The dissertation aimed to develop a practical method for analyzing non-normal data in the Agricultural Research Department. The researchers modified a common statistical algorithm to handle a wider range of random structures in their models. By using a new estimation procedure called IRREML, they were able to extend the capabilities of generalized linear models to better fit their data. This approach involved updating weights and using restricted maximum likelihood estimation. The researchers successfully implemented this method in Genstat 5 software, making it accessible for widespread use both within and outside the department.