F om Agr no i a to Far er ecorn en a 0 - ISPCispc.cgiar.org/sites/default/files/pdf/120.pdf · Chapter Eight 63 Variability in ... Experimentation Conduct experiments in farmers'

An Economics Training Manual

F om Agr noFar er ecorn

i a a to•en a 0

An Economics Training Manual

From Agronomic Data toFarmer Recommendations

elM M Y T

The International Maize and Wheat Improvement Center (CIMMYT) isan internationally funded, nonprofit scientific research and trainingorganization. Headquartered in Mexico, the Center is engaged in aworldwide research program for maize. wheat, and triticale, withemphasis on food production in developing countries. It is one of 13nonprofit international agricultural research and training centerssupported by the Consultative Group on International AgriculturalResearch (CGIAR). which is sponsored by the Food and AgricultureOrganization (FAO) of the United Nations, the International Bank forReconstruction and Development (World Bank), and the UnitedNations Development Programme (UNDP). The CGIAR consists of 40donor countries. international and regional organizations, and privatefoundations.

CIMMYT receives support through the CGIAR from a number ofsources, including the international aid agencies of Australia, Austria,Brazil, Canada, China, Denmark, Federal Republic of Germany,France. India, Ireland, Italy, Japan, Mexico, the Netherlands, Norway,the Philippines, Saudi Arabia, Spain, Switzerland. the United Kingdomand the USA, and from the European Economic Commission, FordFoundation, Inter-American Development Bank, InternationalDevelopment Research Centre. OPEC Fund for InternationalDevelopment, Rockefeller Foundation, UNDP, and World Bank.Responsibility for this publication rests solely with CIMMYT.

Correct Citation: CIMMYT. 1988. From Agronomic Data to FarmerRecommendations: An Economics Training Manual. Completelyrevised edition. Mexico, D.F.

ISBN 968-6127-18-6

I Preface

j i i

This document is a completely revised version of theCIMMYT Economics Program manual, From AgronomicData to Farmer Recommendations: An EconomicsTraining Manual, written by Richard Perrin, DonaldWinkelmann, Edgardo Moscardi, and Jock Anderson.Since its publication in 1976 that manual has beenthrough six printings and has been translated into sixlanguages. The manual has been used by countlessstudents and researchers for learning a straightforwardmethod of analyzing the results of on-farm agronomicexperiments and making farmer recommendations.

We approach the revision of such a successful manualwith considerable caution. Our work over the pastdecade has given us a chance to present this material,in the classroom and in the field, to agriculturalresearchers in a wide variety of settings all over theworld. This experience has led us to propose and testsome new ways of explaining and presenting keyconcepts. We gradually began to consider the possibilityof incorporating some of those ideas in a revisedmanual.

One of the first steps in the process was to introduce aset of exercises for classroom teaching, developed byLarry Harrington. Later, Robert Tripp and Gustavo Saindeveloped further exercises and methods of presentationwhich they tested in training courses. Tripp and Sainwrote the first draft of the present document and gUidedits review by the entire staff of the CIMMYT EconomicsProgram.

Just as this revised manual has built on the experienceof hundreds of researchers with the original version, wehope that those who use this new version will prOVidesuggestions for its improvement. We believe it will beuseful in the classroom as well as for individual studyand reference. A book of exercises has been developed toaccompany this manual and can be obtained fromCIMMYT. We hope that the new version of the manualwill find an acceptance as wide as that of itspredecessor.

Derek ByerleeDirectorCIMMYT Economics Program

Acknowledgements

iv

Many people have contributed to the production of thismanual. Jock Anderson and Richard Perrin. two of theauthors of the original manual, were kind enough toread the final draft of this revised version and to offerdetailed comments and suggestions. Miguel Avedillo.Carlos Gonzalez. Peter Hildebrand. Roger Kirkby,Stephen Waddington, and Patrick Wall also read thefinal draft and provided valuable ideas. In addition.participants in the courses and workshops presented bythe CIMMYT Economics Program over the past decadehave made significant contributions.

The document passed through several drafts. whichwould not have been possible without the superborganization and typing of Maria Luisa Rodriguez.Editing was in the very competent hands of KellyCassaday and design was skillfully directed by AnitaAlbert. Typesetting, layout. and production werecarefully managed by Silvia Bistrain R.. Maricela A. deRamos. Miguel Mellado E .. Rafael De la Colina F.. JoseManuel Fouilloux B.. and Bertha Regalado M.

Contents

Part One: Chapter One 1Introduction Overview of Economic Analysis

Part Two: Chapter Two 13The Partial Costs That Vary

Budget Chapter Three 20Gross Field Bene"Fits,Net Benefits, and the Partial Budget

Part Three: Chapter Four 30Marginal The Net Benefit Curve andAnalysis the Marginal Rate of Return

Chapter Five 34The Minimum AcceptableRate of ReturnChapter Six 38Using Marginal Analysisto Make Recommendations

Part Four: Chapter Seven 55Variability Preparing Experimental Results for

Economic Analysis: RecommendationDomains and Statistical AnalysisChapter Eight 63Variability in Yields:Minimum Returns AnalysisChapter Nine 71Variability in Prices: SensitivityAnalysis

Part Five: Chapter Ten 76Summary Reporting the Results

of Economic Analysis

Index 79

v

Part One Introduction

Chapter OneOverview ofEconomic Analysis

1

This manual presents a set of procedures for theeconomic analysis of on-farm experiments. It is intendedfor use by agricultural scientists as they developrecommendations for farmers from agronomic data.Developing recommendations that fit farmers' goals andsituations is not necessarily difficult, but it is certainlyeasy to make poor recommendations by ignoring factorsthat are important to the farmer. Some of these factorsmay not be very evident.

A recommendation is information that farmers can useto improve the productivity of their resources. A goodrecommendation can be thought of as the practiceswhich farmers would follow, given their currentresources, if they had all the information available to theresearchers. Farmers may be able to use arecommendation directly, as in the case of a particularvariety. Or they may adjust it somewhat to their ownconditions and needs, as in the case of a fertilizer levelor storage technique. The agronomic data upon whichthe recommendations are based must be relevant to thefarmers' own agroecological conditions, and theevaluation of those data must be consistent with thefarmers' goals and socioeconomic circumstances.

On-Farm Research

The stages of an on-farm research program areshown in Figure 1.1. The first step is diagnosis. Ifrecommendations are to be oriented to farmers. researchshould begin with an understanding of farmers'conditions. This requires some diagnostic work in thefield, including observations of farmers' fields andinterviews with farmers. The diagnosis is used to helpidentify major factors that limit farm productivity and tohelp specify possible improvements.

The information from the diagnosis is used in planningan experimental research program that includesexperiments in farmers' fields. The on-farm experimentsshould be planted on the fields of representativefarmers. After the first year. the experimental resultsform an important part of the information used forplanning research in subsequent crop cycles. Otherdiagnostic work continues during the management ofthe experimental program as researchers continue toseek information about farmers' conditions andproblems which will be useful in planning futureexperiments.

Figure 1.1. Stages of on-farm research_______________~ ____J

On-Fann Research

1. DiagnosisReview secondarydata. informal andformal surveys

~

Policy

National goals.input supply.credit. markets.etc.

Choosetargetfarmersandresearchpriorities

IdentifypoUcyissues

2

2. PlanningSelect priorities forresearch and designon-farm experiments

3. ExperimentationConductexperiments infarmers' fields toformulate improvedtechnologies underfarmers' conditions

14. Assessment-Farmer assessment-Agronomic evaluation-Statistical analysis-Economic analysis

15. RecommendationDemonstrate improvedtechnologies tofarmers

Newcomponentsfor on-farmresearch

JdentJfyproblemsfor stationresearcb

ExperimentStation

Develop andscreen newtechnologicalcomponents

3

The results of the on-farm experiments must beassessed. There are several elements in such anassessment. First, researchers must discuss the resultswith farmers to get their opinions of the treatments theyhave seen in their fields. The farmers' assessment isvery important. The experimental results must also besubjected to both an agronomic evaluation and astatistical analysis. Finally, an economic analysis of theresults is essential. The economic analysis helpsresearchers to look at the results from the farmers'viewpoint, to decide which treatments merit furtherinvestigation, and which recommendations can be madeto farmers. The procedures for carrying out such aneconomic analysis are the subject of this manual.

The results of an assessment of on-farm experimentscan be used for several purposes. First, they may beused to help plan further research. Some experimentswill have as their goal the clarification of productionproblems: Is production limited by the availability ofphosphorus? Will improved weed control give animportant increase in yields? The answers to suchquestions provide researchers with information forfurther work. As Figure 1.1 shows, that information canbe used to plan subsequent experiments. It also mayhelp orient work on the experiment station.

Second, the results may be used to makerecommendations to farmers. Some experiments willcompare possible improvements to farmers' currentpractices: Which level of phosphorus should be applied?Which weed control method gives the best results? Theanswers to these questions provide information to guidefarmers in their management decisions.

Finally, the results of on-farm experiments maysometimes be used to proVide information topolicymakers regarding current policy toward suchmatters as input supply or credit regulations.Experimental results can be used to help analyze policyimplementation: Given a significant response tophosphorus, is the appropriate fertilizer available? Dolocal credit programs allow farmers to take advantage ofnew weed control methods? Although the emphasis inthis manual will be on the economic analysis of on-farmexperiments for gUiding further research and makingrecommendations to farmers, at several points linksbetween on-farm research and policy implementationwill be mentioned.

4

Goals of the Farmer

In order to make recommendations that farmers willuse. researchers must be aware of the human elementin farming. as well as the biological element. They mustthink in terms of farmers' goals and the constraints onachieving those goals.

In the first place. many farmers are primarily concernedwith assuring an adequate food supply for their families.They may do this by producing much of what theirfamily consumes. or by marketing a certain proportionof their output and using the cash to obtain food. Farmenterprises also provide other necessities for the farmfamily. either directly or through cash earnings. Inaddition. the farm family is usually part of a widercommunity. towards which it may have certainobligations. To meet all of these requirements. farmersoften manage a very complex system of enterprises thatmay include various crops. animals. and off-farm work.Although the procedures of this manual concentrate onthe evaluation of improvements in particular cropenterprises. it is essential that these new practices becompatible with the larger farming system.

Second. whether farmers market little or most of theirproduce. they are interested in the economic return.Farmers will consider the costs of changing from onepractice to another and the economic benefits resultingfrom that change. Farmers will recognize that if theyeliminate weeds from their fields they will likely benefitby harvesting more grain. On the other hand. they willrealize that they must give up a lot of time and effort forhand weeding. or that alternatively they must give upsome cash to buy herbicides and then expend sometime and effort to apply them. Farmers will weigh thebenefits gained in the form of grain (or other usefulproducts) against the things lost (costs) in the form oflabor and cash given up. What farmers are doing in thiscase is assessing the difference in net benefits betweenpractices-the value of the benefits gained minus thevalue of the things given up.

As farmers attempt to evaluate the net benefits ofdifferent treatments. they usually take risk into account.In the weed control example just mentioned. farmersknow that in the case of drought or early frost they mayget no grain. regardless of the type of weed control.Farmers attempt to protect themselves against risks ofloss in benefits and often avoid choices that subject

1

them to these risks, even though such choices may onaverage yield higher benefits than less risky choices do.That farmers may prefer stable returns to the highestpossible returns is referred to as risk aversion.

Another factor in farmers' decision making, related torisk aversion, is the fact that farmers tend to changetheir practices in a gradual, stepwise manner. Theycompare their practices with alternatives, and seek waysof cautiously testing new technologies. It is thus morelikely that farmers will adopt individual elements, orsmall combinations of elements, rather than a completetechnological package. This is not to say that farmerswill not eventually come to use all the elements of apackage of practices, but simply that in offeringrecommendations to farmers it is best to think of astrategy that allows them to make changes one stepat a time.

Characteristics of On-Farm Experiments

What are the characteristics of agronomic experimentsthat will allow an appraisal of alternative technologies ina way that parallels farmer decision making? Thefollowing are five requirements of on-farm experimentsthat must be fulfilled if the procedures described in thismanual are to be useful:

The experiments must address problems that areimportant to farmers. It may be that farmers themselvesare not initially aware of a particular problem (e.g., anutrient deficiency or a disease), but if research does notlead to possibilities for significantly improving farmproductivity, it will neither attract the interest offarmers. nor merit assessment. Thus the experimentsdemand a good understanding of farmers' agronomicand socioeconomic conditions.

2 The experiments should examine relatively few factorsat a time. An on-farm experiment with more than, say,four variables will be difficult to manage and may beinappropriate given farmers' stepwise adoption behavior.

3 If researchers are to compare the farmers' practice withvarious alternatives in order to make a recommendation,then the farmers' practice should be included as one ofthe treatments in the experiment. The farmers will wantto see this comparison in any case.

1/ Once this work has beendone. and there is evidencethat farmers will adopt thenew insect control method.it could be used as anonexperimental variable inthe fertilizer experiments.as long as it is understoodthat the fertilizerrecommendation to bedeveloped from such trialsdepends on the farmers firstadopting the insect controlmethod.

4

5

6

The nonexperimental variables of an experiment shouldreflect farmers' actual practice. It is sometimes temptingto use higher levels of management for thenonexperimental variables to increase the chances ofobservable responses to the experimental variables. Thistype of experiment may certainly be justified in somecases, but the results usually cannot be used to makerecommendations to farmers.

An example may be useful. Assume that researcherswish to carry out a fertilizer experiment in an areawhere insects cause crop losses but farmers do notcontrol insects. There are four possibilities:

• Carry out the fertilizer experiment with good insectcontrol. The experiment will give interestinginformation on fertilizer response but will probablynot prOVide a relevant fertilizer recommendation forfanners who do not use insect control. An analysisof this experiment using the procedures in thismanual will give misleading results.

• Carry out the fertilizer experiment without insectcontrol (the farmers' practice). The results can beanalyzed, using the procedures in this manual, todecide what level of fertilizer is appropriate. givenfarmers' current insect control practices.

• If insects are indeed a very serious problem, it maybe better to experiment first with insect controlmethods before experimenting with fertilizer. Thediagnosis and planning steps of on-farm research aremeant to help set these priorities. The methods ofthis manual could then be used to help identify anappropriate insect control method forrecommendation to farmers.lI

• If insects and fertility are both serious problems,then an experiment can be designed which takesboth insect control and fertilizer as experimentalvariables. As long as one treatment represents thefarmers' practice with respect to both insect controland fertility, the experiment can be analyzed usingthe procedures in this manual.

Finally. not only must the management ofnonexperimental variables be representative of farmers'practice, but the experiments must be planted atlocations that are representative of farmers' conditions.

Recommendation domain

7

If most of the farms are on steep slopes, the results ofexperiments planted on an alluvial plain will probablynot be relevant. Similarly, if most farmers plant onecrop in rotation with another, experiments from fieldsthat have been under fallow may provide little usefulinformation. More will be said in the following sectionabout selecting locations.

Experimental Locationsand Recommendation Domains

The development of recommendations for farmers mustbe as efficient as possible. The conditions under whichfarmers live and work are diverse in almost everyrespect imaginable. Farmers have different amounts andkinds of land, different levels of wealth, differentattitudes toward risk, different access to labor, differentmarketing opportunities. and so on. Many of thesedifferences can influence farmers' responses torecommendations. But it is impossible to make aseparate recommendation for each farmer.

As a practical matter, researchers mustcompromise by identifying groups of farmers whohave similar circumstances and for whom it islikely that the same recommendation will besuitable. In this manual such a group of farmers iscalled a recommendation domain. Recommendationdomains may be defined by agroclimatic and/or bysocioeconomic circumstances. The definition of therecommendation domain depends on the particularrecommendation. For example, a new variety may beappropriate for all farmers in a given geographical area,whereas a particular fertilizer recommendation may beappropriate only for farmers who follow a certainrotation pattern or whose fields have a certain type ofsoil. Thus the recommendation domain for varietywould be different from the recommendation domainfor fertilizer.

