in four parts
-O AITS I auill.
Foreign Agricultural Service April 1954
U.S.DEPARTMENT OF AGRICULTURE
This publication has
been prepared for use in
the technical cooperation
program of the Foreign
Actual farms-with their various soil, slope, cropping, and manage-
ment conditions-are the ultimate test of any new or improved practice.
a guide to
EXTENSIVE TESTING ON FARMS
by Henry Hopp
in 4 parts
Part I: Introduction
Part II: Result Tests
Foreign Agricultural Service
UNITED STATES DEPARTMENT OF AGRICULTURE
Washington, D. C.
Part I. Introduction .. . . .....
When do you make extensive tests on farms ?
How long should extensive tests last? ....
What are the kinds of extensive tests ? .....
Result tests ....... .......
1. Delimit the regions ........
2. Decide on the number of farms
3. Select the farms ..........
4. Decide if you need check plots .
5. Put in the plots . . ...
6. Collect the data . ..
7. Interpret the results . .
. . 13
. . 16
S ~~ ~ 2
PART I. INTRODUCTION
"Farmers generally would not change their prac-
tice from observing what could be done on farms op-
erated at public expense. There must, therefore, be
demonstrations carried on by farmers themselves on
their own farms and under ordinary farm conditions."
-A. C. True.
Like the farmer, the technician also demands
demonstration: before he recommends a practice, he
wants proof that it is applicable to the farms in his
Since World War II the concept of technical service to farmers as a
tool for agricultural progress has grown rapidly in acceptance through-
out the world. Programs to give this type of service are being set up at
an unprecedented rate. Naturally these programs look to older ones for
The older technical-service programs generally operate through
two well-recognized organizational entities: An experiment station and
an extension service. In its simplest form, the experiment station is a
place at which experiments are conducted to develop better practices
for farms, and the extension service is a group of advisers who help
farmers adopt these practices.
But these definitions tend to obscure a vital step in the process of
agricultural advancement: the whole complex of testing and proving op-
erations that goes far beyond the intensive research at the experiment
station itself and is necessary before an improved agricultural practice
is ready for recommendation to farmers. This complex involves out-
lying substations, experimental farms, experimental plots on farms,
farmer cooperators, and often commercial organizations such as seed
companies and farm-implement manufacturers. In fact, a considerable
part of the work of experiment stations consists of the extensive testing
that is carried on outside the physical confines of a main station.
Each of the two levels of testing serves a distinct purpose: intensive
research, at the experiment station, discovers a good practice; exten-
sive testing, on farms, determines the applicability of that practice over
a region. Intensive research gives the first clues as to what is a better
practice but usually is not a sound basis for a flat recommendation to
farmers. Only after an extension agent is armed with satisfactory in-
formation on applicability, can he confidently recommend a practice,
knowing in advance what results farmers should get.
How extensive testing works in an agricultural-service program is
illustrated by the cooperative activities in California. The foreword to
Research Program of the Experiment Station, 1950-1951, published by
the University of California's College of Agriculture, states, "The Ag-
ricultural Extension Service and the Agricultural Experiment Station
have over 7,000 tests, demonstration and research plots scattered over
the state each year. They work closely together." And J. Earl Coke,
formerly director of agricultural extension in California and now as-
sistant secretary of the United States Department of Agriculture, wrote
to the author on September 2, 1952, as follows-
"Because of the variety of climate, soils and crops grown
in California, it is necessary to conduct test plots in the
field to determine actual adaptability of crops under local
conditions. Therefore, for many years the Extension Ser-
vice has conducted many tests, in most cases with the co-
operation of the Experiment Station. For example, grain
variety trials are conducted throughout the State by the
farm advisors. In the majority of the cases the seed for
these variety tests comes from the Experiment Station,
which also threshes, weighs, and tabulates results from
many of the plots. In addition to the agricultural exten-
sion test plots, field tests are conducted by the Experi-
ment Station in various counties, in which in most cases
the farm advisors are involved in some way. The farm
advisor may assist in the selection of the cooperator and
the plots; he may also assist in the application of various
treatments used in obtaining the results. This works
very much like the tests run by the Extension Service ex-
cept that the leadership comes from the Experiment Sta-
We see, then, how intensive research at an experiment station is not
enough support for an extension program and must be reenforced by ade-
quate extensive testing. Herein lies guidance for organizers of a new
agricultural-service program. Although at the beginning the organiza-
tion will have a much smaller extensive testing program than the State
of California, it should nevertheless recognize the place that each of
three steps has in agricultural development: (1) Intensive research at
the experiment station to develop practices, (2) extensive testing of
these practices to determine applicability under farmers' conditions,
and (3) assistance to farmers in putting the practices into effect. Fail-
ure to provide for the second step tends toward research without oppor-
tunity for practical outlet, and toward extension without a proved tech-
This Guide has been prepared to aid the agricultural technician who
is specifically concerned with extensive testing on farms. When a new
practice is suggested to him, either by the experiment station or by his
own experience, he must still find answers to several questions before
he can recommend that practice: How will the practice work in the dif-
ferent parts of my area? How must it be modified for farm use? Will
it yield enough benefit to be worthwhile? How much will the benefit vary?
