Figure 4. Comparison of the discriminating power of two models deprived (model 1) or not (model 2) of Starmaze tests equated for the number of variables. In both models, the 6 most discriminating variables from the preliminary discriminant analysis conducted on all variables were entered in the definitive discriminant analysis [Model 1: (1) FAB, (2) MMSE-R, (3) FCSRT-CS, (4) CBT-B, (5) RCFT copy, (6) FCSRT-TR; Model 2: (1) temporal memory score, (2) FAB, (3) percentage of direct paths on learning trial 11, (4) MMSE-R, (5) FCSRT-CS, (6) FCSRT-TR). Top, Classifier performances for both models performed on the 60–80, AD and FTLD groups. Graphs represent individual classifications along both factors extracted from the discriminant analysis. The more the groups are separated, the more the model discriminates. Bottom, Discriminant analysis on pairs of groups. In each pairwise model of models 1 and 2, significant variables are extracted and, below, classifier performances are shown. FR, Free recall; TR, total recall; CS, sensitivity to cueing (%); W. L., partial Wilks lambda.