Ed Adams, Paul Peterson, and Russ Karow
(also in pdf format)
provides growers already doing on-farm testing (OFT) with a tool
to refine their testing system. It also provides growers considering
on-farm testing for the first time a basic background in testing
theory and procedures. The guide is narrowly focused on one type
of test--replicated strips. This type of test is similar to testing
done by university researchers and is the easiest to statistically
analyze and interpret.
Before we discuss
on-farm testing, you must understand the following terms and concepts
used in field testing:
A check plot
or control represents your current practice. It does not receive
the new technology being tested. This might be your conventional
tillage practice, fertilizer applied in the usual manner, the variety
you currently grow, or a crop not receiving a fungicide application.
The check plot and the treated plot differ only in the specific
treatment comparison being made. Aside from this, plots must be
managed exactly the same to avoid biasing the results.
In a tillage
experiment, the check plot could be your normal plow operation and
the new technology might be a sweep chisel. Aside from the two tillage
treatments, all other production practices must be the same: planting
date, fertilizer rate, variety, weed and disease management.
In some situations,
the new technology incorporates several practices. For example,
if a conventional till-plant operation is compared to no-till seeding,
different tillage, fertilization and seeding systems are being compared.
A fair comparison can only be made between the two complete systems,
not any given part of either system.
meaning repitition, is used to determine whether the difference
between plots is due to chance variation always present in fields
or caused by the treatment(s) being evaluated. Through replication,
average treatment effect values can be obtained. Comparisons between
average values are always more accurate than those between single
plots. Replicating your check and treatment plots at least three
or more times will give you much greater confidence in your results.
generally replicated in both space and time. Replication in space
means that several strips of each treatment are placed in a field
(replication on-site) or that single strips of each treatment are
placed in several fields across the farm (replication over sites).
Replication in time is repeating the trial over several years. Climatic
conditions, soils, and other factors can change significantly from
location to location and year to year. It is critical that final
conclusions about a new practice be made only after being evaluated
over several years and/or at several locations.
assures that any one treatment is not biased or favored in any way.
To randomize a trial, randomly mix the order and placement of replicated
check plots and treatments (Fig. 1). You may draw treatment numbers
out of a hat or flip a coin as you assign treatments to plots. If
treatments are assigned to plots in a nonrandom fashion, you may
unknowingly introduce bias.
An example of a completely random plot design, with a check plot
(C), a treated plot (T), and 3 replications. Notice the entire trial
area is kept within a uniform soil condition. Other plot arrangements
example shows how these basic ideas are used in an on-farm testing
The steps involved
in laying out an on-farm test are:
your goal and objectives.
what treatments you will use and what your check plot or control
how best to lay out your plots on the selected site.
what data you will collect and how you will collect it.
how the data will be evaluated.
how the data will be shared with others.
Every OFT project
must have a goal and specific objectives. Goals are statements of
the overall theme of your experiment. A goal, for example, may be
to reduce soil erosion on your farm.
are statements of the problem you wish to evaluate in your project.
These are the ideas you want to test or questions you want to answer.
Objectives are measurable and relate to your overall goals. The
objectives will determine what is measured and the type of data
you will collect during the project. For example, you might postulate
that a no-till seeding system will leave more surface residue than
the seeding system you now use. The trial you establish would involve
a comparison between the two systems. One objective will be to determine
if the residual levels are different between the two seeding systems.
Residue levels would be one type of data collected as part of the
treatments for a trial, keep them simple and few, no more than 3,
including the check plot. As treatments increase in number, so do
the number of plots, and the complexity of the OFT project.
comparisons that represent significantly different production practices.
Until you have a significant amount of experience in on-farm testing,
avoid making treatment comparisons of minor production practices.
an appropriate check or control plot. For instance, you might wish
to compare deep placement of fertilizer versus broadcast application.
Knowing that surface applied fertilizer is often less efficient
than deep placed fertilizer, it may be tempting to increase the
surface applied rate in order to try to equalize treatments; however,
this would confound the study. If placement is the main objective,
different fertilizer rates should not be included as treatments.
If is very important that production inputs other than the treatments
being tested remain constant. If management inputs are changed between
treatments, the results may be biased due to the input differences.
