anal nun-I u

----

 

 

 

 

 

   

     
 
  

     
  
 
  
   
  
       

~ -____- fi \ — ‘ I AA A 'c - ‘ ¥
. 'T—W - _ “w ‘.‘u , “ulc— _ —--o.
01-. ' u L m-—..~W" 'Qw'flw Wr‘i‘. ' "vs fi.."““" "" A f '0. . - - \. ‘ . . .
.J 0 .q." . "-0-... -- ,‘ ... _' . .- -.. .. . .- ,... . _ -. ' ._ - . __
,. ._—,—.-N....__~ _—:I‘-I-_“-I~. _ _ . . ..,__ - _‘ ~-... . , _ -- .. . -. Q 5 .l . _ O . -' = - .V' . - . _ . .
.‘. ‘ . ' - . o... ' ' 'I o ‘ . '-."‘ ’ — ‘ ' " . - A . .. '3 . -.. _ " _‘ , . ' ‘
g . - t _ ' — ' ‘ 9.. o I ‘ -- ‘ . . . ' " 0 .- - .. ‘ '
_ ‘ - a . . - Oo- ' - .- o u . o ' - ‘3'“ . . o 1 - . .. ..
; . ... . ' . . - . ' - '. - - ." . - ... . ‘ . _ ' -'---
-p~ o, " - .' . -‘ - ~ - 0 ' . ' - Y Q ' - 'o v o 7 ‘ u _
r — ._ - _ ' - . - _ . . -..
1 . ' - o .- . 0 . -‘ - --. - .
-g _ . - - .- . - .
‘ . .‘ ‘ . U u u 0 -
. . .. ..
. u - O - ‘ ‘ ‘ o .
o ' h . _ - . . . ~ - ' ’
c ’ ‘ -- ‘. ' . - ‘- ' ' . ' .- . ‘ ‘ .- . ‘ ' 1‘
9 n » " 1‘ l 4. ’ - - Q . ' ‘
. ‘ a - ’ .. ’ . ’ ' ' . o . '-- ‘ I ~ '
. .. .
.7 ‘ .. .__ v - . . . .
- . - . _ . ‘ - . ‘ . v _ _
. o . ‘
‘ - . _ -.. . n _
‘ .o. . . . . . .‘ . . “ a v -
O I. - ' C .
o I .
1 . . fi 0- -¢ _
- ‘° '. 0 . ‘. - - F - .
. l _ . . . . -
' ' . ‘ . - n -.. ' n
. . . .
- ' 0 ‘ , ' ' ‘ -
‘ . _ -‘ . ‘ ' . . ' o.
I Q . '-
_ . ' ' » a . , . ‘ - '
- . . - . . - . ' ’3 '
a .. - - - . .
. . a. . a , . . . . _
we . ‘ _ . 4-. . . . 0- 1 .. -
. . . V ' ' I - .' Av
. 4 _ '.o . . . . . . .
- o ‘ . ' ' - ' ' '
o ‘ . ~ . - ' 7
A ,. I4- - ‘ g
o . I n c ' a ' . ‘ .
.I . '~ . J ‘ . n s _
‘ ‘ o . . o ‘ ‘. .
. ' . ‘I ’ . . o. .
. . o .
_' o -., ' u. -. . . ".
‘ ‘ o .- ‘_ '- ‘ u o' ‘.‘
_ -~ ' o ‘ A ° .. ‘ . 0 _ ' '
6 .I O . - - n . . ‘
. .. . _
. “ v . D . . - g . ‘
‘ . - - .‘ . ‘ 9' p ' . . a .
-. _. - . - .- . _ . ' s . ~
- . - .
. g ' ‘ u . .
' o ' ‘ v n O . . . . .
v _ _ . . .. . _ . . .
. . 0... ~ , . I. I - . . I i
- o . ~ ' .
' o - \ a ' c _ - ' . . ‘ n .
a u -I . . . u . .. .. -
. . ‘. _ . .
4: . '7 . ‘ - I
‘ ' '-. ' §
- . . u
. _ . A . . . . . .. .
O I I I . ‘
.. o - - I . 3
I - - O u I
. o
- o i ' .- ' ' ‘ ' .. ' | . '
. .
. 'I . _ . - - .
g ‘ . ' - -
- ‘, _ ‘ _ - . . ' .
- ' v . - ' . u -
O _ Q -
° 0 ‘ n ' - . . . . ' 5 ' .
.. - . . . . _ . . .
, ' I ' I 'a' - ,i
‘ o I ‘ ' n a . ' . . ‘ .
a o . . I . .. . . .
g I . ‘ - ' . ” . a' o'
- ’ . - - . . -‘
o . O ' .. 0| o _ - .' ‘3 - .
. ' t ‘ . ' . - . . ."" I. . 2‘ I -
. I . . . ' _ . . .
- 0.. _ . q» - ' I .5 . A .‘
_ . . _ ‘. . . ' '- . 4 o. ‘ o.
l ' . . _ - - - " »- _ _ .
, . . - . - . .. . . .o I- —~ . . . ~ . .,
- , ‘ - - ~ . . - .
. , . . . . . x A . . - . ‘-~. A - .
. ' . ' ' ~ - ' - - v' 2' ‘ .' ‘ ' ‘ ° -- - :"""'
" _: 0 . - , ‘ ’ ‘- . . .I .-.o'.l.‘_‘._.'. I
' ~ . n _ . p . .- ' ‘ " I_. ~ I . \- ‘ — .. . -0.... .. ‘
. ' . .,. . .. . .
o
I
,
p
o
.
- .
. .5
- g.
C
. O
.
v
.
.
;
_ 0'
- u
. s
u
.
E:
-
u

.
o
o -
'0
. s
n .-
2 ‘I‘
. 9:-
c
x.
O - a
V n

0 -.
I
I-
- -:.'
'D’
t- "
PI' up
n y'

L“ 6‘

 

v
0

c .:
-: q

' ‘
- .‘L
I

 

 

{‘G-

G

c

‘

5.1!.

   
 

 

-
L I
’ v'
' "
, . .
. .
. f
,_V . .
c .
0', ,
-. . .
0‘ I
9
-.. ,
n ,V,
1‘ u‘.
. .5
. I
-
O .‘s
‘u. ’
C .
’\ D
. __.¥
" oa~
. ..
s
f . -
.
' V
‘ a~
\ - uv'
it ~\
I» v
. ‘-
1' .‘I‘
d‘ .
‘5 3-"
.. '
w. . .c'
a 0 %
up 0.
-o as
Us ..
a Q'
~ ' .5
fi . .
Q:- . . "v a. 0.0
.- fv Fl-v‘t.

        
   
    

  

o" I
D.-

    

“nan
m6

  

—‘< c
.

  

- 3‘ .
J.- .-r
:b‘ 1.1%.

    
  
 
 
  

 

nm‘
a». ”1.13“:-

1..' .0
vs x 10"" :.‘

 

    
   
    
  

    
 
  
  
    
 
 
  

 
 

‘ i
‘ ‘ 7-... . pp. .»
‘ “’- ‘z u.“

. . J

    
 

4- -.,...V.
_ . 1
J- .._.
. r-‘Iv pI.Io

"he or

 

a: "

'1

           

     
  

             

-"‘. G
~‘.':O.. . . _.. .4.) 4......” ‘
I_- ‘r-. o. ‘ 0 V"- w-Ui-o .7._.

' h-h-‘.-‘ I. 0 .‘v
.- .,.,__. w ..

      

  
  

 

  
  
  
 
 
 

O '1“.
v--» «c .
.L. - Macaw:
I v..- r g . th .
"N A

  
 
  

       
     
 
  
  

, ' on v
' o A. 0.2-. '.“"‘~
l “'0‘: my. _‘ “"~'0
4 Q‘ 0‘ I» , I ' ,_
1". ”.I Dr "bdrffg. . '0- .u
to- R;(”“' ' " """"‘-.C'vo\"ul:a$m' ' U"""‘
- M,..-, ,.‘, .'::.". """ 0r 0.1-.