Answering these and similar questions on applicability involves a
kind of testing that is different from the intensive research done at the
experiment station. For one thing, it requires the cooperation of farm-
ers; for another, it has to be conducted in many locations and under ac-
tual farming conditions. Sometimes, to get valid measures of applica-
bility, the technician must go even further than locating the tests on
farms: sometimes he must have the farmers themselves do the work,
with their own implements and in their own way.
The technician concerned with testing a practice for adaptability
finds himself in the intermediate role of researcher and extension man.
In fact, since extensive testing is a step that lies between the intensive
research at the experiment station and true extension work on the farm,
it is a step that can be taken either by the researcher as a sequel to his
station experiments or by the extension man as a preliminary to his
blanket recommendation of a practice.
WHEN DO YOU MAKE EXTENSIVE TESTS ON FARMS?
The concept of extensive testing requires that you carefully assess
the real reason for your test. If you decide specifically how you want
to use the information developed by the test, you will usually be able to
decide whether you need an intensive test, an extensive test, or possibly
both in sequence. When you intend to use the information to make a
final recommendation for farmers, you probably need a test of an ex-
In conventional research, the technician is concerned with a number
of practices-sometimes a very large number-and with exact measure-
ment of the responses. For these reasons, he generally lays out his
tests in replicated, or repeated, plots on one or more experimental
areas. In extensive testing the situation is somewhat different: there
the technician usually begins with a practice that is reasonably certain
.:- ***,- iE
J'" r '
I -, /
.' j , ;, '-
Experimental plots at a research station. Practices proved superior
here usually need extensive testing on farms before they can be recom-
mended for actual farm conditions.
- Z M
to have a beneficial effect, but he is trying to determine its applicability
To make clearer the distinction between questions answered by re-
search and questions answered by extensive testing, we might point out
broad problems that typically involve applicability and therefore are
solved by extensive tests on farms and not by experiments at the re-
To determine conditions in an area.-Some tests are undertaken to
determine conditions in an area for which the technician is responsible.
An example is a test to determine the average response to fertilizer.
Problems of this kind might be looked upon as a survey conducted by
means of tests. The answer cannot be obtained through research at one
or two research stations; the validity of the answer depends on getting
an adequate number of tests over the area.
To find responses for different regions.-Another group of problems
concerns different responses to a practice in the different regions of the
technician's area. A region may be a soil series, a geographic complex,
or even a particular class of farmers. You may have a lead on a prac-
tice from research at the experiment station, but you may still need to
delineate the regions where the practice applies or where variations need
to be made in the practice. For example, in one soil region a 10-6-4
fertilizer mixture may be the best; in another, a 5-10-5. Such a prob-
lem cannot be solved at one location; the tests will have to be made at a
number of places in each region.
To find responses under actual farming conditions.-Research is
often conducted under "ideal" conditions, or at least under conditions
not usual on actual farms; and the clue to a better practice thus comes
out of nonrepresentative conditions. The problem still remains of de-
termining what results the practice will give on the farm. Response of
a new crop variety at the research station, on soil that has been man-
aged well, may be far different from the response on farms, where fer-
tilizer practices might be variable and quite different from those at the
Or it may be a question not of differences in land conditions but
rather of farmers' operations. Then the testing has to be done not only
on the farms but also by the farmers as well-each with his own imple-
ments and in his own way. When this latter concept is involved, research
tests may indeed be quite inapplicable.
Sukhatmelf cites an outstanding example in India of a comparison
between the results obtained on farms and at experiment stations. His
data show the uncertainty of using research-station results as a measure
/ T. V. Sukhatme. "Assessment of Additional Food Production"
(report of sample survey in Madhya Pradesh, 1949). Agr. Sit. in India
of farm response. He found (1) that the land to which farmers applied
new practices was much better than the average farm land of the area,
and (2) that, despite this fact, farmers' results were much less than an-
ticipated from research-station results. Some causes of such discrep-
ancies between farmer and research-station results are that-
1. Yields from small experimental plots are usually
greater than yields from farmers' fields.
