Prediction of the voluntary intake of grass silages by beef cattle. 2. Principal component and ridge regression analyses

Abstract

ABSTRACTData on individually recorded silage intake (SDMI), concentrate intake and live weight of steers and data on silage composition including toluene dry matter, pH, total nitrogen, ammonia nitrogen, volatile fatty acids, digestibility and fibre measures obtained from experiments at three sites were used.Correlation and principal component analyses indicated that there was severe collinearity among a number of the variables, particularly among various fermentation characteristics, between different measures of digestibility and between different measures of fibre. This collinearity was shown to have caused instability in least-squares multiple regression coefficients of silage intake on other variables previously obtained from the same data (Rook and Gill, 1990). Principal component regression and ridge regression allowed the derivation of new coefficients which were more stable and more in line with a priori expectations from experimental results. Linear functions of the original variables and the use of models containing fewer variables also proved effective in overcoming the problems of collinearity.Fermentation characteristics were shown to be the most important silage factors affecting intake and butyric acid was shown to be more important than other volatile fatty acids or ammonia nitrogen. Neutral-detergent fibre was found to be better related to intake than other fibre or digestibility measures.