A handbook on the

Methodology
for an integrated
experiment-
survey on rice
yield constraints
S.K. DE DATTA, K. A. GOMEZ, R. W. HERDT, and R. BARKER

1978
THE INTERNATIONAL RICE RESEARCH INSTITUTE
Correct citation: De Datta, S. K., K. A. Gomez, R. W.
Herdt, and R. Barker. 1978. A handbook on the
methodology for an integrated experiment-survey on rice
yield constraints. The International Rice Research
Institute, Los Baños, Laguna, Philippines.

The International Rice Research Institute receives support

from a number of donors including the Ford Foundation,
the Rockefeller Foundation, the European Economic
Community, the United Nations Development
Programme, the United Nations Environment Programme,
the Asian Development Bank, the International
Development Research Centre, the World Bank, and the
international aid agencies of the following governments:
USA, Canada, Japan, the United Kingdom, the
Netherlands, Australia, the Federal Republic of
Germany, Iran, Saudi Arabia, New Zealand, Belgium,
Denmark, and Sweden.

The responsibility for all aspects of this publication rests

with the International Rice Research Institute.

First Printing
November 1978
Second printing
November 1981
Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Background and objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

A team approach to constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Expected use of the research results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

The model ..................................... 7

Objectives of the methodology . . . . . . . . . . . . . . . . . . . . . . . . . 8
Actual and potential farm yields defined . . . . . . . . . . . . . . . . . . . 10

Choice of study area and sites . . . . . . . . . . . . . . . . . . . . . . . . . . 15

Choosing the study area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Selecting experimental sites . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

Experimental procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Selecting the treatments to be tested . . . . . . . . . . . . . . . . . . . . . . . . . 20
Agronomic data to be recorded . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Unique procedural considerations . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Survey procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
Intensive data parcel concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

Analysis of constraints data . . . . . . . . . . . . . . . . . . . . . . . . . . 45

Estimation of yield gap and contribution of the test factors . . . . . . 45
Economic analysis of a constraints experiment . . . . . . . . . . . . . . . . . . . 52
Socioeconomic constraints analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 56

References cited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Preface

A multitude of factors have been identified as possible causes of low

levels of rice production in Asian countries, but few have been em-
pirically verified. One difficulty in such verification is the diverse
nature of the constraints – some relate to the rice plants and the
field in which they grow, some to the farmer who grows the rice,
some to the village or province in which the farm is located, and
some to national policies.
Between 1973 and 1977, IRRI agronomists, economists, and
statisticians developed and implemented a multidisciplinary research
project to study rice yield constraints, and encouraged colleagues
in countries throughout Asia to undertake similar projects. The
International Development Research Centre of Canada helped sup-
port the coordinated research effort.
This handbook presents the basic methodology resulting from that
experience. It examines constraints to increased rice production in
farmers’ fields when there is reason to believe that unexploited
technology could give higher yields. Underlying the approach is the
hypothesis that a technology will be widely adopted if it provides
farmers with substantial economic benefits. The yield constraints
project methodology presented in this handbook will give researchers
answers to the following questions:
1. What is the gap between yields with farmers’ present practices and
the highest yield possible in their fields after predetermined pro-
duction factors are modified?
2. How much of the yield gap can be attributed to each of the pre-
determined production factors?
3. What are the economic costs and returns associated with the dif-
ference between farmers’ present practices and the maximum yield
level of the selected factors?
4. How much of the yield gap can be profitably recovered?
5. If the inputs that are most profitable differ from the actual
inputs used by farmers, what personal, social, institutional, or
political factors keep farmers from using the most profitable
levels?
 This methodology is appropriate for determining the importance
of several factors contributing to a yield gap in areas where tech-
nology that is capable of giving substantially higher yields exists but
is not used by farmers. The methodology is not designed to measure
the effect of all possible production constraints. It cannot, for
example, explain why some farmers grow one rice crop a year while
others grow two. It is not a technique for understanding why some
governments can effectively implement fertilizer distribution schemes
while others cannot, and it does not examine why some governments
have high rice prices while others have low.
This handbook, based on 4 years of experience in the constraints
research project by IRRI and by six national teams, provides a ready
reference for anyone interested in the methodology of the yield
constraints project. The project was an evolving research effort, and
the methodology resulted from a trial-and-error process over many
locations and seasons. The handbook presents the central core of
techniques and designs that are used to answer the questions posed
above.
The methodology described does not contain empirical results,
but some illustrative data are used. Empirical results are presented
in IRRI annual reports from 1973 through 1977, and in the IRRI
publication Constraints to High Rice Yields on Asian Rice Farms: An
Interim Report.

2 CONSTRAINTS METHODOLOGY HANDBOOK

Background and objectives

An estimated 30% of tropical Asia’s rice land was planted to semi-

dwarf rice varieties in 1978, and some countries had 70% of their
rice land in semidwarfs. These modern varieties have helped signi-
ficantly in feeding the growing populations of Asia, providing 1 t/ha
higher yields than the traditional varieties in many places, but their
average farm yields are still far below those demonstrated possible
by research scientists. Far from reaching the researchers’ yields of
6 to 8 t/ha, good farmers obtain 3 to 4 t/ha, while many are lucky
to get 2 t/ha.
The situation suggests two questions:
• Why have some farmers still not accepted the modern varieties?
• Why are many farmers unable to achieve the potential high yields
of the modern rice technology?
The first question has received considerable research attention. But
few researchers have asked the second question, which means there
is still little quantification of the factors responsible for the low
on-farm yields of the modern varieties.
The factors that keep farmers from attaining high yields with
modern varieties may be physical, economic, social, or any com-
binations of the three. Physical conditions on some farms may
prevent the farmer from exploiting the full potential of the tech-
nology. High yields may be physically possible but economically
unprofitable. Lack of credit may prevent farmers from using available
inputs. In still other cases, social or institutional problems may exist.
And there is always the possibility that aspects of the technology
may not be understood by the farmers or by those directly advising
them.

Because of the possible combination of factors that can keep farmers A TEAM
from getting high yields from the new rice technology, a team APPROACH
approach is required. Both agronomists and economists are needed TO CONSTRAINTS
to make an effective study. An experimental technique for identifying
 and quantifying yield constraints in farmers’ fields was developed
and tested by Gomez et al (1973). It measures the potential yield,
the actual yield, and the yields corresponding to the inclusion or
withdrawal of test factors over and above the farmer’s levels in plot
experiments on sample farms. The relative contribution of each com-
ponent to the difference between the potential yield and the farmer’s
yield was then assessed.
De Datta (1974) compared a series of combinations of inputs of
increasing intensity (management packages) to establish the yield
and profitability of several different combinations and to indicate the
approximate intensity most attractive to farmers.
Although experiments on farmers’ fields provide evidence of the
performance of the new rice technology, an understanding of man-
agerial ability, input availability, and profitability requires data that
are obtained only through farmer surveys. Hence, we integrate field
experiments and socioeconomic surveys of farmers, the study area,
and the institutions serving them.

EXPECTED USE The results of the research are potentially useful in at least three
OF THE RESEARCH ways:
RESULTS 1. The discovery of a large unexploited gap that can be economically
recovered is a clear indication that more or better extension work
is needed. If the gap is large and cannot be economically overcome,
there is a need to seek a cheaper technology. If the researcher’s
technology does not increase farm yields under farmers’ condi-
tions, then either farmers have adequate levels of technology or
the existing (research) technology doesn’t work for the farmers.
In the first case, scientists can give their attention to other pro-
blems; in the second case they should try to improve the existing
technology.
2. Clearer understanding of how the physical and institutional envi-
ronments restrict yields can suggest modifications to make the
new technology more usable in those environments. It may also
pinpoint situations where institutions or policies can be changed
to make the technology more attractive.
3. Improved understanding of constraints will enable extension
workers to help farmers make better choices among the pro-
duction techniques currently available to them, given their pres-
ent resources.

4 CONSTRAINTS METHODOLOGY HANDBOOK

 Each use of constraints research results can help to increase rice
yields. But the constraints evaluated must be representative of a
significant geographic area of rice production in a country in order
to provide enough data for national program changes.
Our primary attention is on the difference between the yields that
we can achieve under farmers’ conditions, and the yield farmers
actually obtain in the same areas. We believe that an understanding
of this difference or yield gap will lead to faster progress toward
increased rice production.

BACKGROUND AND OBJECTIVES 5

The model

Figure 1 demonstrates the concept of the difference between actual

farm yield and maximum experiment station yield. The first com-
ponent, yield gap I, exists mainly because of environmental differ-
ences between experiment stations and the average rice farm. The
technology that gives high yields on experiment stations may not
give as high yields in the less favorable environments that exist in a
large part of the rice-growing area of Asia. There may also be com-
ponents of experiment station technology that are not transferable
to a farmer’s field.

