PLS Results - Distances Tab

Partial Least Squares (PLS)

Select the Distances tab of the PLS Results dialog box to access options to evaluate the distances (from the origin, i.e., where the values for all dimensions are equal to zero) for predictor variables (columns in the design matrix), dependent (response) variables, and for observations (based on residuals), given the currently specified Number of components (see PLS Results). These observation plots can be helpful in determining major contributors to the prediction of the conceptual variable(s) (plotting weights) as well as outliers that have a disproportionate influence (relative to the other observation) on the results (plotting residual values).

Note that there are two sets of buttons; click the left buttons to display results in a spreadsheet, click the right buttons to display results in a graph. For additional details about these options, see the Introductory Overview or the Partial Least Squares (PLS) Index.

Sample. Select one of the option buttons under Sample to specify which type of sample to base the predicted and residual values.

Analysis/Cross-validation/Both/Prediction. You can display and plot predicted and residual values for all observations that were used to compute the current results (select Analysis), all observations that were not used to compute the current results, but have valid data for all predictors and the dependent variables (select Cross-validation), both the Analysis and Cross-validation samples (select Both), or display and plot predicted values and related statistics for all cases that have valid data for the predictor variables, but missing data for at least one dependent variable (select Prediction). If these options are not available, no valid cases in the prediction sample were found, and/or no cross-validation sample was specified via the PLS Quick Specs - Options tab or via the SAMPLE keyword in the PLS Analysis Syntax Editor.

X weight dist. Click the X weight dist. button to display a spreadsheet (left button) or graph (right button) of X weight distances, which show the Euclidean distances of the predictor variables (effect columns) from the origin computed from the X weights over the current Number of components (see PLS Results). Specifically, each distance is the square root of the sum of squared X weights over the current Number of components.

Loadings dist. Click the Loadings dist. button to display a spreadsheet or graph of the loading distances, which show the Euclidean distances of the predictor variables (effect columns) from the origin computed from the X weights over the current Number of components (see PLS Results). Specifically, each distance is the square root of the sum of squared loadings over the current Number of components.

X resid. dist. Click the X resid dist. button to display a spreadsheet or graph of X residual distances, which show the Euclidean distances of the observations from the origin computed from the X residuals over the predictor columns (in the design matrix), for the current Number of components (see PLS Results).

Y resid. dist. Click the Y resid dist. button to display a spreadsheet or graph of Y residual distances, which show the Euclidean distances of the observations from the origin computed from the Y residuals over the current number of dependent (response) variables, for the current Number of components (see PLS Results).