Recommendation domains are identified. defined. andredefined throughout the process of on-farm research.They may be tentatively described during the firstdiagnosis. Experimentation adds precision to thedefinition of domains. The final definition may not bedeveloped until the recommendation is ready to bepassed to farmers.

When interpreting agronomic data to make theirrecommendations, researchers must have a fairly clearidea of the group of farmers who will be able to use thisinformation. Researchers must consider not only theagroclimatic range over which the results will berelevant, but also whether such factors as differentmanagement practices or access to resources will beimportant in causing some farmers to interpret theresults differently from others.

For the purposes of this manual, it is important that theon-farm experiments be planted at locations that arerepresentative of the recommendation domain. Theeconomic analysis is done on the pooled data from agroup of locations of the same domain. The economicanalysis of results from a single location is not veryuseful because, first, researchers cannot makerecommendations for individual farmers, and second,one location will rarely provide sufficient agronomicdata to be extrapolated to a group of farmers. Thus all ofthe examples in this manual will represent data fromseveral locations of one recommendation domain.

Introduction to Basic Concepts

To make good recommendations for farmers,researchers must be able to evaluate alternativetechnologies from the farmers' point of view. Thepremises of this manual are:

1 Farmers are concerned with the benefits and costs ofparticular technologies.

2 They usually adopt innovations in a stepwisefashion.

3 They will consider the risks involved in adoptingnew practices.

These concerns will be treated in subsequent sections ofthe manual. Part Two describes the construction of apartial budget, which is used to calculate net benefits.Part Three presents the techniques of marginal analysis.This is a way of evaluating the changes from onetechnology to another by comparing the changes incosts and net benefits associated with each treatment.Part Four describes ways of dealing with the variabilitythat is characteristic of farmers' environments.Variability in results from location to location and from

8

9

year to year, and in the costs of the inputs and prices ofcrops, is of concern to farmers as they make productiondecisions. Part Five summarizes the first four sectionsand provides general gUidelines for reporting researchresults.

The following sections offer a brief overview ofthese topics.

The Partial Budget

Partial budgeting is a method of organizingexperimental data and information about the costs andbenefits of various alternative treatments. As anexample, consider the farmers who are trying to decidebetween their current practice of hand weeding and thealternative of applying herbicide. Suppose that someexperiments have been planted on farmers' fields, andthe results show that the current farmer practice ofhand weeding results in average yields of 2,000 kg/ha,while the use of herbicides gives an average yield of2,400 kg/ha.

Table 1.1. Example of a partial budget

Herbicide

Average yield (kg/hal 2.000 2,400Adjusted yield (kg/hal 1.800 2.160Gross field benefits (S/hal 3.600 4.320

Cost of herbicide (S/hal 0 500Cost of labor to apply

herbicide (S/hal 0 100Cost of labor for hand

weeding (S/hal 400 0

Total costs that vary (S/hal 400 600

Net benefits (S/hal 3.200 3.720

Table 1.1 shows a partial budget for this weed controlexperiment. There are two columns, representing thetwo treatments (hand weeding and herbicide). The firstline of the budget presents the average yield from alllocations in the recommendation domain for each of thetwo treatments. The second line is the adjusted yield.

Although the experiments were planted onrepresentative farmers' fields. researchers have judgedthat farmers using the same technologies would obtainyields 10% lower than those obtained by theresearchers. They have therefore adjusted the yieldsdownwards by 10% (yield adjustment will be discussedin Chapter 3).

The next line is the gross field benefits, which valuesthe adjusted yield for each treatment. To calculate thegross field benefits it is necessary to know the field priceof the crop. The field price is the value of one kilogramof the crop to the farmer. net of harvest costs that areproportional to yield. In this example the field price is$2/kg (i.e., 1,800 kg/ha x $2/kg = $3.600/ha).~

Farmers can now compare the gross benefits of eachtreatment, but they will want to take account of thedifferent costs as well. In considering the costsassociated with each treatment, the farmers need onlybe concerned by those costs that differ across thetreatments. or the costs that vary. Costs (such asplowing and planting costs) that do not differ acrosstreatments will be incurred regardless of whichtreatment is used. They do not affect the farmers'choices concerning weed control and can be ignored forthe purpose of this decision. The term "partial budget"is a reminder that not all production costs are includedin the budget-only those that are affected by thealternative treatments being considered.

In this case, the costs that vary are those associatedwith weed control. Table 1.2 shows how to calculatethese costs. Note that they are all calculated on a per-

Table 1.2. Calculation of costs that vary

2/ The use of the S sign in thismanual is not intended torepresent any particularcurrency, and severaldifferent currencies areassumed in the examplesthat follow, Additionalabbreviations used in thismanual are: hectare (hal,kUogram (kg), and liter (I).

10

Price of herbicideAmount usedCost of herbicide

Price of laborLabor to apply herbicideCost of labor to apply herbicide

Price of laborLabor for hand weedingCost of labor for hand weeding

$250/12 l/ha

$500/ha

$50/day2 days/ha$lOO/ha

$50/day8 days/ha$400/ha

11

hectare basis. The total costs that vary for eachtreatment is the sum of the individual costs that vary.In this example. the total costs that vary for the presentpractice of hand weeding is $400/ha, while the totalcosts that vary for the herbicide alternative is $600/ha.

The final line of the partial budget shows the netbenefits. This is calculated by subtracting the total coststhat vary from the gross field benefits. In the weedcontrol example, the net benefits from the use ofherbicide are $3,720/ha. while those for the currentpractice are $3,200/ha. Net benefits are not the samething as profit, because the partial budget does notinclude the other costs of production which are notrelevant to this particular decision. The computation oftotal costs of production is sometimes useful for otherpurposes. but will not be covered in this manual.

A partial budget, then, is a way of calculating the totalcosts that vary and the net benefits of each treatment inan on·farm experiment. The partial budget includes theaverage yields for each treatment. the adjusted yieldsand the gross field benefit (based on the field price ofthe crop). It also includes all the costs that vary for eachtreatment. The final two lines are the total costs thatvary and the net benefits.

Marginal Analysis

In the weed control example. the net benefits fromherbicide use are higher than those for hand weeding. Itmay appear that farmers would choose to adoptherbicides. but the choice is not obVious. becausefarmers will also want to consider the increase in costs.Although the calculation of net benefits accounts for thecosts that vary. it is necessary to compare the extra (ormarginal) costs with the extra (or marginal) net benefits.Higher net benefits may not be attractive if they requirevery much higher costs.

If the farmers in the example were to adopt herbicide. itwould require an extra investment of $200/ha, which isthe difference between the costs associated with the useof herbicide ($600) and the costs of their currentpractice ($400). This difference can then be compared tothe gain in net benefits. which is $520/ha ($3.720-$3,200).

In changing from their current weed control practice toa herbicide the farmers must make an extra investmentof $200/ha; in return, they will obtain extra benefits of

12

$520Iha. One way of assessing this change is to dividethe difference in net benefits by the difference in coststhat vary ($520/$200 = 2.6). For each $l/ha on averageinvested in herbicide, farmers recover their $1, plus anextra $2.6/ha in net benefits. This ratio is usuallyexpressed as a percentage (i.e., 260%) and is called themarginal rate of return.

The process of calculating the marginal rates of returnof alternative treatments, proceeding in steps from theleast costly treatment to the most costly, and deciding ifthey are acceptable to farmers, is called marginalanalysis.

Variability

In addition to being concerned about the net benefits ofalternative technologies and the marginal rates of returnin changing from one to another, farmers also take intoaccount the possible variability in results. Thisvariability can come from several sources, whichresearchers need to consider in makingrecommendations.

Experimental results will always vary somewhat fromlocation to location and from year to year. An agronomicassessment of the trial results will help researchersdecide whether the locations are really representative ofa single recommendation domain, and can therefore beanalyzed together, or whether the experimentallocations represent different domains. This type ofagronomic assessment helps refine domain definitionsand leads to more precisely targeted recommendations.

Another source of variability in experimental resultsderives from factors that are impossible to predict orcontrol, such as droughts, floods, or frosts. These arerisks that the farmers have to confront, and if theexperimental data reflect these risks, they should beincluded in the analysis.

Finally, farmers are also aware that their economicenvironment is not perfectly stable. Crop prices changefrom year to year, labor availability and costs maychange, and input prices are also subject to variation.Although such changes are difficult to predict withaccuracy, there are techniques that allow researchers toconsider their recommendations in view of possiblechanges in farmers' economic circumstances.

Part Two The Partial Budget

Chapter TwoCosts That Vary

Costs that vary

Opportunity cost

Field price (of an input)

13

The first step in doing an economic analysis of on-farmexperiments is to calculate the costs that vary for eachtreatment. Costs that vary are the costs (perhectare) of purchased inputs, labor, and machinerythat vary between experimental treatments.Farmers will want to evaluate all the changes that areinvolved in adopting a new practice. It is thereforeimportant to take into consideration all inputs that areaffected in any way by changing from one treatment toanother. These are the items associated with theexperimental variables. They may include purchasedinputs such as chemicals or seed, the amount or type oflabor, and the amount or type of machinery. Thesecalculations should be done before the experiment isplanted, as part of the planning process, to get an ideaof the costs of the various treatments that are beingconsidered for the experimental program.

In developing a partial budget, a common measure isneeded. It is of course not possible to add hours of laborto liters of herbicide and compare these with kilogramsof grain. The solution is to use the value of thesefactors, measured in currency units, as a commondenominator. This provides an estimate of the costs ofinvestment measured in a uniform manner. It does notnecessarily imply that farmers spend money for labor orreceive money for grain. Neither does it imply thatfarmers are concerned only with money. It is simply adevice to represent the process that farmers go throughwhen comparing the value of the things gained and thevalue of the things given up.

An important concept in these calculations is that ofopportunity cost. Not all costs in a partial budgetnecessarily represent the exchange of cash. In the caseof labor, for instance, farmers may do the workthemselves, rather than hire others to do it. Theopportunity cost can be defined as the value of anyresource in its best alternative use. Thus if farmerscould be earning money as laborers, rather thanworking on their own farms, the opportunity cost oftheir weeding is the net wage they would have earnedhad they chosen not to stay on the farm and weed. Theconcept of opportunity cost will be discussed at severalpoints in the following sections.

The field price of a variable input is the valuewhich must be given up to bring an extra unit ofinput into the field. The field price is expressed inunits of sale (e.g., $ per kilogram of seed, liter ofherbicide, day of labor, or hour of tractor time).

Field cost

14

The field cost is the field price multiplied by thequantity of the input needed for a given area. Thusfield costs are usually expressed in $ per hectare. If thefield price of herbicide is $1011, and if 3 lIha arerequired, then the field cost of the herbicide is $30/ha.In both cases, the emphasis is on the field, Le., what thefarmers pay to obtain the input and to transport it totheir farms. Such field prices may be quite differentfrom official prices.

Identifying Variable Inputs

To identify which inputs are affected by the alternativetreatments included in an experiment, researchers mustbe familiar with farmers' practices as well as thepractices used in the experiment. They must then askthemselves which operations change from treatment totreatment and make a list of these.

For example. consider an experiment in which twodifferent fungicides (A and B) are being tested, alongwith the farmers' practice of no fungicide application.There are thus three treatments in the experiment. Thelist of variable inputs is as follows:

• Fungicide A• Fungicide B• Labor to apply each fungicide• Labor to haul water for mixing with each fungicide• Rental of sprayer to apply each fungicide

This list includes purchased inputs (the fungicides),labor, and equipment (a sprayer). The following sectionsexplain how to calculate the costs for all of these.

Purchased Inputs

Purchased inputs include such items as seed, pesticides,fertilizer. and irrigation water. The best way to estimatethe field price of a purchased input is to go to the placewhere most of the farmers buy their inputs and checkthe retail price for the appropriate size of package. Forinstance, if farmers normally purchase their insecticidein I-kg packets in a rural market, that is the price thatshould be used, rather than the price of insecticide in25-kg sacks in the capital city.

In some situations, the farmers will be selecting seedfrom their previous crop, rather than buying seed. Thisseed is not without cost. The best way to estimate theopportunity field price of the farmers' own seed is to usethe price that farmers pay if they buy local seed, eitherfrom each other or at the market.

The next step is to find out how the farmers get theinput to the farm. Such inputs as insecticides andherbicides, which are not bulky, can be carried by thefarmers and transportation costs are probably notimportant. But for fertilizer and perhaps seed, this is notthe case. Usually the farmers must use a truck oran animal to get the input to the farm. If this is so, atransportation charge must be added to the retail price.As many farmers pay others to transport such items forthem, it is not difficult to learn what the normal chargesare. In general, it is best to be guided by the practicethat is followed by the majority of farmers in therecommendation domain.

For example, if a 50-kg sack of urea costs $375 in themarket, and it costs $25 to transport the 50 kg to thefarm, then the field price of urea is calculated as follows:

$375+$ 25

$400

cost of 50 kg urea in marketcost of transporting 50 kg to farm

field price of 50 kg urea

15

or $40050 kg = $8/kg, field price of urea

Very often fertilizer experiments, especially those in theearly stages of investigation, use single-nutrientfertilizers. The treatments are usually expressed interms of amounts of the nutrient (e.g. 50 kg N/ha or 40kg P20S/ha). In these cases, it is useful to go one stepfurther and calculate the field price of the nutrient. Thiscan be done by simply dividing the field price of thefertilizer by the proportion of nutrient in the fertilizer. Inthe case of urea, which is 46% nitrogen,

$8/kg urea------- = $17 .4/kg N, field price of N0.46 kg N/kg urea

The field cost of 50 kg N in a particular treatment usingurea would be 50 x $17.4, or $870/ha.

16

This should be done only when working with single­nutrient fertilizers. and it presumes that the field priceof nitrogen (for instance) is roughly the same in anynitrogen fertilizer available. If it is not, researchersshould of course be aware of this and take thesedifferences into account when considering whichfertilizer to experiment with and recommend.

A final point about purchased inputs is in order. Thisdiscussion has assumed that the inputs in theexperiments are available in local markets, or can bemade available. If this is not the case, then theeconomic analysis of experiments involving such inputsmay be of little immediate use to farmers. The resultsmay be used. however. to communicate to policymakersthe possible benefits of making a particular inputavailable.

Equipment and Machinery

Some experimental treatments may require the use ofequipment not reqUired by other treatments. It is thennecessary to estimate a field cost per hectare for the useof the equipment.

The easiest way to estimate the per-hectare field cost ofequipment is to use the average rental rate in the area.For example, if farmers rent their sprayers for $20/dayand if the sprayer can cover 2 ha in one day. then thefield cost may be taken as $IO/ha. When estimating thefield cost of tractor-drawn or animal-drawn implements,or small self-powered implements. the average rentalrate in the area can also be used. This is especiallyappropriate if most farmers are renting the implementsanyway. but even for farmers who own their equipment,the rental rate is a good estimate of the opportunity fieldcost. In certain cases a pro-rated cost per hectare can becalculated. using the retail price of the equipment andits useful lifetime, but this calculation involves factorssuch as repair costs, fuel costs, and the possibility thatthe equipment will have other uses on the farm. If a pro­rated field cost is to be calculated. it is best to consultan agricultural economist who is familiar with theequipment and costing techniques.

LaborIt is very important to take into consideration all of thechanges in labor implied by the different treatments inan experiment. Estimates of labor time should come

17

from conversations with farmers and perhaps directobservation in their fields. Information about labor usederived from the experimental plots is not very useful,however, if small plots are used in the experiments. Thebest way to get this information is to visit with severaldifferent farmers. Each will have an opinion as to thetime reqUired for a given operation, but a number closeto the average of these opinions will proVide a goodestimate. Not all farmers take the same amount of timefor a given task, so the estimates will only beapproximate. For new activities with which farmers arecompletely unfamiliar, some educated guesses will have tobe made until more reliable estimates can be developed.

If farmers hire labor for the operations in question, thefield price of labor is the local wage rate for day laborersin the recommendation domain, plus the value ofnonmonetary payments normally offered, such as mealsor drinks. This wage rate can be estimated by talking toseveral farmers. The field cost of labor for a particulartreatment is then the field price of labor multiplied bythe number of days per hectare reqUired.

