input reduces output. Within stage II, the profit-maximising input level is where the value of the marginal product equals the input price, the rule used to find the economic optimum fertiliser dose.
Farm costs are fixed (land rent, depreciation of machinery and buildings, interest on fixed capital, which do not vary with output in the short run) or variable (seed, fertiliser, pesticides, hired labour, fuel). Farm management surveys also distinguish paid-out costs from imputed costs of owned resources, such as the value of family labour and the rent of owned land, because a farm can show a cash profit while failing to pay its own labour a market wage. Gross margin is gross value of output minus variable costs; net farm income deducts fixed costs as well. The benefit-cost ratio (BCR) of a practice or project is the present value of benefits divided by the present value of costs; a BCR above 1 means the benefits exceed the costs.
Agricultural markets show particular features: seasonal price swings tied to harvests, price instability from weather shocks and inelastic demand (a small surplus can cause a large price fall), long marketing chains, and perishability. The price elasticity of demand is the percentage change in quantity demanded divided by the percentage change in price; for staple foods it is low. The marketing margin is the difference between the consumer price and the farm-gate price, and the producer's share is the farm price as a percentage of the consumer price: if farmers receive 30 currency units per kilogram and consumers pay 50, the producer's share is 60%. Agricultural price and cost statistics are the subject of the economic statistics and index numbers subject.
Agricultural extension carries research results to farmers and farmers' problems back to research. Its methods are individual (farm and home visits, office calls), group (demonstrations, field days, farmer field schools, group meetings) and mass (radio, television, printed material, mobile-phone advisory services and social media). A result demonstration shows the result of a proven practice on a farmer's field; a method demonstration shows how to do something. Farmer field schools, developed in South-East Asia in the late 1980s for integrated pest management in rice and later spread across Africa and Latin America, teach through season-long group experimentation. The training and visit system, promoted in many countries from the 1970s to the 1990s, organised extension staff around a fixed schedule of training and farm visits.
Adoption is the decision of an individual farmer to use an innovation; diffusion is its spread through a population over time. Rogers described adoption as a sequence of stages, commonly given in the older form as awareness, interest, evaluation, trial and adoption. He classified adopters by when they adopt, using the standard deviations of a normal distribution of adoption times: innovators (the first 2.5%), early adopters (the next 13.5%), early majority (34%), late majority (34%) and laggards (16%). The cumulative number of adopters over time typically follows an S-shaped (logistic) curve.
The rate of adoption depends on the innovation's relative advantage, compatibility with existing practices, complexity, trialability and observability, and on farmers' resources, information, credit and risk. Statisticians measure adoption through surveys (adoption rate: the percentage of farmers or area using a practice; adoption intensity: the share of a farm's area under it) and model it with binary and limited-dependent-variable regressions. Because adopters differ from non-adopters in many ways, comparing their yields does not measure the effect of the innovation, a point taken up in the chapter on programme evaluation.
Most countries organise agricultural research through a national agricultural research system (NARS): public research councils or institutes, agricultural universities, private companies, non-governmental organisations and farmer organisations. The balance between them differs, but three models recur.
In the land-grant model, research, teaching and extension are combined in universities. The United States set it up through the Morrill Act of 1862, which endowed colleges of agriculture, followed by federal funding of state agricultural experiment stations and of cooperative extension through the universities; a federal research service runs national laboratories alongside. Agricultural universities in several Asian and African countries were modelled partly on it.
In the research-council model, a national council or corporation funds and manages a network of research institutes and coordinates university research. Examples include the national agricultural research councils and corporations of India, Brazil, Kenya, China and France, which run commodity and regional institutes, multi-location coordinated trials and national germplasm collections. Brazil's public agricultural research corporation, created in the 1970s, is often cited for its role in making the tropical savanna (cerrado) productive.
