Recommendation Domains: A Framework for
CIMMYT Economics Program Working Paper 02/84
* Economics Program, CIMMYT. The opinions expressed are not
necessarily those of CIMMYT
In cooperation with researchers in many national agricultural
research programs, CIMMYT has sought to develop procedures which help to
focus agricultural research squarely on the needs of farmers. The
process involves collaboration of biological scientists and economists to
identify the groups of farmers for whom technologies are to be developed,
determining their circumstances and problems, screening this information
for research opportunities, and then implementing the resulting research
program on experiment stations and on the fields of representative
CIMTYT's Economics Program has emphasized developing procedures for
the first stage of this process, through to establishing experiments.
The evolution of the procedures, now synthesized in a manual "Planning
Technologies Appropriate to Farmers: Concepts and Procedures" has been
strongly influenced by collaborative research with many national programs
and with CIMMYT's wheat and maize training programs.
There is a need to synthesize the experiences of those working in
on-farm research in order to provide more detailed guidelines on
particular concepts and issues. One example is the present paper, which
summarizes experience with the concept of the "recommendation domain" and
provides guidelines for applying this concept to a research program. We
believe it will be useful to anyone interested in on-farm research.
As with other working papers we will appreciate comment and
criticism in order that we might improve the paper and the procedures.
We would be especially grateful for comments and observations from those
who have used the concept in orienting their own work.
Donald L. Winkelmann
Director, Economics Program
We wish to acknowledge the contribution of many colleagues in the
development of this paper. The phrase recommendationn domain" was
originally used in the first CIMMYT Economics Manual (Perrin et al,
1976). Don Winkelmann saw the need for a further elaboration of the
concept of recommendation domain, and has guided this paper through
several drafts. Earlier versions of the paper have also profited from
comments and suggestions by Poniah Anandajayasekeram, Malik Ashraf, Kwasi
Bruce, Derek Byerlee, Edward Clay, Mike Collinson, Greg Edmeades, Dan
Gait, Edith Hesse, Alberic Hibon, Peter Hobbs, Roger Kirkby, Ron Knapp,
Allan Low, Juan Carlos Martinez, A.F.E. Palmer, David Rohrbach, Gustavo
Sain, Pat Wall, and Michael Yates. We are, of course, responsible for
remaining errors or deficiencies. Maria Luisa Rodriguez cheerfully typed
numerous drafts and this final version.
1.0) Introduction............................... ................... 1
2.0) Recommendation Domains in the CIMMYT On-Farm Research Strategy. 2
3.0) Definitions................................................. 5
4.0) Guidelines for Domain Formation.............................. 7
5.0) Issues and Complications ....................;................. 21
6.0) Summary ....... .............................................. 24
Bibliography............................... ..... ........ 26
oe~eee ee oeo ee e o oe oeo eo 6 -eee oeo-oee oooe 2
Many national agricultural research programs are moving toward the
adoption of on-farm research techniques. This implies location-
specific research for representative farmers. Among the challenges that
scientists face in this type of research are the identification of
priority themes for investigation, the selection of representative sites
for on-farm experimentation, and, most important, the definition of the
clientele for whom the recommendations are to be developed. The concept
of the recommendationn domain" is a powerful tool for resolving these
problems and for organizing an efficient on-farm research program.
The term "recamnendation domain" was first introduced in the CIMMYT
Economics manual on the use of partial budgets for economic analysis of
agroncmic data (Perrin, Winkelmann, Moscardi and Anderson, 1976). In
this manual, the recommendation domain was described as follows:
"It is impossible to conduct experiments on each farm to make
recommendations tailored to each farm. Instead, you must define a
target group of farmers, conduct experiments under conditions
representative of their farms, and make recommendations which are
applicable to the entire group. We shall call such a group of
farmers a recommendation domain. Generally, a recommendation domain
will consist of farmers within an agroclimatic zone whose farms are
similar and who use similar practices..." (p.l).
Further discussion of the recamendation domain concept was
presented in the second CIfMYT Economics manual, on assessing farmers'
circumstances (Byerlee, Collinson, et al, 1980). In this manual, the
recommendation domain was defined as "a group of roughly homogeneous
farmers with similar circumstances for whom we can make more or less the
same recommendation" (p.71).
The aim of this paper is to discuss in more detail the concepts and
procedures associated with forming recanrendation domains. First, the
need for domains will be discussed, with emphasis on their operational
SThe term "on-farm research" will be taken to mean "research with a
farming systems perspective, using on-farm research techniques".
For a discussion of concepts and vocabulary associated with on-farm
research, see Byerlee, Harrington and Winkelmann (1982).
role in OFR. Then the recommendation dcmain concept itself will be
examined and techniques for forming domains will be presented, followed
by a discussion of issues and complications involved in the practical use
of domains in on-farm research.
2.0) Recamnendation Domains in the CIMMYT On-Farm Research Strategy
Over the past several years, CIMMYT agronomists and economists have
developed a set of procedures for multidisciplinary, on-farm research
with a farming systems perspective./ These procedures are designed to
be used by biological scientists, social scientists and farmers, in order
to derive appropriate recommendations. They include the following series
of steps: the diagnosis of farmers' circumstances, the design and
management of on-farm experiments, the analysis of experimental results,
and the presentation of recommendations to farmers. The concept of the
recommendation domain is vital to every one of these steps of on-farm
The early stages of on-farm research are concerned with diagnosing
farmers' practices and problems and identifying opportunities for on-farm
experimentation. The diagnosis begins with a review of secondary data
and talks with local officials, extension agents, etc. Then researchers
carry out an informal exploratory survey of farmers. This may be
followed by a formal survey with a short questionnaire. During this
diagnosis researchers must propose at least tentative recommendation
domains. The delineation of the domains helps address the following
questions: What are the principal research opportunities in this area?