1. The concept of yield

gaps between an experiment
station rice yield, the
potential farm yield, and the
actual farm yield (Gomez
197 7).
 The constraints research project focuses on yield gap II. Gap II is
the difference between the potential and actual yields in farmers’
environments. By definition, this gap exists because farmers use
inputs or practices that result in lower yields than those possible on
their farms. Gap II breaks into two areas of investigation.
• One area identifies what biological or physical inputs or cultural
practices account for the gap. The most critical factors differ from
one region to another, but before any remedies can be taken
(e.g., recommending a package of improved practices) the biological
nature of the gap must be understood. Experiments in farmers’
fields are essential to obtain this information.
• The other area identifies why farmers are not using the inputs or
cultural practices that would result in higher yields. The importance
of these factors will differ from area to area, but understanding
them will help in designing programs that incorporate the missing
biological components for overcoming the yield gap. Economic
analysis of the experiment and farm surveys provide the main
research inputs for this area.

OBJECTIVES OF The general objective of the methodology is to identify the factors

THE METHODOLOGY that explain the difference between actual and potential rice yields
in selected farm environments. The specific objectives follow:
• to measure the on-farm gap between the yield with production
techniques or input levels for maximum yields and the yield of
representative farmers in their physical environments;
• to determine the contribution of each tested technical factor (i.e.
input or cultural practices) to the gap between actual and potential
yield;
• to determine the extent to which use of each tested technical
factor can be profitably increased;
• to determine what social and institutional factors prevent farmers
from using technology that gives higher, more profitable yields;
• to determine whether it is possible to change the physical and
socioeconomic factors constraining yields, and the difficulty, if
any, of making such a change.
Figure 2 diagrams the constraints research process and the relation-
ship of the methodology to the goal of agricultural development –
increased production, profit, and rural employment. Attaining that
goal requires that new technology appropriate to the environment

8 CONSTRAINTS METHODOLOGY HANDBOOK

be extended to farmers. An opportunity for profitable sales of farm
output must also exist. If production does not increase, or does not
increase at an appropriate rate, the reasons are sought by asking the
following questions:
• When the new technology is used in farmers’ fields, are yields
increased over those with use of the farmers’ technology? If not,
what factors restrict yields? Answers here provide feedback to
researchers and enable them to design more appropriate technology.
• Is the new technology more profitable than the technology now
used by farmers? If not, it is not reasonable to expect it to be
adopted.
• Is the level of profitability restricted by government policies? If so,
can the policies be changed, or can the technology be redesigned?
Answers here provide feedback to policy makers and researchers.
• Do farmers have a level of knowledge adequate for effective use
of the technology? Are the necessary inputs and cash or credit
required to use the technology available to farmers? If not, how
can programs ensure the availability of such inputs? Answers here
provide feedback to government programs, and to the extension 2. Constraints research in the
system. development process.

THE MODEL 9
ACTUAL AND Actual farm yields are easy to define. They are what farmers get.
POTENTIAL FARM They may be measured for a field, a farm, a village, a region, or a
YIELDS DEFINED nation. In the constraints research we measure farm yields from the
farmers’ own field production or from experiments in farmers’ fields
simulating farmers’ practices. For purposes of the project we define
potential yield as the highest yield obtained in farmers’ fields with
improved technology.

Field experiments and treatments

Three types of field experiments are conducted: complete factorial,
minifactorial, and supplemental. The field experiments accurately
estimate the potential farm yield, the actual farm yield, and inter-
mediate yield levels representing varying combinations of input use.
The specifics of each type of experiment are discussed later. Two
sets of treatments are tested in the experiments: the factorial com-
ponent and the management-package component.
Factorial component. Treatments in the factorial component are
either complete or incomplete factorial combinations of n test factors
(they may be practices or production inputs), each at two levels. The
test factors included are those that researchers hypothesize to be the
major causes of low yields in farmers’ fields. The two levels of each
factor are
• the farmer’s practice, and
• the maximum-yield or high practice.
The farmer’s practice is what the farmer actually does in the current
crop season, and it varies from farm to farm. The high practice is the
level of each factor anticipated to give the maximum yield, and is
fixed for all farms in a given environment. All practices and inputs
not considered as test factors are held constant at the farmer’s level.
The factorial component supplies data on actual farm yield, poten-
tial farm yield, and varying yield levels resulting from a systematic
increase or withdrawal of one or more of the test factors. The data
are used to estimate yield gap II, and the individual contribution of
the test factors to the gap. The precision of this estimate of the yield
gap depends on the choice of the test factors. Theoretically, as the
number of test factors increases, the accuracy of the estimate of
potential farm yield increases because more complementary factors
are included. Consequently, the yield gap estimate is more accurate.
As test factors are added, however, the experiment becomes more
difficult to conduct.

10 CONSTRAINTS METHODOLOGY HANDBOOK

 Management-package component. Treatments are designed to re- The farmer’s yield is the
base for measuring the
present intermediate levels between the farmer’s practices and the yield gap. For example,
anticipated-maximum-yield practices. The incremental steps between fertilizer should be
treatments usually involve a simultaneous change in more than one spread before rice
exhibits severe nitrogen
input. The management-package component tests the different input deficiency.
combinations selected to represent different yield levels and pro-
duction costs. It does not measure the contribution to the yield gap
of a particular input (which is the purpose of the factorial com-
ponent), but it allows a meaningful look at the question of costs and
returns for each management package. One or more management
packages could be suitable for immediate recommendation to farmers.

Experimental technique
The high level of each test factor and of each management package is
specified before the start of the experiments. The farmer’s level of
each test factor, on the other hand, is usually not known in advance,
but is obtained through observation during the cropping season.
Because a farmer’s practices may vary from paddy to paddy, a com-
parable paddy technique is used to specify the farmer’s level of each

THE MODEL 11
 test factor, i.e. the same paddy where the experiment is located, or
a nearby paddy, is chosen as the comparable paddy before the
experiment is set up. Whatever the farmer does in the comparable
paddy becomes the farmer’s level in the experiment.
In the experiment, all management practices and cultural practices
except the test factors are the same as those used by the farmer
on whose farm the experiment is located. As much as possible,
they are managed and applied by the farmer himself in the same
manner in which he treats the rest of his farm. They are the fixed
factors in the experiment.
In each experiment data on grain yield, insect and disease inci-
dences, weed incidence, rat damage, and water conditions are col-
lected. In addition to records of the farmer’s level for each test
factor, records of all his other management and cultural practices
are taken. When possible, data on the farm’s physical environments
(climate and soils) are also gathered. Such information may help
explain the difference in experimental farm yields and in yield gaps
over farms, seasons, and years.

Surveys
The farm survey part of the study
• supplies preliminary information on the probable farm practices
that need to be improved to get higher yield in the study area;
• expands the area of coverage from that possible through field tests
(field experiments are relatively more costly to conduct and thus
are not expected to cover a sufficiently large part of any target
area); and
• describes the socioeconomic conditions of the farms (age, sex,
education, employment of household members, farmers’ attitudes
and perception, and existing farm practices), and the related econ-
omic environment (marketing, prices of inputs and produce, input
suppliers, and credit opportunity).
A preliminary survey and a follow-up survey are required. Inter-
views are used to obtain the required information, and the crop-
cutting technique is used to determine yields when the interview
method is deemed inaccurate.
The preliminary survey is done before the experiments are estab-
lished. Its purpose is to obtain general information within the study
area on farm size, tenure status, productivity, technology awareness,
irrigation facilities, and perceived constraints. Where such information

12 CONSTRAINTS METHODOLOGY HANDBOOK

is available a preliminary survey may not be necessary. Such inform-
ation provides the basis for
• a hypothesized set of production constraints to be used as test
factors in the experiment, and
• sampling design for the selection of experimental farms, and
a
sample farms in the follow-up survey.
The number of questions in the preliminary survey is generally small
so that a large number of farms can be covered.
The follow-up survey is done at the end of the crop season of the
experiments. The information pertaining to the crop season gathered
is:
• farmer’s levels of the factors tested in the experiment,
• farmer’s variety, management, and cultural practices,
• yield,
• farmer’s perception of factors that limit yield, and
• cost of individual inputs and price of rice.
The follow-up survey sample is selected from those farmers included
in the preliminary survey. It includes all the farmers on whose fields
experiments were conducted, and at least an equal number of others
selected at random.

THE MODEL 13
Choice of
study area and sites

The key to a successful yield constraints study is thorough planning.

The major planning activities include choosing the study areas, select-
ing the experimental factors, and designing experiments and surveys.

