Simulated On-Farm Research: A 15-Day Class Exercise
Peter E. Hildebrand1
Food and Resource Economics Department
University of Florida, Gainesville FL 32611-0240
A simple, rapid and inexpensive procedure for simulating on-farm research is described. The exercise has
been used for many years in a graduate-level farming systems methods course and has been found to be
effective in giving students a feel of the need for and the requirements of on-farm research. An example
analysis is included.
Very often, particularly in developing countries, but also in North America and Europe, farmers do not
adopt new farm technologies because the indicated responses are not achieved when applied on farms and
under farm conditions and farmer management. In order to make realistic estimates of expected responses,
and to help sales persons and extension agents make more specific recommendations, on-farm research is
an accepted procedure. However, there is often a temptation by researchers accustomed to experiment
station conditions to either reject outright data from farms because the trials cannot be controlled like they
can on stations, or to think of farms as replications and analyzing results by analysis of variance (ANOVA)
as if they were from station data.
The purpose of on-farm trials is not to determine which factors are responsible for different responses such
as is desired from on-station trials, but rather to ascertain under what biophysical and socioeconomic
conditions different technologies are more appropriate, therefore making specific recommendations
feasible. Treatment by environment interaction requires a very different approach, design and analytical
procedure for on-farm research. Because of the diversity among farms, it is necessary to sample a wide
range of biophysical and socioeconomic conditions in on-farm trials. A relatively simple procedure for
designing and analyzing on-farm trials to achieve this is "Adaptability Analysis" (Hildebrand and Russell,
A simulated on-farm trial based on Adaptability Analysis and its predecessor, Modified Stability Analysis
(Hildebrand, 1984; Hildebrand and Poey, 1985), has been used for more than 20 years in the farming
systems research-extension methods graduate course, ALS 5813, at the University of Florida. For several
years, nitrogen response of radishes, with a five-week requirement between planting and harvest was the
type of trial used. Based on student evaluations, it was one of the highlights of the course even though it
required two Saturday mornings outside normal class time to complete.
The result of one student's PhD research (Parera, 1990) convinced the professor that it would be more
efficient to use several cultivars or lines of sweet corn and measure only seedling emergence. This cut
duration of the trial to two weeks, reduced significantly the amount of land required, eliminated the cost of
purchasing fertilizer, and allowed the trials to be installed and harvested during the two-hour, regularly
scheduled class period instead of an entire Saturday morning.
To simulate the wide biophysical diversity found among farms, the exercise requires an area of which part
has shade in the morning, part has shade in the afternoon, part has no shade, and part has no sun. An area
with trees on the east, south and west (in the northern hemisphere) works well, although shade screen could
also be used. However, screen would require annual installation and removal, plus added cost for materials.
Each of these sun/shade environments represents a village or community from which several farms are
incorporated in the exercise. To generate socioeconomic diversity of farm environments, half of the farms
in each village are irrigated and half are rainfed. Depending on the number of students in the class, they are
divided into teams of one or three persons to install the trial, take data and analyze the results. Each team
represents a farm in each of the four environmental (sun/shade) locations. One trial for a class of 12
students divided into four teams of three persons each is described for one year as follows:
"Sweet corn is an important seasonal staple in the northern part of the country of
Floriland. The producers eat it, but there is also a very active market for urban
consumption. Sweet corn is best when eaten shortly following harvest because the quality
and taste decline rapidly as sugar in the kernels is converted to starch. Sweet corn
varieties containing the supersweett" gene have a high sugar content at edible maturity
and a low sugar to starch conversion rate, both of which increase shelf life, postharvest
eating quality and market value. However, the use of supersweett" varieties has been
limited in Floriland due to poor germination and subsequent poor stand establishment
attributed to low seed vigor and susceptibility to soil borne diseases (Parera 1990). Poor
germination and emergence in the supersweet varieties are further aggravated by stressful
environmental conditions. To compensate, more seed could be planted per ha, but
supersweett" seed is expensive, and Floriland's farmers have little cash available at time
of sweet corn planting.
"On-farm" trials will be conducted in Floriland to test the effects of cultivar and
environment on sweet corn emergence. The research domain consists of four villages
situated on the 1) western, 2) eastern, 3) northern and 4) southern slopes of the only
mountain in Floriland. Because Floriland is in the northern hemisphere, the villages
receive: 1) AM Shade, 2) PM Shade, 3) Full Shade, or 4) No Shade, respectively.
Included in the trials will be a hybrid cross of a variety popular many years ago (Golden
Cross Bantam), the most common local hybrid (Silver Queen), a related yellow hybrid
(Golden Queen), and one material with the supersweet gene (Florida Staysweet). Four
farmers will be selected in each of the villages to participate in the trial (each team in the
class will represent a "farm" in each village). Because some fields in each village are
rainfed and some are irrigated, two rainfed and two irrigated fields will be selected in
each village in order to help assure sampling a wide range of environments.