What are the target crops that deserve first attention? Are target crops
and opportunities the same throughout the area, or are there significant
differences? And most importantly, on what themes should research
concentrate in order to derive useful recommendations for farmers in the
shortest time possible?
SSee Byerlee, Collinson et al. (1980); Palmer, Violic and Kocher
(1982); Perrin et al. (1976); and Violic, Kocher and Palmer (1981).
2.2) Design of experiments
Once an experimental program has been defined there are a number of
issues with regard to actually setting up on-farm experiments that must
be addressed through reference to recommendation domains.
What is a representative site and a representative farmer cooperator
for an on-farm trial? No farmer or site is ever completely
representative but many mistakes in selecting cooperators may be avoided
by careful attention to the current characteristics and practices of
farmers. Clearly, a site should be representative of the conditions of
the recacmendation domain that is being studied.
At what levels should non-experimental variables (fixed factors) be
set? In any experiment, experimental variables are distinct from fixed
factors; the former vary over treatments within an experiment but the
latter do not. Nonetheless the (unvarying) level of each fixed factor
must be chosen. CIMMYT OFR procedures (e.g. Palmer, Violic and Kocher,
1982, p. 12; Moscardi et al, 1982) advocate setting fixed factors at
"representative" levels, so that on-farm experiments may measure the
yields and profits that farmers can expect when they superimpose one or
more of the treatments on top of their own current practice. Once again,
"representativeness" can be defined with reference to a given
recommendation domain: Fixed factors should be set at levels
representative of those for the domain being studied./
Researchers must analyze the experimental data in order to formulate
farmer recamendations. Three kinds of analysis are usually needed: (1)
agronomic analysis (how may observed responses be explained in terms of
biological and physical processes?) (2) statistical analysis (are
observed responses real or due to random chance?) (3) economic analysis
SThis does not mean that fixed factors are necessarily uniform for
experiments in one domain. Farmer practice will vary somewhat
within a domain, and the experiments may want to sample this
variation. For more discussion of issues related to site selection
and the level of non-experimental variables, see Kirkby, et al.
(1981) and Tripp (1982).
(which alternative technologies will be preferred by farmers?). -
In doing these kind of analyses pooling the data is generally
recommended. Data from trials within the same domain should be pooled
but data from different domains should be analyzed separately.
The ultimate purpose of on-farm research is of course to derive
recommendations for farmers. If the concept of the recommendation domain
has been followed faithfully, then by the time recommendations are ready
for farmers, extension agents know exactly who their targets are. Using
recommendation domains helps avoid two equally unpalatable alternatives:
(1) offering a different recommendation for each farmer (too expensive)
(2) offering a single reccmrendation for the whole farmer population,
despite differences among farmers (inappropriate for many farmers).
Instead, recommendations are derived and offered with well-defined groups
of farmer clientele in mind.
2.5) The policy context
At any point in the on-farm research process the use of
recommendation domains allows researchers to be able to spell out for
which groups of farmers they are working, approximately how many farmers
are in each group, what are their principal practices and problems, and
what types of recommendations are likely to be produced. This is a great
help in developing good relations between researchers on the one hand,
and institutional or national policy makers on the other. Not only does
this kind of information help researchers in the allocation of their own
research resources but it also gives them useful information to offer to
those who set research policy.
SWe have emphasized that farmers will prefer technologies that are
compatible with their circumstances. An understanding of these
circumstances, both socio-economic and biological, should be
accomplished in the diagnostic, planning, and early experimental
phases of on-farm research. Economic analysis will then be carried
out on technologies otherwise acceptable to farmers. For a review
of partial budgeting techniques for economic analysis of agronomic
data, see Perrin et al. (1976) and Harrington (1982).
Recommendation domain has already been defined as a group of farmers
whose circumstances are similar enough so that they are all eligible for
the same recommendation. It should be emphasized that the domain is a
group of farmers, not a geographical area or land type. Domains are
composed of farmers because farmers, not land types, take decisions on
new elements of technology. Defining domains in terms of groups of
farmers underlines the possible importance of socioeconomic criteria in
domain identification. It also allows the possibility of domain
distinctions that are not amenable to mapping (neighboring farmers can
belong to different domains or, as well, a given farmer can belong to
more than one domain).
It usually happens that there are a number of research opportunities
for a particular camrodity, or even for several commodities, that a group
of farmers have in common. These opportunities should of course be
considered together, taking account of their interactions and relative
importance as plans for a research program evolve. It is natural to
think of the group of farmers that share these opportunities as a single
recommendation domain. But because two groups of farmers may share same
opportunities, but not others, it is well to remember that a
recommendation domain is really specific to a particular enterprise and a
particular research problem. That is, our interest is in defining the
group of farmers for whom a specific recommendation is applicable.
Research area in this paper will simply mean the area in which
investigation is to take place. This is usually defined by the research
institution and may have administrative or agroclimatic boundaries.
Although the concept of recommendation domain is often quite helpful in
refining these boundaries, we will assume here that the research area is
given. Our job is to take the mandated research area and decide how it
should be divided into recommendations domains.
Farmers' circumstances are used in order to identify recommendation
domains. They are defined as "all those factors which affect farmers'
decisions with respect to use of a crop technology. They include natural
factors such as rainfall and soils, and socioeconomic factors such as
markets, the farmers' goals and resource constraints" (Byerlee,
Collinson, et al. (1980): 70). Figure 1 shows how circumstances may
affect farmers' practices and their abilities to adopt new
Figure 1 Farmers Circumstances
Source: Byerlee, Collinson, et al. (1980).