When members of the farm family will do the work, it isnecessary to estimate the opportunity cost of familylabor. This is the value which is given up to do the workand thus represents a real cost. For example, if farmersmust take a day off from working in town to do extraweeding, they will give up a day's wages in town. Thisopportunity cost is just as real as paying a laborer to dothe work. And even if the farmer has nothing to do butsit in the shade, the opportunity cost is not zero, asmost people put some value on being able to sit in theshade rather than work in the sun.

The best place to start in estimating an opportunity fieldprice for family labor is the local wage rate (plusnonmonetary payments). It is not unusual to find therate higher during some periods of the year than others,and this must be taken into account.

It is sometimes difficult to estimate an opportunity costof family labor, especially if local labor markets arepoorly developed. Labor availability may varyseasonally, or across different types of farm households.Labor aVailability and labor bottlenecks are two of themost important types of diagnostic information that aidin selecting appropriate treatments for experiments andin defining recommendation domains. If labor is scarceat a particular time, extreme caution must be used inexperimenting with technologies that further increase

Total costs that vary

the labor demand at that time. In cases such as this. itis reasonable to set the opportunity cost of labor abovethe going wage rate. On the other hand. if additionallabor is to be used during a relatively slack period. anopportunity cost below the going wage rate might beappropriate. But in no case should the opportunity costof labor be set at zero.

In situations where most farm labor is provided by thefamily. and where the new technologies beingconsidered change the balance between cashexpenditures (Le.. for inputs) and labor. special caremust be taken in estimating labor costs. If a particulartreatment involves a large change in labor input.relatively small differences in the opportunity cost oflabor will have significant effects on the estimation ofthe cost of the treatment.

Total Costs That Vary

Once the variable inputs have been identified. their fieldprices determined. and the field costs calculated. thetotal costs that vary for each treatment can becalculated. The total costs that vary is the sum ofall the costs that vary for a particular treatment. Adescription of a weed control by seeding rate experimentis provided in Table 2.1; the calculation of costs thatvary is shown in Table 2.2; and the calculation of thetotal costs that vary is shown in Table 2.3.

Table 2.1. Weed control by seeding rate experiment (wheat)

Treatment

l~

234

cY Farmers' practice

Data

Weed control

No weed controlHerbicide (2 l/ha)No weed controlHerbiCide (2 l/ha)

Seeding rate

120 kg/ha120 kg/ha160 kg/ha160 kg/ha

18

Field price of seedField price of herbicideField price of laborField price of sprayerLabor to apply herbicideLabor to haul water

$20/kg$350/1$250/day (local wage rate)$75/day (rental rate)2 days/haOne laborer can haul 400 l/day(200 I water/ha are reqUired forthe herbicide)

Table 2.2. Calculation of costs that vary

Cost of seed Treatments 1 and 2: 120 kg/ha x $20/kg = $2,400/ha

Treatments 3 and 4: 160 kg/ha x $20/kg = $3,200/ha

Cost of herbicide Treatments 2 and 4: 2 l/ha x $350/1 = $700/ha

Cost of labor to apply herbicide Treatments 2 and 4: 2 days/ha x $250/day = $500/ha

Cost of labor to haul water Treatments 2 and 4: 200 I required x $250/da = $125/ha400 l/day y

Cost of sprayer Treatments 2 and 4: 2 days/ha x $75/day = $150/ha

Table 2.3. Total costs that vary for weed control by seeding rate experiment

Treatment

1 2 3 4

Seed ($/hal 2,400 2,400 3,200 3,200Herbicide ($/hal 0 700 0 700

Labor to apply herbicide ($/hal 0 500 0 500Labor to haul water ($/hal 0 125 0 125

Sprayer ($/hal 0 150 0 150

Total costs that vary ($/hal 2,400 3,875 3,200 4,675

The perceptive reader will have noticed that not all ofthe costs that vary have been treated in this chapter.There are two important exceptions. Costs associatedwith harvest and marketing are discussed in the nextchapter, where they are included in the field price of thecrop. Costs associated with obtaining working capital,such as interest rates, are discussed in Chapter 5.

19

Chapter ThreeGross Field Benefits,Net Benefits,and the PartialBudget

20

There are several steps involved in calculating thebenefits of the treatments in an on-farm experiment:

Step 1. Identify the locations that belong to the samerecommendation domain. The economicanalysis is done on the pooled results of anexperiment that has been planted in severallocations for one recommendation domain.

Step 2. Next, calculate the average yields across sitesfor each treatment. If the results of theseexperiments are agronomically consistent andunderstandable, do a statistical analysis of thepooled results. If there is no reasonableevidence of differences among treatment yields.researchers need only consider the differencesin costs among the treatments. But if there arereal yield differences. then researchers shouldcontinue with the partial budget.

Step 3. Adjust the average yields downwards. if it isbelieved that there are differences between theexperimental results and the yield farmersmight expect using the same treatment.

Step 4. Calculate a field price for the crop and multiplyby the adjusted yields to give the gross fieldbenefits for each treatment.

Step 5. Finally. subtract the total costs that vary fromthe gross field benefits to give the net benefits.With this calculation the partial budgetis complete.

Pooling the Results From theSame Recommendation Domain

The first line of a partial budget is the average yield foreach treatment for a particular experiment for alllocations for a recommendation domain. Recall that arecommendation domain is a group of farmers whosecircumstances are similar enough that they will all beeligible for the same recommendation. Tentative

21

identification of recommendation domains begins duringthe diagnostic and planning stages of on-farm research.This tentative identification is used for selectinglocations for planting experiments. The recommendationdomain for a fertilizer experiment, for example, mightbe defined in terms of farmers who plant the targetcrop, whose fields have certain types of soil, and whofollow a particular crop rotation. Locations for theexperiments are chosen to represent farmers with thoseparticular circumstances. Upon analyzing the results itmay be found that a factor not previously considered,such as the slope of the field, is responsible for differentresults between locations. In such a case, theexperiments from the tentative domain would not all becombined for economic analysis. Instead, they would bedivided into two domains (further defined by slope, inthis case), and two separate analyses would be carriedout. More detail on how and when to pool experimentalresults is presented in Chapter 7.

It should be noted here that although locations can beeliminated from analysis if it can be demonstrated thatthey do not belong to the recommendation domain inquestion, this does not hold for locations where trialswere severely damaged by drought, flooding, or otherenvironmental factors that are not predictable. Suchlocations must be included in the economic analysisbecause they are outcomes that farmers will experience,too. Further discussion of risk analysis is to be found inChapter 8.

Assessing ExperimentalResults Before Economic Analysis

Before doing an economic analysis of the pooled resultsof an experiment for a particular recommendationdomain, researchers must assess the experimental datato verify that the observed responses make sense froman agronomic standpoint. Researchers must also reviewthe statistical analysis of the experimental data.Performing an economic analysis on experimental datathat researchers do not understand, or do not haveconfidence in, is a misuse of the techniques presented inthis manual.

If statistical analysis of the results of an experimentindicates that there are no relevant differences betweentwo treatments, then the lower cost treatment would bepreferred. If researchers have evidence that treatment

yields are probably about the same. the gross benefitsfor these treatments will also be similar, and the lowestcost method of achieving those benefits should bechosen. If two methods of weed control give equivalentresults, for instance, the method with the lower coststhat vary should be chosen (for further experimentationor for recommendation) and no further economicanalysis is needed.

More details on the relation of statistical analysis toeconomic analysis are given in Chapter 7.

Average Yield

When the recommendation domain for a particularexperiment has been established and agronomic andstatistical assessments have indicated that it isworthwhile to proceed with a partial budget, the averageyields of each treatment are entered on the first line ofthe partial budget.

Table 3.1 shows the results from five locations in onerecommendation domain of the weed control by seedingrate experiment described in Tables 2.1-2.3. There weretwo replications at each location. Note that the resultsfrom location 5, which was affected by drought. areincluded in the average.Q/

Table 3.1. Yields (kg/hal for weed control by seeding rate experiment

Treatment1

No weed control120 kg seed/ha

(farmers' practiceI

ReplicationLocation 1 2 Avg.

',2.1 Note that the individualtreatment yields arereported to the nearest10 kg/ha. to reflect thereliability of the data. Itshould be remembered thatneither the average yieldsnor any of the results ofcalculations done with themcan be more precise thanthe original yield data onwhich they are based. Thusthe final digit reported inthe average yields is notsignificant and ismaintained in the partialbudget for convenienceonly.

22

1

23459/

Average yield

!Y Affected by drought

2,1802,8001,720

2,680530

2,2202,6401,8802,620

670

2,2002,7201,8002,650

600

1.994

Adjusted yield

The average yields for the four treatments are reportedon the first line of the partial budget (Table 3.2, p. 27).

Adjusted Yield

The next step is to consider adjusting the averageyields. The adjusted yield for a treatment is theaverage yield adjusted downward by a certainpercentage to reflect the difference between theexperimental yield and the yield farmers couldexpect from the same treatment. Experimentalyields, even from on-farm experiments underrepresentative conditions, are often higher than theyields that farmers could expect using the sametreatments. There are several reasons for this:

Management. If they manage the experimental1 variables, researchers can often be more precise and

sometimes more timely than farmers in operations suchas plant spacing, fertilizer application, or weed control.Further bias will be introduced if researchers managesome of the nonexperimental variables.

Plot size. Yields estimated from small plots often2 overestimate the yield of an entire field because of errors

in the measurement of the harvested area and because

in one recommendation domain

Treatment Treatment Treatment2 3 4

Herbicide (2 I/hal No weed control Herbicide (2 I/hal120 kg seed/ha 160 kg seed/ha 160 kg seed/ha

Replication Replication Replication

1 2 Avg. 1 2 Avg. 1 2 Avg.

3,030 2.570 2,800 2,440 2,180 2.310 3.200 3,060 3.1303,090 3,410 3.250 2.790 3.010 2.900 3,410 3.510 3,4602.200 2.180 2,190 1,820 1.680 1,750 2,410 2.230 2,3203.270 3.090 3.180 2.950 2.770 2.860 3,400 3,480 3,440

860 740 800 700 500 600 620 680 650

2,444 2,084 2,600

23

3

4

24

the small plots tend to be more uniform thanlarge fields.

Harvest date. Researchers often harvest a crop atphysiological maturity, whereas farmers may notharvest at the optimum time. Thus even when theyields of both researchers and farmers are adjusted to aconstant moisture content, the researchers' yield maybe higher, because of fewer losses to insects, birds,rodents, ear rots, or shattering.

Form of harvest. In some cases farmers' harvestmethods may lead to heavier losses than result fromresearchers' harvest methods. This might occur, forexample, if farmers harvest their fields by machine andresearchers carry out a more careful manual harvest.

Unless some adjustment is made for these factors, theexperimental yields will overestimate the returns thatfarmers are likely to get from a particular treatment.One way to estimate the adjustment reqUired is tocompare yields obtained in the experimental treatmentwhich represents farmers' practice with yields fromcarefully sampled check plots in the farmers' fields.Where this is not possible, it is necessary to review eachof the four factors discussed earlier and assign apercentage adjustment. As a general rule. totaladjustments between 5 and 30% are appropriate. Ayield adjustment of greater magnitude than 30% wouldindicate that the experimental conditions are verydifferent from those of the farmers, and that somechanges in experimental design or management mightbe in order. Many of these problems regarding yieldadjustment are eliminated if the farmers manage theexperiment. Decisions regarding experimentalmanagement will depend on several factors, but wherepossible the farmers should certainly manage thenonexperimental variables. As the experimentationmoves into later stages, farmers should also manage theexperimental variables.

In the case of the weed control by density experiment inwheat, researchers estimated that their methods ofseeding and of herbicide application were more precisethan those of the farmers, and so estimated a yieldadjustment of 10% due to management differences. Plotsize was also judged to be a factor, and a further 5%adjustment was suggested. Since the plots wereharvested at the same time as those of the farmers, noadjustment was needed to account for differences in

Field price (of output)

25

harvest date. However, the plots were harvested with asmall combine harvester, while the farmers used largermachines, and the difference in harvest loss was judgedto be about 5%. Thus the total yield adjustment for thisexperiment was estimated to be 20%. The second line ofthe partial budget (Table 3.2) thus adjusts the averageyields downwards by 20%. For instance, the averageyield for Treatment 1 is 1,994 kg/ha and the adjustedyield is 80% x 1,994 or 1,595 kg/ha.

It is obvious that this type of adjustment is not precise,nor does it pretend to be. The point is that it is muchbetter to estimate the effect of a factor than to ignore itcompletely. As researchers gain more experience in anarea they will have better estimates of the differencesbetween farmers' fields and the experiments, and yieldadjustments will become more accurate. The yieldadjustment, although approximate, should not be lookedupon as a factor to be applied mechanically. Each typeof experiment, each year, should be reviewed beforedeciding on an appropriate adjustment. If this is done,researchers will be able to make decisions about newtechnologies with a realistic appreciation of farmers'conditions.

Field Price of the Crop

The field price of the crop is defined as the valueto the farmer of an additional unit of production inthe field, prior to harvest. It is calculated bytaking the price that farmers receive (or canreceive) for the crop when they sell it, andsubtracting all costs associated with harvest andsale that are proportional to yield, that is, coststhat can be expressed per kilogram of crop.

The place to start is the sale price of the crop. This isestimated by finding out how the majority of thefarmers in the recommendation domain sell their crop,to whom they sell it. and under what conditions (suchas discounts for quality). Crop prices often varythroughout the year, but it is best to use the price atharvest time. It is the amount that the farmer actuallyreceives, rather than the official or market price of thecrop. that is of interest.

Next, subtract the costs of harvest and marketing thatare proportional to yield. These may include the costs ofharvesting, shelling, threshing, winnowing, bagging, andtransport to the point of sale. These costs have to be

26

estimated on a per-kilogram basis. In the case ofharvesting or shelling, for instance, this may requirecollecting data on the average amount of labornecessary to harvest a field of defined size and yield, orshell a given quantity of grain. Again, these may becash costs or opportunity costs.

• If farmers sell maize to traders for $6.00/kg,

• and they incur harvesting costs of $0.30/kg,

• shelling costs of $0.20/kg,

• and transport costs of $0.20/kg,

• then the field price of an additional unit of maize is:$6.00 - ($0.30 + 0.20 + 0.20) == $5.30Ikg.

It is important to account for these costs because theyare proportional to the yield; the higher the yield of aparticular treatment, the higher the cost (per hectare) ofharvesting, shelling, and transport. That is, the cost ofharvesting, shelling, and transporting 200 kg is almostexactly twice the cost of performing the same activitiesfor a harvest of 100 kg. As these costs will differ acrosstreatments (because the treatment yields are different),they must be included in the analysis. It is convenientto treat these costs separately from the costs that vary(described in Chapter 2) because, although they do varyacross treatments, they are incurred at the time ofharvest and thus do not enter into the marginal analysisof the returns to resources invested. That is, farmershave to wait perhaps five months to recover theirinvestment in purchased inputs, but only a few days torecover harvest-related costs.

If there are costs associated with harvest or sale that donot vary with the yield, then these should not beincluded in the field price, nor in the partial budget. Inthe example of the weed control by s<.:eding rateexperiment, the farmers sell their wheat in town for$9/kg. The harvesting is done by combine, and theoperators charge on a per-hectare basis (regardless ofyield), so harvest cost is not included in the calculationof field price.

• There is a bagging charge of $O.lO/kg,

Gross field benefits

• transport charge of $0.50/kg,

• and a market tax of $0.40/kg,

• so the field price of the wheat is:$9.00 - ($0.10 + 0.50 + 0.40) = $8.00/kg.

The gross fJ.eld beneflts for each treatment iscalculated by multiplying the field price by theadjusted yield. Thus the gross field benefits forTreatment 1 is 1,595 kg/ha x $8/kg = $12,760/ha.

Although the field price is based on the sale price of thecrop, the concept can normally be used even insituations where farmers do not produce enough fortheir own needs. An alternative would be to calculate anopportunity field price for the crop, based on the moneyprice the farm family would have to pay to acquire anadditional unit of the product for consumption (see note5, p. 35). But under most conditions use of the fieldprice is adequate for estimating the value of the productto farmers, even when the product is not sold, and thisis the approach that will be followed in this manual.

