NUMBER OF COLLECTIONS OF MILK YIELD ON PREDICTION OF MODELS TO ESTIMATE AVERAGE DAILY MILK YIELD OF NELLORE BEEF COWS

Abstract

The objective of the experiment was to evaluate models with different number of collections of milk yield to estimate average daily milk yield in Nellore beef cows until 217 days of lactation. The eight evaluations of milk yield were taken from 29 cows, by direct method through milking, at 14, 42, 70, 98, 126, 154, 182 days and weaning (217 days). For the selection of multiple linear regression models, with objective of estimate the average daily milk yield according to number of milk collects, Stepwise procedure of SAS was used. The data were submitted to residual analysis by tests of heterocedasticity of variance (?² statistic), normality (W statistic of Shapiro-Wilk) and diagnosis of outliers (three observations were excluded from original n=30), beyond multicolinearity diagnosis. For validation of the selected regression models, the prediction error sum of squares statistic (PRESS), was used. The selected days of milk yield collections were: 98 for 1 collection; 70 and 98 for 2 collections; 14, 70, and 98 for 3 collections; 14, 70, 98 and 154 for 4 collections; 14, 70, 98, 154 and 182 for 5 collections; 14, 42, 70, 98, 154 and 182 days for 6 collections and 14, 42, 70, 98, 126, 154 and 182 days for 7 collections. The adjusted R² for models with 1; 2; 3; 4; 5; 6 and 7 collections were, respectively, 0.7900; 0.8718; 0.9064; 0.9596; 0.9698; 0.9875 and 0.9972. All the predicted models were satisfactory to estimate average daily milk yield. In a possibility of to execute various collections of milk yield during lactation period, three milk collections chosen in strategic dates during lactation period were sufficient to estimate with high precision the average daily milk yield.

KEY WORDS: Lactation, maternal phase, multiple linear regression, selection of independent variables, residual analysis, techniques of validation of regression models.