2. Cultural operations (land preparation, seeding, culti-
vating, and harvesting) in research stations are usu-
ally more timely in application and are more thor-
oughly and expertly done than on farms. Besides,
some of the farm implements used at the station may
be different from those used by farmers.
3. Research personnel usually apply a practice more
adequately and with more ample consideration of the
factors that produce success than do farmers, who
perform the practice as a part of their farming op-
To measure the profitableness of a practice.-Before a practice can
be recommended to farmers, the question of its profitableness often
must be answered. For example, you might have this problem: What
quantity of fertilizer is the most profitable for farmers to use, consid-
ering that there may be many unknown factors, other than nutrient de-
ficiency, that are limiting crop growth under actual farming conditions ?
The best quantity of fertilizer for farmers to use, year in and year out,
may often be less than the quantity that gives the maximum yield response.
To measure the variability of benefit.-In determining whether farm-
ers in a certain region should adopt a practice, you must be concerned
not only with the average benefit in the region but also with the consist-
ency of the benefit to the individual farmers. A practice that gives large
Benefits on some farms but none on others is not as safe to recommend
as a practice that gives consistent results everywhere.
To assess a practice when there is no single check.-You know that
experiments at research stations often include comparisons with present
practice. These comparisons are simple to make when there is a single
present practice: the experimenter simply includes a check plot in the
test. But sometimes there is no single present practice. For example,
if each farmer grows his own corn, the experimenter would need as many
checks as there are farms. Or, if farmers have different breeds of cows,
a large number of checks would be needed in order to test, let us say, a
new feeding practice at the research station. It might then be easier to
conduct the tests directly on farms, and the checks will be the different
actual practices on the several farms.
HOW LONG SHOULD EXTENSIVE TESTS LAST?
For some practices the results are influenced as much by weather
as by soil. A practice that gives consistent benefit in both good years
and bad is a safer practice to recommend than one that works only in
good years. Therefore, if the consistency of a practice from year to
year is in question, you will have to repeat the test for several years.
Some practices are likely to be much influenced by yearly variabil-
ity in weather; others are not. Thus, large applications of fertilizer may
be profitable only in seasons with ample rainfall; and an extensive test
to determine the benefit year in and year out would have to be repeated
several years before final recommendations could be made. On the other
hand, feeding an improved ration to calves is likely to have benefits that
are fairly independent of weather; test of such a practice could probably
be accomplished in one year, or even less.
If the extensive test has to be repeated for several years, no hard
and fast rule can be given as to how many years are required: two years
may be enough, but several more may be needed. Your aim should be
to run the test for enough years to take in reasonable extremes of weath-
er. Long-term weather records will help you determine when you have
struck extremes; but, if such are not available, you can resort to the
opinions of extension people and farmers, who have reason to remember
WHAT ARE THE KINDS OF EXTENSIVE TESTS?
For convenience, it seems advisable to distinguish between two kinds
of extensive tests on farms-the result test and the farm experiment.
The result test is undertaken to find the result of applying a single
practice under farm conditions; it is simple, requires little knowledge
of research methods, and has high demonstrational value for farmers.
You are already familiar with a related term, "result demonstration."
A result demonstration is used to demonstrate a result; a result test is
used to test a result. You perform a result demonstration only after the
practice has been proved; you perform a result test before it is proved.
Often, however, you may have to test more than one practice. Then
you will have to conduct a more complex kind of test-the farm experi-
ment. This kind is undertaken to find which of several practices-for
example, several fertilizer formulations-is best under farm conditions.
It requires more application of statistical methods and has less clear-
cut demonstration value. The designing of it requires the cooperation
of technicians with experience in research.
PART II. RESULT TESTS
"It has been said that farmers
are a hard class to reach and im-
press. That is not my experience.
They are the most tractable of
people if you have anything sub-
stantial to offer-but they want
-Seaman A. Knapp.
The result test is closely related to the result demonstration, a tech-
nique already in wide use by extension workers. The result demonstra-
tion helps a community of farmers to learn the benefit of a recommended
practice. Its purpose and technique have been summarized as follows:
"It is used to prove the practical application of basic
facts to farm and home problems and is in no sense ex-
perimental except possibly in the mind of the demon-
strator. With this method the extension worker can uti-
lize the results secured from the adoption of a farm or
home practice or a combination of practices to prove by
comparison the value of the new method. "/
The result demonstration and result test are often quite similar when
carried out in the field. The reason is obvious: often the same kind of
proof may be required to show the farmers the benefit of a practice as
is required to show the technicians. Besides, a result test usually has
fine demonstrational value; it is an effective tool for teaching as well
as for learning.