Table 3.2. Partial budget, weed control by seeding rate experiment

Treatment

1 2 3 4

Average yield (kg/ha) 1,994 2,444 2,084 2,600Adjusted yield (kg/ha) 1,595 1,955 1,667 2,080

Gross field benefits ($/ha) 12,760 15,640 13,336 16,640Cost of seed ($/ha) 2,400 2,400 3,200 3,200

Cost of herbicide (S/ha) 0 700 0 700

Cost of labor to applyherbicide (S/ha) 0 500 0 500

Cost of labor to haulwater (S/ha) 0 125 0 125

Cost of sprayerrental (S/ha) 0 150 0 150

Total costs that vary (S/ha) 2,400 3,875 3,200 4,675

Net benefits (S/ha) 10,360 11,765 10,136 11.965

27

Net benefits

1./ It is important to rememberthat the net benefits do nothave greater precision thanthe original yield data(which in this case werereported to three significantdigits in Table 3.1). Whenusing a calculator forfurther operations (such ascalculating the marginalrates of return). it isconvenient to take thenumbers as they appear inthe partial budget. but forfinal reporting researchersmay wish to round the netbenefits (e.g.. SIl.800instead of S 11.765 inTreatment 2).

28

Net Benefits

Table 3.2 is a partial budget for the weed control byseeding rate experiment. The final line of the partialbudget is the net beneflts. This is calculated bysubtracting the total costs that vary from thegross field benefits for each treatment.1/

Including All Gross Benefits in the Partial Budget

The examples discussed above have assumed that asingle product is the only thing of value to the farmersfrom their fields. This is often not the case. In manyregions crop residues have considerable fodder value, forinstance. The procedure for estimating the gross fieldbenefit for fodder is exactly the same as that forestimating the value of grain. First estimate production(by treatment) and adjust the average yields. Thencalculate a field price. Of course "harvesting" becomes"collecting," "shelling" becomes "baling," and so forth.It is important to consider each activity that isperformed (for instance, is maize fodder chopped?).Multiplying the field price of the fodder by the adjustedfodder yield gives the gross field benefit from fodder,and this should be added to the gross field benefitfrom grain.

Another important example is that of intercropping. Ifthe majority of farmers in the recommendation domainintercrop, then experiments should reflect that practice.(Intercropping experiments may of course includeindividual treatments with a single crop as well, if thatis considered a possible alternative.) It may be that theexperimental variables affect only one crop, but iffarmers intercrop maize and beans, for instance, then afertilizer experiment on maize should include beans, ora disease control experiment on beans should be plantedwith maize. The yields of both crops should bemeasured, since treatments may have a direct orindirect effect on the associated crop. The partial budgetwould then have two average yields, two adjustedyields, and two gross field benefits.

The total costs that vary would be subtracted from thesum of the two gross field benefits to give the netbenefits. Table 3.3 gives an example.

Table 3.3. Partial budget for an experiment on bean density and phosphorus application in a

maize-bean intercrop

Treatment

1 2 3 4

Bean density (plants/hal 40.000 60,000 80,000 80,000

Phosphorus rate (kg P205/hal 30 30 30 60

Average bean yield (kg/hal 650 830 890 980

Average maize yield (kg/hal 2,300 2,020 1,700 1,790

Adjusted bean yield (kg/hal 553 706 757 833Adjusted maize yield (kg/hal 1,955 1,717 1,445 1.522

Gross field benefits.beans ($/hal 17,143 21,886 23,467 25.823

Gross field benefits.maize ($/hal 14.663 12,878 10,838 11,415

Total gross field benefits ($/hal 31.806 34.764 34.305 37,238

Cost of bean seed ($/hal 900 1,350 1,800 1,800

Cost of labor for plan tingbeans ($/hal 450 675 900 900

Cost of fertilizer ($/hal 1,050 1,050 1,050 2,100

Total costs that vary ($/hal 2,400 3.075 3,750 4,800

Net benefits ($/hal 29,406 31.689 30.555 32,438

29

Part Three Marginal Analysis

Chapter FourThe Net Benefit Curveand the MarginalRate of Return

Dominance analysis

30

In the previous chapter a partial budget was developedto calculate the total costs that vary and the net benefitsfor each treatment of an experiment. This chapterdescribes a method for comparing the costs that varywith the net benefits. This comparison is important tofarmers because they are interested in seeing theincrease in costs required to obtain a given increase innet benefits. The best way of illustrating thiscomparison is to plot the net benefits of each treatmentversus the total costs that vary. The net benefit curve(actually. a series of lines) connects these points. Thenet benefit curve is useful for visualizing the changes incosts and benefits in passing from one treatment to thetreatment of next highest cost. The net benefit curvealso clarifies the reasoning behind the calculation ofmarginal rates of return. which compare the incrementsin costs and benefits between such pairs of treatments.Before proceeding with the net benefit curve and thecalculation of marginal rates of return, however. aninitial examination of the costs and benefits of eachtreatment. called dominance analysis, may serve toeliminate some of the treatments from furtherconsideration and thereby simplify the analysis.

Dominance Analysis

Table 4.1 lists the total costs that vary and the netbenefits for each of the treatments in the weed controlby seeding rate experiment from the previous chapter.

Notice that the treatments are listed in order ofincreasing total costs that vary. The net benefits alsoincrease. except in the case of Treatment 3, where netbenefits are lower than in Treatment 1. No farmer wouldchoose Treatment 3 in comparison with Treatment 1,because Treatment 3 has higher costs that vary. butlower net benefits. Such a treatment is called adominated treatment (marked with a "D" in Table 4.1),and can be eliminated from further consideration. Adominance analysis Is thus carried out by firstlisting the treatments In order of Increasing coststhat vary. Any treatment that has net benefitsthat are less than or equal to those of a treatmentwith lower costs that vary is dominated.

This example illustrates that to improve farmers'incomes it is important to pay attention to net benefits.rather than yields. Notice (from Table 3.2) that theyields of Treatment 3 are higher than those ofTreatment 1. but the dominance analysis shows that the

value of the increase in yield is not enough tocompensate for the increase in costs. Farmers would bebetter off using the lower seed rate, provided they arenot using herbicide.

Table 4.1. Dominance analysis, weed control by seeding rateexperiment

WeedTreatment control

1 None3 None2 Herbicide4 Herbicide

Net Benefit Curve

Seedingrate

(kg/hal

120160120160

Total coststhat vary

($/hal

2,4003,2003.8754.675

Netbenefits($/hal

10.36010.136 D11,76511.965

Net benefit curve

31

The dominance analysis has eliminated one treatmentfrom consideration because of its low net benefits. but ithas not provided a firm recommendation. It is possibleto say that Treatment 1 is better than Treatment 3, butto compare Treatment 1 with Treatments 2 and 4further analysis will have to be done. For that analysis.a net benefit curve is useful.

Figure 4.1 is the net benefit curve for the weed controlby seeding rate experiment. In a net benefit curveeach of the treatments is plotted according to itsnet benefits and total costs that vary. Thealternatives that are not dominated are connectedwith lines. The dominated alternative (Treatment 3)has been graphed as well. to show that it falls below thenet benefit curve. Because only nondominatedtreatments are included in the net benefit curve, itsslope will always be positive.

Marginal Rate of Return

The net benefit curve in Figure 4.1 shows the relationbetween the costs that vary and net benefits for thethree nondominated treatments. The slope of the lineconnecting Treatment 1 to Treatment 2 is steeper thanthe slope of the line connecting Treatment 2 toTreatment 4.

4,5003.000 3,500 4.000

Total costs that vary ($/ha)

Figure 4.1. Net benefit curve. weed control by seedingrate experiment

11,000

10,500

11.500

Net benefits($/ha)

12.000

The purpose of marginal analysis is to reveal just howthe net benefits from an investment increase as theamount invested increases. That is, if farmers invest$1,475 in herbicide and its application, they will recoverthe $1,475 (remember, the costs that vary have alreadybeen subtracted from the gross field benefits), plus anadditional $1,405.

Marginal rate of return

An easier way of expressing this relationship is bycalculating the marginal rate of return, which isthe marginal net benefit (i.e .. the change in netbenefits) divided by the marginal cost (i.e .. thechange in costs), expressed as a percentage. In thiscase, the marginal rate of return for changing fromTreatment 1 to Treatment 2 is:

$11.765 - $10,360 = $1,405 = 0.95 = 95%$ 3,875 - $ 2,400 $1,475

This means that for every $1.00 invested in herbicideand its application, farmers can expect to recover the$1.00, and obtain an additional $0.95.

32

The next step is to calculate the marginal rate of returnfor going from Treatment 2 (not 1) to Treatment 4.

$11,965 - $11,765 = $200 = 025 = 259f$ 4,675 - $ 3,875 $800' 0

Thus for farmers who use herbicide and plant at a rateof 120 kg seed/ha, investing in the higher seed ratewould give a marginal rate of return of 25%; for every$1.00 invested in the higher seed rate, they will recoverthe $1.00 and an additional $0.25.

The two marginal rates of return confirm the visualevidence of the net benefit curve; the second rate ofreturn is lower than the first. It is possible to do amarginal ar:talysis without reference to the net benefitcurve itself (Table 4.2). Note that the marginal rates ofreturn appear in between the two treatments. It makesno sense to speak of the marginal rate of return of aparticular treatment; rather, the marginal rate of returnis a characteristic of the change [rom one treatment toanother. Because dominated treatments are not includedin the marginal analysis, the marginal rate of return willalways be positive.

Table 4.2. Marginal analysis, weed control by seeding rate experiment

2.400J,----__ 1,4753,875 J 8004,675

1O,360JI----1,40511,765--,11,965JI---- 200

95%

25%

Marginal rateof return

Marginalnet benefits

($/ha)

Netbenefits

($/ha)

Marginalcosts($/ha)

Coststhat vary

($/ha)

124

The marginal rate of return indicates what farmers canexpect to gain, on the average, in return for theirinvestment when they decide to change from onepractice (or set of practices) to another. In the presentexample, adopting herbicide implies a 95% rate ofreturn, and then increasing seed rate implies a further25 %. As the analysis in this example is based on onlyfive experiments in one year, it is likely that theconclusions will be used to select promising treatmentsfor further testing, rather than for immediate farmerrecommendations. Nevertheless, a decision cannot betaken regarding these treatments without knOWing whatrate of return is acceptable to the farmers. Is 95% highenough? What about 25%? The next chapter explainshow to estimate a minimum rate of return.

33

Chapter FiveThe MinimumAcceptableRate of Return

Working capital

Cost of capital

34

In order to make farmer recommendations from amarginal analysis, it is necessary to estimate theminimum rate of return acceptable to farmers in therecommendation domain. If farmers are asked to makean additional investment in their farming operations,they are going to consider the cost of the moneyinvested. This is a cost that has not been considered inprevious chapters. Because of the critical importance ofcapital availability it is treated separately. Workingcapital is the value of inputs (purchased or owned)allocated to an enterprise with the expectation of areturn at a later point in time. The cost of workingcapital (which in this manual will simply bereferred to as the cost of capital) is the benefitgiven up by the farmer by tying up the workingcapital in the enterprise for a period of time. Thismay be a direct cost. as in the case of a person whoborrows money to buy fertilizer and must pay aninterest charge on it. Or it may be an opportunity cost,the earnings of which are given up by not puttingmoney, or an input already owned. to its bestalternative use.

It is also necessary to estimate the level of additionalreturns, beyond the cost of capital, that will satisfyfarmers that their investment is worthwhile. After all,farmers are not going to borrow money at 20% interestto invest in a technology that returns only 20% andleaves them with nothing to show for their investment.In estimating a minimum acceptable rate of return,something must be added to the cost of capital to repaythe farmers for the time and effort spent in learning tomanage a new technology.

There are several ways of estimating a minimumacceptable rate of return (or. more simply, a minimumrate of return).

A First Approximation ofthe Minimum Rate of Return

Experience and empirical evidence have shown that forthe majority of situations the minimum rate of returnacceptable to farmers will be between 50 and 100%. Ifthe technology is new to the farmers (e.g.. chemicalweed control where farmers currently practice handweeding) and requires that they learn some new skills. a100% minimum rate of return is a reasonable estimate.If a change in technologies offers a rate of return above

5/ In cases where theopportunity field price isused to calculate gross fieldbenefits. the estimation ofthe minimum rate of returnshould be based on theperiod from planting to thetime when the householdmakes Its principalpurchase of the commodity.This Is generally much laterthan harvest. and thus theminimum rate of return inthese cases will be higherthan when the field price isused to calculate gross fieidbenefils.

35

100% (which is equivalent to a "2 to I" return, ofwhich farmers often speak), it would seem safe torecommend it in most cases.

If the technology simply represents an adjustment incurrent farmer practice (such as a different fertilizer ratefor farmers that are already using fertilizer), then aminimum rate of return as low as 50% may beacceptable. Unless capital is very easily available andlearning costs are very low. it is unlikely that a rate ofreturn below 50% will be accepted.

This range of 50 to 100% is rather crude but it shouldalways be remembered that the other agronomic andeconomic data used in the analysis will be estimates orapproximations as well. This range should serve as auseful gUide in most cases for the minimum rate ofreturn acceptable to farmers. It is important to note thatthis range represents an estimate for crop cycles of fourto five months. If the crop cycle is longer, the minimumrate of return will be correspondingly higherQ/. In areaswhere the inflation rate is very high. this range shouldbe adjusted upward by the rate of inflation over theperiod of the crop cycle as well. (For more informationon inflation, see pp.71-72.)

The Informal Capital Market

An alternative way of estimating the minimum rate ofreturn is through an examination of the informal capitalmarket. In many areas. farmers do not have access toinstitutional credit. They must either use their owncapital, or take advantage of the informal capitalmarket, such as village moneylenders. The interest ratescharged in this informal sector provide a way ofbeginning to estimate a minimum rate of return.Informal conversations with several farmers who arepart of the recommendation domain should giveresearchers a good idea of the local rates of interest. "Ifyou need cash to purchase something for the farm. towhom do you go?" and "How much does this personcharge for the loan of the money?" are examples ofrelevant questions.

If it turns out that local moneylenders charge 10% permonth. for instance, then a cost of capital for fivemonths would be 50%. To estimate the minimum rateof return in this case, an additional amount would haveto be added to represent what farmers expect will repaytheir effort in learning about and using the new

technology. This extra amount may be approximated bydoubling the cost of capital (unless the technologyrepresents a very simple adjustment in practices). Thusin this example. the minimum rate of return would beestimated to be 100%. Again. it should be emphasizedthat this is simply a way of deriving a rough estimate ofthe level of returns that farmers will require.

The Formal Capital MarketIt is also possible to estimate a minimum rate of returnusing information from the formal capital market. Iffarmers have access to loans through priv~te orgovernment banks. cooperatives, or other agenciesserving the agricultural sector, then the rates of interestcharged by these institutions can be used to estimate acost of capital. But this calculation is relevant only if themajority of the farmers in fact have access toinstitutional credit. If they do not, then they willprobably face a cost of capital different from that offeredthrough relatively cheap institutional credit. In somecases. it may be that farmers with otherwise similarcircumstances must be divided into two groupsaccording to their access to one or the other type ofcredit. These two groups of farmers would face differentminimum rates of return and may well represent twoseparate recommendation domains.

In other cases, institutional credit may be available tofarmers, but only for certain crops or in the form ofrigidly defined credit packages. If institutional credit isnot likely to be available for the recommendations beingconsidered. then the cost of capital in these creditprograms is not relevant to the estimation of aminimum rate of return. This is another example of howon-farm research can proVide information topolicymakers, in this case by interacting with creditinstitutions to assure that their services are directed tofarmers in as efficient a manner as possible.

If farmers do have access to institutional credit, the costof capital can be estimated by using the rate of interestcharged over the agricultural cycle. That is, the rate ofinterest should cover the period from when the farmersreceive credit (cash or inputs) to when they sell theirharvest and repay the loan. In addition, it is necessaryto include all charges connected with the loan. Thereare often service charges. insurance fees, or even

37

farmers' personal expenses for things like transport totown to arrange the loan. that must be included in theestimate of the cost of capital.

