The stupidity, the criminal vandalism, the wanton destruction of information involved in dichotomisation

2026-02-14 · Source: Statistical Modeling, Causal Inference, and Social Science · Field: Science & Research — Mathematics & Computational Sciences, Health & Medical Research, Research Methodology & Innovation · Depth: Advanced, quick

Summary

A study analyzed 21,435 unique randomized controlled trials (RCTs) from the Cochrane Database of Systematic Reviews (CDSR), finding that 7,224 (34%) had continuous outcomes and 14,211 (66%) had binary outcomes. Trials with binary outcomes exhibited larger average sample sizes, yet also showed larger standard errors and fewer statistically significant results. The researchers concluded that while sample sizes are increased to offset the lower information content of binary outcomes, this compensation is often insufficient. Many binary outcomes result from the avoidable dichotomization of continuous data, a practice termed "responder analysis," which leads to a significant and unnecessary loss of information. This method is deemed wasteful, costly, and unethical, as it burdens more participants than required.

Key takeaway

For AI Scientists designing or evaluating clinical trials, you should critically assess any proposed dichotomization of continuous outcomes. This practice significantly reduces statistical power and inflates required sample sizes, leading to less efficient and potentially unethical research. Prioritize direct analysis of continuous data to preserve information and ensure robust conclusions, potentially using tools like the described Shiny app to demonstrate the impact of dichotomization.

Key insights

Dichotomizing continuous outcomes in clinical trials leads to significant information loss and inflated sample sizes.

Principles

• Binary outcomes reduce information content.
• Dichotomization inflates standard errors.
• Increased sample size does not fully compensate.

Method

The study used a simple method to approximate information loss across clinical trials in the CDSR. A Shiny app calculates information loss from "responder analysis" and determines required sample sizes for continuous vs. dichotomized outcomes.

In practice

• Avoid dichotomizing continuous outcomes.
• Use the Shiny app to quantify information loss.
• Calculate sample sizes for continuous data.

Topics

• Dichotomization
• Clinical Trial Design
• Statistical Information Loss
• Sample Size Determination
• Biostatistics

Best for: AI Scientist, Data Scientist, Research Scientist, Domain Expert

Related on AIssential

• Survey Statistics: Individualism and the CV Noise Problem — Individual-level log loss often fails to differentiate substantively important model improvements in binary data.
• 2-Step Agent: A Framework for the Interaction of a Decision Maker with AI Decision Support — Misaligned prior beliefs in AI decision support can degrade outcomes, even with rational agents and perfect models.
• Causal-Privacy Audit Workflow for Synthetic and Distilled Data in Dropout Support — Generated student data requires joint causal-privacy audits to ensure decision-making fidelity, treatment-effect preservation, and privacy.
• Causal Decomposition Analysis with Synergistic Interventions: A Triply-Robust Machine Learning Approach to Addressing Multiple Dimensions of Social Disparities — An extended causal decomposition analysis evaluates synergistic effects of multi-domain interventions using a triply robust machine learning estimator.
• Recent discoveries on the acquisition of the highest levels of statistical fallacies — Selection bias can create spurious negative correlations among highly selected groups.
• Influential study touting ChatGPT in education retracted over red flags — A highly influential study published in Springer Nature's *Humanities & Social Sciences Communications*, which claimed OpenAI's ChatGPT positively impacts student learning, has been retracted nearly a…

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Statistical Modeling, Causal Inference, and Social Science.