Estimating and Predicting Feed Conversion in Broiler Chickens by Modeling Covariance Structure

Abstract

Modeling covariance structure was used to estimate and to predict feed conversion in broiler chickens from one experiment under repeated-measures design. Eight treatments that consisted in a combination of four strains (Arbor Acres, Ag Ross 308, Cobb and RX) and two sexes were evaluated at six ages (7, 14, 21, 28, 35 and 42 d) in two blocks with three replicates per block. Feed conversion was subjected to a mixed model, MIXED procedure in SAS^® software, where was modeled covariance structure using ten types. Also, it was obtained a correlogram and analyses of variance for each structure. Means±standard errors were estimated and polynomial trends were assessed using linear, second- and third-order to predict the trait over ages. First-Order Autoregressive Moving-Average was chosen the best covariance structure that is extremely important to obtain more accuracy of estimate (from 1.048 at 7 days to 1.703 at 42 days for Cobb) and predicted (from 1.061 at 7 days to 1.577 at 42 days for Cobb) means on feed conversion, which is better predicted when linear effect is used because it presented very closely both estimate and predicted means at all ages, except at 14 days. Modeling covariance structure allowed us to choose the best model to estimate and to predict feed conversion, opening the windows to understand a little more about its trend over ages and also to purpose nutrition managements that may be adopted to maximize growth and to minimize total cost of poultry production.