Once the cost of capital on the formal market has beencalculated, an estimate of the minimum rate of returncan be obtained by doubling this rate. This will prOVidea rough idea of the rate of return that farmers will findacceptable if they are to take a loan to invest in anew technology.

Summary

It is necessary to estimate a minimum rate of returnacceptable to the farmers of a recommendation domain.In most cases it will not be possible to provide an exactfigure, but experience has shown that the figure willrarely be below 50%. even for technologies thatrepresent only simple adjustments in fanner practice.and is often in the neighborhood of 100%, especiallywhen the proposed practice is new to farmers. If thecrop cycle is longer than four to five months, theseminimum rates will be correspondingly higher. Wherefarmers have access to credit, either through theinformal or formal capital markets. it is possible toestimate a cost of capital (or an opportunity cost ofcapital) and use it to estimate a minimum rate of return.But even in these cases. it must be remembered that thefigure will be apprOximate. The next chapter explainshow to use the estimates of the minimum rate of returnto judge which changes in technology will be acceptableto farmers.

Chapter SixUsing MarginalAnalysis to MakeRecommendations

Marginal analysis

38

Chapter 4 demonstrated how to develop a net benefitcurve and calculate the marginal rate of return betweenadjacent pairs of treatments. Chapter 5 discussedmethods for estimating the minimum rate of returnacceptable to farmers. The purpose of this chapter is todescribe marginal analysis, which is the process ofcalculating marginal rates of return betweentreatments, proceeding in steps from a lower costtreatment to that of next higher cost, andcomparing those rates of return to the minimumrate of return acceptable to farmers. It should beemphasized again that this type of analysis is usefulboth for making recommendations to farmers, wherethere is sufficient experimental evidence, and for helpingselect treatments for further experimentation. Threeexamples of marginal analysis follow.

Weed Control by Seeding Rate Experiment

It might be best to start by returning to the example ofthe weed control by seeding rate experimentsummarized in Figure 4.1. After the dominance analysisthere were only three treatments left for consideration,and the marginal rates of return were calculated. IfTreatment 1 represents the farmers' practice,will farmers be willing to adopt Treatment 2 orTreatment 4?

Farmers should be willing to change from onetreatment to another if the marginal rate of returnof that change is greater than the minimum rate ofreturn. In this case, if the minimum rate of return were100%, the farmers would probably not be willing tochange from their practice of no weed control,represented by Treatment 1. to the use of herbicide.represented by Treatment 2, because the marginal rateof return (95%) is below the minimum. If the minimumrate of return were 50%, then farmers would be willingto change to Treatment 2. Only if the minimum rate ofreturn were below 25% (which is unlikely) would thefarmers be willing to change from Treatment 2 toTreatment 4. As long as the marginal rate of returnbetween two treatments exceeds the minimumacceptable rate of return, the change from one treatmentto the next should be attractive to farmers. If themarginal rate of return falls below the minimum, on theother hand. the change from one treatment to anotherwill not be acceptable.

Fertilizer Experiment

Figure 6.1 shows the results of a nitrogen experiment inmaize. Table 6.1 gives details on the experimentaldesign and costs that vary. The yield data are theaverage of 20 locations from three years ofexperimentation. Table 6.2 is a partial budget for theexperiment. Figure 6.2 shows the net benefit curve andTable 6.3 shows the marginal analysis (one of thetreatments is dominated).

For the recommendation domain where theseexperiments were planted. researchers estimated thatthe minimum rate of return for the crop cycle was100%. With 20 experiments over three years,researchers felt that they were ready to make a nitrogenrecommendation to farmers, who are currently not usingnitrogen fertilizer on their crop. What should be therecommendation? Or, in other words, if farmers areconsidering investing in nitrogen fertilizer and the laborto apply it, what should be the recommended levelof investment?

Figure 6.1. Vields from nitrogen experimentYield

(kg/ha)

3,500

3,000

2,500

2,000

40* 80* * 120** 160**

39

*Kg N/ha

single application. * * = split application

Table 6.1. Nitrogen experiment data

TreatmentNitrogen(kg/hal

Number ofapplications of N

Average yield (kg/halfor 20 locations

over 3 years

lW 02 403 804 1205 160

!JJ Farmers' practice

Data

o1

222

2,2222,8673,2563,4443,544

Field price of N = $0.625/kgField price of maize = $0.20/kgCost of one fertilizer application $5.00/haYield adjustment = 10%Minimum rate of return = 100%

Table 6.2. Partial budget. nitrogen experiment

Trefltment

1 2 3 4 5o kg 40 kg 80 kg 120 kg 160 kgN/ha N/ha N/ha N/ha N/ha

Average yield (kg/ha) 2,222 2,867 3,256 3,444 3,544Adjusted yield (kg/ha) 2,000 2,580 2,930 3,100 3,190Gross field benefits ($/ha) 400 516 586 620 638Cost of nitrogen ($/ha) 0 25 50 75 100Cost of labor ($/ha) 0 5 10 10 10Total costs that vary ($/ha) 0 30 60 85 110Net benefits ($/ha) 400 486 526 535 528

This analysis should always be done in a stepwisemanner, passing from the treatment with the lowestcosts that vary to the next. If the marginal rate of returnof the change from the. first to the second treatment isequal to or above the minimum rate of return, then the

40

Figure 6.2. Net benefit curve. nitrogen experiment

10020 40 60 80Total costs that vary ($/ha)

o

520

540

Net benefits

($/ha)13illl!lllllli,1111111111111'1111111

Table 6.3. Marginal analysis. nitrogen experiment

Total costs Netthat vary benefits Marginal rate

Treatment ($/hal ($/hal of return

1 o kg N/ha 0 $400

2 40 kg N/ha $ 30 $486287%

2 80 kg N/ha $ 60 $526133%36%

4 120 kg N/ha $ 85 $535X

5 160 kg N/ha $110 $528 D~

gI Treatment 5 is dominated

next comparison can be made. between the second andthird treatments (not between the first and third). Thesecomparisons continue (Le.. increasing the level ofinvestment) until the marginal rate of return falls belowthe minimum rate of return. If the slope of the netbenefit curve continues to fall. then the analysis can be

41

42

stopped at the last treatment that has an acceptable rateof return compared to the treatment of next lowest cost.If the net benefit curve is irregular. then further analysismust be done. (See the next example. p.43).

In the nitrogen experiment. the marginal rate of returnof the change from 0 kg N/ha to 40 kg N/ha is 287%.well above the 100% minimum. The marginal rate ofreturn from 40 kg N/ha to 80 kg N/ha is 133%. alsoabove 100%. But the marginal rate of return between 80kg N/ha and 120 kg N/ha is only 36%. So of thetreatments in the experiment. 80 kg N/ha would be thebest recommendation for farmers.

There are a couple of things to notice about thisconclusion. First. the recommendation is not(necessarily) based on the highest marginal rate ofreturn. For farmers who use no nitrogen. investing in 40kg N/ha gives a very high rate of return. but if farmersstopped there. they would miss the opportunity forfurther earnings. at an attractive rate of return. byinvesting in an additional 40 kg of nitrogen. Farmerswill continue to invest as long as the returns to eachextra unit invested (measured by the marginal rate ofreturn) are higher than the cost of the extra unitinvested (measured by the minimum acceptable rateof return).

The second thing to notice is that the recommendationis not (necessarily) the treatment with highest netbenefits (120 kg N/ha). If instead of a step-by-stepmarginal analysis. an average analysis is carried out.comparing 0 kg N/ha with 120 kg N/ha. the rate ofreturn looks attractive (Le.. (535-400)/(85-0) = 159%).but this is misleading. The average rate of return of159% hides the fact that most of the benefits werealready earned from lower levels of investment. Thisaverage rate of return lumps together the profitable andthe unprofitable segments of the net benefit curve. Themarginal analysis indicates acceptable rates of return upto 80 kg N/ha. If the farmers are to apply 120 kg N/ha.the analysis shows they would only get a marginal rateof return of 36% on their investment of the last $25. Itis likely that they would be willing to invest their moneyin nitrogen up to 80 kg N/ha. and then ask if there isnot some other way of investing that final $25 (a littleextra weeding. fencing for animals. etc.) that would givea better rate of return than 36%.

Yield(kg/ha)

In summary, the recommendation is not necessarily thetreatment with the highest marginal rate of returncompared to that of next lowest cost, nor the treatmentwith the highest net benefit, nor the treatment with thehighest yield. The identification of a recommendationrequires a careful marginal analysis using anappropriate minimum rate of return.

Tillage Experiment

This example illustrates some additional aspects ofmarginal analysis and the selection of recommendations.Figure 6.3 presents yield data from a tillage experimentin wheat. Table 6.4 gives details of the design and thecosts that vary. The yield data are the average of sixlocations from one year of experiments. Table 6.5 showsthe partial budget. Figure 6.4 shows the net benefitcurve and Table 6.6 shows the marginal analysis.

Figure 6.3. Yields from tillage experiment

---==----======~

4.400i-'iiiiiiiiiiiiiiiiiiiiiiiiii

4.200

4,000"~======-

3,800 .......:==

3,600

3,400 'i...iiiiiiiiiiii":

1

43

2 3

Treatment4

Table 6.4. TIllage experiment data

1~/ None

2 None3 Chisel4 Mold board

!Y Farmers' practice

TreatmentType of

plowNumber ofcultivations

2o22

Seeding method

By handZero-till planterBy handBy hand

Average yield (kg/ha)for 6 locations

3,8004,0804,3004,470

Data

Tillage costs:CultivatorChisel plowMold board plow

Zero-till planter

$7/ha$16/ha$22/ha

$20/ha

Cost of seeding by handField price of wheatYield adjustmentMinimum rate of return

$2/ha$0.08/kg

20%80%

Table 6.5. Partial budget. tillage experiment

Treatment

1 2 3 4

Average yield (kg/ha) 3,800 4,080 4,300 4,470

Adjusted yield (kg/ha) 3,040 3,264 3,440 3,576

Gross field benefits ($/ha) 243 261 275 286

Cost of plowing ($/ha) 0 0 16 22

Cost of cultivation ($/ha) 14 0 14 14Cost of seeding ($/ha) 2 0 2 2

Cost of zero-till seeding ($/ha) 0 20 0 0

Total costs that vary ($/ha) 16 20 32 38

Net benefits ($/ha) 227 241 243 248

44

40, i :20 25 30

Total costs that vary ($/ha)

Figure 6.4. Net benefit curve. tillage experiment

230

235

2402451~i~li1E~

Net benefits

($/ha)l~i~lllIllIr

Table 6.6. Marginal analysis. tillage experiment

Total costs thatTreatment vary I$/ha)

1 16

2 203 32

4 38

Net benefitsI$/ha)

227

241

243248

Marginal rate ofreturn

350%

17%}83% 39%

First. it should be noted that this tillage experiment isdifferent from the nitrogen experiment in that it testsfour distinct treatments. rather than the continuousincrease of one factor. It is impossible to use 80 kg ofnitrogen without using 40 kg of nitrogen. but using onetillage method does not require first using a lower costmethod. There are four different options. arranged onthe net benefit curve in order of increasing costs. Themarginal analysis is simply a way of examining various

45

46

alternatives for tillage (in this case). The comparisonsare made. as always. in a stepwise manner between onealternative and the next. in order of increasing costs.until an acceptable recommendation is identified.

Second. the situation is a bit different from the previousexample in that only six locations from one year areavailable for analysis. Thus the analysis will be used tohelp plan further experiments, rather than to makefarmer recommendations.

Finally. the shape of the net benefit curve is differentfrom the previous example. The marginal rate of returnin going from Treatment 1 to Treatment 2 is 350%. wellabove the minimum. Therefore Treatment 2 is certainlya worthwhile alternative to the farmers' practice. Next.the marginal rate of return in going from Treatment 2 toTreatment 3 is 17%. and below the minimum.Treatment 3 can therefore be eliminated fromconsideration. But the marginal rate of return betweenTreatments 3 and 4 is 83%. and above the minimumrate of return of 80%. In such cases as this. where themarginal rate of return between two treatments fallsbelow the minimum. but the following marginal rate ofreturn is above the minimum. it is necessary toeliminate the treatment(s) that are unacceptable andrecalculate a new marginal rate of return. In thisexample. it is necessary to calculate a marginal rate ofreturn between Treatment 2 and Treatment 4. Theresult is 39% (248-241 = 39%) • which is below the

38-20minimum rate of return. Thus Treatment 4 is alsorejected. If this last marginal rate of return had beenabove 80%. however. Treatment 4 would have been thebest treatment.

In this case researchers should continue to experimentwith Treatment 2 (the zero-till planter), which seems tobe a promising alternative to the farmers' practice oftwo cultivations before seeding. Treatments 3 and 4 givehigher yields. but their costs are such that they do notprovide an acceptable rate of return. Researchers mustdecide if there is sufficient evidence to eliminate thesetreatments from future experimentation. or if anotheryear of testing is worthwhile.

§I For the purposes of thismanual the term "residual"is used in a special way. toindicate the differencebetween the net benefitsand the cost of theinvestment. The readershould note that the termhas other meanings. both ineconomics and in otherfields.

Analysis Using Residuals'-----

The conclusions of a marginal analysis can be checkedby using the concept of "residuals. "§/ Residuals (as theterm is used here) are calculated by subtracting thereturn that farmers require (the minimum rate of returnmultiplied by the total costs that vary) from the netbenefits. Table 6.7 illustrates this method, using thedata from the nitrogen experiment (Table 6.3).

Table 6.7. Analysis of nitrogen experiment using residuals

(1 ) (2) (3)Return

Total costs Net required Residualthat vary benefits [100%x(1)J [(2) - (3))

Treatment ($/ha) ($/ha) ($/ha) ($/ha)

1 o kg N/ha 0 400 0 4002 40 kg N/ha 30 486 30 4563 80 kg N/ha 60 526 60 466~

4 120 kg N/ha 85 535 85 450

fil Maximum residual

The treatments are listed. as usual, in order of totalcosts that vary. Column 1 gives the total costs that varyand column 2 gives the net benefits. Column 3 is theminimum acceptable rate of return multiplied by thecosts that vary, and represents the return that fannerswould require from their investment in order to changetheir practice. For instance, if 40 kg N/ha has costs thatvary of S30/ha, and if the minimum rate of return is100%. this means that farmers would ask for returns ofat least an additional S30/ha before investing in 40 kgN/ha. Finally. the residual (column 4) is the differencebetween net benefits (column 2) and the return thatfarmers require (column 3). Of course this residual is notthe profit, and it is the comparison between theresiduals, rather than their absolute value, that isof interest.

Farmers will be interested in the treatment with thehighest residual. In this case, the treatment with thehighest residual is 80 kg N/ha, which is the sameconclusion that was reached in the previous analysis.Stopping at 40 kg N/ha denies the farmers thepossibility to earn more money per hectare. Going on to120 kg N/ha implies a loss, after accounting for thereturn that farmers require.

47

48

Residuals can also be used to check the conclusions ofthe marginal analysis of the tillage experiment (Table6.6). Table 6.8 shows the results; Treatment 2 is the onewith the highest residual.

Table 6.8. Analysis of tillage experiment using residuals

(1) (2) (3) (4)Return

Total costs Net required Residualthat vary benefits 180%x(1)) ((2) - (3))

Treatment I$/ha) ($/ha) ($/ha) ($/ha)

1 16 227 13 2142 20 241 16 225~

3 32 243 26 2174 38 248 30 218

~ Maximum residual

This method of calculating and comparing residuals willalways give the same conclusion as the graphicalmethod of marginal analysis shown earlier. The methodof using reSiduals, however, requires an exact figure forthe minimum rate of return, whereas the graphicalmethod allows comparison of the marginal rates ofreturn with various assumptions about the minimumrate of return. Thus it is advisable to use the graphicalmethod first and then, if necessary, check theconclusions with respect to a particular minimum rateof return by calculating residuals.

