Replication Rate by Author Overlap
Do replications involving the original authors succeed more often than fully independent ones? This page plots the replication rate against the number of authors shared between the original study and its replication. Authors are matched by disambiguated SciSciNet/OpenAlex author identity, with ORCID checks and normalized-name matching as fallbacks (see methodology below).
Rate = success / (success + failure + reversal); inconclusive and unrecorded outcomes excluded, where success means the outcome recorded in the database. Unit: replication effect (n = 7,015 — originals matched in the author-overlap dataset). How this is defined.
Replication rate by number of shared authors
Method details
Author matching. Both papers of each (original, replication) pair are resolved by DOI to their author lists, primarily from a SciSciNet-v2 snapshot derived from OpenAlex (snapshot 2026-01-29), falling back to the live OpenAlex API for papers missing from the snapshot, and to the database’s own author-name fields as a last resort. Two authors count as the same person when they share a disambiguated OpenAlex author ID, or when their normalized names match (diacritics stripped, surname + first initial, “Last, First” ordering handled, single-character typos tolerated for full given names).
Why differing OpenAlex IDs don’t veto a name match. OpenAlex’s author disambiguation errs heavily toward splitting: the same person often carries several author IDs across their papers (initials-only bylines, typos, name variants, affiliation changes), while wrongly merging two people into one ID is rarer. Sharing an ID is therefore good evidence two byline entries are the same person, but carrying different IDs is weak evidence they are different people. Meanwhile, the same surname and initial appearing on a paper and on its direct replication in the same literature is strong contextual evidence of identity. So a name match is allowed to bridge differing IDs, with ORCID as the tie-breaker where available: different ORCIDs veto the match, identical ORCIDs confirm it, and missing ORCIDs let the name match stand. (A real example from this database: the first author of an Alzheimer’s cohort study appears under one OpenAlex ID on the original and under a second ID — via the typo “Leonared” — on its follow-up; only name matching recovers that the two papers share their lead author.)
Strict matching. The “strict matching” toggle counts only ID-verified shared authors and discards every name-based match. Because ID splits then misclassify some genuinely author-involved replications as independent, strict mode is a deliberate lower bound on overlap — biased toward the “independent” side. Its purpose is falsification: if the gap between independent and author-involved replication rates were an artifact of same-name false positives, it would shrink under strict matching. A gap that survives the toggle cannot be explained by name-matching errors.
Not causal. Original-author involvement is not randomly assigned. Original teams may choose to (help) replicate the effects they are most confident in, while independent teams may target suspect findings; original authors also bring materials, expertise and — possibly — allegiance effects. The gap between independent and author-involved replications reflects all of these at once.
Team-size confound. Large consortium replications (Many Labs and similar) mechanically dilute the fraction views: one original author among dozens of replicators is a small fraction but full “involvement”. Compare the count, fraction and lead-author views before drawing conclusions.
Matching limitations. OpenAlex author disambiguation is imperfect (it can split one person into several IDs or merge namesakes), ORCID coverage is sparse for older papers, and name matching can miss authors who changed names or publish under different transliterations. Overlap counts are therefore best treated as a lower bound with occasional false positives; the build script logs every name-only match for manual spot-checking.
Coverage. 7,909 of 8,445 replication rows (93.7%) have author data for both papers; rows missing a DOI on either side, or whose papers resolve nowhere, are excluded from the charts. 7,950 rows have both DOIs.
Units. The chart counts every determinate replication attempt once. Reversals count as determinate non-replications; inconclusive rows are excluded. Overlap is counted on the original paper’s byline, so it never exceeds the original’s author count.
Data: replications_database_2026_08_01_120356.csv. Author lookup generated 2026-07-25.