<%BANNER%>
Evaluating a method for defining recommendation domains
CITATION THUMBNAILS PAGE IMAGE ZOOMABLE
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/AA00008198/00001
 Material Information
Title: Evaluating a method for defining recommendation domains a case study from Kenya
Physical Description: 23, 3 leaves : ill., maps, ; 28 cm.
Language: English
Creator: Franzel, Steven
Publisher: s.n.
Place of Publication: S.l
Publication Date: 1985?
 Subjects
Subjects / Keywords: Agriculture -- Research -- On-farm -- Kenya   ( lcsh )
Genre: bibliography   ( marcgt )
non-fiction   ( marcgt )
Spatial Coverage: Kenya
 Notes
Bibliography: Includes bibliographical references (leaf 21)
Statement of Responsibility: Steven Franzel.
General Note: Caption title.
General Note: Typescript.
 Record Information
Source Institution: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: oclc - 756841418
ocn756841418
Classification: lcc - S540.O53 F7 1985
System ID: AA00008198:00001

Full Text

o/. ~lz~


EVALUATING A METHOD FOR DEFINING RECOMMENDATION DOMAINS: A CASE
STUDY FROM KENYA

Steven Franzel
Development Alternatives, Inc.

Introduction

In recent years, the concept of "recommendation domain"
has emerged as an important tool in farming systems
research and extension (FSR/E). A recommendation domain (IM)) is
defined as "a group of roughly homogenous farmers with similar
circumstances for whom we can make more or less the same
recommendation" (Byerlee, et. al., 1980). One of FSR's principal
distinguishing characteristics is that research must be location
and farmer-group specific. Defining recommendation domains in an
area assists scientists to focus their efforts on the particular
problems of different farmer groups, who may vary according to
physical (e.g., soil and rainfall), biological (e.g., crops and
pests) or socio-economic (e.g., income and food preference)
variables.

There are varying opinions in the literature as to when and
how researchers should define recommendation domains. Most
researchers recommend making hypotheses concerning domain
boundaries at the beginning of the research process, based on
reviews of secondary data and interviews with key informants or
during the exploratory survey [1] (Harrington and Tripp, 1984;
Hildebrand and Poey, 1985). These hypotheses are then tested and
refined during the exploratory survey, formal survey, and
experimental program which make up the on-farm research
process. Collinson (1982) proposes a distinct survey
exercise preceding the exploratory survey in which researchers
administer a brief questionnaire to extension agents. The
purpose of this exercise is to elicit information for making a
preliminary delimitation of recommendation domains.

The purpose of this paper is to evaluate the quality of a
brief survey of extension agents, called an RD-definition
exercise in this paper, similar to that proposed by Collinson.
The principal advantage of this method is that it enables
researchers to collect a great deal of information about a large
area in a short period of time. The method also helps
researchers to decide where and for which farmers they want to
focus their efforts, thus increasing the efficiency of the
research process. However, conducting a questionnaire


[14 A formal survey is a survey in which data are
collected by means of a questionnaire, which is administered by
enumerators to a randomly selected sample of farmers.
Exploratory surveys, also called rapid reconnaissance surveys,
informal surveys, or sondeos, are surveys in which researchers
themselves interview farmers using informal, unstructured
techniques in order to encourage dialogue and probing of issues.








survey of extension agents before conducting a survey of farmers
is open to a broad range of error, since researchers know little
about the farmers and area and the extension agents may lack
knowledge or give biased responses. If the errors are
significant, the exercise can be wasteful and misleading.

This paper evaluates the quality of information obtained in
an RD-definition exercise in Kenya by comparing the information
obtained with that of a formal survey of farmers carried out in
the same area during the same cropping season. In the first
section of this paper, the concept of recommendation domain is
discussed, including how recommendation domains differ from
conventional zoning exercises. Next, the research design is
presented, including the design, implementation, and results of
the RD-definition exercise and the informal and formal farmer
surveys that followed. Finally, data from the RD-definition
exercise are compared to data from the formal survey. This
comparison is used for evaluating the usefulness of the exercise
for (1) demarcating RD boundaries and (2) for providing
preliminary information about farmers in each.


Recommendation Domain Concept

The concept of grouping farmers in order to focus
agricultural research is not new. Many national agricultural
research programs have delineated zones in their countries
according to rainfall, altitude, or agro-ecological variables in
order to better focus their research and extension programs. For
example, the delineation of altitude zones in Kenya has
facilitated the development of hybrid maize varieties suitable
for each zone (National Agricultural Research Station, Kenya,
1979). In Rwanda, researchers have divided the country into 12
agro-ecological zones according to soil types, altitude and
rainfall (USAID, 1983).

Zoning represents a compromise which researchers make between
two extremes in deciding how to make their research efforts most
cost-effective. At one extreme, one could argue that each farm
is different and that researchers should thus work with each
individual farmer to design packages to fit his or her own
circumstances. However, this approach is not practical, given
the high cost of conducting research and the scarcity of trained
researchers in third world countries. At the other extreme, one
could argue that research funds are best economized by drawing up
a single research agenda to encompass an entire country.
However, this approach is clearly inappropriate, since farmers in
different areas of a country have different environments and
different needs. The compromise which researchers arrive at in
zoning exercises is to group areas according to those variables
considered most important in determining the growth of the crop
they are working with.

However, the concept of recommendation domain has one








principal characteristic which distinguishes it from traditional
concepts of zoning. Whereas zoning approaches have traditionally
focused on biological and physical aspects of geographical
areas, recommendation domains focus on groups of
farmers, not geographical areas. This distinction is important
because it is farmers, not geographical areas, which decide
whether or not to adopt a new technology. The implication is
that socio-economic variables play an important role in
determining adoption just as do climatic, geographical and
ecological variables. Thus, changes in a particular socio-
economic variable, such as income level, may be an important
determinant in the adoption of a specific variety, just as a
change in rainfall from one area to another influences the
variety which farmers choose to grow. Other socio-economic
variables, which may play an important role in influencing the
appropriateness of a recommendation include access to markets and
inputs, access to cash, power source, food preferences, and farm
size (Harrington and Tripp, 1984). Thus, researchers defining
recommendation domains in an area need to consider socio-economic
variables as well as physical, climatic, and biological
variables.

