Tree-Based Methods as an Alternative to Logistic Regression in Revealing Risk Factors of Crib-Biting in Horses

Abstract 

Determining the risk factors might help in designing prevention of crib-biting. Logistic regression is a commonly used statistical method for finding risk factors, but tree-based methods are also getting more popular. An important difference between these two statistical approaches is that logistic regression makes a number of assumptions about the underlying data, whereas tree-based methods do not. Another difference is that logistic regression can be used to derive odds ratios for the significant risk factors, whereas tree-based methods create a tree where the ramifications represent the risk factors. The probability of occurrence is assigned to each end of branch in the tree. Data of horses used for noncompetition purposes were analyzed with three statistical approaches: logistic regression, classification tree, and conditional inference tree methods. By this, we compared the advantages and disadvantages of these statistical methods. No difference was found between the two tree-based methods regarding the structure and prediction accuracy of the trees. Compared to them, logistic regression revealed fewer risk factors, and also the number of the stereotypic horses classified correctly by the model was less. The representation of the tree-based methods is closer to medical reasoning and also high-order interaction of the risk-factors can easily be visualized. Our results suggest that tree-based methods can be a new alternative in revealing risk factors, even if used alone or together with logistic regression.

Keywords: Crib-biting, Risk factors, Classification tree, Conditional inference tree, Logistic regression