Th Adopio of Agrclua
A Guid fo Suve Deig
The Adoption of
A Guide for Survey Design
CIMMYT Economics Program
CIMMYT is an internationally funded, nonprofit scientific research and training organization. Head-
quartered in Mexico, the Center is engaged in a research program for maize, wheat, and triticale, with
emphasis on improving the productivity of agricultural resources in developing countries. It is one of 17
nonprofit international agricultural research and training centers supported by the Consultative Group
on International Agricultural Research (CGIAR), which is sponsored by the Food and Agriculture
Organization (FAO) of the United Nations, the International Bank for Reconstruction and Development
(World Bank), and the United Nations Development Programme (UNDP). The CGIAR consists of some
40 donor countries, international and regional organizations, and private foundations.
CIMMYT receives core support through the CGIAR from a number of sources, including the interna-
tional aid agencies of Australia, Austria, Belgium, Brazil, Canada, China, Denmark, Finland, France,
India, Germany, Italy, Japan, Mexico, the Netherlands, Norway, the Philippines, Spain, Switzerland, the
United Kingdom, and the USA, and from the European Economic Commission, Ford Foundation, Inter-
American Development Bank, OPEC Fund for International Development, UNDP, and World Bank.
CIMMYT also receives non-CGIAR extra-core support from the International Development Research
Centre (IDRC) of Canada, the Rockefeller Foundation, and many of the core donors listed above.
Responsibility for this publication rests solely with CIMMYT.
Correct citation: CIMMYT Economics Program. 1993. The Adoption of Agricultural Technology: A Guidefor
Survey Design. Mexico, D.F.: CIMMYT.
AGROVOC descriptors: Innovation adoption, agricultural development, farmers, technical progress,
AGRIS category codes: E14, E10
Dewey decimal classification: 338.064
Editing: Kelly Cassaday
Design: Miguel Mellado E.
Layout: Eliot Sanchez P. and Miguel Mellado E.
Printed in Singapore.
Why Study Adoption?
Ways of Studying Adoption
Describing Adoption: The Logistic Curve
The Role of Adoption Studies in Assessing the
Returns to Research and Extension
The Factors That Influence Farmers' Decisions
Farm Resource and Farmer Characteristics
The Farming System
Post-Harvest Utilization and Markets
Timing of the Survey
Methods for Analyzing Adoption Patterns
This manual presents methods for designing formal surveys that measure
and analyze the adoption of agricultural technology. First it discusses the
rationale for undertaking an adoption study and the issues involved in
determining the study's audience and scope. The manual then reviews
factors that influence adoption patterns and should be considered in
designing the survey: farm resources (e.g., farm size) and farmer
characteristics (e.g., age, gender, education), the farming systems, post-
harvest utilization and markets, and farmers' sources of information about
new technology. Next, the organization of adoption surveys is discussed,
giving attention to sampling issues and the timing of the survey as well as
to overall survey design and implementation. The final section describes
several analytical techniques that are useful for interpreting the results of
adoption studies. Numerous examples from adoption studies conducted in
Asia, Africa, and Latin America illustrate the main points.
This document is the latest in a series of manuals on research methods
produced by the CIMMYT Economics Program. Our previous manuals
have addressed themes of on-farm research planning and analysis,
whereas this manual is concerned with methods for monitoring and
evaluating the results of agricultural technology generation and transfer
The manual focuses on the design of surveys that assess and analyze the
adoption of agricultural technology. We are convinced that not enough
effort has gone into carrying out and utilizing formal adoption studies. We
have found that although there are a number of good sources on survey
design and analysis, there is little available that could help researchers
organize and plan a survey whose purpose is the measurement and
analysis of technology adoption patterns.
As more attention is paid to the evaluation of agricultural research, and as
research budgets come under more scrutiny, it is imperative that
researchers have the capacity to measure the outcomes of their work and to
use this information to refine their strategies and to document impact. We
believe the methods described in this manual will be useful not only for
agricultural researchers, but also for technology transfer agencies, rural
development programs, and non-governmental organizations, all of whom
need to be able to monitor and assess the progress of their work.
The first draft of the manual was written by Robert Tripp, who also
managed the review process. Paul Heisey collaborated on subsequent
drafts, which also benefited from contributions by Daniel Buckles, Michael
Morris, Martien van Nieuwkoop, Larry Harrington, Miguel L6pez-Pereira,
Wilfred Mwangi, and Gustavo Sain. In addition, Louise Sperling (of the
Centro Internacional de Agricultura Tropical, CIAT) and Willem Janssen
(formerly of CIAT) provided thorough reviews and suggestions for
improving the document.
We hope the manual will prove useful for helping to understand the
adoption process. We welcome suggestions and comments from those who
CIMMYT Economics Program
Why Study Adoption?
There is no more distinctive feature of agriculture than its dynamism.
Farming practices change continually. Farmers build on their own
experience and that of their neighbors to refine the way they manage their
crops. Changes in natural conditions, resource availability, and market
development also present challenges and opportunities to which farmers
respond. In addition, farmers learn about new technologies from various
organizations, programs, and projects dedicated to research, extension, or
rural development. These organizations develop and promote new
varieties, inputs, and management practices. It is essential that such
organizations be able to follow the results of their efforts and understand
how the technologies they promote fit into the complex pattern of
agricultural change in which all farmers participate.
This manual is focused on one method for studying agricultural change. It
describes the design and management of farm surveys used to document
and explain farmers' adoption of agricultural technology. Too little
attention has been paid to this aspect of the process of agricultural
technology development. The aim of this manual is to help strengthen
institutional capacity to carry out such adoption studies. This capacity is
important for agricultural research organizations that develop innovations
for farmers, extension institutions that promote new technology, various
types of rural development projects that introduce changes in agricultural
technology, and a range of non-governmental organizations (NGOs) and
community level efforts that are working to improve farming practices.
There are several reasons to invest in studying the adoption of agricultural
technology. These include improving the efficiency of technology
generation, assessing the effectiveness of technology transfer,
understanding the role of policy in the adoption of new technology, and
demonstrating the impact of investing in technology generation. Each of
these is now discussed in more detail.
aBWs Monitoring and feedback in technology generation
Any program that attempts to develop and promote improved farming
practices should be able to assess progress and use that information to
make future actions more effective. One of the principal incentives behind
the development of adaptive research methods such as farming systems
research (FSR) or on-farm research (OFR) was the criticism that much
agricultural research was being done on experiment stations, isolated from
the fields, problems, and perspectives of client farmers. Many national
agricultural research programs have now established location-specific
adaptive research capacity that includes diagnostic surveys and on-farm
experimentation. But few of these organizations regularly monitor
technology adoption to improve the efficiency of adaptive research. It is
not uncommon to find that a well-conceived program of agricultural
research and extension has been carried out for a number of years in a
given area but that none of the personnel involved can give any more than
anecdotal evidence of changes that have taken place in farmers' practices,
let alone the reasons for these changes. In many cases, the adaptive
research is in danger of straying far from farmers' needs unless researchers
have a way of monitoring farmers' experience with the new technologies
being generated. For example, a new variety may have been tested on
farmers' fields and promoted by extension. But unless some sort of
monitoring is carried out, researchers will not know the degree to which
the variety is actually being used. In addition, it will be very helpful to
know what farmers see as the advantages of the new variety and what they
perceive as its drawbacks, in order to provide feedback to plant breeders
for refining their selection criteria.
JiBia The effectiveness of technology transfer
Most extension services are actively engaged in promoting new
technologies with farmers. Resources are invested in various extension
activities, such as field days or demonstrations, and the extension service
may undergo considerable reorganization, such as with the training and
visit (T&V) system (Benor and Harrison 1977). But only infrequently are
resources reserved for monitoring the outcome of these extension efforts
and using the analysis to understand why some recommendations or
extension techniques are more successful than others. For example, if the
extension service is recommending a green manure crop, it will be very
useful to know what proportion of farmers are using the new practice. For
those farmers who have not adopted, do they find disadvantages with the
new practice, is the practice too far removed from farmers' knowledge
base, or has the extension methodology not been effective in acquainting
these farmers with the new technique?
Governments or donor agencies sponsor rural development projects in
which the introduction of new agricultural technology plays an important
role. Although these projects may represent a large investment of funds,
the capacity to monitor progress is rarely a focus of project design. For
instance, a development project may operate under certain assumptions
about the possibility of improving tillage operations, and the effects this
will have on yields and income. It is important to follow the degree to
which project participants are actually changing their tillage practices and
identifying any problems that have occurred.
It is often difficult to draw a line between technology generation and
technology transfer. An effective adaptive research effort should involve
both researchers and extension agents, for instance. There is growing
emphasis on participatory research as well, in which farmers assume more
responsibility for the identification and dissemination of new technologies.
One good example is the efforts of NGOs to improve agricultural practices
at the community level.
Many of these projects are effective at identifying the priorities of farmers
and enlisting widespread participation of community members to
investigate and promote innovations. But these projects rarely go to the
trouble of documenting their results and assessing their progress to make
future actions more effective. A project may generate considerable
enthusiasm about the importance of improved crop storage, for instance,
and several options may be available for farmers to try. But it is important
to follow up on the actual number of farmers who make a change, to
analyze which of the storage options they find most attractive, and to
understand farmers' choices.
Thus there is a widespread need to place additional emphasis on
monitoring the results of technology transfer and eliciting farmers'
feedback. Organizations responsible for developing new technology need
to know if the transfer process is functioning. Organizations responsible for
promoting technology need to know if their message is being heard. And
community or regional development efforts need to judge to what extent
technological change is contributing to their goals.
-- The role of policy in technology adoption
Monitoring progress is necessary not only to improve the internal efficiency
of research and extension efforts, but also to improve the effectiveness of
interactions with other institutions, particularly those responsible for
policy. Very often a research or extension effort falls well short of its goals
because of lack of coordination between institutions. Adoption studies may
show the potential for technology diffusion by demonstrating progress in
areas where institutional coordination is good, or may analyze the
problems in areas where technology diffusion has been slow.
Adoption studies are also useful for illustrating the degree to which
acceptance of new technologies is limited by insufficient inputs, credit, or
marketing infrastructure. If it appears that farmers are unable to take
advantage of a new technology because they lack inputs, this information
can be presented to policymakers who have responsibility for the
agricultural inputs that are available and the way they are distributed. If an
adoption study shows that access to credit significantly influences the type
of technology that farmers use, then this information may be presented to
those responsible for designing and funding credit programs. Similarly,
adoption studies may be used to highlight marketing bottlenecks that limit
the acceptability of new technologies.
Effective communication between researchers and policymakers is not very
common. It will take more than a few adoption studies to establish good
links among researchers, extension personnel, national policymakers, and
public interest groups. But the information from a well-conceived and
effectively presented adoption study can be very useful for improving this
type of communication.
UI Measuring the impact of technology generation and transfer
Another important use of the information from adoption studies is to assess
the impact of agricultural research and extension and to measure the
returns to investments in these activities. Research and extension
institutions are often engaged in a battle to maintain their budgets, and this
implies the necessity for demonstrating results. Adoption studies are an
important tool for measuring and assessing impact. They also provide data
that can be used to estimate the returns to investment in research or
extension. Such an analysis may be used to justify further investment in
these sectors or to help identify the most productive opportunities for
investment within research or extension.
An important question on the minds of policymakers is who benefits from
new technology. Adoption studies may be designed to document what
kinds of farmers and what areas of the country have profited most from the
development of a particular technology.
The evaluation of impact and returns to investment is also a common
feature of rural development projects, but these evaluations are often done
without access to solid data on adoption. Even NGO projects need to spend
more time documenting progress and analyzing the effectiveness of their
investments. As more donor attention is directed to the option of NGO
contributions to agricultural change, these organizations will come under
increasing pressure to present well-documented evidence of their
l Contributing to the literature on adoption
There is a very large body of literature on the adoption of agricultural
innovations (Rogers 1983; Feder, Just, and Zilberman 1985). The methods
described in this manual can add to that literature, although their principal
purpose is to serve institutions involved in promoting agricultural change
rather than to contribute to the theory of adoption. Many academic studies
on adoption assume that the technology is appropriate and tend to
concentrate on identifying the characteristics of farmers who are likely to
adopt. The kind of study described in this manual faces the more difficult
challenge of not only describing patterns of adoption but also
understanding whether or not the technology and its institutional
environment are adequate to the needs and resources of farmers. This type
of adoption study must be done with an open mind; its purpose is not to
promote a particular technology but rather to help research and extension
be more effective in responding to farmers' needs.
The following chapters are concerned with adoption studies that help
describe, evaluate, and understand the process of technological change. It
is assumed that the principal audience for the studies will be staff of
research and extension institutions (public or private), and of other
institutions at the national level that are responsible for assessing or setting
policies and allocating funds that determine the scope and direction of
Ways of Studying Adoption
- The process of agricultural research
Although this manual focuses on the design and analysis of formal
adoption surveys, it would be misleading to think of adoption as
something that is the subject of a single study conducted at the end of a
research effort. Monitoring changes in farming practices and assessing the
adoption of new technology should be important elements of the entire
When a program of research or extension is being planned, it is essential to
get a clear idea of what type of changes or technologies would be
acceptable to farmers. Diagnostic surveys provide information on farmers'
current practices and concerns (Byerlee, Collinson, et al. 1980). To the
extent that these surveys assess the distribution and rationale for farmers'
present use of technology, they can be thought of as "adoption studies" for
previous technology generation efforts. Information from such surveys and
from other sources needs to be carefully considered in planning a research
agenda (Tripp and Woolley 1989), and a growing number of techniques are
available for improving farmer and community participation in the
planning process (Farrington and Martin 1988).
As research is carried out, and especially as experiments are planted in
farmers' fields, it is essential to obtain continuous feedback from farmers. It
is a waste of resources to conduct several years of research on a technology
only to discover that farmers find it unacceptable. There are several ways
of monitoring on-farm experiments. One basic strategy is simply to make
sure that farmers are consulted when researchers or extension agents visit
field sites and that farmers' opinions are recorded and analyzed (Tripp
1982). Other techniques are available for ensuring that farmers participate
in the evaluation of technologies being tested. Guidelines have been
designed to help elicit and utilize farmers' reactions to experimental
treatments (Ashby 1990) (see Box 1 at the end of this chapter). Farmer
groups have also been helpful in planning and monitoring on-farm
experiments (Norman et al. 1988).
NWM Methods for assessing adoption
Although constant monitoring of farmers' opinions and experience is
essential during the design and testing of agricultural technology, it is also
necessary to carry out some sort of assessment after a new technology has
been recommended or introduced. The type and the timing of the
assessment will depend on the purposes of the study.
This manual describes the design and analysis of formal survey
instruments for assessing the adoption of agricultural technology. A formal
survey of technology adoption is one of several kinds of studies that can be
done to assess adoption (Box 2). As mentioned in the previous section, it is
important to have a continual interchange between farmers and researchers
as technology is being developed and tested, and this interaction provides
the first indication of whether or not a new technology is acceptable.
Another way of assessing a technology's acceptability is by following up on
what farmers who have hosted experiments do the following year.
Once a technology has been released or an extension program has been
initiated, it is possible to study a random sample of farmers to analyze the
degree of adoption. An informal survey (similar to the informal diagnostic
surveys used to help set priorities for a research program) is very useful for
providing researchers with preliminary feedback about the acceptability of
a technology. It can also provide information about policy-related problems
that may impede the spread of a technology. An informal survey may be
sufficient for analyzing adoption patterns, but more often the type of
formal survey described in this manual is necessary. Formal surveys
generate quantitative information that is useful to decision makers and are
better able to explore some of the complex issues in understanding
variability in adoption among farmers. But it is assumed that such a survey
will be carried out as part of a research or extension effort that has been
well planned and executed and has included various opportunities for
assessing farmers' opinions and practices along the way. It is also assumed
that the design of the questionnaire is preceded by a good informal survey
that helps researchers identify key issues to be pursued in the
The results of a formal adoption study can be combined with other data on
changes in farm production, farm incomes, or consumer gains to develop a
more complete impact study. There are also other ways of studying the
spread of a new technology. Data from an agricultural census may provide
some idea of the degree to which farmers use a particular technology. If a
new technology involves purchased inputs, for instance, surveys of input
merchants may be useful for assessing the spread of the technology.
All institutions that are involved in generating agricultural technology
should have the capacity to carry out studies that document the degree of
adoption and help explain the rationale for farmers' decisions. Adoption
studies can be useful for several purposes, and a decision about the
audience for the study must be taken before the study is designed.
The information from an adoption study can be used to:
1) provide feedback from farmers that is helpful in refining the
technology generation effort;
2) assess the effectiveness of a technology transfer strategy;
3) improve the flow of information between research and extension, on
the one hand, and policymakers, on the other; and
4) document the impact of a technology generation or extension effort.
This manual describes the design of formal surveys for studying adoption.
A formal survey is a particularly useful method, but it must be seen as part
of a wider range of activities that are used to make technology generation
as efficient and as responsive to farmers' needs as possible.
