By Wayde Looker
Today’s farmer encounters yield trial results nearly everywhere.
Last fall, I sat in the combine, harvesting soybean variety trials, while listening to seed commercials on the radio. Upon my arrival home, I turned on RFD-TV and saw more advertisements highlighting a specific seed’s advantage over its competition.
Just a few days later, I met with my seed dealer and chose hybrids based on third-party and private industry yield data. Simply put, yield data and variety trials are found in every available media outlet. Interpreting and comparing the various trials can be confusing, leading growers to question what yield data are reliable. Fortunately, there are some basic statistics that can make this much easier.
How many treatment replications?
One of the most important things to consider when evaluating any type of research is the number of replications of treatments within the study.
In the case of hybrid or variety trials, the treatment is simply the variety or hybrid being tested. The more times the treatments are replicated, the more confident you can be that the results are accurate and not the result of a fluke.
Agricultural research typically has three to six replications of treatments. Again, this helps to protect against unintended factors, like field variability, influencing the results of the study. Furthermore, treatments should be randomly assigned throughout the field to ensure there is no bias in treatment placement. This randomization also helps reduce the chance that field variability influences the results.
Least significant difference
Least significant difference (LSD) is an extremely important factor to account for when analyzing yield data. It is linked to the confidence level (see below) and will be increased as confidence increases. LSD describes the minimum difference in yield for two varieties to be considered statistically different from one another.
In other words, if an LSD is 5, then a soybean variety that yielded 64 bushels per acre would statistically be the same as a variety that yielded 69 to 59 bushels per acre (64 +/- 5). This is because the difference in yield is likely due to some other effect rather than the genetic difference between varieties.
Yield differences greater than the LSD value mean that genetic factors likely account for the difference in yield, and not just environmental factors. Using the previous example, a soybean yielding >69 or <59 bushels per acre would be statistically different from a soybean yielding 64 bushels per acre.
LSD is closely associated with another important term: confidence level. The confidence level is used to describe how certain the results of a study are attributed to the actual genetic differences in the varieties tested, versus other environmental impacts. Basically, how certain are we that this information is correct?
This term is also what determines statistical significance. For example, if a study uses a 90% confidence interval and has a 5-bushel-per-acre LSD, then one can be 90% sure any two varieties that are separated by more than 5 bushels per acre would be different because of genetic differences instead of environmental variability. Most confidence intervals used in agriculture fall between 70% and 95%.
Coefficient of variance
Coefficient of variance (CV) is another tool used to describe data in yield trials. CV is an indicator of data uniformity. Larger CVs usually indicate that there was more environmental variability within the study, and a lower CV indicates a study with more uniformity and less environmental variability.
Some factors that can influence CV are: different soil types, uneven rainfall or drainage, differences in elevation or even planting conditions.
The basic understanding of statistics outlined above equips producers with some tools to carefully evaluate yield trials and the differences between them. Replications and randomization reduce the likelihood of an outlier or fluke data point influencing the entire study. The confidence interval and LSD help growers know which varieties outperformed others, and how sure they can be those differences are correct. Finally, the CV describes the variability of the study.
Overall, the best studies will report the confidence intervals, have low LSDs, a low CV, randomized treatments and a greater number of replications. Understanding these basic statistical terms can help producers make more informed seed selection decisions.
Looker is a certified crop adviser and a soybean research associate at Ohio State University. Reach him at 740-637-6815 or firstname.lastname@example.org.