1

2

SOME QUESTIONS ABOUT MARGINAL ANALYSIS

Is marginal analysis the "last word" for makinga recommendation?Marginal analysis is an important step in assessing theresults of on-farm experiments before makingrecommendations. But agronomic interpretation andstatistical analysis are also part of the assessment, aswell as farmer evaluation. As researchers conduct on­farm experiments, they must constantly solicit farmers'opinions and reactions. Alternatives that seem to bepromising both agronomically and economically mayhave other drawbacks that only farmers can identify. Tothe extent possible, screening treatments forcompatibility with the farming system should take placebefore experiments are planted. But farmer assessmentof the experiments is also essential. It is the farmerswho have the last word.

How precise is the marginal rate of returnas a criterion?It is important to bear in mind that the calculation ofthe marginal rate of return is based on yield estimatesderived from agronomic experiments and on estimatesof various costs, often opportunity costs. Furthermore,the marginal rate of return is compared to a minimumrate of return which is only an approximation of theinvestment goals of the farmers. Discretion and goodjudgment must always play an important part ininterpreting these rates and in makingrecommendations. If the marginal rate of return iscomfortably above the minimum, the chances are goodthat the change will be accepted. If it is close to theminimum rate of return then caution must be exercised.In no case can one apply a mechanical rule torecommend a change that is a few percentage pointsabove the minimum rate, or reject it if it is a few pointsbelow. Making farmer recommendations requires athorough knowledge of the research area and theproblems that farmers face, a dedication to goodagronomic research, and the ability to learn frompreVious experience. Marginal analysis is a powerful toolin this process, but it must be seen as only a part of theresearch strategy.

3 Can the marginal rate of return be interpreted ifthe change in costs that vary is small?Certain experiments, such as those that look at differentvarieties or perhaps modest changes in seeding rate,

49

involve changes in costs that may be quite small. If theyield differences are at all substantial. the resultingmarginal rate of return can be very large. sometimes inthe thousands of percent. In these cases the marginalrate of return is of little use in comparing treatments.Thus it is usually not worthwhile calculating marginalrates of return for variety experiments. unless there aresignificant differences in cost between varieties (e.g..local maize variety versus a hybrid), or in the marketvalue of the varieties (e.g.. because of consumerpreference).

4 Is it really possible to make recommendations.using marginal analysis. without considering all thecosts of production?Remember that the starting point in on-farm research isthe assumption that it is much better to considerrelatively small improvements in farmers' practices.rather than propose large-scale changes. The idea isthus to ask what changes can be made in the presentsystem. and to compare the change in benefits with thechange in costs. Because the focus is on the differencesbetween two treatments. rather than their absolutevalues. costs that do not vary between treatments willnot affect the calculation of the marginal rate of return.Table 6.9 shows two cases. both using the same yieldsand costs that vary. For the partial budget. the marginalrate of return is calculated in the usual way. Thecomplete budget includes all of the costs of production;they are of course constant ($300/ha) for eachtreatment. When the marginal rate of return is

Table 6.9. Marginal analysis using a partial budget and a complete budget

Partial budget 1 2 Complete budget 2

Gross field benefits Gross field benefits(S/ha) 500 650 (S/hal 500 650

Total costs that vary Total costs that vary

IS/ha) 100 200 (S/hal 100 200

Net benefits (S/hal 400 450 Total of costs that do notvary (S/hal 300 300

Total costs (S/hal 400 500

Net benefits (S/ha) 100 150

Marginal rate 450 - 400 50% Marginal rate 150 - 100 50%of return 200 - 100 of return 500 - 400

50

5

Netbenefits

($/ha)

51

calculated using benefits and total costs, the result isthe same.

Is the correct strategy always to considersmall changes in farmers' practices?Experience has shown that farmers are much morelikely to adopt new practices in small steps rather thanin complete packages. But in following this strategy itshould be realized that farmers can (and do) eventuallyadopt a new set of practices over a period of severalyears of testing. The complexity of the individual stepsdepends on the nature of the agronomic interactionsamong the elements being tested and on the resourcesavailable to farmers.

It is often possible to take advantage of this sequentialadoption pattern in making recommendations. Initialsteps may be intermediate between farmers' practiceand the recommendation that would be selected bymarginal analysis. Figure 6.5 is the net benefit curve for

Figure 6.5. Net benefit curve, weed control by fertilizerexperiment

Total costs that vary ($/ha)

a weed control by fertilizer experiment. The curveshows that a combination of improved weed control andfertilization should be the recommendation.

Nevertheless, it is possible to first promote anintermediate recommendation of improved weed controlonly and then add fertilization later. The curve allowsresearchers to trace out an efficient set of technologiesfor recommendation as farmers increase expenditurelevels. In this case, further analysis would indicate thatadopting fertilizer first, without improved weed control,would not be a worthwhile option.

More complex changes, such as the introduction of newcrops or cropping patterns, are of course possible aswell. But such changes require extremely carefulplanning and analysis which are beyond the scope ofthis manual.

6 What is the difference between a marginalanalysis and a continuous analysis of data?Agronomists often estimate response functions forfactors such as nutrients, and economists use similarcontinuous functions to select economic optima. Yet themethodology of this manual uses a marginal analysis forsets of discrete alternatives. There are three reasons foremphasizing the latter method. First, marginal analysis,using discrete points, can be used for any type ofexperimentation, whereas continuous analysis is onlyapplicable to factors that vary continuously, such asfertilizer rates or seed rates. Second, the computationalskills and facilities necessary for estimating responsefunctions are not always available. Finally, greatprecision is not required for farmer recommendations(e.g., for fertilizer levels) because farmers will adjustthem to their individual conditions.

A continuous economic analysis may be very useful incertain situations, however. But if it is done, it requiresthe same degree of care in estimating the benefits andcosts that farmers face that has been emphasized in thismanual for constructing a partial budget and conductingmarginal analysis. The sophisticated analyses that areoften done with unrealistic assumptions about farmers'yields, field prices, or minimum rate of return do notgive useful conclusions.

7 Does the marginal analysis assumethat capital is the only scarce factor for farmers?In the marginal analysis, all factors are expressed inmonetary units. This does not necessarily mean thatfarmers think of all costs and benefits in monetaryterms, or that cash is necessarily the limiting factor.Marginal analysis may be used, for instance. in anexperiment that compares treatments which differ onlyin the amount of (unpaid) family labor utilized on a cropwhich is not sold. To decide whether extra amounts oflabor would be effectively invested to produce extraamounts of the crop, opportunity costs and prices canbe assigned and the comparison made.

Nevertheless. in cases where family labor is thepredominant source of labor, and experimentaltreatments involve significant changes in labor use. caremust be taken in valuing labor. If, for instance, achange from one treatment to another implies areduction in family labor and an increase in cashexpenditure. a modest increase in total costs that varymay in fact represent a significant increase in cashoutlay (balanced to some extent by a reduction in labor"costs"). In cases where family labor is a particularlyimportant factor in farmer decision making regardingnew technologies. a careful analysis must beundertaken. This is complicated by the fact that theopportunity cost of labor is sometimes difficult toestimate. Different members of the household (men.women. children) will likely have different opportunitycosts of labor. and the time of the year (slack season.peak season) will also affect the estimate.

One possibility is to do a sensitivity analysis (Chapter 9).which involves doing several marginal analyses usingdifferent estimates of the opportWlity cost of labor.Another technique involves estimating the returns tolabor for the treatments and comparing the marginalreturns to labor between two treatments with variousestimates of the opportunity cost of labor. This is areminder that there are often alternative analyticaltechniques, beyond the scope of this manual, whichmay be useful in making decisions about theappropriateness of a particular technology.

8 Can the concept of marginalanalysis be used for planning experiments?It is common to consider a change in farmers' practiceby doing a quick calculation of how much additionalyield would be needed to pay for the extra costs of the

53

new practice. If an extra 100 kg of fertilizer costs$1,000. and wheat is selling for $5/kg. then the estimatemight be that the farmers would need an extra 200 kgof wheat ($1.000/$5) in order to "repay the fertilizer."However, there are three errors in this kind ofcalculation.

The first error is in using market prices for fertilizer andwheat, rather than field prices. The second is notincluding the labor or machinery costs associated withthe use of fertilizer. The third is in not including theminimum rate of return. The following formula correctsthose errors, and provides a useful way for helping toconsider practices that are proposed forexperimentation.

6.Y = 6.TCV(l + M)p

where 6.Y = minimum change in yield required6.TCV = change in total costs that vary

P = field price of productM = minimum rate of return (expressed

as a decimal fraction)

In the example just mentioned. if the additional fertilizerplus the labor to apply it is worth $1,200. the field priceof wheat is $4/kg. and the minimum rate of return is50%. then:

6.Y $1,200 (1 + 0.5)$4

450 kg of wheat

Thus. given current prices. the minimum yield increaserequired by farmers from the addition of an extra100 kg of fertilizer is 450 kg of wheat, not the 200 kg inthe original calculation. The use of this type ofcalculation before designing an experiment helps ensurethat the treatments include an economically realisticrange of levels.

9 Can marginal analysis be usedwhen yields are variable or prices change?Yields in agronomic experiments are usually quitevariable. and prices often change. Methods foraccommodating this kind of variability to marginalanalysis are discussed in Chapters 7, 8, and 9.

Part Four Variability

Chapter SevenPreparingExperimental Resultsfor Economic Analysis:RecommendationDomains andStatistical Analysis

55

Marginal analysis for a particular experiment should bedone on the pooled results from at least several locationsover one or more years. To prepare the experimentalresults for this type of analysis, several steps must betaken. First, researchers must review the purpose of theexperiment in order to decide whether the results of theanalysis are to be used for making recommendations forfarmers or for gUiding further research. Second, a reviewof results from the different locations will indicatewhether all of the locations belong to the samerecommendation domain and can therefore be analyzedtogether. Finally, a combination of agronomic judgmentand statistical analysis will lead to a decision regardingthe yield differences among treatments in theexperiment. If researchers have little confidence thatthere are real differences in yields, then the total coststhat vary of each treatment can be compared; thetreatment with the lowest costs will generally bepreferred. If, on the other hand, researchers believe thatthe differences observed represent real differencesamong treatments, then a marginal analysis shouldbe done.

Reviewing the Purpose of the Experiment

Each experimental variable in an experiment has apurpose, and researchers should review the objectives ofthe experiment before thinking about an economicanalysis. Some experimental variables are of anexploratory nature; they are meant to proVide answersregarding response (e.g., is there a response tophosphorus?) or to elucidate particular productionconstraints that have been observed (e.g., is the lowtillering observed in the wheat crop due to a nutrientdeficiency or to the variety?). These variables are meantto prOVide information that can be used in specifyingproduction problems and designing solutions for them.The treatments in these exploratory experiments arechosen to detect the possibility of responses, and thusneed not be designed to represent economically viablesolutions to a particular problem. Researchers must bearthis in mind when considering the economic analysis ofexperiments with this type of exploratory variable. If theexperimental results prOVide clear evidence that aparticular production problem exists, the economicanalysis may help to select possible solutions for testing.If a high level of an insecticide in an exploratoryexperiment prOVided evidence of a response, but if the

56

marginal analysis then showed an unacceptable rate ofreturn, researchers would want to examine lower levelsof insecticide or less expensive insect control methods insubsequent experimentation.

Other experimental treatments test possible solutions towell-defined production problems. The solutions willhave been selected for testing not only because theypromise economically acceptable returns. but becausethey are compatible with the farming system and do notrepresent special risks to farmers. When there are yielddifferences among treatments in these cases. themarginal analysis should be more rigorous. because arecommendation may be made to farmers.

The marginal analysis should be done on the pooledresults of a number of locations, usually over more thanone year. No strict rules can be given here, but thenumber of locations should be sufficient to giveresearchers confidence that the results fairly representthe conditions faced by farmers in the recommendationdomain. A very rough rule of thumb might be to includeat least 20 experimental locations (in relativelyhomogeneous environments) over two years for eachrecommendation domain. The exact number of test sitesreqUired will depend on the Variability (across sites'andacross years) in the recommendation domain and on thetechnology being tested. For instance. fertilizerrecommendations usually require a fairly large numberof locations to adequately sample the range of responseby soil type. rotation, and so forth. Insect controlrecommendations may require several years of evidenceto sample year-to-year Variability in insect populations,especially in the case of routine preventive treatments.

Once recommendations are derived they are oftenpresented to fanners through demonstrations, whichmay involve one or more large plots shOWing variousalternatives next to a similar plot with the farmers'practice. As a way of follOWing up on therecommendation the results of these demonstrationplots should also be subjected to an economic analysis.preferably as part of the demonstration.

57

Tentative Recommendation Domains

Whether the experiments are of an exploratory nature orare testing possible solutions. they should be planted inlocations that represent the tentative definition of therecommendation domain. Recall that a recommendationdomain is a group of farmers whose circumstances aresimilar enough that members of the group are eligiblefor the same recommendation.

An example may help. In a particular research areathere is experimental evidence of a response to nitrogenin maize. Farmers currently use no fertilizer. and anexperiment is designed to test various levels of nitrogen.Most of the farmers plant maize under rainfedconditions. although a few have access to irrigation.Because the response to nitrogen may differ underrainfed and irrigated conditions. and because of thesmall number of farmers with irrigation. only farmerswith rainfed fields are considered. (If there were morefarmers with irrigation. experiments might be plantedwith them as well. but they would almost certainly be aseparate recommendation domain.) Most of the farmerswith rainfed fields have land with sandy to sandy-loamsoils. Locations are chosen to represent this range of soiltypes. and careful note is taken in the field book of thesoil type at each location. The tentative definition of therecommendation domain includes the range of soiltypes, but the experimental results may distinguishseparate domains. Nonexperimental variables. such asvariety. planting date. and weed control are left in thehands of the farmers. A certain range in these practicesis present in the recommendation domain. and theactual practices at each location are noted in the fieldbook. The researchers do their best to reject locationsthat represent very unusual practices or conditions(such as a few farmers who plant a special maize varietyto sell as green maize.)

The tentative definition of the recommendation domainfor the fertilizer experiment is thus: "All farmers in thearea who plant maize under rainfed conditions on sandyto sandy-loam soils." This definition allows for somevariability in conditions and practices. and the selectionof experimental sites tries to represent this range. butavoids obvious extremes.

1

Notice that the recommendation domain is defined forthe particular experimental variable. A differentexperimental variable (say, a disease-resistant variety)might be tested in a domain of a different definition. Inthis case, the variety might be tested on both irrigatedand rainfed fields, if no difference in its diseaseresistance capacity were expected.

Reviewing Experimental Results

The results of each experiment at each location in thetentative recommendation domain must be reviewed.Inconsistencies in results between locations can be dueto one of three causes:

Redefinition of the recommendation domain. In theabove example, soil type was being considered as apossible means of subdividing the recommendationdomain. If the responses are very different at locationswith sandy soils and those with sandy-loam soils, thenthere may be two separate recommendation domains(and two separate economic analyses). Or it may be thatan unexpected characteristic is of importance. Suppose,in this same example, that some farmers plant a maize­maize rotation, while others rotate their maize withfallow. If the responses to nitrogen are different on thesetwo types of fields, the original recommendation domainmay be refined (by eliminating the rotation thatrepresents a minority of the farmers) or divided (byrotation, if both rotations are of importance in the area).

The important point is that researchers must have aclear and consistent definition of the recommendationdomain whose experiments will be submitted toeconomic analysis. Domain definitions are reviewed andrefined during the experimental process. As the numberof possible defining characteristics for domains is greaterthan the number of locations to be planted, carefulselection of experimental locations is important. Theroutine collection of information adequate to describeeach location (e.g. elevation, soil, cropping history.management practices) is a most important activity,without which across-location interpretation isimpossible.

58

2

3

59

Improper experimental management. At times theexperimental results at a location may differ from theothers because of problems in experimentalmanagement. This may include errors by theresearchers (such as applying the wrong dosage of achemical), or factors related to the farmer (such as acow destroying part of the experiment. or the farmerfailing to weed because of a misunderstanding). In suchcases the location can be eliminated from the analysisand the researchers will gain a bit more experience-inthe management of chemicals, in locating experimentswhere there is little chance of animal damage. or incarefully discussing with farmers their responsibilities inthe management of an experiment. Part of experimentalmanagement includes the selection of locations. Iflocations have to be eliminated because they havecharacteristics well outside the normal range of therecommendation domain (such as very late plantingdates) this too is an indication of the necessity toimprove experimental management.