SOCIO ECONOMIC CIRCUMSTANCES
Income, food preferences,
Land, labor, capital
Overall Farming System
ping Pattern, Rotations, Food
Supply, Labor Hiring, etc.,
---oCircumstances which are often major sources of uncertainty for decision-making.
A recommendation is a description of a new element or elements in a
production technology (an improved variety, a new chemical, a different
practice, a change in the timing of an operation, etc.) which researchers
believe farmers will find useful. In the case of the on-farm research
paradigm described here it is derived from an understanding of farmers'
problems and a thorough testing under farmers' conditions.
Recamendations are sometimes made in groups or "packages", as when a new
variety is recommended along with a certain planting density, insect
control and fertilizer level. This is particularly important when there
are strong interactions among several elements. The emphasis, however,
should always be on recommendations that farmers can adopt in a step-wise
fashion. There is now considerable evidence that farmers are more likely
to adopt simple recommendations and make changes gradually, rather than
make abrupt, large-scale changes in their practices (e.g. Byerlee and
Hesse de Polanco (1982)). Thus on-farm research identifies and tests
technologies with a limited number of new elements under farmers'
conditions, to find out which recommendations can be accommodated by
4.0) Guidelines for Domain Formation
The process of domain formation is usually a gradual one, as
researchers gain more experience in their area. Although there is no
unique formula for determining domains there are a set of guidelines that
can be used. These are discussed in the following sections.
4.1) Principles of recommendation domain formation
Recommendation domains are formed based on the researcher's
understanding of farmers' circumstances and practices. Sometimes the
identification of domains can be achieved in the early stages of
diagnosis, after examination of secondary data and conversations with
extension agents, for instance. At other times they are not really well
defined until after a formal survey, and it is not unusual that the final
delineation of recommendation domains must await the results of a year or
more of experimentation. But from the very beginning of the process
researchers should at least begin forming impressions about possible
domains. These impressions are tested and refined as the on-farm
research progresses, until a final definition of the domains in the
research area is established.
The concept of farmers' circumstances is used both for identifying
opportunities for investigation and for forming recommendation domains.
An understanding of farmers' circumstances allows the researcher to
explain current farmer practices, identify key problems and propose
improvements that can be tested on farmers' fields. It also provides the
researcher with an idea of whether or not a particular improvement is
appropriate for all farmers in the research area or only for some.
There is a sense in which the formation of recommendation domains is
related to the statistical concept of stratification. The statistician
stratifies a sample in order to eliminate certain types of variability
and better concentrate on the particular factors under study. In forming
recommendation domains we are grouping farmers who have roughly
homogeneous circumstances and whose needs for technology are thought to
be similar. Through that grouping we are able to develop technology more
appropriate to those specific groups, at a considerable saving in
Recommendation domain formation can be thought of as a process of
considering all the various circumstances that might affect farmer
practices and deciding, for each one, if it is the basis of significant
differences in practices and possibilities within the research area. One
way of making this operational is to think of a checklist, such as that
in Table 1, which lists major categories of circumstances that may be
used to define recommendation domains. The list is by no means complete,
and researchers working in different areas will surely add other factors
to this list. It will also be appreciated that many of these factors are
interrelated: altitude affects temperature and frost incidence, for
instance, and rainfall affects weed population.
Several examples may make clear how the variables on this checklist
can be used to define recommendation domains. Consider the case of soil
differences, which are often important in determining farmer practices.
In one-research area in southern Veracruz, Mexico, there were two basic
soil types. Farmers in the river flood plain had alluvial soils and grew
wet season vegetables and dry season maize. Neighboring farmers had
sandy, acid soils and grew pineapple and maize, in the wet season only.
The difference in soils is responsible for very different maize practices
and problems with respect to such factors as moisture stress, disease and
insect incidence, and fertility requirements. Recommendations about
maize appropriate for one group would not likely be appropriate for the
other. Thus we have two separate recamrendation domains for maize, in
this case determined by soil type.
Table 1 Variables Often Considered in Forming Recammendation Domains.
Risk of drought
Risk of flooding
Nutrient supply capacity
Access to markets and inputs
Access to family labor
Access to other labor
Access to credit
Access to cash
Access to markets for
Access to irrigation
Off-farm labor opportunities
Food preferences and diet
Community customs and
There are often differences in soils within a research area. Does
this mean they will always correspond to different recommendation
domains? No, not at all. In another part of Mexico, in a highland
barley area, soil type varied from clay loam to sandy loam, and
researchers hypothesized that these might cause different domains. But a
closer study of the area revealed no significant differences in farmers'
practices or problems by soil type, and researchers realized that they
were either dealing with a single domain, or that another circumstance on
their checklist besides soil type might be used for distinguishing
Another natural circumstance that may lead to significant
differences in practices and research opportunities is altitude. In part
of the Callejon de Huaylas in Peru, maize researchers identified two
recommendation domains, based on altitude. In the lower .dmain, from
2,600 to 3,000 meters, farmers could plant two crops a year and had
serious problems with leaf fungus diseases. In the higher domain, above
3,000 meters, only one crop a year was possible and one of the principal
problems that farmers faced was frost damage to their maize. Altitude
here served to distinguish two domains, with different maize practices,
problems, and opportunities for research. Again, altitude will not always
serve to distinguish recommendation domains. If variation in altitude is
not associated with significant differences in farmer practices or
biological response then it can be crossed off the checklist.