But the result demonstration, in its strict sense, is used only with
practices of which the technician already knows the true benefit. His
purpose is to give the farmers enough concrete experience with the prac-
tice to convince them of its effectiveness. In the result test the same
technique is used: trials are made on farms and results are evaluated.
But the primary objective is distinctly different. A result test is under-
/ Lincoln D. Kelsey and C. C. Hearne. Cooperative Extension Work,
p. 345. Ithaca, N. Y., 1949.
taken to determine the benefit or effectiveness of an improved practice
under the farming conditions of the community. The result test precedes
issuance of a recommendation; the result demonstration follows it.
the result demonstration the number and location of trials are de-
cided by the need for teaching impact on the community; in the result
test, by the need for measurement of the practice. Usually, though not
always, a larger number of trials will be required for a result test than
for a result demonstration.
The proper design of a result test is important to assure its validity.
We shall therefore point out in distinct steps the procedures you should
follow in setting one up.
STEP 1. DELIMIT THE REGIONS. Begin by deciding on the regions
within your area to which the practice should apply. This is a point that
is easily overlooked; too often we
get into the details of making the
test before we ever decide on the
regions of application.
Of course, if your particular
area is one in which approximate-
ly the same conditions-of soil,
climate, altitude, and so forth-
prevail throughout, you will plan
to issue only one set of recom-
mendations for the whole area and
will therefore have no reason to
delimit subdivisions. But if your
area is made up of dissimilar
regions, each of which may call
for a different set of recommen-
dations, you must define these
Generally you will start with
a map of the entire area in which
you are working. On it you will
mark off the regions where the
new practice is to be tested, i. e.,
the subdivisions for which you
will issue separate recommendations if results justify. These subdivi-
sions are called test regions. In deciding how many of these to have,
you must temper your subdividing with a consideration of practical lim-
itations: The cost of testing in, and the complexity of recommending for,
many diverse regions.
Let us see how this is done. The map shows the subdivisions decided
upon for a corn-variety test. In this hypothetical example the corn-grow-
ing area was first marked off into two main regions-one in the west and
another in the east. These were set up as separate test regions because
each had different growing conditions and would doubtlessly call for sep-
Delimitation of area of application for a proposed
practice, and subdivision of the area into test re-
gions (I, II, III, IV).
In the western region two further divisions were made according to
altitude-the coastal plains (I) and the mountains (II). This was done be-
cause research was finding that not all altitudes favor the same varieties.
Region II contains three soil types, but these were not designated as
separate test regions because they were not extensive and because sep-
arate recommendations for each soil would be too complicated to follow.
The eastern region was divided according to farming practice-dry
farming (III) and irrigated farming (IV). This was done because the vari-
eties that will prove the best under one practice will probably not prove
the best under the other and because farmers can easily follow the dif-
ferent recommendations for irrigated and nonirrigated land.
No further subdivisions were made because the number of tests re-
quired for these four regions was all that the organization could carry out.
Now, after these four regions were delimited, a complete result test
was laid out in each region.
STEP 2. DECIDE ON THE NUMBER OF FARMS. Experience has shown
that you should have 15 to 30 farms in a result test. Fewer farms will
probably give you insufficient in-
formation for a sound recommen-
dation to farmers. On the other
hand, more than 30 farms will
rarely be necessary for a good
appraisal of a practice.
You may wonder why the num-
ber of farms is important.
"Wouldn't one really good test on
a single farm do?" you may ask.
A moment's thought will show
you how dangerous that idea is.
Look at it this way. The farm-
ers of your area, with your guid-
ance, are trying to determine
what benefit they will get from
the new practice. You want to
know this so definitely that you
can issue a clear-cut recommen-
dation to everyone. If you conduct
a test on just one farm, even if
it is a very good test, it will ap-
ply only to that one farm. After
all, soils vary a great deal over a region, and farmers vary too. You
cannot be sure, from the one farm, what results farmers in general
will get. Obviously, you are on dangerous ground in trying to reach a
general conclusion from one case.
Then there's another point to bear in mind if you are considering a
test on just one farm. The farm you use for the test may show a marked
benefit from the practice, may even give you an increase of 50 percent
or more. But no matter how promising the results on that one farm may
be, they still cannot tell you how much the benefit will vary from farm
to farm. The very practice that proved so successful on one farm, may
give highly variable results when tried on other farms; and a practice
that gives highly variable results is not as safe to recommend to farm-
ers as one that gives consistent results. It is sobering to consider that
many a potentially good practice could be thrown completely "out of the
running" just because it failed in a one-farm test.