In many countries, extension is run separately by the ministry of agriculture, and in others it has been partly privatised or contracted out, with input companies, cooperatives and non-governmental organisations advising farmers. Agricultural education runs from vocational schools and diploma colleges through universities offering bachelor's, master's and doctoral degrees in agriculture, veterinary science, forestry, fisheries and agricultural engineering.
Agricultural statistics are produced by national statistical offices and ministries of agriculture, sometimes with a specialised agricultural statistics research institute. They include agricultural censuses, crop area and yield surveys, livestock surveys, and price and cost data. Statisticians in research institutes design and analyse experiments for scientists; statisticians in official statistics run surveys and censuses; both roles appear in the examinations for which this manual is written.
International agricultural research is led by CGIAR, a global partnership founded in 1971 that funds a network of international research centres. Several of them began in the 1960s, when the rice research centre in the Philippines and the maize and wheat centre in Mexico bred the semi-dwarf varieties of the Green Revolution. Present centres work on rice, maize and wheat, potato and sweet potato, dryland crops, tropical agriculture, livestock, fish, forestry and agroforestry, water management and food policy, and they are located across Asia, Africa, Latin America and the Middle East. They hold major international gene banks, breed improved germplasm that national programmes adapt and release, and run large multi-site trials and household surveys whose data are increasingly published openly.
The Food and Agriculture Organization of the United Nations (FAO), founded in 1945, sets international standards for agricultural statistics and compiles FAOSTAT, the global database of production, trade, prices, land use and food balance sheets. FAO's World Programme for the Census of Agriculture, issued for each ten-year round, provides the concepts, definitions and recommended core items for national agricultural censuses, so that censuses in different countries can be compared. The Global Strategy to improve Agricultural and Rural Statistics, and its successor programmes, support countries in building integrated survey systems. Other relevant bodies include the International Fund for Agricultural Development (IFAD), which finances rural development projects; the World Bank, which funds agricultural projects and household survey programmes that include agriculture; the World Organisation for Animal Health (WOAH), which sets standards for animal disease reporting; and the Codex Alimentarius Commission for food standards.
Regional bodies coordinate research across groups of countries, such as forums for agricultural research in Africa, Asia-Pacific and Latin America. For a statistician, these organisations matter because they define the concepts and standards in which agricultural data are collected and reported, publish open data sets used in research, and employ statisticians in survey methodology, trial networks and impact evaluation.
Rural development programmes aim to raise rural incomes, reduce poverty and hunger, and improve services. They take many forms. Income-support and social-protection programmes include cash transfers, conditional cash transfers tied to school attendance or health visits (Mexico's programme, launched in 1997, and Brazil's family grant are widely studied examples), and public works programmes that offer paid employment on rural infrastructure (for example the employment guarantee scheme in India and Ethiopia's productive safety net programme). Agricultural programmes subsidise inputs such as fertiliser and seed (input subsidy programmes in Malawi and other African countries), provide credit and insurance, build irrigation and watershed works, strengthen extension, and support farmer organisations and value chains. Infrastructure programmes build rural roads, electricity, water supply and markets. Many countries also run land reform, tenure registration and rural livelihood programmes for women's self-help groups.
Each programme asks statisticians to measure inputs (budget spent, fertiliser distributed), outputs (kilometres of road, number of farmers trained), outcomes (adoption, yields, incomes, food consumption) and impacts (poverty, nutrition, resilience). Standard indicators include the poverty headcount ratio (the proportion of people below a poverty line, national or international), the poverty gap, the prevalence of undernourishment, child stunting and wasting rates, the Food Insecurity Experience Scale, dietary diversity scores, household income and consumption per adult equivalent, yields and productivity, and asset indices. The Sustainable Development Goals include agricultural indicators such as the volume of production per labour unit by size of enterprise and the proportion of agricultural area under productive and sustainable agriculture, for which FAO is the custodian agency.
