CONDUCTING ON FARM RESEARCH IN FSR
MAKING A GOOD IDEA WORK
Farming Systems Support Project
Institute of Food and
University of Florida
Gainesville, Florida 32611
Office of Agriculture and
Office of Multisectoral Development
Bureau for Science and Technology
Agency for International Development
Washington, D.C. 20523
NETWORKING PAPER No. 13
CONDUCTING ON FARM RESEARCH IN FSR
MAKING A GOOD IDEA WORK
Tacloban City 7101
Ithaca, N.Y. 14853
*Clive Lightfoot is a Research Associate and Randolph Barker is a Professor
of Agricultural Econanics, Cornell University. The authors are indebted to
Tully Cornick for comments. All three are currently involved in the
Philippines FSR, Farming Systems Development Project Eastern Visayas.
Networking Papers are intended to inform colleagues about farming
systems research and extension work in progress, and to facilitate the
timely distribution of information of interest to farming systems
practitioners throughout the world. The series is also intended to invite
response front the farming systems network to help advance FSR/E knowledge.
Comments, suggestions and differing points of view are invited by the
author or authors. Names and addresses of the author or authors are given
on the title page of each Networking Paper.
Networking Papers do not necessarily present the viewpoints or opinions
of the FSSP or its affiliates. Readers wishing to submit materials to be
considered for inclusion in the Networking Paper series are encouraged to
send typewritten, single-spaced manuscript, ready for publication. The
FSSP does not edit Networking Papers, but simply reproduces the author's
work and distributes it to a targeted audience. Distribution is determined
by geography and subject matter considerations to select a sub-group from
the FSSP mailing list to receive each Networking Paper.
Conducting on Farm Research in FSR -
Making a Good Idea Work
Clive Lightfoot and Randolph Barker
Over the past decade a wide spread interest has developed in Farming
Systems Research (FSR) in the International Agricultural Research Centers
(IARCs), in the national agricultural research and extension systems of
developing countries, and in academic circles in many developed countries.
Among practitioners there has been general agreement on the broad
philosophical approach. In fact, it has been often stated that FSR is a
philosophy rather than a methodology. As a consequence, nearly any research
activity that is seen as farmer oriented and interdisciplinary is labeled FSR
if for no other reason than to attract donor funding. Wooed by the rhetoric,
donor agencies such as the World Bank and USAID have made substantial
investment in FSR projects.
Major attempts have been made in the literature to clarify the concepts of
FSR. This, for the most part, has led to more acronyms (FSR&D, FSR/E, FSIP,
and OFR/FSP to name a few) and more confusion, as various authors have given
us their own perceptions. Most recently the World Bank hired Norman Simmonds
to tour the world and unravel the mysteries of FSR. Simmonds's report makes
Clive Lightfoot is a Research Associate and Randolph Barker is a Professor
of Agricultural Economics, Cornell University. The authors are indebted to
Tully Cornick for comments. All three are currently involved in the
Philippine FSR, Farming Systems Development Project Eastern Visayas.
- 1 -
an important contribution in that it presents the broad perspective of FSR
with great clarity. But, as with most of the literature, the methodological
issues are scarcely addressed. With most of the attention devoted to
clarifying the philosophy and concepts and a lack of focus on development of
methodology, it is not surprising to find a growing concern among the donors
and practitioners alike that FSR is not improving the efficiency of our
research extension effort. FSR is not leading to more rapid adoption of new
technology and significant gains in agricultural production, productivity and
farm family welfare. Indeed, many of the problems experienced arise from
this lack of focus which in turn explains the weak development of methods
that exploit the comparative advantage of FSR. It is the experience of many
projects that initial methodological approaches to FSR, both surveys and
experiments in farmer's fields, have been for the most part inappropriate.
FSR methodologies are slowly evolving in a number of projects and
institutions, which take into account the limited resource endowments and
exploit the comparative advantage of national research and extension
networks. Our objective in this paper is to identify a set of methods and
procedures that allow FSR projects to immediately increase their efficiency
in terms of developing technologies that farmers adopt. In doing this we
glean from the works of others and our own experiences. This task has been
made difficult by the paucity of material submitted to academic journals.
