An Evaluation of Estimators of Location for Maize Yield Comparisons1

Abstract

The arithmetic mean's performance as an estimator is well regarded under optimal circumstances, but has received criticism as being nonrobust under less than ideal conditions. This study evaluated how well eight estimators of location performed in comparing maize (Zea mays L.) hybrid yield performance at early stages of testing. Data from five hybrids grown together in 123 tests were taken from commercial yield trials in 1981 and 1982. Eight statistics based on five trials were used to estimate four orthogonal comparisons of hybrid yields and were evaluated for their ability to predict mean differences based upon 118 other trials. The arithmetic mean, the 10% trimmed mean, and a rank‐order weighted mean were the best estimators in the study. The 20% trimmed mean, an adaptive estimator, Huber P15, and a squared deviation weighted mean performed almost as well as the three best statistics. The median was the worst estimator. This study supports the use of the mean, compared with the other seven estimators investigated, for comparing maize hybrid yields at early stages of testing.