Consider an example in which Practice A and Practice B are both
tried on an adequate number of farms. As the following diagram shows,
each gives the same average yield, 50 bushels per acre. Judging from
that fact alone, one practice seems just as good as the other. But when
you examine each practice for consistency of result, we see that Practice
B is obviously the safer one to recommend. The lowest yield we get with
B is 40 bushels, but with A we get a yield as low as 10:
Lowest Average Highest
yield yield yield
10 bu. o | 50 bu. 9 0 90 bu.
Lowest Average Highest
yield yield yield
40 bu. 50 bu. 4. 60 bu.
In other words, if you had made just a one-farm test, both practices
might have given the same increase in yield. Here is where testing on
a single farm really falls down: it shows absolutely nothing about the
variability of the benefit over the region. The only way to determine the
variability is to try the practice on a sufficient number of farms.
Although we know from past experience that 15 to 30 farms are re-
quired for most tests, these numbers are, after all, a rather offhand
specification. To be exact, the number of farms you will require in or-
der to have confidence in the answer from your test will depend on two
considerations: (I) How great benefit you expect from the new practice
on the average and (2) how much variability you expect the results to
show. If you expect a large benefit from the practice and fairly consist-
ent results throughout the region, you can cut down on the number of
The little table on the next page will help you decide how many farms
to have. Suppose, for example, that you want to test a new variety of corn
for your region. Let us say that you expect it to give so large a benefit
that it doubles the yield, that is, gives an average increase of 100 per-
cent. Let us say, too, that you expectthe increase to be quite variable
in the region. Then, referring to the table, you will find that 15 farms
or so should participate. Of course it is well to put the test on a few
extra farms besides, since on some farms the test may not be carried
through to the end.
This is about as close as we can come at this point to determining
the number of farms that are required. As you might suspect, there are
more precise methods for making this determination. But they are
If you expect an And, if you expect Then you should
average increase the increase in have this num-
of- the region to be- ber of farms:
Quite variable 10
200% LFairly consistent 7
SQuite variable 15
10% Fairly consistent 10
SfQuite variable 25
50% 7Fairly consistent 15
2% Quite variable 30
z 5 LFairly consistent 20
If you ever get into a costly result test, in which it is important to
determine the best number of farms quite accurately, you might want
to use a more precise method. The procedures are given in Parts III
and IV of this Guide. But for most tests the foregoing table is adequate.
STEP 3. SELECT THE FARMS. The purpose of your test is to obtain
an answer that applies to all farms in the region. Yet it is obviously imn-
possible to have all farms partici-
pate in a result test. The crite-
rion whereby the participating
farms are chosen is important:
the participating farms must be
representative of all the farms.
If they are not, you will never
get results that are correctly ap-
plicable to the region, no matter
how carefully you conduct the test.
One method of selecting the
farms-a method that is no longer
considered valid-is to choose
farms that are thought to be
"typical" of the region. Some-
times you will hear a technician
say, "I selected this farm for the
test because it is typical." Get
away from that idea, "typical."
It is fraught with danger.
First of all, the concept of a
"typical" area is fallacious.
There is no such place A region
:1 ah ^ ..;. p I
laflP~ff* ^a' -^^< t-. --1;;' L~r*-^*^- t*t .^-L-; L~~? ri -* -;~.YL^a r^ A
~, _* *- l-ll *J -. _. .. -* ,
iL< -a2IL4 r *"**ae gnr g r.. *.. .---
Modern and primitive cultivating methods practiced in the same commun-
ity point up the need for tests of applicability. To make sure that results
of a test apply to a whole community, the technician must put the test on
a representative cross section of all farms.
is certain to include many variations-in topography, soil, previous crop-
ping practice, farm size, method of cultivation, farmers' abilities, and
so forth. There is, in truth, no way to define what is typical. No single
plot of ground can represent an entire area, with its many diversities.
There is another reason for getting away from the idea, "typical."
Even if you could select typical farms, you would not be doing anything
worthwhile. For then the results of the test would apply only to these
typical, or average, farms. But in a result test, not only do you need to
know the average benefit of the practice; you need to know also how con-
sistent the benefit is over the range of conditions in the region. To find
this out, you will have to set up the test on a range of farms. Be careful
not to overlook this point: Variability of the response is just as important
as the average of the response.
In view of the nonexistence of typical sites and the need to test a prac-
tice on a range of sites, the only valid procedure is to select farms for
the test that are a fair cross section of all the farms. Only then will you
have farms participating that are a true representation of the farms in
Now, how can you get a fair cross section? The ideal way would be
for you to put all the farmers' names in a hat and, blindfolded, to draw
out the required number. In practice, however, it is usually impossible
to select the farms in so random a manner. For one thing, it may be in-
convenient to obtain a list of all the farmers in the area. Besides, it
may not be feasible to include certain farms.
