Comparison of Linear and Curvilinear Decreasing Terms in Logistic Flock Egg Production Models

Abstract

The Adams-Bell (logistic-linear) and logistic-curvilinear flock egg production models were fitted to weekly hen-day egg production data and to the first 24 wk of data from 45 first-cycle flocks using the Marquardt method of the NLIN (nonlinear) procedure of SAS. Future egg production was predicted by extrapolating from 24 wk fits for the nonlinear models and from linear regression from peak production to 24 wk. Mean R2 value was significantly higher for the Adams-Bell model than for the logistic-curvilinear model using all data, but mean R2 was not different for fits to 24 wk. Mean percentage for error of prediction was significantly higher for the logistic-curvilinear model than for Adams-Bell, indicating a high prediction bias. Mean percentage error for Adams-Bell was significantly higher than for linear regression. Mean absolute error for the logistic-curvilinear model was significantly higher than for Adams-Bell, which was not different from linear regression. For predicting future egg production, linear regression after peak production is the simplest method and works as well or better than nonlinear models. For modeling the entire flock egg production curve, including the exponential rise to the peak, the Adams-Bell model is preferable, because its linear decreasing term provides better fits than the curvilinear (e–bt) decreasing term.