Confirmatory factor analysis and structural equation models to dissect the relationship between gait and morphology in Campolina horses

• Published in: Livestock science 2022-01, Vol.255, p.104779, Article 104779
• Main Authors: de Oliveira Bussiman, Fernando, Silva, Fabyano Fonseca e, Carvalho, Rachel Santos Bueno, Ventura, Ricardo Vieira, Mattos, Elisângela Chicaroni, Ferraz, José Bento Sterman, Eler, Joanir Pereira, Balieiro, Júlio Cesar de Carvalho
• Format: Article
• Language: English
• Subjects: Causal inference; Confirmatory factor analysis; Horse genetics; Latent traits; Structural equation models
• ISSN: 1871-1413; 1878-0490
• DOI: 10.1016/j.livsci.2021.104779

•Morphology and Locomotion are not directly measurable, they are latent traits.•Morphology has a positive causal effect on Locomotion.•Selection for Morphology and Locomotion will led to horses with good quality.•Confirmatory factor analysis and structural equations allowed causal inference. It is well known that locomotion and morphology are related. Several studies have been performed over the years showing the genetic correlation between them in horses. However, none could reveal the possible causative link between those traits in this species. The aim of this study was to apply statistical causal inference to dissect the possible causative links between morphology and locomotion traits in Campolina horses. It was used 49,994 horses with records for 28 traits. All phenotypic records were adjusted for fixed (contemporary group, stud, and age) and random (technician) effects except the animal genetic effect. All missing observations were removed, and the final dataset was left with 640 adjusted records for all traits. A confirmatory factor analysis (CFA) was conducted to build a measurement model for three latent traits: Morphology (M), Morphological quality (Q) and Locomotion traits (L). Here it was assumed that morphology and locomotion cannot be directly measured, but only morphological and locomotion aspects. A multitrait animal model was performed to obtain breeding values (EBV) for the three latent traits, and factor scores from CFA were adjusted for the EBV. The adjusted factor scores were used as input in structural equation models to investigate all possible causative links among latent traits. The measurement model showed acceptable fit to the data, and the differences between sample and model-implied (co)variance matrices were low (0.01). L had low heritability estimate (0.03) whereas M and Q had moderate (0.21 and 0.36, respectively). All genetic correlations were negative, but those between Q and the others were favorable for selection. There is a causative direct effect of Q over M (-0.30) and an indirect over L (-0.10). It was found a causative effect of M on L (0.35). Knowing causal paths among traits will help breeders to take selection decisions and may have important implications for strategies in horse improvement.