Fiction, future, and prediction

I post about statistics and data in medical research. Every post ends with a section “Just one more thing” to provide a concrete data example with code in the R programming language that the reader can try out and examine.

Immortality

Immortal time bias
Survival analysis

Immortal time bias occurs when patients must survive a certain period to receive treatment, making them “immortal” during that time. This can falsely suggest a survival benefit. From Oscar winners to cancer patients, this subtle bias can distort results, reminding analysts that measuring survival accurately requires careful attention to time.

Jan 14, 2026
Simon Schwab

Try, try, try

Randomized controlled trial
Replication

A randomized controlled trial is testing an intervention or trying it out. Unfortunately, it is not uncommon for subsequent studies to shrink initial effects towards zero, and there are reasons for that. The situation may simply highlight the natural evolution of scientific evidence in medicine, where initial excitement often gives way to more realistic assessments over time.

May 19, 2025
Simon Schwab

Land of confusion

Confounding
Statistical modeling
Multivariable models

Confounding can mislead medical research, obscuring true relationships between variables. This post explores how confounding biases arise, illustrating with real-world examples. Model-based approaches, like multivariable regression models, adjust for multiple confounders and offer a more accurate framework for interpreting complex data and making informed clinical decisions.

Mar 30, 2025
Simon Schwab

True colors

p-values
Research credibility
Research waste

Statistical significance is often misunderstood. Even when a low p-value is obtained, it does not guarantee the result is true; in fact, it may still be wrong. This claim can be effectively illustrated using colors, revealing how many studies were really true and how many were false. Without skepticism and blindly focusing on the p-value while ignoring other factors, I might just as well believe in Santa Claus.

Dec 15, 2024
Simon Schwab

1901

Statistical computing
Linux

While cleaning out my basement, I found a 20-year-old IBM computer. First, I wanted to get rid of it because the computer has become too old and useless. Instead, I installed a lightweight Linux operating system, R, Quarto, and RStudio. I then successfully analyzed the Titanic dataset as a showcase. It worked extremely well. Linux can bring new life to old computer hardware.

Oct 20, 2024
Simon Schwab

I don’t sleep, I dream

Animal studies
Repeated inspection of data
p-values

A 2018 sleep study increased its sample size until achieving statistical significance—a flawed practice known as p-hacking. This post explains why such practices are problematic, how they inflate false positives, and what that means for research credibility. Using data simulations, I highlight the pitfalls of data peeking and make the case for pre-specified statistical analysis plans in preclinical animal research.

Aug 18, 2024
Simon Schwab

Fake plastic trees

AI
Big data
Biomarkers
p-values
Univariable screening

In today’s data-driven world, more data doesn’t always mean better decisions. We often overvalue data and undervalue statistics. Excessive, poorly guided analyses, especially with high-dimensional data, risk overfitting and false discoveries. The post warns that data is meaningless without solid statistical methods and expert oversight, and cautions against trusting AI-driven analyses.

Jul 9, 2024
Simon Schwab
No matching items