Packet 6 Concept Map

Packet 6 Concept Map

1. Multicollinearity

   • EFFECTS
     • Artificially increases standard error
       • Decreases test statistic & increases p-values
         • Results of test not valid
   • Some predictors are correlated
     • Information overlaps/is redundant - we don't want double counting
   • WAYS TO DEAL WITH MC
     • Variable Selection (StepAIC(), R^2 Adjusted)
     • Principal Component Analysis (PCA)
       • KNOW THE FOUR STEPS TO PCA
     • LASSO/Ridge Regression
   • Detecting MC
     • VIF = 1/(1-R^2)
       • If >10 there is a MC problem
       • Fit a model that predicts one predictor variable from other predictor variables

2. One Qualitative Independent Variable

   • Categorical Predictors=Factors
     • Different categories =levels
   • General Interpretation of Beta with binary variable
     • When x=1, we expect y to be Beta higher on average than when x=0

3. Variable Screening Method

   • Want simple powerful models
   • Criterion-based variable selection searches through all positive 2^q models
     • picks the model that provides the best value of the particular criterion
     • Adjusted R^2 looks for maximization
     • AIC-smaller=better
       • Removes variable with the lowest AIC until it is minimized

4. Possible Regression Selection Procedure

5. Danny Lipa, Mark Moran, Grant Worthington, Kieran Wurl