The most critical issue in defining recommendation domains
is how to decide which variables are most important in delimited
farmers. Recommendation domains are differentiated according to
the expected responses of groups of farmers to recommendations,
that is, suggested solutions to their problems (Flora, 1985).
However, this presents researchers with an apparent paradox: how
can we be sure of the constitution of a recommendation domain
before the recommendation has been made? As Harrington and
Tripp, 1984, point out, the answer is that we cannot be
completely certain, but as the process of on-farm research passes
from diagnosis through experimentation to recommendations
researchers become more and more confident of the boundaries of
their domains. At the onset of the research process, before
actual recommendations or even target crops are identified,
farmers may be grouped into recommendation domains according to
their principal circumstances and problems. This can be done
because farmers with similar circumstances and problems are
likely to have similar development opportunities; a given
recommendation will likely be more or less applicable to all
farmers in the domain. As the research process unfolds,
researchers learn more about the farmers they are working with
and conduct experiments to formulate recommendations for farmers.
These activities permit researchers to further refine the domains
to ensure that the domain represents a group of farmers who will
respond favorably to a $iven recommendation. Harrington and
Tripp, 1984, provide guidelines and numerous examples of how to
select variables for defining recommendation domains.

Defining recommendation domains in a particular area serves
several purposes. First, before beginning diagnostic surveys,
defining recommendation domains in a region can help researchers
to decide where and for which farmers they want to focus their
efforts. For example, in 1979, CIMMYT assisted the Ministry of









Agriculture And Water Development, Zambia, to define
recommendation domains in Central Province, which has an area of
116,000 square miles and 349,000 people (CIMMYT Eastern Africa
Economics Program, 1979). Researchers conducted a brief survey
of extension agents and used the information to define eight
recommendation domains; policy makers and researchers used this
framework to help them decide which recommendation domains to
focus their research on.

A second example further illustrates the utility of defining
recommendation domains before beginning farmer surveys. In
Guinea in 1985, the newly constructed Tindo research center was
assigned the task of serving an area known as "Upper Guinea",
which makes up one-quarter of the country. However, the research
center has neither the staff nor the resources to work in all
areas of Upper Guinea. The center has decided to begin work in a
pilot area near the center but it is not known which if any of
the nearby areas are representative of larger areas of Upper
Guinea. Therefore, a brief study similar to the one conducted in
Zambia has been proposed in order to define recommendation
domains in Upper Guinea, This exercise will assist the research
center to select a few of the defined domains to focus their
efforts on.

Second,. during the FSR/E stages of diagnosis, design,
testing and extension, defining recommendation domains assists
researchers to assure that their activities are targeted towards
the needs and circumstances of specific groups of farmers. For
example, in an area suitable for vegetable production,
researchers may delineate recommendation domains according to
access to transportation; due to the high perishability of most
vegetables, they should not be promoted in areas without reliable
road access. Promotion of vegetables would then be targeted
only for good-access domains with reliable road access. During
the diagnostic stage, researchers identify the problems and
circumstances of a specific group (or specific groups) of farmers
and propose experiments to solve these problems. During the
design and testing stages, researchers conduct on-farm
experiments with farmers and at sites which are representative of
the recommendation domain they are working in. Data from the
trials are analyzed and recommendations are then formulated and
extended with well-defined groups of farmer clientele in mind
(Harrington and Tripp, 1984).

Third, recommendation domains are useful groupings for
policy makers and planners to use in designing and implementing
their programs. Collinson, 1982, claims that the domain
framework may be used by policy makers as an interface between
national and local interests in guiding decisions and clarifying
the implications of decisions concerning rural development
policies and programs. For example, greater emphasis on foreign
exchange earnings as opposed to food self-sufficiency at the
national level will require policies and resources to be shifted
to those domains which have greatest potential for producing
export crops.









Research Design


In 1981, the medium altitude area (1100 to 1400 meters) of
Kirinyaga District, called Middle Kirinyaga in this paper, was
selected as the study area for a farming systems diagnostic
survey to be conducted by the Scientific Research Division,
Ministry of Agriculture, in collaboration with the CIMMYT Eastern
Africa Economics Program. The area was selected because it was
felt that an adaptive production research program could offer
substantial benefits to local farmers, as well as contribute to
several of the Kenya government's policy objectives. First, the
area is low income without major cash crops and the government
places a high priority on increasing incomes in low-income, small
farmer areas. Second, maize is an important crop in the area and
achieving self-sufficiency in maize, Kenya's principal food
staple, is another important policy objective. Third, the area
has significant potential and good infrastructure yet adoption
rates for recommended inputs and practices are low. Thus
researchers felt that revised research recommendations based on
on-farm experiments could contribute significantly to increasing
production.

Next, a brief exercise to define recommendation domains was
conducted which took approximately two weeks and involved
interviews with extension agents and local subehiefs (local
administrators). The purpose of this exercise was to define
recommendation domains in the study area and the exercise was
considered important for two reasons. First, by defining
recommendation domains in the- area, researchers could then
systematically select those domains which they wanted to focus on
in their surveys and experimental programs. Second, by defining
boundaries between domains, they could be sure that farmers
outside the domains under study were not included in the surveys
and experiments, thus wasting valuable time and resources.

Two separate tasks were required in the RD-definition
exercise. One was to identify geographical, or across-area,
differences among farming systems. Across-area differences are
generally caused by physical factors--e~g., climate and
altitude--but may also be a result of historical or socio-
economic differences--e~g., ethnic group or government settlement
schemes.

The second task was to examine within-area differences in
farming systems, that is, to check whether two or more
recommendation domains may be interspersed in a particular
zone. Within-area differences are usually caused by
socioeconomic and historical factors such as ethnic group,
income, or participation in a government credit program, but may
also be caused by physical factors, such as soil type.

