TC-03 statcheck (recomputation of reported p-values)
Recomputes p-values from reported test statistics and degrees of freedom and produces a list of inconsistencies and gross inconsistencies.
What it does
Recomputes p-values from reported test statistics and degrees of freedom and flags mismatches. Artefact: a list of inconsistencies and gross inconsistencies, where a gross inconsistency changes the significance decision at the 0.05 level.
When to use it
Use when you want to check that the p-values in a set of articles match their reported test statistics.
When it does not fit
Fits text with results in American Psychological Association style. Misfit: tables, other formats, or results with an effect size between statistic and p-value. It cannot tell which element is wrong.
Result format
The result is a list of reported tests, each with the test statistic, degrees of freedom, the reported p-value and the recomputed p-value. It marks each inconsistency, and marks as gross the ones where the significance decision changes at the 0.05 level.
Requirements on the material
| Property | Requirement type | Requirement |
|---|---|---|
| Content type | Always applies | requires=text |
| Unit of analysis | Always applies | requires=claim |
| Amount of material | Applies in some cases | prefers=needs automation |
| Appraises source credibility | Applies in some cases | no |
Not restricted on: Structure, Variable types, Time structure, How the sample was chosen, Level of detail available, Number of sources, Whether sources can be joined, Known data problems, Producer interest, Source dependence, Origin, Document genre, Language, Personal information, Security classification, Licence.
| Minimum material | Test statistics, degrees of freedom and p-values reported in American Psychological Association style text. |
|---|---|
| Additional sources needed | none |
| Not for (rules) | none |
| Not for (text) | Wrong choice for results in tables or other formats, or where an effect size sits between the statistic and the p-value. It cannot tell which element is wrong, and it may overstate inconsistencies (9.0 against 4.3 per cent by hand). |
Choosing and ordering
| Kind of technique | Verification technique, runs in step 6 |
|---|---|
| Phase tag | validate |
| Other phase tags | verify and prepare |
| Kind of finding | contradiction |
| Question answered | what |
| Strongest statement it supports | describe |
| Needs as input |
|
| Makes as output |
|
| Must exist first | None |
| Loop target | proceed 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: deterministic.
| Stakes | Check | Person involved | Counts as a different method |
|---|---|---|---|
| exploratory | input validation | none | no |
| decision-informing | input validation | none | no |
| decision-informing | code execution for arithmetic | none | no |
| decision-informing | spot check | optional | no |
| high-stakes | input validation | none | no |
| high-stakes | code execution for arithmetic | none | no |
| high-stakes | spot check | optional | no |
| high-stakes | human review sample | needed | no |
| high-stakes | independent review | needed | no |
Failure modes and limits
| Model suitability | model plus code |
|---|---|
| Typical model failure | Misses results that are not in the expected format and reports a mismatch caused by rounding or a one-sided test as an error. |
| Error sources | input data; data transformation |
| Certainty limits | A mismatch can come from rounding, one-sided tests or typing errors. The software extracted 67.5 per cent of results found by hand, so a clean result covers only the extracted part. |
Related techniques
- pairs with TC-01 GRIM test (granularity-related inconsistency of means): GRIM checks reported means while statcheck checks reported test statistics
- pairs with CK-18 Numeric and attribution recomputation by code (numeric fidelity check; deterministic recomputation): recomputation of figures covers numbers outside the test statistic format
Evidence and tag basis
Tested. It extracted 67.5 per cent of results found by hand in a 49-article validation and may overstate inconsistencies (9.0 against 4.3 per cent by hand). Across 30,717 articles, 49.6 per cent of those with such results had at least one inconsistency and 12.9 per cent at least one gross inconsistency. (Nuijten et al., Behavior Research Methods 2016)
Tag basis: input profile inference; selection inference; sequencing inference; robustness inference. All tag values are inference unless a source is named.