Latent variable models on performance tests in guide dogs : II SEM (structural equation modeling) and path diagrams.

Aim: Compare two reduced models to a saturated one by Structural Equation Modeling. Data: Data on a previous study conducted by factor analysis on 11 behavioral test on 143 guide dogs reared in the National Guide Dog School of Scandicci (Firenze, Italy), were analyzed using Structural Equation Modeling on three separated models. Design: In the first model (Fig. 1; Full-model), each latent variable is influencing all observed variables. In the second model (Fig. 2; Model-2), individual latent variables are influencing only observed variable with high loadings, with exception of a variable with a minimum loadings, which is affected by both latent variables. Third model (Fig. 3; Model-3) excludes minimum loading variable from Model-2. Findings: Produced results are in linea with a previous study where: - first latent variable, L1, associated to characters of reaction to external stimuli, was identified as fearfulness/curiosity; and - second latent variable, L2, related to characters of sociability and to relationship with the handler, was identified as dominance/submissiveness. Analysis of differences shown in chi- square-fitting values of models and in others fitting indexes such as Akaike Information Criterion and Root Mean Square Error of Approximation, allows to accept hyphotesis of equivalence of models. Conclusions: Use of SEM and path diagrams results intuitive and flexible allowing a simpler interpretation of relation between variables in respect to factor analysis.