Definition
Plain language
A way of adding up separate small effects, assuming each one contributes the same amount no matter what the others do.
As stated in the literature
A regression form in which the predicted output is a baseline plus independent per-factor contributions, with no interaction terms; here fit on log-odds to test whether prompt fragments contribute separably.
Also called: additive fit, additive structure, additively
Why it matters: It is the simplest way to check whether many small influences stack up predictably, and how much it fails to explain tells you where hidden combined effects are lurking.
For example, an additive model assumes that adding a polite greeting shifts the answer by the same amount whether the question is about chemistry or ethics.
Heard on the show
“And on that log-odds ruler, the additive fit holds up on prompts it never saw.”Episode 243 — How a Hundred Meaningless Word Choices Add Up to Flip a Model's Answer