Agreed there have been reviews on FSR, notably Shaner et al 1979, Norman
.1979, Gilbert, Winch and Norman 1981, and most recently Simmonds 1983; also
some IARCs' have produced training manuals, notably Collinson, 1980, Perrin
et al 1979, and Zandstra et al 1982. We make no grandios claims for these
procedures since we are still at the point of testing them. Unfortunately,
like most other practitioners in the field we are more familiar with vw-h-
does not work than what does work.
Part 1. Understanding the Existing Farming System
and Identifying Problems
An important first step in FSR is to select sites in which the research will
have a significant impact. The recommendation domain and target group of
farmers must be identified and the farming systems described in order to be
able to understand and identify the important problems.
Site selection is frequently not accomplished by those who are to carry ovut
the research but rather by those who prepare the project proposals. TI
normally involves a rather unsystematic mixture of political, socioeconoma~
and technical judgement. The governments usually mandate the beneficiaries
of research in broad general terms such as "subsistence farmers," "smalI
farmers" or "resource poor farmers." The task of characterizing a"s
selecting research participants should begin with site selection. It should
involve the systematic use of secondary data including soil maps and census
data to consider in both geographic and demographic terms the potential
beneficiaries based on site selection. In short, at the very conception of
the project an effort should be made to define in general terms the
recommendation domains and target beneficiaries.
In many farming systems (and cropping systems) projects, the site selection
has been followed by detailed in-depth benchmark and or multiple visit
surveys to obtain the necessary information to fully describe the farming
system. Eicher and Baker (1982) have the following comment on the efficiency
of this approach: "Moreover, it often requires 6-12 months to plan a cost
route study, a year to carry it out, and sometimes 2-3 years to analyze and
publish the results. Concern with the cost of cost route surveys and the
need to generate rapid results has led to a search for survey methodologies
which can produce results in a few months rather than 2-3 years."
The failure of the large survey approach led to the development of shorter
and more informal survey procedures, of which CIMMYT's exploratory surveys
and ICTA's sondeo are perhaps the most popular. These procedures come under
the heading of a class of activities known as "rapid rural appraisal" and
frequently earn the additional title "quick and dirty." The problem
suggested by this latter title is that these approaches do not provide
adequate information on which to design appropriate research activities. The
problems are identified at too general a level, i.e., soil fertility, soil
erosion, or livestock nutrition.
Much more must be known about the current range of farmer knowledge and
experience and resource potentials and constraints. Here, we propose a
diagnostic procedure which combines the quick survey approach with a much
more detailed monitoring and measurement in specific problem areas or areas
that appear to offer potential for research. These two levels of
investigation are described in the sections that follow.
Level one is the sondeo or exploratory survey activity. The sondeo is a
survey conducted by an interdisciplinary group without the use of a formal
survey lasting a period of several days. The details of this procedure are
fully described by Hildebrand (1979). The purpose is to more clearly define
the recommendation and target group of farmers and to identify the majoi
problems and potentially researchable issues. While the sondeo team normally
represents several disciplines, even more important than the disciplinary
composition is the choice of individuals. Two types of people are needed;
those who are capable of identifying research problems and issues and those
who know the region. While the latter group is likely to be composed
entirely of the FSR team the former group may include people outside the
project with on-farm research experience. Many of these experienced
individuals are busy with other work but are able to commit a few days to
participate in a sondeo.
We should emphasize the fact that the success with which the sondeo can
identify key problem areas depends a great deal on the quality and experience
of the team. The usual project situation is one of doing the sondeo with a
fresh group of people who spend as much of their time trying to work together
as they do learning about the system.
How do we know that from the sondeo that we have identified the right
problem areas? At the end of the sondeo when the report has been completed,
a dialogue must be held with farmers, probably on a group basis, to discuss
the sondeo team findings. It is very important at this early stage in the
project that the farmers and scientists agree on the problems.
