Definition
Plain language
A way of expressing a difference in terms of how much the numbers normally bounce around.
As stated in the literature
Effect expressed in units of the within-subject SD (a Cohen's-d-style standardization), allowing comparison of effect magnitude across differently scaled outcome measures.
Also called: standard deviations, SD units
Why it matters: It lets you compare the size of effects across measures with completely different scales, so a raw jump of two points can be judged as large or trivial.
For example, saying a change improved scores by half a standard deviation tells you it moved results about half as much as people naturally differ from each other.
Heard on the show
“The authors run their own SFT and SFT-then-RL with three seeds and report standard deviations, which is responsible.”Episode 009 — How Two Silent Library Bugs Quietly Invalidated a Wave of Reasoning Papers