The exercise to define recommendation domains in Middle
Kirinyaga included three methods. First, secondary data on the










area were assembled. Unfortunately, these proved to be of
limited use because most of the data had been averaged across the
entire district, which ranges from 300 to 5,000 meters in
altitude. Second, researchers, government officials, and local
leaders familiar with the area were informally interviewed about
across-area and within-area differences. Third, a short, two-
page questionnaire, presented in Appendix 1, was administered to
extension workers and sub-chiefs in each sublocation, the
smallest administrative unit, of Middle Kirinyaga. In this
exercise, the respondents were asked about the principal
characteristics of the farming systems in their area and
characteristics that differentiated farmers between and within
areas. Separate questionnaires were completed for low income and
high income farmers, since it was hypothesized that there were
significant differences in the circumstances of each group. The
questionnaire was drawn up following discussions with key
informants; individual questions concerned such aspects as
physical environment, cropping pattern and practices, livestock,
and sources. of income.

The interviews were conducted by the researcher; each
interview lasted about 40 minutes. Twelve questionnaires
covering 15 sublocations in Middle Kirinyaga and adjoining areas
were completed in one week; most of these interviews were
_completed when extension agents attended a monthly meeting at the
district agricultural office. Data tabulation and analysis
took another. three days. Data were tabulated on large sheets
with questions listed down the side and areas listed across the
top (see Figure 1). This method assisted researchers in
identifying important differences among areas. More details on
methods to use in RD-definition exercises are found in CIMMYT
Eastern Africa Economics Program, 1979 and Franzel, 1981b.

The results of the exercise to define recommendation domains
are reported in Franzel (1981a). The exercise was followed by an
informal survey and a formal survey, both executed during the
same season as the RD-definition exercise. During both of these
surveys, modifications were made in the definition of
recommendation domains based on available information. The
results of the exercise, the across-area and within-area
differences in farming systems in Middle Kirinyaga, are described
in the two following sections.

Across-Area Differences

Map 1 presents the boundaries of Middle Kirinyaga, based on
the results ~of the exercise. The initial area surveyed was about
300 square kilometers inhabited by 55,000 people; the area
defined as Middle Kirinyaga was composed of about 170 square km.
with a population of about 35,000 people. Three other
recommendation domains bordering on Middle Kirinyaga were also
defined. Principal agro-ecological and socio-economic
characteristics of Middle Kirinyaga included:

a) 1,100 mm. of rainfall distributed into two seasons





Srvc~t
r+F~tj~5 ~


FJ. -9' r Red /
7. .I


' 8,


ri


; i, t-. 7 u '
I~l~LI _


f;~ A~J


Gde sl.,
6 Fltp


fl-t-
yeth sov


~Srp- !`;cr
~561-


c
s

IJ;b)r~p Z ~tr f i hrN


_ _ __. I


I' --i
'"~sIr~


_


I- --J


A $..~ 3, 30J -Jo


~L_ ___ _~ ~_


_OL" 'lf llo ) I1; clS a 1r;--j-4


_ __ T


t~


_ _~__


_ _1 _


C----~-------~ -


SarreS

.,


--- -----f---r---i~ ---+---i


__ __ _2_


1


f I-


*c *


- -- --- -- - --


-I
i-


L


~f_ I __ 1 L 1_1


Example of Part of a Tabulation Sheet from
Exercise to Define Recommendation Domains,
Middle Kirinyaga.


Figure 1:


3 & .. e M

Q~~T~~I 5 I O ~~oI1K 9 j
Skit &>< trus / Met


~- CO' L;~~~rlr-:L, IHC *J1~
r


ilrto '1


6C
1/4Y 0C


ri



?i


< o


100 a


toof ;


irlfz;c~z~c
c;CS. IAIG


- -10 b i


GiF/n
k, at


Gr-4~ si
Ptd


tfGA :


fl~bj


~af ~iJ **0 ,Iboo


ILSD-lro


libe-1 3as


A++tr- s *;.. Jr.


"KV ~fd
Ic anh I~


)T** Pr'
*5~Y90 fir.J


~L(~


( er) .ro (Il,.,rC)


~r2rcy


f-7 est


f-f ar


f o/n sixt


r--a


SArc- cu t


i 0ro,


h ; n, r.ilc1


I
; bt-
t


**-So./. -


A.,+ Zaz/e.


Je 4,,e C


2t~o/ cr


F. Jm;
f (75 & ,-
---1- ~~.- ~---t--
,(-ltl~c IYI'Z1ca


.<



^u e


Av,,t


r., u
Es ~.


S4 f


FS/,...!F:
fra~ lr6 G1)~ Ker


1kc 1GAC~ ~c


UK/uz -


ht're
C he.' ft ter


~~L~~le- klc,-c ----
i;--
-w .df ) npg
0 .,- o 5 ft
.rr-f~ 6 E .f


~I~i;;'C~. o l~~i;









4 4 I _-
-. go rb-~ k~ NDUINI r
u ,KIANJANG'A uso4aa **
JuE )K yet a Kut .*ko
St Kiombwe q U Klramm Sch .
spKIbIngott ,I II Sch Is rr

nK IBINGOTI '- Nd *b
~r*lrr I. "
botse K AIT 3RUI W
UI I Kiondat L 1 ND Ong 8
Kingo o






GA ARO a Me one;
echoru a K ug

ch KEY

IrGAC~KI YA R K
Bunare of Midl Kirinyagaq,IC~


1 n Arena incued in Middle Kirinyaga

SAG gon A c ` ** following RD-identification
sesns ch. f exercise but excluded later *****~
.-pI~~ an., ./ Kondon\
AGANA TOWN





Map 1. Middle Kirinyaga








b) red loam soils and a flat to mildly sloping terrain
c) farm size is small averaging 2-4 ha
d) principal food and cash crops are maize and beans, which
are almost always intercropped
e) some farmers own ox-plows while others rent
f) most farmers own a few head of cattle, for milk and/or
for draft
g) The busiest period of the year is the period of March
through May, when farmers prepare their land, plant, and weed
their maize and beans

The boundaries of Middle Kirinyaga represent fairly distinct
changes in the agro-ecological and socio-economic environment.
These differences, in turn, cause differences in the way farmers
manage their farms. For example, to the north of middle
Kirinyaga, altitude increases, rainfall is higher and the terrain
is steeper. Population density is high, farm sizes are smaller,
and coffee and dairy are important cash enterprises. Exotic
breeds of cattle replace local breeds as one moves north, since
the climate of the higher areas is more suitable for the exotic
breeds. Hand hoes replace ox plows as the principal method of
preparing land in the higher areas since terrain is steeper, and
fields of annual crops are smaller. Maize takes longer to mature
in the higher areas; thus only one crop is grown per year whereas
in Middle Kirinyaga, two crops are grown.