After defining the audience for an adoption study, the next step is to
carefully define what changes are going to be measured. The following
chapter describes various factors related to measuring adoption.
Cassava Varieties Open Evaluation
HA5 H/GH STARCH, NOT "WATER"~ "DRY", 1$ "FLOURY". T7E "SKIN" IS WHITE, AND FLESH
"CREAMY", A DISADVANTAGE BECAUSE PINK SK/N IS GETT/N6 A BETTER MARJXET PRICE,
THIS PLANT /S "MEDIUM" IN HE/16T
"I LiKE TH/5 BECAUSE VERY TALL PLANTS ARE DIFFICULT TO HARVEST. BUT /T BRANCHES
VERY CLOSE TO THE GROUND. TA/S MAKES WEEDING DIFFICULT. TH7/ WILL HAVE TO
PLANTED FURTHER APART TO MAKE WEEDIN6 EASIER., SO THE PRODUCTION WILL BE
LOWER. "7T/5 HAS A &OOD NUM/3ER OF ROOTS THE YIELD WILL sE ~oOD'O
"T /5 DIFFICULT TO HAR VEST. LOOk' AT THE BROKEN ROOTS".
/ LIKED :(CAUSES STORAGE LOSSES DUE TO
ROT WiEN ROOT /S DAMAGED)
I WILL NOT PLANT THIS A6AIN BECAUSEE
Y/ELD WILL 1E LOW AND THERE WILL
BETTER HAYVEST LOSSES "
Code for Comments:
(b) Plant height "MEDIUM"
(c) Height of branching LOW DIFFICULT TO WEED
(d) Resistance (disease/pest)
(e) Period(s) for harvest
(f) Root appearance
(g) Root rot
(h) Starch content DRY, FLOURY
(i) Color of epidermis WHITE
(j) Color of flesh (pulp) CREAMY
(k) Root position on stem NO PEDUNCULE -ATTACHED TO STEM
(I) No. of roots ASSOCIATED WITH H16H YIELD
General evaluation DISLIKED-LOWBRANCHIN6 (YIELD)
Source: Ashby (1990).
Source: Ashby (1990).
opinions of technology;
farmer participation in
Follow-up on accept-
ability with farmers who
have participated in
or group interviews)
Informal survey of
Formal survey of
Studies of technology
use based on secondary
data (e.g., agricultural
Interviews with input
suppliers (e.g., seed
2-4 years after
release of technology
and/or initiation of
2-4 years after
release of technology
and/or initiation of
2-5 years after
release of technology
and/or initiation of
2-4 years after
release of technology
and/or initiation of
2-4 years after
release of technology
and/or initiation of
10-20 Refine research objectives
to meet needs and
conditions of farmers
10-20 See if farmers keep using
technology. Identify whether
there are problems with its
20-40 Provide feedback to researchers
on feasibility of technology
and feedback to policymakers
on accessibility of technology.
The study is a necessary step
for designing a formal survey.
60-120 Provide feedback to
researchers, information for
policymakers. Contribute to
60-120 Combine data on adoption
from formal survey with
estimates of yield/income
gains and estimates of research
and/or extension program costs.
n.a. Use secondary data (such as
agricultural census) to
assess spread of new
5-20 Estimate demand for technology.
Detect bottlenecks in input
Source: This table borrows from a similar one developed by W. Janssen for training courses at CIAT.
a These sample sizes are only suggestive and may vary outside the ranges listed here, depending on the
purpose of the survey and the proposed analysis.
n.a. = not applicable.
One of the most important issues in designing an adoption study is the
definition of criteria for adoption. If we are interested in the diffusion of a
new variety, for instance, what constitutes adoption? Are farmers who
plant even a few rows of the new variety considered adopters, or do they
have to plant a certain minimum proportion of their fields with the new
variety? If we are interested in the adoption of crop management practices,
how closely does the farmer have to follow a recommendation before being
considered an adopter? Is any fertilizer use to be counted as adoption, for
instance, or does the rate and timing of application have to fall within
Although these may seem to be definitions that can be decided after the
survey is completed, they need to be discussed beforehand because they
can influence the sorts of questions asked to the farmer. An example of
definitions of adoption for a survey that examined changes in weed control,
planting practices, and tillage is shown in Box 3 at the end of this chapter.
In defining the criteria for adoption, it is also important to remember that
although recommendations may be presented to farmers as a package of
several practices, some components of the package may be adopted first,
others may be adopted later, and some may never find widespread
acceptance. The adoption study should therefore ask specifically about
each component of the package, bearing in mind that individual
components may be adopted at different times or under different
In many cases a survey will examine technological change in circumstances
where farmers have several options from which to choose. For instance, the
objective of an extension program may be to acquaint farmers with the
principles of erosion control. Farmers may be able to take advantage of
several appropriate crop management practices, and the survey should
explore which ones farmers are using and the rationale for their choices.
Similarly, farmers may make their own modifications to a new technology
(such as a storage technique or a piece of machinery), and an adoption
study needs to pay careful attention to this type of farmer innovation.
The adoption of a new technology may have implications for the rest of the
farming system, and these attendant changes may be examined in an
adoption study. Researchers will be pleased to see the widespread
adoption of a new variety, for instance, but what effects does this change
have on the use of other varieties and the genetic diversity in farmers'
fields? In other cases the adoption of a new variety may bring about
significant changes in other management practices. An example is shown in
Another issue in measuring adoption is the fact that farmers often have
several fields that may be subject to different management practices.
Researchers need to decide whether to assess adoption on all fields or only
the largest field, or on fields that have characteristics relevant to the new
technology (e.g., examining soil conservation practices only on sloping
fields). The answer to this question depends in part on whether it is
necessary to estimate the total area where a particular technology is in use.
If, for instance, researchers wish to estimate the proportion of crop area in a
given region planted to improved varieties, then either all fields should be
included in the survey, or fields rather than farmers should be randomly
sampled. Asking farmers to estimate total area planted to different
varieties is usually not difficult, but if the practice under investigation is
more complex, such as the rate and timing of fertilizer application, then
estimating the range and average for these practices will be more time
consuming. If the average rate of input application is reported in the
survey analysis, it should be clearly stated if the rates are only for those
farmers who use the input, or are the average across users and non-users.
Investigating practices by individual field also has other advantages. One
of the important ways of explaining differences in adoption behavior
between farmers is to look at related aspects of crop management. It may
be hypothesized that a particular weed control practice is used only in
intercropped fields, for instance. This hypothesis on adoption can be tested
only with field-specific information.
Sometimes there is more than one agricultural season per year. In this case,
adoption may be assessed for only the most important season or for all
seasons (Box 5).
In summary, the adoption of a new technology can be defined in several
ways. In all cases, the definition of "adoption" needs to be agreed upon.
Sometimes it may be sufficient simply to report on the proportion of
farmers using the technology (at some defined level). In other cases, the
actual proportion of fields or crop area under the new technology will need
to be estimated. An example of several ways of reporting these results is
shown in Box 6.
Describing Adoption: The Logistic Curve
Many adoption studies go beyond an analysis of current practices and
attempt to document adoption history. Information about past seasons
requires more time to obtain, but can be very useful. Ideally, information
on past practices and adoption history would come from baseline surveys,
but such information is often not available.
Not all adoption studies will want to analyze historical change, but such
analysis can be useful for several purposes. It may help project future
demand for inputs, determine whether extension needs to be strengthened,
or quantify the change in the number of technology users over time to
It is useful to distinguish between adoption, which is measured at one
point in time, and diffusion, which is the spread of a new technology across
a population over time (Thirtle and Ruttan 1987). Much of the literature on
diffusion assumes that the cumulative proportion of adoption follows an
S-shaped curve in which there is slow initial growth in the use of the new
technology, followed by a more rapid increase and then a slowing down as
the cumulative proportion of adoption approaches its maximum (which
may be well below 100% of the farmers).
The most common function used to portray the curve is the logistic
function. For technology adoption, the y-axis represents the proportion of
farmers or area adopting a technology and the x-axis represents time.
This curve can be described mathematically:
Yt= K /( + e -bt),
Yt = the cumulative percentage of adopters or area at a
K = the upper bound of adoption;
b = a constant, related to the rate of adoption; and
a = a constant, related to the time when adoption begins.
If we have sufficient observations on Yt, we can estimate the three
unknown parameters K, a, and b with a non-linear regression. For practical
purposes, however, this very difficult technique can be replaced with an
ordinary least squares regression if we have at least three observations on
Yt and we can estimate K (the maximum adoption expected)
independently. In this case, we note that the equation of the logistic curve
can be transformed to:
In( ) = a + bt.
Simple ordinary least squares regression of the transformed variable
In [Yt /(K Yt ) on a constant and time will then yield estimates of a and b
(Griliches 1957). This kind of calculation is easy to do with many
spreadsheet packages. One could also fit a curve without regression with
only two observations, although the information from only a few
observations is likely to be limited (see below).
There are several methods of estimating K. The first is simply to plot the
data and to choose the level that appears to be the upper bound of
adoption. A second method is to run the regression using different values
of K and choose the one that maximizes R2. (This is also readily
accomplished with most spreadsheets. It should be noted that in general, R2
and t-statistics from these regressions have no statistical meaning. This
technique only helps to choose K to get a fairly close fit to the data.) To
reduce the time spent selecting K, a combination of simple plotting and
experimental regressions might be used.
Although the logistic curve is the most common way of representing
technology diffusion, it is well to remember that it is based on certain
assumptions about diffusion, and that the fixed parameters estimated for
the curve imply that the relevant price ratios, infrastructure, and the
technology itself have remained constant over the period when the curve is
fitted. An example of a logistic curve is shown in Box 7.
It is also important to remember that not only is the diffusion of a new
practice among farmers a gradual process, but that individual farmer
testing of technology may follow the same type of curve. If possible, a
farmer will test a new technology on a small part of the farm, and if the
results are positive will gradually increase the use of the technology. Box 8
shows this process for a new variety.
If an adoption study examines a number of different practices, whether or
not they have been presented to farmers as a technological package, it is
important to consider the relationships among the adoption patterns. In
some cases, different elements may be adopted independently, while in
other cases there may be a sequential adoption pattern, as shown in the
example in Box 7. Sometimes certain elements will likely be adopted
together, either because of biological complementarities between them or
because farmers are provided incentives (e.g., a credit package).
Although it is important to remember that actual diffusion patterns may
not follow the smooth theoretical curves, historical data on adoption
provide valuable information about trends and prospects for a new
technology. These data allow one to see when adoption began and to judge
the degree to which research or extension programs were in fact
responsible for the introduction or spread of the technology. This
information also allows for an estimate of the rate of adoption and
predictions about future progress.
One problem with these estimates, however, is that they assume
cumulative adoption that is, once a farmer begins using the technology,
he or she will keep using it. In some cases this is not correct, and many
farmers may have one or more years of experience with the technology
only to have subsequently abandoned it. One way of investigating this
phenomenon is to compare current use with past use (Box 9). It may be that
a significant proportion of farmers has experience with the technology but
very few currently use it. If this is the case, it is worth trying to get
information on why farmers have stopped using the technology. This
comparison between past and current use is relatively straightforward.
More detailed information on historical patterns of adoption, use, and
disadoption requires questioning on year-by-year use of a technology,
which is often quite difficult.
Besides looking at past patterns of adoption, it is sometimes tempting to try
to assess future trends by asking farmers their plans for using a technology
the following year. Although it is often valuable to obtain farmers' opinions
about the feasibility of using a technology and identifying what its
attractions and drawbacks might be, this information cannot be used to
assess adoption. Statements about what a farmer would like to do, is
interested in, or hopes to do, are not substitutes for data on actual
The adoption literature also refers to differences between early and late
adopters (Rogers 1983). In the case of technologies that depend on
purchased inputs, for instance, the first farmers to adopt a new technology
may be larger-scale farmers or those with more resources or capacity to
experiment with new practices. In some cases a technology may be
appropriate only for this type of farmer and does not diffuse any further. In
other cases farmers who have fewer resources also adopt the technology, or
it may be that the technology is in fact more appropriate for these farmers.
In any case, it may be important to draw a distinction between earliness of
adoption and the current degree of adoption (Box 10).
Earlier sections discussed ways of estimating the degree of adoption of a
new technology. These included measures of the proportion of farmers,
cropped area, or harvest. If the adoption study has been done to help
provide some measure of the impact or importance of the research or
extension effort, it will be helpful to convert these figures so that the actual
amount or value of the increased production (or other benefits) resulting
from adoption can be estimated. This may require some additional
questions on the survey, as well as complementary information.
If the benefits of the new technology are largely expressed as increased
yield, the first step is to estimate yield changes due to adoption. There are
several ways of doing this. The adoption survey itself may be a source of
yield data. Farmers may be asked to estimate their yields from particular
fields, or crop cuts may be taken as part of the survey. Reliable yield
estimates, either by farmers' reports or by crop cuts, are not particularly
easy to obtain, however, and the reader is referred to Poate and Casley
(1985) for advice on appropriate methods. Even if good yield estimates are
obtained from the survey, it will be difficult to find farmers who manage
comparable fields in which the only difference is the adoption of the
technology under study, or to find comparable farmers who use and do not
use the technology, to provide firm estimates of yield differences that can
be attributed to adoption. Year-to-year variations caused by climatic factors
make it very difficult to use data from the same farmer across several years
to estimate yield changes due to technological change.
A better way of obtaining data on yield differences that have occurred
because of the new technology is through experimental data. If the
recommendations have been derived from on-farm experiments, then yield
estimates should be available comparing farmers' practice with the new
practice. Caution must be exercised in using experimental data, however,
to ensure that the yield estimates for the new technology were obtained
under typical farmers' management, rather than researchers' management.
Comparisons between new and traditional technology under researchers'
management often give misleading results.
Once the yield difference has been estimated, it is possible to assign a value
to the increased yield and calculate the total value of increased production
resulting from adoption in the study area. The simplest approach is to
assume that widespread adoption has not affected prices, but when this is
not the case, price effects must be accounted for. If a diffusion curve has
been calculated, this can be used to estimate the stream of benefits over
time. This figure will provide some idea of the value of the product of the
research effort. It may also be important to obtain an estimate of the
increased income for farmers who have adopted the new technology. Such
an estimate will require good data on the variable costs of the technology.
Estimates of the benefits of a new technology should be balanced against
possible costs implied by changes in other parts of the farming system (for
example, if a new technology leads to a change from intercropping to
monocropping). The long-term sustainability of a new practice may also
need to be examined when considering costs and benefits.
If the research program has been planned carefully, with due attention to
developing technologies appropriate for specific groups of farmers
("recommendation domains"), then the adoption study may hold few
surprises regarding the distribution of the technology. But often
unexpected or unexplored factors influence the actual distribution of a new
technology or of its benefits. Although the "average" impact in terms of
yield or income gains may be impressive, which farmers are able to take
advantage of the change? Are many excluded from using the new
technology? Answers to these questions may be sought by ensuring that
the adoption survey covers a wide range of farmers, placing special
emphasis on the resources available to farmers. Thus the survey should
seek to assess the experiences of larger- and smaller-scale farmers, those
who have access to credit and those who do not, and so forth. These factors
are discussed in more detail in Chapter 3.
More complex questions may also be asked about the distributional
impacts of a new technology. Not only is it important to understand how a
new technology is used by different types of farmers, it is also important to
see how the benefits of the technology are distributed among various
sectors of the population. Is it farmers or consumers who gain most? In the
farming sector, how is the extra income divided among landowners,
tenants, and laborers? Do male farmers gain at the expense of female
farmers? Does the technology increase or decrease the demand for labor,
and how does that affect the incomes of the poorest sectors of the
population? An example of this kind of analysis is shown in Box 11. The
answers to most of these questions go beyond the basic adoption study
described in this manual, but such adoption studies are a necessary part of
research on the distributional impacts of technological change. Further
discussion of these issues can be found in Barker, Herdt, and Rose (1985,
Chapter 10) and Lipton with Longhurst (1989).
The Role of Adoption Studies in Assessing
the Returns to Research and Extension
One of the reasons for doing an adoption study is to provide evidence of
the returns to a research or extension effort. This analysis is done by
comparing the investment in the technology development effort to the
value of the results, measured in terms of yield or income gains. Although
this manual does not pretend to provide guidance for this complex type of
study, it should be obvious that the results of a good adoption study are an
essential element for a benefit-cost analysis of an agricultural technology
If the adoption study is to be used for this purpose, it is quite likely that
additional information, beyond the degree and distribution of adoption,
will be needed. In particular, the survey will need to collect evidence that
allows the change in technology to be attributed clearly to the research or
extension program under examination. The information required includes
evidence of the similarity between the farmers' new practice and the
recommended technology, assurance that farmer adoption took place after
the recommendations became available, and evidence that the information
utilized by farmers had its origin in the research or extension program. If
the only change being examined is the adoption of a new crop variety, then
it is usually easy to collect this information and attribute the change to a
particular research program. But obtaining similar information for crop
management practices may be considerably more difficult (Box 12).
