startthat.work home

CK-15 Sample representativeness audit (coverage check; selection bias check)

Compare the analysed sample with the population it claims to describe and with the steps that selected it, and produce a comparison table on known variables and a list of exclusion steps.

Check extracted resultsEvidence: DocumentedVerification technique, runs in step 6Strongest statement: describe

What it does

Compares the analysed sample with the population it claims to describe and with the steps that selected it. Artefact: a comparison table on known variables and a list of exclusion steps. Read the gaps between sample and population.

When to use it

Use when results from a sample will be read as describing a wider population or a full set of documents.

When it does not fit

Misfit: no known population benchmark. In LLM pipelines, retrieval or truncation steps can create selection silently (inference).

Result format

The result has a comparison table of the analysed sample against the population on known variables, with the gap in each row, and a list of the exclusion steps that selected the sample. It states where the gaps between sample and population are largest.

Requirements on the material

PropertyRequirement typeRequirement
Content typeAlways appliesrequires=text; requires=tabular
Unit of analysisAlways appliesrequires=dataset
Level of detail availableAlways appliesrequires=record-level
Appraises source credibilityApplies in some casesno

Not restricted on: Structure, Variable types, Time structure, How the sample was chosen, Number of sources, Whether sources can be joined, Amount of material, Known data problems, Producer interest, Source dependence, Origin, Document genre, Language, Personal information, Security classification, Licence.

Minimum materialat least one population benchmark on a known variable and a record of how items were selected
Additional sources neededcomparator
Not for (rules)none
Not for (text)Wrong choice when no population benchmark exists. In language model pipelines, retrieval or truncation steps can create selection without being noticed.

Choosing and ordering

Kind of techniqueVerification technique, runs in step 6
Phase tagvalidate
Other phase tagsverify and prepare
Kind of findinggap; risk
Question answeredwhat; who
Strongest statement it supportsdescribe
Needs as input
Makes as output
Must exist firstNone
Loop targetproceed with limit stated

Library checks by stakes

Each stakes level includes the checks of the level below it. Checks that need a person apply at high stakes only, so the process leaves them out at lower stakes (decision 38). Repeatability: judgement dependent.

StakesCheckPerson involvedCounts as a different method
exploratoryspot checkoptionalno
decision-informingspot checkoptionalno
decision-informingcode execution for arithmeticnoneno
decision-informingassumption and limitation statementnoneno
high-stakesspot checkoptionalno
high-stakescode execution for arithmeticnoneno
high-stakesassumption and limitation statementnoneno
high-stakessensitivity analysisnoneno
high-stakesindependent reviewneededno

Failure modes and limits

Model suitabilityhuman judgement required
Typical model failureSummarises the documents it was given as if they were all the relevant documents.
Error sourcesinput data; source selection
Certainty limitsCertainty drops when the benchmark is out of date or on different definitions (indirectness) and when selection steps are undocumented (risk of bias).

Evidence and tag basis

Documented. Only the point that validation sets should be random or stratified so that they represent the corpus was verified. (Grimmer and Stewart 2013, Political Analysis)

Tag basis: input profile inference; selection inference; sequencing inference; robustness inference. All tag values are inference unless a source is named.