The same holds true for other natural circumstances. In their study
of farmers' practices and problems researchers will want to ask whether
such things as rainfall pattern, slope, or pest incidence can be used to
define different recommendation domains. Important factors are of course
not limited to natural circumstances, and Table 1 presents a number of
socioeconomic circumstances that may also be useful in identifying
domains. An example or two may be helpful.
It is often the case that farmers who share the same natural
circumstances nevertheless have different access to resources which
affects their practices and their ability to adopt innovations. In one
area in Zimbabe maize farmers prepared their plots with ox plows before
planting. As only about half of the farmers owned oxen the rest had to
rent them. The renters were delayed in their planting, which affected
their production through drought risk, disease, and other factors
specific to late planted maize. There are a series of research
opportunities for animal renters which are not applicable to owners, and
thus it is worth considering two recommendation domains, distinguished by
Another case will provide a counter-example. In a research area on
the north coast of Honduras most farmers controlled weeds in the maize
crop with herbicides, but only one-third of the farmers owned backpack
sprayers. Researchers believed there might be a difference in weed
control practices between sprayer renters and owners. But a survey
showed no differences in weed control practices or timing between the two
groups and revealed that the rental market for backpack sprayers was
quite adequate. Thus access to a sprayer did not affect farmer practices
and did not serve to distinguish recommendation domains.
Farmers can also often be distinguished by access to land.
Differences in farm size may not only directly affect the type of
practice that a farmer follows, but may also be correlated with many
other differences, such as access to equipment, credit, or marketing
facilities. At times these distinctions are quite clear and are
responsible for the formation of different recommendation domains. In
parts of the highlands of Ecuador small and large wheat farmers occupied
the same natural environment, but their socioeconomic circumstances were
quite different. The former relied on animal traction and had no access
to credit, while the latter used tractors and credit facilities (which
lowered their costs for obtaining fertilizer.) This led to quite
different practices (e.g. different rotations and fertilizer treatments)
and these in turn indicated different research opportunities. The result
was two recommendation domains in a biologically homogeneous area one
of small wheat farmers (under 5 ha) and the other of large wheat farmers.
It is of course not always the case that farm size is a determining
factor for domain formation. Researchers want to ask if two farmers in a
given region with different sized farms use essentially the same
technology for a particular enterprise and if they have access to the
same type of resources and markets. Do they use the same variety, the
same seeding techniques, the same seeding dates, the same fertilizers,
etc.? If there are differences, then there may be two domains. If there
are no significant differences, then farm size will not be used in
defining domains. In this case researchers will go on and ask the same
questions about other natural and socioeconomic circumstances on their
checklist (Table 1). If farm size is not important, does altitude or
soil type or land tenure serve to distinguish farmers' practices and
problems? If not one or more of these factors, what else on the
checklist might define different domains? As researchers gain more
experience in domain formation they will probably rely less on a formal
checklist./ But the process is always the same, considering how a
series of circumstances affect how a farmer undertakes a particular
In the examples considered so far a single factor (e.g. altitude or
farm size) has been used to divide a research area into recommendation
domains. But it is not always the case that only one factor influences
farmer practices and research opportunities. Researchers must exhaust
the possibilities on their checklist in the search for relevant
circumstances for defining domains. An example of maize research in Peru
was discussed earlier, in which researchers identified two domains, based
on altitude. In fact, the actual situation was more complicated, as
there were other important differences in farmers' circumstances in the
research area. In the lower zone there were two principal farm types -
small farms averaging less than 2 hectares and large farms averaging 40
has. These two farm types had quite different patterns of rotation,
input use, varietal requirements and maize sales. In the higher zone
there were not such marked differences in farm size, but some farmers had
Researchers will have their own preferences on how to think about
these factors during diagnosis. Collinson (1979), for instance,
suggests first considering agroecological factors and then moving to
"hierarchical" divisions due to socioeconomic differences.
access to irrigation while others did not (all farmers in the lower zone
had irrigation). This was responsible for significant differences in
rotations and input use. Thus there were actually four different
recommendation domains in the research area, based on altitude, farm size
and access to irrigation.
In order to form recommendation domains researchers must study the
circumstances and practices of farmers in their area. Using a checklist
of circumstances they can consider in turn various possibilities for
defining recommendation domains. It may be that the area is homogeneous
enough to constitute a single recommendation domain. If not, there are
usually one or at most a few key circumstances that can be used to define
domains. This is not to say that the differences between the domains are
necessarily simple, but only that there should be a relatively
straightforward way of identifying and describing them.
In the case of the two domains formed by differences in altitude,
researchers are not so much interested in altitude per se but rather in
the way altitude is responsible for determining two quite different,
complex patterns of disease and pest incidence, cropping cycle and
varietal preferences. It is these factors that determine the practices
that farmers follow and the innovations that they are likely to adopt.
It is these factors that dictate two separate sets of on-farm experiments
for researchers. Delimiting the two domains in terms of altitude is
simply a convenient way of identifying the domains and helping
researchers to plan their work. It may be that the distinction between
altitude zones is even correlated with other factors such as human
population density, with lower densities at the higher elevations. This
would lead to differences in rotation patterns and soil fertility between
the two domains, even though there is no a prior relationship between
altitude and rotation. Again, the denomination of the high elevation and
the low elevation domains is a convenient way of describing a whole
series of different circumstances among two groups of farmers.