In some cases it will be necessary to distinguish between returns to
research and returns to extension. If a study is to estimate the returns to
one or the other of these activities, particular care must be taken in
separating the two, and in examining to what extent research and
extension are substitutes or complements. Assessing the returns to
extension programs is quite challenging; a review of recent literature and
advice on the organization of such studies can be found in Birkhaeuser,
Evenson, and Feder (1991).
Once the extent and value of the technological change has been
documented and it is possible to attribute a definite proportion of this
change to the research or extension program, the benefits of this change are
compared to the costs of that program. A standard reference for benefit-
cost analysis is Gittinger (1982), and a review of methods for evaluating the
returns to agricultural research is presented by Norton and Davis (1981).
In designing an adoption study, care must be taken to define precisely
what technologies are being considered. It is likely that changes in farmers'
practices will represent a combination of farmer adaptations to new
technology, farmer innovation, as well as other changes external to the
technology generation effort.
Decisions must also be taken regarding how to measure adoption. Is the
objective to see if farmers are using a practice on any part of their farms, or
is it necessary to quantify the area planted or to measure the proportion of
yield produced with the new practice?
What time frame is of interest? Is the objective to assess adoption in the
current year, to try to establish a history of first use, to explore the
diffusion process, or to examine patterns of continued use or abandonment
If the objective is to help assess impact, the study needs to be able to
demonstrate a connection between changes in farmers' practices and the
research, extension, or community development effort. The survey should
develop information that establishes a causal link. The survey may also
need to provide evidence regarding the adoption of the technology among
different classes of farmers, in order to further analyze impact. The survey
may also provide an opportunity to develop some of the additional
information necessary for assessing impact, such as yield or income
changes, as well as evidence of other changes attendant on the new
technology, such as changes in labor patterns or land use.
Besides clearly defining and documenting the degree to which farmers
have changed their practices, an adoption study is also useful for
understanding the rationale behind these changes. The following chapter
examines some of the factors that help us to understand adoption patterns.
For each of three technological alternatives in a maize on-farm research program in Panama, a series of
acceptance criteria were defined. These criteria allow for a range of definitions of adoption (e.g., it is
possible to distinguish those farmers that have adopted the herbicide only from those that use the
herbicide at the recommended time and rate).
Chemical 1. Chemical weed control
2. Type of product
3. Application time
4. Application rate
1. If the farmer uses chemical
2. If the farmer uses
Gesaprim or Gramoxone
3. i) Gesaprim: 0-5 days
ii) Gramoxone: 0-35 days
4. i) Gesaprim 1-3 kg/ha
ii) Gramoxone 1-3 It/ha
Spacing 1. Planting arrangement 1. If planting is done in rows
2. Density 2. 45,000 60,000 plants/ha
Zero tillage 1. Tillage system 1. If the farmer does not use
2. Application of herbicides 2. If the farmer applies
herbicides prior to planting
Source: Martinez and Sain (1983).
In some cases, the adoption of a new technology leads to other changes in the farming system. If these
changes are of interest to researchers, they can be analyzed in an adoption study. The example below
shows that many farmers in Peru who adopted a new bean variety also changed some of their
management practices because of the new variety's characteristics.
Percentage of farmers who changed their crop management after adopting the bean variety
Plant seed less deeply 24 0 30
Reduce planting distance 16 0 0
Increase fungicide use 2 6 24
Increase fertilizer use 7 6 24
Change from broadcast to row planting 1 36 18
Plant beans closer to maize in intercrop 6 36 0
Change to monocrop beans 0 3 6
Source: Adapted from Table 15, Ruiz de Londorfo and Janssen (1990).
Farmers in Bumbogo, Rwanda grow beans during two seasons. Researchers were interested in assessing
the degree to which farmers had switched from bush beans to climbing beans. The data below show that
the relative importance of improved climbing beans is considerably higher in the long rainy season,
where they account for one-quarter of the bean area and half of the bean production.
Adoption of different bean types in the short and long rainy seasons, Rwanda
Local bush beans
Improved climbing beans
Local climbing beans
.28 (74) 166 (56)
.05 (13) 77 (26)
.05 (14) 52 (18)
.38 (100) 295 (100)
.07 (54) 23 (27)
.03 (25) 42 (49)
.03 (21) 21 (25)
.13 (100) 86 (100)
Source: Sperling et al. (1992).
A survey in three regions of Malawi examined the adoption of hybrid maize. Questions were asked
about varietal use on each field that the farmer managed.
It was possible to calculate several estimates of hybrid adoption. Not only does adoption vary by region,
but the different estimates provide a complex picture of hybrid adoption. Only 12% of total maize area is
in hybrids, although 27% of farmers grow some hybrid maize. But despite these relatively low figures,
35% of total maize production comes from hybrid maize.
Adoption of different maize types by farmers in three areas of Malawi
Percent farmers growing
Local maize 97 99 97 98
Hybrid maize 14 33 38 27
Other maize 8 11 14 10
Percent aggregate maize area
Local maize 91 84 74 85
Hybrid maize 6 13 22 12
Other maize 2 3 4 3
Percent aggregate maize output
Hybrid maize 18 44 47 35
Source: Smale et al. (1991).
The following graph shows the actual adoption history and fitted logistic curves
for three technologies (variety, fertilizer, and weed control) that were presented to
barley farmers in Mexico as a package but had independent adoption patterns.
Farmers tended to adopt the variety first, then improved weed control, and finally
o 0 60
60 64 68 72 76 80
Logistic curves for the adoption of three technological components in the
wet zone, Central Mexico.
Source: Byerlee and Hesse de Polanco (1986).
A survey in Peru that examined the adoption of a new bean variety asked farmers
how much of their farms they planted to the new variety in the first year that they
used it and in subsequent years. The results show a gradual increase in the use of
the variety as farmers gain confidence.
Use of bean variety Gloriabamba, by province, Peru
First season 0.11 0.28 0.13
Second season 0.24 0.40 0.40
Third season 0.43 0.57 0.50
Fourth season 0.43 1.06 0.83
Source: Ruiz de Londofio and Janssen (1990).
One valuable contribution to assessing adoption patterns is to compare the
proportion of farmers who have ever used a technology with the proportion
currently using it. If the latter is much lower than the former, there is an indication
that although farmers have tried the technology they have encountered difficulties
The following example comes from Nepal and shows that different types of
farmers have very different experiences with respect to improved maize varieties.
The majority of large commercial farmers have tried these varieties and continue
to use them. Almost half of small commercial farmers have also tried these
varieties, but only a small proportion continue to use them. Some part-time and
subsistence farmers have also tried the varieties, but almost none currently use
Current and past use of improved maize varieties by type of farm, Nepal
Source: Ashby (1982).
The farmers who adopt a technology first are not always the ones who find it most
useful. The following example from Nepal shows that although larger farmers
were the first to adopt improved (high yielding) varieties (HYVs) of rice, it is the
smaller farmers who have the highest current usage.
Adoption of improved rice varieties (HYVs) by farm type, Nepal
Large commercial 8.3 31
Small commercial 7.6 60
Part-time 5.1 72
Subsistence 5.8 66
Source: Ashby (1982).
A study of mechanical wheat harvesting in Pakistan examined the advantages of the new harvesting
method for farm owners in terms of time saved, ability to plant the following crop earlier, and yields.
Although mechanical harvesting was not a practice recommended by the research service, it was
spreading, and researchers wished to study the implications. The following figures show substantial
advantages for farmers.
Farmers' estimates of differences in supervision time, harvest duration, and yields for harvesting
by hand and with a combine, Pakistana
Supervision (mean) h/day
Harvest duration (mean) days
Time available to prepare
for next crop (mean)
Yield (grain recovery) (mean)
Source: Smale (1987).
a Holding input levels, variety, and harvested area constant.
However, an accompanying study showed the importance of the wheat harvest as a source of income for
landless laborers. The following data show how a shift to mechanical harvesting could affect the incomes
of the poor.
Percentage distribution of laborers by most important income source, rabiand kharif seasons,
All agricultural income
All non-agricultural income
Source: Smale (1987).
a Rabi is the period from wheat sowing to wheat harvesting.
b Kharif is the period after wheat harvesting through rice harvesting.
These contrasting sets of data are valuable to research leaders and policymakers as they decide on
guidelines for research and the possible promotion of mechanical harvesting.
The following data come from a study of wheat technology adoption in northern Mexico (Traxler and
Byerlee 1992). Surveys were carried out in several different years (results from 1981 and 1989 only are
shown here). There is evidence of change in several crop management practices. An analysis was also
done on changes in recommendations and evidence of causality between the recommendations and
farmers' new practices. It can be seen that although there is evidence of change in four practices, only
two of those changes can be attributed to research and extension.
Evidence of causality between changes in recommendations and farmers' practices, northern
Land Percent farmers
32 23 No Minor
5 Dec. 6 Dec.
Phosphorus Percent farmers
176 230* Yes
59 78 Yes Significant
8 33 Yes Significant
82 56* Yes Significant
Source: Traxler and Byerlee (1992).
* = difference significant at 1%.
The recommendations for nitrogen use changed only slightly between 1981 and 1989, but farmers
significantly increased the amount of nitrogen they applied, mostly in response to economic factors. The
proportion of farmers using phosphorus increased significantly as well, but there is little evidence that
they responded to a recommendation for soil testing for phosphorus. On the other hand, ridge planting
was developed by the research service and the survey showed that adopting farmers learned of the
technique through demonstrations and field days. Similarly, the survey showed that the farmers who no
longer use insecticide now take advantage of the information provided by an integrated pest
management program developed by local research and extension personnel. It is these latter changes
whose contribution to increased wheat yields needs to be examined in an investigation of returns to
research investment. The costs of the entire investment in crop management research over the period,
including research in components not adopted by farmers, must be compared to the benefits of those
elements that have actually been adopted.
The Factors That Influence Farmers' Decisions
Besides documenting the degree and scope of adoption of a technology,
many adoption studies also try to understand the patterns of adoption that
are observed. Chapter 1 emphasized that this type of analysis can be useful
for helping to refine the objectives of a technology generation program, as
well as for improving interactions between technology generation and
various aspects of national agricultural policy. Our task in analyzing
adoption patterns is to see what is acceptable and useful to farmers,
identify what is not, and suggest ways for improving the
The introduction mentioned that there is a large literature of technology
adoption studies, many of which pay little attention to the technology itself
but rather concentrate on characteristics of the farmer. In some cases it is
the attitude or personality of the farmer that is analyzed, with adopters
being considered more "progressive" or "modern." In other cases,
socioeconomic characteristics, such as wealth, landholding, or education,
are used to explain the differences between those who adopt and those
who do not.
Although such approaches may be interesting, the type of analysis
proposed in this manual requires a more careful examination of the
interaction between the characteristics of the technology and the
characteristics of the farmers and farming systems that might accommodate
the technology. This analysis of adoption patterns should be a logical
continuation of the research planning process. When technologies are being
planned and tested, priorities are set on the basis of potential benefits for
farmers (including a consideration of profitability and risks) and the ease
with which farmers may be able to adopt the technology (including
compatibility with their farming system, the possibility that they can test
the new technology themselves, and the availability of institutional
support) (Tripp and Woolley 1989).
An adoption study can be seen as another phase of the technology
generation process and an opportunity to look in more detail at how
researchers help adapt technology to farmers' needs, and how farmers
adapt their practices and conditions to take advantage of new technology.
This chapter will examine several sets of factors that influence adoption
patterns, including: the degree to which the technology is appropriate for
farmers' conditions; the compatibility of the technology with the local
farming system; how the technology is supported by markets; and how it is
presented by extension and other information systems.
The list of factors that may influence adoption is long. Researchers will
want to review this list before planning an adoption study and decide
which factors should be included in their analysis. Researchers will use
their knowledge of the types of farmers potentially affected, a comparison
of the characteristics of the new technology with farmers' traditional
practices, and an understanding of the technology diffusion process, to
decide which factors deserve attention in a particular adoption study.
Farm Resource and Farmer Characteristics
The first set of factors that we will explore are those characteristics of the
farmer and farm resources that can be used as explanatory variables in
understanding adoption patterns. These include factors such as education
or wealth that may predispose a farmer to take an interest in a new
technology, and resources such as amount of land or access to credit that
may make it easier or more profitable for a farmer to change practices.
Such farmer and farm resource characteristics occupy a major part of the
literature on adoption. Analysis of this kind of factor can be directed at
either of two audiences. It can be used for an assessment of the impact and
distributional consequences of adoption; is the new technology restricted to
certain sectors of the farming population? As well, an analysis of farm and
farmer characteristics may provide feedback to research itself for refining
the technology; is it only appropriate for farmers with certain resource or
skill levels, and what can be done to make the technology more widely
A~ Characteristics of the farmer
Much of the literature on adoption assumes that new technology is
necessarily "good" and concentrates on analyzing those characteristics of
individual farmers that make them more receptive to these innovations.
For the purposes of a technology generation program, however, it is much
better to examine the correspondence between the recommendation and
farmers' conditions, without assuming that the technology is perfectly
appropriate or that those farmers who adopt ought to be called
"progressive." Our purpose is rather to identify specific conditions that
make a technology more, or less, acceptable to farmers and that can be
addressed through research, extension, or agricultural policy to make
technology generation more efficient.
Education. Many adoption studies examine the relation between a farmer's
formal education and adoption behavior. Education may make a farmer
more receptive to advice from an extension agency or more able to deal
with technical recommendations that require a certain level of numeracy or
literacy. These skills are of course not necessarily perfectly correlated with
years of schooling, and some adoption studies go so far as to include a
small test of farmers' skills (for example, of the mathematics necessary to
calculate a dosage of herbicide). Informal education may be important as
well, and in certain cases adoption studies enquire about such things as
attendance at short courses organized by the extension service (see the
section on "Information," p. 41). It may also be interesting to ask about the
farmer's own history of innovation and of trying new ideas.
Many (but certainly not all) adoption studies show some relationship
between technology adoption and the educational level of the farmer. The
more complex the technology, the more likely it is that education will play
a role. Thus the diffusion of a new variety among farmers may not depend
at all on their education level, while the diffusion of a chemical input may
be more rapid among farmers who have at least some minimum amount of
schooling. The important point is that if such a relation is demonstrated,
we must then ask how strong that relationship is and what the practical
implications might be. If a particular technology finds its way
predominantly to farmers with a certain level of education, then several
options should be considered. One is to try to simplify the technology (or
develop alternatives) so that it is more accessible. Another option is to
concentrate extension resources on farmers with less education and to train
them in the use of the new practice. And a third option is to use this result
in making a case for more investment in extension services, training, or
rural schools to accelerate the use of agricultural technology-which is
becoming ever more complex.
Age. Another farmer characteristic that is often examined in adoption
studies is age. A farmer's age may influence adoption in one of several
ways. Older farmers may have more experience, resources, or authority
that would allow them more possibilities for trying a new technology.
Experience in a particular farming area or with a given crop may not be
strictly correlated with age, however, and it may be worth asking more
specifically about experience. On the other hand, it may be that younger
farmers are more likely to adopt a new technology, because they have had
more schooling than the older generation or perhaps have been exposed to
new ideas as migrant laborers.
In either case, it is unlikely that the demonstration of a relation between
age and adoption per se will be of immediate utility. It is more important to
see if the relationship is due to farmers' experience or education or if the
association with age is more a reflection of characteristics of the farm
household, including the distribution of authority, labor availability, or
sources of income. An example of this type of analysis is shown in Box 13
at the end of this chapter.
Gender. Women farmers are often forgotten in official agricultural
statistics. Because women play a key role in most agricultural systems, it is
important that adoption studies consider the degree to which a new
technology reaches women farmers. This requires careful planning of the
survey. In the first place, researchers should have a good idea of the crops,
cropping systems, or farm operations that are important for women
farmers and should make sure that the survey is designed to obtain the
relevant information. Perhaps more important, the survey sampling and
interviewing strategies should be planned so that women are interviewed
where they are the decision makers or have considerable knowledge about
a particular subject.
If the survey results show a significant difference in adoption rates
between men and women farmers, there are at least two hypotheses to be
explored. The first relates to the type of farming system managed by
women. It may be that the recommendations being examined are less
appropriate for the crops grown or crop management practiced by women,
and decisions would have to be taken regarding a reorientation of the
research program. The second hypothesis related to differential adoption
between men and women farmers is that women farmers are less likely to
command the resources (such as land, credit, or information) to take full
advantage of the technology. In such cases the conclusion might be to place
more emphasis on technology development that is appropriate to the
resources available to women, or to address policy changes that might
make services such as credit or extension more available to women
farmers. Box 14 provides a few examples of how gender analysis may
contribute to an adoption study.
Ethnic, religious, and community factors. In many cases a technology is
introduced to an area that includes farmers of different customs and
traditions. These differences may be most notable between communities or
between members of several groups living in the same community. It will
not be surprising to find that adoption patterns may differ among these
groups. Although these differences may be easy to demonstrate, again we
need to ask whether this information can be used to make the research or
extension program more effective. In many cases, differences among
groups arise from differences in the resources they manage (for example,
one group may have access to more or better land than another) or to
differences in farming systems or practices among the groups. In these
cases, it is a question of tailoring the recommendations for the conditions of
the different groups. Another possible explanation for such differences is
that one group may have better access to government services. If this is the
case, the conclusion may be one of reorienting government policy.
