Society of Actuaries (SOA) PA Practice Exam 2025 – 400 Free Practice Questions to Pass the Exam

The method used for backward selection in the context of Bayesian Information Criterion (BIC) on training data is indeed stepAIC. Backward selection is a technique where you start with all candidate predictors and progressively remove the least significant variables based on a specific criterion, such as BIC, which helps in model selection by penalizing the complexity of the model.

The stepAIC function specifically implements both backward and forward selection and optimizes for criteria like AIC (Akaike Information Criterion) and BIC. When applying backward selection through stepAIC with BIC, the method examines the effect of removing each variable and selects the model that provides the best balance between model fit and complexity.

Other options on the list serve different purposes: glmnet is used for fitting generalized linear models via penalized likelihood with elastic net regularization but does not focus explicitly on backward elimination. cv.glmnet is used for cross-validation with glmnet to find the optimal regularization parameter, thus not focusing on backward selection. Lastly, dummyVars is used for creating dummy variables from categorical data and is not related to model selection methods like stepwise regression. Therefore, stepAIC is clearly positioned as the appropriate choice for executing backward selection while considering BIC.