PROC HPREG is run with SELECTION=STEPWISE and the default entry/removal significance levels. A variable 'credit_score' is added at step 2 but then removed at step 6 when 'payment_history' enters. At step 9, 'credit_score' re-enters after 'debt_ratio' removes 'payment_history'. What is the most appropriate modeling response to this stepwise cycling?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Here's the deal — b is correct because stepwise cycling is a diagnostic signal of multicollinearity — when predictors are correlated, their partial contributions change as correlated predictors enter/leave the model. The appropriate response is to investigate the correlation structure (VIF, correlation matrix), consider regularization alternatives (ridge, LASSO), or use subject matter expertise to select among correlated predictors.
Full explanation below image
Full Explanation
B is correct because stepwise cycling is a diagnostic signal of multicollinearity — when predictors are correlated, their partial contributions change as correlated predictors enter/leave the model. The appropriate response is to investigate the correlation structure (VIF, correlation matrix), consider regularization alternatives (ridge, LASSO), or use subject matter expertise to select among correlated predictors. A is wrong; cycling indicates instability, not definitive importance. C is wrong; sample size does not eliminate the fundamental collinearity issue. D is wrong; tighter thresholds may reduce cycling but can also exclude important predictors and don't address the underlying collinearity.