Topic: Purpose-Driven Model Specification

Prespecify Candidate Terms Before Examining Associations

Match Candidate Complexity to Sample Size and Intended Use

Keep Explanatory and Predictive Specifications Distinct

Include Interactions Only for Stated Scientific Questions

Generate Candidate Predictors from Subject-Matter Knowledge

Hold the Outcome Definition Constant Across Candidate Models

Distinguish Constructs from Their Measured Proxies

Document the Rationale for Included and Excluded Terms

Define the Prediction-Time Information Set

Check What a Candidate Formula Implies