In selecting the test farms, then, you will have to compromise on the
principle of randomization. You may have to confine them to farms lo-
cated along accessible roads, or to farms with which cooperative rela-
tions can be easily established. The less you confine your choice to a
particular class of farms, however, and the more you take the farms
"as they come," the closer you will be to having a true representation.
Remember this: Every departure from randomization imposes the danger
of bias in the applicability of the results. But this is a risk you have to
take. It will be up to you to decide how far you can wander from random
in selecting the farms without seriously detracting from the applicability
of the results to the region.
STEP 4. DECIDE IF YOU NEED CHECK PLOTS. You must now decide
whether your test requires one or two plots per farm.
If you intend that farmers
should substitute the new practice
for some practice they are already
^ using, you should have two plots
on every farm: one for the new
practice and one for the old. The
following are examples: (1) Sub-
stituting a new variety for an old;
S_ H (2) using fertilizer instead of none;
/f Wi (3) using an insecticide instead of
Snot; (4) trying a new ration for
/cows instead of the old one.
14- The old practice is called the
check, or control. It is the con-
S edition against which the new prac-
tice is to be compared.
SYou may note that the term
S"plot" does not seem to be the
7 one to use for animal tests or tests
in the farm family. You can still
use the term, however, except that
you should realize that in such a
circumstance the plots are groups
of animals or people, one receiv-
ing a new practice and the other receiving the check practice.
Some kinds of comparisons do not permit you to have the check on
the same farm as the new practice; for these you will have only one plot
per farm. For example, if you are testing the effect of pest control on
cattle, you might have to treat a whole herd. Then the new practice is
applied to one farm, and another farm serves as the check. Each farm
becomes a plot; and two farms serve as a single comparison, just as in
the foregoing example, where the two plots were on the same farm.
Sometimes the test is not comparative; the practice under test is not
intended to replace what farmers had previously been doing. Possible
1. Finding out how a new vegetable does in a region.
2. Finding out if rabbit culture pays,
3. Making a survey of yield of wheat.
In such cases you have just one plot per farm. The information ob-
tained from each farm will be the production of the plot rather than the
advantage it has over a check plot. Here, again, if the test is with ani-
mals or human beings, the word "plot" means the group being tested.
ID & I
A farmer and his family join the technician in making plans. Extensive
testing requires farmer cooperation and, often, farmer participation.
STEP 5. PUT IN THE PLOTS. So far you have decided on the number
of farms, selected them, and fixed the number of plots per farm. Now
you are ready to start the actual
field work of putting in the plots.
SChoosing the location of the
one or two plots on a farm will be
influenced considerably by the de-
sires of the cooperating farmer.
He may want to have the plots in
to a road. This is all right.
There is just one point to remem-
ber: Do not select the piece of
ground solely because it is the
best piece of ground on the farm.
Rather, select it without any judg-
ment one way or the other as to
its relative fertility. Your aim
should be to locate the test plots
at random, without any purpose
for your choice as far as soil con-
ditions are concerned. Follow
this thought in locating the plots
on every farm; then the test as a
whole will be a true cross section
of the conditions in the region.
Some plots will be on good ground, some on poor. All in all, you will
have a fair representation of the soils in the region.
When your test requires two plots on a farm-a check plot and a new-
practice plot-you must be careful to guard against bias. Do not pick the
place for the new-practice plot so that it is on more fertile land than the
check. Doing so would be an error; it would invalidate the honesty of the
test, for the results might show very much in favor of the new practice.
The test would be biased and would not show the true benefit of the prac-
tice. You will use the results of the test as a guide in making recommen-
dations to farmers, and you do not want the difference in yield between
the new and old practices to be biased by differences in soil fertility.
You would like, of course, to have the two plots fall on land that is
exactly alike in fertility and other respects. But this coincidence would
be quite a difficult thing to accomplish, if not impossible.
Have the plots as close together as possible and choose them accord-
ing to convenience rather than relative fertility. Rely on repetition of
the plots from farm to farm to equalize whatever fertility differences
may exist between the plots on each farm individually.
The size of the plots is a further consideration. You know that at
research stations most experimental plots are rather mall and all of
the same size. But in most result tests you will make'the plots large;
what is more, you need not have the"im the same gise on all farms. Ob-
serve this rule: Make the plots large enough to permit following the us-
ual farming methods and to make the results plainly esible. Anything
larger than this is unnecessary, and will not make re~lts more precise.