Unexplained or unpredictable sources of variation.After eliminating locations from the analysis becausethey do not represent the recommendation domain. andeliminating sites where the management of theexperiment is responsible for unrepresentative results.there may still be considerable variation in the resultsfrom the remaining locations. This may be due tofactors that are not understood (and may be the focus offurther agronomic investigation and/or discussion withfarmers). Or it may be due to factors that are understoodbut not predictable. and hence not eligible for defining arecommendation domain, like drought or frost. Thesesites must be included in the economic analysis. unlessresearchers are able to identify particular areas wherethe factor is more likely to occur. It may be. forinstance, that the research area can be divided intomore and less drought-prone domains. But if drought (orfrost or insect attack) cannot be associated withparticular areas. then the results of the affectedlocations must enter the analysis. More will be saidabout treating these risk factors in Chapter 8. but it isimportant to emphasize that locations that have beenaffected. or even abandoned, because of these factorsmust be included in the marginal analysis.

60

Statistical Analysis

In Chapter 3 it was pointed out that the economicanalysis of an experiment should be done only afterreviewing the agronomic assessment and statisticalanalysis. If after reviewing the statistical analysisresearchers do not have confidence that there are realdifferences among treatments, then they need to takeanother look at the experiment. If the averagedifferences among treatments are large relative to theyields obtained by farmers (e.g.. 5-10% or more ofaverage farmer yields), but there is insufficient evidencethat these differences are real, then researchers maywant to review the design or management of theexperiment and perhaps repeat it the next cycle. If thedifferences among treatments are small in relation tofarmers' yields. and researchers have no confidence thatthe differences are real, then they need consider onlythe differences in costs among treatments and choosethe one with lowest costs.

Cases where no significant yield differences exist and nomarginal analysis is required are not necessarily trivial.If experimentation leads to recommendation of apractice that lowers the costs of production whilemaintaining yields. the gains in productivity of farmerresources are as legitimate as those from a higheryielding (and higher cost) treatment. One commonexample is that of substituting some form of reducedtillage for mechanical tillage. This often results inconsiderable cost saVings. although yields may notbe affected.

In experiments with factorial designs. an examination ofthe statistical and agronomic analyses will help pointthe way to the most appropriate type of economicanalysis. For example. in an experiment with twofactors, one factor may be responsible for yielddifferences although the second factor is not (and thereis no interaction between them). In that case. the yieldsfor levels of the first factor should be the average foreach level over all levels of the second factor. Such acase occurs in a nitrogen by tillage experiment in whichthere is a response to nitrogen, but not to tillage (Table7.1). The tillage method to be chosen for furtherexperimentation is the one that costs the least. Thepartial budget for such an experiment will then have

Table 7.1. Yield data for a nitrogen by tillage experiment

1 502 503 1004 100

Average yield: 50 kg N/ha100 kg N/ha

TillagemethodTreatment

Nitrogen(kg/ha)

Average yield(kg/ha)

"A" 2.560"B" 2,300"A" 3,120"B" 3.200

2.430 kg/ha3,160 kg/ha

Average yield: tillage method "A"tillage method "B"

2,840 kg/ha2.750 kg/ha

61

only two columns. corresponding to the two nitrogenlevels (50 kg/ha and 100 kg/hal. The yields for the twonitrogen levels will be the average yields across tillagetreatments (to take advantage of all the data aVailable.which should give a better estimate of real differences inyields between nitrogen levels). The first line of thepartial budget ("Average yield") will thus have 2,430and 3,160 kg/ha. The costs that vary will include thoseassociated with the change in nitrogen level (fertilizer,application costs), but not those associated with tillage.The marginal analysis of the partial budget will examinethe marginal rate of return of changing from onenitrogen level to another.

The economic analysis of factorial experiment isconcerned only with factors that exhibit responses orare involved in interactions. Therefore the interpretationof experiments including several factors is oftensimplified because some factors may be dropped fromthe analysis. In the example above, for instance, tillagewas not included in the analysis. But if there had beenan interaction between tillage and nitrogen, the partialbudget would have had four columns (with all possiblecombinations of tillage and nitrogen) and the costs thatvary would have reflected both factors.

62

In the early stages of on-farm experimentation there areoften experiments with a large number of treatments (12to 15 or more) examining several variables. Thestatistical analysis of such experiments may be quitecomplex, and its relation to an economic analysis at firstsight may be unclear. The point to remember is that thepurpose of those experiments is to characterize asqUickly as possible the responses and the interactions ofseveral factors. Once that is accomplished, a smallnumber of possible solutions can be tested. If the resultsof such an exploratory experiment are agronomicallyclear (and the statistical analysis can only help inmaking this decision), then the next year's experimentswill certainly be simpler, and a marginal analysis willhelp to select a reasonable range of treatments for thoseexperiments. If the results are not clear agronomically,then further exploratory work is needed, and there isless that a marginal analysis can contribute to theselection of treatments for future experiments.

Chapter EightVariability in Yields:Minimum ReturnsAnalysis

63

Assigning experimental locations to differentrecommendation domains and reviewing themanagement of the experiments (Chapter 7) helpaccount for some of the variability in experimentalyields. After doing this, however, some variability willcertainly remain, and farmers and researchers will takethis into account when making decisions aboutalternative practices. Some variability in theperformance of particular treatments will beunexplained, whereas some may be due to identifiablefactors such as drought, frost. or flooding. In either case,farmers will want to know how this variability mightaffect their welfare, and what undesirable outcomes arepossible if they adopt a recommendation. One methodfor analyzing experimental data in this way is known asminimum returns analysis.

Dealing with Risk in On-Farm Research

Recall that the objective of an on-farm research programis to improve the productivity of farmers' resources.Besides improving the production of target crops oranimals. this may also include lowering the costs ofproduction or increasing the stability of production. Thelatter is an important factor for many farmers, whosepractices often reflect attempts to reduce the risks offailure. Common examples of such practices includestaggering planting dates to minimize the risk of losingan entire crop to drought. or investing extra labor todouble over the maize plants before harvest in areaswhere there are strong winds.

Risk has three important implications for an on-farmresearch program. First, new technologies that areproposed for testing should be compatible with farmers'practices to reduce risk. Before proposing a technologythat relies on a uniform planting date, for instance,researchers should take account of farmers' rationale forstaggered planting dates. Technologies that do not takeaccount of farmers' attempts to reduce risk have littlechance of being adopted.

The second implication is that the risks faced byfarmers may suggest opportunities for developingrecommendations to help stabilize farm production.Drought risk may be reduced with moistureconservation techniques, and losses from high windsmay be reduced with shorter varieties. Thus in settingpriorities for an experimental program, researchers

64

should include the possibility of testing alternatives thatmay not necessarily increase average benefits, butinstead help to reduce their year-to-year variability.

The third implication is that researchers will want to becareful in evaluating how new recommendations modifythe risks currently borne by the farmers in arecommendation domain. The amount that farmers arewilling to give up (in terms of average net benefits) toreduce the effects of an uncertain environment is ameasure of their degree of risk aversion. The degree offarmers' risk aversion may depend on several factors,but in general it can be said that most farmers indeveloping countries are moderately averse to risks. It isnot easy to specify the degree of risk aversion, but it issomething that should be considered when proposingnew recommendations.

Risk and Data From On-Farm Experiments

The source of risk is often thought of as beingsusceptible to quantification. Thus it is possible to saythat the probability of less than 400 mm of rainfall inthe grOWing season is 0.2 (Le., one year in five). Ifresearchers have information about the probability ofoccurrence for a particular event, then those data maybe used in interpreting experimental results. If, forinstance, it is known that there is a drought on theaverage of one year in five, causing a certain percentageof crop loss, that information can be factored into ananalysis of the results of the on-farm experiments,whether or not they were conducted during a droughtyear. But this type of precise data is not usuallyencountered, and researchers need a more useful way oflooking at the variability in their own experimental data.Even if the source of variability is well specified (e.g.,midseason drought), probabilities may not be available.Often the variability observed in experimental resultsand in farmers' fields is due to several sources. Thus theminimum returns analysis presented here is not, strictlyspeaking, a method of risk analysis, but rather a way ofassessing the variability due to unpredictable and attimes unexplained causes.

65

The Farmers' Point of View

Before minimum returns analysis is done to look atvariability the way that farmers do, it is useful toconsider how in fact farmers approach this problem.

First, recall that the marginal analysis is based on theaverage yields from a number of locations. If a proposedrecommendation gives an average yield of 3,000 kg/ha,it is certain that it will have yielded more than 3,000kg/ha in some locations and less in others. If thefarmers' practice yields an average of 2,000 kglha, it toowill exhibit some variation. And if the marginal analysisindicates that the proposed recommendation has anacceptable marginal rate of return, when compared tothe farmers' practice, it is a rate of return based onthese average yields. Minimum returns analysis will notlook at averages, but rather at the results fromindividual sites. Looking at across-location and across­year variability is one way of estimating the risks forfarmers associated with the proposed recommendation.The careful definition of recommendation domainsattempts to eliminate across-location variability as muchas possible. Across-year variability, on the other hand, isestimated here based on the results of only two or threeyears, and tends to underestimate the year-to-yearvariability that farmers face. Nevertheless, a carefulminimum returns analysis is a useful way of examiningthe variability associated with different technologicalal ternatives.

Second, note that farmers are more interested invariability in benefits than variability in yields. Aminimum returns analysis looks at variability innet benefits.

If the results of a set of on-farm experiments show thattwo treatments have the same average net benefits, butone treatment's results are more variable than theother's, it is likely that farmers will prefer the treatmentthat is more consistent, rather than the one thatsometimes gives very high net benefits but at othertimes gives very low net benefits.

But variability per se is not the only factor that farmerswill take into account when deciding among treatments.If one treatment always gives higher net benefits than

Minimum returnsanalysis

another treatment, it may not matter if the first exhibitshigher variability than the second. As long as marginalanalysis shows that it gives an acceptable rate of return,and farmers are assured that even in the worst cases itgives higher net benefits than the alternative. thenfarmers will be interested in adopting it.

The most difficult decisions must be taken when theaverage net benefits for one treatment are higher thanthose for another. but in some locations the net benefitsare lower than those of the alternative. The marginalanalysis (on average results) shows the treatment to beacceptable. but there are some individual cases wherethe benefits are lower than those of the alternativetreatment. Should the farmers choose the treatment thatis better on average. or the one that offers less chance oflow net benefits? It is here that a minimum returnsanalysis is most helpful.

Prerequisites for a Minimum Returns Analysis

A minimum returns analysis is a way of screening datafrom on-farm experiments in order to give farmers (andresearchers) additional information about the variabilityin returns implicit in a proposed recommendation incomparison with the farmers' practice. A minlmumreturns analysis compares the average of thelowest net benefits for each nondominatedtreatment. For the analysis to be relevant, severalprerequisites must be met:

1 The marginal analysis must have been done on alllocations for a given experiment and for all years. Itshould include all locations deemed to belong to therecommendation domain, including locations with poorresults or those that have been abandoned. A marginalanalysis done only on locations with "good" results willnot be of much use to farmers. At times it is tempting toremove a particularly poor location from the analysis. Iften locations were planted in the recommendationdomain, and one location had poor results because offrost damage, the analysis of the remaining nine willgive farmers an idea of what returns they can expect ifthere is no frost. This may not be very usefulinformation. If nine locations were damaged by frost, noone would propose analyzing only the single good one!Thus minimum returns analysis assumes that alllocations have been included in the marginal analysisdone preViously.

66

2

3

4

67

A minimum returns analysis should be done only onexperimental treatments that are being considered forrecommendation. That may include not only thefarmers' practice and the treatment that has beenjudged acceptable on the average by marginal analysis.but also other nondominated treatments that mayprovide alternatives if the tentative recommendationproves unsatisfactory.

Minimum returns analysis presumes that researchershave tried to explain the reasons for the variability theyobserve, rather than assuming it is simply bad luck. Themore precise an idea of the sources of observedvariability. the more useful the information from theminimum returns analysis will be for farmers.

Minimum returns analysis is most useful whenrecommendations are being considered. Although it doesnot pretend to be mathematically precise. it does try toassess the effects of variability, and this is bestestimated from a large number of results. Minimumreturns analysis is most relevant when done on theresults of at least 20 locations from at least two years.The results should be from enough locations and yearsto fairly represent the variability that farmers in therecommendation domain are likely to face.

Minimum Returns Analysis

For simplicity. the steps in the minimum returnsanalysis will be illustrated for a comparison betweenonly two treatments. Table 8.1 lists the yield data from20 locations over three years of the "0 kg nitrogen"(farmers' practice) and "80 kg nitrogen" treatments in afertilizer experiment. The 80 kg N/ha treatment gives.on the average. higher yields than the 0 kg N/ha.although there is considerable variability for bothtreatments. The marginal analysis of the average yielddata showed 80 kg N/ha gives an acceptable rate ofreturn (see Table 6.3).

68

Table 8.1. Yields by location for Treatments 0 kg Nand 80 kg N

Yield (kg/halLocation o kg N 80 kg N

1 2,450 3,9702 2,840 3,9303 2,130 1,8704 2,170 3,720• • •• • •• • •

20 2,570 1,780

Average of20 locations 2.222 3.256

The first step is to calculate the net benefits at each oneof the locations for each one of the treatments. This isnot as time consuming as it sounds. In the case of the80 kg N treatment, the necessary calculations areshown below:

Net benefits =: (Y x A x P)-TCV,

where

Y =: yield at one locationA =: I-the yield adjustmentP =: field price of crop

TCV =: total costs that vary for the treatment

If A =: 0.90, P = $0.20/kg, TCV = $60/ha

then the net benefits for treatment 80 kg N for eachlocation will be:

(Y x 0.9 x $0.20) - ($60)

or 0.18 Y - 60.

Because Treatment 0 kg N has no costs that vary, theformula for calculating the net benefits is even easier(0.18 Y). The net benefits for each location are shown inTable 8.2.

To do the minimum returns analysis, select the(approximately) 25% lowest net benefits for onetreatment and compare their average with that of the

Table 8.2. Net benefits bylocation for Treatmentso kg Nand 80 kg N

25% lowest net benefits for the alternative. The fivelowest net benefits representing the 25% worst cases foreach treatment are marked in yellow in Table 8.2.

Net benef"rts($/hal

o kg N BOkgN

441 655511 647383 277391 610250 593322 619490 660458 600180 162250 612542 562512 681285 291387 578375 230494 661485 660295 480485 683463 260

400 526

location

123456789

1011121314151617181920

Average

Averageof fivelowest 252 244

69

If the average of the lowest net benefits for the tentativerecommendation is higher than the average of thelowest net benefits for the farmers' practice. then therecommendation should be made. because even in theworst cases the recommendation does better than thefarmers' practice.

But if the average for the tentative recommendation islower than that for the farmers' practice. then a decisionmust be made. The average of the five lowest netbenefits for 0 kg N is $252. whereas the average for thefive lowest for 80 kg N is $244. The absolute value ofthese net benefits has little meaning but the differencebetween the two should be examined. If the difference issmall. then farmers will probably be willing to acceptthis risk. knowing that over the long run they will comeout ahead with the recommendation. In this case. thedifference is only $8, and is small in relation to theaverage increase in net benefits ($126). So it is likelythat farmers will be willing to accept this risk. But if thedifference is large. representing a sum equivalent to asignificant part of farmer income or a quantity thatwould put farmers in serious debt to a bank or amoneylender. then it would be best to reconsider therecommendation. Perhaps an alternative could be found(in this case it would be worth doing the minimumreturns analysis on 40 kg N as well). If no less riskyalternative is available, then the farmers' practice is tobe preferred.