Wealth. Wealthier farmers may be the first to try a new technology,
especially if it involves purchased inputs. This may be because wealthier
farmers are more able to take risks or have better access to extension
information or to credit, or they may be able to use their own cash
resources to experiment with a new technique. Whether or not this pattern
persists, and wealthier farmers are the ones that are the major adopters and
users of a new technology, may be an important issue for an adoption
survey. The targeting and organization of the original research is also an
important factor, especially the degree to which the research was aimed at
A survey might use any of several methods for estimating a farmer's
wealth, and the method chosen will depend primarily on the hypotheses
that researchers wish to explore. Many times it is farmers with more
resources (land, labor, capital) who are able to take advantage of a new
technology. These factors are discussed in the following section. In other
cases a farmer's wealth may be a proxy for education (discussed above) or
connections with extension (discussed below in the section on
"Information," p. 40). Or in some cases farmers with a more commercial
orientation, who sell a large proportion of their harvest, are the ones who
adopt particular technologies (see "Post-Harvest Utilization and Markets,"
p. 38). Wealth per se is a difficult parameter to measure on a survey,
although it is sometimes a useful concept in explaining adoption. A method
of "wealth ranking," in which knowledgeable members of a community are
asked to divide households into groups according to locally recognized
wealth standards, is described in Grandin (1988).
Farm size. Farm size is a common variable examined in adoption studies
and is often a good proxy for wealth. It is often assumed that larger-scale
farmers will be more likely to adopt a technology, especially if the
innovation requires an extra cash investment. It may be that a certain
threshold farm size is necessary before the investment in a technology is
worthwhile. Or it may be that on larger farms different management
practices (e.g., mechanization) are used, making a recommendation more
appropriate for them. On the other hand, certain technologies are more
appropriate for the intensive management characteristic of smaller farms
(or at least of farms with a higher ratio of labor to land). Finally, farm size
may be related to access to information or credit that would facilitate the
adoption of a recommendation. The distribution of a new technology
among smaller and larger farms is valuable information for assessing
impact, but in order to refine a technology and ensure that it is appropriate
for a range of farm sizes, more specific information about the relation of
farm size to farm management is usually required (Box 15).
Labor. Technologies have different labor characteristics; some reduce the
amount of labor required for growing a crop, while others significantly
increase it. In planning an adoption study, researchers need to identify the
labor implications of each technology being examined. Are changes in labor
required, and if so, how great are they, and at what time of the crop season
do they occur? How does the technology change the division of labor
between men and women? If the new technology requires a significant
amount of extra labor, it may be necessary to develop a rough labor profile
for the farm household (see the next section, 'The Farming System"). How
many laborers are available? When are the peak labor periods during the
year? Do some household members work off of the farm and, if so, during
what periods? In addition, the survey may need to obtain information on
the availability of hired labor during the period relevant to the
recommended technology. An example of how labor availability affects
technology adoption is shown in Box 16.
Credit. Credit may be an important factor in determining adoption. If a
recommendation implies a significant cash investment for farmers, its
adoption may be facilitated by an efficient credit program. If the majority of
adopters use credit to acquire the technology, this is of course a strong
indication of credit's role in diffusing the technology. Similarly, many
farmers who do not adopt may complain of a lack of cash or credit as the
principal factor limiting their adoption. But the interpretation of this data is
aided by a knowledge of the sources of credit available to farmers. What
are the rules and regulations associated with official credit programs in the
research area? What procedures do farmers have to go through to obtain a
loan? Is the credit provided in cash or as inputs? The workings of the local
informal credit market should also be investigated. Such information
should be analyzed before the questionnaire is designed so that the survey
can distinguish accurately between farmers who may or may not have
access to credit.
Rather than facilitating access to new technology, credit programs are
sometimes responsible for obligating farmers to use a particular
technology. The credit may be offered as a package that provides a set of
inputs to farmers. Parts of the package may be "adopted" simply because of
this obligation, although farmers may feel that they are inappropriate or
unprofitable. If this is the case, the adoption study can provide valuable
data for refining and making more efficient both the recommended
package and the credit program.
Equipment and machinery. Farmers' ownership of equipment or
machinery may influence their ability to adopt technology. Farmers who
own draft animals or tractors can be more flexible in changing their tillage
34 practices than farmers who must rent or borrow equipment (Box 17). If a
recommendation involves a new type of equipment or machinery, the
degree of adoption may depend on the number of farmers who are able to
acquire the equipment and whether or not an effective rental market
develops. An adoption study that examines technology related to
equipment and machinery may need to ask how and where farmers rent
the necessary equipment.
Land tenure. An issue that is much debated in the adoption literature is the
degree to which land tenure affects a farmer's ability to adopt. This issue is
of practical interest for an adoption study if it helps us understand the
degree to which all farmers are able to take advantage of a new technology,
and whether different technologies are required for farmers without secure
access to their land.
Researchers will need to be acquainted with the specific details of rental or
sharecropping arrangements. In the case of sharecropping, how are
obligations divided between sharecropper and owner, and how is the
harvest divided? Different arrangements may make a technology more or
less attractive for the sharecropper.
Another important element of the land tenure issue is the fact that in many
cases renters or sharecroppers will be less interested in technologies that
have long-term effects, such as soil fertility maintenance or enhancement,
because they have no guaranteed access to the land. In other cases, rental or
sharecropping may go beyond a simple economic contract between owner
and tenant and involve particular obligations that restrict the tenant from
using a new technology, such as when a tenant is required to plant varieties
that provide crop residues for the owner's animals to graze after harvest.
The Farming System
One of the basic principles of on-farm research is that technologies must be
compatible with the farming system if they are to find acceptance. For this
reason, a good part of the diagnostic and planning phases of on-farm
research is devoted to examining the possible interactions between a
proposed technology and the management of the crops and animals that
constitute the farming system. Experience has shown that many times a
technology that appears as a reasonable innovation is rejected by farmers
not because of any intrinsic quality of the technology itself, but because it
conflicts with other elements of the farming system. An adoption study
should examine the degree to which that technology is consistent with the
rest of the farming system.
The principal audience for this aspect of an adoption study is researchers
themselves. The idea is to take this opportunity to monitor the 35
compatibility of the technology with the farming system and, if any
problems are detected, to use this information to refine the technology.
We will consider below some of the parameters associated with an analysis
of farming systems: the allocation of labor among various enterprises in the
system, the management of other crops planted in the same field or in
rotation, the biological conditions of the field, the soil conditions, and
climatic factors. In addition, we will examine the concept of risk. Much of
this type of analysis is useful only if it is done with reference to a specific
field, so it is important that the survey be organized to collect this
information on particular fields and for particular years.
., Labor in the farming system
A basic element of farming systems diagnosis is the development of a labor
calendar that gives an idea of how farm household labor is allocated over
the year. It is essential to know if the labor demands of a new technology
conflict with a particularly busy time of the year for farmers or rather take
advantage of a period when labor is available. It is important to remember
that this labor profile is determined not only by operations on the target
crop but also by demands from various other enterprises in the farming
system. Thus it may seem that farmers should be able to do an additional
weeding on the target crop, for instance, but actually at that time they are
busy planting another crop.
An adoption study is not the place to develop a complete labor profile. This
should have been done previously, in the earlier stages of research. But
questions can be asked about the availability of labor and/or competing
activities during those periods of peak labor demand for the technology
being promoted. Simply asking the farmers their perceptions of the labor
requirements of the new technology also can be useful (Box 18).
Other crops in the system
Intercropping and relay cropping are common practices in many farming
systems. New varieties or practices for one crop may thus have to be
compatible with the presence and management of other crops. This may
affect the choice of variety (e.g., does the new maize variety support the
intercropped climbing beans?) or management practice (is the new weed
control method compatible with the intercrop?). In some cases, what
appear to be "weeds" are actually volunteer or sown species that may be
used as food or fodder. The adoption study should pay attention to the
degree to which recommended practices are being used for the target crop
in various intercropping systems. Significant differences may signal the
necessity to target the recommendations better.
36 Crop rotations are also an important part of farming systems, and the
acceptance of the recommended technology may be affected by practices or
conditions of the preceding or following crop. A rotation pattern may affect
the timing of operations in the following crop (see Box 19). Carryover
effects from fertilizers or herbicides may also affect the management of the
following crop. Cropping history itself is important, and the practices
appropriate for newly cleared fields may be quite different from practices
suitable for fields with a longer cropping history. Again, the study of
rotation patterns is part of farming systems diagnosis, but the adoption
study should pay careful attention to the relative use of recommended
practices as part of various rotation patterns.
--- Biological circumstances
The adoption of technology may be affected by the weeds, diseases, and
insect pests prevalent in the area or in specific fields. Weed control
techniques will be more or less appropriate depending on the weed
populations or the presence of particular problem weeds in the field. New
varieties may be more or less susceptible to diseases or insect pests, and
certain management practices (such as planting time) may reflect farmers'
attempts to avoid these problems. The examination of the interaction of
these factors with technology acceptance will require that researchers are
familiar with local names for the weeds, diseases, or insects.
Land quality and soil type may be important factors influencing the
acceptance of a new technology. The selection of sites for on-farm
experiments needs to take account of variations in soil type or land type in
the research area. Researchers should have a good idea of the response and
appropriateness of the new technology under the major soil conditions of
the area. Not only may management practices differ by type of soil, but
other conditions, such as slope or moisture retention capacity, are often
important as well. To examine the influence of these factors on the
acceptance of technology in an adoption study, researchers should be
familiar with local terms for variations in soil or land type.
Climatic factors play an obvious role in the management of farming
systems. Rainfall patterns limit the crops that can be grown and regulate
planting and harvesting schedules. The possibility of drought or flooding
makes farmers wary about investing in some technologies (see next
section). Seasonal temperature changes also regulate cropping patterns, as
when frosts at the end of the growing season dictate early planting and/or
use of early maturing varieties. By the time an adoption study is carried
out, the researchers should have analyzed available secondary data on
climatic variables and have a clear idea of how they may interact with the
new technology. Climatic factors in one area may set limits on the
acceptability of a technology, and farmers may be asked their opinions and 37
experiences in this matter. Or adoption patterns may differ significantly
between two research areas, based in part on differences in climate (Box
20). Researchers should also be aware of any climatic abnormalities during
the year of the adoption survey and understand how these may affect
either the short-term or long-term use of the technology.
The failure of a new technology to be accepted is sometimes attributed to
risk aversion on the part of farmers, but this is a difficult parameter to
assess, especially in an adoption study. One important element of risk is
certainly the compatibility of the technology with climatic circumstances
described above. If secondary data show that there is a high probability of
mid-season drought at the end of June, it is not surprising that farmers
would find a maize variety that flowers in late June to be unacceptable.
Farmers would be able to articulate this reasoning in response to a question
on a survey.
There is also the perception that some farmers are more willing to take
risks than others. Besides the obvious fact that wealthier farmers will
almost certainly be more willing to invest in testing a new technology, there
is little in the literature that gives us a firm grounding for comparing risk
attitudes to adoption behavior. Certainly in many cases resource-poor
farmers have come to adopt practices, such as growing high value cash
crops, that entail considerable risk.
Another element related to risk is the variability in prices. Farmers'
experience may be that input or product prices are so variable that the
adoption of a particular technology presents unacceptable risks.
In adoption studies that seek to explore the influence of risk on the
acceptance of a technology, it is best to either use what is known about
climatic factors from secondary data or ask farmers specifically to give their
perceptions of the advantages and disadvantages, including risks, of the
Post-Harvest Utilization and Markets
The adoption of a new technology can be hindered or enhanced depending
on whether it is in accord with the system of post-harvest utilization and
marketing and the organization of input markets. A technology may lead to
significant increases in crop production, but if the inputs needed are not
available or if the extra production cannot be utilized effectively, then the
technology may be rejected. An examination of these issues may be helpful
for two potential audiences. The first is research itself; if a technology is
incompatible with post-harvest conditions, or if farmers cannot obtain the
required inputs, then the technology may have to be adjusted to these
conditions. But if the technology is underutilized because of problems with
input or product markets, then the results of an adoption study may be
used to demonstrate to policymakers the advantages of improving these
4 Post-harvest utilization and output markets
The introduction of new crop varieties to resource-poor farmers requires
an understanding of their food consumption patterns and preferences. If a
new variety is to be used for home consumption, it will have to be
acceptable for local processing and cooking practices. This issue needs to
be explored from the early stages of research, so as not to waste time on
varieties that are incompatible. Farmer participation in the research is
essential, and simple taste tests or panels may be devised to sort out which
varietal characteristics are crucial for achieving farmer acceptance.
One difficulty, however, is that farmers' tastes are subject to change as well,
and a variety that is not ideal for certain purposes may have agronomic or
other characteristics that offset the disadvantages. Thus at times a period of
testing is necessary, when farmers get better acquainted with the variety
and see to what degree it can be used. An adoption survey offers a chance
to monitor farmers' experience with a variety that has been released and to
provide feedback to plant breeders regarding acceptable and unacceptable
characteristics (Box 21).
If farmers market a considerable proportion of their harvest, then the
acceptability of the new variety on the market needs to be investigated as
well. The characteristics required for market acceptance may be
considerably different from those that determine acceptability in the
household. Researchers will want to talk to both farmers and merchants to
identify the principal characteristics that regulate the market price of a crop
variety. The survey can get more information about how and where
farmers sell particular varieties, and can explore differences in price among
A more general look at crop marketing may also be called for. Not only do
markets affect the acceptability of a new crop variety, they may also
influence farmers' interest in any technology that promises higher yields. If
markets are inefficient, there may be little incentive to invest in improved
technology. In addition, characteristics such as seasonal variation in market
prices may affect the acceptability of technologies that change the timing of
harvest (e.g., a technique that allows earlier planting).
Storage practices may also deserve attention in a varietal adoption survey.
Indeed, new storage techniques may be one of the subjects of such a
survey. In addition, crop storage practices may affect the acceptability of
technologies that increase the harvest or change its timing. New varieties
may have quite different storage characteristics, and this may be a limiting
factor for acceptance. An adoption survey can explore the degree to which
such storage problems are best addressed through plant breeding or
through improved storage techniques.
Crop production may not only be important for human consumption, but
may also be important for animal production as well. The place of animals
in the farming system needs to be explored in relation to new technologies.
New varieties may have less acceptable forage qualities, for instance. New
crop management techniques related to plant populations, weeding, or
harvest may increase grain yield but affect the production of byproducts
destined for animals. The sale of crop byproducts may be an important
source of income. The use of damaged or spoiled grain or tubers for on-
farm animal feed may diminish farmers' interest in certain crop protection
S l Input markets
If a recommendation involves the use of purchased inputs, the adoption
study can check to see if farmers are able to acquire them. Preliminary
investigation should develop information on points of sale, available
supplies, and form of sale, and the survey can then compare this to farmers'
experience. Do farmers know where the inputs are sold? Are there
difficulties in getting to the location? Are the inputs always available when
they are needed (Box 22)? What price does the farmer have to pay? Is input
The use of new crop varieties and seeds requires particular attention to the
source of the input. Seed is frequently saved from year to year, exchanged
among farmers, or purchased at a local market, besides being acquired
from official sources. Not only is it important to ask the source of seed, but
also the number of years since the farmer first obtained seed from that
This type of information can be valuable for developing communication
with policymakers and demonstrating ways in which relatively minor
changes in input policy may significantly affect the use of new technology
For farmers to adopt a technology they must first know about it. The
information may come from several sources. It is important to explore the
degree to which farmers have received the necessary information. This will
help in analyzing the degree to which low adoption may not be a function
of the technology itself, but rather of the information that is available. This
analysis is useful for improving extension policies and programs.
In order to explore these issues on a survey, researchers need to be familiar
with existing extension programs and other media for providing
information to farmers. They need to know whether field days,
demonstrations, farm visits, or other methods are used by the extension
service, and which terms farmers use to refer to them. Radio, television,
newspapers, or magazines may be important sources of information in
One simple way of addressing this question is to estimate the degree of
contact that each farmer has had with extension activities and to compare
this with adoption behavior, in order to see if the extension contact has had
an influence. Another strategy is to ask farmers where they learned about a
technology (Box 23), although farmers' memories often are not perfect on
It also must be borne in mind that much information comes from other
farmers. This makes studying the effect of extension difficult, because the
information may well have originated with an extension agent but passed
to the respondent through another farmer. This kind of problem led
Birkhauser, Evenson, and Feder (1991) to suggest that a study designed to
measure the actual impact or effectiveness of an extension program is best
organized by comparing communities that have had access to an extension
program with those that have had no contact, rather than trying to make
comparisons between individual farmers in one community.
