BCFC "A Guide to On-Farm Demonstration Research ~ How to Plan, Prepare and Conduct Your Own On-Farm Trials" (2017)
A Guide to On-Farm Demonstration Research (full pdf version)
1.1 What Is Demonstration Research?
FARMER DIRECTED AND GOAL SPECIFIC
Before we can discuss ‘demonstration research’ we must first outline the difference between traditional agricultural demonstration studies and research studies. Typically, demonstration studies are used by Agriculture businesses to demonstrate the benefits of using a new practice or product. The research behind the new practice or product has already been conducted and demonstration studies are installed to show the local producers what could be expected by adopting the new practice or product. Research studies are hypothesis driven experiments that follow somewhat standardized protocols that include replication and randomization.
What is demonstration research (DR)? Within this document, DR refers to using a combination of demonstration styled studies with research elements to answer a simple question with some confidence. The benefit of DR is that it is farmer directed, it can be carried out independently, and it uses the resources a typical farmer would have on-hand. Demonstration research allows a farmer to use a small portion of their land to test and identify ways to better manage their resources in order to increase productivity, or to achieve any farming goal they may have.
While the results of DR are not intended to be published or to undergo rigorous statistical review, it is important for those wishing to try DR to understand the foundations of research and how variability influences results. An understanding of the foundations of research will enable you to achieve the best results with your demonstration research.
1.2 Principles Of Research
SORTING TREATMENT EFFECTS FROM BACKGROUND NOISE
Research is about predicting future responses. For example, rather than observing that Variety A outperformed Variety B last year; research allows a farmer to state with confidence that it is highly likely that Variety A will outperform Variety B every time they are planted under the same conditions.
One of the challenges of demonstration research is to sort out the true effects caused directly by the research treatments versus the effects caused by “background noise”.
Within a field, or barn, or soil profile there exists a population, or an entire group of similar individuals. The population can be alfalfa plants in a field, cows in a barn, or microbes in the soil. Usually the population is so large that it is not possible to examine every individual; therefore, researchers randomly select a sample of the population and that subset is used to represent the entire population.
How well a population can be represented by a sample depends on the sample size. The larger the sample size, the better it will represent the population.
If you are researching a population of 1,000 individuals, would you trust one person, chosen at random, to provide an accurate, representative measurement of the entire population?
Chances are a single individual would not accurately represent the population. What if your sample contained ten people randomly chosen from the population? That would be better than the one, but not as good as a sample of 100. As the sample size approaches the size of the population, the sample will more closely represent the population. The only way to get completely accurate results is to measure every individual in a population; however this would be time consuming and expensive. Instead, we sample populations and make the assumption that if we sample enough, we will have a fair representation of the whole population.
The second challenge for researchers is related to naturally occurring variability.
In order to reduce the effects of variability, each treatment you compare in an experiment should be done more than once, or repeated (statistically known as ‘replication’). There are different ways to replicate an experiment. One way would be to replicate the exact same experiment on many farms at the same time. Or, you could replicate across time, performing the experiment on the same farm year after year.
FOR EXAMPLE, if you conducted an experiment just once, you might wonder "did I get those results because it was a wet/dry/hot/cold year?", or "were those results specific to this field? What would happen if I conduced this research at different locations?".
Replication across the landscape, or over time, helps you to determine if results are due to your research treatments or due to naturally occurring variation. If one practice is superior to another, it will become evident if you make enough repeated comparisons. In fact, the benefit of one practice compared to another has to be significant enough to overcome the effects of natural variability in order to be worth considering. Figure 1 demonstrates one example of natural variability. Within a uniform field, under identical management, there was up to 9 bu/acre difference between strips adjacent to one another, and more than 15 bu/acre across the whole field. In this example, the variability may be due to differences in soil nutrients, soil moisture, or some other factor.
As Figure 1 demonstrates, there can be a lot of variability within your field due to factors such as differences in soils, topography, or historical management. It is not practical to try to avoid this variability because some level of variability will always be present; instead you can incorporate natural variability into your DR.
The most effective and practical way to reduce the impact of natural variability is to have a long strip for each treatment area. Generally, the larger, in particular longer, the treatment area is, the better the results are likely to be. IF possible, 500 ft or longer is recommended for each treatment area.
Within this document, we will outline simple research designs intended to test one treatment against another, or A versus B. Farmers wanting to design complex demonstration research or incorporate replication within their field are advised to consult a government agent, university researcher, or a consulting scientist for guidance with experimental design and statistical analysis of data collected.
To read the full 68 page document, please click on the link at the top of this page.