Level two is the diagnostic, monitoring and measurement activities. We
have already learned from the failure of the large survey approach that we
cannot gather and analyze data in a reasonable time period in all aspects of
the farming system. The informal survey work has helped us to identify the
major livestock and crop activities and the major problem areas. We must now
limit the number of data gathering activities and see that they are clearly
First on the agenda is the need for a survey to describe the target group
of farmers in more quantitative terms. This survey should concentrate on
about a dozen key variables such as farm size and tenure, family size and
occupation, land use including crops grown, and livestock enterprises. While
the survey procedures can be standardized across sites, in our view, the
formal questionnaire must be kept extremely short and when possible should be
analyzed at the site. The advent of the microcomputer unfortunately, once
again, has strengthened the notion that the size of the survey can be
increased and the data brought to a central location for rapid processing and
analysis. A more appropriate alternative is to strengthen the capacity of
the site team to conduct their own analyses' using calculators, sorting
strip, etc. Site researchers need such information quickly, for example, to
assist them in selecting farmer co-operators who are representative of the
The survey information provided in the above activities should be adequate
to allow the researchers to identify the major problem areas. It is,
however, unrealistic to expect that after only a few short months on
location, the research team is prepared to design experiments. The
experiments conducted in the first year of the project are normally of little
value because they do not, or more correctly cannot, address the important
issues. More formal monitoring activities are needed. Once the two or thre-
key areas are identified, an in depth investigation must be undertaken with a
view to understanding the farmers knowledge and experience, the range of
environmental variability, and the factors explaining variability in
performance among farmers. The specific purpose of this analysis is to
quantify the production and management experience across farms and to
identify potential areas of research impact. This requires a careful
monitoring of both physical and socio-economic factors.
For example, if improvement in cattle production is identified as a likely
area for technical impact, one must examine the existing production system oe
a sample of farms. Information must be gathered throughout the year on feel
supplies and feeding practices, animal health, labor requirements and
purchase of inputs if any. We need to know why some farmers are doing better
than others and what researchable topics might lead to significant gains I.
How is the monitoring to be organized? Approximately 20 farms should be
adequate for the task. Monitoring activities might combine occasional
surveys to establish labor requirements, livestock inventories, animal health
etc. with frequent visits to monitor feeding practices and seasonal
variations in feed supplies. It might be necessary to obtain laboratory
analyses of indigenous forages to establish their nutritive value. At the
end of the monitoring period, a report could be prepared on "potentials for
increasing cattle production in the farming system researchable issues."
Knowing that this report is the objective of their efforts will help the site
team researchers to focus their work. One could visualize a similar type of
monitoring activity for other problem areas such as crop production, erosion
control, or soil fertility.
The danger of reducing the monitoring to say a single enterprise such as
cattle production is that we may fail to emphasize the linkage of the
enterprise to other components of the farming system. We must be careful to
guard against this, spelling out clearly the way in which the cattle
enterprise competes for feed supplies, labor, and other inputs within the
farming system. Alternatively, the failure to sharply focus our monitoring
activities also has its price. If we choose to monitor all livestock
activities, then it will be difficult or impossible to obtain the depth of
understanding we need to identify the researchable issues.
In summary, through the diagnostic analysis, every effort should be made to
quantify the parameters to minimize subjectivity in identifying and
specifying researchable areas. The site staff have to take an active role in
the diagnostic measurements, monitoring, and other forms of data collection.
An assessment must be made of those problems which can be solved by carrying
out simple experiments conducted on farmers' fields by the site teams in
conjunction with farmers, and those problems which require more complicated
in-depth investigation. Selecting and designing the innovations for on-farm
investigation is the subject of the next section.
Part 2. Selecting and Designing the Innovations
Worthy of On-farm Investigation
In the usual framework of farming systems research, design is recognized as
the second stage. The selection of innovations to be tested on farms and
their design usually falls to the technical scientists working on the
research stations. Typically, the influence of the information gathered
during the initial surveys on selection and design is weak. Consequently
what ends up being tested are largely the current interests or
recommendations of the research institutions. For example, where soil
fertility is seen as the problem, researchers will want to run tests on
chemical fertilizers even when these can not be purchased by farmers. This
is not to say that these scientist are disinterested in the relevance of vow_?
to the. farmer, but that they do not have enough detailed information to dc
the job properly. The point here is that these recommendations are seen to
be relevant because any innovation can be said to address the general problc
of low production described by the surveys. The cursory nature of thi,
process that is customary in FSR springs also from a confusion about this
stage in the overall framework of research.
The stage of design as presented in Norman's four stages of FSR can and has
been interpreted in two ways. The original and intended interpretation was
that design be a process of refinement of technology packages to fit the
farming conditions by on-farm researcher managed experiments. When practiced
in the field it did not take long before the stages of design and testing
became impossible to distinguish. Now that experience has indicated farmers
disinterest in packages, a second interpretation has emerged, primarily
through the work of Collinson. He interprets design to be an intellectual
process where all possible technical solutions are screened and 'prioritized'
by technical and social scientists. Farmers also should be included in this
Five steps are enumerated in the CIMMYT manual for the conduct of design
(Byerlee, Collinson, et. al. 1980). The first step is identification by
technical scientists of the biological problems encountered in the initial
surveys. Each problem is then examined to define the possible causes which
for example may be related to the farmers objectives or limited resources.