Farmers can be asked about specific production problems to see if they are
familiar with recommendations (Box 24). Finally, it is worth remembering
that knowledge and use are two different things. An adoption study can
examine the distribution of farmers who don't know about a technology,
those who know about it and don't (or no longer) use it, and those who
know about it and use it. This contrast will provide a clear indication of the
degree to which knowledge is a factor in explaining adoption patterns.
The following is a checklist of factors that help explain adoption patterns.
No adoption study will include all of these factors, and it is important to
decide which ones to focus on. This decision will depend to a large extent
on the purpose of the study. Because an adoption study can have several
potential audiences, it is important that the audience be clearly defined 41
before the survey is designed.
A checklist of factors that are important for understanding adoption
Education Measure years of schooling, Types of training available
participation in other to farmers. '
Age Age of decision maker. -'
(Effect may be related to
Gender (Effect may be related to Sampling strategy to
type of farming system or include women farmers.
access to resources such What crops, systems,
as credit, extension, etc.) operations do women have
Ethnic (Effect may be related to What are the different groups
group, type of farming system or in the research area? How
etc. access to resources.) do they refer to themselves?
Wealth Can be estimated by land
holding, income, or other
factors. (Effect may be
related to specific
Farm size (Effect may be related Local measures of area.
to farming system or
Labor Amount of labor available Labor requirements of
force in household; hiring the technology.
Credit Participation in credit What credit programs exist
programs; source of loans for farmers? What are
for farming. their requirements? Who
Equipment Ownership of equipment; Equipment and machinery I
and experience with rental. requirements of the
machinery technology. Rental markets
in the area.
Land Decision makers' access What are local customs and
tenure to specific field, vocabulary for renting,
Labor in Amount of labor available Timing and amount of labor
farming in household; hiring required by new technology.
system practices; other labor Cropping calendar.
demands in system.
Other Intercropping and relay Local practices for
crops in cropping in specific intercropping, relay
system field. Rotation history cropping, rotation.
of specific field.
Biological Characteristics of specific Local vocabulary for weeds, 4
factors field, by season. pests, diseases. Suscep-
tibility of new technology.
Soils, Soil type, slope, other Local vocabulary for land 4
land type characteristics of types. Experimental
specific field, evidence of different
responses of technology.
Climate Farmers' opinions about Secondary data on climate. 4
timing of technology,
Risk Farmers' opinions of risk Secondary data on climate. 4
of technology. Other
indicators of farmers'
attitude to risk.
Home Opinions on acceptability Local food preparation, 4
consump- of new variety. Changes types, and methods.
tion in food preparation.
Output Farm crop sales; amounts, How and where are crops 4 4 4
marketing timing. marketed in area? How do
prices vary by season?
Storage How do farmers store the Local storage techniques. 4
crop? Opinions on
characteristics of new
technology related to type
or timing of storage.
Fodder Timing of fodder Importance of animals and 4
needs requirements. Types of fodder in farming system.
Input Where farmers obtain Local sources of inputs; 4 4
markets specific inputs. Knowledge availability. Local
of input type, source. vocabulary for inputs.
Inform- Contacts with extension Types of extension
ation activities. Knowledge activity in the research area.
of technology. Source of
knowledge. Time of first
use of technology.
Once the purpose and audience for an adoption study have been defined, the definition and
measurement criteria for changes in farmers' practices have been identified, and hypotheses have been
listed regarding reasons for differences in adoption patterns, the adoption study can be designed. The
next chapter discusses issues related to choosing a sample, designing questions, and implementing the
Although the age of the household head or decision-maker may influence adoption patterns, other
household characteristics may play an even more important role. Anthropologists use the term
"household development cycle" to describe the way households evolve over time, often growing larger
and more complex and then declining. This is not merely a function of household size, but also reflects
changes in access to resources (for example, a son may gradually acquire rights to his father's land),
changing sources of income (younger households may earn more money from off-farm labor), and
various kinship obligations.
The table below compares the adoption of hybrid maize in Swaziland for three types of household. It
was found that hybrid adopters were no more likely to sell maize than non-adopters, and the question
was then asked why some farmers would grow hybrids for subsistence and others would not. The
argument made here is that younger households (type 1) have higher consumer/worker ratios and
require more efficient technology for producing maize. More mature households (type 2) have lower
consumer/worker ratios but have significant cash earnings to support the improved technology.
Households that are in decline (type 3) have the lowest consumer/worker ratios and thus the least need
for more productive technology. In addition, their wage earners earn less than those from other
household types because they tend to be older and less educated. It can be seen that this classification by
household type is more successful at explaining adoption than a simple analysis by age of household
hybrid maize 57 64 37
Average age of
household head 40.3 56.7 59.7
ratio 2.2 1.9 1.7
wage earners 1.0 2.2 1.4
Source: Calculated from Tables 7.17, 7.19, 8.12 in Low (1986).
The adoption of technology by women farmers may be quite different from that by men. The planning of
an adoption survey may need to include a careful examination of how responsibilities for different
agricultural activities are divided between men and women. The following data from Nepal show that
women participate in most agricultural decisions, and in cases related to choice of variety they are the
Farm management decisions in Nepal (%)
What crop to plant 18.0 30.2 51.8 100.0
What seed to use 20.7 60.4 18.9 100.0
Amount and kind of fertilizer 32.5 39.7 27.8 100.0
Source: Acharya and Bennett (1982).
One of the reasons that women farmers may adopt technology at a lower rate than male farmers is that
often women are not as well served by extension services as men are. The following data from Malawi
show the strong tendency for extension to have less contact with women farmers.
Type of extension contact for male household heads (MHH), wives, and female household heads
(FHH), Malawi (percentages of total in category)
Personal visit 41 28 23 28 12 4
Group meeting 66 44 49 43 12 8
Demonstration 13 6 6 5 1 0
Field visit 13 9 6 15 5 2
Source: Spring (1985).
Women farmers may also have different farming systems. In a project in The Gambia, the provision of
new pump-irrigated rice technology benefited men more than women farmers.
Women's participation in different rice technologies in The Gambia
Fields under women's control (%) 10.0 77.0 91.0
Yields per hectare 5.9 2.5 1.3
Input cost (US$/ha) 294.0 154.0 20.0
Labor input by women (in %
of unpaid family labor) 29.0 60.0 77.0
Source: von Braun and Webb (1989).
A study in Eastern Province, Zambia, showed that adoption of new technology was related to farm size,
but the relationship depended on the technology. The adoption of hybrid maize, a cash crop, is very
dependent on farm size. Ox cultivation varies as well with farm size, partly because farmers who have
more land are more likely to own oxen. Fertilizer use, on the other hand, is only weakly related to farm
size, and fertilizer rates do not vary across different farm size classes.
Size of farms and adoption of technology in the Plateau Zone of Eastern Province, Zambia
Households (%) 23.6 31.7 15.4 18.7 10.6
Average size (ha) 0.6 1.5 2.4 3.8 8.0
Farmers using hybrids (%) 6.9 30.8 42.1 52.2 96.1
Maize area under hybrids (%) 3.1 15.1 17.6 24.2 49.2
Cultivation with oxen
Farmers using oxen (%) 25.9 56.4 65.8 87.0 96.2
Farmers owning oxen (%) 17.2 17.9 36.8 58.7 76.9
Farmers using fertilizer
(percent) 51.7 67.9 65.8 67.3 100.0
Crop area fertilized
(percent) 45.5 49.8 52.3 43.5 72.3
Average rate of
application (kg of plant
nutrients per fertilized
hectare) 102.5 92.2 94.6 104.5 93.9
Source: Jha et al. (1991).
A study of the adoption of dry-seeded rice (DSR) in the Philippines examined a number of factors. One
of the most important was labor availability in the household, because the dry-seeding technology
demands more labor for weeding. Farmers who adopt dry-seeding must either use more household
labor or hire labor.
A logit regression showed that labor availability, knowledge, and farm area had significant associations
with the adoption of dry-seeding. The first table shows the output of the logit regression. (For more
information on logit regressions, see pp. 76-81.)
Logit regression analysis for adoption of DSR in Iloilo, Philippines, 1982
bo Constant -1.0154 1.1258 -0.902 n.s. a
X1 Farmer's age -0.0143 0.0160 -0.895 n.s.
X2 Education -0.0623 0.0539 -1.156 n.s.
X3 Labor index 0.3132 0.1545 2.027 5%
X4 Draft index -0.2643 0.2331 -1.134 n.s.
X5 Extension visits 0.00373 0.00829 0.451 n.s.
X6 Knowledge 0.4922 0.1109 4.437 0.1%
X7 Credit -0.4662 0.4190 -1.113 n.s.
Xg Tenancy 0.000130 0.00417 0.031 n.s.
Xg Rainfed lowland area 0.0000558 0.0000227 2.455 5%
Log likelihood ratio statistic [-2(log Lo log Lmax)] = 44.6 (9 d.f.)
Cases predicted correctly = 70.5
Sample size = 173
Source: Denning (1991).
a n.s. = not significant.
The second table looks more specifically at the independent effects of the three factors on the adoption of
the DSR technology. It uses the results of the logit regression to calculate the probability of adoption for
different types of farmers. With respect to labor availability, it shows that the higher the labor index, the
more likely farmers are to adopt DSR, independent of the effects of knowledge or farm area.
Predicted probabilities of DSR adoption for an average farmer, for different sizes of rainfed
lowland area being farmed and different levels of labor availability and farmer knowledge, Iloilo,
Mean knowledge level 0.45 0.53 0.69 0.59 0.66 0.79 0.72 0.78 0.87
High knowledge level 0.59 0.66 0.79 0.71 0.77 0.86 0.81 0.86 0.92
Source: Denning (1991).
The availability of machinery or equipment may affect farmers' ability to follow recommendations. The
following data from Kenya show that machinery owners (who tend to be larger farmers) are able to
prepare their land for wheat closer to the recommended time than those who must rent machinery.
Relationship between wheat planting date and machinery ownership, Kenya
Small (n = 24)
Medium (n = 18)
Large (n = 17)
Source: Hassan, Mwangi, and Karanja (1992).
a Numbers in brackets refer to the week (e.g., first, second, fourth) of the month.
Farmers must often make trade-offs in the management of their resources, and labor often presents a
particular challenge. Farmers in western Ethiopia were encouraged to switch from broadcasting maize to
row planting. About half of the farmers changed to row planting, but a survey that included questions on
farmers' opinions of the recommendations showed that the greatest limitation to further adoption was
the extra labor required by row planting.
Farmers' opinions on row planting, western Ethiopia (n = 64)
Labor saving 7.8 92.2 0.0
Seed saving 98.4 1.6 0.0
Convenience for oxen-cultivation 90.6 4.7 3.1
Reduce lodging 96.9 0.0 3.1
Yield 93.8 3.1 1.6
Source: Beyene Seboka et al. (1991).
Adoption of technology for one crop may be affected by interactions with another crop in the rotation. In
the example below, wheat that is planted after fallow in the Punjab of Pakistan is much more likely to be
planted at the recommended (early) date than wheat that follows cotton, since the cotton harvest can
extend into the time for preparing land for wheat. Other practices, such as number of plowings and
fertilizer use, tend to differ somewhat between the two systems as well.
Major differences in production practices in two cropping patterns in cotton-wheat areas,
Percent planted before 30 Nov. 5 51
Percent planted during 1-15 Dec. 27 31 .00
Percent planted after 15 Dec. 68 18
Number of plowings and plankings 6.9 8.0 .09
Average fertilizer use (kg/ha)
Nitrogen 98 86 .09
Phosphorus 50 40 .08
Number of irrigations 6.0 5.9 .69
Average wheat yield (kg/ha) 2,178 2,401 .04
Source: Akhtar et al. (1986).
The adoption of technology is influenced by climatic circumstances. In an adoption study in Central
Mexico, it was found that barley farmers in a higher rainfall area were more likely to adopt
recommendations than those in a lower rainfall area. In addition, the technological components were
adopted differently in each area. The majority of farmers in both areas adopted improved barley
varieties, but the use of herbicide and fertilizer was much higher in the higher rainfall area, where
response to these inputs was greater. In addition, farmers in the wet zone tended to adopt herbicide
before fertilizer, while those in the dry zone, where weeds were less of a problem, tended to adopt
Adoption of three technological components in two zones of Central Mexico, 1975 and 1980
Improved varieties 76 91 96
Herbicide 77 74 82
Fertilizer 46 62 65
Improved varieties 29 61 61
Herbicide 11 13 17
Fertilizer 14 26 30
Source: Byerlee and Hesse de Polanco (1986).
It is often useful to ask farmers their opinions on various characteristics of a new variety. Rarely will a
new variety be judged superior to a local one on all counts, but farmers are able to balance the
advantages and disadvantages of a new variety in making adoption decisions.
The data below show that a new maize variety in Ghana is judged to have a number of superior
agronomic traits, but is found to be deficient in both storage and cooking quality. These deficiencies did
not keep the variety from being widely adopted, but more attention to storing and cooking qualities in
breeding future maize varieties would certainly lead to even higher acceptance of new varieties.
Farmers' opinions on local and new maize varieties
Yield without fertilizer 66
Yield with fertilizer
Source: Tripp et al. (1987).
In order to understand the input distribution system it is useful to ask farmers where they acquired their
inputs. In a study on wheat varietal adoption in Pakistan, researchers found that very few farmers were
using government seed depots or merchants as sources of seed.
Sources of wheat seed of new varieties and other popular varieties, Mardan, Pakistan
Own 35 53 56 81
Otherfarmers 35 35 37 19
Seed depot 15 12
Shopkeeper/grain merchant .... 7
Total 100 100 100 100
Source: Heisey (1990).
When asked about the location of the seed depots, only about half of the farmers could say where these
were located. These data provided evidence to show that seed was not being distributed from seed
depots in the way that policymakers believed.
Percentage of farmers who knew seed depot location and had visited a depot, Pakistan
Knew correct location 51 46 52
Stated other location 14 11 38
Did not know location 35 44 10
Visited depot 38 36 21
Source: Heisey (1990).
a Includes only farmers with correct knowledge of depot location.
Knowing where farmers get information about new technologies may be important. If one of the
purposes of a particular adoption study is to evaluate information transfer, this sort of data is essential.
The table below shows farmers' answers to a question on how they learned about a new maize variety,
row planting, fertilizer, and fertilizer application methods. Although extension is the most important
source of information, the degree to which farmers also obtain information from other farmers,
especially in the case of new varieties, is noteworthy.
How farmers first learned of technology (Ghana, 1990)
Visit by extension (%)
Other extension method (%)
(Total extension) (%)
From another farmer (%)
Other or don't know (%)
Source: Ghana Grains Development Project (1991).
It may be useful to ask about farmers' knowledge of specific recommendations. The data below are
from a study that examined the effectiveness of a training and visit extension program in India. A
number of questions were asked regarding technologies presented by the extension program, and the
answers of contact farmers were compared to those of non-contact farmers. In some cases contact
farmers were more aware of the recommendations than other farmers, while in other cases there was no
Farmers' knowledge and adoption of selected extension recommendations on groundnut in
Dhenkanal District, Orissa, India (recommended practices indicated by an asterisk)
1. What can be done
to control the attack
of termites in
2. What can be done
to control the Tikka
3. What kind of seed
treatment should be
given to groundnut?
4. Do you practice the
seed treatment of
5. What is the best
in groundnut (erect,
(+ 20 cm)
Seed and soil treatment*
Soil treatment and fertilizer
Do not know
Seed treatment and spray*
Do not know
Do not know
Correct (17 < X < 25cm)
Do not know
Source: Hoeper (1988).
Organi- Chapter 1 emphasized that the study of adoption is an integral part of the
nation technology generation process. Most of the factors that affect the potential
acceptability of a new technology must be identified and monitored while
agricultural research is being planned and carried out. The study of
adoption is not something that can be left until the last minute.
Although a number of methods can be used to assess adoption once a
technology has been released (see Box 2, p. 9), this manual focuses on the
management of formal farmer surveys. There is an ample literature on
survey techniques and design, and this chapter will emphasize only those
factors that are of particular relevance to adoption surveys.
Before approaching the adoption survey design, researchers need to
address several issues. First, researchers must carefully define the nature of
the technology change they hope to analyze. What types of changes in
farmers' practices do they propose to study? Second, they need to identify
the audience for their study. Is the purpose to provide feedback from
farmers to the research or extension program to make it more effective? Is
the audience another institution that helps determine the policy
environment in which the technology generation process operates? Is the
purpose to document the impact of a research or extension effort? And
third, a decision on the purpose of the study will determine whether the
study will emphasize documenting the degree of adoption (Chapter 2) or
whether it will also explore the reasons for the observed pattern of
adoption (Chapter 3).
The adoption studies described in this manual are aimed at providing
information to improve the efficiency of research and extension activities.
This information can also be used to assess the effectiveness of particular
investments in research or extension. Although donor development
projects or other special investments are often subject to outside evaluation,
in the majority of cases studies of agricultural technology adoption will be
carried out by the institutions that developed the technology. This places a
special responsibility for objectivity on those doing the study. It must be
remembered that the study is done to assess and to understand, rather than
to "prove" success. The sampling for the study must be undertaken with
special care, avoiding biases that may show the institution in an
unreasonably favorable light and making sure not to ignore issues that
56 should be addressed. For instance, if the adoption of a new technology may
Sbe more rapid with farmers who have better access to roads, extension
services, or markets, it is important that the sampling for a study be carried
out so as to avoid favoring such farmers and giving the impression that
adoption is more widespread than it really is.
