• TABLE OF CONTENTS
HIDE
 Front Cover
 Title Page
 Table of Contents
 List of Tables
 List of Figures
 Objectives
 Basic considerations
 Key terms
 Summary of the steps in analysis...
 Introduction
 Response of treatments to different...
 Steps in the analysis and interpretation...
 Design of on-farm trials
 Reference
 Appendix






Group Title: Staff paper - Food and Resource Economics Department, University of Florida - SP95-4R (revised)
Title: Steps in the analysis and interpretation of on-farm research-extension data based on adaptability analysis (AA)
CITATION PAGE IMAGE ZOOMABLE PAGE TEXT
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/UF00054817/00001
 Material Information
Title: Steps in the analysis and interpretation of on-farm research-extension data based on adaptability analysis (AA) a training guide
Series Title: Staff paper
Physical Description: iii, 25, 18 p. : ill. ; 28 cm.
Language: English
Creator: Hildebrand, Peter E
Bastidas, Elena P
University of Florida -- Food and Resource Economics Dept
Publisher: Food and Resource Economics Dept., Institute of Food and Agricultural Sciences, University of Florida
Place of Publication: Gainesville Fla
Publication Date: [1995]
 Subjects
Subject: Agriculture -- Research -- On-farm   ( lcsh )
Field experiments   ( lcsh )
Agricultural innovations -- Research   ( lcsh )
Agricultural systems -- Research   ( lcsh )
Agricultural extension work   ( lcsh )
Genre: government publication (state, provincial, terriorial, dependent)   ( marcgt )
bibliography   ( marcgt )
non-fiction   ( marcgt )
 Notes
Bibliography: Includes bibliographical references (p. 24).
Statement of Responsibility: Peter E. Hildebrand and Elena Bastidas.
General Note: "May 1995."
Funding: Staff paper (University of Florida. Food and Resource Economics Dept.) ;
 Record Information
Bibliographic ID: UF00054817
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved, Board of Trustees of the University of Florida
Resource Identifier: aleph - 002026389
oclc - 33056549
notis - AKL3969

Table of Contents
    Front Cover
        Front Cover
    Title Page
        Title Page
    Table of Contents
        Page i
    List of Tables
        Page ii
    List of Figures
        Page iii
    Objectives
        Page 1
    Basic considerations
        Page 1
    Key terms
        Page 2
    Summary of the steps in analysis and interpretation of on-farm research-extension data
        Page 3
    Introduction
        Page 4
    Response of treatments to different environments
        Page 4
    Steps in the analysis and interpretation of on-farm research-extention data
        Page 4
        Calculate a measure of environment: Environmental index, EI
            Page 4
            Page 5
        Relate treatment response to enviroment
            Page 6
            Page 7
            Page 8
        Access interaction of treatment with enviroment
            Page 9
        Characterize the enviroments
            Page 10
        Define recommendation domains
            Page 11
            Page 12
            Page 13
            Page 14
        Compare results repeating steps 2 through 5 using alternative evaluation criteria
            Page 15
            Page 16
            Page 17
        Create extension recommendations for each of the recommendation domains
            Page 18
            Page 19
    Design of on-farm trials
        Page 20
        Nature of on-station and on-farm trials
            Page 20
        On-farm trials
            Page 20
        Designing on-farm trials for adaptability analysis
            Page 21
            Page 22
            Page 23
    Reference
        Page 24
    Appendix
        Page 25
        Page 26
        Page 27
        Page 28
        Page 29
        Page 30
        Page 31
        Page 32
        Page 33
        Page 34
        Page 35
        Page 36
        Page 37
        Page 38
        Page 39
        Page 40
        Page 41
        Page 42
        Page 43
Full Text







Staff Paper Series


FOOD AND RESOURCE ECONOMICS DEPARTMENT

Institute of Food and Agricultural Sciences
University of Florida
Gainesville, Florida 32611


L--
-J


















STEPS IN THE ANALYSIS AND INTERPRETATION
OF ON-FARM RESEARCH-EXTENSION DATA
BASED ON
ADAPTABILITY ANALYSIS (AA):
A TRAINING GUIDE

PETER E. HILDEBRAND1
AND
ELENA BASTIDAS


STAFF PAPER SP95-4R
(Revised)


MAY 1995


Staff papers are circulated without formal review by the Food and Resource Economics
Department. Contents are the sole responsibility of the author.
















SProfessor, Food and Resource Economics Department, Institute of Food and Agricultural Sciences, and Graduate
Assistant, Agricultural Education and Communication Department, University of Florida, Gainesville, Florida
32611-0240, USA.















TABLE OF CONTENTS



OBJECTIVES ................ ............................................ 1

BASIC CONSIDERATIONS .................................................. 1

KEY TERMS .......................................................... 2

SUMMARY OF THE STEPS IN ANALYSIS AND INTERPRETATION
OF ON-FARM RESEARCH-EXTENSION DATA .......................... 3

INTRODUCTION ........................................................ 4

RESPONSE OF TREATMENTS TO DIFFERENT ENVIRONMENTS .................. 4

STEPS IN THE ANALYSIS AND INTERPRETATION OF ON-FARM RESEARCH-
EXTENSION DATA ................................................. 4
Calculate a measure of environments: Environmental Index, El ................... 4
Relate treatment response to environment .................................. 6
Assess interaction of treatments with environment .......................... 9
Characterize the environments .......................................... 10
Define recommendation domains ........................................ 11
Compare results repeating steps 2 through 5 using alternative evaluation criteria ..... 15
Create extension recommendations for each of the recommendation domains ........ 18

DESIGN OF ON-FARM TRIALS ....................... ........................ 20
Nature ofon-station and on-farm trials .................................... 20
On-farm trials......................................................20
Designing on-farm trials for Adaptability Analysis ............................ 21

REFERENCES ........................................................... 24
APPENDIX ............................................................. 25























List of Tables



Table 1. Response of maize (t/ha) to three soil amendments and
the farmers'practices from on-farm research results ................................... 5

Table 2. Response of maize (t/ha) to three soil amendments and
the farmers' practices from on-farm research results (sorted data) ........................ 5

Table 3. Environmental characteristics of the sites for the on-farm maize trials .................... 10

Table 4. Risk calculations comparing CM and TSP for land class PF, .......................... 13

Table 5. Response of maize (kg/$ cash cost) to three soil amendments
and the farmers' practices .................................................. 15

Table 6. Risk calculations comparing CM and TSP for land classes: PF2, SF2 and WL............... 17


















List of Figures


Figure 1. Observed response in t/ha of the TSP treatments to environment (E) ................... 6

Figure 2 a. Estimated linear response in t/ha of the TSP treatment to environment (El). ................ 7

Figure 2 b. Comparison of linear and quadratic response in t/ha
of the FP treatments to environment (EI) ............................................. 7

Figure 2 c. Comparison of linear and quadratic response in t/ha of the CM treatment
to environment (EI) .................................................... 8

Figure 3. Estimated responses in t/ha of the four treatments to environment (E) ................... 9

Figure 4. Relationship of soil pH to El of the sites for the on-farm maize trial .................... 11

Figure 5. Tentative recommendation domains and the technology recommended for maize,
based on environmental factors (land class) and the evaluation criterion (t/ha) ............ 12

Figure 6. Risk levels for treatments CM and TSP for Land class PF,. Evaluation criterion t/ha ....... 14

Figure 7. Estimated responses in kg/$ cash cost of the four treatments to environment (EI) ......... 16

Figure 8. Risk levels for the criterion kg/$ cash cost and the treatments CM and TSP ............. 17

Figure 9. Summary of the recommendation domains and the technology recommended ............. 18

Figure 10 Summary of the recommendation domains and the recommended technology based on
environmental characteristics (land type) and two evaluation criteria, t/ha and kg/$ ........ 19


iii









STEPS IN THE ANALYSIS AND INTERPRETATION
OF ON-FARM RESEARCH-EXTENSION DATA
BASED ON
ADAPTABILITY ANALYSIS (AA):
A TRAINING GUIDE

Peter E. Hildebrand
and
Elena P. Bastidas'

Modified Stability Analysis or MSA (Hildebrand, 1984) is a procedure for designing, analyzing and
interpreting on-farm trials conducted to assess new technologies and disseminate the resulting
recommendations. This farmer-participatory method now called Adaptability Analysis (AA) can
serve as a basis for an entire research-extension program (Hildebrand and Russell, draft). This guide
demonstrates some of the basic procedures for analyzing and interpreting data taken from an on-farm
research example for which an appropriate design was used (Singh, 1990).

