Principal component analysis of body weight prediction from morphometric variables in Ethiopian indigenous sheep

Accurate estimation of body weight (BW) is essential for effective sheep management, though direct weighing is often impractical in smallholder systems where scales are unavailable. Predictive models using linear body measurements (LBMs) are common but often compromised by multicollinearity among traits. This study employed principal component analysis (PCA) to identify the underlying structure of morphometric traits and develop prediction model for BW in Ethiopian indigenous sheep, comparing the predictive performance of PCA-based generalized regression (GR) against GR using original LBMs. Data on BW and 16 LBMs were collected from 249 ewes and 57 rams aged 20–60 months. PCA with Varimax rotation was performed separately for each sex, with the number of principal components (PCs) determined by parallel analysis. PCA revealed a unidimensional structure in rams (one PC, explaining 31.51% of variance) and a multidimensional structure in ewes (three PCs, explaining 40.54% of variance). GR models using original LBMs outperformed PCA-based models within the analyzed dataset, with higher generalized R² values (rams: 0.765 versus 0.680; ewes: 0.442 versus 0.313) and lower AICc values. Key predictors were body length and cannon bone length for rams, and heart girth, hair length, and tail width for ewes. These findings suggest that while PCA effectively reduces dimensionality, it may sacrifice trait-specific predictive information critical for accurate weight estimation. We recommend LBM-based GR models using sex-specific key predictors as practical tools for BW estimation in smallholder sheep production systems.