Post

The idea is simple. "Is there a difference?" is almost never the question we actually care about. How big is it? How sure are we? Does it matter? DABEST answers those directly, guiding the eyes towards the quantified effect sizes and confidence intervals instead of an asterisk(or 2 or 3 asterisks).

The bootstrap-enabled uncertainty quantification is especially transparent. Unlike error bars derived from theoretical distributions, you just resample your data to form the interval empirically.

What delights me most is how far the idea travels. I keep meeting the same problem far outside neuroscience and biology: how much did a model improve, how large was a drug's effect, what should we make of a shift in the numbers recorded from our own bodies?

The modern world is saturated with automatically collected metrics. Across science and beyond, many of us are trying to understand not merely whether a number moved, but by how much, with what uncertainty, and whether the change is large enough to matter.