OBJECTIVES:

The use of this guide should provide the user with:

1. The steps to follow to make technology recommendations for specific bio-physical and
socioeconomically-created environments and tailored to the various desires, needs and
constraints of specific farmers.

2. A basic understanding of the design requirements for on-farm research to make it amenable
to analysis by AA.

BASIC CONSIDERATIONS:

1. It is assumed that the user of this guide is basically familiar with AA.

2. While Analysis of Variance (ANOVA) can be used in conjunction with AA (Stroup et al.,
1993), it is not necessary. In this guide, ANOVA is not used.

3. The analyses can be done manually, with a calculator, or on a computer with SAS,
spreadsheet, or other type of analytical program. The degree of sophistication depends on
the user's capabilities and availability of equipment. The processes discussed here, with the
exceptions noted, are independent of the resources used to conduct the analysis.




1 Professor, Food and Resource Economics Department, Institute of Food and Agricultural Sciences, and
Graduate Assistant, Agricultural Education and Communication Department, University of Florida, Gainesville, Florida
32611-0240, USA.









STEPS IN THE ANALYSIS AND INTERPRETATION
OF ON-FARM RESEARCH-EXTENSION DATA
BASED ON
ADAPTABILITY ANALYSIS (AA):
A TRAINING GUIDE

Peter E. Hildebrand
and
Elena P. Bastidas'

Modified Stability Analysis or MSA (Hildebrand, 1984) is a procedure for designing, analyzing and
interpreting on-farm trials conducted to assess new technologies and disseminate the resulting
recommendations. This farmer-participatory method now called Adaptability Analysis (AA) can
serve as a basis for an entire research-extension program (Hildebrand and Russell, draft). This guide
demonstrates some of the basic procedures for analyzing and interpreting data taken from an on-farm
research example for which an appropriate design was used (Singh, 1990).

OBJECTIVES:

The use of this guide should provide the user with:

1. The steps to follow to make technology recommendations for specific bio-physical and
socioeconomically-created environments and tailored to the various desires, needs and
constraints of specific farmers.

2. A basic understanding of the design requirements for on-farm research to make it amenable
to analysis by AA.

BASIC CONSIDERATIONS:

1. It is assumed that the user of this guide is basically familiar with AA.

2. While Analysis of Variance (ANOVA) can be used in conjunction with AA (Stroup et al.,
1993), it is not necessary. In this guide, ANOVA is not used.

3. The analyses can be done manually, with a calculator, or on a computer with SAS,
spreadsheet, or other type of analytical program. The degree of sophistication depends on
the user's capabilities and availability of equipment. The processes discussed here, with the
exceptions noted, are independent of the resources used to conduct the analysis.




1 Professor, Food and Resource Economics Department, Institute of Food and Agricultural Sciences, and
Graduate Assistant, Agricultural Education and Communication Department, University of Florida, Gainesville, Florida
32611-0240, USA.











KEY TERMS

Confidence interval The probability that the selected evaluation criterion (for example, t/ha) will
fall within a certain range above and below the mean. It is calculated by the formula 2:
R (t. s/Vn)

Diffusion domains Informal and naturally occurring interpersonal communication networks for
diffusion of information on agricultural technology. Often specific to the commodity or
product involved.

Environment The natural biophysical and socioeconomically modified or created conditions existing
for plant or animal growth in the location of the trial. Includes influences of any differences
in management practices not part of treatments.

Environmental Index, El A convenient measure of the environment at the location of the trial. For
a specific environment, it is the average response of all the treatments for that environment,
usually based on physical yield per hectare.

Evaluation criterion The measure or measures used to compare the treatments in a trial. Can reflect
a researcher's concern (t/ha), for example, or a farmer's concerns and/or needs (kg/kg seed,
among many others).

Extension recommendations An extension message that includes the description of the technology,
as well as the specific environment and specific evaluation criterion for which it is being
recommended. Can be designed differently for use in specifically defined diffusion domains
within a single recommendation domain.

Recommendation domains The situations for which specific treatments or technologies will be
recommended. They are defined by a combination of environmental factors and evaluation
criteria.

Research domain The range of environments over which a trial is conducted. Ideally it represents
a wide set of biophysical and socioeconomic conditions.

Risk The probability (or percentage of time) that the selected evaluation criterion, such as t/ha, will
fall below a certain level.

Trial Usually refers to a set of treatments being evaluated over a range of environments. Can also
refer to the set of treatments at each environment. This double definition is seldom confusing
in context.

2-
2 Y= mean yield (or other measure of the criterion) for the treatment; t== the value from a "t" table for
probability level alpha; s = sample standard deviation; n = number of observations.











SUMMARY OF THE STEPS IN ANALYSIS AND INTERPRETATION
OF ON-FARM RESEARCH-EXTENSION DATA

To thoroughly analyze and interpret on-farm research data it is necessary that the design of the trial is
appropriate and amenable to this type of analysis. The design of on-farm research will be discussed later
in the guide. The steps are:

1. Calculate the environmental index, EI.

2. Relate treatment response to environment.

2 a. Plot all observations (data points) for each treatment against the environmental index on
a graph. This is an important step that is often overlooked but should not be. Ignorance
of the nature of the relationship can lead to inappropriate conclusions.

2 b. View the observations demonstrating treatment response to El and make an estimate of
the relationship of each treatment to environment. This can be done with linear or
curvilinear regression or by drawing a line.

2 c. Evaluate data quality.

3. Assess treatment interaction with environment comparing the responses of all treatments to EI.

4. Characterize the environments. Data for this step are often missing. Trial design and data
recording procedures should include adequate opportunity for collecting necessary data such as:
soil type, pH, planting date, etc.

5. Interpret results and define recommendation domains:

5 a. Define tentative recommendation domains.

5 b. Assess the risk associated with new technologies within the tentative recommendation
domains as compared with the farmers' own technology using the distribution of
confidence intervals.3

5 c. Define final recommendation domains. The persons involved in the on-farm trial process
(researchers, extension workers and farmers) are in the best position to use their
imagination, knowledge and judgement to interpret the results and to convert them into
useful recommendations (Andrew and Hildebrand, 1993).

6. Compare the results repeating steps 2 through 5 using alternative evaluation criteria.

7. Create extension recommendations for each of the recommendation domains and formulate
messages appropriate to each of the diffusion domains.


3 ANOVA can also be used within tentative recommendation domains to determine the significance of the
differences among or between the treatments. This analysis is not included in this guide.











ANALYSIS AND INTERPRETATION
OF ON-FARM RESEARCH EXTENSION DATA


INTRODUCTION

On-farm research extension trials can have various functions and can be managed by researchers,
extension workers and/or farmers (Hildebrand and Poey, 1985). The most appropriate design for
incorporating farmer participation is a simple (few treatments), non-replicated trial which has one to
three treatments to be compared with the farmers' own technologies. For purposes of demonstration,
a trial conducted in the Amazon basin of Brazil (Singh, 1990) with four treatments on eight
environments, without replication, will be used. Additional information on design will follow
discussion of the steps in the analysis.


RESPONSE OF TREATMENTS TO DIFFERENT ENVIRONMENTS

The term "environment" is used here rather than "farm" or "location" because on a single farm, or
even in a single field, more than one environment can exist for the production of the livestock or
cultivation of crops. Similar farmers do manage different environments differently just as different
farmers may manage similar environments differently. Furthermore, making technologies conform
to varying environments, rather than making the environments conform to the technology, reduces
the need for off-farm resources and is more in keeping with sustainable agriculture.