In the third step a wide a range of apparent solutions to each problem are
generated. Typically a narrow range of direct solutions are entered here
when the farmer needs a wide range of direct and indirect (i.e. solutions
that exploit system interactions) solutions. These solutions are then
screened by technical scientists who pose the questions: Will this biological
relationship hold in the farm situation? What are the husbandry requirements
for success? Concurrently the economists ask: Are the infrastructural
support requirements feasible? Do the farmers have sufficient resources?
Will it use resources more profitably?
Step five, priorities the technical options in terms of potential impact,
ease of adoption, and ease of research effort. When establishing priorities
the adoption concern should be the single most important criteria. The
importance of this final step cannot be overemphasized because it controls in
large measure the successful implementation of experiments, the degree of
- 10 -
adoption, and thus the validity of this approach to research. By this token
it is essential to involve the farmer in all the steps outlined above ane
especially the screening process. Here, it may be useful to present a wide
range of possible solutions to the farmers so that they can pick those most
suited to their circumstances. This will be especially true where
researchers have difficulty in answering the questions posed in step four and
even more so when assessing step five's ease of adoption.
Part 3. Using Farmers to Conduct the
The range of on-farm experiments found in FSR programs encompass the most
complex replicated factorials to simple two plot demonstrations. This broer'
range has been divided into three types by the level of researcher and farmy
involvement. The most complex trials such as the IRRI component tests ~-
CIMMYT exploratory and levels tests are classified as researcher managed and
executed. Intermediate levels such as IRRI's superimposed cropping patterns
are classified as researcher managed and farmer executed. The least complex
trials often termed demonstrations or by CIMMYT verification tests are
classified as farmer managed and executed. The intention is that
technologies move from the most complex type one to the least complex type
three which implies that on-station basic research occurs before type one.
Thus type one tests research generated technologies for biological
performance in the farm setting. The second type of trial exposes the
technologies to farmer levels of management and farmer opinion. Finally,
predictable technologies are tested in the type three trials for performance
over a wide range of farming conditions. Although this wide range of trials
- 11 -
are talked about in practice, most of the research manpower, particularly at
the IARCs, is tied up in the complex trials and with intermediate level
trials which although simple in design are costly in terms of supervision and
FSR recognizes two distinctly different bodies of knowledge the knowledge
which comes from basic scientific research 'and the knowledge which is
acquired through time by farmer experience. This latter body of knowledge is
not formalized nor does it appear in the literature. It can only be captured
through direct interaction between researchers and farmers. Even though feed
back loops are built into the conceptual diagrams of FSR, in practice the
ability to draw on the farmers traditional body of knowledge remains weak.
This is in large measure because FSR remains very top-down in its
FSR field workers typically have borrowed designs wholesale from
conventional on-station experiments. Most programs expend all their energies
on the more complex types of researcher-managed work because they are very
demanding to implement. These conventional experimental methods were
primarily developed to determine "site effects" on largely unknown biological
parameters. They are largely miniaturized research station experiments
having randomized block designs of two or more replicates with four or more
treatments which in case of CIMMYT levels trials are factorially structured.
The analytical procedure is to conduct analysis of variance or response
functions analysis and to apply standard statistical tests of significance.
Adequate precision in the data are guaranteed by enormous research input into
- 12 -
the management and implementation of each experiment; a resource level thbt
IARCs command but that is rarely found in national institutions.
Researchers in the national level programs, despite the great difference in
resource endowments have generally followed the lead of the IARCs adopting
the same methodological procedures. While economists have struggled to
analyze the benchmark survey, agronomists have borrowed the IARC designs and
set out component and cropping pattern trials. The project researchers have
been overwhelmed by the sheer magnitude of the problems involved in managing
data and controlling experiments. Only a few experiments are conducted &t
each site on a handful of farms. In the variable farm environments these few
observations stand little chance of detecting real effects, even if one werc
to assume that the experiments have been properly managed, which is normal-
not the case. These facts notwithstanding, standard statistical test
typically are applied to the data and recorded in the results. The farmers
involvement in these experiments only extends to the lending of land, an,
perhaps land preparation with some weeding. In the experiments, yield pe-
hectare is taken as the principle criterion of evaluation, and the most
commonly used test inputs are variety, chemical fertilizers and
insecticides. Inputs such as seeds and chemicals are supplied by the
researchers and for this the farmers are only too willing to co-operate.