These changes, of course, occur rather gradually and at
different rates as one moves northward. However, the boundary
drawn between Middle Kirinyaga and areas to the north broadly
reflects the above mentioned distinctions. It is clear that
circumstances are so different between the two areas that
problems and research opportunities will differ considerably.
For example, a longer season maize variety is required for the
upper zone; a short-cycle drought-avoiding variety would be more
appropriate for Middle Kirinyaga.

In contrast to the northern boundary, the southern boundary
of Middle Kirinyaga is a fairly distinct line representing a
change in soil type from red loam to heavy, black clay. The
black soils area is characterized by somewhat different crops,
such as cotton, sunflower, and green gram. There are also
significant differences in the cropping calendar, larger farms,
and greater numbers of cattle than in Middle Kirinyaga. To the
east and west of Middle Kirinyaga are dry, hilly areas with less
fertile soils. These areas have lower population densities, and
more uncultivated land than Middle Kirinyaga.

Within-Area Differences

The task of assessing within-area differences among farmers
in Middle Kirinyaga was more difficult than identifying across-
area differences. Two factors were considered in the
investigation:

1. Characteristics of the Farming System: Researchers








sought to identify whether there were important differences in
the way farmers managed their farms--their priorities, the
resources they used, their constraints, and the strategies and
practices they employed to use available resources to best meet
their priorities. Important differences in management, it was
felt, would probably reflect important differences in research
needs and development opportunities.

2. Potential for Change: Researchers were also concerned
with the reaieptnilfrchange among farmers in the area.
Two farmers may be operating their farms in the same manner, but
have different potential for change because of different resource
availability.

In Middle Kirinyaga, it appeared that access to cash income
was an important distinction among farmers determining both the
characteristics of the farming system and the potential for
change. For example, access to cash influenced whether the
farmer undertook certain enterprises, such as owning exotic-breed
cattle. Managing these cattle requires substantial cash and
other resources for purchasing feed, transporting water to the
home, and protecting the animals against disease. Fur their, it
appeared that low income farmers made much less use of purchased
inputs such as hybrid maize seed and maize insecticide than
higher income farmers. Also, high income farmers tended to own
oxen or were able to hire oxen as soon as the rain started, thus
taking full advantage of the brief rainy periods for growing
their crops. On the other hand, fewer low-income farmers own
oxen and those without oxen tended to plant late, relying on
social contacts rather than cash payments in order to secure an
ox-plow team.

It was also hypothesized that income level was associated
with many other aspects of the farming system aside from
enterprise choice and crop husbandry. For example, food security
appeared to be an entirely different problem for each of the two
groups. Nearly all high income farmers appeared to obtain a
regular flow of cash from a non-farm enterprise or a farm
enterprise such as dairy; therefore cash was always available for
purchasing food when required. However, low income farmers were
forced to hire out their labor or sell other resources, such as
livestock, when their food supplies ran short.

These differences in the way farmers operate their farms
affect the type of experiments to be planned for the two domains.
For example, the levels of non-experimental variables in maize
experiments for the two domains should reflect differences in the
practices of the two groups, e.g., time of planting, variety, and
plant population. Further, the number of and range in
experimental variables could generally be greater for high income
farmers, since low income farmers lacked cash for purchasing
improved inputs.

Two further issues concerned the number of income groups to
establish and how to draw lines between them. We decided to








classify farmers into two groups--those who could afford modest
investment in their farms and those who could not, because it was
thought that this division would cover most of the variation
among farmers in the area. A set of proxies for income were
drawn up-to differentiate high income farmers from low income
farmers for the informal survey. The proxies for high income
farmers included ownership of exotic breeds of cattle, house
type, past land purchases, and type of off-farm income. A
subjective weighting of these variables was used to allocate
farmers between the two groups. In only a few cases, was there
any uncertainty as to which group a farmer belonged.


Effectiveness of the Exercise for Defining
Recommendation Domains

This section focuses on evaluating the exercise for
defining recommendation domains and is divided into three parts.
First, sources of error in the exercise and measures taken to
reduce their effect are examined. Next, the quality of data
obtained in the exercise is evaluated by comparing data obtained
with data collected in the formal survey. Finally, the
delineation of RD';s, based on the RD definition exercise, is
compared with the groupings arrived at following the formal
survey.

Sources of error

Eight potential sources of error which were encountered in
the RD-definition exercise are listed below. It is important to
examine these sources so as to understand why some estimates were
incorrect and so as to avoid or correct for these in future
exercises. Five of the sources stem from the respondents, i.e.,
extension agents and sub-chiefs, and three from the researchers.

A. Respondent-based sources of error.

1. Bias towards "progressive" farmers.

Respondents tended to frame their answers about farmer
circumstances with the high income, "progressive" farmers in
mind. The respondents, most of whom are extension workers, tend
to have more contact with such farmers (Leonard, 1977).
Furthermore, when asked about average or typical farmers, the
respondent frequently interpreted these as meaning "best"*
farmers. In order to reduce these biases in the Middle Kirinyaga
RD definition exercise, the respondent was repeatedly reminded of
the difference between "progressive" and typical farmers.

2, Bias due to attempt to impress investigators.

It is natural for many extension agents to try to impress an
outsider by telling him how many recommended innovations their
farmers have adopted. Further, they may feel that the researcher
is actually evaluating their work. It is difficult to








differentiate this bias from the above one, but both lead to
similar results--a bias towards high income, progressive farmers.
To counter this bias, respondents were reminded that the
researchers goal was not to evaluate their work, but to
understand farmer circumstances. Respondents were also reminded
that low adoption rates may be caused by many factors other than
poor extension.

3, Inaccurate information due to inability of
respondent to deal with percentages.

Many respondents have only primary school education and
some, particularly older ones, are not comfortable dealing with
percentages. Some progress can be made by rephrasing questions
in layman's terms, e~g., "if I were to select ten persons by
chance at a village meeting where all heads of household are
present, how many would have exotic breeds of cattle?" Further,
a continue of all-most-half-some-few may be substituted for
percentages. However, even these terms may confuse the
respondent.