STEPS IN THE ANALYSIS AND INTERPRETATION OF ON-FARM RESEARCH-
EXTENSION DATA

Calculate a measure of environments: Environmental Index. El Step .

The factors that influence the environment for raising crops or livestock are many and complex and
are very difficult to assess. A convenient substitute measure of the quality of each environment where
a trial has been conducted is the average "yield" of all the treatments included when, and only when,
the same treatments are included in all the sampled environments. Thefirst step is to calculate this
index, EI, which provides an effective measure of the environmental differences in the research
domain represented by the range of Els.

The data that will be used to demonstrate this analysis are summarized in Table 1. To facilitate
further analyses, it is convenient to sort the data by descending (or ascending) values of this index.
The data in Table 2 are sorted in descending order with respect to the environmental index EI.











ANALYSIS AND INTERPRETATION
OF ON-FARM RESEARCH EXTENSION DATA


INTRODUCTION

On-farm research extension trials can have various functions and can be managed by researchers,
extension workers and/or farmers (Hildebrand and Poey, 1985). The most appropriate design for
incorporating farmer participation is a simple (few treatments), non-replicated trial which has one to
three treatments to be compared with the farmers' own technologies. For purposes of demonstration,
a trial conducted in the Amazon basin of Brazil (Singh, 1990) with four treatments on eight
environments, without replication, will be used. Additional information on design will follow
discussion of the steps in the analysis.


RESPONSE OF TREATMENTS TO DIFFERENT ENVIRONMENTS

The term "environment" is used here rather than "farm" or "location" because on a single farm, or
even in a single field, more than one environment can exist for the production of the livestock or
cultivation of crops. Similar farmers do manage different environments differently just as different
farmers may manage similar environments differently. Furthermore, making technologies conform
to varying environments, rather than making the environments conform to the technology, reduces
the need for off-farm resources and is more in keeping with sustainable agriculture.


STEPS IN THE ANALYSIS AND INTERPRETATION OF ON-FARM RESEARCH-
EXTENSION DATA

Calculate a measure of environments: Environmental Index. El Step .

The factors that influence the environment for raising crops or livestock are many and complex and
are very difficult to assess. A convenient substitute measure of the quality of each environment where
a trial has been conducted is the average "yield" of all the treatments included when, and only when,
the same treatments are included in all the sampled environments. Thefirst step is to calculate this
index, EI, which provides an effective measure of the environmental differences in the research
domain represented by the range of Els.

The data that will be used to demonstrate this analysis are summarized in Table 1. To facilitate
further analyses, it is convenient to sort the data by descending (or ascending) values of this index.
The data in Table 2 are sorted in descending order with respect to the environmental index EI.











ANALYSIS AND INTERPRETATION
OF ON-FARM RESEARCH EXTENSION DATA


INTRODUCTION

On-farm research extension trials can have various functions and can be managed by researchers,
extension workers and/or farmers (Hildebrand and Poey, 1985). The most appropriate design for
incorporating farmer participation is a simple (few treatments), non-replicated trial which has one to
three treatments to be compared with the farmers' own technologies. For purposes of demonstration,
a trial conducted in the Amazon basin of Brazil (Singh, 1990) with four treatments on eight
environments, without replication, will be used. Additional information on design will follow
discussion of the steps in the analysis.


RESPONSE OF TREATMENTS TO DIFFERENT ENVIRONMENTS

The term "environment" is used here rather than "farm" or "location" because on a single farm, or
even in a single field, more than one environment can exist for the production of the livestock or
cultivation of crops. Similar farmers do manage different environments differently just as different
farmers may manage similar environments differently. Furthermore, making technologies conform
to varying environments, rather than making the environments conform to the technology, reduces
the need for off-farm resources and is more in keeping with sustainable agriculture.


STEPS IN THE ANALYSIS AND INTERPRETATION OF ON-FARM RESEARCH-
EXTENSION DATA

Calculate a measure of environments: Environmental Index. El Step .

The factors that influence the environment for raising crops or livestock are many and complex and
are very difficult to assess. A convenient substitute measure of the quality of each environment where
a trial has been conducted is the average "yield" of all the treatments included when, and only when,
the same treatments are included in all the sampled environments. Thefirst step is to calculate this
index, EI, which provides an effective measure of the environmental differences in the research
domain represented by the range of Els.

The data that will be used to demonstrate this analysis are summarized in Table 1. To facilitate
further analyses, it is convenient to sort the data by descending (or ascending) values of this index.
The data in Table 2 are sorted in descending order with respect to the environmental index EI.











ANALYSIS AND INTERPRETATION
OF ON-FARM RESEARCH EXTENSION DATA


INTRODUCTION

On-farm research extension trials can have various functions and can be managed by researchers,
extension workers and/or farmers (Hildebrand and Poey, 1985). The most appropriate design for
incorporating farmer participation is a simple (few treatments), non-replicated trial which has one to
three treatments to be compared with the farmers' own technologies. For purposes of demonstration,
a trial conducted in the Amazon basin of Brazil (Singh, 1990) with four treatments on eight
environments, without replication, will be used. Additional information on design will follow
discussion of the steps in the analysis.


RESPONSE OF TREATMENTS TO DIFFERENT ENVIRONMENTS

The term "environment" is used here rather than "farm" or "location" because on a single farm, or
even in a single field, more than one environment can exist for the production of the livestock or
cultivation of crops. Similar farmers do manage different environments differently just as different
farmers may manage similar environments differently. Furthermore, making technologies conform
to varying environments, rather than making the environments conform to the technology, reduces
the need for off-farm resources and is more in keeping with sustainable agriculture.


STEPS IN THE ANALYSIS AND INTERPRETATION OF ON-FARM RESEARCH-
EXTENSION DATA

Calculate a measure of environments: Environmental Index. El Step .

The factors that influence the environment for raising crops or livestock are many and complex and
are very difficult to assess. A convenient substitute measure of the quality of each environment where
a trial has been conducted is the average "yield" of all the treatments included when, and only when,
the same treatments are included in all the sampled environments. Thefirst step is to calculate this
index, EI, which provides an effective measure of the environmental differences in the research
domain represented by the range of Els.

The data that will be used to demonstrate this analysis are summarized in Table 1. To facilitate
further analyses, it is convenient to sort the data by descending (or ascending) values of this index.
The data in Table 2 are sorted in descending order with respect to the environmental index EI.












Table 1. Response of maize (t/ha) to three soil amendments and the farmers' practices from on-farm
research results in Amazonas, Brazil (Singh, 1990).

t/ha


Farm No.


1TS


CM


4-


OP


Average 0.7 0.8 2.8 3.2 1.9


FP = farmers' practices, PCW = processed city waste (from Manaus), TSP = triple super phosphate


and CM chicken manure.


Table 2. Response of maize (t/ha) to three soil amendments and the farmers' practices from on-farm
research results in Amazonas, Brazil (Singh, 1990). Data sorted in descending order with respect to
the environmental index EI.


Farm No.
7
6
2
8
5
4
1
3

Average


PCW
1.4
1.0
1.1
0.7
0.7
1.1
0.2
0.0

0.8


t/ha
TSP
4.5
4.2
3.4
3.5
3.4
1.6
1.3
0.2

2.8


CM
4.0
3.6
4.4
4.0
3.6
2.8
2.8
0.6

3.2


FP = farmers' practices, PCW = processed city waste (from Manaus), TSP = triple super phosphate and CM chicken manure.










Relate treatment response to environment


The yield data for each treatment should be related to the environmental index. The secondstep is
to view the observations by graphing the results of one treatment against El as in Figure 1.4 It is
necessary to decide if the relationship is linear or curvilinear, and some estimate of the relationship
must be made. It is satisfactory simply to draw a line or a curve through the data. With practice, this
can become fairly precise. Linear regression can be accomplished easily with many inexpensive
calculators, and linear or curvilinear regression can be estimated with a computer. The estimated
relationship shown in Figure 2 a is from linear regression. Figures 2 b and 2 c show comparisons of
linear and curvilinear regression for FP and CM. For treatments FP and CM it is fairly evident that
curves represent the nature of the data better than straight lines. Therefore, for the remainder of this
analysis, curves will be used for these two treatments. For PCW and TSP straight lines are adequate.


Plot observations


Step 2 a


RESEARCHER'S CRITERION
MAIZE, MANAUS, BRAZIL, 1989


II
0 0.5 1 1.5 2 2.5 3 3.!
ENVIRONMENTAL INDEX, El


rI]
TSP


Figure 1. Observed response in t/ha of the TSP treatment to
Amazonas, Brazil (Singh, 1990).


environment (EI) for maize in


4 For this step all graphs should have identical axes so they can be compared easily by placing one graph over
the other. This will also facilitate the comparison of treatment responses to environment in the next step.


Ster 2











View Observations


RESEARCHER'S CRITERION
MAIZE, MANAUS, 1989


rO-
TSP


0 0.5 1 1.5 2 2.5
ENVIRONMENTAL INDEX, El


Figure 2 a. Estimated linear response in t/ha of the TSP treatment to environment (EI) for maize
in Amazonas, Brazil (Singh, 1990).

RESEARCHER'S CRITERION
MAIZE, MANAUS, 1989


0 0.5 1 1.5 2 2.5
ENVIRONMENTAL INDEX, El


m
FP

QUAD

LIN


3 3.5


Figure 2 b. Comparison of linear and quadratic response in t/ha of the FP treatments to environment
(EI) for maize in Amazonas, Brazil (Singh, 1990).5


5 In the case of the farmer's practice the estimated linear response to Els below EI= 1 and the estimated
curvilinear response to Els below EI= 1.6 present negative yield values. These negative values will be ignored in step 3
when comparing the responses of all treatments to El because they do not represent true values.


....................................................................................................................................1

-~ F
. 1 . 1 . .


Step 2 b











RESEARCHER'S CRITERION
MAIZE, MANAUS, 1989


0 0.5 1 1.5 2 2.5
ENVIRONMENTAL INDEX, El


CM

QUAD

LIN


3 3.5


Figure 2 c. Comparison of linear and quadratic response in t/ha of the CM treatment to environment
(EI) for maize in Amazonas, Brazil (Singh, 1990).


Evaluate data quality.


Step 2 c


Three criteria help evaluate data quality. The first relates to the range of environments sampled, the
second to the general conditions of the year in the research domain, and the third to distribution of
the environments in the research domain.

1) The range of the environmental index, EI, should be at least as large as the overall
mean EL If this criterion is violated, it usually means the research domain included only the
best environments (perhaps only "progressive farmers" were involved), or else that the year
was exceptional and resulted in high yields throughout the research domain.

2) The range and distribution of yields of the farmers' practices should be similar to
expected yields over a series of years. If the year was particularly good or bad overall, or
only very good or very poor sites were chosen, this criterion could be violated.

3) The distribution of the Els should be reasonable across the range of environments
sampled.











The data in Table 2 fairly well satisfy the three criteria. The range of Els (3.1 0.2 = 2.9) is greater
than the overall mean El (1.9), easily satisfying the first criterion. The range of yields of the farmers'
practices (FP) represents the normally expected range of yields under these conditions, satisfying the
second criterion. The distribution of Es, shown in Figures 1 to 3, is also quite reasonable, satisfying
the third criterion. Therefore, even though the number of environments is quite low (8), it should be
expected that the relationships among the treatments over various environments represented in Figure
3 will be stable over time, should the trial be repeated in this research domain (not necessarily the
same farms or sites). It also means that the persons involved in the trial can have confidence in
making recommendations to farmers in the specified recommendation domains (see step 6) based on
only this one year's data.

Assess interaction of treatments with environment Step 3

When all treatments have been related to, or regressed on EI, the third step is to assess the response
of the treatments to environment, as in Figure 3. No interaction exists if all the lines are parallel. If
no treatment by environment interaction exists (in practice this seldom occurs), the treatment which
is greatest over all environments is the best for the criterion used, here t/ha. However, if the lines
are not parallel, such as in this case, and is most common in practice, treatment by environment
interaction exists and different treatments may be best for different environments. Notice that the
values of the El are shown in the lower part of Figure 3. This will be useful in characterizing
recommendation domains. They also help evaluate data quality.


RESEARCHER'S CRITERION
MAIZE, MANAUS, 1989



4 ---------------------...............................---..................--. .............--... FP
3 -_ --------,----- ---- ^ F-- --- ----------- -P W
3 ........... ,
STSP

.... .CM


0 Y .1 IE sII I
0 0.5 1 1.5 2 2.5 3 3.5
ENVIRONMENTAL INDEX, El




Figure 3. Estimated responses in t/ha of the four treatments to environment (EI) for maize in the
Amazonas, Brazil (Singh, 1990).













Characterize the environments


Environments can be characterized using both biophysical and socioeconomic factors that may, at the
same time, be both quantitative and qualitative in nature. Data obtained for the environments in the
Amazon example include soils characteristics and a category called "Land class", Table 3. The soils
characteristics are self explanatory. Land class refers to the kind of forest that was cleared (P =
primary, S = secondary) and the number of years it has been cropped (1 = first year, etc.). The term
WL refers to land that had been cleared by bulldozer at the time of colonization and is, essentially,
waste land.




Table 3. Environmental characteristics of the sites for the on-farm maize trials in Amazonas, Brazil
(Singh, 1990).


Land class


PFI
PF,
SFI
PF1
PF2
SF2
SF2
WL


4.21
3.45
2.29
2.26
2.45
3.12
1.99
1.35


Alsat

58.3
69.1
91.7
79.2
80.0
94.8
90.7
94.8


7.4
7.1
4.5
6.8
5.0
2.8
2.0
0.1


Because the data in Table 3 have been sorted by EI, it is easy to assess the relationship between El
and these characteristics. Lower Els are associated with lower pHs, lower phosphorus levels, lower
ECECs and higher aluminum saturation. If desired, these relationships can also be graphed and/or
estimated by regression with El being the dependent variable as was done with pH in Figure 4.

Perhaps the most useful for farmers and extension agents is the Land class characteristic, because
farmers in these conditions seldom, if ever, have detailed soil information on their fields. It can be
seen that both the nature of the forest that was cleared for the field and the number of years in use
are closely associated with EI (Table 3).


Step 4














ENVIRONMENTAL CHARACTERIZATION
MAIZE, MANAUS, 1989


Z 2.5 ..............................................................................
0
S2.5
-J
2:
Z
w




0.5 -

0
3.8 4 4.2 4.4 4.6
pH


OBS


R2=0.95


4.8 5 5.2 5.4


Figure 4. Relationship of soil pH to El of the sites for the on-farm maize trial in Amazonas, Brazil
(Singh, 1990).




Define recommendation domains Step 5


Define tentative recommendation domains


Step 5 a


The recommendation domains depend on the environmental characteristics and selected evaluation
criteria, in this case, t/ha. Using this researcher criterion, TSP would be recommended for the highest
Els and CM would be recommended for all the other land (Figure 3). Notice that the two highest
Els are PFi, but the fourth highest is also PF1 (see Table 3). The difference between these
environments is that the higher ones have a pH greater than 5.0 and a phosphorus level above 7.0
ppm. However, farmers usually will not have this detailed information about their soils, so it may be
necessary to group all the PF, in a single recommendation domain and all the other classes in another.











The persons involved in the trial, including the farmers, should make this type of judgement to
facilitate the dissemination of results6. By grouping all PF1 into one domain, two tentative
recommendation domains have been defined for the criterion t/ha (Figure 5). If t/ha is a relevant
criterion, it would apply that either TSP or CM could be recommended for a field taken from primary
forest and in the first year of production and that CM should be recommended for all the other fields.
A final recommendation for PF, will depend on risk analysis.


Recommendations
Land Class Evaluation Criterion:
t/ha
PF1 TSP or CM


............... ........... .....
................ ................ .....;...............
PF2

L ~ :. :** ::: ::::.:: :. .I ..::;: . :::.:

................Source: Singh, 1990. .


Source: Singh, 1990


Figure 5. Tentative recommendation domains and the technology recommended for maize, based
on environmental factors (land class) and the evaluation criterion (t/ha ), Rio Preto da Eva,
Amazonas, Brazil.


Determine risk associated with the new technology


Step 5 b


The probability of low values (a measure of risk) of the selected criterion for any of the technologies
being assessed in the trial can be estimated by means of a distribution of confidence intervals based
on the treatment results in the tentative recommendation domain.




6 If for some reason research-extension personnel feel that they can include such characteristics as pH, ECEC,
and/or Al saturation, perhaps these recommendation domains could be redefined. Alternatively, it may be necessary to
confirm if farm 8 was really PF, or that it is, perhaps SF, and misclassified.











The formula:


y (t. s//n)


gives the confidence interval for the level of probability from a two-tailed "t" table for n-1 degrees
of freedom and where s = sample standard deviation for the observations in the tentative
recommendation domain. In the two-tailed "t" table, a probability level of 0.4 means that 40% of the
values lie outside the interval (either above or below) and 60% lie inside the interval defined by the
formula.
The lower value of this formula:


y- (t, s//n)


provides information on the probability that values lie below the confidence interval and is a measure
of the level of risk associated with the technology in this tentative recommendation domain. In this
step we consider the choice between CM and TSP for PF,. Table 4 is a summary of the calculations
and Figure 6 shows graphically the risk levels for the two treatments.


Table 4. Risk calculations for t/ha comparing CM and TSP for Land class PF, using the formula
Y (t, s//n) with Y = 3.9 and s = 0.2 for CM; Y = 4.1 and s = 0.5 for TSP; n = 3 and df= 2, and
using a from a one-tailed "t" table.'


probability
a of a lower value tf-2 CM TSP

0.2500 25.00 0.816 3.76 3.83
0.2000 20.00 1.061 3.73 3.75
0.1500 15.00 1.386 3.68 3.66
0.1000 10.00 1.886 3.62 3.51
0.0500 5.00 2.920 3.48 3.20
0.0250 2.50 4.303 3.23 2.79
0.0100 1.00 6.965 2.94 2.00
0.0050 0.50 9.925 2.54 1.13
0.0005 0.05 31.598


7 If only a two-tailed "t" table is available, the a value for a one-tailed "t" test is half the a value for the two-
tailed test. That is, an a value of 0.4 in a two-tailed "t" table, indicating that 40% of the values lie outside the interval
corresponds to an a value of 0.2 in a one-tailed table indicating that 20 % of the values lie below the interval.










RISK ESTIMATION
LAND CLASS PF,


CM

TSP


5 10 15 20
RISK (% times below Y-axis value)


Figure 6. Risk levels for treatments CM and TSP for Land class PFi. Evaluation criterion t/ha,
Amazonas, Brazil (Singh, 1990).


Values on the X-axis represent the percentage of times that yields of the indicated treatments fall
below the values represented on the Y-axis. In this case there is no significant difference associated
with the risk of TSP and CM for maize planted in PF,, either could be recommended .


Define final recommendation domains


Step 5 c


Based on the risk analysis of results in PF, we can confidently group all the PF, in a single
recommendation domain. In this way, the tentative recommendation domains indicated in Figure 5
become definitive or final recommendation domains for the criterion t/ha.











Compare results repeating steps 2 through 5 using alternative evaluation criteria Step 6

The evaluation criterion used so far has been t/ha, commonly used by agronomists in crop trials and
appropriate in the majority of cases to calculate EI. Nevertheless, few farmers use this criterion when
making production decisions. Farmers' evaluation criteria depend on resource scarcity and the
product the farmers want to maximize from their crop or livestock in question. If seed, labor or cash
are scarce, the most appropriate criteria are: kg/kg seed, kg/days labor, or kg/$ cash cost,
respectively. With AA it is easy to analyze data using alternative criteria.

The sixth step is to compare treatment results using alternative evaluation criteria. In this case we
will use kg/$ cash cost (Table 5). Figure 7 is based on the analysis using this criterion. Note that the
same El has been used regardless of the criterion being evaluated. The El values used to form the
X-axis do not change. The criteria used on the Y-axis do change. The same procedures were used
to obtain these relationships as were used to obtain the relationships based on the researcher's
criterion, t/ha.
Cash costs of the treatments were FP = $12, PCW = $208, TSP = $98 and CM = $127. Notice that
very different conclusions result when the evaluation criteria change. This is important because it
relates to the recommendations that will be made.


Table 5. Response of maize (kg/$ cash cost) to three soil amendments and the farmers' practices
from on-farm research results in Amazonas, Brazil (Singh, 1990).

kg/$ cash cost

Farm No. FP PCW TSP CM EI
7 208.3 6.8 45.9 31.5 3.1
6 183.3 4.8 42.9 28.4 2.8
2 0.0 5.3 34.7 37.7 2.2
8 16.7 3.4 35.7 31.5 2.1
5 12.5 3.4 34.7 28.4 2.0
4 20.8 5.3 16.3 22.0 1.4
1 12.5 0.8 13.3 22.4 1.1
3 0.0 0.0 1.5 5.1 0.2


FP = farmers' practices, PCW = processed city waste (from Manaus), TSP = triple super phosphate and CM chicken manure.


Many evaluation criteria may apply to the same set of on-farm research data and analyses should be
made for each criterion. In this example, two evaluation criteria have been demonstrated: t/ha in
Figure 3 and kg/$ cash cost in Figure 4. Once again, note that the El remains constant and does not
change for different evaluation criteria.












FARMER'S CRITERION
MAIZE, MANAUS, 1989


200 --

I-
W150 -
O
0




50
50 --... .. ... .. .. ... .. .. ... .. .. ... .. .. ... .. -- ----------- --- :---- -



0 0.5 1 1.5 2 2.5 3 3.5
ENVIRONMENTAL INDEX, El


FP

PCW

TSP


CM

E
El


Figure 7. Estimated responses in kg/$ cash cost of the four treatments to environment (EI) for
maize in Amazonas, Brazil (Singh, 1990).


Define tentative recommendation domains


Step6a


Based on analysis of the data of Table 5 and Figure 7 and the characterization of the environments
we can say that for the farmers' criterion (kg/$ cash cost), none of the amendments are better than
the farmers' practices for PF1 and SFI. Thus none of the amendments would be recommended for the
production of maize when the land has been taken out of primary or secondary forest and is in the
first year of use. After the first year, if maize must be produce, either CM or TSP could be
recommended. The option would depend on risk analysis.


Determine risk


Step6b


Table 6 shows the calculations and Figure 8 graphically shows the risk levels for CM and TSP for the
farmers' criterion, kg/$ cash cost, and for maize cropping on land after the first year of use, using
equation (2). In this case, CM is less risky (has a lower probability of low values) than TSP and
would be recommended for maize being planted on land in the second year of use.












Table 6. Risk calculations comparing CM and TSP using the equation Y (t, s/Vn) with a one-
tailed "t" table, when Y = 19.5 and s = 10.0 for CM; Y = 16.5 and s = 13.7 for TSP; n = 4
environments (land classes: PF2, SF2 and WL); and degrees of freedom = 3.


Probability of
a a lower value t=3 CM TSP

0.2500 25.00 0.765 15.7 11.2
0.2000 20.00 0.978 14.6 9.7
0.1500 15.00 1.250 13.2 7.9
0.1000 10.00 1.638 11.3 5.2
0.0500 5.00 2.353 7.7 0.3
0.0250 2.50 3.182 3.6
0.0100 1.00 4.541
0.0050 0.50 5.841
0.0005 0.05 12.941


RISK ESTIMATION
POOR ENVIRONMENTS: PF, SF, AND WL


CM

TSP


5 10 15 20
RISK (% times below Y-axes value)


Figure 8. Risk levels for the criterion kg/$ cash cost and the treatments CM and TSP from the maize
trials in the Amazonas, Brazil (Singh, 1990).











