Access Metrics
Download counts indicating intent to use
Justification
Download counts indicate intent to use data, a stronger reuse signal than views. Downloads represent a concrete action versus passive discovery.
Practical Guide
Track downloads. Strongest outcome predictor — 41.5x citation lift.
Download counts are the strongest single predictor of citation impact in our data. Datasets with any downloads receive 41.5x more citations (RR = 41.50, p < 0.001). Downloaded data gets cited — this validates the entire SHARE framework. The 1.9% of Zenodo datasets with zero downloads represent abandoned or inaccessible records.
For Repositories
- Implement COUNTER-compliant download tracking
- Distinguish unique downloads from bot traffic
- Report standardized download metrics via Make Data Count
For Depositors
- Monitor downloads as the best proxy for actual data reuse
- If downloads are low, check that your data files are accessible and clearly described
- Downloads validate that your deposit-time metadata is working
Outcome metric — strongest predictor in the entire framework. Cannot be controlled at deposit time but validates that good metadata drives downloads.
Standards Sources
Convergence score: 1/4 independent sources —
| Standard | Field / Property | Obligation Level |
|---|---|---|
| COUNTER Code of Practice | Dataset downloads | Standard |
| Make Data Count | Standardized usage metrics | Standard |
FAIR Principle Alignment
Primary mapping: Outcome metric (not FAIR-derived)
This is an outcome metric not derived from FAIR principles. The R (Reuse) bucket intentionally measures realized impact rather than metadata quality, enabling validation that deposit-time signals predict downstream use.
How This Signal Is Measured
Total unique downloads. Binary for v1: any downloads = 1.
Empirical Evidence (Zenodo, n=1.3M)
Per-signal statistics use Zenodo as the primary validation source because it is the largest general-purpose repository with structured DataCite metadata, natural variance across all 25 signals, and available citation/usage data. Domain-specific repositories exhibit ceiling effects or restricted variance that preclude per-signal discrimination. Cross-repository validation is reported separately.
Prevalence
98.1%
of Zenodo datasets
Citation Lift
41.1x
vs. datasets without
Data Source
Zenodo (CERN)
1,328,100 records analyzed
Interpretation: Near-universal (98.1%). The 1.9% without any downloads represent abandoned or inaccessible records. Downloads are the strongest single outcome predictor — 41x citation lift confirms that downloaded data gets cited.
Quantitative Evidence
Scoring Formula
log₁₀(downloads + 1) × (4 / log₁₀(max_downloads))
Contribution: 4 of 100 points · Reuse bucket (0–20)
With Signal Present
1,303,290
datasets (98.1%)
μ = 0.249 citations/dataset
Without Signal
24,810
datasets (1.9%)
μ = 0.006 citations/dataset
Rate Ratio
41.50
95% CI: [35.34–48.73]
P-value
< 0.001
z = 45.45
Significance
Method: Poisson rate ratio · Source: Zenodo (n = 1,328,100)
Note: Strongest single outcome predictor (41.5× lift). Scored on continuous log scale. Downloaded data gets cited.
R — Reuse Bucket
All signals in this bucket: