Analysis Methods for Assessing Risk in Agronomic Decision Making.

Understanding the risk involved with agronomic decisions is as important as identifying which treatment is best. Agronomists are challenged with datasets of increasing complexity that are often unbalanced and conducted over numerous years. Specific examples of long-term, complex, continuous variable treatments include plant density and date of planting, while discrete variable treatments include tillage and cropping systems decisions. Larger data sets across numerous environments allow for better estimates of risk. Farmers are usually most interested in the mean. The mean describes the first moment of a normal distribution, but three other moments, variance, skewness and kurtosis, are as important and describe the risk involved with the decision. Variance is the “spread” of the data around a treatment mean. Skewness is the frequency of “upside” versus “downside” events. Kurtosis is the frequency of “extreme” events. Most farmers would like to make decisions picking treatments with the best mean and low risk i.e., one with low variance, an equal chance of beating the average, and stable performance. MIXED analysis procedures deal with the complexities of unbalanced data collected over years. For example, in Wisconsin plant densities of 85,000 plants ha^-1 produce greatest corn yields, but university recommendations and most farmers grow corn at 67,000 plants ha^-1. Variance is linear and at 67,000 plants ha^-1 is ± $44 ha^-1 and at 85,000 plants ha^-1 is ± $65 ha^-1. Skewness and kurtosis are equal across the range of plant density treatments indicating an equal probability of upside/downside risk and extreme events. The grower decision is whether $21 ha^-1 at higher plant density is worth the risk. Treatments may have greater return, but risk can be greater as well.