Define final recommendation domains


Ifkg/$ cash cost is relevant, as it would most likely be to these farmers, what they normally do (FP)
is the best on all land in first year of use. If farmers want or need to produce maize a second year,
CM would be recommended. Figure 9 gives a summary of the recommendations for this evaluation
criterion.


Recommendations:
Land Class Evaluation Criterion:
kg/$ of cash cost
PF,
FP
SF,

PF2

SF2 CM

WL

Source: Singh, 1990


Figure 9. Summary of the recommendation domains and the technology recommended for maize,
Rio Preto da Eva, Amazonas, Brazil, based on environmental factors and evaluation criteria.


Create extension recommendations for each of the recommendation domains


and formulate messages appropriate to each of the diffusion domains


Step 7


The results of using AA in on-farm research facilitate the creation of multiple recommendations
tailored to specific environments and for different farmers' criteria. Many different farmers included
in the research domain can benefit from this type of research-extension program. Figure 10
summarizes the recommendation domains and the resulting recommendations based on the results,
analysis and interpretation of the example used in this guide.


Step 6 c












Having specific recommendations facilitates making messages for use by extension. Extension
messages are oral (used, for example, in radio programs or with person-to person contact) or written
(extension bulletins) communications that can be generated based on each recommendation for
different clients (commercial producers, small farmers, different language groups, etc.).


Recommendations:
Land Class Evaluation Criterion: I Evaluation Criterion:
I
t/ha I kg/$ of cash cost
PF, TSP or CM
FP
SF1
SF ..... ..." ....


PF2 "/

SF2 CM
Zo /rc / "/990 / /
WL ,,

Source: Singh, 1990


Figure 10. Summary of the recommendation domains and the recommended technology for maize
based on environmental characteristics (land type) and two evaluation criteria, t/ha and kg/$, Rio
Preto da Eva, Amazonas, Brazil.










DESIGN OF ON-FARM TRIALS

Designing on-farm trials is very different from creating experimental designs for a trial conducted on
an experiment station. The reasons for undertaking the trials also are usually quite different. This
means that the nature of the results will also be different and the uses to which the results are put will
also vary. In turn, this means that the analysis of the data will need to be different.

Nature of on-station and on-farm trials

Traditional experiment station trials are set up to assess the effect of a set of treatments. In a
complete trial the treatments are replicated (repeated in different blocks) to facilitate a statistical
analysis to help determine if any differences among treatments can be considered real or just a matter
of chance. Controls in an on-station experiment often are zero-level treatments. Such traditional
experiment station trials are enumerative in nature and their purpose is descriptive. Most of the
common statistical procedures such as ANOVA are enumerative in nature as well. When a sufficient
number of replications have been used in the experimental design, ANOVA will help estimate
differences among treatments-were these differences from the control or from each other real or did
they result from chance? When significant differences exist, the researcher can describe them and
relate them to the conditions at the trial site on the station and for the year when the experiment was
conducted. The results apply to those specific conditions and are not meant to be extrapolated to
other locations, conditions or years. The same kinds of conclusions would be relevant to trials
conducted on an individual farm using the same kind of replicated design.

Experiment station trials are not usually meant to be predictive. From the results of the experiment
described above, it would not be reasonable to predict that the same results would be achieved at
other locations, conditions or years. In order to make a statement about what might happen in the
future (prediction) the same experiment would have to be repeated at the same location over a
number of years. Then it would be possible to describe what happened at that location and over that
period of years. Prediction would be based on the premise that the same results could be expected
in future years.

On-farm research is usually conducted for the purpose of making recommendations to a broader
group of farmers than just those who participated in the trial. The recommendation is, in fact, a
prediction that the same results will occur on specified kinds of farms or fields if the practice is
followed by farmers with those conditions. Statistical procedures that involve prediction are analytic
in nature. "Ad hoc statistical procedures are common in analytic studies... [because] traditional
methods [such as the most common forms ofANOVA] simply cannot accommodate the complexity
of on-farm trials." (Stroup et al., p. 160)

On-farm trials

The design of on-farm trials for the purpose of making recommendations to a much larger number
of farmers must take into account this difference in the nature of the two research approaches. In










DESIGN OF ON-FARM TRIALS

Designing on-farm trials is very different from creating experimental designs for a trial conducted on
an experiment station. The reasons for undertaking the trials also are usually quite different. This
means that the nature of the results will also be different and the uses to which the results are put will
also vary. In turn, this means that the analysis of the data will need to be different.

Nature of on-station and on-farm trials

Traditional experiment station trials are set up to assess the effect of a set of treatments. In a
complete trial the treatments are replicated (repeated in different blocks) to facilitate a statistical
analysis to help determine if any differences among treatments can be considered real or just a matter
of chance. Controls in an on-station experiment often are zero-level treatments. Such traditional
experiment station trials are enumerative in nature and their purpose is descriptive. Most of the
common statistical procedures such as ANOVA are enumerative in nature as well. When a sufficient
number of replications have been used in the experimental design, ANOVA will help estimate
differences among treatments-were these differences from the control or from each other real or did
they result from chance? When significant differences exist, the researcher can describe them and
relate them to the conditions at the trial site on the station and for the year when the experiment was
conducted. The results apply to those specific conditions and are not meant to be extrapolated to
other locations, conditions or years. The same kinds of conclusions would be relevant to trials
conducted on an individual farm using the same kind of replicated design.

Experiment station trials are not usually meant to be predictive. From the results of the experiment
described above, it would not be reasonable to predict that the same results would be achieved at
other locations, conditions or years. In order to make a statement about what might happen in the
future (prediction) the same experiment would have to be repeated at the same location over a
number of years. Then it would be possible to describe what happened at that location and over that
period of years. Prediction would be based on the premise that the same results could be expected
in future years.

On-farm research is usually conducted for the purpose of making recommendations to a broader
group of farmers than just those who participated in the trial. The recommendation is, in fact, a
prediction that the same results will occur on specified kinds of farms or fields if the practice is
followed by farmers with those conditions. Statistical procedures that involve prediction are analytic
in nature. "Ad hoc statistical procedures are common in analytic studies... [because] traditional
methods [such as the most common forms ofANOVA] simply cannot accommodate the complexity
of on-farm trials." (Stroup et al., p. 160)

On-farm trials

The design of on-farm trials for the purpose of making recommendations to a much larger number
of farmers must take into account this difference in the nature of the two research approaches. In










DESIGN OF ON-FARM TRIALS

Designing on-farm trials is very different from creating experimental designs for a trial conducted on
an experiment station. The reasons for undertaking the trials also are usually quite different. This
means that the nature of the results will also be different and the uses to which the results are put will
also vary. In turn, this means that the analysis of the data will need to be different.

Nature of on-station and on-farm trials

Traditional experiment station trials are set up to assess the effect of a set of treatments. In a
complete trial the treatments are replicated (repeated in different blocks) to facilitate a statistical
analysis to help determine if any differences among treatments can be considered real or just a matter
of chance. Controls in an on-station experiment often are zero-level treatments. Such traditional
experiment station trials are enumerative in nature and their purpose is descriptive. Most of the
common statistical procedures such as ANOVA are enumerative in nature as well. When a sufficient
number of replications have been used in the experimental design, ANOVA will help estimate
differences among treatments-were these differences from the control or from each other real or did
they result from chance? When significant differences exist, the researcher can describe them and
relate them to the conditions at the trial site on the station and for the year when the experiment was
conducted. The results apply to those specific conditions and are not meant to be extrapolated to
other locations, conditions or years. The same kinds of conclusions would be relevant to trials
conducted on an individual farm using the same kind of replicated design.

Experiment station trials are not usually meant to be predictive. From the results of the experiment
described above, it would not be reasonable to predict that the same results would be achieved at
other locations, conditions or years. In order to make a statement about what might happen in the
future (prediction) the same experiment would have to be repeated at the same location over a
number of years. Then it would be possible to describe what happened at that location and over that
period of years. Prediction would be based on the premise that the same results could be expected
in future years.

