Stratification based on key variables is a common practice in clinical trials, as achieving a balanced distribution across known baseline factors is often considered essential for the credibility of study outcomes. However, the actual advantages of this approach are subject to debate in the literature. Some researchers have demonstrated that it fails to enhance efficiency in large samples and only marginally improves it in smaller ones. This study explores various subgroup analysis methods, paying special attention to the potential benefits in terms of inferential accuracy between pre-stratification and post-stratification. The merits and drawbacks of population-based versus randomization-based inference are evaluated, considering the influence of treatment-by-covariate interactions and patient response variability. Our findings indicate that pre-stratification yields minimal benefits and may even be detrimental, particularly in randomization-based procedures when there’s a chronological bias. Moreover, even in the presence of treatment-by-covariate interaction, pre-stratification could significantly diminish inferential accuracy.