4. Inaccurate information due to respondent'is lack of
knowledge.

Two types of responses may be included in this category. In
some cases, respondents may fear telling the investigator that
they do not know the answer to a question concerning farmers in
their district because it makes them look like they do not know
their job. Thus, they hazard a guess and give a false answer.
In other cases, the respondent thinks he knows the answer to a
question but is mistaken.


5. Bias because answer refers only to area which
respondent knows best.

Some respondents may give an answer which applies to the
area which the respondent knows best, usually the area around his
home, but not to other areas of the region he serves. In the RD
definition exercise in Middle Kirinyaga, the effect of this bias
was reduced by first identifying on a map the respondent's area
of work, home and various landmarks within the area. The
interviewer consistently referred to these landmarks to try to
keep the respondent's focus on the whole area and not just the
section he knew best.

B. Researcher-based sources of error.

1. Bias due to attempt to compensate for perceived
respondent biases.

In some cases, the researcher feels he must discount or
alter an extension agent's response because it is inconsistent
with the responses of other respondents, other observers familiar
with the area, or because it contradicts the researcher's own








observation. However, these compensatory biases may be incorrect
or may overcompensate for respondent bias.


2. Improper phrasing of question or inappropriate
narrowing of possible responses.

In the early stages of an investigation, researchers lack
sufficient knowledge about an area to appropriately phrase
questions; they thus may inadvertently guide answers away from
realistic possibilities. For example, in this study, the
researcher asked respondents to name the methods which farmers
used to prepare their land, e.g., own oxen, hired oxen, tractor,
etc., and estimate the percentage of farmers using each method.
Unfortunately, the researcher never considered the possibility
that some farmers may not prepare their land at all; that many
farmers did not was discovered during the informal survey. It is
not clear that extension agents were aware that some farmers
practice no-till but the manner in which the question was phrased
did not encourage them to offer this information.

3. Bias due to incorrect assessment of composition of a
particular RD.

Most respondents were asked to give answers to two
questionnaires-- one for high income farmers, and one for low
income farmers. At the time of questioning, the researcher
believed that most high income farmers were in-migrants and thus
orientated the questions around in-migrants. However, later it
became clear that most high income farmers were natives of the
area. In-migrants differ from other high income farmers in some
important respects; responses concerning such characteristics
were biased towards in-migrants.

Evaluation of Data Describing Recommendation Domains

The system of accurday ratings for parameter estimates was
developed based on how close an estimate in the RD-definition
exercise was to sample estimates from the formal survey. The formal
survey data are assumed to be accurate, on the grounds that the
survey was (1) administered to a random sample of farmers, (2)
used a questionnaire, thus a set of standardized questions for
all respondents, and (3) was preceded by a comprehensive informal
survey, which was important for developing an understanding of
the local terms of reference critical for questionnaire
development.

"Closeness" was measured in three different ways, depending
on the type of parameter estimated. If the parameter estimated
in the RD-exercise was the percentage of farmers with a
particular characteristic, say percentage of farmers growing
coffee, then the following system was used: a margin less than
10 percentage points away from the sample percentage in the
formal survey was considered to be highly accurate, a margin of
between 10 and 20 percentage points was considered to be








moderately accurate, and a margin of greater than 20 percentage
points was considered to be of low accuracy.

The particular boundaries between the ratings are arbitrary
but broadly reflect the degree of closeness and the effects which
an error would have on the understanding of the farming system
and the defining of recommendation domains. For example, the
formal survey indicated that 85% percent of low income farmers
intercrop their maize and beans. If the RD definition exercise
estimate had been off by ten or lesa percentage points it would
have made little or no difference in the understanding of the
farming system or demarcation of recommendation domains. The
conclusion would still be that the overwhelming majority of all
low income farmers intercrop. But if the RD exercise estimate
had been off by 30 percentage points -- i~e., if it had been
estimated that 55% percent, or a substantial proportion of
farmers, do not intercrop -- researchers may have decided to
demarcate recommendation domains according to whether or not
farmers intercrop. This would have been a fundamental error,
since the number of farmers who do not intercrop is so small that
this variable should not receive high priority in defining
recommendation domains.

For parameter estimates of quantities, such as maize yields
per hectare, an estimate within 10 percent of the sample estimate
was considered to be accurate, 10 to 20 percent of moderate
accuracy, and off by more. than 20 percent, of low accuracy. For
nominal data, such as the two principal months in which farmers
experience labor shortages, a correct estimate was considered to
be highly accurate. A partially correct estimate, say estimating
March and April to be the principal months of cash shortage when
the formal survey indicates that April and May are the principal
months, is considered to be moderately accurate. Estimates which
are not even partially correct are considered to be highly
inaccurate.

Table 1 presents the accuracy ratings and sources of
inaccuracy for 21 parameter estimates for each of two income
groups from the exercise to define recommendation domains and
from the formal survey. The table shows that over one-third of
the estimates made in the exercise to define recommendation
domains were highly accurate and three-quarters were of high or
moderate accuracy for both income groups. The data are derived
from data appearing in the Appendix, which presents the 21
parameter estimates made following the exercise to define
recommendation domains and data for the same parameters from the\
foralsurvey results.

Surprisingly, the levels of accuracy are nearly identical
for each income group. Extension workers tend to have more
contacts with "progressive" and high income farmers; thus it had
seemed reasonable to hypothesize that the accuracy ratings in the
RD-definition exercise would be higher for high income
farmers than for low income farmers. Indeed, extension agents,
ierespondent-based sources, were responsible for twice as
















High Income Low Income
Farmers Farmers


Number Percent Number Percent
of Estimates of Estimates


Level of Accurac a

High

Moderate

Low

Total No. of Estimates


Sources of Inaccurac b

Respondent-Based

Lack of knowledge

Progressive Farmer Bias


aThis table is derived from Appendix 2. Accuracy means
degree of correspondence with results from formal survey.
Accuracy categories are defined as follows. Three different
kinds of parameters were estimated:


TABLE 1


Accuracy of Estimates from Exercise to Define
Recommendation Domains, Middle Kirinyaga, 1981


100


100


Researcher-Based


Phrasing

Compensating for perceived
progressive farmer bias

Composition of Recommenda-
tion Domains

Total of inaccurate
estimates :


100C


100C








1. Percent es o th roup wt a articular
characteristic: Here, high means estimate was within +
10% of formal survey result, moderate = within + 20%
and low means estimate was beyond this range.

