Classification
The classification table shows the results of the Naive Bayes model. For each case, the predicted response is Yes if that cases's model-predicted probability of Yes is greater than the model-predicted probability of No.
- Cells on the diagonal are correct predictions.
- Cells off the diagonal are incorrect predictions.
153 of the 403 people who responded to the offer are classified correctly. 3851 of the 4597 nonresponders are classified correctly. Overall, 80.1% of the cases are classified correctly. The low percentage (38.0%) of correctly classified respondents is somewhat disappointing, but typical in situations where the outcome of interest has a low probability. One way to capture more respondents is to classify cases with a less than 50% probability of being a respondent as respondents.
To accomplish this, from the menus choose:
- Type predresponse as the target variable.
- Type PredictedProbability_2 > 0.30 as the numeric expression.
- Click OK.
The variable predresponse bases predicted response upon whether the model-predicted probability of response is greater than 0.30.
To see a classification table based upon this new definition of predicted response, from the menus choose:
Figure 3. Crosstabs dialog 
- Select Predicted Probability for response_01 = 1 as the row variable.
- Select predresponse as the column variable.
- Click OK.
With the new cutoff, 202 of the 403 (50.1%) people who responded to the offer are classified correctly. This comes at the cost of fewer non-respondents that are correctly classified; only 3483 of the 4597 are now correct. Overall, only (3483+202)/5000=73.7% of the cases are classified correctly, but if the goal of identifying respondents is more important than the goal of identifying non-respondents, you can accept a drop in the overall classification rate.