Prism offers four ways to graph results of logistic regression.
Name |
X axis |
Y axis |
Actual vs Predicted |
Actual (Entered) Y |
Predicted probability |
ROC Curve |
1 - Specificity |
Sensitivity |
Logistic Plot |
Actual (Entered) X |
Actual (Entered) Y |
Proportion Correct vs Cutoff |
Cutoff value |
Proportion classified correctly |
Actual vs. Predicted
The Actual vs Predicted plot generates a violin plot with two groups: one group contains the observations entered in the data table with a dependent (Y) value of 0, and the other with a Y value of 1. For each of these groups, Prism plots the corresponding predicted values (from the model) as a violin plot.
ROC Curve
The ROC curve evaluates classification performance. As is typical with ROC curves, the values are plotted on axes of Sensitivity and 1-Specificity for every possible cutoff value, where (for any given cutoff value):
•Sensitivity is the number of values entered as 1 and correctly classified as 1, divided by the total number of values entered as 1
•Specificity is the number of values entered as 0 and correctly classified as 0, divided by the total number of values entered as 0
Logistic Plot
The Logistic plot option is only available when a single predictor (X) variable is included in the model. This generates the typical S-Curve (or part of an S-Curve) that represents the predicted probability of Y=1 at the given X value. If two models are being compared (via the Compare tab), this option will be available only if both models contain only one predictor (main effect) in addition to an intercept term.
Proportion Correct vs. Cutoff
The Proportion Correct vs Cutoff graph is an alternative to looking at the ROC curve. Similar to the ROC curve, it investigates every possible cutoff value (the X axis for this graph), and plots the corresponding proportion of observations that were correctly classified for that cutoff.
