4/9/2009

Chapter 9: Overview of Model Selection

• Strategy for Building a Regression Model: See Figure 9.1
1. Data collection and preparation
• What kind of data is this?
• Which variables are essential? Which ones can be screened out?
• Data errors
• Extreme outliers

What are the important interactions? Higher-power terms?
What are the relationships between variables?

2. Reduction of explanatory variables
   How do we decide what subset of variables is “best?”
• All possible subsets
• Automatic search procedures
  
  
3. Model refinement and selection
• Compare models: interaction effects, fit
• Further refinements
  
  
4. Model validation
• Comparison to new data
• Comparison to theory or simulation
• Cross-validation
  
  
• Example: Surgical unit data (Surgery-Ch9.sas)
• Criteria for Model Selection
1. R-sq
2. adjusted R-sq
3. Mallows’ Cp
4. AIC and SBC
5. PRESS
• Search procedures
1. All possible subsets
2. “Best” subsets
3. Stepwise Regression

For Tuesday: Read the rest of Chapter 9, and prepare the following problems for a Problem Session: #15, 16, 19, 26, 28a.