Moreover, the check plot on a farm need not be the same size as the
new-practice plot. As the diagram shows, the new-practice plot can be
just a segment of a field, with the remainder serving as the check.
Possible sample I
harvested in the
check plot L J New-practice plot
In result tests, the check plot may be much larg-
er than the new-practice plot. Sometimes just a
sample of the check plot may be harvested for the
Then, when harvesting, determine the areas of the two plots and con-
vert the yields to an equal-area basis.
In getting data, it may be more convenient to harvest just a sample
out of the check plot. Try to locate the sample next to the new-practice
plot and have it about the same size.
If you harvest just a sample out of the check plot, be careful again to
guard against bias. Do not deliberately select a poorer spot. The way
to avoid bias is to decide on the location of the sample at the very start
of the test, before you see the results. For example, you might decide
on having the sample of the same dimensions as the new-practice plot
and adjoining it on the west.
To be safe, stipulate the position of the sample when you write the
test plan-even before you start the field work.
%~t '4 r-
The farmer measures the benefit of a new practice: "Good program
building provides for evaluation of results."-Kelsey and Hearne.
STEP 6. COLLECT THE DATA. You should make every effort to get
numerical data on the results.of the test; only with such data will you be
able to tell the farmers what a-
mount of benefit they can expect
to receive from the new practice
on their own farms. Try to get
actual measurements. Do not be
content to take merely your im-
pression of the results.
Although you may make a
number of intermediate observa-
tions that are valuable, the ulti-
mate information you want is the
actual yield or other benefit of
the practice. When you harvest
the test plots, be sure that you
also make a record of the size
of each plot. Then you can con-
vert the yields from plots of
different sizes to yields from
plots of the same size.
SHere are some examples of
data that have been collected in
Example 1. Data from part of a corn test.
Example 2. Data from part of a feeding test on milk cows.
New feed Old feed -Production per cow
Farm Production Number of cows Production Number of cows New d Old feed
(Pounds) (Pounds) (Pounds) (Pounds)
Jones 89 4 653 36 2Z 18
Smith 627 20 576 20 31 29
Doe 148 10 211 19 15 11
Example 3. Data from part of a test of kudzu production.
Farm Yield Plot size Yield per acre
Jones 1.79 tons 10,000 sq. ft. 7.8
Smith 598 lbs. 1/10 acre 3.0
Doe 8584 lbs. 0.8 acre 5.4
STEP 7. INTERPRET THE RESULTS. The ultimate outcome of a re-
sult test is the issuance of a recommendation, a statement on how well
the new practice works in the
region. The data you have col-
lected from the several test plots
will now have to be interpreted
t for regional application. How
valid this interpretation will be
depends partly on how valid was
the cross section of farms that
was selected in the first place.
The interpretation of the re-
S- suits involves a number of statis-
tical calculations, but fortunately
none of these are difficult or com-
plicated. Consider that much of
the effort and planning you have
expended thus far will have been
wasted if you do not now bring the
test to its proper conclusion with
a proper statistical analysis of
the data. In fact, much of the va-
lidity of the recommendations
that ensue from the test will de-
pend on your making this analysis.
Table 1 shows how to make
an analysis for a test that is comparative. The data are from a test of
a corn variety. First list the yields from the new and check practices
for each location. The benefit is obtained by subtracting the check yield
from the new-practice yield (note that farm 13 gave a minus benefit).
Enter the squares of the benefits in the last column. Add all columns.
The means, or averages, are obtained by dividing the sums by the num-
ber of farms.
Table l.-Data from a corn variety test on 25 farms
Farm Y.. Benefit Square of
New practice Check benefit
Bushels per acre Bushels per acre Bushels per acre
1.... 23 16 7 49
2.... 37 26 11 121
3.... 24 17 7 49
4.... 20 14 6 36
5.... 28 20 8 64
6.... 39 28 11 121
7.... 8 6 2 4
8.... 17 12 5 25
9.... 28 20 8 64
10 ... 25 18 7 49
11 ... 22 16 6 36
12 ... 22 16 6 36
13 ... 11 13 -2 4
14 .. 21 15 6 36
15 ... 18 14 4 16
16 ... 17 12 5 25
17 ... 37 26 11 121
18 31 22 9 81
19 ... 14 10 4 16
20 ... 21 15 6 36
21 28 20 8 64
22 ... 24 17 7 49
23 ... 22 16 6 36
24 ... 28 21 7 49
25 .. 26 19 7 49
Now you use the sum of the squares of the benefits, 1,236, to obtain a
term known as the standard deviation (s.d.). The following calculations
show how this is done:
Sum of squares of benefit -- -. ------ 1,236
(Sum of benefit) (162) 262,144 00
= --- 1,050
Number of farms 25 25
Difference= 1236- 1050 ---------------- 186
Number of farms minus 1
Standard deviation = square root of 7.75 -- ---
sad. x 100
Standard deviation in percent = check mean
2.8 x 100
= 17.2 = 16%
You are now prepared to consider three pertinent questions in issuing
recommendations to farmers.
