Fig 1.
Accuracies for latent discriminant analysis without cross-validation on simulated data for 8, 12 and 19 predictors with increasing n.
The averaged accuracy is displayed on the y-axis and the increased sample size for the simulation on the x-axis.
Fig 2.
Accuracy differences (y-axis) between traditional training set optimisation and leave-one-out cross-validation compared to independent test set validation for increasing sample sizes (x-axis).
The dashed horizontal grey lines indicate the upper and lower boundary of the [-0.05; +0.05] stability corridor. The vertical coloured line indicates the sample size points-of-stability for the training set optimisation technique. Inset plot: accuracy difference scores zoomed in for sample size between 40 and 240.
Table 1.
Illustration how the dominant practice (linear discriminant analysis with training set optimisation) can lead to an erroneous conclusion.
Table 2.
Suggestions for the improvement of the accuracy estimation in the predictive analysis in verbal credibility assessment research.