Definition
Plain language
A way to show two things really are about the same, rather than just failing to spot a difference.
As stated in the literature
Statistical procedure (e.g. TOST) that pre-specifies a smallest effect size of interest and rejects the presence of effects larger than that bound in either direction, converting a non-significant result into positive evidence of equivalence.
Also called: equivalence testing, TOST
Why it matters: It stops teams from treating a small or underpowered study as proof that two systems perform the same when the data simply couldn't tell them apart.
For example, if two versions of an assistant score within a hair of each other, an equivalence test lets you say 'these are genuinely interchangeable' instead of merely 'we didn't find a difference.'
Heard on the show
“" Which is why the equivalence test is worth one sentence.”Episode 249 — The Chatbot Knows Your Facts And Still Won't Mention Them