Pasture & Range: January 2003

In my last article, I pointed out that variability is a characteristic of all biological systems. This characteristic arises from genetic variability inherent in all plants and animals, even among the same species. For instance, think about the frame size variability that might exist in 100 head of steers of the same breed. This variability, or lack of uniformity, also exists in the environment — for example, consider the high degree of variability that exists across the landscape in most soils.

In a simple example, we might want to determine if a new rye variety yields more forage than an old standard variety. If we plant the new and the old varieties in side-by-side, identically sized plots, and then measure the forage yields, what would we measure? We would measure the difference in forage yield between the varieties, plus the difference in the environment and soils of each plot, plus any differences in the way we conducted the work, including planting, fertility, clipping height of the forage, etc.

Since these other factors affect yield, we must use a procedure that can separate the yield difference from other sources of variation. That is, we must be able to design an experiment that allows us to decide whether the observed difference is caused only by the rye varieties or by other factors. In this example, with only two plots planted to different rye varieties, we do not have enough information to determine if the measured yield differences are real or are caused by environmental variation or by a lack of uniformity in how we conducted the work.

The way to decide if the difference is real is relatively simple. The yield of these two adjacent plots will only be considered different in their yield character if the observed yield difference is larger than the variation that would be expected if both plots were planted to the same variety. This latter difference is a measure of what we call experimental error. Experimental error is the difference among experimental units treated alike, where experimental units might be plots of land, an individual animal, or groups of animals. We must design an experiment that allows us to obtain an unbiased estimate of the effect of interest (e.g. forage yield) and an estimate of the experimental error. By obtaining estimates of both of these measures, we will be able to separate the experimental error from the answers we seek. In order to accomplish this, our design must include replication and randomization.

Replication is necessary to obtain multiple experimental units treated alike, which, as mentioned previously, allow us to measure the experimental error. The random assignment of treatments to experimental units is necessary to obtain unbiased estimates of both the effect of interest (e.g., forage yield) and the experimental error. The term "unbiased" is another way of saying we are going to give each treatment a "fair" chance. By randomly assigning treatments to the experimental units, we do not expect to favor any treatment. We would obviously favor, or bias, the results if we selectively planted one rye variety in the "best" soil and the other in "poor" soil areas.

A well-designed experiment will incorporate all possible ways of minimizing experimental error, because the ability to detect differences among treatments increases as the size of the experimental error decreases. For example, if the true yield difference between the old and new rye varieties is 50 pounds per acre, and the experimental error is 100 lb/acre, you will never detect the yield difference.

It is important to keep in mind that the results from research are only as good as the integrity of the researcher conducting the work. There is no question that proper experimental design is important, but it is also true that proper analysis and interpretation are equally important. It is my goal to ensure that the results we report come from experiments that are well designed and properly analyzed and interpreted. In my next article, I will give you some information regarding the interpretation of results so that you will be able to evaluate research findings.