On-farm research is usually conducted for the purpose of making recommendations to a broader
group of farmers than just those who participated in the trial. The recommendation is, in fact, a
prediction that the same results will occur on specified kinds of farms or fields if the practice is
followed by farmers with those conditions. Statistical procedures that involve prediction are analytic
in nature. "Ad hoc statistical procedures are common in analytic studies... [because] traditional
methods [such as the most common forms ofANOVA] simply cannot accommodate the complexity
of on-farm trials." (Stroup et al., p. 160)

On-farm trials

The design of on-farm trials for the purpose of making recommendations to a much larger number
of farmers must take into account this difference in the nature of the two research approaches. In










most settings where on-farm research is being conducted, farmers want and need information and
recommendations rapidly. On the other hand, research resources, whether from governmental or
non-governmental organizations or from the farmers themselves, are usually very scarce. Therefore,
research methods and designs that are efficient yet effective and permit broad extrapolation beyond
just the farms or environments involved in the trials are a necessity. Designs must be flexible and
permit farmer and researcher innovation yet maintain the power to achieve confidence in the
recommendations resulting from the trials. Above all, the designs used for on-farm research must be
adaptable to many conditions and uses.

Designing on-farm trials for Adaptability Analysis

Adaptability Analysis provides the basis for the design of on-farm trials and meets the criteria set forth
above. Because farmers should be active participants in the trials and involved in all aspects including
selection of treatments and the nature of the environments used for the trial, it is necessary to be very
clear on the differences between treatments and environments.

Treatments and replications versus environments

In a few words, in an on-farm trial using Adaptability Analysis, anything that is not a treatment
becomes a factor affecting the environment for each of the locations. Consider a 2 x 2 factorial
crop trial with a randomized complete block arrangement but with only one block at each location.
Treatments are 1) each farmer's own local varietys with no synthetic fertilizer, 2) an improved variety
with no synthetic fertilizer, 3) the farmer's local variety with synthetic fertilizer, and 4) the improved
variety with synthetic fertilizer. Because farmers in this research domain seldom use synthetic
fertilizer with their local varieties, the local variety with no fertilizer can be considered as the check
or control and should be exactly the same as what each participating farmer does on the rest of his
or her field.

Except for these four treatments, there is no reason why each farmer participating in the trial must
follow the same cultural practices so long as the treatments are the same. Early or late planting, one
or two weedings, use of manure, irrigated or rainfed, low or high densities, etc., or even additional
fertilizer applications, are all factors that influence the nature of the environment in which the
crops or livestock are being grown. These factors need to be documented for environmental
characterization, but they are not different treatments, and actually have a positive effect on the
analysis by AA rather than a negative effect. Because these cultural practices represent those of
different farmers and are not controlled by the researcher, it is difficult for a researcher trained only
in experiment station research to accept. This is because in on-station research, the researcher is
interested in knowing the effect of only those variables being studied, i.e. the treatments. If other
factors are allowed to vary, their effect on the production environment can confound the effect of the
treatments. In on-farm research, however, the purpose of the trial is to find out how the treatments


8 It is acceptable for the farmers' own varieties to differ if they normally use different varieties.









What would happen to the above design if the farmers planted the two varieties correctly, but each
farmer used a different level or mix of synthetic fertilizer (which originally were thought of as
treatments)? The remaining factors that are still common to all locations (the varieties) are the only
remaining treatments and the different fertilizer applications become factors affecting the
environments. This, then, would leave a simple two treatment design. The different fertilizer
applications of the individual farmers would be recorded and used to characterize environments. A
similar design would have resulted if the farmers decided not to use the improved variety but rather
just used the same fertilizer application on part of their fields and no fertilizer on the rest. Fertilizer
becomes the treatment (if the level applied was the same in all cases).

Treatments

Treatments should be few in number to facilitate participation of farmers in the trial. This enhances
diffusion of acceptable results but also helps research and extension workers understand the farmers'
evaluation criteria, needed for analysis of the data. All environments should have the same
treatments, or at least a common sub set of treatments. Note that differences in farmer management
practices become factors affecting environment and have a positive rather than a negative effect on
trial design. The farmers' practices may differ from farm to farm to reflect each farmer's
individual management practices. These differences, of course, must be carefully documented to
serve in characterizing the environments of each farm. Furthermore, in most cases, one plot which
contains only the individual farmer's usual practices will serve as the control treatment in an on-farm
trial. This is because the farmer (as well as the research and extension personnel) must judge any new
technology against what the farmer already is doing to determine if the new technologies are better.
It is not adequate to determine only which of the new treatments is better than the other new
treatments.

Replications

Only one block at each location is needed for analysis by AA. If more than one block (multiple
replications) are desired by the research or extension worker to help assure that a trial will not be lost
at a specific location, two blocks is a sufficient number. Usually the blocks will be averaged for
analysis by AA. More than two blocks per location is not an efficient use of resources unless each
block is considered a different environment.

Environments

The number of environments is more important than the number of replications in each environment.
Based on Stroup et al. (p. 172) a simple rule can be devised:










The rule of 48 The number of treatments times the number of environments should equal
approximately 48 so long as the number of environments does notfall below about 12.
Thus, for 4 treatments, 12 environments would be sufficient; for 3 treatments, 16
environments; and for 2 treatments, 24 environments.9

Finally, in order to increase the probability that the first of the three criteria for confidence will be
met, the design should include a wide range of environments, including different kinds
of farmers and physical settings, and these should be distributed as well as possible to help satisfy the
third of these criteria. The second criterion depends largely on natural conditions beyond
the control of the persons doing the on-farm research.






































9It is recognized that in the example used in this guide, the number of environments is less than 12. It was used
because it facilitates calculations, but it does result in a limited number of observations upon which to base environmental
characterization.











REFERENCES

Andrew, C.O. and P.E. Hildebrand. 1993. Applied agricultural research: Foundations and
methodology. Westview Press. Boulder, Colorado.

Hildebrand, P.E. 1984. Modified stability analysis of farmer managed, on-farm trials. Agronomy
Journal, 76:271-274.

Hildebrand, P.E. and F. Poey. 1985. On-farm agronomic trials in farming systems research and
extension. Lynne Rienner Publ. Inc., Boulder, Colorado.

Hildebrand, P.E. and J. T. Russell. 1995 (Book draft). Adaptability analysis: A method for the
design, analysis and interpretation of on-farm research-extension.

Singh, B. K. 1990. Sustaining crop phosphorus nutrition of highly leached oxisoils of the
Amazon Basin of Brazil through use of organic amendments. Unpublished Ph.D.
Dissertation, University of Florida, Gainesville.

Stroup, W. W., P.E. Hildebrand and C.A. Francis. 1993. Farmer participation for more effective
research in sustainable agriculture. Chapter 12 In: Technologies for sustainable
agriculture in the tropics. American Society of Agronomy, Madison, Wisconsin.


Guidc/Juncl7






























APPENDIX






















Table 1. Response of maize (t/ha) to three soil amendments and the farmers' practices from on-
farm research results in Amazonas, Brazil (Singh, 1990).



t/ha

Farm No. FP PCW ISP CM El

1 0.2 0.2 1.3 2.8 1.1
2 0.0 1.1 3.4 4.4 2.2
3 0.0 0.0 0.2 0.6 0.2
4 0.2 1.1 1.6 2.8 1.4
5 0.2 0.7 3.4 3.6 2.0
6 2.2 1.0 4.2 3.6 2.8
7 2.5 1.4 4.5 4.0 3.1
8 0.2 0.7 3.5 4.0 2.1

Average 0.7 0.8 2.8 3.2 1.9

FP = farmers' practices, PCW = processed city waste (from Manaus), TSP = triple super phosphate and CM = chicken manure.
