It is often asked if this process of domain formation is adequate
for covering all farmers in a research area. Will there not be a few
farmers in the high elevation domain, for instance, whose practices are
different from the rest? Or might there not be some farmers whose
circumstances are in between the two altitude zones? There may well be,
but are there enough such farmers to make it worthwhile to form separate
recommendation domains? Recall that domains are formed so that
researchers can effectively deal with the majority of farmers in a
particular area. The selection of good criteria for domain formation
will result in a few large domains, each roughly homogeneous with respect
to major research opportunities and current production practices, with
distinct differences between domains. There may be some farmers who are
not covered by the definitions, but forming special domains for them
might not be a wise use of research resources.
4.2) Policy variables in recommendation domain formation
The question of which farmers should be addressed by an experimental
program is not only related to research efficiency,l but also to policy.
If several domains have been identified it is often necessary to decide
which ones will receive attention. Very often national policy will
contribute to making these decisions, as priority may be given to certain
types of farmers (small farmers, connercially-oriented farmers, etc.) or
to certain types of crops (basic grains, cash crops, etc.). As research
policy is usually concerned with obtaining high benefits from a given
research investment, this also often implies concentration on domains
that contain the largest numbers of farmers and present the most
promising opportunities for improving productivity.
The relationship between policy and on-farm research is not one-way,
however. There are substantial opportunities for feedback from on-farm
research to policy makers. In the case of recommendation domains there
is the opportunity for providing policy makers a much clearer idea of the
nature of the farming population. Very often policy mandates are stated
in terms of "target groups" whose definition (e.g. "the small farmers of
region X") masks considerable variation in circumstances and potential.
Dividing the research area into recommendation domains can contribute to
much more precise targeting.
4.3) Acquiring data for domain formation
There is an apparent paradox in the definition of a recommendation
domain. If it is defined as a group of farmers whose circumstances are
similar enough to make them eligible for the same recommendation, how can
we be sure of the constitution of the domain before the recommendation
has been made? The answer is that we often cannot be completely certain,
but as the process of on-farm research passes from diagnosis through
experimentation to recommendations researchers become more and more
confident of the boundaries of their domains. From the beginning of the
research process hypotheses are formed about possible domains. These
hypotheses are tested during surveys and the conclusions are used in the
design of an on-farm experimental program. At times it is only after a
year or more of experiments that researchers are able to make the final
adjustments in their domains.
In order to acquire information useful for domain formation adequate
data collection methods are required. The initial diagnosis must be done
rapidly and efficiently, so that on-farm experiments can be planted as
quickly as possible. Thus elaborate studies which collect great amounts
of detailed information are not appropriate. The idea is to identify
research opportunities and likely recommendation domains and use this
information to begin experiments. Procedures for assessing farmers'
circumstances are described in Byerlee, Collinson et al. (1980). These
procedures include a review of secondary data, an exploratory survey and,
often, a short, well-focused formal survey.
Initial hypotheses on variables for dividing farmers into domains
may be developed during a review of secondary data for the research area.
Keeping in mind the checklist (Table 1) of circumstances which may affect
domain formation, the researchers can examine the secondary data with an
eye towards identifying possible key factors. Soils maps, census reports
or other data may suggest possible sources of variation in farmers'
practices. Conversations with local extension staff can also be quite
valuable. With the initiation of the exploratory survey the evaluation
of these hypotheses may carmence. For example, if census data indicated
three major land tenure classes in the research area the exploratory
survey could be used to ascertain whether these tenure classes had any
important effect on farmers' practices or problems. The exploratory
survey is the time when the checklist is most fully utilized. By talking
to farmers and observing their fields researchers have the opportunity to
decide which circumstances on the list are likely determinants of
differences in farmers' practices.
During the exploratory survey, development of hypotheses on
recommendation domains and hypotheses on research opportunities proceed
together. Researchers strive to understand how different circumstances
lead to different practices and problems, and whether or not these
differences are relevant to the research opportunities that have been
identified. For example, if the important research opportunities in a
maize area appear to be insect control and disease-resistant varieties,
then soil differences may not define recommendation domains. If, in the
same area, the principal research opportunities turn out to be
fertilization and moisture conservation then the difference between maize
farms on sandy soils and those on heavier soils is probably enough to
determine two separate recommendation domains.
There need not be any difference in current farmer practice in order
for a particular research opportunity to divide an area into different
domains. In one area in Honduras both land owners and renters had
similar maize practices, using a maize-fallow rotation which allowed
several years between crops of maize on one piece of land. Research
opportunities for weed control and variety were the same for both groups.
But in thinking about the possibility of intensifying the system by
introducing a cover crop of velvet beans which would allow several years
of continuous maize plantings, the difference between owners (who had
assured access to their plots over time) and renters (who did not) became
important, and defined two different domains with respect to this
opportunity. The interaction between research opportunities and domain
boundaries is therefore quite important.
At the end of the exploratory survey the checklist has been
significantly reduced so that researchers generally have only a few
possible candidates for defining recommendation domains. The exploratory
survey is often followed by a formal survey in which randma sampling and
a short, well-focused questionnaire are employed. Samples for the survey
should be drawn so that each tentative recommendation domain is
represented by at least 25-30 farmers. During the survey, information
should be collected on the "short list" of variables that are proposed
for defining recarrmendation domains as well as on variables that measure
key aspects of farner practice (i.e., practices related to important
research opportunities). Cross-tabulation of "short list" variables by
farmer practice variables will indicate which criteria most strongly and
consistently influence the farmer practice.