Data for these indicators come from household surveys (living standards, income and expenditure, demographic and health surveys), agricultural surveys, programme administrative records (management information systems) and, increasingly, remote sensing and mobile-phone data. Administrative records report what the programme did; they rarely show what would have happened without it.
Monitoring tracks a programme's implementation and outputs against targets, using administrative data and periodic surveys. Evaluation asks whether the programme caused the changes observed. The central difficulty is the counterfactual: what would have happened to participants without the programme. A simple before-after comparison attributes every change in the period, including weather and prices, to the programme; a comparison of participants with non-participants is biased if participants differ, as they usually do, because programmes select poorer, better-connected or more motivated people, or people select themselves.
Impact evaluation designs address this. A randomised controlled trial assigns the programme at random among eligible households or villages (or randomises the order of roll-out), so that the two groups are comparable; the evaluation of Mexico's conditional cash transfer used randomised phasing-in across villages and became a model for the field. Where randomisation is impossible, quasi-experimental methods are used: difference-in-differences compares the change over time in participants with the change in non-participants, assuming parallel trends without the programme; propensity score matching compares participants with non-participants of similar observed characteristics; regression discontinuity compares units just above and just below an eligibility cut-off such as a landholding limit; and instrumental variables use a factor that affects participation but not outcomes directly. These methods are covered in the subject on causal inference.
A baseline survey before the programme starts, with the same questionnaire repeated at follow-up, is the foundation of credible evaluation. Statisticians decide the sample size needed to detect a given effect (power calculations, accounting for clustering within villages), the sampling design, the indicators and questionnaires, the treatment of attrition (households lost between survey rounds), and the analysis. They also report the cost-effectiveness of programmes and the distribution of benefits, because an average effect can hide large differences between poorer and richer households or between women and men.
Agricultural statistics are compiled from records in many units, and errors of conversion are among the commonest in published figures. The International System (SI) units are standard for official and scientific work: hectare (ha) for area, tonne (t) and kilogram (kg) for weight, litre for volume.
Area: 1 hectare = 10,000 m² = 0.01 km²; 1 acre = 0.4047 ha, so 1 ha = 2.471 acres; 1 km² = 100 ha. Many countries also use local land units of varying size, which must be converted using locally established factors. Weight: 1 tonne = 1,000 kg; 1 quintal = 100 kg; 1 pound = 0.4536 kg. Grain trade in North America uses the bushel, a volume unit converted to weight by standard weights per crop: a bushel of wheat or soybeans is 60 lb (27.2 kg) and a bushel of maize 56 lb (25.4 kg). Yield conversions combine both: 1 t/ha = 1,000 kg/ha = 10 quintals/ha = 100 g/m²; 1 bushel of maize per acre is about 62.8 kg/ha.
Moisture matters for any harvested product. Grain yields are reported at a standard moisture content, which differs by crop and country (values around 12% to 15.5% are common for cereals). To convert a weight W at actual moisture M_a to standard moisture M_s, use W_s = W × (100 − M_a)/(100 − M_s). A plot harvest of 5.0 kg at 20% moisture becomes 5.0 × 80/86 = 4.65 kg at 14%. Dry-matter yields (0% moisture) are used for forage and in physiological work.
Processing conversion factors link primary and processed products: paddy to milled rice, seed cotton to lint, sugarcane to sugar, live weight to carcass weight, milk to milk products. These factors depend on variety, technology and quality; for example, the milling outturn of rice is commonly in the range of about 60 to 70%. FAO publishes technical conversion factors for many commodities, and national statistical systems use their own measured values, which should be documented with each series.
Plot yields are scaled to per-hectare yields by dividing by the harvested (net) plot area. A net plot of 3 m × 4 m = 12 m² yielding 7.2 kg of grain gives 7.2/12 = 0.6 kg/m², or 6.0 t/ha. Using the gross plot area, including border rows, understates yield; harvesting border rows overstates it because of edge effects.
