Statistical analysis of lamb survival during the rearing period with the use of classification trees and logistic regressions

Abstract

The aim of the presented research was to statistically analyse the survival of 44 133 Polish Merino and Polish Merino in Old-Type lambs between birth and the 100th day of their life, using classification trees and logistic regression. The study included lambs born between 2008 and 2017 and used in 43 flocks in Pomerania and Kujawy region (Poland). The results showed that 9.27% of all controlled lambs did not survive till the 100th day of life. The statistical analysis of the case of lambs’ death during their first 100 days of life was carried out using multiple logistic regression as well as classification trees, using two algorithms CART and CHAID. The quality of multiple regression and decision tree models was compared considering the following criteria: percentage of misclassifications, average squared error and the area under the Receiver Operating Characteristic curve. The calculated quality criteria for tree models that were created during the research suggested that the classification trees formed based on CART algorithm were the most accurate in defining the variability of studied characteristics, i.e. survival of lambs up to the 100th day of age. For the best available classification model, the ranking of variable importance, developed based on the “Importance” measure, allowed to conclude that the type of lamb’s birth, season, following by the year of birth, subsequent lambing, lamb’s sex and its breed were the most significantly differentiating factors. It was noted that the tree built with the use of CART algorithm was composed of 30 leaves. It was also shown that the highest frequency of lamb’s death during the rearing period was to be expected among triplets born in winter or summer (37.14% of all deaths), while the highest chance (98.42%) of surviving till the 100th day of life showed singletons, born from their mother’s 3rd to 6th litter, in the spring-winter season in the last year of the present research.