It is important to emphasize that this type of analysisassumes that all locations are representative of a singlerecommendation domain. and that there is nothingspecial about any individual location. The poor resultsfor one treatment mayor may not be in the samelocation as the poor results for another treatment. Thusin Table 8.2 the farmers' practice does much better thanthe recommendation in location 3. whereas in location 5the reverse is true. But it is assumed that these

70

locations passed through the analysis described inChapter 7. The explanation for these peculiar resultsmay be a specific factor, such as flooding, or it may bean undetermined cause. But the decision has been takenthat they both fairly represent the recommendationdomain, should be included in the marginal analysis,and then included in the minimum returns analysis.

Finally, it should be noted that the minimum returnsanalysis is done with actual location by location data.No attempt is made to fit the data to standard frequencydistributions. The rule of thumb of looking at the worst25% of cases for each treatment is a gUideline only.Experimental results unfortunately do not always givesmooth curves and normal distributions. The key tominimum returns analysis, as with the other analyticaltechniques described in this manual, is a commonsenseexamination of the data from the farmers' point of view.

Chapter NineVariability in Prices:Sensitivity Analysis

71

Experimental yields are not the only element of thepartial budget that is likely to vary. Input and productprices are subject to change as well. Researchers needsome way of deciding which prices to use in a partialbudget when making recommendations. At times it isdifficult to predict where prices might be a year orseveral years in the future, or difficult to estimate theopportunity cost of a particular input such as labor. Inthese cases, researchers need a way of estimating therange of prices under which a given treatment may berecommended. A method for doing this is calledsensitivity analysis.

Which Costs and PricesShould Be Used in the Partial Budget?

Chapters 2 and 3 emphasized that the partial budgetshould use the costs and prices that farmers actuallyface, rather than those announced in the newspaperor set by the government. But beyond this rule thereare still a number of questions that may be asked abouthow to select the appropriate price. The price of the cropmay vary considerably within one year, or betweenyears. Both crop and input prices may be subject toinflation. And both may be affected by governmentpolicies. What prices should be used in these cases?

It is not uncommon for crop prices to vary within ayear, rising just before harvest and then falling afterharvest. Even if all the farmers in a recommendationdomain store their crop after harvest to sell it at a laterdate, it is usually most convenient to base the fieldprice of the crop on the market price immediatelyafter harvest.

If crop (or input) prices vary from year to year, it ispossible to use the average price over the past, say,three to five years as a basis for calculating field prices.If researchers have access to price data from ten yearsor more, a trend price may be estimated. Very often,however, these "trends" are due to inflation. Althoughinflation is a serious problem for any country, it neednot be an impediment to the marginal analysis. If thecalculations of the costs that vary are based on theinput prices that the farmers will face at the beginningof the cycle, and if the field price of the crop used forcalculating gross field benefits is based on the crop pricethe farmers will receive at the end of the cycle, and ifthe minimum rate of return includes the rate of inflation

72

(which it should if it is based on the rate of interest inthe informal capital market, or in the unsubsidizedformal capital market), then the comparison of themarginal rate of return to the minimum rate of return isvalid. Alternatively, if input prices and product pricesare taken at one point in time, then the inflation ratedoes not have to be included in the minimum rateof return.

In some cases, prices are controlled by the government,either directly or through certain policies that affect theoperation of market forces. If input prices aremaintained at low levels through subsidies of some kind(or if crop prices are maintained at high levels), caremust be taken in using these prices in the economicanalysis of experimental results. If the analysis is to beused for making recommendations to farmers for futureyears, a judgment must be made as to whether thegovernment can maintain such subsidies. If it seemsunlikely, then it will be better to use more realisticprices in the calculations.

If, on the other hand, farmers are adversely affected bygovernment policy, if crop prices are controlled (andfarmers have no alternative markets) or inputs are soldat higher than world market prices, then there are twopossible lines of action. First, over the short term,recommendations will have to be based on the pricesthat farmers face under these policies. But second, if itis felt that there is something to be gained by prOVidingpolicymakers with information about the consequencesof their current policies and the possible advantages of achange, the same analysis can be done usin!5 estimatesof undistorted prices and be presented to policymakers.Thus the same set of experiments can be analyzed intwo different ways, for two different audiences; usingcurrent prices for short-term farmer recommendations,and using alternative prices for contributing to theconsideration of policy options.

Sensitivity Analysis

Markets, inflation, and policies are often unpredictableenough that, short of access to a crystal ball, there is noway for researchers to predict prices with any certaintya few years in the future. Recommendations ofteninvolve an investment in extension agents' time, fielddays, pamphlets, or radio programs, and researcherswould like to feel that a recommendation will be able to

withstand any likely changes in prices of inputs or cropsfor at least a few years.

Sensitivity analysis

The best way to test a recommendation for its ability towithstand price changes is through sensitivity analysis.SensItIvIty analysIs simply implies redoing amarginal analysis with alternative prices. If, forinstance, a fertilizer recommendation is made usingcurrent fertilizer prices, but there are indications thatthose prices may increase, a reasonable estimate of thenew prices may be substituted in the analysis. Table 9.1illustrates such a situation. In the original analysis (caseA), a field price for nitrogen of $0.625/kg was used. Therecommendation of 80 kg N was made, assuming aminimum rate of return of 100%. If the field price ofnitrogen increases to $0.75/kg, would the samerecommendation hold? Redoing the partial budget (caseB) with the higher price of nitrogen shows that therecommendation of 80 kg N is now in doubt, becausethe marginal rate of return of changing from 40 kg N to80 kg N is just equal to the minimum rate of return.Any higher nitrogen prices would necessitate loweringthe fertilizer recommendation.

Table 9.1. Sensitivity analvsis for nitrogen experiment-----====;-;::=:::==-

Case A(Current field priceof N = $0.625/kg)

Case B(Future field priceof N = $0.75/kg)

601070

516

80 kg N2.930

586

40 kg N2.580

516

305

35481

o kg N2.000

400

ooo

400

80 kg N2.930

586501060

526

o kg N 40 kg N2,000 2.580

400 516o __....;2;:;,;;5;.".-_o 5o 30

400 486

Adjusted yield (kg/halGross field benefits IS/halCost of ~ rtUtzer ($/halCost of labor (S/halTotal costs that vary ($/halNet benefits ($/hal

Marginal rates of return

o kg N LO 40 kg40 kg to 0 kg

= 2 7%= 133%

o kg40 kg

to 40 kgLO 80 kg

231%100%

If the minimum rate of return does not change, and theprice of labor and the field price of maize remainconstant, how high can the field price of nitrogen gobefore even 40 kg N ceases to be a viablerecommendation? Such questions can be answered by

73

l:l.Y

l:l.TCV

M

the formula in Table 9.2. (This is the same formula usedin Chapter 6, p. 54, to help in selecting economicallyviable treatments for experimentation). The change inthe total costs that vary will depend on the field price ofN (n) and the labor costs of applying 40 kg N/ha ($5).The calculation shows that the nitrogen field price canrise to $ 1.33/kg before 40 kg N ceases to be a profitablepractice for farmers.

Sensitivity analysis can also be used to examineassumptions about opportunity costs. particularly thoseof labor. At times a partial budget is developed whichuses an opportunity cost of labor that is only a roughestimate. If the treatments involve significant changes inlabor, an inaccurate estimate of the opportunity cost oflabor may lead to erroneous conclusions. Otheropportunity costs of labor can be substituted in thepartial budget to give an idea of the range over which agiven recommendation would be acceptable to farmers.

Table 9.2. Calculation of maximum acceptable field priceof nitrogen

change in adjusted yieldchange in total costs that varyminimum rate of return(expressed as a decimal fraction)

P = field price of productl:l.TCV(l + M)

l:l.Y = P

or

l:l.TCVP x l:l.Y

l+M

40 n + 5 =

74

Example

Increase in adjusted yield betweeno kg Nand 40 kg N = 580 kg/ha

Cost of labor to apply fertilizer = S5/haMinimum rate of return = 100%Field price of maize = SO.20/kg

To calculate the maximum acceptable field price of nitrogen(n) in order for the application of 40 kg nitrogento be economic:

0.2 x 5802

n = S1.33/kg

Suppose experimental evidence shows that a certainherbicide gives the same average yield as the farmers'hand weeding. A comparison of costs that vary is thusthe only economic analysis necessary for making therecommendation. Table 9.3 shows these calculations. Incase A. the researchers have assumed an opportunitycost of labor of $l/day. The total costs that vary of usingthe herbicide are lower than those of hand weeding. andtherefore the herbicide should be recommended. But ifthe opportunity cost of labor is only $O.50/day, thenhand weeding is the preferred alternative. (Calculationsshow that as long as the opportunity cost of labor isabove $O.56/day. the herbicide is to be recommended.)This illustrates the necessity of carefully studying theaVailability and utilization of labor before makingrecommendations for something like weed control.

The discussion of sensitivity analysis serves as areminder that farmer recommendations may change asprices change. Agronomic data regarding responses to afactor are valid as long as the biological environmentand farming practices do not change. The economicinterpretation of that data will depend on changes inprices. There is thus the need to continually reviewfarmer recommendations. based on past agronomicexperiments, in the light of present (and future)economic circumstances.

Table 9.3. Sensitivity analysis for weed control experiment

Case A Case B(Opportunity cost (Opportunity cost

of labor = $1.00/day) of labor = $O.50/day)

Costs that vary Hand weeding Herbicide Hand weeding Herbicide

Herbicide ($/ha) 0 8 0 8Sprayer ($/ha) 0 1 0 1Labor cost ($/ha) 20 4 10 2

Total costs that vary ($/ha) 20 13 10 11

75

Part Five Summary

Chapter TenReporting the Resultsof Economic Analysis

This manual has presented a set of procedures for doingan economic analysis of on-farm agronomicexperiments. The careful use of these procedures willhelp in selecting treatments for further experimentationand for developing farmer recommendations. Whenresearchers report the results of on-farm experiments, asummary of the results of the economic analysis shouldbe included. The following points are a checklist fororganizing a report of the economic analysis.

1 Review Objectives of ExperimentBefore beginning any analysis, review the objectives ofthe experiment. Include a review of the previousdiagnostic and experimental evidence that was used inplanning the experiment and a review of the tentativedefinition of the recommendation domain. The purposeof each variable in the experiment should also bereviewed. Does it represent a possible alternative to thefarmers' practice, or is it meant to provide initialevidence about the importance. interactions or causalityof particular production constraints? In other words, dotreatments represent possible farmer recommendations,or are they being used to help design furtherexperiments which will lead to such recommendations?

2 Review Experimental Design and ManagementReview the design and management of the experiment.The marginal analysis presented in this manual isuseful only when applied to on-farm experiments withparticular characteristics. The nonexperimentalvariables must be at levels representative of farmers'practice in the recommendation domain, and onetreatment must represent the farmers' practice withrespect to the experimental variable(s).

3 Calculate Total Costs That VaryIdentify the variable inputs for each treatment in theexperiment. Make sure that all inputs that vary acrosstreatments are included. paying particular attention tochanges in labor. Calculate the costs that vary for eachtreatment, on a per-hectare basis. For purchased inputs.base the costs on realistic field prices that farmers in therecommendation domain must face. For nonpurchasedinputs, develop realistic opportunity costs. Sum the totalcosts that vary for each treatment. (A preliminarycalculation of these costs should have been done whenthe experiment was being planned.)

78

4 Calculate Average YieldsReview the results of the experiment at each location.These may be the results of a single year, or of severalyears. Decide if all the locations represent a singlerecommendation domain. Decide if any locations shouldbe eliminated because of errors in experimentalmanagement. Report the reasoning behind thesedecisions. Use statistical analysis to help decide if thereare any differences in response among the treatments.Locations with results that were affected by unexplainedor unpredictable factors must be included in thestatistical analysis.

5 Decide If a Partial Budget Should Be Presenteda) If there are no yield differences among treatments,the one with lowest total costs that vary should bechosen for further experimentation or, if there issufficient evidence, for recommendation.

b) If there are yield differences among treatments, thena partial budget will have to be developed.

6 Calculate Adjusted YieldsThe first line of the partial budget should show theyields for each treatment averaged over all locations inthe recommendation domain. The second line showsadjusted yields based on differences between theexperiments and the farmers' fields with respect to trialmanagement, plot size, or time or method of harvest.

7

8

77

Calculate Gross Field BenefitsCalculate the field price of the crop. Remember, anexperiment may involve more than one crop, and/ormay involve crop by-products, such as fodder, which areof importance to farmers. The field price of a crop is theprice that farmers receive, less all costs of harvestingand marketing that are proportional to the yield. Thegross field benefits for each treatment are the adjustedyields times the field price.

Calculate Net BenefitsList the costs that vary, and the total, for eachtreatment. Calculate the net benefits for each treatment.The partial budget should contain only yield, cost, andbenefit figures. Assumptions about field prices, yieldadjustments, etc. should be presented beneath thepartial budget as footnotes. Details on experimentaltreatments should be clearly presented elsewhere in thereport, in the discussion of the experiment.

9 Do a Dominance AnalysisArrange treatments in order of ascending total costs thatvary. with corresponding net benefits. Eliminatedominated treatments.

10 Estimate a Minimum Acceptable Rate of ReturnEstimate a minimum rate of return for a crop cycle. Inmost cases the minimum rate of return will probably bebetween 50% and 100% for a crop cycle.

11

12

Do a Marginal AnalysisA marginal analysis presents the nondominatedtreatments on a net benefit curve and calculates themarginal rates of return between pairs of adjacenttreatments. Compare the marginal rates of return to theminimum rate of return in order to select acceptabletreatments. Present the results of the marginal analysisin the report.

Draw Conclusions From the Marginal Analysisa) If the results of the experiment are being used to helpplan further experimentation. then the results of theeconomic analysis should be discussed in the report inlight of the choice of appropriate treatments forexperiments in the next cycle.

b) If the economic analysis is being done to develop arecommendation. then the report should contain adiscussion of the evidence that has been used to makethe recommendation.

13 Before Making a Recommendation,Do a Minimum Returns AnalysisIf data from enough locations and years are available, doa minimum returns analysis on all the experimentalresults to examine the implications of the variability inthe results for farmer welfare.

14 Before Making a Recommendation,Do a Sensitivity AnalysisIf variability in prices or costs is expected, carry out therelevant sensitivity analysis and include the results inthe report.

78

Index

79

References to definitions of terms are printed inboldface type.

Adjusted yield. 10. 23-25Adoption of recommendations. 5, 51-52Agronomic assessment. 3. 12.21, 58, 62Average yield. 9, 22-23Continuous analysis. 52Cost of capital, 34-37Costs that vary, 10. 13-19Dominance analysis, 30-31Experimental variables, 5-6, 55Farmer assessment, 3. 49Field cost, 14Field price (of an input). 13-16Field price (of output). 10.25-27,71Gross field benefits. 10. 27-28Inflation, 35, 71-72Labor. 16-18, 53, 74-75Management of experiments. 5-7, 23-25. 59Marginal analysis, 11-12. 38-46Marginal rate of return, 12. 32-33. 49Minimum rate of return. 34-37,48, 71-72Minimum returns analysis. 66-70Net benefits, 4. 11, 28Net benefit curve, 31-32, 41, 45Nonexperimental variables. 6, 23-24. 57On-farm experiments, 5-7On-farm research. 1-3Opportunity cost, 13. 16-17. 34, 53. 74-75Opportunity field price (of an input), 15Opportunity field price (of output). 27. 35 (footnote)Packages of practices, 5. 51-52Partial budget. 9, 27-29Policymakers. 3, 16, 36, 72Recommendations, I. 49. 51-52Recommendation domain, 7-8,20-21.57-58Residuals. 47-48Risk. 4-5. 59. 63-66Sensitivity analysis, 53. 73-75Statistical analysis, 3, 21-22. 60-62Total costs that vary, 11. 18-19Working capital, 34

ISBN 988-8127-18-6

CENTRO INTERNACIONAL DE MEJORAMIENTO DE MAIZ Y TRIGOINTERNATIONAL MAIZE AND WHEAT IMPROVEMENT CENTERUsboa 27 Apartado Postal 6-841 06800 M6xico. D.F.• M6xico