Choice of the study area depends on the individual researchers. While CHOOSING THE
the goal is to cover enough area to be able to gather valid data on STUDY AREA
important rice-producing regions, the area will usually be limited.
The area in which the study will be carried out should be
• representative of areas where a large proportion of the rural pop-
ulation is dependent on rice and where total rice production is
important;
• one where there is reason to believe that a significant and im-
perfectly understood gap exists between potential and actual rice
yields; and
• accessible to the researchers.
For the preliminary survey the farmers are selected by a stratified
sampling procedure. The strata may include units such as agroclimatic
area, municipality, and village, perhaps with subvillage as the final
stratum. Because the objective is to represent an entire area, and not
specific villages or subvillages, proportional sampling may be used.
The total number of farmers interviewed in the preliminary survey
should be roughly twice the number that will be interviewed in the
follow-up survey. There should be a minimum of 100 farmers for
the preliminary survey in each area. Enough units of final strata are
chosen to represent the diversity in the study area. Farmers are
sampled at random from the final strata. A preliminary survey
questionnaire is in Table 1.
16 CONSTRAINTS METHODOLOGY HANDBOOK
Experimental sites should be as representative as possible of the SELECTING
study area, but random selection of experimental sites is impractical EXPERIMENTAL
because of the need for farmer cooperation and accessibility. Never- SITES
theless the researcher should identify an accessible site that is repre-
sentative of a substantial number of paddy fields in the area, taking
into account, soil, landscape position, and water flow.
Water, the most critical environmental factor in rice production,
is important in the selection of the research sites. Some research sites
should be in areas where water control is poor so that constraints
there could be measured. But not all sites should be concentrated
in such areas because it is questionable whether potential yields are
much higher than actual yields there. Sites that cover a range of
water control conditions must be chosen. Research sites may include
One method of selecting experimental sites is to divide farms into rainfed farms where
low, medium, and high yielding groups on the basis of the preliminary drought damage is often
severe such as is seen
survey, and to select an equal number of experimental farms from here (Camarines Sur,
each group. Philippines, 1981).

CHOICE OF STUDY AREA AND SITES 17

Experimental procedures

The researchers decide which test factors to include in constraints

experiments, what levels of those factors to test, and what com-
binations of the factors to use. Four factors are the maximum
practical number that can be included in the experiment using the
complete factorial treatment combinations; three factors use about
as much land as most small farmers are willing to give up. A larger
number are easier to handle with minifactorial or other treatment
combinations.
If study of a location-specific factor is important, some experi-
ments can be placed in one type of location and some in another, so
that the factor can be studied in a slightly different way. For
example, a problem soil may be a locational factor that can be
studied through experiments in areas with good soil and with the
problem soil.
Selecting test factors. The chosen experimental factors are the
three or four most important aspects of modern rice technology that
are required for high yields but are not used by farmers, or are used
by farmers at inadequate levels.
Constraints researchers should be able to identify a package of
high-yield technology that can be separated into three or four com-
ponents or test factors. Each factor is included separately in the
factorial component and in combinations at various levels in the
management package component.
If the researchers want to find out what yields farmers could
achieve and the reasons they are not being achieved, the inputs used
in each test factor should be purchasable in the country, i.e., they
should be potentially available to farmers. Two specific examples
illustrate this concept:
1. An unreleased breeding line should not be contrasted with the
farmers’ variety. Variety can, however, be a variable factor if an
improved variety is available but has not been adopted by farmers.
2. Only chemicals that are commercially available should be used in
the experiment, not materials available only to research workers
or materials especially imported for the experiment.
SELECTING THE The choice of the treatments depends on the resources available to
TREATMENTS TO the researcher and on the depth of information desired. A large
BE TESTED experiment gives more information but requires more people, time,
land, and money than a small experiment. A smaller experiment
represents the agroclimatic area better because it permits more sites
to be covered. The combination of designs we discuss is a feasible
compromise between cost and completeness.
Three types of trials in farmers’ fields are conducted in a study
area. All include the same test factors. The high level of the three or
four test factors is set at the same maximum yield level in all trials.
Three types of trials are conducted on a minimum of 20 farms –
complete factorial on 4 farms, minifactorial on 4 other farms, and
supplemental on 12 others.

Complete factorial trial

In a complete factorial trial, a complete factorial combination of the
test factors, each at two levels (Gomez and Gomez, 1976), and three
management packages (De Datta et a1 1976) are tested. The number
of treatments is equal to 2n + 3 (where n is the number of test
factors). Factor levels for each treatment are shown in Table 2
for the case of three test factors.
In the factorial, the test factors are kept at the farmer’s level
(level 1) and the maximum yield level (level 5) as defined below:
Level 1: actual farmers’ practices. These differ from site to site,
depending on the practices followed by the particular farmer.
Level 5: the level of test input needed to get the maximum yield,
with good weather for the entire crop season. No cost limit is
involved.
In the management packages, the test factors are combined in
incremental steps between the farmers’ level and the high level. These
packages are therefore combinations of the test factors. No fixed
number of packages is required. Three, in addition to the levels in
the factorial, are described below.
Level 2: a modest level of inputs that may be chosen to get the
greatest added yield per added unit of input.
Level 3: a level of inputs which has a cost about midway between
the cost of levels 2 and 4. This might be the technology package cur-
rently recommended to farmers in the area.
Level 4: the level of inputs that is expected to give maximum
profit in good weather, and will not result in excessive lushness,

20 CONSTRAINTS METHODOLOGY HANDBOOK

Table 2. Treatment included in each of the three types of experiments conducted in farmers' fields in constraints research,
assuming three test factors.

Level of a Treatment included in

Treatment Variety Fertilizer Insect All other Complete Mini- Sup-

(Factor 1) (Factor 2) control factors factorial factorial plemental
(Factor 3)

Factorial
1 1 1 1 1 X X
2 5 1 1 1 X
3 1 5 1 1 X
4 1 1 5 1 X
5 5 5 1 1 X X
6 5 1 5 1 X X
7 1 5 5 1 X X
8 5 5 5 1 X X X

Packages
9 5 2 2 1 X
10 5 3 3 1 X
11 5 4 4 1 X Xb

a Level 1 = farmer’s level. Level 5 = maximum yield level, detailed specification given. The factors must be chosen for each
site by researchers at the site; variety, fertilizer, and insect control are merely illustrative.
b Optional.

lodging, or other yield reductions if poor weather occurs during the

crop season.
If desired, an additional treatment that tests a high level of cul-
tural practices together with the high level of all test factors may
be added. All treatments are randomized together, two replications
are used, and plots are between 18 and 20 m2, with a harvest area of
10 m2. The farmers’ level of each test factor is simulated using the
comparable paddy technique (see p. 22). Yield of the simulated
farmers’ plot (treatment 1) is compared with a crop-cut yield estimate
from the comparable paddy to assess the accuracy of the simulated
farmers’ treatment.

Minifactorial trial
In a minifactorial trial, the number of treatments equals two more
than the number of test factors – one with all factors high, one with
all factors at the farmers’ level, and the others, each with all but one
factor at the high level (Table 2). Two replications are used and plot
size is 18-20m2 with harvested area of 10 m2. If desired, additional

EXPERIMENTAL PROCEDURES 21
plots with cultural practices at the maximum yield level or one plot
based on management package level 4 in order to evaluate profit-
ability may be added. The farmers’ level of each test factor is
simulated by the comparable paddy technique; the yield of the
simulated farmers’ plot is compared with yield estimated by crop
cut on the comparable paddy.

Supplemental trial
In a supplemental trial, one plot has all the test factors at the high
level. This gives the estimated potential farm yield. The plot is 50 m2
in size and is not replicated. Farmers’ yields are measured by crop-
cutting the farmers’ field (five cuts of 2 m2 each are suggested). The
crop cuts must be on the same paddy as the supplemental trial. A
second test plot with cultural practices and test factors at the maxi-
mum yield level may be added, if desired.
The supplemental trials need not be surrounded by levees (IRRI
1978). If the center portion, say 10 m2 , of the 50 m2 plot is used to
estimate potential yield, plots without levees provide the same degree
of accuracy as plots with levees, and give fewer water control
problems.
Levees are also unnecessary in the complete and minifactorial
experiments if plot size is kept at 20 m2 , and no water flows through
the plots when chemicals are applied. In test plots without levees,
the separate plots must be carefully identified and treatments must
be imposed accurately. Despite the fear of researchers, it seems that
with a moderate degree of water control, separate fertilizer treat-
ments can be maintained with unleveed plots (IRRI 1977). When
variety, spacing, seedling, land preparation, and other similar factors
are being tested, levees are not needed.

Comparable paddy technique

There is invariably some difficulty in knowing what the actual
farmer’s practices will be on a particular farm. Although the farmer is
interviewed about his farm practices before the trial, the chances are
that he will not do exactly what he plans to do. Periodic follow-up
is, therefore, essential. But because it is time consuming and is a
nuisance to the farmer, and because a farmer may not use the same
practices uniformly throughout the farm, a comparable paddy tech-
nique is suggested. It involves the following procedures:
• A nearby paddy that is quite similar to the paddy with the experi-

22 CONSTRAINTS METHODOLOGY HANDBOOK

 ment is chosen. It is referred to as the comparable paddy. The
timing and amount of the test factors used by farmers on the com-
parable paddy are simulated by the researcher in appropriate treat-
ments in the experiment.
• The researcher should inspect the comparable paddy at every visit.
If the test factors are weeding or irrigation, the researcher should
be able to detect the accomplishment of those factors by visual
observation. For test factors like fertilizer or insecticide application,
for which visual observation is not possible, markers may be used.
A bamboo stick with a bright color on one end may be placed at
one corner of the comparable paddy; the farmer turns the marker
upside down whenever chemicals are applied. That is the signal for
the researcher to contact the farmer for necessary information on
the specific operation.