The survey analysis should seek to identify a small number of
domains, each as homogeneous as possible, which allow efficient research
on the highest priority themes. The survey may, for instance, define two
recommendation domains with very distinct research opportunities, as in
the example of domains defined by altitude in Peru. In that case,
research on maize varieties (one of several opportunities) was oriented
by farmer responses to a question on principal problems. Those at the
lower altitude indicated a problem with leaf fungus disease, while those
at the higher altitude expressed interest in maize of a shorter cycle
because of frost damage.
In other cases, two domains may share at least some research
opportunities, but require experimentation under different conditions.
Domains that are defined by access to irrigation, for instance, may share
chemical weed control in maize as a research opportunity, but different
products, levels and application methods may be indicated for each
domain. The survey is used in this case to define the circumstances that
are representative of irrigated and non-irrigated domains, in order to
choose the levels of non-experimental variables for each domain.
Particular care must be taken when proposed criteria for domain
formation are proxies for actual practices and conditions. Analysis of
the survey should lead to major, as opposed to merely statistically
significant, differences between domains. For example, in one survey in a
barley producing area tractor ownership was proposed as a criterion for
distinguishing recommendation domains. Analysis of the survey showed
several differences in land preparation between tractor owners and
renters. In the case of early harrowing before ploughing, for instance,
54% of the owners, but only 41% of the renters, did a pre-plough
harrowing. The difference was statistically significant (at 5%) and
showed, not surprisingly, a tendency for tractor owners to do a more
thorough job of preparing their fields than tractor renters. Differences
of this magnitude were observed for several other land preparation
methods. They were not, however, sufficient to define recommendation
domains. Whether or not the farmer did an early harrowing was affected
by competing labor demands, previous crops, soil conditions, and several
other factors besides machinery ownership. Thus more effort should be
made to specify the complex of circumstances that conditions land
preparation methods. The single factor of tractor ownership identified
in the survey, although responsible for statistically significant
differences in practices, is not sufficient to divide the research area
into two clearly distinguishable domains. In the meantime, if research
opportunities are identified which interact with land preparation
(seeding methods and timing, for instance), then research should be
carried out for the major categories of land preparation, using land
preparation itself as a defining characteristic of the domains, rather
than tractor ownership, which is only a weak proxy.-
4.4) Using domains as a framework for on-farm research
Once recommendation domains are identified they are used as the
basis for the on-farm experimental program. Experiments are designed for
specific recommendation domains; the exact number of a certain experiment
to be planted in one domain depends largely on the type of experiment.
If it is an experiment of an exploratory nature then it may be repeated
only a few times, while if it is a verification experiment (the stage
just before demonstration) then it will be very widely distributed within
6/ This example assumes that land preparation itself is not an
opportunity for investigation, which of course may not be the case.
the re nation domain.
the recanmendation domain.-
It should be noted that the number of experiments required for a
given domain does not depend on the size of the domain. It was pointed
out earlier that one can think of domains as statistical strata, and
on-farm experimentation can be considered an exercise in sampling. Each
experiment measures the effect of new elements of technology on crop
yields, income and risk for the respective cooperating farmer. The
benefits of a particular element may be estimated for the target farmer
population by averaging the results of several trials. When strata are
internally homogeneous (as recommendation domains should be), a small
sample from each is sufficient to obtain a precise estimate of the
stratum mean. This is because the sample size needed to achieve a
desired level of precision at a certain level of probability does not
depend on the population size, but rather on its variability.
The experiments are of course planted under conditions
representative of the recommendation domain. If the domain is defined as
all farmers who have less than 10 hectares, have fields between 2,600 and
3,000 meters above sea level, and do not have access to irrigation, then
the experiments for this domain must be planted under these
circumstances. Beyond this, the survey will have specified what the
representative farmer practices are for the particular domain.
Non-experimental variables are usually set at the farmer's level, unless
there is the expectation that farmers will soon adopt a new practice
which warrants being included as a non-experimental variable.
Although recommendation domains can usually be identified before
planting the first year's experiments it occasionally happens that the
results of the experiments themselves are useful in refining domain
definitions. If a domain is a group of farmers who face similar
circumstances, follow similar practices and share similar opportunities,
then one would expect similar results from experiments planted with
SFor more on the stages of on-farm experimentation, see Violic,
Kocher and Palmer (1981).
different members of the same domain. In terms of analysis of variance,
"site by treatment interaction" should not be consistently significant.
When this interaction term is significant (at an appropriate level)
researchers should see if this is merely random variation (e.g. because
of rainfall differences), or if there is a constant factor (e.g. a
previously unidentified difference in soil types) which is responsible.
In the latter case, this may lead to a division of what was formerly one
domain into two or more. Similarly, when experimental results are
consistently uniform across two domains, researchers may consider
combining them into one.
Once domain boundaries are firmly identified, the agronomic,
economic and statistical analysis of experiments proceeds by pooling the
data within each domain. The results are then used to form
recommendations for the domain.
4.5) Preliminary zoning -
At times research programs wish to use a set of tentative domains to
organize OFR in a large area. Senior research planners may feel, for
example, that one domain may occur in numerous small defined areas (each
handled by a different OFR field team). To reduce duplication of effort
in on-farm trials, these senior researchers may wish to make a first
"rough cut" at domain formation, assigning each OFR field team to work
with one or two of them.
In these cases, "zoning" procedures can be used. Specifically,
formation of numerous tentative domains in a large area can be initiated
by means of a very brief formal survey with local extension personnel
that provides data for grouping together farmers with similar farming
systems. As OFR teams are assigned to initiate fieldwork, they can
accept or adjust the tentative domains identified in the zoning process.
SThis section draws heavily on the experience of M. Collinson (1979)
and S. Franzel (1981) in East Africa.