’ .. , 'M' W?"

        
   
 
  
 
    

t '0'
.u'

 

‘.
"' ‘r-rrac‘v ‘

v‘au4 - ‘
O'_o.‘- .‘

o"
n

 

A...

I
a:
X

   
 

    
        

_ . '.‘ 0“ J «u’;
*‘V‘o - .9 .1 . .

W I .I, . a c .‘

32.3 o“. r In". 0-2“ "m. o a.

(do: n ..',!’I. “a” An.

' 1‘ u- o. l‘ r‘ ':0.r ‘

      
 
  

t
"O

    

  
  
 
 

   

-o .1...
1 3’.

      

     

b - .

n. :gv-z -

~~p...I,. I“
o.-

   
 

33:731.}: : 7::

 
   

11--

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIRC/DatoDuepSS-p. 1 5

 

 

vV/

ll

///§QTA COLLECTION SYSTEMS AND METHODS IN ON-FARn/FARM{NG SYSTEMS
RESEARCH: A CRlTlCAL Revlen;//

Derek Byerlee and Robert Tripp**

October 198l

* Incomplete first draft for comments. Little attempt has been made in
' this draft to integrate contributions from the literature,

** Economics Program. International Maize and Wheat improvement Center

(CIHMYT), Mexico.

Thesis
ID‘i
OLN
THS

(iv

introduction

That: or CONTENTS

vvv—w

A Framework for Evaluating Data Collection Systems and Methods in FSR

- The Economics of lnformation_

- FSR as a Data Collection System

- Characteristics of Data Collection Methods

-Degree of Observation

Degree of Participation of Researchers
Frequency of Data Collection

Use of Written Recording of Information
Degree of Structure and Specificity

Degree of Quantification

Degree of Representativeness

Characteristics of the Situation in which FSR is to be implemented

- Characteristics of the Study Area

- Characteristics of the Research Project

ClMMYT Experiences in Designing Data Collection Systems

- Evolution of the CIMMYT Methodology
- Production Research Program in Ecuador

- Ghanaian Grains Development Project

- Hheat in the Yaqui Valley, Mexico

Conclusions

JI

'iNTRODUCTION

On-farm or Farming Systems Research11 has been increasingly promoted
in recent years as a means of increasing the effectiveness of agricultural
research systems and delivering appropriate technologies to farmers. Most
FSR programs consist of at least two basic components. The first is a
diagnostic stage in which farm surveys and other methods are used to identify
flsv we...

constraints in the system and choose possible solutions to those constraints

that would be feasible to farmers. The second is an experimentation stage
. ,,..-—e _

f

in which possible solutions to these problems are’tested under farmers'
conditions. The ultimate end product of this process is information usually
in the form of a recommendation to farmers on an improvement or improvements
in the technology for a particular enterpriseCSI in the system. To be effect-
ive, this information should be quite specific and deal with only a limited
number of priority changes in farming methods, since farmers because of risk,
capital shortage and learning by doing, will only make gradual changes in
their farming methods.

-There are also other users of information generated by FSR. Hany FSR .
programs have identified serious limitations in technologies being generated
on experiment stations and this information is fed back to experimentation
station researchers to help reorient their research more closely to farmers'
problems. Also farming systems researchers are increasingly directing their
efforts toward providing information to policy makers on'problems or inconsist-

encies in the policy environment in which farmers operate. In both these

 

lIWe use the terms On-Farm Research (OFR) and Farming Systems Research (FSR)

interchangeably in this paper since they basically have the same objectives
and use the same methodologies (see Harrington).

f].

case the Information provided is again qUIte specific « o.g. information

to plant breeders on the characteristics of a desired variety or information
to policy makers on the pay-offs to making a specific input more readily
available.

in each of the above cases, FSR is essentially a data collection
system to generate information for quite specific objectives. Nonetheless,
data collection methods for FSR have until recently been derived from
traditional farm management research which has usually had quite different
objectives. Following its origins in developed countries, farm management
research originally focused on collection and analysis of input-output

type data to improve resource allocation of individual farmers. More recently

with recognition that this is not a cost-effective approach in developing

'countries (Collinson), farm management research has broadened its objectives

to providing information to improve the micro-level basis for policy decisions
(Taylor). However, as practiced, the emphasis is still placed on generating
input-output type information at one point in time to have available to analyse

policy questions as they arise. Data collection methods in farm management

.research reflect this emphasis on general input-output information. Some .

modifications have been made to fit the situation in developing countries,
usually by more reliable (and more costly) methods, such as frequent visits
or field measurement, of obtaining input-output data (Norman, spencer).

The application of data collection methods of farm management research
to FSR has at least two important limitations. First, farm management research
usually generates an information "bank” at one point in time, or at regular
intervals to be called on for general use. Secondly, data collection is usually .

usually confined to input-output information and attempts to understand the

’ ‘9
l is:
I ;"

systems are made 35.22§£3_with the aid of farm models constructed with this
input-output data.

More recently. economists involved in FSR have experimented with alter-
native data collection methods not normally employed in farm management res
search. One method that has been adopted is the use of unstructured informal
interviews conducted by the researchers themSelves. This method was initially
adopted in efforts to reduce costs of following traditional survey methods
- particularly the costs of training interviewers, interviewing subStantial
numbers of farmers and analysing data (Collinson, Hildebrand). However,
other advantages of this method are being increasingly appreciated, particularly
the ability for researchers to obtain at first hand a rapid understanding of
farming systems.

Survey techniques nouibeing used in FSR programs cover this wide range
of approaches from informal interviews to multiple visits or cost route surveys
- sometimes with direct field or yield measurement. The role of experimenta-
tion in the data collection system and the methods used also vary widely.-

Some of this variation reflects differences among FSR programs in the objectives,
research resources, time horizon and the farming systems under study. But a. .
large part of the variation, we believe, arises from diffEIEEEflEQDESPF‘QQEs-

of an efficient data collection system to achieve the specific objectives of

.___—____—— ,..—---‘-— w - .. _______.'

 

_ __ ._ a. _ .-
‘d-u-och-s - —..—q I. -—..--.- -—

FSR.

The purpose of this paper is to develop a conceptual framework for
designing efficient data collection systems in FSR and to illustrate the
design of data collection systems in specific situations. The framework draws

on the theory of the economics of information to establish some principles for

evaluating the efficienty of a particular system. Within this framework, a
data collection system or process is a combination of various methods or
techniques of getting data each of which are used at various points in

the process,usually in a sequential manner. Each of the data collection
methods is defined by various characteristics which“ identify and
discussed. However, the design of a data collection method must be specific
to the particulan research situation at hand. ways in which the research
situation affect the choice of data collection'systems are briefly listed.
The following section then presents the general approach advocated by CIMMYT
for efficiently generating information. Examples are given of how the;
approach was modified to fit specific situations where we have worked.
Finally, we make some general recommendations for improving data collection

systems and methods in FSR.

A Framework for Evaluating Data Collection Systems and Methods in FSR

The Economics of information

The theory of the economics of information while difficult to apply
formally to the design of data collection systems in FSR, does provide
principles which are useful in making decisions on alternative systems and
methods. First, information is generated in FSR because it has value to
the user - i.e. the farmer, the experiment station researcher- or policy
maker. In many cases it may be possible to put an economic value on this
information. For example in an area where farmers are regularly using both
nitrogenous and phosphorous fertilizer, a FSR project might provide informa«
tion that phosphorous need only be applied on a certain soil type and within
a specific rotation. Thenqif this information is transferred effectively to
farmers who follow the recommendation, the value of the information is the
money saved from reducing phosphorous applications. Likewise, if information
is made available from FSR that emphasis in maize breeding should be changed
to earliness rather than yield since farmers with an early variety could
plant a second crop, then we can value the information in terms of the
research resources programmed for expenditure on maize varieties which in all
likelihood would not have been accepted by farmers.

