Investigation into the use of Residual Maximum Likelihood (REML) for the statistical analysis of winter oilseed rape variety trials

Abstract

In 1994 HGCA commissioned a statistical review of the UK Recommended List winter oilseed rape trials programme. The final report of the review contained recommendations relating to the analysis of the data. The major of these recommendations was that the Residual Maximum Likelihood (REML) algorithm in Genstat should replace fitting constants as the preferred method for the statistical analysis of the major variates, and in particular for yield. However, problems were anticipated with the implementation of these recommendations and the routine use of REML.

This project addresses those problems and evaluates the effect that changing to REML has on variety performance ratings and statistical errors.