Direct Validation of AMMI Predictions in Turfgrass Trials

Abstract

Previous studies using turfgrass quality (TQ) data from the National Turfgrass Evaluation Program (NTEP) used cross validation with additive main effects and multiplicative interactions (AMMI) models to demonstrate accuracy gain within a single trial. The objective of this study is to test AMMI predictive accuracy for new years or new years × locations. Five winners recommended by the raw data AMMIF and five winners recommended by the more accurate AMMI5 model were selected, resulting in a 10‐entry roster at six validation sites for evaluation over 5 yr. Average TQ for AMMI5 selections was 5.57 and was significantly greater than AMMIF selections averaging only 5.31. In five of six validation locations, higher TQ was observed by planting AMMI5 selections over AMMIF. The correlation between the 60 validation observations and AMMI5 predictions was 0.628 but for AMMIF was 0.479. Better predictive success indicated by a significantly lower mean squared deviation (MSD) was observed with AMMI5 (0.453) than AMMIF (0.608). The AMMI5 model predicted winners were five times as effective as AMMIF in predicting observed winners from the top group (rank one and two). There was decisive evidence of a parsimonious AMMI model increasing predictive success across years despite a meager 1.24 statistical efficiency with AMMI5.