There are a number of principles which hold in placing a value on in-
formation ln FSR. First, there is usually a time lag between expenditure of
resources for generating information and the impact of the information on
decisions. Part of this time lag is due to the time needed to generate

information. For example, in a dry area, information on the effect of

climatic variability on fertilizer response may require three or more years of
experimentation. There is also a time lag when information must diffuse to

a large number of decision makers as in the case of technological re-
commendations to farmers. The value of information from FSR must therefore

be discounted by these time lags. Efficient information generation procedures
will however recognize this discount factor and will place more value on
information that allows farmers to increase their incomes by l0 percent
annually beginning now than information that will allow income increases of

50 percent in five years time. - ,

Second, information clearly has higher value for some variables than
others. Given costs of generating information, the data collection system
should be able to efficiently identify those variables about which information
has most value, in order to focus research resources to obtain more valid
estimates of those variables. Many times, the approach is adopted that if
it doesn't cost much to collect additional information, then why not do it,
The problem with this approach is that it often ignores the opportunity cost
in terms of time and effort in analysing data on a large number of variables
all of which are implicitly treated as having equal value. Also information.
is usually transferred to farmers more effectively if it focuses on only a
few'variables.

Third, increasing amounts of information on a specific variable usually
lead to diminishing returns. That is, increasing confidence in the value of
a specific variable such as the performance of a new wheat variety relative

to the farmers' variety may be of decreasing value to decision makers.

- However, information with a higher degree of quantification and confidence

etr'i

w

intervals is sometimes easier tg_t:ansfer £9 gggisignemakers, For example;
by informally talking to farmers in an area, we may believe quite confidently
that farmers working with the official credit bank are receiving inputs late
and delaying planting even though we don't have a specific quantitative
estimate of this problem. However to communicate this information to a
relevant bank official we may be more effective if we can say that the
planting date of farmers who work with the bank is on average l0 days later
than other farmers and that there is only a 6 percent chance that this
difference is due to chance alone. This same problem. also arises when
FSR research is conducted as part of a University research program. Here
presentation of quantitative tests of specific hypotheses is usually required
fOr acceptance by university peer groups.

implicit in the above diacussion is that data collection also has a
cost. These consist of financial costs (including fixed investments) and
human resource cost. Most FSR projects at the national level are severely
limited in both of these aspects so that cost-effective techniques of genes
rating information become an overriding priority: Costs are associated with
both the gathering and the analysis of data, Too often, researchers do not ’
consider the cost of the data analysis a particularly in terms of time.
Again costs also increase with the quantity of information generated. For
example Figure __ shows sample size (assumed to be proportional to costs)
in relation to the size of the confidence interval on the variable estimated.

The final principle that emerges from information theory is that infor-

mation generation can be sequential with data collected in one stage of the

process being used to make decisions about thevalue of further data collection.

This principle has primarily been applied in Bayesian sampling methods where
prior information on the mean and variance of a particular variable is used
to make a decision about further sampling from a population given expected

costs and returns from the additional information. This process may be
. (’___~ ._.v-m-

an“-..

generalized to a sequential data collection system in which information from

_..-

M

one stage is used to make_declsions about which variables should be emphasized

.n——«...—-.—— —u--'- —

 

in further data coliggtiQnand how much information should be obtained for
__flflflflflflfiwwfl . _,. ., _ _,fl _ -_ .

each variable. This sequencing method is important in designing an efficient
¢_._,,,aa,h.

data collection system in FSR since the ultimate objective is to narrow down

I

from a wide array of possibilities to a few variables on which information

will be transferred to the user.

FSR as a Data Collection System
We have argued that FSR is a system for generating information for
various users, most importantly farmers. The process usually employs

several different methods such as assembly of secondary data,qunstructured

M

informal farmgfiuiulerviews, formal surveys with_questionnaires and experiments.

_.....,-*~—

-——--r~.-—.—. ——

 

The task of an FSR program is to allocate the limited financial and human .
resources among various methods over time to most efficiently arrive at
specific and useful information on a few variables to be communicated to

the the userCs).

-9.

l

yhiven limited resources, an FSR program must also decide between allocating.

i
l

! all resources to a specific study area versus spreading resources over a larger 3
i area resulting in less information being obtained for each area but with
[potentially wider impacts.

FSR may have both short run and longeterm objectives. In the short run,
providing some information quickly to farmers that allows an improvement on
their existing system will usually be a priority objective. in the long run,
the FSR program will usually consider other users of the information particularly
experiment station researchers and policy makers, FSR programs may also consider
development and training in the necessary skills of FSR in order to increase the
capacity for future work. Finally ESR programs may have to generate political
support within the research institution for continuation of their research.

This again argues for some short term results to help establish credibility.

Survey researchers in general always have the temptation to obtain as
much information as possible with the idea that it might be useful in the
future. This is particularly true in FSR programs precisely because FSR
advocates a much broader, multifactorial approach to agricultural research
that it is liable to bite off more than it can chew in the limited time and .
resources defined by a particular'project's budget. There is, as Clay points
out, danger in over-emphasizing "the system" and attempting to work on every-
thing at once. To realize that a small farmers' management of his crops,
animals and offefarm enterprises is a complicated business and that methods
for its improvement are neither straightsforward nor obvious, is a vital step

forward. But it is something else to propose that we attack on all fronts

simultaneously. in the real world, people must work in limited time frames

with well-defined objectives and budgets, and demonstrate success in as short
a time as possible. If FSR cannot accomodate itself to those requirements
then it has little chance for acceptance..

An efficient process will then be based on methods for diagnosing
the farming systems that allow researchers to identify a fig variables

that promise immediate improvements.in the system. This argue§_for a

H

sequential process in which a broad study of the system gradually narrows

_.

down_to intensive work on priority areas. It is an iterative model, in the

M

sense that information gathered at early stages is used to generate hypotheses

on key relationships which are then tested at later stages. It is iterative

-_... v-u—q

 

also in that it allows identificatiqn of new problems and new hypotheses as the
research continues. These will be basis for future data collection efforts.
Finally, an efficient process will attempt to incorporate already known
information into the data generation process. In particular, we would argue
that the “body of knowledge” available within research disciplines and from
experiment station research must be integrated. All relevant disciplines
that are available should be involved from the beginning, each offering its
own point of view, prior information and suggestions for work. Through this,
process research questions‘become identified. '
We do not want to give the impression that data collection system for

FSR is a complicated affair. it is only to point out that FSR is an attempt

to reorganize research on a permanent basis and that data collection should be

..—-
u—u—avd'"

viewed in the context,9f_a,long_term process. A wide range of data collection
-- ,..u an we --

methods may be applied. Their inclusion in the data collection system must

however contribute to the overall efficiency of the system in meeting specific

objectives.

Characteristics of Data Collection hethods_

Typically, data collection methods have been classified in terms of

4

increasing costs - from informal interviews or ”sondeos” to structured

frequent visit methods. However this classification-system is difficult to

apply in practice.because there are a continuous array of methods. For

example, some methods might combine elements of informal surveys such as

unstructured farmer interviews with elements of formal surveys such as

random samples. Also we have argued that an efficient data collection system

may consist of several methods upto and including the experimentation stage.

in

methods.

a)

b)

c)

d)

this paper we choose to classify characteristics of data collection
The characteristics that we discuss are as follows:

The degree of direct observation - that is whether a variable such
as yield is measured directly or is obtained by less direct methods
such as by asking farmers.

Degree of participation by researchers, This is measured by the.
extent to which researchers have first hand contact with farmers
and their fields.

