Model Fit & Concepts of Interaction

Review of Logistic and Proportional Hazards Regression Model Selection:

You can look at changes in the deviance (-2 log likelihood change)

• Deviance - Residual sum of squares with normal data
  • Problem: Deviances alone do not penalize "model complexity" (hence need LRT - but onlu applies to nested models)
• AIC, BIC - More commonly used
  • Larger BIC and AIC -> worse model
  • BIC is more conservative
  • Both based on likelihood
  • An advantage is that you do not need hierarchical models to compare the AIC or BIC between models
  • A disadvantage is there is no test nor p-value that goes with comparison of models
  • A model with smaller values of AIC or BIC provides a better fit
  • Used to compare non-nested models
    
    • A non-nested model refers to one that is not nested in another; the set of independent variables in one model is not a subset of the independent variables in the other models
    • The data must be the same