Techniques for drawing samples for surveys are described elsewhere and
will not be discussed here in detail (see Casley and Kumar 1988, Scott
1985). It is assumed that the adoption study will look at a relatively small
number of technological changes in a well-defined area of a country.
It is important to define the population to be sampled. Is it, for instance, all
the farmers in a specified region? All the rice farmers? All farmers who
grow at least a half hectare of rice? All rice farmers who have participated
in a special extension program? The answers to these questions will depend
on the purpose of the study, but a clear definition of the population is
essential. Once this decision has been taken, the next step is to decide what
units will be sampled. This may prove more difficult than it sounds. If
households are the units, how do we define a household? In some cases it
may consist of members of an extended family who share certain
agricultural resources and responsibilities but who also manage other
agricultural enterprises independently. Are the interviews to be conducted
with the head of the household? In many cases the household head may be
a male, but females may take key decisions or have significant
responsibilities for crop management. In other cases, certain fields or crops
may be the responsibility of particular household members. Preliminary
investigation vill be necessary to decide how respondents are to be chosen
and who will be the subject of the interviews.
Once the study population has been defined, it is necessary to identify a
sampling frame. The sampling frame is constructed so that random
sampling can be carried out (in other words, so that theoretically -
every unit in the population has a known chance of being chosen for the
survey). In many of the procedures described here, the probabilities of
selection for each farmer will be equal. The sampling frame is, ideally, a list
of all the units that could be sampled. In many cases, however, such a
complete list will not exist and may be impossible to assemble. In these
cases, alternative procedures can be defined that still allow for random
selection. One alternative is to draw a map of the households in the study
area. Another possibility is to devise rules for sampling; for instance, a
number of randomly selected points on a map may be chosen and
enumerators are instructed to take the first (or nt) unit in a randomly
A common strategy when complete sampling frames are not available is to
do two-stage sampling. If the study area consists of a large number of
villages, for instance, the first stage of sampling would be a random
selection of a certain proportion of those villages. The second stage would
be to assemble complete lists of farmers or households in the selected
villages and then draw samples from those lists. Special care must be taken
in assembling the lists to ensure that all farmers (e.g., including all women
farmers) are represented. Two-stage sampling may be the only feasible
alternative in areas where complete population or residence lists are
unavailable. It requires careful planning, especially if the first-stage units
(e.g., villages) are very different sizes. Scott (1985) discusses two
approaches to this problem. One is to recombine the first-stage units so
they are of roughly equivalent size. The second is to select the units with a
probability approximately equivalent to their size, and then to choose equal
numbers of farmers from each sampled unit.
The size of the sample will depend to some extent on the nature of the
study and the resources available. The appropriate sample size is also
determined by the amount of variability in the population. If the study area
farmers are very heterogeneous in their practices or adoption behavior, a
larger sample will be required. One way of reducing some of the sampling
error related to this variability is to stratify the sample. If part of the
variability in the sample results from differences between, let's say, farmers
who have access to irrigation and those who do not, it will be more efficient
to stratify the sample by access to irrigation, rather than to rely on purely
random sampling to provide sufficient numbers of each class of farmer for
the analysis. When a stratified sample is used, however, researchers need
to be able to estimate the proportion of each category in the general
population so that they can assign weights to each subsample when
estimating important parameters, such as adoption rates, for the entire
population. The adoption studies described in this manual may be carried
out with as few as 50-60 respondents, but the complexity of the adoption
process is such that 80-120 respondents will be a more usual sample size.
At times the existence of previous studies may help determine the nature of
the sampling frame. If a formal survey was conducted in the area several
years earlier as part of the diagnosis for the research program or to provide
baseline data, an adoption study carried out using similar sampling
procedures will provide valuable comparative data. In some cases, an
adoption study with the same sample used earlier may be useful. In some
cases a "panel" of farmers is visited every few years to follow technological
change. There are both advantages and disadvantages to repeatedly
surveying the same farmers (see Scott, 1985). Although this is at times an
attractive option, in most cases the sample for an adoption study will be
drawn without benefit of previous surveys.
Sampling for comparisons
In many instances an adoption study attempts to answer questions about
differences in the rate of adoption between groups of farmers. Have larger-
or smaller-scale farmers tended to adopt the technology? Are participants
in extension programs more likely to adopt? Does distance from markets
make a difference in adoption behavior? These and similar questions may
be addressed in two possible ways. In the majority of cases it is likely that
these characteristics will appear with sufficient frequency in a randomly
drawn sample that the appropriate comparisons can be made. But if a
particular comparison is important enough for the purposes of the study,
and if it is not certain whether a randomly drawn sample will provide a
sufficient number of each category, then the sampling may be done in a
purposive manner. For instance, if it is important to judge whether the
participants in an extension program have reacted differently to a new
technology than non-participants, then part of the sample may be drawn
from a list of program participants and part may be drawn from the
general population. In this example the sampling is fairly straightforward
because a list of program participants exists. But in other cases (such as
large-scale versus small-scale farmers) it may take considerably more effort
to develop appropriate samples, and researchers will have to decide if this
is worthwhile. As with stratified sampling, for each purposively chosen
group, researchers need to know the proportion of farmers in order to
know what weight to assign each subsample to assess overall adoption
Besides deciding on an appropriate sampling frame for farmers, it may also
be necessary to define a sampling procedure for fields. If farmers in the
study area have several fields, researchers will have to decide whether they
are going to assess adoption of the technology on all of the farmer's fields,
on fields with certain characteristics, or only on the largest field. This
decision will depend partly on the definition of adoption selected for the
study and the purpose of the study. If the purpose is to inventory the use of
a particular technology, then all fields may have to be sampled. If the
purpose is more to understand the context in which a technology is used,
then a subsample of fields may be sufficient, or the farmers themselves may
be the only sampling unit and be classified as adopters or non-adopters.
Timing of the Survey
In most cases the best time to do an adoption study will be shortly after the
harvest, when farmers may have more time to answer questions and when
they will be able to report on their experience during the season. If more
than one season per year is being studied, the timing question is more
complex. The timing may also depend on the specific purpose of the
survey. For certain surveys it may be useful to include field observations in
the study (to observe the effectiveness of a weed control technology, for
instance), in which case the survey will be done during the crop season. Or
it may be that storage or marketing issues are important, and researchers
may wait until some time after harvest and then survey farmers about their
marketing experience. The timing of the survey may be affected by natural
circumstances. Although it is often impossible to carry out a survey in the
proverbial "normal" year, unusual conditions such as a severe drought may
mean that conditions are not appropriate for assessing farmers' experience
with a particular technology, and the survey might have to be postponed
or at least redesigned. In certain areas where the performance of a
technology is chronically variable (because of uncertain rainfall, for
example), it may be necessary to assess adoption and technology
performance over several years.
This manual discusses the development and analysis of short, well-focused
questionnaires for studies of adoption. Such questionnaires can be used to
document and quantify the degree of adoption of particular technologies
and to collect information that helps explain the patterns and extent of
adoption. It is very important that the purpose of the survey be identified
clearly so that the questionnaire is an efficient instrument for collecting
priority information, not a device for exploring a wide range of issues.
The section on sampling discussed the importance of an objective approach
to assessing adoption. The same advice is relevant when designing the
questionnaire. The questions should be presented to the farmer in as
neutral a manner as possible. The best way to do this is simply to
document the relevant practices that the farmer is using, much as would be
done in a diagnostic survey of farming practices. The interviewers should
explain to the farmers that they are interested in farming practices and
problems, not that they are trying to see if farmers are following
recommendations. Thus questions on varietal use should not begin with,
"Do you use new variety X?", but rather with more general questions that
ask farmers to list the varieties that they grow. It is very important that no
leading questions are asked and that the farmers do notfeel they are being tested
with regard to their knowledge or use of recommendations.
The survey questions will be determined by the objectives of the study. In
documenting adoption, for instance, the goal may simply be to estimate the
proportion of farmers using a practice, or it may be to estimate the
60 proportion of area or production associated with the practice. The goal
chosen for the survey will determine the type of question to be asked.
Similarly, if one goal is to understand why some farmers have adopted a
practice and others have not, researchers will need to form specific
hypotheses to guide the types of questions that are asked. These
hypotheses may be pursued directly, by asking farmers' opinions, or
indirectly, by statistical comparisons between adoption behavior and
specific characteristics of the farmers' circumstances (Chapter 5).
There are a number of good sources available on survey design, and only
general guidelines will be offered here. The reader is referred to Byerlee,
Collinson, et al. (1980), Casley and Kumar (1988), Bernsten (1980), Alreck
and Settle (1985), and Murphy and Sprey (1982) for more detail.
1. Open versus closed questions
Questions may either be preceded so that the farmer's response should
correspond to one of a limited number of choices, or the space for the
response on the questionnaire may be open and the interviewer records
whatever the farmer says. There are advantages and disadvantages for
each of these types of questions, but for an adoption study it is to be
expected that the vast majority of questions will be closed. This is not
only more efficient for analysis, but also reflects the fact that the
adoption study will be quite well focused; will be carried out in an area
where researchers should have a good idea of local practices,
vocabulary, and so forth; and will be designed after a good informal
2. Leading questions
No survey should contain leading questions, but this is particularly
important for an adoption study. There should be no "hints" or
encouragement that might lead a farmer to respond positively about
particular practices. For instance, farmers should not be told what is
"correct" and then be asked if they follow that practice. Similarly,
farmers should be allowed to give the names of varieties or purchased
inputs that they use and should not be provided these names
beforehand. If farmers cannot remember the name of the input, this fact
should be recorded on the questionnaire.
Questions about adoption should be as specific as possible. "Do you use
fertilizer?" is rarely a sufficiently precise question, for instance. The
fertilizer use should be specified by season and field. If input rates are
required for the study, questions should be asked about specific fields
and the interviewers will have to be familiar with local units of measure
for both inputs and field size.
4. Order of questions
The questionnaire should follow a logical order. The introductory
questions should be of a general nature. These would be followed by
questions about specific aspects of crop management. Sensitive
questions, such as those about use of credit, are best left until near the
end. As well, if the study requires information on farmers' knowledge of
recommended practices, this type of question should be left until after
the farmer's actual practices have been recorded.
,4ii, a Examples of questions for an adoption study
Appropriate questions for a questionnaire should be based on under-
standing developed during an informal survey. No manual can provide
"standard" questions to be used on a survey. The following examples are
presented only to show some appropriate formats for questions.
A. Many questions regarding the adoption of technology can be asked in
the context of an inventory of practices by field. It is almost always
preferable to ask about practices in specific fields, rather than in general.
If complete information on the area of adoption is required, the question
can be repeated for all fields. In other cases, questions about the major
field may suffice.
The following table illustrates how questions about weeding practices in
cassava might be organized:
Size of field (no. of acres)
Date of planting cassavaa
Name of cassava variety(ies)
Date of planting intercropa
Date of 1 st weeding
Method of 1st weeding
Date of 2nd weeding
Method of 2nd weeding
a Week and month, e.g., 2/4 = 2nd week of April.
b 0 = monocrop, 1 = maize, 2 = sorghum, 3 = other.
c 1 = machete, 2 = hoe, 3 = plow.
B. Measurements for input use may be necessary if it is important to
estimate exact dosages. This may require using local measures and
In an area where farmers measured herbicides with small cans, the
following questions were asked:
If you used herbicide to control weeds, can you tell us about your
Size of field: "tareas" (local measure)
Days before (-) or
after (+) planting
Name of herbicide
Cans of commercial
product per sprayer tank
Number of tanks per
C. Farmers' opinions on a new technology are often best assessed by
asking about specific characteristics of the technology. The farmer may
be asked to compare two or more technologies (e.g., varieties) and be
asked which is best or worst.
The following is an example of questions to elicit farmers' opinions on
1. We have heard that the cover crop that farmers call "fertilizer bean"
has some advantages. Which of the following are important for you?
Improves soil fertility
Makes land preparation easier
Helps control weeds
Helps conserve moisture
1 = Very important advantage 3 = Not important
2 = Some advantages 4 = Don't know
2. We have also heard that the "fertilizer bean" has some
disadvantages. Which of the following are disadvantages?
Planting the cover crop means losing
one cycle of maize
Brings more insects
Brings more rats
Causes landslides on steep slopes
1 = Very important advantage 3 = Not important
2 = Some advantages 4 = Don't know
D. Studying varietal use may require careful questioning regarding the
source of the seed.
1. Name of principal variety planted in this field:
2. Where did you obtain the seed that you planted this year?
1. From last year's harvest
2. Purchased from the seed depot
3. Purchased from the market
4. Purchased from another farmer
5. Gift from another farmer
3. If the seed is from last year's harvest, in what year did you first
acquire this variety?
4. How did you first acquire the seed?
1. Purchased from the seed depot
2. Purchased from the market
3. Purchased from another farmer
4. Gift from another farmer
E. Extension contact and source of knowledge.
To be able to attribute a change in farmers' practice to an extension
program, questions need to elicit when the farmer first began using the
practice and the source of the materials and knowledge required to
carry out the practice.
1. If you are using leucaena for alley farming on part of your farm, in
which year did you begin this practice?
2. Where did you acquire the seed?
1. Another farmer
2. The farmer's club
3. An extension agent
3. How did you learn how to plant the leucaena?
1. From another farmer
2. Attending a field day organized by farmer's club
3. Attending a field day of extension service
4. Visit from extension agent
5. Reading extension bulletin
6. Own experiments
4. How did you learn how to prune the leucaena?
1. From another farmer
2. Attending a field day organized by farmer's club
3. Attending a field day of extension service
4. Visit from extension agent
5. Reading extension bulletin
6. Own experiments
Guidance on survey implementation can be found in a number of sources
(Byerlee, Collinson, et al. 1980; Casley and Lury 1981). Any questionnaire
should go through several drafts as it is reviewed by colleagues and pre-
tested with farmers. It is difficult to overemphasize the importance of
careful questionnaire testing. The field testing of the questionnaire should
be done by the people who will work as enumerators, so that they can
participate in developing the final version and have a thorough
understanding of the questions. Besides participating in the field testing of
the questionnaire, the enumerators should undergo additional training to
ensure that they are able to present and interpret the questions correctly.
The logistics of carrying out the survey need to be carefully thought out.
Enumerators are assigned specific portions of the sample, and are given
explicit instructions regarding rules for replacing sample farmers who are
unavailable. The enumerators may be deployed singly, or in teams of two,
with one responsible for asking the questions and the other for recording.
Completed questionnaires should be turned in to a supervisor every
evening, and they should be checked so that gaps or inconsistencies can be 65
addressed immediately. -
The coding and analysis should begin immediately after the completion of
the survey. For an adoption survey to be useful, a report should be written
and distributed as soon as possible.
The choice of sample for an adoption study will be determined in part by
the objectives of the study. The study area needs to be defined (political
boundaries, ecology, etc.). Farmers may be selected at random from the
study area, or the sample may include farmers with particular
characteristics. The sample size will depend on the variability of the
farming population being studied and the resources available. Decisions
will also have to be made about sampling within the farm: are all fields to
be studied, or only particular ones?
The design of questions for an adoption study should follow the rules of
survey design in general, with attention to clarity, specificity, and logical
ordering. In an adoption study, particular care must be given to ensuring
that the questions do not bias the farmer towards positive responses
regarding the technologies being studied.
The survey should be implemented as efficiently as possible, and timely
analysis and write-up are crucial. The following chapter provides some
guidelines for analyzing the results of an adoption survey.
The results of an adoption survey require careful analysis and
presentation. The form and focus of the final report will depend very much
on the purpose of the study. If the purpose is only to document the degree
of adoption, then thought must be given to the factors discussed in Chapter
2 regarding the definition of adoption, the best way to present the degree
of present adoption, and the necessity of analyzing historical adoption
patterns. This type of analysis is useful for considering the rate of progress
of a technology generation effort and for helping to investigate the impact
of such an effort.
But very often we also wish to analyze the adoption patterns and to try to
explain why a technology may be reaching certain farmers and not others.
Some of the possible reasons for differences in adoption were outlined in
Chapter 3. Explaining adoption patterns requires additional analysis,
which is the subject of the present chapter.
There are two principal strategies for helping us understand why farmers
accept or reject a particular technology. One is to seek the opinions and
observations of the farmers, and the second is to do a statistical comparison
of adoption behavior with the characteristics of the farm, farmer, or
institutional environment. Both of these are valid aspects of survey
analysis. This chapter discusses the analysis and presentation of this type
Farmers usually know what they like and what they do not like about a
new technology and are able to express their opinions. These opinions will
reflect their own experience. If the adoption study encourages farmers to
provide an honest assessment and does not make them feel that the
questioning represents a test of knowledge of "modern agriculture" or
"recommended practices," then very useful information will be obtained.