Degree to which written recording is used in the data collection
method. This may range from no written recording all the way to
the use of a questionnaire to record questions and answers in an
interview.

Degree of structure and specificity. This reflects the extent to
which a given method elicits specific information in a particular

sequence or is more open-ended and iterative.

-12-

e) Degree of quantification. This varies depending on the extent to
which a given variable is quantified.

f) Degree of confidence. This relates to the standard concept of
the confidence intervals to be placed on estimates of variable or
tests of a hypothesis. This may be subjectively measured
depending on researchers' strength of belief in their conclusions
or objectively measured if a specific variable is quantified in

a random sample.

A given data collection method is defined by the combination of these characters.

For example, a sondeo or exploratory survey usually involves a high degree of

yparticipation by researchers and little structure and specificity and a low

degree of quantification. The degree of observation and written recording
varies widely. The degree of confidence, here measured subjectively, may also
vary. Observations in farmers' fields employ a high degree of observation but
vary substantially with reSpect to other characteristics. we turn now to a

discussion of the merits of each characteristic.

Degree of Observation. Many (but not all) of the phenomena of interest to ,
FSR are, at least in principle, subject to direct observation and measurement.
Data collection methods can be arranged in order of the degree of observation.
At the bottom of the scale are techniques which draw*upon others' observations
of farmers' practices. These would include such things as the review of
secondary data (censuses, past surveys, etc.) and visits to local officials

to get their observations about what farmers do. The next step up the scale

includes all of the various interview methods which ask a sample of farmers to

describe their own behavior. Next are methods in which the investigator him“
self or an assistant is the observer, directly observing variables such as
inputs used or types of weeds in a field. And finally, the most controlled
type of observation is the field experiment,s in which the investigator
\___flwe—fl

directly observes or measures a range of data under controlled conditions
through his management or participation in the experiments,f

In most cases, the higher the degree of observation in a survey method,
the greater the validity of the data, but the more expensive and/or time

l/

consuming the method- . Researchers must take this tradesoff into account
when considering appropriate techniques.: In some cases direct observation
is the only way of obtaining reasonable data if it is needed (the occurrence
of a plant disease which the farmer does not recognize) while in other cases
observation is virtually impossible (yearly income) or very costly (labor
time allocation). On-farm experiments, if they satisfy some of the other
characteristics such as representativeness (see later section), provide some
of the most valid types of information, but usually at high cost. Their

expense argues for careful selection of experimental variables before

beginning. Observation is also valuable when the farmer is unable to describe'

b "‘
-‘r

particular parameters. in these cases there are sometimes proxies that can

 

“M'-

be devised which increase the validity of the data without increasing the

eXpense through direct observation. Besides proxies, the farmer's own

 

l7V

--In some cases however, too much observation can affect the validity of the
measurements. This is the case when the observer's presence affects the
phenomena under consideration - “the observer effect". Following a farmer
around all day with a stopwatch may be an example of this; time allocation
data of equal validity may be obtained by a multiple visit method in which
the subject reports his activities to the investigator (Tripp).

pl“?

observational skills can be increased, either by aiding him (Showing him
pictures of common weeds as part of a survey, providing him with scales to
measure his harvest) or by visiting him more often so his memory will be
fresher (next section).

Participant observation is a technique with potential in FSR. it
has been most often used in anthropology, and in its classic form the
investigator lives in a community for a long period of time, learns the
language, participates in the daily activites of community members, and
observes matters as unobtrusively as possible, gradually building up a corpus
of data which can be analyzed and reported. This type of observation is a
luxury-that few FSR programs can afford, but certain aspects of the method
are of greater relevance. Using participation in community activities as a
way of observing things and developing people‘s confidence in the researcher
is of'value - helping to plant a trial in a farmer's field and visiting it
frequently is an excellent oppOrtunity for precisely this type of participant
observation as the researcher uses the trial visit as a pretext for casually
discussing a range of other matters with farmers and spending some time walking
through the community (Tripp). This is a way of gaining insights into many .
matters that formal interviews or observational techniques could not touchl!.

Observation of farmer's fields also has potential for wider use in FSR.

At a minimum it allows researchers a first hand appreciation of farmers' yield

limiting factors such as weeds and diseases which cannot readily be obtained

 

2-/i"articipation observation is also helpful in cases where the phonemenon in
question cannot be measured. Clay speaks of the necessity of spending time
in communities in Bangladesh to discover how water rights are managed, some-
thing that would be very difficult to ascertain through interviews. Attendance
at meetings is also sometimes valuable for getting an idea about distribution

of political power and for hearing opinions expressed and debated.

l 'J

in farmer interviews. Quantitative collection of data from farmers' fields

may also substitute for experimental observations. For example yield or a

‘weed problem may be related to specific practices used by farmers. Given

sufficient variability in those practices in the sample, this method may
enable quite valid estimates of particular biological relationships that
would normally be estimated by experimentation (see for example Pinstrup-
Andersen; Byerlee, Whitehead). in general however, the method is costly,
especially if formal measurement is employed or fields are not readily
accessible. Some less costly observation methods based On visual estimates
of variables such as yields or weed damage have been used quite successfully
(Byerlee, Harrington and Marko; Byerlee).

Observation techniques are however rarely sufficient even for the
measurement of a very specific variable or'relationship. in particular,
observations to truly reflect variability over time must be taken at several
points in time. This greatly adds to costs. Alternatively observations at
one point in time must be supplemented by farmer interviews to enable the
VGSU'tS to be placed in the perspective of the year to year variability. Fi-

nally, interpretation of observations is always necessary and farmers must be

approached for this. Reliance on pure observation leads, it has been said,
to the conclusion that "every morning the women of the village go down to the

river bank and try to break rocks with wet clothes".

igegree of Participation of Researchers. At one extreme, a survey can be
completely designed in an office, given to a group of enumerators hired for
the purpose, and the results analyzed back in the office. This level of

researcher (non) participation is sometimes encountered but we would argue

"1...,

.15..

that the questions of the researchers! active field involvement is particularly
important in FSR. In particular, we have argued that FSR, at least initially
should be open-ended and not prejudge the variables of interest. A high
degree of participation allows researchers to know at first hand the study
area, formulate hypotheses and diagnose priority problems.

Also, the researchers' expertise is (or should be) greater than that
of any intermediary that might be hired and therefore the researcher is likely
to exercise care and judgement in identifying and recording data.

On the other hand, it is usually not a wise decision to have senior

I

researchers Spending significant amounts of time administering questionnaires

to a large sample. Where specific information is being tested it may be best

for researchers to design and test the instrument and have others obtain the
data. Also, in certain situations researchers who are outsiders may have
difficulty at least initially, in obtaining information '3 a locally recruited
intermediary may be needed. This might occur where there is,a high degree of
suspicion about government officials or other strangers, where there are
language differences between researchers and farmers, or where male researchers

attempt to get information from female farmers or vice versa. .

Frequency of Data Collection. Norman has suggested that data can be classified
as single point if an event occurs over a short period of time or continuous

if it occurs over an extended period of time. Data can also be classified as
SEE:EEEEEEL-if farmers readily conceptualize it (e.g. inorganic fertilizer
inputs) or nonregistered if farmers do not normally conceptualize or measure

V
a unit (e.g. family labor input or organic manure). Multiple visit data

-17..

collection methods help to reduce the time over which farmers must recall an
event and are often employed for continous and/or unregistered data. Data
taken by observation may also be done in multiple-visits,especially
observations such as the incidence of insect problems,in a growing crop.

Observation can often substitute for multiple-visits interviews.

