Coefficient of Variation in ANOVA
The coefficient of variation (CV percent) is a normalised measure of trial precision. In agronomy it is reported alongside every ANOVA table because it tells the reader how noisy the experiment was relative to the response level. Acceptable thresholds vary by crop and trait.
How CV is computed
CV percent equals the square root of MS_error divided by the grand mean of the response, multiplied by 100. It is unitless. A CV of 5 percent means the residual standard deviation is 5 percent of the average response, which is small. A CV of 30 percent is large and signals either a noisy trait or a poor trial.
Field trial benchmarks
Conventions vary by crop and reference, but agronomy textbooks (notably Gomez and Gomez, Statistical Procedures for Agricultural Research) describe CVs around 6 to 8 percent as acceptable for cereal grain yield trials, around 10 to 12 percent for vegetables, and 15 percent or higher as a warning sign for any controlled trial. Breeding nurseries with single-row plots commonly run CVs of 12 to 18 percent and that is treated as the cost of low replication.
What to do if CV is high
A high CV is not by itself a reason to reject a trial, but it should prompt investigation: check for outliers and recording errors first, then consider whether blocking was effective, whether plot size was adequate, whether the response should be transformed (square root or log for counts and skewed traits), and whether the trial should be repeated. Reporting CV transparently is a sign of a careful analyst.