5.0) Issues and Complications
As researchers deal with domain formation in their study areas
several issues and complications tend to arise. These include questions
of domain size, domain permanence and others. The purpose of this
section is to discuss these questions and show how they may be addressed.
What is the appropriate size of a recommendation domain? There is
no set answer to this question, but obviously the larger the domains the
more cost-effective will be the research program.
Domain size is influenced by the heterogeneity of the area. In
places where there are many different microclimates and great variations
in the socioeconomic circumstances of farmers a relatively large number
of domains are likely to be identified. In other places, vast areas may
be subject to similar circumstances and farmer practices, and a few
domains will suffice.
Domain size is also determined by the availability of research
resources. More resources allow the exploration of nore research
opportunities and thus the delineation and management of more and smaller
domains. At tines these factors may have contradictory influences on
domain size, as when work is carried out in a very complex, heterogeneous
target area with very few resources available for implementing OFR. In
these cases a decision is often taken to carry out research in only a few
high-priority domains, selected according to research opportunities,
farmer characteristics or national policy.
Domain size is thus bounded on the small side by expected returns to
research expenditures. Domains should not be so small that benefits from
new technology for that domain are less than corresponding research costs
(or better, less than the expected returns from alternative uses of
research resources). Domain size need not be bounded on the large side.
In fact, domains should be as large as possible, with the condition that
farmers in the domain can still be expected to adopt recommendations
arising from work on major research opportunities. Large domains allow
the fixed costs of on-farm trials to be spread over a wider number of
In practice, domain sizes demonstrate considerable variation. They
have ranged from a few thousand farmers to several tens of thousands, or
more. There is clearly no "best" size for a recommendation domain.
5.2)The permanence of domains
We have already seen that the definition of domains may be refined
during the process of on-farm research. As workers become better
acquainted with the area, their perspective of research opportunities and
agronomic responses will change, leading at times to redefinition of
Similarly, we have seen that domain definitions are linked to
research opportunities, and as research themes tend to shift over time
these shifts often require adjustments in domain boundaries. In one
research area in Ecuador, for example, where maize was the principal
crop, farmers with and without complementary irrigation constituted a
single domain for maize research, as no significant differences in
practices could be detected between these two groups of farmers. But as
research progressed, and especially as an early-maturing maize was
released for farmers which allowed new rotation patterns, the difference
in access to irrigation became important. Rotation possibilities that
included crops grown in the dry season were much different between these
two groups of farmers, and where previously there had been a single
domain two were formed as research advanced.
It must be kept in mind that the simplified, shorthand definitions
of domains really serve to summarize researchers' perceptions of how a
complex of farmer practices and circumstances influence the
identification and development of research opportunities. As these
opportunities change and evolve so do domain definitions. Domains may be
joined, split or otherwise redefined, and researchers should do so when
research efficiency may be improved.
5.3) Correspondence between domains and on-farm experiments
Recommendation domains are formed in order to help researchers
define different experimental programs. At times the difference between
domains in a given research area may be extreme, including different
target crops and completely different research opportunities. In this
case the on-farm experiments planted in the two domains would bear no
relationship to each other. Even when the target crops are the same,
research opportunities sometimes differ so greatly that the maize
experiments (for example) in one domain are totally different from those
Because recommendation domains are partially determined by research
opportunities, it sometimes happens that two domains (with respect to one
opportunity) are included in, and share the experiments of, another
larger domain. As an example, in one wheat area two domains were based
on soil type. The soil type determined land preparation and crop
rotation possibilities and hence strongly influenced the nature of the
weed population. Thus separate sets of weed control experiments were
planted in the two domains. Soil type had no influence on varietal
requirements however, so the area constituted only one domain with
respect to variety, and the number of variety trials planted (across the
two soil types) was appropriate for a single domain. If there were
reasons to suspect an interaction between varietal performance and soil
.type (or the practices determined by soil type), however, two sets of
variety trials would be indicated.
In certain cases, the experimental program may be exactly the same
between domains. In the example of the two domains distinguished by soil
type it may be that similar fertilizer trials should be planted in both
domains in the initial stages of experimentation although the agroncmic
responses and final recommendations will likely be different. Or in the
case of domains distinguished not by soil type but rather by land tenure,
the same fertilizer trials may give the same agronomic response, but
lower net benefits to sharecroppers than to owners will mean somewhat
different fertilizer recommendations for the two domains. In this latter
case, a single set of experiments may suffice for deriving the two
As national agricultural research programs move toward on-farm
research, the need grows for a way to specify the clientele for that
research. The recommendation domain concept can fill this need.
Conceptually, a domain is a group of farmers with similar
circumstances who are eligible for the same recommendation.
Operationally, domains are formed around farmers with similar practices
for a given enterprise and for whom researchers see similar opportunities
for the improvement of these practices. Such farmers can be grouped
together in terms of biological and/or socioeconomic variables.
Recommendation domains are useful as a framework for on-farm
research. As researchers strive to select the few most important
experimental variables and then study them under representative
conditions, domains provide the necessary context for defining
"important" and "representative". Recommendation domains also provide a
criterion for pooling the data obtained from on-farm trials, thus
resolving the classic problem of extrapolating research results beyond
the farms on which trials are conducted.
Domains are formed by considering farmers' circumstances. As
researchers begin their work, they are interested in how these
circumstances affect farmers' practices and how they condition research
opportunities. As ideas for research opportunities emerge so do clear
definitions of recommendation domains. Beginning with a comprehensive
list of farmers' circumstances researchers conduct an informal survey
which helps to eliminate many of these as potential criteria for defining
domains. A reduced number of possibilities may be tested through a
formal survey to see if they are in fact useful for dividing farmers into
roughly homogeneous groups who could benefit from the same
recommendation. This information is utilized in the design and planting
of on-farm experiments. It may be that the final domain boundaries are
not decided upon until the experimental results are analyzed. In any
case, by the time recommendations are ready, they are already targeted to
well-defined groups of farmers.