A crop calendar gives, for each crop and region, the typical dates of sowing, main growth stages and harvest. FAO and international crop-monitoring initiatives publish crop calendars by country and region, and national ministries publish their own. Calendars differ between hemispheres: wheat is harvested in May to August in much of the Northern Hemisphere and November to January in Argentina and Australia. They differ within countries with altitude and rainfall, and many crops have more than one season.
Crop calendars drive the timing of agricultural statistics. Area surveys must be done after sowing is complete and before the crop is harvested; yield surveys based on crop-cutting experiments must be done at harvest, plot by plot; forecasts are issued at fixed stages during the season (sowing, mid-season, pre-harvest) and revised as the season advances. Remote-sensing classification of crops depends on images taken at stages when crops look different from one another, which requires knowing the calendar.
Statistics are reported for agricultural years or marketing years rather than calendar years. An agricultural year groups the seasons that belong together (for example, from the wet-season sowing of one year to the dry-season harvest of the next), and a marketing year runs from the start of the main harvest of a crop. A crop harvested across two calendar years must be attributed consistently to one reporting year; mixing agricultural, marketing and calendar years is a common source of apparent discrepancies between national and international data.
For trial planning, the calendar sets the window for each operation: sowing date experiments must fall within it, and data collection (germination counts, flowering dates, disease scoring, harvest) must be scheduled to stages, not dates, because development varies with season. Days to 50% flowering and days to maturity are standard trial records precisely because stage timing differs between varieties and environments.
A mathematics or statistics graduate entering agricultural statistics uses this manual to answer the agriculture unit of the entrance examination and, more importantly, to talk to agronomists, breeders and economists in their own terms. In a first trial or survey project the student must know why border rows are discarded, what a yield component is, how disease is scored, and why a pond rather than a fish is the experimental unit. Without this, statistical advice is either ignored or wrong.
A statistician five years into work at a research institute, statistics office or development agency uses agricultural knowledge daily: choosing plot sizes and doses for a fertiliser trial and fitting response curves with economic optima, converting survey yields to standard moisture and consistent units, timing crop-cutting experiments to the calendar, designing livestock and fishery surveys around herd dynamics and seasons, and building impact-evaluation samples for an input-subsidy or irrigation programme with sensible indicators, and explaining results to agronomists and economists in their own terms.
A senior statistician with twenty-five years of experience sits on committees that set national methods: agricultural census content following international guidelines, conversion factors and standard moisture contents for official series, the indicators for monitoring development goals, and protocols for variety-release trials. Such a person explains to policy makers what the numbers mean and do not mean, resolves conflicts between administrative and survey estimates, and builds and mentors teams that combine agricultural and statistical expertise in research and official statistics.
A professor or adviser with a PhD and long experience shapes how agriculture and statistics are taught together, designs bridge courses for mathematics graduates and statistics courses for agriculture graduates, and advises international organisations and national systems on the design of integrated agricultural survey programmes, trial networks and impact evaluations. Such a person also leads research on new data sources, such as remote sensing, sensors and farm records, and on how they can be combined with traditional surveys and designed experiments.
The manual is an introduction for statisticians, not an agronomy, animal science or economics course. It does not give fertiliser recommendations, pesticide doses, feeding standards or typical yields, because these vary by crop, breed, soil, climate and country and must come from local sources. Shares of agriculture in GDP and employment, crop-loss percentages and the livestock share of output are given as broad global magnitudes; current values should be taken from FAOSTAT, the World Bank and national accounts. Institutional details of individual countries' research systems and named rural programmes are given only as brief examples; their structures change and should be checked against current official sources. Soil classification systems, livestock unit coefficients and standard moisture contents differ between countries and are described, not tabulated.
Original practice questions written to the syllabus and pattern of the exams named in each question's examTags. They are not past papers and do not reproduce any board's questions.
These notes are free to read here, with no account.
What the Laisha app adds:
Free for ten days, then a subscription. We would rather write that here than have you find it later.