Selection Methods for Models to Predict Feedstuff Associative Effects in Goats

Abstract

Abstract Animal nutrition models can be useful to gain an understanding of factors responsible for and predict biological responses. However, specific models developed can vary among statistical methods employed, especially with relatively small datasets. In this meta-analysis study, independent variables selected by regression tree, stepwise regression, and Least Absolute Shrinkage and Selection Operator (LASSO) analyses were compared. The database consisted of 135 treatment means (weighted by the number observations) from 25 publications in which goats consumed forage ad libitum with or without supplementation and was divided into three subsets of forage with a crude protein (CP) concentration < 6% (Low; n = 46), 6–10% (Moderate; n = 50), and > 10% (High; n = 39). Regression tree analysis was conducted with rpart of the R statistical programming language and stepwise regression (proc stepwise) and LASSO (proc glmselect) were conducted with SAS. The target variable was forage metabolizable energy (ME) intake relative to metabolic body weight (BW0.75) and potential predictor variables were supplement ME intake also scaled to BW0.75, forage organic matter (OM) digestibility and neutral detergent fiber (NDF) concentration, and supplement concentrations of ME and CP. As shown in Table 1, supplement ME intake was selected with each method. Based on the order of selection, forage NDF concentration had larger impact than supplement ME intake with the Low dataset, whereas supplement ME intake was most important with Moderate and High datasets. Among the statistical methods evaluated, selected variables varied most with the Low dataset, were similar for the Moderate dataset, and were the same for Stepwise and LASSO approaches with the High dataset. In conclusion, model development to predict feedstuff associative effects in goats requires careful attention to variable selection that is impacted by statistical method and varies with forage composition.