Methods
Stratified versus monetary-unit sampling in messy ledgers
Textbooks present a tidy fork. Live UK ledgers offer missing cumulative fields, negative lines, and “other” buckets that swallow risk. Here is how Schema Analytics tutors choose — and when we refuse both defaults.
Start with the question, not the button
If your objective is to give larger items a higher chance of selection while still covering the volume of smaller postings, monetary-unit sampling (MUS) is often honest. If your objective is to inspect qualitatively different groups — retail vs wholesale, automated vs manual journals — stratification usually earns its paperwork.
The failure mode we see most: teams pick MUS because the transaction sampling audit app defaults to it, then invent a purpose paragraph afterwards.
When stratification earns the hours
Stratify when risk stories differ by band or segment and you can define those bands without circularity. Document residual strata explicitly. A leftover “everything else” pile is fine if you say what risk it still carries and how you sized it.
When MUS is the cleaner story
MUS shines when value correlates with exposure and you can obtain a reliable book amount field. If the export cannot produce a stable cumulative total, do not paper over it with manual sorts you cannot reproduce. Fix the frame or choose another approach.
A short decision rule
Write one sentence: “We are sampling to learn ___ about ___.” If the blank needs segments, lean stratified. If it needs proportional value coverage, lean MUS. If the sentence is vague, pause — size calculators will not repair an unclear objective.