\ _ M. q
H ‘5‘ ____.___ ,____ ._..-—- - "' \gl‘.‘ w ' _-_l - _ '--,
‘w

Cassava yields - usually a continuous non-registered variable in West Africa

—- —__.—————.._______ .——— C -

can be measured directly in crop cuts or by multiple visits to farmers. Labor
time allocation can be estimated either through observation or interviews, at
one point in time or more often. Time allocation could be estimated (very
poorly) by a single visit to a locality in which various activities were
observed and recorded. This could be improved upon by more frequent visits,
and further still by a series of random visits to a selected sample of house“
holds (Johnson). At the extreme, observation could be continuous over a period
of days (and at various time of the year) with researchers following farmers
through their daily routines. This is obviously the method with the highest

frequency of observation and of course the most demanding in terms of research

'resources. Similarly, interview methods for estimating time allocation vary

from single interviews in which the farmer is asked to estimate time spent
during the year at various tasks (or proxies, such as asking what periods of
the year are the busiest - Collinson) to more'frequent visits in which farmers
are asked only to recall activities in the time period since the previous
visit or in a recent period of time.

Thus there are a series of considerations in deciding how frequently '

’_.

data should be collected, How readily can the farmer understand what is being

asked fori How good is he at recalling the information? Over what time period

does the phenomenon occur? Can observation be used in place of interviews?
On the other side, the investigator must ask what the costs might be in terms
of time, money, and imposition on farmers, and if they are justified by the
extra validity gained with higher frequency.

in general, we believe that there are very few variables which are
important in FSR which require the expense of multiple-visits. Many multiplee
visits surveys have been set up to measure seasonal labor inputs and bottle-
necks (e.g. Spencer and Byerlee) where single-visit methods (e.g. asking
farmers their opinion on the busiest season) would give most of the information

/

at much lower costs.

Use of Written Recording of Information. Data collection methods vary
substantially in the degree to which information is recorded in writing.
Some methods may depend entirely on oral interviews or visual observations.
information obtained In this way may or may not be recorded after the interview.
At the other extrme are written questionnaires used to administer questions
and record information dggigg an interview.

' we believe that whatever method is employed that regular recording of
information in writing is desirable in order to preserve as much informatioe
as possible for future reference either by the same researcher or by other ‘
researchers. For example, an agronomist who visits an on-farm experiment on
a regular basis learns a great deal both from visual observation of the
experiment and casual conversation with the farmer, that is useful in inter-

preting the experimental results (Tripp). This information should be recorded

regularly since the researcher himself may want a reference in future years

Ila.

when combining results across sites and years and in any event changes in
research staff in a FSR are often quite high, and written information can
easily be passed from one researcher to another.

'Emphasis on writing responses during an interview may however have
a negative influence on farmers' cooperation, especially if sensitive informa-
tion is being recorded. This can be avoided by writing down information after
the interview. With experience, researchers can even administer a short
questionnaire without using written questions or notes in the interview,
provided information is recorded immediately afterwards. However, with the

C_.__.....-—-.

development of appropriate interview skills, there is usually little difficulty

I...

in recording information in the interview for all but the most sensitive

information.

I '-
f-P"
e ,‘u—v

Degree of Structure and Specificity. These two characteristics are usually
but not necessarily associated in data collection methods. A questionnaire
is a structured method in which specific questions are asked following the
order of the questionnaire. Sondeos or exploratory surveys usually employ
unstructured interview methods with questions being chosen depending on the

farmers interest and the interviewers judgement. Unstructured methods also'

”——

allow an iterative interview method where questions are formulated depending

 

on farmers' responses or observations in the farmers' field. Questions

,s- u / """
in these interviews are often not very specific - that is, the same questions
are not asked to each farmer.

The degree of structure and specificity required in a data collection

method depends on the stage in the data gathering process. ’ddentification of 7

\. - _

'.’arbbiéms"5hagf6&w&iatian of hypotheses is best done with fairly unstructured /

—-..--u-""

ods; On the basis of this information, it is possible to use structured

-——————~_./'

methods to obtain specific information or test specific hypotheses. i
One of the principal strategies of FSR is to narrow a field of enquiry
untila‘éw hypotheses remain to be tested, through structured surveys, observa-
thpns or field trials. In many cases researchers are probably at the point
where they should be concerned "rather more with identifying the important

P.—

variables rather than measuring their movements" (Hill). As much flexibility
;; possible should be maintained at the beginning of the research process so
that these important variables can indeed be identified. This does not mean
collecting large quantities of survey data, but rather spending enough time
pursuing a variety of matters and gaining a general understanding of the
farming system, and then narrowing the field as rapidly as possible. This

argues for the importance of open-ended questions and unstructured interviewing

.—.r*
u.

situations.

Degree of Quantification. Once variables have been specified there is the
question of the degree of quantification required to describe them. Some“
variables are of course not quantifiable (land inheritance systems) and others
are quantifiable only with difficulty (degree to which an opinion is held),

but most variables are at least in principle. able to be quantified. All things

being equal, the more quantification the better, but several considerations

are important here. First is the time and expense required to achieve the

 

extra quantification and to use it in analysis. Second is the ability to

accurately sample and measure the variable. And third is the degree to which
‘6’”

the extra quantification is really needed in the analysis.

Thus the researcher should consider the range of quantification necessary
for his purposes. One way of thinking about this is to consider whether nominal,
ordinal, or interval scales are required for recording the data. in measuring
the presence of a certain weed in farmers' fields for instance, one might
choose to report simply yes/no; another possibility might be to rank the amount
of the weed present on a scale of l to 5; finally, sample areas of fields may
be chosen for an actual counting of the number of weeds per square meter.

The first method is obviously the easiest, but may give misleading conclusions z
about the importance of the particular weed.The final method is very time consuming.

Many times,results from FSR are such that statements like, ”Only l3.22 I
of fafmers use chemical fertilizer, with an average application rate of 23.7
kg of N per hectare", are in fact not actually that precise and of no more
use than the statement that, "Very few»farmers use fertilizers and those
that do apply fairly low rates of N". But more quantitative reports are often
used to secure support for further research or to convince policy makers on
certain points, and in almost all cases such quantitative statements carry.
more weight. General observations, estimates, trends, opinions’are all
valuable but sometimes less effective to report; a nice table brings the .
message across more effectively.

But the issue remains, "What are you going to use this data fori”.
Excessive emphasis on quantification may draw the researcher's attention away
from more important issues. In the example in the previous paragraph, it may

be argued that if preliminary work in the area indicates that fertilizer use

is generally low despite obvious nutrient deficiencies, it may bewmorg_gseful

fw—

.to devote time not In trying to quantify the exact amount used but rather

trying to understand who uses fertilizer and under what circumstances.

._ _..—-~-—'

,

Degree of Representativeness. An important consideration in deciding what
degree of specification and quantification should be sought during data
collection is the possibility of achieving a truly representative sample.

A variety of techniques is available for sampling in specific situations

(see for instance Kearle, Ch. ill, for a discussion of methods for drawing
probability and non-probability samples). However, degree of representative-
ness is not only a problem in formal survey techniques where a random sampling
method may be employed. Indeed one of the prInpral_problems with informal

-—.~

techniques such as sondeos in FSR is the probability of sampling bias.

-‘
q—

"h".

Some of the problems have been identified by Chambers in his work on ”de-

velopment tourism", where he points out that quick visits by researchers
to farming areas are often heavily biased in favor of farmers who are near
towns and paved roads, part of special projects, males, elites, and high.
users of services and inputs. Thus the use of non-random methods to iden-
tify farmers will be in danger of producing very misleading reports unless

researchers make a considerable effort to overcome-these biases;

Representativeness is also important in on-farm experimetnation. The
validity of the results from experimentation is potentially higher than other
data collection methods but only if efforts are made to test the experimental
variables in fields of farmers that represent the variability of farmers in
the area and under management that represents farmers' management.