What was the average increase in yield from the practice?
Solution: Divide the mean benefit by the mean yield of the
check and multiply by 100; i.e.,
x 100 = 38 percent
Answer: Farmers on the average can expect an increase
of 38 percent or 6.5 bushels per acre.
What is the minimum increase in yield that farmers can ex-
pect 3 out of 4 times ?
Solution: Multiply the standard deviation in percent by 0.73
and subtract the product from the average increase
in percent; i.e.,
16 x 0.7 = 11.2 percent
38 11.2 = 27 percent
Answer: Three out of four times, the increase in yield will
be at least 27 percent.
2/ The number 0.7 here is a mathematical constant used in statistics.
What percent of the farmers are likely to get no increase in
yield from the ne ,practice?
Solution: Divide the mean benefit by the standard deviation
to obtain a ratio, i.e.,
Look up answer for this ratio in the following
table, interpolating if necessary:
Fewer than *
About 1 percent of the farmers can normally ex-
pect to get no increase in yield from the new prac-
The statement you can make for farmers as a result of the test might
be something like this:
"What we consider a good cross section of
farmers have participated in our test, and their
results indicate that the new variety will increase
yield in our region by 6.5 bushels per acre, or
38 percent, on the average; three-quarters of the
farmers can expect at least a 27 percent increase;
and fewer than 1 farmer out of 100 should fail to
get an increase."
We might now illustrate the procedure for interpreting results when
the test is not comparative-when you have data on a practice that does
not compare with an old or check practice. In table 2 we have some data
on the yield of kudzu. Interpreting these yield results is not so very dif-
ferent from interpreting comparative benefit results. Enter the squares
of the yields in the last column and add both columns. The mean yield,
5.5 tons per acre, is the sum of the yields, 99.1, divided by the number
of farms, 18.
Table 2.-Data for a kudzus test on 18 farms
Farm Yield Square of yields
Tuon per acre
1- .-------------- 7.8 60.84
2 --------------- 3.0 9.00
3--------------- 5.4 29.16
4--- -- --------- 5.9 34.81
5 ---------------- 4.2 17.64
6--------------- 5.6 31.36
7--------------- 6.5 42.25
8--------------- 4.9 24.01
9- -------------- 5.4 29.16
10 - -------4,4 19.36
11 ---------- --- 5.1 26.01
12 -------------- 6.8 46.24
13 ----- --------- 5.9 34.81
14 -------------- 4.8 23.04
15 ------------ 6.1 37.21
16 .--- -- ------ --- --5.3 28.09
17 - ----- 6.2 38.44
18 ..- --- -- ------- -5,8 33.64
Sum -- ----------- -99.1 565.07
Mean ------------ 5.5
The standard deviation is found in very much the
same way as in the
Sum of squares of yield ----------- -
(Sum of yield)2 (99.1)2 9820.81
Number of farms 18 18
Difference = 565,07 545.60 -- -- -
Number of farms minus 1
= - -
Standard deviation = square root of 1.1453 - -
s.d. x 100 1.07 x 100
Standard deviation in percent .d = 107 x. 00 = 19%
With data that are not comparative, we might also consider three
pertinent questions about the practice.
What was the average yield of kudzu in the region?
Answer: 5.5 tons per acre.
What is the least yield that three out of four farmers can
Solution: Multiply the standard deviation by 0.7 and sub-
tract the product from the mean yield; i.e.,
1.07 x 0.7 = 0.749 tons per acre
5.5 0.749 = 4.8 tons per acre
Three times out of four, the yield will be at least
4.8 tons per acre.
What is the minimum yield farmers can expect?
Solution: Multiply the standard deviation by 4/ and sub-
tract the product from the mean yield; i.e.,
1.07 x 4 =4.28 tons per acre
5.5 4.28 = 1.2 tons per acre
Answer: It would be very unusual for the yield to be less
than 1.2 tons per acre.
From these results, you would be justified in issuing a recommenda-
tion to the community along these lines:
"Kudzu, when established on a representative
cross section of farms in the region, yielded an
average of 5.5 tons of forage per acre. While
there is some variability over the region, three
out of four farmers can expect at least 4.8 tons
per acre; and practically all farmers should get
1.2 tons per acre as a minimum."
/ The 4 here is another mathematical constant.