Evaluation of modeling procedure for fitting in situ feed degradation profiles
2009
Zanton, G.I. | Heinrichs, A.J.
In situ degradation of feeds is a common methodology used for describing the digestion rate and potential for many chemical components of ruminant feedstuffs. The inferences involving the parameters of these procedures may be of questionable accuracy and precision because nonlinear regression techniques that are employed to fit these models do not produce unbiased, normally distributed, and minimum variance parameter estimates until the sample size becomes very large. Therefore, the objective of this experiment was to determine the accuracy and precision of parameter estimates determined by a 2-stage approach (TS; in which individual kinetic parameter estimates are obtained in the first stage and population estimates are obtained by arithmetic mean in the second), a 2-stage, geometric mean approach (GEO), and by nonlinear mixed modeling (NLMM) using simulation techniques. An experiment using corn silage NDF was used as a reference for simulating parameter and error structures. For all analyses, 500 experiments were simulated with sampling time points of 1, 2, 4, 8, 16, 24, 48, and 72 h. Increasing levels of animal and bag replication were evaluated at 2 levels of residual variability [residual SD of 2.95 (observed from the data; R² = 0.985) and 5.90 (2x observed; R² = 0.950)]. Evaluation of the first-order exponential model with discrete lag revealed that median bias of TS and GEO parameter estimates were more sensitive to increased residual variation and degree of replication than NLMM. When residual variation was less, precision of parameter estimates was less affected by estimation procedure; when residual variation was greater, precision of parameter estimates was generally improved by NLMM. From the results of this experiment it is concluded that, in most cases, NLMM is equal or better than TS or GEO from the perspective of minimizing bias and maximizing precision of parameter estimates.
显示更多 [+] 显示较少 [-]