Now let's go to "Multibase_Result" and check the analytical report.
First, let's see contribution table at the top of the sheet. PRESS (Predicated Residue of Estimated Sum of Squares) judges an appropriate number of components. Though the more the PLS extracts components, the stronger the relation between X and Y becomes, the regression model becomes unstable and low-predictablei. In order to avoid such over fitting, PRESS predicts an appropriate number of component by building partial models using limited samples and evaluating their reliability using the other (non-used) samples (evaluation sets). If the estimated Y variables of evaluation sets fit the observed Y variables well, the model is accepted and another component is added. Repeating these procedures, PRESS judges the reliable number of the components. In this case, PRESS in Multibase extracts one component and shows it as red cell color.
The estimated and observed values of dependent variables (Y), whiskey and cheese, are displayed as shown below.
Variable Importance (VIP) and single correlation factors are displayed as shown below. VIP means contribution of each independent variable for model, whereas single correlation factor shows correlation between each Y and X variable. Comparing VIP with single correlation factors, you can understand the difference between whole and partial correlations.
2-3-1 Marunouchi Chiyoda Tokyo Japan