Finally, an important issue in representativeness is the confidence

._.._..--

interval that is placed on measurement of specific variables. Again

quantitative estimates of his interval may be useful for communication of
information to others even though it may not change the degree of confidence
researchers themselves have on a specific variable. Quantitative estimates

of confidence intervals of course depend on the sample size in relation to

the variability in the area. Two principles are however important in choosing
sample size. First, variability is specific for a given variable and hence
different sample size may give equal confidence in the measurement of a vari-
able. Second, confidence intervals are often fixed arbitrarily (e.g. 952)-

in traditional statistical methods and this may not bear any realationship

to the economics consequences of wrong decisions based on the information

a

available. Hence confidence intervals should vary for different variables

W! .

. depending on their economic consequences. For example, the decision to
launch a full scale extension program based on information from an FSR pro-
gram may require fairly high confidence in the information.’ The decision to

include a particular variable in an experimental program may require less

confidence on the part of researchers.

Characteristics of the Situation in which FSR is to be Implemented

The design of a data or information generating process is a function

—-n—_.
‘5‘“

of the specific research situation. This research situation can be
fl 'F ‘\ A

fi‘“
‘_——_--._._-.— ~

 

characterized by features of a) the research area, b) the research project.

KM """""

.—.

Characteristics of the Study Area"
FSR is by definition farmer oriented and hence focused on specific

groups of farmers in a given area. These farmers operate a given farming

 

system‘which reflects the natural and socio-economic environment of the
area. The resulting farming system varies substantially In degree of
complexity. A complex farming system typically includes a number of

different enterprises, including crops (often lntercropped), nonefarm.and

livestock enterprises, and in many cases more than one management strategy
for a given enterprise - such as use of two or more varieties or planting
dates for a given crop enterprise. This complexity is often greatest

for small farmers who produce largely for subsistence purposes with limited
capital resources and with substantial weather-induced risk (Byerlee).
Understanding such a farming system and measuring data on specific variables
will often be more difficult than in a simple or monocrop situation. With
substantial year to year variability, caution is also needed in interpreting
information obtained for one cropping season. _Moreover because of the wide
range of activities in the farming system, it may be dificult to efficiently
narrow down to the key variables. To overcome this problem in CIMMYT, we
have proposed that FSR initially focus on a predetermined crop in the system

chosen because it is a major user of the farmers' resources (Collinson).

i

I ‘-
‘t

Nonetheless, a more complex farming system implies a greater investment of
time and resources to understand the system and identify the variables for
experimentation. .

The complexity of the study area is also determined by the XifiifPilitY
-of the farming system within the 3599:. Some researchers argue that an effort
be made éafly inthepfesearch process to group farmers into relatively
.Rhomogeneous groups or recommendation damains.. The collection of data would
then initially focus on one of these groups. This is a desirable strategy
especially when the classification of farmers can be made on the basis of
prior information. However, in many cases variability in an area is due to
a number of different environmental.factors and a good deal of information

must be generated in order to be able to decide what is the really important

_.W—-’—"

variability from the point of View of designing improvedftechnologies and

9"“

what is the best dividing line between groups of farmersll. It is our{

”fig—.-

ffl

experience that in a heterogeneous study area, quantitative data is often
useful in understanding how variation across the area affects farmers‘
practices.

Characteristics of the farmer also affect the choice of data collection
methods. Level of education and familiarity with local or standard units of
measure are of course important in making decisions about direct observation
versus farmers' estimation of variables such as yield, area and labor inputs

per unit of anea. .Also the differences iTiianguages in a study area and

 

l/

- For example in areas with substantial variability in farm size, Information
on variability of production practices by farm size is often needed to be
able to decide if stratification by farm size is needed, and if so, at
what farm size(s) is this division(s) made.

and especially between farmers and researchers affects but does not exclude

the use of unstructured methods which depend on direct researcher-farmer

dialogue. Farmers' previous experiences with external contacts, especially

government officials may also be a factor in choice of data collection methods.

WheEe there is substantial suspicion of outsiders, use of methods that do not

depend on written recording of information may be more appropriate.
Ea§g_gf‘t:avel in the study area is another factor affecting decisions

on data collection. where good road systems exist, it may be quite cheap to

make field observations on randomly selected fields on a frequent visit basis

or to use area sampling methods. At the other extieme, very poor communications

in the area may make random selection of farmers very difficult even if a

cluster sampling approach is used. , \
Finally, the availability of existing secondary data will be an importanti

factor in designing the data collection process. Available secondary data,

including previous surveys and experimentation in the area, often helps to focus

the data collection more quickly. 'Secondary information such as lists of

farmers is also important in deciding on a sampling method.

Characteristics of the Research Project

FSR projectsvary substantially in the available human and financial
resources, the place of FSR in the overall research Institution and the
working environment provided for individual researchers. Probably the most
important factor in this mix are the available human resources in the form
of trained and experienced researchers. In many cases FSR is introduced

precisely because researchers have had little experience in working directly

with farmers to solve specific problems. Hence the design of the data
collection process must be based on fairly simple methods which can be used
by less experienced researchers. In particular, few agricultural research
systems employ social scientists and most FSR projects must begin with
limited social science capability. Typically then an FSR. project must
"think small" and adopt a long run strategy of building up the capability
of the researchers in data collection methods through informal on-the-job-
training and possibly formal training courses.

The logistical support of a project will also be a factor in designing
the data collection process. Clearly support in the form of tranSportation
to move about the study area is almost an essential prerequisite for a
, successful FSR program whatever methods-are used. In other cases, the
method may be chosen depending on available support, For example, few
projects have access to efficient data processing facilities so that
collection of specific, quantitative data should be restricted to a minimum.
kImany cases collection of large amounts of data without adequate facilities.
for data processing, is a major bottleneck in developing an effective FSR
project.

The institutional structure in which FSR is situated has a number of
implications for the design of the pfocess. If FSR is imposed, for example
with external support, the project will need to demonstrate early payoffs in
order to attempt to build up internal support and continuity for the project.
This further reinforces the need to design the data collection systems to
efficiently focus on a few high priority problems that promise immediate

solution. Finally the incentive structure facing individual researchers is

~28»

often a limiting-factor in the success of FSR. Researchers require incentives
to direct their efforts to farmer problems rather than traditional disciplinary
oriented research efforts. And to effectively work on farmer problems, requires
substantial field work off the research station. That is, incentives are needed
(or at least removal of disincentives) to ensure that researchers' travel and
lodging costs for field work are adequately covered. Since all'effective data
collection methods require field participation by the researchers, there is
really no alternative to the design of an appropriate incentive system to

encourage this participation.

9299

CIHMYT Experiences in Designing Data Collection Systems

Evolution of the CIMMYT Methodology

CIMHYT first became actively involved in OFR programs in about l975.
At that time a concerted effort was made to switch attention from Egyggstg.
adoption studies (Perrin and Winkelmann (1976)) to gx_gn£é_involvementof
economists in designing technologies. The data collection.system employed
initially emphasized a oneevisit formal survey to obtain information on
practices and problems in the target crop (e.g. Mamafiya; Hoscardi; Perrin).
The survey design and implementation in this initial work was largely the

I

province of the economist.

i

In the next stage, an informal presurvey was introduced (Collinson).
The purpose of this presurvey was largely to help orient the formal survey by

improving researchers' familiarity with the study area and enable better design

(as

of the questionnaire and sampling techniques. The method involved researchers

conducting some farmer interviews fairly informally as they traveled through
the study area over a short period of days. Researchers involved in these
presurveys soon recognized the advantages of the informal interviews conducted
by researchers themselves as a means for obtaining a rapid understanding of the
syStem. Accordingly,the role of the presurvey or exploratory survey as it was

renamed, was elevated in the hierarchy of data collection methods. Primary

f—m

inportance was then attached to the exploratory survey and the formal survey
became the mechanism for verifying specific information obtained in the
exploratory survey.