The most common
problem with on-farm trials is lack of recognition that field variation
can mask or conceal treatment differences. Take special care to
plan and organize the field plot layout to assure that all treatments
have an equal opportunity to perform. Choose a field site with the
greatest possible uniformity. Regardless of whether you have a 40-ft
by 100-ft plot or a 100-foot by quarter-mile plot, a uniform field
location is critical. When choosing a site consider previous crop
history (fertilizer rates, herbicides, tillage, etc.), drainage,
soil texture, soil depth, topography, pest infestations, and bordering
influences such as trees, runoff from neighboring fields, lack of
fencing from animals, and other factors. Avoid placing trials in
runoff areas, near fence lines or in field corners. These areas
are often subject to multiple or irregular applications of fertilizer
of a uniform field site depend on the type of test being conducted.
Pay particular attention to things that strongly influence your
treatments. For example, when testing a soil-applied herbicide,
soil organic matter content pH, and texture consistency are important.
In a fertility trial, the soil must have uniform drainage, soil
depth, organic matter content, and soil nutrient levels. For variety
or tillage comparisons, the overall soil productivity level should
be constant within the field site. Use your county soil survey maps
of the fields being considered to help you select the site.
access when selecting a plot location. Is the site easily accessible
for mid-season treatment applications and data collection? If early
or differential harvest is likely (such as with an alternate crop),
can you get at the site with harvest equipment without destroying
other crops? Will you hold a tour of your site? If so, is there
ready access for visitors and their vehicles?
a site, actual plot layout must be considered. Two different layouts
or designs, the completely random design and randomized complete
block design, are commonly used. Completely random designs are used
if the test site is known to be very uniform, without differences
in soil characteristics, fertility levels, slope, and previous crop.
Layout of the plots might look like those in Figure 1. In this design,
all treatments have an equal chance of being assigned to any given
plot. It is possible to have identical treatments side by side.
Use a randomized
complete block design when it is not possible to obtain a uniform
test site. For example, the test site may have different slopes,
previous crops, soil depths, etc. In this case, treatments are grouped
into sets called replications. Each replication contains a complete
set of treatments. Each replication is placed in a uniform area.
Using such an arrangement allows all treatments to have equal potential
to perform. Through this design, the effects of replications can
be removed or "blocked out" when analyzing the data. Plot
layout might look like that in Figure 2.
An example of randomized complete block design with a check plot
(C), treatment plot (T), and 3 replications. Each block of treatments
(replication) is kept within a uniform soil condition.
approach is to include variation equally across all the treatments
in a test. As an example, field strips can include the field variation
by running the strips perpendicular to the variation. A layout of
this type is shown in Figure 3. You must be extremely careful to
include the same variation equally across all treatments to have
a valid test. Establishing treatments within a uniform area is still
the best method.
An example of a replicated strip trial, using a completely random
design. Strips have been laid out so that each treatment has the
same amount of soil variability.
Plot size is
determined by field size, uniformity of the field, equipment used
and area needed to carry out a particular treatment. Adjust plot
lengths so that each treatment is within a reasonably uniform area
or so that each uniformly covers the field variation as discussed
above. Strip plot width is determined by the width of equipment
used to apply treatments (e.g., planter, sprayer, etc.) and/or harvest
plots. The width of the established treatment should be larger than
the harvest width. This way there will be a uniform harvest width
and errors in harvesting will not affect side by side treatments.
Typical treatment plots are between 1/10 and 1/2 acre.
COLLECTION AND RECORD KEEPING
of steps taken in conducting on-farm trials are important for two
reasons. First, detailed records are often required to interpret
data. Sometimes the results are unclear. Thinking about them further
and looking over documentation usually brings explanations to light.
Second, written documentation preserves the details of your OFT
project so you can share information with others.
sheets are included here in pdf format to show the type
of records you should keep: 'On-Farm Research Field Record Log',
'On-Farm Research Management Summary', 'On-Farm Research Costs Summary',
'On-Farm Research Income Summary', and 'Rainfall Record Sheet.'
Also, keep required record sheets on any pesticide applications.
information includes the baseline data needed to document and interpret
a valid, unbiased test. This information is easily entered on the
Description. Clearly state the goals, objectives, treatments
and experimental design of the trial.
Field History. Record differences in soil type and other obvious variations within
the test site and the previous cropping history. Include crop rotation,
tillage practices, previous crop and variety, fertilizer and pesticides
applied. Make a diagram showing the layout of the field trial.
and Fertility Program. Sample soil from the intended harvest
areas using university guidelines. Send samples to a reputable laboratory
for analysis. Make fertilizer applications based on soil test results.
Record the quantity and form of fertilizer used.
at Seeding. If your soil samples are taken near the time of
seeding, record the depth to moisture and depth of moisture. Have
your soil samples evaluated for available soil moisture.
