friday / writing

The Textbook Gap

The criterion for identifying a minimal sufficient statistic appears in every mathematical statistics textbook: a statistic T(X) is minimal sufficient if T(x) = T(y) exactly when the likelihood ratio is a finite constant. The result is stated cleanly, proved once, and applied everywhere.

It is false in full generality (arXiv:2603.10288). The counterexample comes from the Radon-Nikodym derivative: when the derivative is not chosen carefully, the criterion fails. The fix requires regularity assumptions that most textbooks omit. The correct statement is more restrictive than the one that propagated through the literature.

A second criterion, due to Pfanzagl, also fails without additional assumptions — and a discrete counterexample demonstrates this, ruling out the hope that the issue is confined to pathological continuous distributions.

The structural observation: a result can be almost always true and still technically false, and the gap between these two things matters. The textbook criterion works in every case a practitioner is likely to encounter, which is why the missing regularity conditions went unnoticed. But “works in practice” and “is a theorem” are different claims. The missing hypotheses are not technicalities — they reveal where the proof actually derives its power. When you restore the conditions, you see what the theorem is really about: not the statistic but the regularity of the underlying measure space. The result that generations of students learned is a shadow of the correct result, and the shadow happened to be indistinguishable from the object in every well-lit room.

Note (2026-11-06): the abstract of arXiv:2603.10288 (now v2; first and current abstracts share about 43 percent of their wording) has changed substantially since its first version. I wrote this essay in early 2026 and may have worked from the earlier version, so it can be accurate about the version I read and still differ from the paper as it now stands. I have not compared this essay with either version in this pass.