-~- ”‘ The current methodology comprises an integrated data collection system

including secondary data analysis and exploratory surveys, a formal survey and

experiments (Byeriee, Collinson et l). The general elements of this methodology

are shown in Figure‘___, In this approach the major strategy in efficient
data collection is the sequential process of rapidly narrowing down to
focus on the most important information needed for solving a very few high
priority farmer problems. The characteristics of each data collection method
are shown on Table __3 As the information is narrowed down, information
collected tends to be based more on observation_ with more structured
specific methods with higher level of quantification and representativeness.
This general approach is, of course, evolving over time as new
experiences are accumulated and must be modified to suit the characteristics
of each research situation. In particular, there has been considerable debate
as to whether a formal survey is really needed. 0n the one hand, we have
found that agronomists are quite interested in participating in the exploratory
survey whereas they had previously had less interest in the formal surveys.
Hence, the need for quantitative information to convince_agronomists on
experimental content has largely been negated. Secondly, a number of
experiences where we have conducted a formal survey following an exploratory
survey have indicated that the formal survey provided little new information.
In three cases, we conducted an exploratory survey developed a questionnaire.
and then noted what we expected the responses to be. In one case, in a
fairly homogeneous area we predicted accurately the outcome of the most
important questions. In a second case where there was some heterogeneity
to ease of access to villages, we found that we had considerably overestimated

the use of improved maize technology in the exploratory survey because we had

emphasized more accessible villages. However, information from the formal survey did
_no

“31a

change our conclusions on experimental content.a In the third case where
there was a great deal of variability by farm size and agro-climatic
characteristics,the information from the formal survey on some important
variables was quite important in understanding how these factors affected
farmers' practices and in dividing farmers into groups for the purpose of
on-farm experiments. '

Recent experiences reflect efforts to further improve the efficiency
of data collection systems and tailor them more closely to the specific
research situation at hand. We briefly describe three of these experiences

I

below:

Production Research Program in Ecuador

In Ecuador, the national agricultural research institute (INiAPi has
a separate program which is responsible for on-farm research. Members of the
program are stationed in ten different areas of the country. In each of the
areas work is carried out on the principal crop or crop mixture which is _
grown by small farmers. Nhen work is integrated in a new area an informal
survey is carried out by members of the various crop breeding programs and
support departments which will collaborate on trials. They visit the area, '
talk to farmers, visit fields, and discuss issues with some officials.
This generally takes one to two weeks. After this, a formal survey is carried
out which emphasizes qualitative data, in order to further refine knowledge
of the area and select questions to be addressed in trials,€’The enumerators

for these surveys are always INIAP employees, and include the agronomists from

the on-farm research program who will be stationed in the area, the technicians

(high school degree) who will be assisting them, and other members of

experiment station programs and departments who will be collaborating in the
' research. The results are analyzed and reported as quickly as possible.

The sample size is generally small enough (80-i50) and the questionnaire
[straightfonward enought that the analysis can be done by hand, with the
ihelp of a calculator. As the program is a new one, researchers have found it

useful to have these survey results to present to the extension service of

the Ministry of Agriculture, the Rural Development Secretariat, and other
institutions with whom they must collaborate, as well as for use by station

scientists. I

Researchers plant between ten and thirty on-farm trials in a given
area each year. Information is derived not only from observations of the
trials but from conversations with farmer collaborators as well. In one of
the areas, a field book is used which records a wide range of information
about collaborators and their management practices. The questions in the
field book are redesigned each-year as certain problems are resolved and '
other appear. The question in the field book serve as a guide, and are

' generally not administered formally to the collaborator. The field book is .
rather filled in gradually during the course of the season, based on informal
conversations between the researcher and farmer. As considerable care is
practiced in trying to plant trials with a representative sample of farmers.
the group of collaborators serve in same sense as an on-going panel for the
study of various problems.

On-farm research is a dynamic process. As time goes on, trial results

lead to recommendations for farmers, but other results, combined with observa-

tions and conversations with farmers, lead to further hypotheses. Some of

I
f
.1

these can be tested through trials, others through informal surveys or
questioning, and some require formal survey work. in one of the research
areas, maize and beans were the original focus of work. After four years,

recommendations had been produced for an early-maturing maize variety that

' was finding high acceptance among farmers, and attention shifted to other

crops that could be grown in rotation or association with the new maize. As
the original survey had not dealt with these crops, it was thought useful to
design a small survey whose results could help_guide work in future seasons.
Because the researchers had considerable experience in the area, they were
able to select a series of nine communities which’were representative of the
major variations in the research area for the survey. As they well well known
in most of these communities it was easy to obtain, or construct, lists of
farmers, make a random selection of lo farmers for each site, and carry out a
small survey directed at cropping patterns. Collaboration was excellent
because farmers understood by this time the nature of the work and had been
its results. The survey was carried out by the staff who were working in the
area, plus two members of the INIAP agricultural economics program, and it was

.completed, analyzed, and ready for use in trial planting within four months,.

Ghanaian Grains Development Project

Dn-farm research on maize and cowpeas has been conducted in Ghana under,
a 'CIDA/CIMMYT agreement with two Ghanaian institutions ~— the Crop Research institute
and the Drains and Legumes Development Board. The farmer has responsibility for
most of the on-station research, especially breeding work while the latter has

conducted most of the on-farm research.

Prior to the project, some on-farm experimentation had been done in
Ghana especially on fertilizer trials in maize and verification and demonstrav
tion of improved maize technologies. However, these trials were not a part of
a sequential data collection system and in some cases, especially in spacing
thinning, the technology emerging was inconsistent with a farmers' constraint
of a seasonal labor shortage. Farm level surveys in Ghana had been conducted \
largely by universities and had emphasized the traditional farm management
approaches of collecting general input-output information and were not aimed
at solving specific farmer problems. i

The current project has developed a sequential data collection system.
Limitations of research resources to one economist less than full time has
prevented application of the full methodology. The_gxplorato:y survey has

“M
been the basic data collection method for identifying constraints and planning

—._-.---

r'" “*MH ._,__..—-

on-farm experiments. These exploratory surveys have been conducted at the #
FQEZFaf two éfiyié? in each of the major ecological regions in which maize is
grown. On-farm experiments have been planted since the first year based on
whatever information was available to help choose variables for experimentation.““
Information from the exploratory surveys have been incorporated and has been.
important in more clearly specifying variables for experimentation and in
evaluating results, The data from the exploratory survey is complemented

during the verification/demonstration experiments when some specific quantitative
information is collected from collaborating farmers on practices,-inputs and

outputs during the season. A special form is provided for the technician to

fill in information as a farmer completes a specific operation.

One structured formal survey has been conducted following an exploratory
survey in order to verify information. However this has not been extended to
other regions for two reasons. First, the researchers' time spent in obtaining
further information by the formal survey in this one region was about five weeks
which had .an opportunity cost of exploratory surveys in _t_wp_ other regions. This
does not include the time needed for data processing which has been delayed
because of lackiafacomputer software program. Second, the formal survey
largely verified and strengthened the results of the exploratory survey.and,
provided little or no EEELinformation that would change the design of the
on-farm experiments planted in the area. /

The exploratory surveys have.built-up a useful base of information on
farming systems and maize production practices in various ecological regions.
in most cases they provide a good deal more description and understanding of
the farming systems and major factors influencing farmers' practices than had
been the case from earlier farm management surveys. Indeed, the exploratory
surveys have also been quite successful in getting reliable estimates of
input—output data for construction of farm and enterprise budgets. These
are being confirmed by the information obtained from farmers cooperating in.
the on-farm experiments. .