Open questions, such as "Do you like variety X?" or "What problems do
you find with reduced tillage?" may not be effective, however. The formal
adoption study will be based on the experience of an informal survey
(where these open questions are more appropriate), and the questionnaire
should refer more to specific characteristics of the technology that have
been identified. Box 25 (at the end of this chapter) shows farmers' opinions
on two technologies; in one case adopters were asked to explain the
reasons for using a technology, while in the second case farmers who did
not adopt were asked to explain their choice.
Farmers' decisions regarding new technology can rarely be reduced to a
simple set of opinions, however. Farmers use a range of criteria in deciding 67
whether to change their practices, and a number of models and theories
have been developed to try to explain farmer decision-making. Such
methods are beyond the scope of this manual, but one of the most
commonly used tools, "decision trees," is illustrated in Box 26.
Although farmers are experts regarding their own fields and practices,
they may not be able to provide much information on factors that affect
other farmers. In addition, the factors that influence the adoption of a
technology may be so diverse or complex that it is unreasonable to ask a
farmer to account for them or to try to explain the pattern of adoption.
For that reason, an understanding of adoption also can benefit from a
statistical comparison of the farmers' adoption behavior with various
characteristics of the farmers' environment. In most instances it is the
farmers themselves who will be asked to describe these characteristics,
but they will not be asked to actually articulate the causal linkage
between the characteristics and adoption. That is the task of the
researcher, and it may draw upon a wide range of tools in statistical
Techniques of survey analysis are well described in a number of texts,
including Casley and Kumar (1988), Alreck and Settle (1985), and
Norusis (1991). The purpose here is only to review briefly a few of the
most common methods that are relevant to the analysis of an adoption
In attempting to explain adoption patterns by statistical analysis, the
most common approach is to compare the characteristics of farmers who
have adopted a technology with those who have not adopted, to see if
some of these differences might offer insights into the rationale for
adoption. Thus if the proportion of farmers who have access to irrigation
is much higher among adopters, we would be led to explore the
possibility that the technology is more appropriate or accessible if a
farmer has irrigation.
Just because these differences can be found does not of course constitute
proof of an association. The differences must be explored within the
context of all of the data available from the survey. Simple methods of
presenting these data and statistical tests help in making the case, but
the major responsibility in interpretation lies with the researcher, who
must provide a consistent, coherent, and logical explanation for the
adoption patterns that are observed.
One of the simplest and most useful ways of examining differences in
adoption patterns is with contingency tables, in which the cells of the table
compare the proportion of adopters and non-adopters with a particular
characteristic. This is particularly appropriate if the characteristic of interest
is a nominal one, that is, one that is represented by non-numerical
categories, such as access to irrigation (yes or no) or previous crop in the
rotation (potatoes, barley, or other). Even in cases where the variable is a
continuous one (such as farm size or number of days between land prepa-
ration and planting), it is sometimes useful to divide it into a few simple
categories (large vs. small; low, medium, high) and develop contingency
tables. The relevant statistical test of this association is the chi-square test.
In the example below, the adoption of row planting for maize is examined
in light of land preparation practices. It can be seen that there is a strong
association. Farmers who use a tractor for land preparation are much more
likely to adopt row planting; 84% of these farmers adopt row planting,
while only 29% of the farmers using manual land preparation adopt row
planting. The chi-square test shows that it is improbable that this
association could occur by chance.
Planting method by land preparation
Random 177 22 199
Row 71 114 185
Total 248 136 384
S= 105.0 d.f. = 1 p<.0001
The contingency table does not tell us why these two factors are related,
however. It could be that tractor preparation makes row planting easier; in
this case tractor preparation is a cause of row planting. It is possible
(although unlikely) that row planting "causes" tractor use, in the sense that
a farmer who decides to row plant will rent a tractor for land preparation.
The point is that we must go beyond presentation of the association and
seek an explanation.
If the variable is continuous, another option is to compare the means for
adopters and non-adopters. An appropriate statistical test in this case is the
t-test. The table below shows that fields where farmers apply fertilizer tend
to be cropped continuously for a longer time than those fields where
fertilizer is not used.
Fertilizer use by cropping history
Without fertilizer (N = 50) 3.14
With fertilizer (N = 58) 6.40
t=3.51 d.f.=106 p<.001.
The t-test shows that it is unlikely that this difference could occur by
chance. The statistical test says nothing about whether this difference might
be important, however. Neither does it say if there is any direct connection
between a farmer's decision to use fertilizer and the cropping history of the
field. It seems logical that cropping history might influence fertilizer use,
but the researcher needs to explore and elaborate this relation.
In the discussion above we have assumed that the adoption variable is
divided into only two categories, adoption and non-adoption. In some
cases, however, adoption may be measured in a more disaggregated
manner. There may be several categories of adoption (adopters, ex-
adopters, and non-adopters; or complete, partial, and non-adopters).
Contingency tables may be expanded easily to accommodate this type of
categorization. If a continuous variable such as farm size is to be compared
to adoption measured in several categories, a one-way analysis of variance
can be performed. In certain cases, adoption itself may be represented as a
continuous variable (such as kilograms of fertilizer per hectare) and other
statistical methods, such as correlation or regression analysis, will then be
It should be emphasized that the statistical tests are simply a way of
providing a quantitative estimate of the likelihood that the association
observed between two variables could have occurred by chance, even if
there were no relation between them. Most statistics texts warn readers of
the difference between statistical significance and importance. A
relationship may be shown to be very significant (i.e., very unlikely to have
occurred by chance) and yet be quite unimportant. We should always pay
attention to the degree of difference as well as the statistical significance.
For instance, a survey may show that the average farm size for adopters is
4.7 ha, while the average farm size for non-adopters is 4.2 ha. Depending
on the variability in the sample and the sample size, this difference might
be shown to be statistically significant. It is probably not surprising that
farm size exerts some influence on a farmer's adoption decision, but it is
doubtful if we would want to give much emphasis in this case to the fact
that adopters tend to operate farms that are about 10% larger than those of
non-adopters. The information probably is not useful to researchers (for
refining their recommendations) or to policymakers (in understanding the
spread or impact of new technology). If the adopters tended to have farms
that were twice as large as those of the non-adopters, on the other hand, we
would have to begin thinking about better targeting or presentation of the
If two factors are associated, there are a number of ways of explaining the
observed association. In most cases, the analysis of adoption surveys will
look for factors that may cause or contribute to the adoption (or non-
adoption) of a technology. There is no standard method for doing this type
of interpretation, and the researcher must be guided by his or her
knowledge of the data and of the farming system. Further detail on this
type of survey analysis can be found in Rosenberg (1968).
Some ways of interpreting and refining the relationship
between two variables in an adoption study
Just because there is a relationship between two variables does not mean
that one causes the other. In studying adoption, we try to identify factors
that influence a farmer's decision to use a technology. The analysis should
provide a clear and logical explanation for the contribution that a particular
factor makes to the adoption decision. This requires a careful analysis of the
relationship between variables. The following examples show how the
relationship between two variables (A and B) can be understood better by
including a third variable (C) in the analysis.
1. Identifying specific components of the causal variable
In some cases a relationship may seem reasonable, but it is possible to
find a more specific explanation. For example, farm size (A) may be
related to the adoption of contour hedgerows for erosion control (B). The
smaller the farm, the more likely the adoption. What is it about farm size
that would influence the adoption of hedgerows? If we find that small
farms have more family labor available per hectare (C), and that there is
also a strong relationship between labor availability and hedgerow
adoption, it may be sensible to emphasize this factor in our analysis of
Proposed relation: A Bi M"!.'iii s B
Revised relation: A(C) Bs4wwi iim,- B
Farm size (A) w. eM.,,im Adoption of hedgerows (B)
Plant Do not plant
hedgerows (n=22) hedgerows (n=29)
Average farm size 1.2 ha 2.3 ha
t-test prob. < .05
Farm size (A) is associated
with labor availability (C) 'v,-,..- Adoption
of hedgerows (B)
Plant Do not plant
hedgerows (n=22) hedgerows (n=29)
Family labor/ha 2.7 persons 1.2 persons
t-test prob. <.05
2. Looking for intervening variables
There is another possibility, closely related to the previous one, where a
search for a more specific or useful explanatory factor leads to
lengthening the causal chain. If it is found that oxen ownership (A) is
related to the adoption of a new, late-maturing maize variety (B), we
need to ask what is the connection. We may find that farmers with oxen
are able to prepare and plant their fields earlier (C) and are therefore
able to use the new variety without worrying about the relatively short
rainy season. The relationship between planting date (C) and variety
adoption (B) is more useful than the relationship between oxen
ownership (A) and variety adoption.
Proposed relation: A B -o-saa B
Revised relation: [A wmrn was, ] C B B
Oxen ownership (A) w wm.w -.-,, Variety adoption (B)
Adopt Do not adopt
new variety variety Total
Own oxen 28 (70%) 12 (30%) 40 (100%)
Rent oxen 7 (24%) 22 (76%) 29 (100%)
X2 prob. <.01
[Oxen ownership (A) ] Early planting (C) Variety adoption (B)
Adopt Do not adopt
new variety new variety
Mean planting date 14 April 2 May
3. Eliminating an extraneous variable
Sometimes we find an association between two variables that we do not
understand. This may be because the two are related only by chance, or
perhaps because the two variables happen to be related to a third. In
one survey an association was found between tenancy (A) and variety
adoption (B); owners rather than sharecroppers were the ones who
adopted the variety. Researchers could give no clear explanation for
this relationship. It was then discovered that sharecroppers were
concentrated in one region where the new variety had not been
promoted. When the analysis was done controlling for region (C), the
relationship between tenancy and variety adoption disappeared.
Proposed relationship A B
A Co C,
Revised relationship C (no relationship) A (no relationship) B A (no relationship) B
Land ownership (A) Variety adoption (B)
Adopt Do not adopt
new variety new variety Total
Owner 159 (60%) 106 (40%) 265 (100%)
Sharecropper 14 (26%) 39 (74%) 53 (100%)
X2 prob. <.01
The association is tested controlling for the third factor:
Region X (C0)
A (no relationship) B
Do not adopt
X2 prob. = n.s.
Region Y (C,)
A (no relationship) B
Adopt Do not adopt
124 (83%) 25 (17%)
4 (67%) 2 (33%)
X2 prob. = n.s.
4. Finding a variable that suppresses a relation
Sometimes we expect to see an association between two variables, but
we find none exists. A third variable may suppress the relation by being
positively correlated with the dependent variable and negatively
correlated with the independent variable. Preliminary analysis of an
adoption survey showed no relationship between farm size (A) and
fertilizer use (B), although researchers expected to see such a
relationship. A third variable, slope of field (C), suppressed the
relationship. More fertilizer was used on hillside fields than on flat
fields, and most large farms were on flat land. Hillside fields were
positively correlated with the dependant variable (fertilizer adoption)
and negatively correlated with the independent variable (farm size).
Proposed relation: A (no relation) B
Revised relation: A
C (*) Co C,
B A A- B A B
Farm size (A) (no relation) Fertilizer use (B)
Small farms (n=53) Large farms (n=38)
Average fertilizer use (bags/ha) 4.6 4.9
t-test prob = n.s.
The association is tested controlling for the third factor:
Farms on flat land (Co)
A ..... ; B
Farms on flat land
t-test prob. <.05
Farms on hillsides (Ci)
Farms on hillsides
t-test prob. <.05
5. Looking for conditional relationships
A causal relation may be valid, but only under certain conditions.
Preliminary analysis of a survey showed that farmers who planted early
(A) were the ones who used a cover crop (B). One explanation was that
farmers who planted late were concerned about losing their investment
in a cover crop if the rains stopped early. This would make sense,
however, only where rainfall was a problem. When the relationship was
tested controlling for the third factor, rainfall zone (C), it was found that
the relationship was evident only in the low rainfall zone.
Proposed relation: A B
Revised relation: CO C,
A." B A (no relationship) B
Early planting (A) .. Use of a cover crop (B)
Plant before Plant after
1 Nov. 15 Nov.
planting cover crop 40% 24%
The association is tested controlling for the third factor:
Low rainfall zone (Co) High rainfall zone (C1)
A w*wmw B A (no relationship) B
Low rainfall zone High rainfall zone
Plant before Plant after Plant before Plant after
1 Nov. 15 Nov. 1 Nov. 15 Nov.
cover crop 34% 12% 48% 42%
X2 prob.<.05 X2 prob. = n.s.
Wawt Logit, probit, and tobit analysis
The examples above emphasize the fact that understanding adoption
behavior often requires that the researcher look beyond relationships
between single variables or searching for simple explanations. Although
much can be accomplished by re-examining two-variable relationships and
by breaking contingency tables into component tables, there are definite
limits on our capacity to manage or interpret multifactorial relationships in
this way. There are a number of well-developed methods for looking at
multivariate relationships. One of the most common is multiple regression
analysis, but this is appropriate only if the dependent variable is
continuous. As many adoption studies will deal with adoption as a
categorical dependent variable (usually "yes" or "no"), other techniques are
required. Two related multifactorial analytical techniques that are
particularly useful for adoption studies are logit analysis and probit
Both probit and logit models are techniques for estimating the probability
of an event (such as adoption) that can take one of two values (adopt, don't
In analyzing adoption, both probit and logit models use a series of
characteristics of the farm or farmer (which may be dichotomous or
continuous variables) to predict the probability of adoption. The basic
difference between the two models is that logit assumes that the dependent
variable follows a logistic distribution while the probit model assumes a
cumulative normal distribution. For most simple problems the
interpretation of the same data, whether estimated by probit or logit, will
be very similar, with noticeable differences occurring only in the tails of the
distribution (in other words only for individuals having extremely high or
extremely low probabilities of adoption).
The following example is a simple logistic regression that examines the
adoption of a new variety. The dependent variable is variety adoption and
the independent variables are the region of the country, land tenure,
cropping practice, and farmers' age. The variables are listed below:
Variety adoption (VAR) 0 = no 1 = yes
Region (REG) 0 = forest 1 = savanna
Tenure (TEN) 0 = sharecrop 1 = owner
Monocrop (MON) 0 = intercrop 1 = monocrop
Age (AGE) (continuous variable)
The results of the logistic regression show a highly significant association
with region, no association with tenure, a significant association with
monocropping, and none with age.
Constant -2.087 0.549 -3.803 0.000
REG 2.332 0.325 7.169 0.000
TEN 0.465 0.400 1.162 0.245
MON 1.111 0.310 3.585 0.000
AGE -0.012 0.010 -1.289 0.197
Sample size = 318
Log of likelihood function = -162.9
Chi-square statistic for significance of equation = 131.1
Degrees of freedom for chi-square statistic = 4
Significance level for chi-square statistic = 0.000
Cases correctly predicted = 79.0%
This example draws on the same data used in the example on page 73. In
that example, there was a significant relationship between variety adoption
and tenure. But when we controlled for region, the relationship
disappeared. The logistic regression does the same thing, and the results
say that when we control for region (and cropping practice and age) there
is no association between land tenure and varietal adoption.
If we were to run the logistic regression without including region, we
would find that tenure appears to have a significant association with
variety adoption (see the following table). The use of multivariate
techniques such as logit models is therefore no guarantee that we will
identify important relationships and discard unimportant ones. The
opposite strategy of throwing as many independent variables as possible
into the equation is not advisable either. The use of such techniques will be
efficient only if we have a clear understanding of the data with which we
Constant -1.367 0.481 -2.842 0.005
TEN 0.893 0.353 2.531 0.011
MON 1.868 0.266 7.016 0.000
AGE -0.012 0.009 -1.445 0.148
Sample size = 318
Log of likelihood function = -192.7
Chi-square statistic for significance of equation = 71.4
Degrees of freedom for chi-square statistic = 3
Significance level for chi-square statistic = 0.000
Cases correctly predicted = 72.4%
One way of presenting information from probit or logit regressions is to
show how changing one independent variable alters the probability that a
given individual is an adopter. In ordinary least squares regression, a
coefficient can be interpreted directly as the change in the value of the
dependent variable associated with a change in one unit of the independent
variable associated with that coefficient. This is not true in profit or logit
regressions; the change in probability of adoption given a change in one of
the independent variables depends not only on the coefficient of that
variable but also on the levels of all the other independent variables.
As an example, the estimated probability of adoption in a logit model is
probability of adoption = F(b'x),
1 + e-b'x
is the cumulative logistic probability distribution. The expression b'x is
b'x = bo + blx + b2x2 + ...+ bkXk
where b0 is the constant, b, b2, ... bk are the other estimated coefficients,
and x1, x2, ... Xk are the values of the independent variables. In the first
model estimated above, the probability of new variety adoption for a forest
zone farmer who is an owner, 41 years old, and planting an intercropped
field is calculated as below:
F(b'x) = + e-bx
1 + e-b'x
Thus, the probability of such a farmer being an adopter is .11. A farmer
with identical characteristics except for planting a monocropped field
would have a probability of adopting a new variety given as follows:
Similar calculations show that a 41-year-old owner in the savanna zone has
a predicted probability of adoption of .55 if he or she intercrops and .79 if
he or she monocrops. In other words monocropping raises the predicted
probability of adoption for otherwise identical farmers in both the forest
and the savanna zone, but the change in predicted probabilities is different
79in the two different environments.
in the two different environments.