(Notes to Table 1 continued)

2. Numerical parameters: high means estimates was within
+ 10% of formal survey result, moderate within + 20%
and low outside this range.

3. Non-numerical parameters: high = estimate is the same
as formal survey estimates, moderate = estimate
overlaps with formal survey estimate, low = estimate is
completely different.


bSources of inaccuracy are discussed in detail in the
text. Respondent-based sources originate with the respondents,
who are extension agents and sub-chiefs. These sources or
inaccuracy include:

1. Progressive farmer bias: biasing estimates towards
progressive farmers.

2. Lack of knowledge: lacks knowledge about the
information requested.

R'esearcher-based sources are caused by the researcher and
include:

1. Phrasing: poor phrasing of questions, i~e., phrasing
so as to exclude some possible responses.

2. Composition of RD's: mistaken about composition of
RD's.

3. Compensation for perceived respondent bias.


CSources do not sum to 100, since more than one type of
bias may be encountered for a particular estimate.








many errors concerning low income farmers as for high income
farmers. This result is as expected; extension workers showed a
greater lack of knowledge and more progressive farmer bias in
responses concerning low income farmers than for high income
farmers. However, researcher-based errors were more numerous for
high income farmers than for low income farmers because of the
researcher's incorrect assessment of the composition of the high
income RD. Thus, whereas respondent-based biases were the
principal factor lowering the quality of information about low
income farmers, investigator-based biases were the principal
source of inaccuracy concerning high income farmers. The net
results were similar levels of accuracy in information
concerning both groups.

Two other, less important, researcher-based biases also
distorted information about farmers. First, inappropriate
phrasing and phrasing questions in a way to exclude certain
appropriate responses were more of a problem in asking about low
income farmers than in asking about high income farmers. This
indicates, not surprisingly, that the researcher knew less about
low income farmers' circumstances and practices than those of
high income farmers.

Second, modifying estimates to compensate for perceived
respondent bias sometimes decreased the accuracy of the estimate.
Five estimates concerning high income farmers and seven
concerning low income farmers were revised for such reasons. In
six cases the modification improved the accuracy rating; in four
cases the modification lowered the accuracy rating; and in two
cases there was no change. Two examples typify the advantages
and disadvantages of this measure. First, extension agents
reported that about half of the low income farmers purchased
hybrid seed each year. However, a seed retailer assured the
researcher that low income farmers rarely, if ever, bought hybrid
seed. In the RD report the estimate was lowered, and formal
survey results confirmed that the modification was correct. A
second example shows how one can easily make mistakes using the
compensatory bias tool. Extension agents estimated that one-half
of high income farmers and one-third of low income farmers grew
coffee. These figures were revised downward, particularly for
low income farmers, because of a perceived progressive-farmer
bias on the part of the extension agents. In fact, the extension
agents' estimates were accurate, as the formal survey confirmed.

A further issue concerns the reasons underlying respondent
inaccuracies. Estimates concerning crops, maize and bean husbandry,
livestock, labor uses, and income sources were generally fair to
good; poor estimates were made on manure use, maize varieties,
methods of land preparation, maize yields, and percentage of hligh
income farmers intercropping. In several instances, it is
possible to explain why some estimates were good and others were
poor. For example, estimates on manure use were poor because of
respondents' lack of knowledge; manure use is not part of the
extension agents' recommendations to farmers and thus, they were
unaware of farmers' manure practices. Progressive farmer bias









and lack of knowledge were important sources of bias for the
other four variables which were incorrectly estimated, and
improper phrasing of questions by the investigator contributed to
inaccuracy in two of the cases.

Accuracy of RD Identification

As discussed earlier, two sets of factors, across-area
differences and within-area differences, were examined in the
process of identifying RD's. Map 1 compares the across-area
boundaries of RD's in Middle Kirinyaga based on the RD
identification exercise with the delineation based on the
formal survey. There are no important modifications of the
preliminary set of boundaries, except that a part of Rukanga Sub-
location was excluded. There were some apparent differences
between Rukanga and the rest of Middle Kirinyaga from the RD
definition exercise but it was thought that these differences
were not important. However, Rukanga was visited during the
informal survey, it was apparent that rainfall was lower and farm
size and livestock numbers were greater than in Middle Kirinyaga.
Thus the are was excluded. As stated in the'previous section,
the boundaries of Middle Kirinyaga represent fairly distinct
changes in eco-climatic and socio-economic environment. Thus, it
is not surprising that the initial attempt to demarcate
boundaries proved to be very accurate.

The attempt to identify within-area differences proved to be
more complicated and somewhat less accurate. Preliminary
information from key informants led to a focus on the differences
between low income residents who had received land in the late
1950's under land adjudication, and high income in-migrants who
purchased land from residents. In-migrants, it was thought, were
high income, progressive farmers who generally had off-farm jobs
and very different farming practices from resident, low income
farmers. The in-migrants were thought to make up the vast
majority of high income farmers. During the RD definition
exercise, this finding was refuted; it was found that most high
income farmers were residents who had acquired land through
adjudication as well. Thus, the report on identifying RD':s in
Middle Kirinyaga distinguished between two RD's: full-time
farmers with limited capital resources, and part-time farmers
with off-farm income. The principal feature distinguishing these
two groups is the amount of cash they have available for
investing in their farms.

Several refinements were added to this distinction following
the informal survey. First, the correspondence between full-time
farming and low-income was dropped, since some full-time farmers
had lucrative operations such as dairy or tobacco and were thus
high income farmers. Second, the correspondence between part-
time farming and high income was omitted because many low income
farmers were found to be farming only part-time; their off-farm
activities were not lucrative enough to make them high income.
However, the principal distinction made in the RD identification
report, that of access to cash resources, was maintained








throughout each ensuing stage of analysis.


In summary, the exercise identifying RD's was reasonably
accurate in defining and demarcating RD's and in providing
preliminary information about the farmers in each RD. This
result is somewhat astonishing, since so little time and
resources were allocated to this exercise. The primary source of
inaccuracies concerning high income farmers were researcher-based
biases, whereas respondent-based biases were the most significant
source of inaccuracy concerning low income farmers.

