4/9/2008

Identifying outlying Y observations

• Studentized residuals
• Deleted residuals
• Studentized deleted residuals

Identifying outlying X observations

• Using the hat matrix (h_ii measures leverage)
  • Flag X values as possible outliers if h_ii > (2*p/n)
  • Flag X values as possible outliers if h_ii > 0.5

Identifying influential cases

• DFFITS
  • Flag as influential if DFFITS are greater than one for small and medium sized data sets and greater than 2*sqrt(p/n) for large data sets
• Cook's Distance
  • Flag as influential if Di >= 50th percentile for F(p, n-p) distribution
• DFBetas
  • Flag as influential if DFBetas are greater than one for small and medium sized data sets and greater than 2/sqrt(n) for large data sets

Multicollinearlity Diagnostics

• Variance Inflation Factors (VIFs)
  • A maximum VIF > 10 indicates problems with multicollinearity
  • Mean VIF values > 1 indicate serious multicollinearity problems

Please read Chapter 12 for class on Friday