Conditions. Record the crop, variety, seeding rate (pounds per
acre and seeds per pound), planting date, soil temperature, type
of planter, seeding depth, row spacing, residue levels and any other
conditions that might influence the stand establishment and crop
and Observations. Record all field operations in diary format.
Take notes on the methods of your field operations, such as the
type of equipment, depth of tillage operations and materials applied
to either the whole field or to just one treatment.
Weather. General observation of growing season weather conditions is all
that is required. If practical, place a rain gauge at the test site.
After each storm, record rainfall in the rainfall record sheet and
empty the rain gauge. A little oil in the rain gauge will prevent
the water from evaporating before you can get out to the field to
Weeds and Disease. Make notes on the presence and density of
insects and diseases, date of infestation, and extent or severity
of damage. Record similar observations for persistent weeds. Note
differences between treatments, if any, due to pests. If pesticide
treatments are being compared, take more detailed data to evaluate
crop injury and level of control of different pest species.
and Development. During the growing season make and record observations
of plant growth and development. Record the date each treatment
reaches a critical growth stage. It is just as important to record
that you see no differences among treatments at a certain growth
stage as it is to record obvious differences. For example, critical
stages in cereal crop development include emergence, tillering,
stem elongation, booting, and heading. Record crop stage at the
time of treatment applications, such as spraying or top dressing.
When abnormal conditions occur, such as drought, note the differences
in plant growth or response among treatments.
It is important
to plan ahead and identify what should be measured, and when and
how to take measurements. What you will measure depends on the project's
If the purpose
is to increase yield, then a measure of yield is required. If the
objective of a new practice is to increase soil moisture, then soil
water tests are needed. If the purpose is to increase net farm profit,
then you must analyze costs and returns (including yield).
If you need
help in deciding what to measure and how to measure it, consult
your county extension agent. Without appropriate data and a method
to measure treatment differences, your trial will have little value
or could lead to inaccurate conclusions. Remember, the more you
plan and document, the greater confidence you can have in your results.
are needed to make production and economic comparisons between treatments.
To be valid, yield measurements must be taken from comparable areas
in each treatment plot. You must measure the size of the harvest
area. Measure plot lengths with a measuring tape or other reliable
measuring device before or immediately after you harvest each plot.
These distances are then multiplied by the width of the combine
header to arrive at the harvested area. Harvested area is used to
calculate yield per acre. An example of yield calculation is shown
in the data analysis section of this manual.
middle portion of each treatment plot. This assures that the yields
are not affected by a condition bordering the treatment. Yields
can be measured with a local truck scale, a weigh wagon, or using
the barrel method. Harvest equipment must be completely empty and
clean before each treatment is harvested.
Save a sample
from each treatment to determine moisture content at harvest and
any other quality factors that may be important such as test weight
and protein content. If moisture contents differ between the treatments,
yield must be corrected to a constant moisture.
method is a slow and time-consuming method to determine yield, but
may be useful if other methods are not available. It involves measuring
the volume of grain yield using a standard 55-gallon barrel. Treatment
yields can be compared using the barrel method as long as the test
weights do not differ between the treatments. For example, with
wheat you can use the following protocol:
in the trial must first be converted to commonly used units prior
to analysis and summary. The following is an example, converting
wheat yields to common units.
|| 20 ft
|| 400 ft
|| 20 ft x 400 ft = 8000 sq ft
|| 8000 sq ft/43,560 sq ft per acre = 0.18 acre
|| 1000 lbs
|| 1000 lbs/60 lbs per bushel = 16.7 bushels
in bushels per acre:
|| 16.7 bushels/0.18 acre = 92.8 bushels/acre
analysis largely depends on how the project was designed and
conducted. Using a uniform field site, simple treatments, and
a randomized complete block design, a research test can be statistically
and the results quickly evaluated and interpreted. Simple statistical
software packages are available through your county extension
agent to do data analysis.
interpret results from your on-farm trial, carefully summarize management
history, data collected, and observations made. Summary forms are
provided in the back of this guide for this purpose. The summarized
results should address your goals and objectives. If your objective
was to reduce costs, equal or even lower yields may be an acceptable
result as long as costs are reduced and the net return has improved.
the time to share the results with your neighbors and county extension
agents. This flow of information and experience is necessary for
the progress of agricultural production and management.
Baird Miller is an extension agronomist with Washington State University. Ed
Adams is a regional Water Quality Coordinator with Washington
State University. Paul Peterson is an area extension agent
with Washington State University. Russ Karow is an extension
agronomist with Oregon State University.