To be most effective, the constraints project should continue for AGRONOMIC

several seasons in a given area. In the course of the experiment, data DATA TO BE
that may prove helpful in interpreting variations in the experimental RECORDED
results are obtained. Some data are recorded on a per-site basis,
others on a per-plot basis. Observations are taken at different times
throughout crop growth. Some data need to be obtained only once;
others need daily records and still others, weekly. Table 3 sum-
marizes the data to be collected.

Preplanting data
Before the installation of the experiment, the researcher obtains a
soil sample for each site and has it analyzed. The maximum inform-
ation necessary is:
• pH • cation exchange • clay minerals
• % organic capacity • extractable phosphorus
matter • % clay • exchangeable
• % sand • % silt potassium
The researcher interviews the farmers and obtains a general record
of the latter’s practices and input use during the preceding two
seasons. A sample record of one farm is in Table 4.

Site-specific crop condition data

Each week during the experiment, site-specific crop data should be
recorded in the appropriate place on a form like that illustrated in
Table 5. The form may also be used for soil test data. The number

EXPERIMENTAL PROCEDURES 23
 of weeks should be adjusted to the duration of the variety being
grown. The major crop growth stage should be entered in the first
column, indicating maximum tillering, panicle initiation, flowering,
and harvest.

Table 3. Frequency of observations and locations for which agronomic data are recorded.

Time of Record data for

Data observation
the site each plot

Soil test Before experiment X

Farmer's past practices Before experiment X
Solar radiation Daily X
Rainfall Daily X
Water adequacy Daily X
Farmer's practices on comparable paddy 2-3 times a week a X
Weed incidence 21 DT, MTb X
Rat damage When it occurs X
Insect and disease incidence When it occurs X
Yield, corrected to 14% moisture At harvest X
a Record the actual date the farmer performs the practice.
b DT = days after transplanting, MT = maximum tillering.

Table 4. A sample record of cooperating farmer practices a and performance the two seasons before the experiment.

Wet season before Dry season before

the experiment the experiment

Land preparation 1 plowing + 4 harrowing 1 plowing + 3 harrowings

Fertilizer (NPK kg/ha) 123–70–0 32–40–0
Weed control HW at 30, 45 DT HW at 30 DT
lnsect control:
Seedbed Folidol at 27 days Folidol
Field Sevidol at 20 DT Furadan
Varieties
Largest area IR26 IR26
Second largest area IR20 IR20
Other varieties IR32 IR30, IR1561
Yield 2.6 t/ha 3.5 t/ha
Cost of production inputs/ha P900 P650
Rice area farmed 3 ha 3 ha
Major yield constraints observed Tungro None
a HW = hand weeding, DT = days after transplanting.

24 CONSTRAINTS METHODOLOGY HANDBOOK

EXPERIMENTAL PROCEDURES 25
 Weekly data. Solar energy values for the experiments may be
measured with the use of a week-long recording actinograph. One
recorder will serve several experiments in close proximity. Sunshine
hours can be used as a proxy variable if a recording actinograph is
not available. If this is done, the table heading should be changed to
reflect the fact. Daily rainfall is measured in standard rain gauges,
and added for each 7-day period to give the week’s total rainfall.
Water adequacy. Because of its overriding importance, the water
condition at the site must be recorded. Several ways are acceptable.
As in the case of all measurements in Table 5, however, the water
data refer to the average for the entire site, a concept that may
require judgment by the researcher. In the complete factorial experi-
ment, an average of three to five water measurements per site may
be needed, even though efforts are made to have level fields. In the
supplemental experiment, one measurement per site is sufficient.
Alternative measurements may include
• average soil moisture tension (in centibars) measured with a tensio-
meter (this measure cannot be used under flooding but may be
best in dry soil, when drought is encountered);
• average depth of water (procedure described below);
• number of nonflooded days – days when there is no standing water
in the paddies; and
• number of days during which the rice plants are submerged.
Sloping fields should be avoided, but any farmer’s field will have
some places higher or lower than others. At the time of transplanting
or just before, observe how level the field is, and note the plots
where water depth is average. Three to five spots within the experi-
ment that represent the water depth of the site should be picked for
monitoring purposes. If possible, they should be in plots adjacent to
the field boundary bund to make recording easier.
boundary bund to make recording easier.
Depth of water. A wooden stake about 2 cm × 5 cm × 75 cm
should be painted white. A ground level mark in black paint should
be made at about 25 cm, and marks at 1-cm intervals above that with
numbers every 5 cm to the top of the stake. The stake is driven into
the soil with the ground level mark just at the level of the soil. Depth
of water can then be conveniently read directly from the stake. The
depth of water is recorded each day, or the fact that the field is dry
is noted. The depth on various stakes can be averaged to get one
depth for the site for each week. It is usually possible to teach

26 CONSTRAINTS METHODOLOGY HANDBOOK

someone who lives in the village to make daily measurements for a
modest fee.

Plot-specific growth and condition data

Plot-specific data should be recorded on a series of forms similar to
those illustrated in Tables 6, 7, and 8.
Treatments. For each site, a detailed plot layout like Figure 3
should be drawn, showing the randomization used. Treatment details
for input levels 2 to 5 should be summarized on a form like Table 6.
In addition, for every site, the specific practices used by the farmer
in the comparable paddy, and in the farmer’s level of the experiment
(level 1) must be recorded on a form similar to Table 6.
On forms like Table 7, used for recording damage, the level of each
experimental factor should be indicated as in Table 2, using 1 to
represent the farmer’s level, 5 the maximum yield level, and 2, 3, 4
the intermediate package levels.
Weed incidence. The weed population of each plot is scored on a
scale of 1 (no weeds) to 9 (maximum possible infestation). It should
be scored at about 21 days after transplanting, and at maximum

3. A sample plot layout

for a complete factorial-
management package
experiment. The
comparable paddy should
be in the same paddy
with the experiment, or
in a nearby paddy.

EXPERIMENTAL PROCEDURES 27
28 CONSTRAINTS METHODOLOGY HANDBOOK
EXPERIMENTAL PROCEDURES 29
30 CONSTRAINTS METHODOLOGY HANDBOOK
tillering before any mechanical weeding operation that is scheduled
for those times. Data are recorded in the Table 7 form.
Pest and disease incidence. Rat damage, insect damage, and disease
incidence for each plot are scored at the times indicated in Table 3.
Degree of infestation is noted on a scale of 1 (no damage) to 9
(complete destruction) on forms like those in Tables 7 and 8. The
IRRI publication Standard Evaluation System for Rice gives useful
guidelines for scoring common rice pests.
Grain yield. Harvest the center 10 m2 area of each plot. Samples
should be threshed, dried, cleaned, weighed, and checked for moist-
ure content. The yields obtained are adjusted to the standard 14%
moisture level. Techniques for obtaining these data are elaborated in
Techniques for Field Experiments with Rice (Gomez 1972).
On the comparable paddy, yields are estimated by crop cutting.
Five cuts of 2 m2 each are taken at random throughout the com-
parable paddy. They are bulked together, threshed, dried, cleaned,
weighed, and checked for moisture content. Yield is adjusted to 14%
moisture.

For yield constraints experiments, special explanation must be given UNIQUE

to the fixed factors. Because the objective is to measure the yield PROCEDURAL
contributions attributable to a few – say three or four – most CONSIDERATIONS
important yield limiting factors under the farmers’ conditions, the
rule is that the fixed factors must be kept at the farmers’ level in the
factorial component. If this rule is not followed, it will be impossible
to attribute the yield effect to the variable factors; instead some of
the yield benefits will be due to the fixed factors and some to the
variable factors.
In some cases, it may appear that other factors besides the few
selected test factors are important for raising farm yield. In such a
case, the solution is to simply add one treatment in which all possible
fixed factors as well as the variable test factors are held at the
maximum yield level. A comparison of yields obtained on such a
plot with yields obtained on the plot with the variable factors all high
and the fixed factors at the farmers’ level will show the combined
yield effect of the fixed factors.
The most practical way of ensuring that fixed factors are held at
the farmer’s level is to have the farmer himself carry out operations
that are not part of the variable test factors. Thus, if seedlings, vari-

EXPERIMENTAL PROCEDURES 31
 ety, or seeds are not a test factor, the farmer’s seedlings are used in
the experimental plots. If weed control is not a test factor, the
farmer should be told to weed the experiment in the same way he
does the rest of his crop, etc.

SURVEY The sampling of farms for the preliminary survey was discussed in
PROCEDURES the chapter Choice of study area and sites. The follow-up survey
requires a sample including all farms with experiments plus at least
twice as many other farms selected from the area.

Design of follow-up instruments

The socioeconomic aspect of the investigation assumes that the avail-
able technology results in higher on-farm yields than what farmers
are getting. The objective is to determine why farmers are not using
all aspects of the new technology to get those high yields. If the
analysis of the experiments in farmers’ fields shows that yields or
net returns with the new technology are no higher than what farmers
are already getting, then socioeconomic reasons may not explain
farmer behavior. However, the experimental and the survey aspects
proceed simultaneously.