As the project approaches the point at which recommendations can
Confidently be made to farmers, the major bottleneck to increased productivity
will be policy issues relating to the allocation of foreign exchange and
distribution of inputs, Accordingly, consideration is being given to more
formal surveys focusing on specific information for policy purposes. it is

envisioned that these surveys will only collect information on a maximum of

ho variables relating to specific policy questions1/.
The project has recently purchased a microscomputer which will be used

to process data from the formal surveys and from the on-farm experiments.

Wheat in the Yaqui Valley, Mexico

CIMHYT and the Mexican national research organization, iNiA, have
worked for over thirty years on wheat research in the Yaqui Valley. in fact
it was in this area that the high yielding semi-dwarf wheat varieties were
developed and first released. ClMMYT and lNiA’have had a committment to
on-farm research, but without the involvement of economists and with no
systematic effort to describe and understand farmer circumstances, A survey
was therefore designed to obtain at relatively low-cost a broad description
and understanding of wheat production practices in the area. it focused on
specific information on the weed problem, the major factor limiting production
in wheat in the area. I

A two stage semifformal sampling procedure was used. The first stage
was irrigation blocks (AGO ha each) which were sampled randomly, in the second
stage we took all farmers in the block if there were less than S farmers with
wheat in that block. If there were more than five farmers in a block, we chose

farmers according to convenience in locating them, provided that a) we seemed

 

l/

«- in the case of fertilizer distribution, a major problem, data will focus on
which farmers are obtaining access to fertilizer, farmers knowledge and
perception of the benefits from fertilizer compared to actual use, past ex-
perience with fertilizer use and factors such as cropping intensity that are
associated with natural fertility. This information combined with data from
experiments on fertilizer will then provide guidance to an improved

distribution system.

to be getting a reasonable crossesection of good, intermediate and poor fields
as determined by visual inspection of all fields in the block and b) we also
obtained a representation of land tenure types in that block (i.e. private
farmers - small and large, individual ejidatarios and collective ejidos).
Unstructured interviews usually including senior researchers were
conducted as informally as possible to obtain the following information (not
necessarily in a fixed sequence). I
a) Specific data on 35 key factural variables e largely specific
practices for a specific field (chosen at random among the farmers'
fields) but also including cropping system,’land tenure, credit source
-and machinery availability. The choice of the 35 key variables was

based on the first two days' workiwhere an_attempt was made.to visit

different types of farmers and different parts of the study area.

b) More indepth conversations about specific themes, depending on
_ the practices, experiences and interest of the farmer. On average,

we treated 3-5 themes in this way in each interview.

~c) Observations were made on each selected field with particular
emphasis on weed problems. The types of weeds. growing in each
field were noted and a subjective estimate of percent yield loss
to grassy weeds was made. Eye estimates of yield loss were calibrated
with the aid of agronomists experiehced in weed problems in the area.

Only a few minutes was required for each observation.

in addition to the "random" sample, we also talked informally to other
farmers we met as well as extension agents, bank officials, etc. as would be

done in an exploratory survey. However, data on the 35 key variables were not

recorded in these interviews..

information was recorded after the interviewron three precoded forms

w‘" R*H_ --—— -- -

-.I-..-q.__ fi-‘...‘ _.

- a closed form for the key variables, an open form for the themes treated in
depth and a closed form for any prices and costs mentioned by the farmer.
Initially we did not take notes during the interview~but soon found that the
complexity of the system required us to note the quantitative data (e.g. date
of planting, quantity and type of fertilizer). These data were recorded in

a notebook.

'Running tabulations of the 35 key variables were made during the
survey. Simple and cross-tabulations were set up’to check the sample and
test hypotheses. in some cases this led- to new hypotheses and areas for
further questioning. Other information provided by the farmer - opinions,.
experiences, etc - was transferred to a notebook organized by theme as in
an exploratory survey. This served as a basis for writing the report.

_ The method had several advantages. it was an efficient use of
researchers' time since field work for a sample of lOO farmers involved
three weeks of senior researchers' time. (An exploratory survey would have
required at least two weeks in this complex area.) Moreover, the preliminary
analysis and write-up was completed in less than a week because tabulations
were already done during the survey and relatively few variables were
tabulated. The quantitative information on the key variables was also valuable
in several respects. First, cross-tabulation was important in understanding

the substantial variability that existed in the area. Secondly, quantitative

data from the field observations correlated very strongly with farmer practices

and substitued for experimentation - some of which involving rotations would
be expensive and time consuming. Finally, input distribution problems
through the official bank were uncovered which can be clearly portrayed

by cross-tabulations of input use by credit source.

There are some possible problems with the approach. The identification
of key variables stabilized rapidly in the first two days but some changes were
-still made after one week of field work. Also there is a tendency for
researchers to trade-off indepth informal interviews to shorter more formal
interviews to obtain information on key variables and meet a daily sample
"quota"’ Also with random sampling some time is lost by senior researchers
in locating farmers. These problems can possibly be reduced by having a
séBSFéce enumerator team to expand sample size for the key variables.

work will continue in 1981/82 with emphasis on experimentation. The
selection of fields for experiments will be based on the farmers selected for
the survey taking into account the practices they were using on wheat production.
informal survey methods will be used to continue to accumulate information on

wheat production practices, using collaborating farmers as sources.

-40-

Conclusions

 

We have argued in this paper that FSR has some specific requirements
that need to be consider’in the design of a data collection system
Firstly, FSR has or should have well-defined objectives - the generation
of quite specific information for specific users. There is a need then to
identifyclearly the user, the important variables for that user and then
focus research efforts on their measurement. We believe the sequencing of
information gathering from a broad understanding of the farming system down'
to the important variables is an efficient method that could be more widely
employed in FSR. As the range of important variables is narrowed, more
information can be obtained on each variable through more structured and
' specific surveys, field observations or experiments. Secondly, the researchers
themselves need to be actively involved in the data collection. Especially
in the initial stages, researchers from various disciplines must dialogue with
farmers and become familiar at first hand with the study area. The unstructured -
non-specific interview-conducted by researchers themselves is an efficient
method of obtaining a rapid understanding of existing farming systems and
practices. Researchers involved in this type of interview need to develop a.
better conceptual framework than is currently provided by farm management
methods which largely emphasize input-output type data. A systems' framework
which interrelates the farmers' objectives and resource constraints with his
natural and socio-economic environment,provides a more flexible model for
organizing information. Once the data collection moves to measurement of
specific variables, the participation of the researchers is less crucial,

. especially where a large sample is to be employed to provide narrow confidence

9h]!

limits on a variable. However, the presence of the researchers in the field
and the use of participant observation methods ensures that researchers are
continually broadening and deepening their understanding of the farming situation
developing new hypotheses and defining problems for future work. -
Thirdly, the quantification of variables with statistically determined
confidence levels should be severely restricted in FSR programs to key
variables. in our experience, this type of quantification is most useful
in two types of situations. in complex areas with substantial heterogeneity
in several dimensions (e.g. highly variable farm size and soil type),
quantification and formal testing of hypotheses is often useful to give
researchers a clearer picture of this variability and how it'affects farmers":
practices. Secondly, quantification is often necessary to convincingly
communicate information to the users - farmers, experiment station researchers
and policy makers. But in each of these cases the range of variables
for which quantification is needed is quite small. in most cases, a formal
structured surveys with a one page.questionnaire and 30 variables will be
quite sufficient. The remaining 300 plus variables often collected in a
several page questionnaire are usually at best only marginally useful in
strengthening the value of information finally communicated to the user.
Fourth, we should not confuse quantification with representatiVeness
of.data. The principal problem with unstructured exploratory surveys is the
lack of a mechanism for ensuring that the data obtained is representative of
farmers in the area. Since inter-village variation is often much more
important than intra-village variation, a sampling method in which at least

. the first stage sampling unit - the village - is chosen at random, could be

lllllHHllHllllHIIHIUIUllHllllH’lllll’lllllllillllllll

 

31293 02236 9080