"In each field, four contiguous 60 x 60 cm plots will be selected. In each plot, 25 seeds
from one cultivar will be planted in a 40 x 40 cm grid (seed spacing is 10 x 10 cm). A
template will be provided so locations for the seeds can be marked with a pencil or stick.
One person from each team should plant all 100 seeds in a singlefield to avoid
unnecessary experimental error associated with different practices in the samefield. Mark
the center of each field with a flag indicating team number (farm name) and whether the
field is irrigated or rainfed. Before moving on to the next village, draw a map of the
village with location of all four fields (by name and whether rainfed or irrigated) included
in the trial. In each field, indicate location of each oftheplots by variety. Planting date is
Seedling emergence (%) by variety, village and field will be recorded four, seven, and
eleven days after planting, on March 20, 23 and 27. The data will be analyzed by
Results and Discussion
Results from the exercise in spring 1998 are reproduced in Table 1. Data are sorted by the environmental
index, EI, from lowest to highest. Following analysis procedures in Hildebrand and Russell, the response of
each variety is regressed on EI. Results are shown in Figure 1.
Based on the regression analysis for percent germination from these simulated on-farm data as well as an
analysis of risk from the same data (Figures 2 and 3) several conclusions can be made and recommendation
domains (See Hildebrand and Russell, 1996) described for 'Floriland.' Golden Cross Bantam is not adapted
to any of the environments sampled on "farms" in Floriland and should not be recommended. On rainfed
fields of the north slope of the mountain (all shade) and for all fields on the west slope (AM Shade)
planting is risky but Silver Queen and Gold Queen can be recommended if farmers want to plant sweet
corn in those locations. However, seed per acre would have to be doubled to achieve an adequate
emergence density two weeks after planting. In general, sweet corn did well in the other three communities
(plus irrigated fields on the north slope) with Silver Queen outperforming Florida Stay Sweet slightly. If
farmers want to plant the super sweet variety, it can be recommended except for the better environments,
but would require about 10 percent more seed per acre. Cost considerations should also be considered.
When coupled with the design and analysis procedures in Hildebrand and Russell, this achieves a very
realistic simulation of an on-farm trial. It allows students to understand the purpose of utilizing a wide
range of environments to help them ascertain under what biophysical and socioeconomic conditions
different technologies are more appropriate, therefore allowing them to make specific recommendations.
Hildebrand, P.E. 1984. Modified stability analysis of farmer managed, on-farm trials. Agronomy Journal
Hildebrand, P.E. and F. Poey. 1985. On-farm agronomic trials in farming systems research and extension.
Lynne Rienner Publishers, Inc.
Hildebrand, P.E. and J.T. Russell. 1996. Adaptability analysis: A method for the design, analysis and
interpretation of on-farm research-extension. Iowa State University Press.
Parera, C.A. 1990. Improved seed germination and stand establishment in sweet corn carrying the sh2 gene.
PhD Diss. University of Florida, Gainesville FL 32611.
D:\Reports\Two week simulated...
Table 1. Percent emergence, spring 1998, sorted by
environmental index, El, in ascending order.
Village Farm RF/IRR SQ GQ GCB FLSS El
All shade 4 RF 28 16 36 16 24
All shade 3 IRR 52 40 4 32 32
All shade 1 IRR 36 76 4 32 37
All shade 2 RF 68 60 4 64 49
PM shade 3 RF 64 56 40 64 56
PM shade 1 RF 48 80 56 48 58
AM shade 4 IRR 68 80 36 68 63
No shade 3 RF 84 60 36 84 66
No shade 4 IRR 80 72 44 84 70
PM shade 4 IRR 84 84 28 88 71
AM shade 1 RF 84 84 24 96 72
PM shade 2 IRR 92 88 40 68 72
No shade 2 IRR 100 84 32 80 74
AM shade 2 RF 84 84 64 72 76
No shade 1 RF 96 84 32 92 76
AM shade 3 IRR 100 84 64 96 86
Figure 1. Regression of individual variety response to environment
as measured by EL
Figure 2. Estimating risk to farmers of low emergence on west facing
slopes and rainfed north facing slopes (recommendation domain A,
comprised of poorer or lower-yielding environments).
Figure 3. Estimating risk to farmers of low emergence on south and
east facing slopes and irrigated north facing slopes (recommendation
domain B, comprised of better or higher-yielding environments).
Sweet Corn Emergence
0 20 40 60 80 1C
Risk to farmers of low emergence:
Low yielding domain (6 lowest environments)
E 30 *- GQ
20 A GCB
) [-o FLSS
0 Ii I
-10 5 10 15 20 25 30
Number of years of lower emergence
out of 100 years
Risk to farmers of low emergence:
High yielding domain (10 high environments)
0 5 10 15 20 25
Number of years of lower emergence
out of 100 years