Issues addressed
In the surveys, attention is given to the economic, institutional,
social, and psychological constraints. The major issues addressed by
the survey and subsequent analysis follow.
1. Is the maximum yield level of one of the management packages
tested in the experiments more profitable than present practices
(given costs, prices, tenure, market discrimination)
• on the farms where experiments are conducted, or
• on other typical farms?
2. Are inputs available
• at the time when needed, and
• in markets accessible to the farmers?
3. Do farmers have the cash needed to use improved technology
• from institutional credit sources,
• from non-institutional credit sources, or
• from their own resources?
4. How well do farmers understand the components of the tech-
nology vis-a-vis

32 CONSTRAINTS METHODOLOGY HANDBOOK

 • familiarity with brands, chemicals, problems;
• familiarity with use of the technology;
• where to get materials, their cost; and
• what inputs to use for what problems?
5. In terms of their own perceptions, why are farmers using their
present level and method of input application, especially the three
or four critical inputs tested in the factorial experiments?
• What are the major factors that farmers believe keep their yields
low?
The follow-up survey focuses on this limited set of questions.
Only data that answer the questions are included. In that respect,
the constraints survey is much more limited than the usual socio-
economic research survey.

In many countries, farmers cultivate several parcels of land that are INTENSIVE
frequently of different types. For example, one may be rainfed, DATAPARCEL
another pump irrigated, and a third supplied with water from a CONCEPT
canal. To generate meaningful data, it is necessary to address certain
questions to every parcel, a process extremely time-consuming and
tedious to the farmer. To reduce interview time and potential con-
fusion that might arise if detailed data were to be collected on every
parcel, one of the several parcels may be designated as the intensive
data parcel (IDP). Several parcels that are close together, have the
same type of irrigation, and are managed and worked as a single farm
unit may be treated as one IDP.
To designate the IDP of a farmer with several parcels, select a
parcel that meets the stratification criteria used. For example, if
dependability of water is a stratification criterion, one-third of the
IDP should be rainfed, one-third with intermediate level of water
supply dependability, and one-third with a very reliable water supply.
If farmers are selected at random, choose the farmer’s largest parcel
as the IDP. If the farmer is one on whose farm an experiment is
located, select the parcel where the experiment is located.

Location, type, and size of parcels

The first page of the questionnaire (Table 9) identifies the respondent,
and clearly identifies the number, size, and extent of his farmland.
It is also where the parcel from which intensive data will be obtained
is identified. Along with the questions in Table 9, a sketch map of

EXPERIMENTAL PROCEDURES 33
the farm is drawn. When the farm map is drawn, the size of, and
source of water for, each parcel is evident. In later stages of the
questionnaire, always refer to this parcel when asking questions
related to the IDP.
The interview is concerned with the farm units, so all parcels of
land managed by the respondent should be included in Table 9.
Areas where farms are highly fragmented may require more than
four lines for listing the parcels. If so, combine all the remaining
parcels in the last line. Be sure to identify the IDP. For convenience
in data processing, it is important to list the IDP first on the form.
• If rice was grown in the season of reference, list the variety. If
another crop was grown in some parcels, list the crop.
• Record the size of the area as accurately as the farmer can deter-
mine it. For example, if he says “1.0 hectare,” ask him if that is
the exact area. In some cases, it may be desirable to measure the
size of the farmer’s parcels or at least the IDP, especially if the
farmer is uncertain about size, or if there is reason to believe he
is not reporting parcel size accurately.
• If the parcel is irrigated, give the source of water. If it is unirrigated,
write “no” in the appropriate column.
• Under tenure, indicate the type of tenure. A subsequent section
has space for details of tenure arrangements.

Production and tenure

Some data on production and tenure (Table 10) are available from
farmers only after harvest, thus the indication PH (postharvest) for
questions asked after the harvest. In our experience, interviewing is
speeded if farmers are initially contacted during the crop season for
some information; a second interview is conducted after harvest to
get harvest data and to double-check other data.
Because farmers sometimes do not include the quantities paid to
harvesters or paid in kind for other purposes, these methods of
disposal are investigated separately. The separate output disposals
are added to obtain total production. Tenure arrangements and
sharing of costs and output are necessary for evaluating the effective
returns to tenants. It is sufficient to obtain cost-sharing data for a
few inputs that make up the most important costs.

Credit and income alternatives

Table 11 shows the questions used to determine the availability of
credit. It specifies Masagana 99 because that is the institutional

EXPERIMENTAL PROCEDURES 35
source of credit for rice production in the Philippines. The credit
questions should be adapted to fit the credit conditions prevailing
in particular sites.
Questions 4, 5, and 6 give some idea of the alternative income
possibilities available to the respondent. If he has attractive alter-
native opportunities for income, he may neglect opportunities for
rice intensification.
To use the data from questions 4 and 5, the researcher should have
some idea of the costs involved in producing other crops and live-
stock, so they may be deducted to get an indication of nonrice
earnings. It may be necessary to supplement these questions with a
few on costs if alternative income sources are significant.

Prices
Price data are obtained through the questions in Table 12. All major
purchased inputs or services are listed. It is important to specify the
units for which the price is being obtained because some inputs are
sold by the kilogram, some by volume measure, and others on an
area basis. If this information is omitted, problems arise when the
data are summarized. Prices of insecticides are included in the section
on insect control.

Fertilizer use, constraints

Table 13 lists the questions asked to determine fertilizer constraints.
Question 1 is asked during the season when the farmer is first con-
tacted. Question 2 is asked during the postharvest interview. The
timing information is double-checked during the second interview.
The types of materials should be modified to include those available
at the study sites,

Insect control
The purpose of the insect control questions in Table 14 is to deter-
mine whether the farmer perceived insect problems, how well they
understood the problems, and what action they took. The timing
questions are designed to lead the farmer back over the season to try
and make sure all applications are included. Question 4 provides price
information and separates the total amount of insecticide purchased
from the amount used on the intensive data parcel.

38 CONSTRAINTS METHODOLOGY HANDBOOK

EXPERIMENTAL PROCEDURES 39
40 CONSTRAINTS METHODOLOGY HANDBOOK
EXPERIMENTAL PROCEDURES 41
Weed control
The questions in Table 15 establish the farmer’s weed control prac-
tices, the costs of weed control, and the extent to which the farmer
perceives weeds as a problem.

Caution
The questions on use of fertilizer, insect control, and weed control
are keyed to the three major factors used in the constraints experi-

42 CONSTRAINTS METHODOLOGY HANDBOOK

EXPERIMENTAL PROCEDURES 43
ments in the Philippines. If some other factor, for example variety,
is included as an experimental factor in a particular site, questions
about that factor should be included in the questionnaire.
If sampling is properly done, the data on levels of inputs applied
by the surveyed farmers can be used to estimate the levels used by
a representative set of farmers. The levels can then be compared with
the levels used by farmers on whose fields the experiments are
located, to determine if the latter are representative.

Other constraints
The questions in Table 16 provide some idea of the extent of water
problems our Philippine respondents had on their farms. The open-
ended questions on factors keeping yield low are used to determine
the farmer's perception of problems in his rice fields.

44 CONSTRAINTS METHODOLOGY HANDBOOK

Analysis of constraints data

The four major types of constraints data analysis usually performed

are discussed in this chapter:
• estimation of yield gap and contribution of the test factors,
• comparison of yield contribution, costs, and returns to determine
relative profitability of the test factors,
• comparison of yields, costs, and returns among the different pack-
ages tested, and
• tabular analysis of farmers’ perception of constraints to input use
and yield, and other socioeconomic phenomena that may explain
farmers’ behavior.

Conventional statistical analysis of variance should be performed on ESTIMATION

the yield data from each of the complete and minifactorial experi- OF YIELD GAP AND
ments to determine whether effects are significant. The procedures CONTRIBUTION
can be found in standard statistical textbooks. The methods we dis- OF THE
cuss here focus only on the determination of the yield gap and its TEST FACTORS
components. Yield constraints experiments (complete factorial, mini-
factorial, and supplemental trials) involving three test factors –
fertilizer (F), weed control (W), and insect control (I) – illustrate
the methods. The treatments tested are those described in Table 2.
Mean yield data from complete factorial trials on five farms are
shown in Table 17.

Yield gap determination

Data from the three types of experiments – supplemental, mini-
factorial, and complete factorial – are used to determine the yield
gap. For each farm, compute the yield gap as the difference between
the yield obtained with all test factors at the high level (i.e., potential
farm yield) and that obtained with all test factors at the farmer’s
level (i.e., actual farm yield).
Based on the treatment description in Table 2 and using Yi to
represent the yield of treatment i, the formula is:

Yield gap = Y8 - Y1
 From data in Table 17, the yield gap for farm 1 is computed as
8,346 - 5,276 = 3,070 kg/ha.
The average of yield gaps over all farms in a study area – those
with complete factorial, minifactorial, and supplemental trials –
represents the estimated yield gap for the area.