By examining the sources of inaccuracy in the exercise, it
is likely that the effects of these sources can be minimized in
future investigations identifying RD's. In addition, three
important lessons arise from our exercise.

First, interviewing local leaders, e.g., sub-chiefs, in
addition to, or in place of, agricultural extension agents, may be
useful in obtaining more valid information about low income
farmers. In the exercise to identify recommendation domains in
Middle Kirinyaga, respondent-based biases were generally caused
by extension agents' lack of orientation towards low income
farmers. In several cases, sub-chiefs were able to provide
better information about their sub-location because (1) they were
natives of the area and, (2) they had less of a stake in over-
reporting the use of recommended inputs than did extension agents.
However, their education level was generally lower than those of
extension agents. Thus, many had difficulties with such
concepts as estimating the proportion of farmers having a
particular characteristic.

Second, more effort should have been made to identify the
possible responses to questions in the questionnaire,
particularly concerning low income farmers. Because of the
biases towards progressive, high income farmers, there were more
phrasing errors concerning low income farmers than high income
farmers.

Finally, the exercise highlights the importance of correctly
defining within-area differences between RD's before administering
questionnaires. In our case, a more thorough investigation of
the composition of the high income RD should have been carried
out before administering the questionnaires. If the researcher
is uncertain about the within-area differences, the data collected m
be of little use. In Middle Kirinyaga, more informal interviews
with persons familiar with the area could have contributed to a
more accurate assessment of the within area differences in Middle
Kirinyaga.

Summary and Conclusions

During the initial stages of a farming systems diagnostic
survey, most researchers use secondary data and informal
interviews with farmers and key informants to develop a
preliminary understanding of farmers' circumstances and to


'10









tentatively define recommendation domains. However, in the
research reported in this paper, an exercise to define
recommendation domains is conducted at the onset of the research
process and involves a brief questionnaire administered to
extension agents. The exercise, which took approximately three
weeks, was found to be reasonably effective for the following:

1. Tentatively classifying farmers into recommendation
domains. Few revisions were made in the domain boundaries
as a result of the informal and formal surveys which followed the
RD-definition exercise.

2. Developing some preliminary information about farmers in
the respective recommendation domains. About-three-quarters of
the estimates of parameters in the RD definition exercise were
moderately or highly accurate.

In addition, the exercise served three other purposes.
First, by defining recommendation domains in the study area,
researchers were able to decide which RD's they wanted to target
during the informal survey which followed. Thus, researchers
decided to focus their efforts on two domains, high and low
income farmers, in a relatively homogenous area of Kirinyaga
District. Second, by delimiting boundaries between the defined
domains, little time was wasted interviewing farmers who were not
in the targeted domains. Third, by including virtually all
extension agents and local leaders in the information gathering
process from the onset of the research process, the researchers
were able to gain their confidence and active participation in
on-farm research activities. For example, the extension agents
and sub-chiefs proved to be very useful in assisting researchers
in organizing subsequent surveys and in conducting on-farm
experiments.

The accuracy of parameter estimates in the RD-definition
exercise was fairly high; this supports the hypothesis that the
RD-definition exercise described in this paper can be useful for
tentatively defining recommendation domains over large areas
before beginning farmer surveys. Thus, in countries where
secondary data and survey results are limited, an RD-definition
exercise can be useful for assisting policy makers to channel
research resources more efficiently. For example, a research
center with scarce resources can systematically decide which
domains to focus their efforts on, e~g., those with the greatest
numbers of farmers, those where incomes are lowest, etc.

'An analysis of the results of the exercise also provides
guidance on possible improvements which could be made in future
efforts. The level of accuracy of parameter estimates in the RD-
definition exercise was approximately the same for both domains
under study in Middle Kirinyaga. But interestingly, the sources
of inaccuracy were different for each domain. The errors
concerning low income farmers were respondent-based; respondents
tended to bias their responses towards "progressive", high income
farmers. In other cases, they simply did not know the answers to









certain questions about low income farmers in their area. In
contrast, the sources of inaccuracy concerning high income
farmers tended to be researcher-based. For example, the
researcher incorrectly thought that most high income farmers were
in-migrants; thus questions concerning in-migrants led to
responses which did not reflect the practices and circumstances
of most high income farmers.

CITATIONS

CIMMYT Eastern Africa Economics Programme, 1979. "Deriving
Recommendation Domains for Central Province, Zambia."
Demonstrations of an Interdisciplinary Approach to Planning
Adaptive Agricultural Research Programmes, Nairobi.

Collinson, M., 1982. "Farming Systems Research in Ea~stern
Africa: The Experience of CIMMYT and Some National Agricultural
Research Services, 1976-81." Michigan State University
International Development Paper No. 3, East Lansing, Michigan.

Flora, C., 1984. "Grouping Farmers and Their Problems" from
"Diagnosis in Farming Systems Research and Extension Training
Manual", Farming Systems Support Project, University of Florida,
Gainesville, Florida.

Franzel, S., 1981a. "Farmer Target Groups in Middle Kirinyaga,
Kenya." CIMMYT Eastern Africa Economics Program, Nairobi.

Franzel, S., 1981b. "Identifying Farmer Target Groups in an
Area: Methodology and Procedures," _Farming Systems Newsletter,
No.. 4, CIMMYT Eastern Africa Economics Program, Nairobi.

Franzel, S., 1983. "Planning an Adaptive Production Research
Program for Small Farmers: A Casse Study of Farming .Systems
Research in -Kirinyaga District, Kenya." Unpublished Ph.;D.
dissertation, Department of Agricultural Economics, Michigan
State University, East Lansing, Michigan.

Harrington,. L, and R. Tripp, 1984. "Recommendation Domains: A
Framework for On-Farm Research". CIMMYT Economics Program
Working Paper 02/84, Mexico.

Hildebrand, P. and F. Poey, 1985. On-Farm Agronomic Trials in
Farming.Systems Research and Extension. Lynne Rienner Publishers
Inc.; Boulder, Colorado.

