Significance testing - are we ready yet to abandon its use?

Area of Science:

• Statistical methods in clinical research
• Biostatistics
• Evidence-based medicine

Background:

• Significance testing, commonly using p-values, is increasingly recognized for its detrimental effects in scientific research.
• Traditional reporting of p-values without dichotomization or reliance on confidence intervals (often misinterpreted) presents significant limitations.
• Existing statistical practices hinder accurate interpretation and decision-making in clinical research.

Discussion:

• Bayesian statistics offer a more intuitive interpretation of results compared to p-values, despite concerns about subjective models.
• Confidence interval functions provide a continuum of estimates, offering a more nuanced view than traditional confidence intervals.
• Alternative statistical approaches address the interpretational challenges and misuse associated with significance testing.

Key Insights:

• Bayesian posterior probabilities are more intuitive than p-values, aligning better with human decision-making patterns.
• Confidence interval functions and Bayesian methods present superior alternatives to traditional significance testing.
• The persistence of significance testing stems from habit and unfamiliarity with advanced statistical techniques.

Outlook:

• Widespread adoption of superior statistical methods requires overcoming resistance to change and promoting education.
• A collaborative effort among researchers, editors, and reviewers is crucial for transitioning away from significance testing in clinical research.
• Future clinical research should embrace more interpretable and robust statistical frameworks for improved scientific rigor.