Calculating the contribution of test factors

The contribution of test factors to the yield gap is determined only
from the factorial component of the complete factorial trial or from
the minifactorial trial.
Two types of contributions are considered: individual contribution
(of each test factor singly) and joint contribution (of two or more
test factors together). When interactions among test factors are not
appreciable, only information on individual contributions is needed
because in that case, the joint contribution of any two factors is the
sum of the individual contributions of the two. On the other hand,
with the presence of interactions among test factors, both individual
and joint contributions must be determined.
Interactions among test factors can be determined only from the
factorial component of the complete factorial trial. For each farm
with a complete factorial trial, the significance of each type of inter-
action is determined from the analysis of variance performed on the
yield data.
Absence of interactions among test factors. In the absence of
interactions, only individual contributions of test factors need be
determined.

Table 17. Grain yield in constraints experiments (average of two replications per farm).

Yield (kg/ha) by farm

Plot no. Treatment Average yield
1 2 3 4 5 (kg/ha)
1 F1 W 1 I 1 5276 4840 4765 1890 4404 4235
2 F1 W 5 I 1 5509 5259 5023 2096 4686 4515
3 F5 W 1 I1 6332 5313 5863 2731 5505 5149
4 F5 W 5 I 1 6789 5533 6113 2733 5597 5353
5 F1 W 1 l 5 6125 5082 5666 2373 5088 4867
6 F1 W 5 I 5 6307 5144 5869 2592 5353 5053
7 F5 W 1 I 5 7756 6082 6958 3396 6216 6082
8 F5 W 5 I 5 8346 5973 6975 3625 6369 6257
9 F2 W 2 I 2 5083 4205 5762 1903 4591 4189
10 F3 W 3 l 3 5562 4680 5089 2173 4731 4447
11 F4 W 4 I 4 6907 4938 6377 2814 5663 5340

46 CONSTRAINTS METHODOLOGY HANDBOOK

 From complete factorial trials, the individual contribution of a
test factor is calculated as the difference between the average yield
over all treatments with that test factor at the farmer’s level and the
average yield over all treatments with that test factor at the high
level.
Based on the treatment description in Table 2, the formulas are:
Y 2 + Y5 + Y6 + Y8 Yl + Y3 + Y4 + Y7
Contribution of variety = –
4 4
Y 3 + Y5 + Y7 + Y8 Y + Y2 + Y5 + Y6
Contribution of fertilizer = – 1
4 4
Y 4 + Y6 + Y7 + Y8 Y1 + Y2 + Y3 + Y5
Contribution of insect control = –
4 4

With the use of data of farm 1 from Table 17, the individual con-
tribution of test factors is computed:

Contribution 6332+6789+7756+8346 5276+5509+6125+6307

= –
of fertilizer 4 4

= 1501

Contribution = 5509+6789+6307+8346 - 5276+6332+6125+7756

of weed 4 4
control
= 365

Contribution 6125+6307+7756+8346 5276+5509+6332+6789

of insect = -
control 4 4

= 1157

Yield gap and factor contributions for the five farms of Table 17
are shown in Table 18. The difference between the yield gap and the
sum of the three individual contributions represents the residual
term.
From minifactorial trials, the individual contribution of a test
factor is computed as the difference between yield of the treatment

ANALYSIS OF CONSTRAINTS DATA 47

 with all test factors at the high level and yield of the treatment with
that test factor at the farmer’s level and all other test factors at the
high level.
Based on the treatment description in Table 2, the formulas are:

Contribution of variety = Y8 — Y7

Contribution of fertilizer = Y8 — Y6

Contribution of insect control = Y8 — Y5

Presence of interactions among test factors. With the presence of

interactions, both individual and joint contributions of test factors
should be determined. The earlier formulas for calculating the indi-
vidual contribution in the case of no interactions are not valid when
interactions exist.
This is illustrated by considering the formula Y8 — Y7 used for com-
puting the contribution of variety in the case of no interactions,
and using data from minifactorial trials (see treatment description,
Table 2). The difference represents the effect of variety at the high
levels of other test factors, namely fertilizer and insect control. When
interactions are present, such an effect is not the same as that at the
farmer’s levels or at any intermediate levels of the other test factors.
It is clear that when interactions are present, Y8 — Y7 no longer
measures the individual contribution of variety (nor the effect of
variety alone) to the yield gap.
The presence of interactions can be determined only from the
complete factorial trial. Because of that a judgment must be made,

Table 16. Yield gap and yield contribution of each factor in constraints experiments on five farms.

Yield (kg/ha) Yield contribution (kg/ha)

Farm High Farmer‘s Residual
Gap Fertilizer Weed Insect
inputs inputs control control
1 8346 5276 3070 1501 365 1157 47
2 5973 4840 1133 614 148 334 7
3 6975 5765 2210 1146 182 926 –44
4 3625 1890 1735 883 162 634 56
5 6369 4404 1965 1039 198 634 91

Average 6257 4235 2022 1042 211 737 31

Average (%) 52 11 37 –

48 CONSTRAINTS METHODOLOGY HANDBOOK

on the basis of the analysis of all farms with complete factorial trials,
on whether interactions are appreciable in the area. If the interactions
are judged appreciable, then the minifactorial trials can be used to
measure only the joint contributions of test factors with the use of
the following formulas (based on treatment description in tables):

• Joint contributions:

Contribution of variety and fertilizer = Y5 – Y1

Contribution of variety and insect control = Y6 – Y1

Contribution of fertilizer and insect control = Y7 – Y1

The complete factorial trials can be used to measure both the joint
contributions (as above) and the individual contributions through
the following formulas:

• Individual contributions:
Contribution of variety = Y2 – Y1

Contribution of fertilizer = Y3 – Y1

Contribution of insect control = Y4 – Y1

To illustrate the case where an interaction effect is present, mean

yield data of a trial involving a highly significant insect control and
fertilizer interaction are shown in Table 19. The nature of the inter-
action is shown in Table 20. Yield increase from the improved insect
control was larger with the high fertilizer rate than with the farmer’s
fertilizer rate (i.e. yield increases of 0.94 t/ha and 0.17 t/ha, respect-
ively). In the same manner, yield increase from the high fertilizer
rate was larger with improved insect control than with the farmer’s
own insect control (i.e., yield increases of 1.11 t/ha and 0.34 t/ha,
respectively).

ANALYSIS OF CONSTRAINTS DATA 49

Table 19. Rice yields under varying combinations of input use and tested in a farmer's field
in Laguna Philippines, 1976 dry season.

Yielda (t/ha)
Fertilizer Weed control
level level Farmer's Improved
insect control insect control

Farmer's Farmer's 3.69 3.76

High 3.40 3.66

High Farmer's 3.80 4.68

High 3.97 4.95

a Source: Gomez 1977. Data are average of two replications.

The calculation of the yield gap, individual contributions, and

joint contributions based on data in Table 19 follows:

t/ha

Actual farm yield 3.69

Potential farm yield 4.95
Yield gap 1.26
Individual contributions:
Insect control 0.07
Fertilizer 0.11
Weed control -0.29
Joint contributions:
Insect control and fertilizer 0.99
Insect control and weed control -0.03
Fertilizer and weed control 0.28

It is clear that when either the high level of insect control or the
high level of fertilizer is applied alone, no significant yield increase

Table 20. Interaction between insect control and fertilizer rate based on data in Table 19.

Yield (t/ha)
Fertilizer
Farmer's Improved Difference
level
insect control insect control

Farmer's 3.54 3.71 0.17

High 3.88 4.82 0.94

Difference 0.34 1.11

50 CONSTRAINTS METHODOLOGY HANDBOOK

is obtained. But when insect control and fertilizer are applied toge-
ther, a yield increase of about 1 t/ha is observed. This means that
the sum of individual contributions of these two test factors is not
the same as their joint contribution. In such cases, both individual
and joint contributions must be determined to obtain the complete
information.
Table 21 presents results on contributions of test factors (when
interactions among test factors are present) computed from a total
of eight farms — four with complete factorial and four with mini-
factorial trials. Note that while estimates of both individual and joint
contributions are obtained from complete factorial trials, only the
estimate of joint contributions is possible from minifactorial trials.
Adjustment of contributions. The contributions of test factors to
the yield gap are estimated only from two types of trials — the
complete factorial and minifactorial trials — whereas the yield gap
is estimated from three types of trials — the complete factorial, the
minifactorial, and the supplemental trials. Hence, the estimates of
contributions must be adjusted toward the estimate of the yield gap.
The adjusted contribution of each test factor is computed as:
(unadjusted contribution) (yield gap A)
Adjusted contribution = ,
(yield gap B)
where yield gap A is the gap estimated from all trials and yield gap B
is the gap estimated only from the complete factorial and mini-
factorial trials.

Table 21. Yield gap and contributions (individual and joint) of three test factors, estimated from complete factorial and
minifactorial trials.

