Latent variable models on performance tests in guide dogs. 1. Factor analysis

Abstract

The research has been conducted on behavioural test results obtained from 143 dogs of pedigreed stock reared in the National Guide Dog School (SNCG) of Scandicci (Firenze, Italy), consisting mostly of Labradors and Golden Retrievers, but also including German Shepherds. All dogs have been reared under quite uniform conditions and tested individually under similar conditions. The results following the 11 administered subtests, [that constitute variables in our analysis], were expressed in scores ranging from 1 to 5 and used after calculation of the rank averaged scores. The analysis of the Pearson and partial correlations between the variables points out a clean distinction in two groups. The first consists of variables related to characters of sociability and to relationship with the han- dler, with expression of dominance/submissiveness and the second to characters of reaction to external stim- uli with expressions of fearfulness/curiosity. Results of factor analysis led us to reject the one factor model and accept a model with two factors, in which: 1) Factor I identifies variables of the group tied to the fearfulness/curiosity; 2) Factor II identifies the vari- ables of the group tied to sociability and to relationship with the handler. The two factors are correlated, indicating the presence of some non negligible, indirect effects. One out of the eleven variables has not shown important evidence of contribution to any of the factors.

Keywords

• Dog, Performance tests, Factor analysis, Latent variable models