The
Model Results
tab contains the following elements:
This top section of the tab echoes the key parameter settings for the
model
fit. It also lists the final selected
variables
and standard model fitting criteria for the training and test sets such as
Root Mean Square Error
, Mean Absolute Error, Area Under the ROC Curve, Accuracy,
specificity
(for binary
dependent variables
),
sensitivity
(for binary dependent variables), category-specific Accuracy (for non-binary categorical dependent variables), and Harrell's C-Statistic (for continuous dependent variables).
This section first plots the
distribution
of the dependent variable. If it is categorical, the distribution of the
posterior probability
of having the true value (
Prob_Actual
) is plotted. Values below
0.5
for this variable indicate misclassifications. For categorical dependent variables, the distribution of the correctness of the prediction (
Correct_Pred
) is present. Note the fraction of "
Yes
" values for this variable is the same as the Accuracy statistic.
For continuous dependent variables, this section contains the distribution of the predicted values and
residuals
.
You should study these distributions carefully to make sure results make sense based on your knowledge of the data and model. Click on histogram bars to highlight
observations
of interest. To view the underlying data table, go to the
Tabs
section in the upper left corner and click
Model Results > View Data
. Press
to find the first selected row.
See
Distribution
for more information.
|
•
|
Contingency Analysis
(for categorical dependent variables)
|
This section includes a
mosaic plot
and contingency table of the
prediction confusion matrix
. It breaks down all correct and incorrect predictions into cells.
You can see which categories are predicted poorly and view patterns of misclassification. As with histogram bars, click on mosaic cells to highlight observations of interest. To view the details in the underlying data table, go to the
Tabs
section in the upper left corner and click
Model Results > View Data
.
|
•
|
Bivariate Fits
(for continuous dependent variables)
|
This section includes
bivariate
plots of both actual value and residuals versus predicted values. The latter is actually a 45-degree clockwise rotation of the former. These plots can help identify the quality of model fit and outliers in the data. The ideal residual plot is random noise around a
zero
horizontal line; the estimated smoothing spline should be close to horizontal in this case. Deviations from this can be a sign of a poor model, missing
predictors
, or a dependent variable that is very difficult to predict.