ABSTRACT:
This study aimed to use multivariate techniques of principal component analysis and canonical discriminant analysis in a data set from Morada Nova sheep carcass to reduce the dimensions of the original data set, identify variables with the best discriminatory power among the treatments, and quantify the association between biometric and performance traits. The principal components obtained were efficient in reducing the total variation accumulated in 19 original variables correlated to five linear combinations, which explained 80% of the total variation present in the original variables. The first two principal components together accounted for 56.12% of the total variation of the evaluated variables. Eight variables were selected using the stepwise method. The first three canonical variables were significant, explaining 92.25% of the total variation. The first canonical variable showed a canonical correlation coefficient of 0.94, indicating a strong association between biometric traits and animal performance. Slaughter weight and hind width were selected because these variables presented the highest discriminatory power among all treatments, based on standard canonical coefficients.
Key words:
canonical discriminant analysis; principal components; sheep production