But, as one project manager observed when asked to write the "success story"
of his project, "when we withdrew the farmers withdrew."
The methods and procedures described above are inappropriate for FSR. The
primary goal of our research is to enhance adoption of new technology, not to
- 13 -
define biological input-output responses at each site. The following on-farm
experimental method is presented to illustrate a more appropriate procedure
for FSR. Briefly, the experimental method entails the over-laying of
treatments on to the appropriate existing crop or soil conditions. The
procedure for over-laying treatments on to the farmers own crops is similar
to the superimposed trials mentioned by Shaner et. al. 1982, Kirkby et.
al. 1981, and CATIE. For example, to test the benefit of nitrogen
top-dressing of maize, an area of healthy maize, that is a crop that farmers
would be advised to top-dress, is identified on a participant's farm. The
farmer is given the fertilizer with instructions to apply it over half the
identified area demarcating the treatment and control plots after
implementation. By contrast, a conventional experimenter would select the
experimental area prior to land preparation, mark out the plots, and
implement the treatments in the appropriate plots.
The simplicity of this experimental method permits farmer implementation.
Ideally, the innovation for example improved seed, or fertilizer is given to
the farmers for them to implement the treatments. The researchers
involvement extends to the selection of area, instruction on implementation,
and some checking on the accuracy with which the experiment were conducted.
Of course, all measurements and data collection are the responsibility of the
The shift in level of involvement from researcher to the farmer in the
over-laid method increases the scope of farmer participation and gives time
for the researchers to contact many more farms. Testing in appropriate
- 14 -
conditions and the use of many more farms can only increase the rigor and
precision with which innovations are assessed. In addition, farmna
participation could be deepened by soliciting their reactions to the
innovations. The depth of such questioning is likely to be greater when
conducted by researchers trained specifically in this area, which also
affords greater interaction among the disciplines.
There are important implications of this methodology for quantitative
analysis. We have shifted from a. few researchers managed experiments
(usually on just four or five farms) to many trials (twenty or thirty) in
which the farmer applies the input. Although our treatments are set out in a
single recommendation domain, this new procedure introduces the variability
in treatment management among farmers. But at the same time, we havy
increased the rigor of the test and precision of the trial by adding many
more replications across farm. The performance of an innovation across the
variety of environmental and management conditions experienced provider
important information about the range of outcomes farmers might expect.
Such information is provided by the analysis of the distribution of the
observations. This analysis should include the mapping of performance date
and the use of scatter diagrams to provide the researcher with a visual image
of the data and make it easy to associate the level of performance with the
location of the trial. Using simple statistical measures such as mean and
standard deviation, the degree of overlap in performance of the two
treatments (farmers level and superimposed) can be estimated, and where
appropriate significance tests can be employed.
- 15 -
Farmers are also interested in knowing about the likely performance of
innovations over a run of years. Some indication of this is provided by the
calculation of confidence intervals. Here farmers can learn the range of
likely outcomes of a particular new technology. The fact that the trials are
run across so many farms with varying conditions strengthens the utility of
this analysis. However, caution must be exercised in interpreting these
results, since factors causing variability in performance across locations in
a given year, and factors explaining variability on a given farm through time
may be very different.
As with the socioeconomic surveys, the analysis of the data is at the level
of the site teams. Thus, the site teams as well as the farmers are more
directly involved in the research process. We have improved the capacity of
both groups to access the utility of a new innovation. In the final
analysis, the results of our quantitative analysis notwithstanding, farmers
will be the judge of appropriateness of the innovation.
