Is the data meta-research deposits ever reused?
Written against the design standard. The FAIR instrument, rfair, is maintained by the author. The author’s published work measured FAIRness of shared data in dental research (PMID 35656591) and COVID-19 research (PMID 39556553) as a prevalence; this asks whether FAIRness is associated with use.
1. The question
Meta-research argues that shared data are valuable because others can reuse them, and it deposits its own data on OSF, Zenodo and similar services. Whether anyone reuses those deposits, and whether the FAIR properties the field recommends make a difference to reuse, has not been measured for the field’s own data.
2. Prior work, and the search behind the novelty claim
Log key P29.
corpus regex: (data (reuse|citation)|dataset citation|reuse of (shared|deposited) data)
PubMed: ("data reuse"[tiab] OR "dataset citation*"[tiab] OR "data citation*"[tiab]
OR "reuse of shared data"[tiab]) AND (repository[tiab] OR repositories[tiab]
OR Zenodo[tiab] OR "Open Science Framework"[tiab] OR Dryad[tiab] OR figshare[tiab])One hundred and thirty-two PubMed records, forty-two in the corpus. The relevant ones relate data properties, curation and funding to reuse in general (PMID 36246529), survey scientists’ sharing and reuse practices (PMID 32160189), and set out a data citation roadmap for repositories (PMID 30971690). None measures reuse of meta-research datasets.
3. Hypothesis
Meta-research datasets with a higher FAIR score receive more citations as datasets, per year since deposit.
Null. The rate ratio per 10-point FAIR score increase is 1.
What would falsify it. A rate ratio at or below 1: FAIRness, as scored, does not predict use.
The share of deposits never cited is reported descriptively. A very high share is a finding in its own right, and it is also a precision problem the analysis has to handle (section 9).
4. Design and unit
Retrospective cohort of deposits. One dataset identifier is one unit.
5. Sample
Dataset identifiers cited in data availability statements of tier A and frame articles with PMC full text, read from ext-link xlink:href and parsed with rfair::id_parse (S2), restricted to repository hosts that mint DOIs. A deposit cited by several articles is one unit, deduplicated by DOI (S5).
6. Exposure
FAIR score from rfair::assess_fair, run once at a dated snapshot, with the metric version recorded (rfair_metric_versions).
7. Outcome and instrument
Citations to the dataset DOI, from DataCite event data and OpenAlex, excluding citations from the depositing article and from other articles by its authors. Exposure time is months since deposit, censored at the snapshot (S10).
8. Validation gate
The author checks 60 deposits by hand: whether the DOI resolves to the data described, and whether the recorded citations are genuine uses. If fewer than 80 percent of recorded citations are genuine, the outcome is restricted to hand-verified citations on a sample.
9. Analysis
In R.
glmmTMB(n_cites ~ I(fair_score / 10) + host + offset(log(months_since_deposit)),
ziformula = ~ 1, family = nbinom2, data = d)A zero-inflated negative binomial, because most deposits are expected never to be cited and a model that ignores that would let the zeros dominate the rate ratio.
10. Threats
| threat | handling | removed or disclosed | standard |
|---|---|---|---|
| Dataset citation is badly recorded in general | Gate on genuine citations; a sample-based fallback | Disclosed | S8 |
| Identifiers read from text lose characters | Read from markup and parsed with rfair |
Removed | S2 |
| Host determines both FAIR score and visibility | Host as a covariate | Disclosed | |
| The author maintains the FAIR instrument | Stated at the top | Disclosed | |
| Self-citation inflates reuse | Depositing article and its authors’ citations removed | Removed |
11. Probes
| probe | computes | goes to | what would change the design | status |
|---|---|---|---|---|
| PA | dataset DOIs recoverable from markup across tier A and frame, deduplicated | 5 | fewer than 200 deposits makes the rate ratio imprecise | not run |
| PB | rfair::assess_fair success rate and runtime on 50 deposits |
6 | failure on more than 20 percent changes the metric version or host scope | not run |
| PC | DataCite event data reachability and coverage for those DOIs, marginally | 7 | no events recorded for any host makes the outcome unmeasurable | not run |
| PD | the author’s 60 hand checks | 8 | under 80 percent genuine triggers the restricted outcome | needs the author |
12. What this cannot show
- Reuse that does not cite the dataset, which is common and invisible here.
- Whether FAIRness causes reuse; FAIR deposits may come from groups whose work is more used anyway.