Analysis of alternative models treating herd × year effects as fixed or random

Abstract

The analysis was conducted of the national database of station tests carried out between May 1996– February 2001, using the Hungarian Large White breed. Days of test, total amount of consumed feed and valuable cuts were taken into the analysis. Using the method of cross validation, small subsets of the data were excluded and then predicted using the remaining part of the data treating herd × year effects either as fixed or as random. The size of the data excluded was 50 or 10 records at a time and the process was repeated 100 or 500 times, respectively. Mean squared error, bias and correlation between the excluded and predicted observations were calculated for all the excluded subsets. There was no significant difference between the fixed and random models but in the case of valuable cuts the random models showed a lower mean squared error and higher correlation between the excluded and predicted observations than the fixed models.