Yield (t/ha) Contribution (t/ha)

Farm no. a Farmer's High Gap Insect Fertilizer Weed I + F I +W F +W

inputs inputs control (F) control
(I) (W)

1 4.86 9.08 4.22 1.08 0.56 0.63 3.83 1.31 0.84

2 4.32 6.24 1.92 0.56 0.72 -0.04 1.72 0.57 0.68
3 3.60 6.18 2.58 0.22 0.91 -0.29 2.91 1.20 1.70
4 3.69 4.95 1.26 0.07 0.11 -0.30 1.00 -0.30 0.28
5 3.26 4.71 1.45 – – – 1.07 0.21 0.61
6 4.85 6.20 1.35 – – – 1.14 0.87 0.74
7 6.11 8.24 2.13 – – – 2.10 0.90 2.24
8 6.27 8.49 2.21 – – – 1.57 0.38 1.51
Average 4.62 6.76 2.14 0.48 0.59 0.00 1.92 0.64 1.06
a Farms 1–4 have complete factorial trials and farms 5–8 have minifactorial trials.

ANALYSIS OF CONSTRAINTS DATA 51

ECONOMIC A basic economic hypothesis is that farmers will not adopt tech-
ANALYSIS nology without an economic incentive, which means the new
OF A CONSTRAINTS technology must carry with it some advantage over farmers’ existing
EXPERIMENT technology. The measure that most accurately reflects incentive is
the return to family-owned resources. Therefore, the returns to
family-owned resources with the actual technology and the tech-
nology necessary to get potential yields are calculated. Price data
from the survey and yield data from the experiments are used to
budget the returns to family-owned resources.

One difficulty in broadly interpreting the profitability of physic-

ally effective technology is the individual variation in economic attri-
butes of different farmers. Farms are large and small, owner-operated
and operated under various forms of rental, some have large and
others have small labor forces. There is no a priori reason for the
physical effectiveness of technology to systematically vary among
farms with different economic attributes, but the returns to family
resources will differ depending on how large a share of the total
resources is contributed by the farm family.
The survey is used to identify and describe the economic attributes
of various types of farms in the area studied. The survey is also used
to determine whether the various types of farms have the resources,
or can obtain the resources, needed to apply the high potential-yield
technology.
The economic analysis of the experiments can provide a measure
of the value of additional yield from high inputs. To determine net
benefits, one must know the costs of each level of inputs. When the
costs of the input packages and of the separate factors are known,
the difference in net returns of various treatments can be calculated.
Costs that are identical for all packages are not included in the
economic analysis because they do not affect the comparison of the
relative profitability of the packages. Hence, the cost of purchasing
seed is not included if variety and seed source are fixed factors. Only
the costs associated with the test factors are included in the econ-
omic analysis, i.e., partial budgeting is used.

Costs of input packages

The following procedure will give the costs of each treatment as
relevant for owner-operators (Table 22).
1. List the input used in each treatment on a per hectare basis (only
two are illustrated). These data come from the records of the

52 CONSTRAINTS METHODOLOGY HANDBOOK

 experimenter (Table 6).
2. Use the average prices paid by farmers, as summarized from data
obtained from Table 13, to calculate the cost of each input on a
per-hectare basis. If the labor required to apply the inputs is
normally hired (Table 12), such as that for hand weeding, then its
cost should be calculated as part of the cost of the package.
3. Add the cost of interest on the purchased inputs at the rate pre-
vailing in the area.
4. Total the costs for all inputs for each package, including the
farmer’s practices.
A share-tenant’s costs are shown in the middle column of Table 22.
They vary because under some conditions part of the tenant’s costs
may be shared by the landlord. The researcher must determine which
costs are shared by tenant and landlord, and the proportion in which
they are shared (information from Table 10). In the example, it is

Table. 22. Costs of two treatments used in constraints experiment.

Cost/ha
Treatment a lnputs/ha
Owner-operator Tenant b Landlord

9 30 kg N as urea = 66.6 kg urea = 1.3 bags 100 50 50

1 hand weeding or rotary weeding 125 125 0
2 foliar sprays on field 150 75 75
Interest on purchased inputs (1%/mo) 22 15 7
397 265 132

8 • Fertilizer (720) (360) (360)

60 kg P from 16-20-0 = 300 kg 16-20-0
=6 bags 360 180 180
120 kg N, 48 kg from 300 kg 16-20-0
72 kg N from urea = 160 kg urea = 3.2 bags 240 120 180
60 kg K from muriate = 100 kg muriate = 2 bags 120 60 60

• Weed control (295) (210) (85)

Benthiocarb/2,4-D (G) 170 85 85
1 hand weeding 125 125 0

• Insect control
4 foliar insecticide sprays, seedbed 20 10 10
1 granular insecticide treatment, seedbed 20 10 10
5 foliar sprays, field 375 188 187
3 granular sprays, field 510 255 255
Interest on purchased inputs (1%/mo) 116 62 54
2056 1095 961
a See Tables 2 and 6.
b Sharing of various types of costs must be determined from the survey. See Table 10.

ANALYSIS OF CONSTRAINTS DATA 53

 assumed that the landlord pays half the cost of purchased inputs and
the tenant pays half plus any labor costs.

Partial budgeting of management packages

To compare the economics of the alternative management packages
and of the farmer's practices, a partial budgeting approach is used.
For each farm with experiments, an analysis is made following the
approach shown in Table 23. The results are averaged across farms.
If experiments were conducted on different types of farms, say irri-
gated and rainfed, and the results differ widely, averages are calculated
for each class separately.
Each line in Table 23 is calculated as follows:
1. List the yield corresponding to each package (obtained from
Table 8).
2. Multiply the yield by price received by the farmer for his rice
variety (Table 12) to get value of output.
3. List costs that are proportional to output such as harvesting costs
(Table 12). They affect the profitability of high yields and must
be accounted for.
4. List the cost of each input package (from calculation like that in
Table 22) and subtract the proportional costs and input costs to
get the net return to each package.
5. Calculate the added output value, added cost, and added net
return over farmer's practices by subtracting value, cost, and net
return of M1 from the corresponding values for M2-M5.
6. Compute benefit:cost ratio of added output by dividing values in
line 6 by values in line 7. If line 7 is negative or zero, do not
compute.

=
Table 23. Budget summary for average of owner-operators (P/ha unless noted otherwise).

M1 M2 M3 M4 M5

1. Yield (kg/ha) 1567 1880 2185 2427 2158

2. Value of output a 1567 1880 2185 2427 2158
3. Proportional costs (harvest) 261 313 364 404 360
4. Cost of input package 400 397 842 1359 2056
5. Return net of package and proportional costs 906 1170 979 664 -258
6. Added output value over farmers' practices 313 618 880 591
7. Added cost over farmers' practices - 3 442 959 1656
8. Added net return over farmers' practices 264 73 -242 -1164
9. Benefit:cost ratio of added output +b 1.4 0.9 0.4
=
a Price in this case is P1/kg.
b Since cost is less than farmer's practices, rate of return is higher.

54 CONSTRAINTS METHODOLOGY HANDBOOK

Table. 24. Economics of the contribution of separate inputs.

=
Package cost (P/ha) Maximum input Ievel compared to farmer's

Farmer's Maximum Increased Increased Increased Increased B:C ratio

Input
inputs inputs yield input cost output valuea net return of increased
(M1) (M5) (kg/ha) =
(P/ha) =
(P/ha) =
(P/ha) cost b

Factor 1 435 720 1002 285 1002 71 7 3.5

Factor 2 53 295 254 242 254 12 1.1
Factor 3 238 925 741 687 741 54 1.1

a Price of palay =P1/kg.

b increased output value + increased input cost.

Farmers who rent their land on a fixed rate basis have the same
incentives as owners do, so the above procedure adequately reflects
their returns. Share-tenants are different. In calculating the returns
of high input use to share-tenants, the procedure must be modified:
1. Complete the first four steps and calculate the remaining value net
of proportional costs. If landlords pay part of the costs, be sure to
reflect this in the data. As illustrated in Table 22, calculate the
tenants' costs and enter them in Table 23.
2. Subtract the appropriate share of rent to the landlord (Table 10).
3. Follow steps 5 and 6.

Economics of separate inputs

While it is not likely that the maximum-yield level of inputs will be
profitable, it is still useful to know the relative benefits per unit of
cost incurred for each input. This gives some indication of which
inputs are most in need of change. Table 24 shows calculations for
owner-operators. To determine the economics of the separate inputs
for the average of a set of farms, the following steps are used:
1. The input cost for each factor is taken (Table 22) for the farmer's
(1) and the high (5) levels of inputs.
2. The yield with farmer's level and the high level of each input is
taken (Table 17). The difference gives the increased yield. In-
creased costs, output value, and net return are calculated by
subtracting the cost, output value, and net returns of level 1 from
the corresponding values of level 5.
3. The benefit : cost (B:C) ratio is found by dividing increased output
value by increased input cost.
In the example, all three factors give an increased net return, but
only factor 1 has an attractive B:C ratio. While the B:C ratio might

ANALYSIS OF CONSTRAINTS DATA 55

 be higher at lower input rates for factors 2 and 3, factor 1 is clearly
more beneficial than the two other inputs. Factors 2 and 3 contribute
254 and 741 kg/ha to yield, but are not economically attractive at
the high level. Thus, not only does factor 1 contribute 50% of the
yield gap; it is the most profitable input level to use.

