Writing

When to use Bayesian vs. frequentist -- a practical guide for clinical researchers

The choice isn't philosophical. It depends on your prior information, regulatory context, and what you're trying to communicate.

Building reproducible pipelines: the stack that doesn't break in six months

Most analysis code is write-once. Here's how to build workflows your team can maintain, audit, and extend.

What to put in your statistical analysis plan (and what to leave out)

An SAP that's too vague gets you in trouble with reviewers. One that's too rigid ties your hands mid-study.

Power calculations are not magic -- a guide for investigators

Most power calculations are optimistic by design. Here's how to pressure-test yours before the IRB does.

Adaptive trial designs: when flexibility helps and when it hurts

Adaptive designs can dramatically reduce sample size -- or introduce bias that sinks your results.

renv + targets: the reproducible R workflow I actually use

Two tools, properly configured, eliminate almost every 'it works on my machine' problem in R-based research.

A distribution that models the Dunning-Kruger effect: the tetration distribution

Iterated exponentiation on the unit interval produces exactly the non-linearity the DK calibration curve requires. Here is the formal model.