In providing a framework for on-farm research, recommendation
domains are a useful tool. Like all tools, however, they are most
helpful when used with imagination and care.
Byerlee, D., M. Collinson, et al. (1980) Planning Technologies
Appropriate to Farmers: Concepts and Procedures Mexico: CIMMYT.
Byerlee, D., L. Harrington and D. Winkelmann (1982) "Farmings systems
research: Issues in research strategy and technology design" American
Journal of Agricultural Economics 64(5):897-904.
Byerlee, D. and E. Hesse de Polanco (1982) "The rate and sequence of
adoption of improved cereal technologies: The case of rainfed barley in
the Mexican Altiplano" CIMMYT Economics Working Paper.
Collinson, M.P. (1979) "Deriving reccamendation Idomains for Central
Province, Zambia. Demonstrations of an interdisciplinary approach to
planning adaptative agricultural research programmes", Report No. 4
CIMMYT Eastern Africa Economics Program.
Franzel, S. (1981) "Identifying farmer target groups in an area:
Methodology and procedures" CIMMYT East African Economics Program.
Harrington, L.W. (1982) "Exercises in the economic analysis of agronomic
data" CIMMYT Economics Working Paper.
Kirkby, R., P. Gallegos, T. Cornick (1981) "On-farm research methods: A
comparative approach. Experiences of the Quiniag-Penipe Project,
Ecuador" Cornell International Agriculture Mimeograph #91. Ithaca.
Moscardi, E. et al. (1982) "Creating an on-farm research program in
Ecuador. The case of INIAP's Production Research Program" CIMMYT
Economics Working Paper.
Palmer, A.F.E., A.D. Violic and F. Kocher (1982) "Relationship between
research and extension services and the mutuality of their interests in
agricultural development" Mexico: CIMMYT.
Perrin, R.K., D.L. Winkelmann, E.R. Moscardi and J.R. Anderson (1976)
Fran Agronomic Data to Farmer Recommendations. An Economics Training
Manual Mexico: CIMMYT.
Tripp, R. (1982) "Data collection, site selection and farmer
participation in on-farm experimentation" CIMMYT Economics Working
Violic, A., F. Kocher and A.F.E. Palmer (1981) "Research and technology
transfer" Paper presented at the IICA/BID/INIA Seminar on Generation of
Information and Transfer of Technology, Vifia del Mar, Chile, 23-27
LIST OF AVAILABLE CIMMYT ECONOMICS WORKING PAPERS
81/1 Kwasi Bruce, Derek Byerlee and G. E. Edmeades, "Maize in the
Mampong Sekodumasi Area of Ghana; Results of an Exploratory Survey".
81/2 Derek Byerlee and Donald L. Winkelmann, "Accelerating Wheat Pro-
duction in Semi-Arid Developing Regions: Economic and Policy Issues".
*81/3 Edith Hesse de Polanco and Peter Walker, "A Users Guide to
FASAP- A Fortran Program for the Analysis of Farm Survey Data".
*81/4 Alan Benjamin, "An Agro-Economic Evaluation of Maize Production
in Three Valleys of the Peruvian Andes".
*81/5 Derek Byerlee, Larry Harrington and Paul Marko, "Farmers' Prac-
tices, Production Problems and Research Opportunities in Barley Pro-
.duction in the Calpulalpan/Apan Valley, Mexico".
81/6 Larry Harrington, "Methodological Issues Facing Social Scien-
tists in On-Farm/Farming Systems Research".
*82/1 Larry Harrington, et al., "Maize in North Veracruz State, Mexico-
-Farmer Practice and Research Opportunities".
*82/2 Larry Harrington, "Exercises in the Economic Analysis of Agro-
**82/3 J. C. Martinez, "Desarrollando Tecnologia Apropiada a las Cir-
cunstancias del Productor: El Enfoque Restringido de Sistemas de
82/4 Robert Tripp, "Data Collection, Site Selection and Farmer Par-
ticipation in On-Farm Experimentation".
82/5 Robert Tripp, "Including Dietary Concerns in On-Farm Research:
An Example from Imbabura, Ecuador".
82/6 Derek Byerlee and Edith Hesse de Polanco, "The Rate and Sequence
of Adoption of Improved Cereal Technologies: The Case of Rainfed
Barley in the Mexican Altiplano".
*01/83 Edgardo Moscardi, et.al., "Creating an On-Farm Research Program
in Ecuador: The Case of INIAP's Production Research Program".
*02/83 J. C. Martinez and Jose Roman Arauz, "Institutional Innovations
in National Agricultural Research: On-Farm Research within IDIAP,
03/83 James B. Fitch, "Maize Production Practices and Problems in
Egypt: Results of Three Farmer Surveys"
04/83 J.C. Martinez and G. Sain, "The Econcmic Returns to
Institutional Innovations in National Agricultural Research: On-Farm
Research in IDIAP Panama"
05/83 Derek Byerlee, "The Increasing Role of Wheat Consumption and
Imports in the Developing World"
01/84 Juan Carlos Martinez, "La Mise au Point D'Une Technologie
Adaptee aux Contraintes et Atouts de L'Agriculteur: L'Approche du
* Available in English and Spanish
** Available in Spanish only
*** Available in French only