SOCIOECONOMIC One of the major purposes of the survey aspect of the research is to
CONSTRAINTS understand why farmers are not using the inputs needed for high
ANALYSIS yields, or are not using them at the recommended rates. This section
demonstrates some ways in which parts of the survey may be used
to determine that information. Farmers' perception of the important
constraints is valuable information for understanding farmers' actions
(Table 25). If a particular factor is not viewed as a constraint, one
would not expect farmers to take action to overcome it.

Representativeness of experimental farmers

It is important to know whether the farms with experiments are
similar to or different from the average farms in the area being
studied because the researcher would like to generalize his results.
A comparison of the type of farm and level of input use and reported
yield of the entire IDP of farms with experiments and farms without
experiments is the best way to determine this.
Table 26 shows data assembled in this way. Tenure of the two
groups is similar. Irrigation status might also be shown. The average
and standard deviation of the farmers' levels of the three test factors

Table 25. Yield constraints perceived by 150 surveyed farmers in three areas of the Philippines, 1975-76.

Percentage of total sample

Reported constraint Wet season 1975 Dry season 1976

Nueva Ecija Laguna Camarines Sur Nueva Ecija Laguna Camarines Sur

Lack of water 7.5 1.9 29.8 19.1 26.6 45.1

Excessive wind, rain,
flood (typhoon) 16.8 27.4 13.4 36.7 5.0 8.0
Too little fertilizer 17.6 0.0 20.8 13.9 5.0 16.1
Insects 11.7 15.6 8.9 4.4 3.3 17.7
Diseases 25.2 7.8 17.9 5.8 8.3 1.6
Rats 7.5 27.4 0.0 7.3 21.6 3.2
Weeds 1.7 0.0 0.0 5.1 10.0 4.8
Others 12.0 19.9 9.2 7.7 20.2 3.5

56 CONSTRAINTS METHODOLOGY HANDBOOK

Table 26. Levels of test inputs used and yields reported by farmers with constraints experiments and by other surveyed
farmers, Nueva Ecija, 1976-77.

Weed Insect
Tenure status (no.)
Fertilizer Control control Yield
Farmer
group Share- Others (kg NPK/ha) =
(P/ha) =
(P/ha) (t/ha)
tenants
x s.d. x s.d. x s.d. x s.d.

Experiments 15 3 82 34 71 41 127 51 4.0 1.1

Others 35 7 87 26 47 25 68 44 3.8 0.7

included in the experiment are calculated. The levels can be tested

statistically to determine whether the differences are significant or
not. If the two groups are substantially different, the researcher must
be cautious in drawing conclusions from the experimental farms to
the other farms.

Reasons for not using fertilizer

In the survey, farmers are asked what level of each test factor they
apply on their farms and why they do not use higher levels (Table 13).
The replies obtained from surveys in Nueva Ecija, Philippines, are in
Table 27.
Two dominant reasons were given — lack of capital and the idea
that they had applied the level recommended by the technician, or
enough. For a better understanding of the situation, the levels of
Table 27. Proportion of farmers a surveyed giving reasons for not using more fertilizer, Nueva Ecija, Philippines, 1976-77.

Farmers surveyed

Wet season Dry season

Reason given
1975 1976 1976 1977

No. % No. % No. % No. %

Lack of capital 43 61 31 62 53 75 26 43
Plants may lodge (soil still fertile to
increase vegetative growth) 3 4 7 14 0 0 0 0
Fear of risk (already damaged by
tungro) 0 0 6 12 0 0 0 0
Used enough (lime recommended by
technician) 14 20 5 10 16 23 28 47
Variety used doesn't need more
fertilizer 1 2 1 2 0 0 0 0
No response 2 3 0 0 1 1 5 8
Others (no available fertilizer) 7 10 0 0 1 1 1 2

Total 70 100 50 100 71 100 80 100

a Seventy farmers were surveyed in 1975 wet season, 60 in 1976 wet, 71 in 1976 dry and 60 in 1977 dry.

ANALYSIS OF CONSTRAINTS DATA 57

 inputs used were tabulated by reasons as shown in Table 28. Those
reporting lack of capital generally used lower input levels than those
reporting they had used enough. Comparison of input levels and
yields of the two groups with the levels shown to be profitable in
the experiments shows, in the case illustrated, that those claiming
to have used enough actually did use more than the optimal M2 level
of fertilizer. In the dry season they used somewhat less than the
optimal M2 level.
Regression analysis can also be used to help understand the
important factors affecting input use. Table 29 shows the results of
one such analysis in which a number of variables constructed from
information gathered in the survey were used to explain fertilizer
use. Irrigation, modern varieties, Masagana 99 membership, share-
tenancy and traditional beliefs all were related to input use, while
past yield variability, fertilizer, experience, education, and off-farm
work were not. Use of other input factors may be analyzed in a
similar way.

Table 28. Average levels of fertilizer and input expenditures of surveyed farmers who used enough fertilizer and those who
lacked capital, compared with intermediate and maximum experimental levels.

Fertilizer Insect Weed control

Yield Fertilizer level (kg/ha)
Reason given cost control cost cost
(t/ha) N P K
= /ha)
(P =/ha)
(P =/ha)
(P

Wet season
Used enough 1.9 64 28 0 378 66 139
Lacked capital 1.3 39 18 1 279 121 89
M2 experiment level 3.6 50 20 10 263 203 150

Dry season
Used enough 3.0 88 44 8 547 81 57
Lacked capital 2.6 55 28 4 351 43 19
M2 experiment level 6.5 150 40 30 798 742 149

58 CONSTRAINTS METHODOLOGY HANDBOOK

Table 29. Regression equation explaining use of fertilizer (kg/ha) by 160 surveyed farmers,
Central Luzon, Philippines, 1974 wet season.

Independent variable Regression coefficients Standard error

• Environmental
Modern varieties 0.23*** 0.08
Irrigation 15.76*** 6.31
Riska 6.62 8.77

• Institutional
Masagana 99 membership 37.19*** 7.01
Share-tenancy b -10.54* 6.82
Fertilizer availability c 0.55 2.22

• Socioeconomic
Experience with fertilizer d 0.58 0.55
Traditional beliefs e 0.78* 0.52
Educational f -1.26 1.19
Farming experience -0.12 0.25
Off-farm workdays g 0.07 0.10
Farm size h -2.45 2.18

Constant 41.33

Source: R. H. Bernsten, 1978. Primary and secondary constraints to higher farm yields in
Central Luzon, Philippines. Paper presented at the Conf. on Farm Level Rice Yield
Const., IRRI, Los Baños, Philippines, 24-26 April.
R2 = 0.35 R -2 = 0.30 F = 6.57
a Risk measured by the yield coefficient of variation.

where sy = variance of previous 4 years' yield and Y = mean yield for previous 4 years.
Based on reported wet season yields, 1971-74.
b Dummy variable.
c Number of years since farmer first used fertilizer.
d Farmer's perception as to availability of fertilizer. Numerical value of variable ranges from
1 to 5, with 1 indicating "none available" and 5 indicating"plenty available."
'Numerical value of variables ranged from 1 to 5, with 5 indicating "strongly agree" and 1
indicating "strongly disagree," with 3 Mating "never heard."
f Number of years of formal education.
g Number of days employed off-farm.
h Hectarage of all parcels farmed.

*Significant at 10% level = (t = 1.282).

**Significant at 5% level = (t = 1.645).
***Significant at 1% level (t = 2.282).

ANALYSIS OF CONSTRAINTS DATA 59

References cited

Bernsten, R. H. 1978. Primary and secondary constraints to higher

farm yields in Central Luzon, Philippines. Paper presented at the
Conference on Farm Level Rice Yield Constraints, IRRI, Los
Baños, Philippines. 24-26 April.
De Datta, S. K. 1974. Use of management practice packages to eval-
uate farm yield constraints. Paper presented at the International
Rice Research Conference, 22-25 April. IRRI, Los Baños,
Philippines.
De Datta, S. K., W. N. Obcemea, W. P. Abilay, Jr., M. T. Villa, B. S.
Cia, and A. K. Chatterjee. 1976. Identifying farm yield constraints
in tropical rice using a management package concept. Paper pre-
sented at the 7th Annual conference of the Crop Science Society
of the Philippines, Davao City, 10-12May.
Gomez, K. A. 1972. Techniques for field experiments with rice.
International Rice Research Institute, Los Baños, Philippines. 46 p.
Gomez, K. A. 1977. On-farm assessment of yield constraints: meth-
odological problems. Pages 1-16 in International Rice Research
Institute. Constraints to high yields on Asian rice farms: an interim
report. Los Baños, Philippines.
Gomez, K. A., D. Torres, and E. Go. 1973. Quantification of factors
limiting rice yields in farmers’ fields. Paper presented at a Saturday
seminar, 24 November. International Rice Research Institute, Los
Baños, Philippines.
Gomez, K. A., and A. A. Gomez. 1976. Statistical procedures for
agricultural research, with emphasis on rice. International Rice
Research Institute, Los Baños, Philipines. 294 p.
International Rice Research Institute (IRRI). 1977. Constraints to
high yields on Asian rice farms: an interim report. Los Baños,
Philippines. 235 p.
International Rice Research Institute (IRRI). 1978. Annual report
for 1977. Los Baños, Philippines. (in press)