Society of Actuaries (SOA) PA Practice Exam

When would one prefer forward selection over backward selection?

1. When aiming for a more complex model

2. When needing a simpler and more easily explainable model

3. When data has numerous influential points

4. When all variables are known to be significant

The correct answer is: When needing a simpler and more easily explainable model

Forward selection is often preferred when the goal is to create a model that is simpler and more easily explainable. This approach starts with no variables in the model and adds variables one at a time based on their statistical significance and contribution to the model's explanatory power. By carefully considering which variables to include in each step, it allows the analyst to focus on the most impactful variables, leading to a more parsimonious model. In contrast, backward selection begins with a full model that includes all potential variables and iteratively removes the least significant variables. This method can result in a more complex model initially, making it harder to interpret, especially if many variables are extraneous. Thus, if simplification and clarity are priorities, forward selection is the preferable method. In the context of the other options, aiming for a more complex model or dealing with numerous influential points does not align with the intention behind forward selection. Similarly, knowing that all variables are significant does not necessitate the stepwise approach inherent in forward selection, as it could lead to redundancy.