This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: Recitation Supplement for September 29, 2003 Fitting and Comparing Candidate Models We will use the same data set, myraw.xls , as was used in the first section. (See recitation supplement for description). We want to build a model that predicts how a donor responded to the campaign based on the previous information in the database about the donor. Go through the steps 1-4 and 6 from the handout for the first recitation. After that: 1. Drag a Data Partition node onto the diagram. Double click to open the Data Partition node. The Partition tab on the pop-up menu should be active. Reset the Percentages for the data partitioning so that the Train and Validation sets get 50% each, and the Test set gets 0%. (We won’t need a final test for the analysis we do here.) 2. Drag a Regression node onto the diagram, and connect it after the Replacement node. Double click to open the Regression node. Click the Model Options tab ( Regression subtab) and note that the type of regression is Linear . When the Variables tab is active, you can add interaction terms (products of different variables) and higher order polynomial terms (squares and cubes of variables) using the Interaction Builder ( Tools . Interaction Builder. . . ). You can try this later. Click the Selection Method tab. This tab enables you to perform different types of variable selection using various criteria. No selection is done by default. You can choose from the following variable selection techniques: (a) Backward - begins, by default, with all candidate effects in the model and then systematically removes effects that are not significantly associated with the target until no other effect in the model meets the Stay Significance Level or until the Stop criterion is met....
View Full Document
- Spring '07
- Statistics, Regression node