Preregistration, Amendment 4 — What standard meta-analytic summaries cannot tell: sign conflict, partial identification, and evidence from Cochrane reviews

Preregistration Amendment 4 of a single-author methods study on what standard meta-analytic summaries (pooled estimate, Q, I², τ², prediction interval) can and cannot identify about sign conflict among the included studies, with an application to Cochrane reviews (Schwab 2020 CDSR extraction, CC-BY 4.0, https://osf.io/xjv9g/; SHA-256 of CDSR.RData e71210eb70a9601d7d3a56fef2b08cb63bf3a346a132c4a2c934de0a45541ba5).

This deposit fixes, before it is run, the analysis plan of Amendment 4 (data hygiene and subsets, a bracket for the bounded non-identified zone, a binomial check of the known-SE approximation, and fair comparators at realistic precision). Amendments 1–3 were committed to git before being run; git timestamps are author-set and are disclosed as such. This is the first amendment with an independent timestamp.

Files: PREREGISTRATION_AMENDMENT_4.md (plan and predictions), THEORY_v0.3.md (theory at the time of registration), code/a4_corpus.py and code/a4_sim.py (analysis code, not yet run). SHA-256: amendment 15c7a50ab2b8254c21fda4bbb3f311a34cb0828796b742e65394722c6ba2563e; theory 92a4d5c88cfd1bb5ef5932847296488b69c45f8876aa10d9ddca3bbdd6feea73; a4_corpus.py e0f8cd1f07f20ffba94e75166e4365f3397f7bb0f6ae89a8e3653e03130e17d2; a4_sim.py e232888c43c8866cd5fccd1a88f00fa96a2d14bd3f57e92ab4d9022da51e2240. Code imported from earlier stages: cdsr_a3.py 30ad51c95c5fce610448bde27f1f36e701cadb8a9d21bcc31f8f5d9b5abc30c7; cdsr.py 8291505e5ad6d67b668fef2636c8dbf8d470e3a66dfb3c6b6f0e1fb2e95b5d05; realdata.py bff88ea52866eda7e45cb0f014f43016133b5401ea8ccc91d7fa961180cbe674.