Leonard, D.K., 1977. Reaching the Peasant Farmer: Organization
Theory and Practice in Kenya. University of Chicago Press,
Chicago.

National Agricultural Research Station, Kenya, 1979. "Field day
Handout". Kitale, Kenya.

USAID, 1983. "Rwanda Farming Systems Improvement Project",
Project Paper, Washington, DC.








Respondent


TOPOGRAPHIY .



SOILS.



RAINFAUG .



SETTLEMENT AND FA\RM SIZE.

% NEWJ SETTLERS AND WHERE.



SETTLEMENT PATTERN.



FARM SIZE.



AREA CULTIVATED LONG RAINS.



AREA CULTIVATED SIOR~T RAINS.



LAND AVAILABLE FOR RENT OR PURCHASE.



CROPS.

FOOD CROPS (1, 2, 3)


Cnst i CrOs (1, 2,3) AND, a GRowING.




FERTILIZER OR MA\NURC~ USE (3, AND) CR~OPS) ___


MEsTHOD OF LAND PREPARA'~n TION A\ND 3 C.AC//..


____


__~_


Appendix 1:Recommendation Domain .Questionnaire


locationn

PHYSICAL.






BUSIEST MONTHS.



HIIRED LADOR (OPERAUTION/CROP) . .


COTTON POTENTIAL;. ''



MAIZE .

VARIETIES USED!%)IN LONG RAINS. .


VARIETIES USED% 'IN SHORT RAINS.



AVERAGE MIAIZE YIELD, GOOD YEAn (L.R. S.R.)


PURE OR INTERCROPPED.



CATTLE.

% WITH GRADE ( No.)



% WITH ZEDU ( No.)


% MAYN AND WIFE ON FAR(M.

3 MAnJOR ASHt SOUR~Ci;S.*




ACCESS OTHERS.


INCOME.

% WITH1 MADA\TT ROOPS.

% GETTING INCOME FROM1 OUTSIDE.

% HAVING REGULAR JODS.


















High Income Farmers Low Income Farmers

Exercise To Exercise To
Define Recarmenda- Formal Define Recoranenda- Formal
tion Domains Survey Report tion Domains Survey Report


APPENDIX 2


COMPARISON OF ESTIMATES OF PARAMETERS FROM EXERCISE TO DEFINE
RECOMMENDATION DOMAINS AND FROM FORMAL SURVEY
MIDDLE KIRINYAGA, 1981


Land


Farm Size (ba.)

Area Cultivated
Long Rains (ha.)

Farmers Who Have
Purchased Their
Land


2.5


1.5


3.0


2.5


2.4


1.4


2.2


40%


23%


crops


Food Crops


Mai ze-Beans


Maize-Beans


Ma ize-eans


Maize-Bean

1) Beans
2) Maize


1) Beans
2) Maize


1) Maize
2) Beans



88%


Cash Crops


Percent of Farmers
Intercropping
Mai ze/Beans

Percent of Farmers
Growing Coffee

Percent of Farmers
Growing Sunflower

Percent of Farmers
Growing Cotton

Percent of Culti-
vated Area Under
Maize

Percent of Cultivated
Area Under Beans


1) Maize
2) Beans


90%


51%


8 5%


25%


20%


5%


5%



90%


91%


78%


80%


80%


80%


71% 90%













High Incomne Farmers Low Income Farmers

Exercise To Exercise To
Define Recommenda- Formal Define Recommenda- Formal
tion Domains Survey Report tion Domains Survey Report


APPENDIX 2 Continued


Land_ Preparation/
Plantn

Method of Land
Preparation



Percent of Farmers
Not Preparing Soil
Before Planting


Maize/Bean Husbandry


Hire Ox 60%;
Own Ox 20%;
Rent Tractor 20%;


Own Ox 43%;
Hire Ox 41%;
Access to Ox 3%;


Hire Ox 50%;
Own Ox 50%
Tractor < 5%


Borrow Ox 33%
Own Ox 27%;
Hire Ox 27%
No-Till 21%


Long Rain Maize
Varieties


Hybrids 67%
Katunani 33%
2nd Gen, H~yb. 10%


Local 59%
Katumnani 49%
Hybrid 36%
2nd Gen. Hyb.

Katumnani 51%
Local 51%
2nd Gen. Hyb.
Hybrid 26%


Local 55%
2nd Gen. Hybrid 30%
Hybrid 5%
Katmnani 10%

Local 50%
Katunani 30%
2nd Gen. Hyb. 30%
Hybrid 0%


Local 77%
Katumani 69%
2nd Gen.Hyb.2
Hybrid 8%

Local 71%
Katumani 60%
2nd Gen.Hyb.2
Hybrid 6%


Short Rains Maize
Var ieties


Katumani 90%
Hybtrid 10%


Percent of Farmers
Using Fertilizer

Percent of Farmers
Using Manure

Average Maize Yield
(kg./Ha. in Good
Season)


iLivestock

Percent of Farmers
With Zebu Cattle

Percent of Farmers
With Exotic-breed
Cattle


20%


18%


82%


< 5%


10%



1,900


2%


49%



1,100


2,500


1,200


69%


71%


56%


75%


56% 15%













High Income Farmers Low Income Farmers

Exercise To Exercise To
Define Recommenda- Formal Define Recommenda- Formal
tion Doa~ins Survey Report tion Domains Survey Report


Source: Survey data, as reported in Franzel, 1983. Date from exercise to define
Irecommendation domains are reported in Franzel, 1981a.


APPENJDIX 2 -- Continued


Percent of Farmers
With Cattle


INCOME

Percent Getting
Income From
Relatives

Percent Having
Regular Jobs
(Salary/Business)


95%


92%


80%


62%


10%


11%


75%


69%


20%


21%


Busiest Months

Activity at
Busiest Time


April-May


April-May


Weed ing


March-May

Plant ing/
Weeding


March-May

Planting/
Weed ing/


Weed ing


90%


10%


17%


97%


72%


100%


50%


Percent Using Hired
Labor (Not in-
cluding for Plows)


Other

Percent With Zinc
Roofs

Principal Distin-
guishing Factor
Between IRD's

Percent of Farmers
in this RD


Part-time farming/
Off Farm Income


Full-time fearing/
No off-farm income


High-Income


Low-Income


80%


60%