On-farm research is nothing new to farmers. However, many times much effort and expense is put forth to conduct comparisons that are not valid. On-farm research must be practical for the farmer by using plots that are field-scale with standard farm machinery. So that the results of the comparison are not misleading, statistically valid designs should be used. In this way, field variation and other factors will not discredit the reliability of the results. Successful on-farm research begins with thorough planning before going to the field. By using the following guidelines, it is hoped that farmers can conduct research on their farm to answer their production and management questions.
Simplify your research objective into a single researchable question that makes a comparison. Many test plots are too complicated because they are looking at too many treatments. The problem enters when field conditions or other variables cannot be separated from true differences in treatments. On-farm research is most successful when comparing just two practices. For example:
"Is the profit from corn fertilized with 200 lbs. nitrogen per acre different from that of corn with 150 lbs. nitrogen." Now the farmer can identify the type of treatment to set up and the data to be collected to answer the question.
Randomizing and replicating are the key to laying out a scientifically valid plot. This procedure is what separates a purely demonstration plot from one which can be used to make valid conclusions. When comparing two treatments, they must be repeated in side-by-side strips across the field. To adequately overcome field variations, each pair of treatments should be repeated four times, although six is better. To further overcome field variations, the treatments should be randomly located within the pair. Always having treatment A on the left and B on the right may favor one treatment. The following example shows one method of proper plot layout.
Pair 1 2 3 4 5 6 ^ ^ ^ ^ ^ ^ A B, B A, B A, A B, B A, A B
Treatment A = 200 lb N,Treatment B = 150 lb Often a planter is used to layout a plot. One treatment or hybrid will be applied on one side of the planter and the other on the other side. To achieve randomization, a few more strips should be planted and then skipped during data collection and harvesting.
Strip Treatment Pair 1 2 3 4 5 6 ^ ^ ^ ^ ^ ^ A B, A B A, B A, B A B, A B A , B A B | | | | S S S S K K K K I I I I P P P P
The above are examples of Randomized Complete Block Design experiments.
Choose a site that is as uniform as possible. Whenever possible, avoid fields that have variable soil types, slopes, irregular boundaries, and tile lines running parallel with the rows. Longer, field-length strips are preferred to reduce variability in the test.
Plots should be as narrow as possible but still be convenient to plant and harvest. Border rows are needed on each side of the plot to avoid edge effects. Be sure to flag the plot and record treatment locations on a plot map.
Plots should be monitored frequently during the growing season. Record and date your observations in a notebook for safe keeping. Much useful information can be gathered including- emergence, stand, weed and insect damage, soil conditions, and weather conditions. All crop inputs also need to be recorded. Additional data to collect may include spring soil nitrate, tissue tests, fall corn stalk nitrate tests, and yield. Farmers need to be aware of problems that arise which may make the plot unusable or eliminate some strips, such as weed patches or misapplied crop inputs. The best designed field research is of little value if data is not collected accurately.
Statistics are used to determine if the data collected from each treatment is due to treatment differences or due to chance. Calculation of the L.S.D. (Least Significant Difference) will show the minimum difference needed between treatment average results to be considered a real difference and not due to chance. The probability that the difference between treatments could occur by chance is called the p-level. A p-value of 95% means that the probability is only 5% that the difference between the treatments could occur by chance. Using the above example of a randomized complete block design, farmers can use a computer statistical program to calculate the L.S.D. A simple, user-friendly program called AGSTATS is available by sending a disk and postage return mailer to Russ Karow, Crop Science Building 131, Oregon State University, Corvallis, Oregon 97331-3002.
If the average difference of the treatments is greater than the calculated L.S.D. then we have some confidence that the difference is real. But is it the best choice economically? Is the advantage given to one treatment worth the cost difference between the treatments. A farmer should analyze his records to determine where cost differences occurred such as seed, fertilizer, pesticides, tillage, and management time between the treatments. The additional income benefit can then be weighed against the possible increased costs for that treatment. Non-tangible benefits such as improved soil quality and environmental improvement also are to be considered. General conclusions are more safely drawn from trials repeated in more than one location and year.
Zaborski,E., Conducting On-Farm Research Trials Using Randomized Complete Block Design, Dept of Emtomology, O.A.R.D.C.,
Janke,R.,Thompson, D.,Cramer, C,McNamara K., A Farmer's Guide to On-Farm Research, Rodale Institute
Nielsen,R,. Fundamentals of On-Farm Research & Demonstration, Agronomy Dept., Purdue University, W. Lafayette, IN
Nafziger,E., Field Comparisons: Two-Treatment Trial Design and Analysis, Dept. Crop Science, Univ. Of Illinois
All educational programs conducted by Ohio State University Extension are available to clientele on a nondiscriminatory basis without regard to race, color, creed, religion, sexual orientation, national origin, gender, age, disability or Vietnam-era veteran status.
Keith L. Smith, Associate Vice President for Ag. Adm. and Director, OSU Extension.
TDD No. 800-589-8292 (Ohio only) or 614-292-6181