Predicted adoption probabilities in the probit model are calculated in
exactly the same way, except that F(b'x) in this case represents the
cumulative standard normal probability distribution. The mathematical
form of the normal distribution is relatively complex, but values of F(b'x)
can be found easily enough in standard probability tables.
In some instances one might want to analyze not only adoption but also the
extent or intensity of adoption. For example, one might want to estimate
both the probability that a farmer uses fertilizer and the rate of application
as well. A commonly used model in this case is the tobit model. Tobit, like
probit, is based on the normal distribution. As in probit or logit, coefficients
are estimated for the independent variables thought to be relevant. Tobit
estimates also include the standard error of the regression s, which is an
estimate of the standard deviation of the error term in the regression
In a tobit model, the predicted probability of adoption for a farmer with
characteristics x is estimated by F(b'x/s), where b, x, and s are defined as
above, and F is the cumulative standard normal distribution function. The
expected value of the application rate Y (or total amount used, depending
on the dependent variable used in the regression) is estimated by
E(Y) = F(b'x/s)(b'x) + sf(b'x/s),
where f is the probability density function for the standard normal
The expected value of Y, the application rate or total amount used, given
that Y>O (i.e., that the farmer is an adopter), is estimated by
E(Y:Y>O) = b'x + s f
As with the cumulative standard normal distribution, the standard normal
density function is also easily found in probability tables.
Maddala (1983) is a good source of information on logit, probit, and tobit
models. Multivariate analytical techniques are available in several software
packages for personal computers. It is thus quite easy to enter data on a
large number of variables and obtain long printouts and an impressive list
of coefficients and related significance levels. But just as with the simpler
analytical techniques, the responsibility for interpreting the results rests
with the researcher. A result that shows that certain variables are
associated with adoption says nothing about the causal links among those
variables. And a high significance level says nothing about the importance
of the relationship. This warning is especially relevant for multivariate
analyses, where computer technology makes it possible to examine dozens
of variables. It must be remembered that a significance level of 10% means
that the observed association has a one in ten chance of occurring at
random. If we were to include 15 or 20 independent variables in an
equation, we would expect that a couple of them would appear
"significant" just by chance, especially if we ran several regressions with
different combinations of variables. (This warning is of course equally
valid for the practice of analyzing 15 or 20 single-variable relationships
until something appears that is "significant.") A summary of some of these
analytical problems related to understanding the adoption of soil
conservation is presented by Lockeretz (1990).
Even sophisticated analytical techniques have severe limitations in their
ability to disentangle complex relationships among a number of variables.
The purpose of analyzing adoption patterns should be to identify a few key
relationships or factors that not only can help us to understand the
adoption process, but can help to make the research and extension effort
more efficient in the future. The emphasis should thus be on simple
analyses and clear presentations (Casley and Kumar 1988). We should also
remember that we are limited in our ability to understand adoption not
only by the statistical techniques at hand, but most of all by the skill with
which we frame the questions in our adoption study and the quality of the
data that are generated.
Adoption studies often attempt to analyze and understand the observed
adoption patterns. In these cases, adoption survey analysis can include
both an examination of farmers' opinions and observations, and a statistical
comparison of adoption measures with characteristics of the farmer or the
farming system. Farmers' decisions regarding a new technology are often
quite complex, and the purpose of an analysis is to try to use a range of
statistical procedures to identify the most important factors that influence
adoption. The analysis will be successful only if the information can be
used to help improve the efficiency of subsequent technology generation
and diffusion efforts.
A. An adoption survey can ask farmers' opinions about a new technology. Farmers may be able to give
both positive and negative characteristics of the technology.
In the example below, farmers in Nigeria who had adopted alley farming were asked what they saw
as its principal advantages.
Alley farmers' perceptions and uses of alley farms,a Nigeria
Trees used for:
Mulch + feed
Mulch + feed + other
Better soil fertility
Better animal performance
Better soil fertility and
Cultivated tree species preferred by farmers:
Reason for preference for species planted:
Liked by animals 75
Grows fast 5
Source: Reynolds et al. (1991).
a Based on 137 farmers. Percentages do not add to 100% because of missing data.
b Includes stakes and firewood.
B. It is equally important to ask farmers why they don't adopt. In the example below, farmers in Ghana
who did not use the recommended maize variety were asked why they did not plant it. The analysis
was divided between farmers who had used the variety in the past but stopped using it, and those
who had never used the variety.
Reasons for not using improved varieties, Ghana, 1990
Seed not available 62 64
Lack of knowledge 29 0
Storage problems 27 39
Marketing problems 20 49
Cooking quality 8 6
Yield 2 6
(Number of farmers) (98) (33)
82 Source: Ghana Grains Development Project (1991).
Note: Percentage total is more than 100% because of multiple answers.
Decision trees are a method of representing a farmer's choice about a new technology as a series of
decisions, arranged in a hierarchy. The method is described in detail in Gladwin (1989).
An example of a decision tree is shown below. It attempts to analyze farmers' decisions regarding the
use of an early-maturing maize variety ("Katumani") in Kenya. According to this analysis, the decision to
plant this variety is based on farmers' need for early maize, the availability of other early varieties, and
experience with yields and dog damage.
Good to have maize that is ready early?
Tried planting Katumani?
Have another variety for
early maize which is better
NO 3 cases
Reasons for not trying
30 NOT Dogs destroy
PLANT most of your
Getting early maize so
important that you
continue to plant
Have another Damage Yields Other
variety for by dogs are too
early maize low
I cI I I 5
DONOT DONOT DONOT DONOT
PLANT PLANT PLAN ANTL RNT
3 cases 3 cases 4 cases 5 cases
Katumani so low yielding
that the benefits of earliness
are outweighed by the
DONOT Can you obtain
PLANT seed for planting
by buying, borrowing,
5 cases or storing own seed?
n= 81 (6 cases not included
due to inadequancy of data)
Source: Franzel (1984).
Conclusions There is an urgent need to develop more productive technology for farmers
in developing countries. The resources available to public and private
institutions, and to the farmers themselves, are very limited, however.
Thus it is imperative that the process of technology generation be made as
efficient as possible. One way of improving the efficiency of agricultural
technology development is to do a better job of describing and analyzing
There is no single method for studying technology adoption. Indeed,
concerns with adoption and acceptability must form a part of the
technology generation process from its early stages. Agricultural research
that does not include a continual dialogue between farmers and
researchers will have little chance of success.
As technology is developed and made available, there are several ways of
following its progress. This manual has described one important method,
the formal survey. Its advantages include the ability to provide systematic,
quantitative information to those who must take decisions about research
or extension efforts, and its ability to formally test hypotheses that explain
Formal adoption studies can provide valuable information to research,
extension, or rural development institutions that wish to assess their
progress and take advantage of farmer experience. Adoption studies are
also valuable tools for improving the efficiency of communication between
institutions responsible for research, extension, and agricultural policy.
And finally, an adoption study can play an important role in
demonstrating the impact of a research or extension effort and in justifying
continued or expanded support from funding sources.
Properly managed, adoption studies can contribute to improving the
efficiency of agricultural research, technology transfer, input provision,
and agricultural policy formulation. Formal adoption studies should thus
be an important part of the methodology of institutions involved in
References Acharya, M., and L. Bennett. 1982. Women and the Subsistence Sector: Economic
Participation and Household Decision-Making in Nepal. World Bank Staff
Working Paper 526. Washington, D.C.: The World Bank.
Akhtar, R., D. Byerlee, A. Qayum, A. Majid, and P. Hobbs. 1986. Wheat in the
Cotton Wheat Farming Systems of the Punjab: Implicationsfor Research and
Extension. Islamabad, Pakistan: Pakistan Agricultural Research Council.
Alreck, P.L., and R.B. Settle. 1985. The Survey Research Handbook. Homewood,
Ashby, J.A.. 1982. Technology and ecology: Implications for innovation
research in peasantagriculture. Rural Sociology 47(2): 234-250.
Ashby,J. A.. 1990. Evaluating Technology with Farmers: A Handbook. Cali,
Colombia:Centro Internacional de Agricultura Tropical.
Barker, R., R.W. Herdt, and B. Rose. 1985. The Rice Economy ofAsia.
Washington, D.C., and Manila, Philippines: Resources for the Future and
the International Rice Research Institute.
Benor, D., and J.Q. Harrison. 1977. Agricultural Extension: Training and Visit
System. Washington, D.C.: The World Bank.
Bernsten, R. 1980. Design and Management ofSurvey Research: A Guide for
Agricultural Researchers. Occasional Paper No. 8. Nairobi, Kenya: CIMMYT.
Beyene Seboka, Asfaw Negassa, W. Mwangi, and Abubeker Mussa. 1991.
Adoption of Maize Production Technologies in the Bako Area, Western Shewa and
Welega Regions of Ethiopia. Addis Ababa, Ethiopia: Institute of Agricultural
Birkhaeuser, R.E., R. Evenson, and G. Feder. 1991. The economic impact of
agricultural extension: A review. Economic Development and Cultural Change
Braun, J. von, and P.J.R. Webb. 1989. The impact of new crop technology on the
agricultural division of labor in a West African setting. Economic
Development and Cultural Change 37(3):513-534.
Byerlee, D., M. Collinson, etal. 1980. Planning Technologies Appropriate to
Farmers: Concepts and Procedures. Mexico, D.F.: CIMMYT.
Byerlee, D., and E. Hesse de Polanco. 1986. Farmers' stepwise adoption of
technological packages: Evidence from the Mexican Altiplano. American
Journal ofAgricultural Economics 68:519-527.
Casley, D.J., and K. Kumar. 1988. The Collection, Analysis, and Use ofMonitoring and Evaluation Data.
Washington, D.C.: The World Bank.
Casley, D.J., and D.A. Lury. 1981. Data Collection in Developing Countries. Oxford, UK: Clarendon
Denning, G.L. 1991. Intensifying rice-based cropping systems in the rainfed lowlands of Iloilo,
Philippines: Results and implications. In Planned Changein Farming Systems: Progress in On-Farm
Research, ed. Robert Tripp. Chichester, UK: John Wiley. Pp. 109-142.
Farrington, J., and A. Martin. 1988. Farmer Participation in Agricultural Research: A Review ofConcepts
and Practices. Agricultural Administration Unit, Occasional Paper9. London, UK: Overseas
Feder, G., R.E.Just, and D. Zilberman. 1985. Adoption of agricultural innovations in developing
countries: A survey. Economic Development and Cultural Change 33(2):255-298.
Franzel, S. 1984. Modelling farmers' decisions in a FarmingSystems Research exercise: The
adoption of an improved maize variety in Kirinyaga District, Kenya. Human Organization 43:199-
Ghana Grains Development Project. 1991. A Study ofMaize Technology Adoption in Ghana. Kumasi,
Ghana: Ghana Grains Development Project.
Gittinger, J.P. 1982. Economic Analysis ofAgricultural Projects. 2nd edition. Baltimore, Maryland:
Gladwin, C. 1989. Ethnographic Decision Tree Modeling. Sage Qualitative Research Methods Series
No.19. Beverly Hills, California: Sage.
Grandin, B.E. 1988. Wealth Ranking in Smallholder Communities: A Field Manual. London, UK:
Intermediate Technology Publications.
Griliches, Z. 1957. Hybrid corn: An exploration in the economics of technological change.
Hassan, R.M., W. Mwangi, and D. Karanja. 1992. Wheat supply in Kenya: Production technologies,
sources of inefficiency, and potential for productive ty growth. Draft paper.
Heisey, P. (ed.). 1990. Accelerating the Transfer of Wheat Breeding Gains to Farmers: A Study of the
Dynamics of Varietal Replacement in Pakistan. CIMMYT Research Report No. 1. Mexico, D.F.:
Hoeper, B. 1988. The training and visit system of agricultural extension in practice: Investigations
on its operations in two Indian states. Quarterly Journal of international Agriculture 27:247-276.
Jha, D., B. Hojjati, and S. Vosti. 1991. The use of improved agricultural technology in Eastern
Province. In Adopting Improved Farm Technology: A Study ofSmallholder Farmers in Eastern Province,
Zambia, ed. R. Celis, J. Milimo, and S. Wanmali. Washington, D.C.: International Food Policy
Lipton, M., with R. Longhurst. 1989. New Seeds and Poor People. London, UK: Unwin Hyman.
Lockeretz, W. 1990. What have we learned about who conserves soil? Journal of Soil and Water
Low, A. 1986. Agricultural Development in Southern Africa. London, UK: James Currey.
Maddala, G.S. 1983. Limited-Dependent and Qualitative Variables in Econometrics. Cambridge, UK:
Cambridge University Press.
Martinez, J.C., and G. Sain. 1983. The Economic Returns to Institutional Innovations in National
Agricultural Research: On-Farm Research in IDIAP, Panama. CIMMYT Economics Program
Working Paper04/83. Mexico,D.F.:CIMMYT.
Murphy,J., and L.H. Sprey. 1982. Monitoring and Evaluation ofAgricultural Change. The Netherlands:
Norman, D., D. Baker, G. Heinrich, and F. Worman. 1988. Technology development and farmer
groups: Experience from Botswana. Experimental Agriculture 24: 321-31.
Norton, G.W. and J.S. Davis. 1981. Evaluating returns to agricultural research. American Journal of
Agricultural Economics 63:685-699.
Norusis, M.J. 1991. The SPSS Guide to Data Analysisfor SPSS/PC+TM. 2nd Edition. Chicago, Illinois:
Poate, C.D., and D.J. Casley. 1985. Estimating Crop Production in Development Projects: Methods and
Their Limitations. Washington, D.C.: The World Bank.
Reynolds, L., C. di Domenico, A.N. Atta-Krah and J. Cobbina. 1991. Alley farming in South-Western
Nigeria: The role of farming systems research in technology development. In Planned Change in
Farming Systems: Progress in On-Farm Research, ed. RobertTripp. Chichester, UK: John Wiley. Pp.
Rogers, E.M. 1983. Diffusion of Innovations. New York, New York: Free Press.
Rosenberg, M. 1968. The Logic ofSurvey Analysis. New York, New York: Basic Books.
Ruiz de Londoflo, N., and W. Janssen. 1990. Un caso de adopci6n de tecnologia: La variedad defrijol
Gloriabambaen Peru. CIATWorking Document No. 61. Cali, Colombia: Centro Internacional de
Scott, C. 1985. Samplingfor Monitoring and Evaluation. Washington, D.C.: The World Bank.
Smale, M. 1987. Wheat Harvest Technology in Punjab's Rice-Wheat Zone: Combines, Laborers and the Cost
ofHarvest Delay. PARC/CIMMYT Paper87-23. Islamabad, Pakistan: Pakistan Agricultural
Smale, M., with Z.H.W. Kaunda, H.L. Makina, M.M.M.K. Mkandawire, M.N.S. Msowoya, D.J.K.
Mwale, and P.W. Heisey. 1991. Chimanga Cha Makolo," Hybrids and Composites: An Analysis of
Farmers' Adoption ofMaize Technology in Malawi, 1989-1991. CIMMYT Economics Working Paper
91/04. Mexico, D.F.: CIMMYT.
Sperling, L., G. Randrianmapita, E. Rutagengwa, B. Ntambovura, L. Mubera, and L. Uwimana.
1992. Etude de adoption et impact de varieties ameliorees en milieu reel. Proceedings ofa Regional
Seminar, Programme pour 'A melioration des Haricots dans La Region des Grands Lacs, Kigali, 21-25
January 1991. CIAT African Workshop Series No. 17. Kigali, Rwanda: Centro Internacional de
Spring, A. 1985. Reaching female farmers through the male extension staff. Paper presented at
Farming Systems Research Symposium, 14 October, 1985, Manhattan, Kansas.
Thirtle, C.G., and V. Ruttan. 1987. The Role of Demand and Supply in the Generation and Diffusion of
Technical Change. New York, New York: Harwood Academic Publishers.
Traxler, G., and D. Byerlee. 1992. Economic returns to crop management research in a post-green
revolution setting. A merican Journal ofAgricultural Economics 74:573-582.
Tripp, R. 1982. Data Collection, Site Selection and Farmer Participation in On-Farm Experimentation.
CIMMYTEconomics Working Paper No. 82/1. Mexico, D.F.: CIMMYT.
Tripp, R., K. Marfo, A.A. Dankyi, and M. Read. 1987. Changing Maize Production Practices ofSmall-
Scale Farmers in the Brong-Ahafo Region, Ghana. Mexico, D.F.: Ghana Grains Development Project.
Tripp R., and J. Woolley. 1989. The Planning Stage ofOn-Farm Research: Identifying Factorsfor
Experimentation. Mexico, D.F. and Cali, Colombia: CIMMYT and Centro Internacional de
I S BN -686127-77-1
Lisbo 27, A o Posta 6-64, 660 M-io F Mexico