The landscape
What the self-identified meta-research literature studies, and what it does not
Every number on this page is derived from the corpus by a rule written in build/lit/describe.py and reproduced by running it. Nothing here was read by a model or chosen by hand. The labels are vocabulary matches, so a label means the record uses a topic’s words, not that a human judged the record to be about that topic.
The field is young and it is accelerating
The corpus holds 1830 records. 1396 of them, 76 percent, were published in 2020 or later.
| period | records |
|---|---|
| 1965-1969 | 2 |
| 1970-1974 | 2 |
| 1975-1979 | 1 |
| 1985-1989 | 4 |
| 1990-1994 | 6 |
| 1995-1999 | 8 |
| 2000-2004 | 8 |
| 2005-2009 | 21 |
| 2010-2014 | 68 |
| 2015-2019 | 314 |
| 2020-2024 | 906 |
| 2025-2029 | 490 |
The earliest records are scattered and mostly editorial. The shape changes around 2015, which is when the vocabulary itself settles: a paper published before then that does this kind of work rarely calls it meta-research, so this corpus does not contain it. That is a limitation of a tier A search and it is stated again in the protocol.
What it studies
A record can carry several labels, so the column sums past the corpus size. 11 records match no topic phrase at all.
| topic | code | records | share |
|---|---|---|---|
| Names itself meta-research | metares |
1812 | 99.0% |
| Reporting quality and guideline adherence | report |
277 | 15.1% |
| Registration, outcome switching and publication bias | reg |
268 | 14.6% |
| Reproducibility and replication | repro |
268 | 14.6% |
| Authorship, citation and metrics | biblio |
163 | 8.9% |
| Statistical practice, spin and p-hacking | stats |
114 | 6.2% |
| Data and code sharing, FAIR | datashare |
84 | 4.6% |
| Retraction, misconduct and integrity | retract |
83 | 4.5% |
| Open access, preprints and predatory publishing | oa |
81 | 4.4% |
| Peer review | peerrev |
76 | 4.2% |
| Conflicts of interest and funding | coi |
64 | 3.5% |
| Research waste and prioritization | waste |
46 | 2.5% |
| Generative AI in research writing and review | aitext |
33 | 1.8% |
| Equity and representation | equity |
21 | 1.1% |
Where the vocabulary is thin
Blocks carry different numbers of phrases and phrases of different breadth, so a count here measures how much of a topic’s vocabulary this corpus uses and not how much attention the topic receives. The counts are not comparable across blocks and a small one is not evidence of a gap.
| topic | records |
|---|---|
| Research waste and prioritization | 46 |
| Generative AI in research writing and review | 33 |
| Equity and representation | 21 |
An earlier version of this page read a field-level conclusion off this table, that meta-research does not study peer review or generative AI. It was wrong, and it was wrong because the labelling vocabulary had been borrowed from the search vocabulary, where narrowness is a virtue. Broad phrases for labelling now live separately in build/lit/describe.py and the counts above are the corrected ones. The episode is left on the page rather than quietly fixed, because a project about research transparency that silently corrects its own record is arguing against itself.
Topics studied together
Pairs of topic labels on the same record, most frequent first. The pairs are what a new study would have to be novel against.
| pair | records |
|---|---|
| reg and report | 59 |
| report and repro | 46 |
| datashare and repro | 40 |
| reg and repro | 40 |
| biblio and report | 31 |
| reg and stats | 30 |
| repro and stats | 28 |
| datashare and reg | 19 |
| datashare and oa | 19 |
| biblio and reg | 18 |
| peerrev and report | 18 |
| biblio and oa | 17 |
| coi and reg | 17 |
| biblio and datashare | 16 |
| biblio and retract | 16 |
| oa and report | 16 |
| repro and retract | 15 |
| reg and retract | 15 |
| report and retract | 15 |
| report and stats | 15 |
Topics over time
Records per half decade, from 2015. A topic that flattens while the corpus grows is losing share of the field’s attention, not gaining it.
| topic | 2015-2019 | 2020-2024 | 2025-2029 |
|---|---|---|---|
| Reporting quality and guideline adherence | 39 | 138 | 96 |
| Registration, outcome switching and publication bias | 37 | 144 | 77 |
| Reproducibility and replication | 52 | 141 | 72 |
| Authorship, citation and metrics | 24 | 83 | 53 |
| Statistical practice, spin and p-hacking | 25 | 53 | 32 |
| Data and code sharing, FAIR | 8 | 55 | 20 |
| Retraction, misconduct and integrity | 9 | 28 | 39 |
| Open access, preprints and predatory publishing | 7 | 51 | 23 |
| Peer review | 12 | 37 | 26 |
| Conflicts of interest and funding | 12 | 34 | 16 |
| Research waste and prioritization | 11 | 25 | 10 |
| Generative AI in research writing and review | 0 | 4 | 29 |
| Equity and representation | 1 | 14 | 6 |
Where it is published
| journal | records |
|---|---|
| Journal of clinical epidemiology | 207 |
| eLife | 74 |
| Proceedings of the National Academy of Sciences of the United States of America | 57 |
| BMJ open | 48 |
| PloS one | 46 |
| BMC medical research methodology | 44 |
| Royal Society open science | 44 |
| BMJ (Clinical research ed.) | 31 |
| Perspectives on psychological science : a journal of the Association for Psychological Science | 24 |
| Systematic reviews | 24 |
| Research synthesis methods | 22 |
| BMC medicine | 20 |
| Research integrity and peer review | 18 |
| BMJ evidence-based medicine | 17 |
| F1000Research | 16 |
| Trials | 15 |
| Cochrane evidence synthesis and methods | 14 |
| PLoS biology | 14 |
| Frontiers in psychology | 13 |
| Journal of dentistry | 12 |
One journal carries more than a tenth of the corpus. A field concentrated in a single venue inherits that venue’s editorial priorities, which is itself a meta-research question and one this corpus does not answer.
How it is indexed
MeSH descriptors on the 1315 records that carry MeSH indexing.
| descriptor | records |
|---|---|
| Humans | 1181 |
| Research Design | 322 |
| Randomized Controlled Trials as Topic | 305 |
| Epidemiologic Studies | 243 |
| Systematic Reviews as Topic | 178 |
| Meta-Analysis as Topic | 155 |
| Bias | 136 |
| Reproducibility of Results | 125 |
| Female | 113 |
| Periodicals as Topic | 110 |
| Biomedical Research | 98 |
| Male | 97 |
| Research Personnel | 80 |
| Treatment Outcome | 74 |
| Clinical Trials as Topic | 70 |
| Cross-Sectional Studies | 68 |
| United States | 61 |
| Publishing | 61 |
| Adult | 58 |
| Research Report | 53 |
| Publication Bias | 53 |
| Research | 52 |
| Animals | 52 |
| Publications | 52 |
| Outcome Assessment, Health Care | 48 |
What kind of studies these are
Publication types, with the container types that describe every journal article removed.
| type | records |
|---|---|
| Review | 242 |
| Meta-Analysis | 191 |
| Systematic Review | 161 |
| Comment | 47 |
| Editorial | 37 |
| Letter | 36 |
| Comparative Study | 32 |
| abstract | 21 |
| Scoping Review | 17 |
| Published Erratum | 17 |
| Historical Article | 15 |
| Observational Study | 12 |
| Preprint | 8 |
| Biography | 7 |
| Network Meta-Analysis | 6 |
Reviews, meta-analyses and systematic reviews together outnumber every other declared type. Many of the records typed as meta-analysis are meta-epidemiological studies, which PubMed has no separate type for, so this table separates declared types and not study designs.