In short, the procedure suggested above provides more time for the site
team researchers to be engaged in the data collection and analysis. However,
greater sensitivity to the analysis and interpretation of data is a product
of more education. Researchers must be willing and able to assume a. certain
freedom from the rigidity of traditional scientific procedures. If the
approach we have outlined is to work, it will require developing capabilities
at the overall project level and at the site team level that do not normally
- 16 -
In this paper we have focused on the methodological issues associated with
farming systems research. We believe that if it is to be effective FSR must
be based not only on a philosophy, but also on an efficient methodology. The
methodologies used in most projects to date have been borrowed from research
stations. They were designed not to enhance the speed of adoption, but to
improve our understanding of physical and biological relationships (e.g.
fertilizer response). These procedures, while useful in their own right, do
not effectively incorporate the experiential knowledge of farmers regarded as
essential in FSR.
Given the objectives of FSR, traditional research procedures appear to b~
inappropriate. Furthermore, the site research teams in the national program
have had neither the trained manpower capacity nor the resources to
successfully manage the experiments in farmers' fields. For those who have
been involved in national farming systems programs, -the problems we have
discussed in using existing methodologies and trying to manage experiments in
farmers' fields undoubtedly have a familiar ring. We have been struck by the
similarity of these problems in locations as afar apart and as different
environmentally and culturally as Botswana, Ecuador, and the Philippines.
We should be quick to acknowledge that despite these problems much has been
learned from FSR to date. Our argument is that it has been learned at a very
high cost, a cost higher than developing countries and donor agencies may be
willing to pay in the future.
- 17 -
An alternative methodology has been described. To a large extent it is not
new. Various components have been described and tested by others who like us
have been concerned with the need to develop more appropriate procedures. In
the descriptive problem identification phase, we suggest the use of the
sondeo coupled with a more in depth diagnostic analysis of the key problem
areas identified in the sondeo. In the design we emphasize the need to
explore in conjunction with farmers the range of relevant alternatives in
order to pick out those to include in the research design. Of critical
importance in these early phases is the need for farmer participation. In
the testing phase, we suggest that where possible, treatments be superimposed
by the farmers themselves, and that 20 to 30 farmers be included in a single
trial. The researchers will then be free to concentrate on the collection
and analysis of data. Development of site team research capabilities, of I
course, will be a major task.
In closing, we stress the need to develop a more efficient FSR -
methodology. We are, however, in the unfortunate position .of knowing what
doesn't work. We will need to test the procedures described above in order
to determine whether what we have proposed is an appropriate alternative.
- 18 -
Byerlee, Derek, Michael Collinson, et. al., Planning Technologie
Appropriate to Farmers Concepts and Procedures (CIMMYT: Mexico, 1980).
Collinson, Michael, Farming Systems Newsletter, East Africa (CIMMYT, 1984).
Eicher, Carl K. and Doyle C. Baker, Research on Agricultural Development in
-Sub-Sahara Africa: A Critical Survey MSU International Development Paper No.
1, (Michigan State University: East Lansing Michigan, 1982).
Gilbert, E. H., D. W. Norman, and F. E. Winch, Farming Systems Research: A
Critical Appraisal, MSU Rural Development Paper No. 6, (Michigan Stat,
University: East Lansing Michigan, 1980).
Hildebrand, P. E., "Summary of the Sondeo Methodology Used at ICTA," ProE<
Conference on Rapid Rural Appraisal, December 4-7, 1979 (IDS: University of
Sussex, Brighton, U.K.).
Kirkby, R., P. Gallegos, T. Cornick, "On-Farm Research .Method: A
Comparative Approach Experiences of the Quimiag-Penipe Project, Ecuador"
(International Agricultural Mimeograph No. 91, Cornell University, Ithaca,
Norman, D. W. The Farming Systems Approach: Relevancy for the Small Farmer,
MSU Rural Development Paper No. 5, (Michigan State University: East Lansing
- 19 -
Perrin, R. K., D. K. Winkelmann, E. R. Moscardi, and J. R. Anderson, From
Agronomic Data to Farmer Recommendations: An Economics Training Manual,
Information Bulletin No. 27 (CIMMYT, 1976).
Shaner, W. W., P. F. Philipp and W. R. Schmehl Farming Systems Research and
Development: Guidelines for the Developing Countries (Westview Press,
Boulder, CO: 1982).
Simmonds, Norman W., "Farming Systems Research: State of the Art" (Draft)
(IBRD: Washington, D.C.: 1983).
Zandstra, H. G., E. C. Price, J. A. Litsinger, and R. A. Morris, A
Methodology for On-farm Cropping Systems Research (International Rice
Research Institute, Los Banos, Philippines: 1981).
- 20 -