Table 2. Response of maize (t/ha) to three soil amendments and the farmers' practices from on-
farm research results in Amazonas, Brazil (Singh, 1990). Data sorted in descending order with
respect to the environmental index EI.



t/ha
Farm No. FP PW TISP CM El
7 2.5 1.4 4.5 4.0 3.1
6 2.2 1.0 4.2 3.6 2.8
2 0.0 1.1 3.4 4.4 2.2
8 0.2 0.7 3.5 4.0 2.1
5 0.2 0.7 3.4 3.6 2.0
4 0.2 1.1 1.6 2.8 1.4
1 0.2 0.2 1.3 2.8 1.1
3 0.0 0.0 0.2 0.6 0.2

Average 0.7 0.8 2.8 3.2 1.9


FP = farmers' practices, PCW = processed city waste (from Manaus), TSP = triple super phosphate


and CM = chicken manure.



























Table 3. Environmental characteristics of the sites for the on-farm maize trials in Amazonas,
Brazil (Singh, 1990).


Land class


PFi
PFi
SFi
PF1
PF2
SF2
SF2
WL


Alsat


4.21
3.45
2.29
2.26
2.45
3.12
1.99
1.35


E2QS


58.3
69.1
91.7
79.2
80.0
94.8
90.7
94.8





















Table 4. Risk calculations for t/ha comparing CM and TSP for land class PFi, using the formula
y (t sA/n) with $ = 3.9 and s = 0.2 for CM; Y = 4.1 and s = 0.5 for TSP; n = 3; and degrees of
freedom = 3; using a from a one-tailed "t" table.


probability
a of a lower value td-2 CM TSP

0.2500 25.00 0.816 3.76 3.83
0.2000 20.00 1.061 3.73 3.75
0.1500 15.00 1.386 3.68 3.66
0.1000 10.00 1.886 3.62 3.51
0.0500 5.00 2.920 3.48 3.20
0.0250 2.50 4.303 3.23 2.79
0.0100 1.00 6.965 2.94 2.00
0.0050 0.50 9.925 2.54 1.13
0.0005 0.05 31.598























Table 5. Response of maize (kg/$ cash cost) to three soil amendments and the farmers' practices
from on-farm research results in Amazonas, Brazil (Singh, 1990).



kg/$ cash cost

Farm No. EP PCW ISP CM El

7 208.3 6.8 45.9 31.5 3.1
6 183.3 4.8 42.9 28.4 2.8
2 0.0 5.3 34.7 37.7 2.2
8 16.7 3.4 35.7 31.5 2.1
5 12.5 3.4 34.7 28.4 2.0
4 20.8 5.3 16.3 22.0 1.4
1 12.5 0.8 13.3 22.4 1.1
3 0.0 0.0 1.5 5.1 0.2


FP = farmers' practices, PCW = processed city waste (from Manaus), TSP = triple super phosphate and CM = chicken manure.






















Table 6. Risk calculations comparing CM and TSP using the equation y (t, s/Vn) with a one-
tailed "t" table, when Y = 19.5 and s = 10.0 for CM; Y = 16.5 and s = 13.7 for TSP; n = 4
environments (land classes: PF2, SF, and WL); and degrees of freedom = 3.



Probability of
a a lower value tf=3 CM TSP

0.2500 25.00 0.765 15.7 11.2
0.2000 20.00 0.978 14.6 9.7
0.1500 15.00 1.250 13.2 7.9
0.1000 10.00 1.638 11.3 5.2
0.0500 5.00 2.353 7.7 0.3
0.0250 2.50 3.182 3.6
0.0100 1.00 4.541
0.0050 0.50 5.841
0.0005 0.05 12.941






















RESEARCHER'S CRITERION
MAIZE, MANAUS, BRAZIL, 1989







m*


i i !. U .

U
. ....I.... I ..I..... .I. I.....................................................................


0 0.5


1 1.5 2 2.5
ENVIRONMENTAL INDEX, El


Figure 1. Observed response in t/ha of the TSP treatment to environment (EI) for maize in
Amazonas, Brazil (Singh, 1990).


TSP



















RESEARCHER'S CRITERION
MAIZE, MANAUS, 1989


TSP


0 0.5


1 1.5 2 2.5
ENVIRONMENTAL INDEX, El


3 3.5


Figure 2 a. Estimated linear response in t/ha of the TSP treatment to environment (EI) for maize
in Amazonas, Brazil (Singh, 1990).



















RESEARCHER'S CRITERION
MAIZE, MANAUS, 1989


FP

QUAD

LIN


0 0.5 1 1.5 2 2.5
ENVIRONMENTAL INDEX, El


3 3.5


Figure 2 b. Comparison of linear and quadratic response in t/ha of the FP treatments to
environment (EI) for maize in Amazonas, Brazil (Singh, 1990).




















RESEARCHER'S CRITERION
MAIZE, MANAUS, 1989


CM

QUAD

LIN


0 0.5


1 1.5 2 2.5 3 3.5
ENVIRONMENTAL INDEX, El


Figure 2 c. Comparison of linear and quadratic response in t/ha of the CM treatment to
environment (EI) for maize in Amazonas, Brazil (Singh, 1990).




















RESEARCHER'S CRITERION
MAIZE, MANAUS, 1989


FP

PCW

TSP


CM
Els
Els


0 0.5


1 1.5 2 2.5 3 3.5
ENVIRONMENTAL INDEX, El


Figure 3. Estimated responses in t/ha of the four treatments to environment (EI) for maize in the
Amazonas, Brazil (Singh, 1990).





















ENVIRONMENTAL CHARACTERIZATION
MAIZE, MANAUS, 1989


x
w
S2.5

S 2
UJ

| 1.5
z
0
1


0 +
3.8
3.8


OBS


R'=.95


4 4.2 4.4 4.6 4.8 5 5.2 5.4
pH


Figure 4. Relationship of soil pH to El of the sites for the on-farm maize trial in Amazonas,
Brazil (Singh, 1990).






















Recommendations
Land Class Evaluation Criterion:
t/ha
PF, TSP or CM

...................................... ...............
,.. .. .... . ... ..- . .
SF1


PF2


SF2


WL S ,9

Source: Singh, 1990
Sour.e.............


Figure 5. Tentative recommendation domains and the technology recommended for maize,
based on environmental factors (land class) and the evaluation criterion (t/ha), Rio Preto da Eva,
Amazonas, Brazil.

















RISK ESTIMATION
FOR LAND CLASS PFi


CM

TSP


5 10 15 20 25
RISK (% times below Y-axis value)


Figure 6. Risk levels for treatments CM and TSP for Land class PF1. Evaluation criterion t/ha,
Amazonas, Brazil (Singh, 1990).




















FARMER'S CRITERION
MAIZE, MANAUS, 1989


2 00 ...............


150 ......
0


< 100-. .


50



0 0.5


FP


PCW

TSP


CM


El


1 1.5 2 2.5 3 3.5
ENVIRONMENTAL INDEX, El


Figure 7. Estimated responses in kg/$ cash cost of the four treatments to environment (EI) for
maize in Amazonas, Brazil (Singh, 1990).



















RISK ESTIMATION
POOR ENVIRONMENTS:PF2,SF2 AND WL


18

16

14
F-
) 12
0
10
O
.110

< 8
01
S6
4

2

0


CM


TSP


5 10 15 20
RISK (% times below Y-axes value)


Figure 8. Risk levels for the criterion kg/$ cash cost and the treatments CM and TSP from the
maize trials in the Amazonas, Brazil (Singh, 1990).


















Recommendations:
Land Class Evaluation Criterion:
k/$ of cash cost
PFi

SFI

PF2

SF2 CM

WL

Source: Singh, 1990











Figure 9. Summary of the recommendation domains and the technology recommended for maize,
Rio Preto da Eva, Amazonas, Brazil, based on environmental factors and evaluation criteria.

























Recommendations:
Land Class Evaluation Criterion: I Evaluation Criterion:
t/ha I k_/$ of cash cost
PF, TSP or CM
FP
..................... ..
SF '..................... .

PF2

/ / .
..................... .
SF2 CM

WL //

Source: Singh, 1990


Figure 10. Summary of the recommendation domains and the recommended technology for
maize based on environmental characteristics (land type) and two evaluation criteria, t/ha and
kg/$, Rio Preto da Eva, Amazonas, Brazil




University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs