Calcuja Research · October 2026 · OpenAlex · seeded samples

Citation Pattern References 2026: What Is Typical in a Field?

Measured field distributions for four citation-pattern signals — peak-year share, annual growth jumps, coauthors per work and top-10 concentration. This dataset is the comparison basis for the Citation Anomaly Check.

Claims like "a 20% self-citation rate is suspicious" or "50 coauthors is unusual" only mean something relative to a field's actual distribution. This dataset measures those distributions: for each of the 11 field groups in our OpenAlex benchmark sample we computed the middle-80% band (p10–p90) of four citation-pattern signals. The Citation Anomaly Check uses exactly these rows as its reference — there are no hidden thresholds.

// Key Findings
  • Concentration is the norm, not the exception: the median researcher draws ~50% of lifetime citations in a single calendar year — most records are small, so one good year dominates. Only the very top tail (>~85–100% depending on field) is unusual
  • Growth jumps of ~3× are ordinary: the median researcher's largest year-over-year jump is 2.5–3.8× across fields; jumps only look unusual beyond ~15–20× (field p90)
  • Team size is a field property, not a personal trait: median coauthors per work run from 1 (humanities, social sciences, economics) to 3 (medicine, chemistry, engineering), with physics and life sciences reaching 7–8 at p90
  • Field differences dominate personal differences — which is exactly why the anomaly check always compares inside a field band and never against a global cutoff
// Short answer

Across all fields, the typical band (p10–p90) is: peak-year share 22%–100%, largest annual jump 0.6×–16.6×, coauthors per work 1–6. A researcher's values are only "unusual relative to field" outside their own field's band — read the tables by row, not the pooled line.

Methodology

  • Author sample: the same seeded OpenAlex sample as our h-index benchmarks — 63,665 authors across 11 field groups (sample+seed, works_count>5, primary field = top topic's field), snapshot October 2026.
  • Growth signals are computed from each author's counts_by_year: peak-year share = citations in the most-cited year ÷ lifetime citations; largest jump = maximum ratio of consecutive yearly citation counts, only over year-pairs where both years have ≥5 citations (authors without a valid pair are excluded — hence the smaller n).
  • Coauthors per work: a separate seeded random sample of 1,000 works per field group (works?filter=topics.field.id:…&sample=1000&seed=<group seed+77000>), each work contributing its authorship count. The distribution is per-work, not per-author — the check compares a researcher's median work against it, which is stated on the result card.
  • Top-10 concentration: for ≈300 authors per field (the first 300 of each group's seeded sample — a deterministic subsample), the sum of citations of their 10 most-cited works ÷ their total citations. Row n is the achieved sample size at publication; rows are refreshed as sampling completes.
  • Self-citation has no reference row in v1: published self-citation thresholds vary by source and field, and we do not publish cutoffs without our own data basis. The check reports the measured share and its sample size only.
  • Reproducibility: the computation script (scripts/citation-pattern-refs.py in the site repository) is deterministic given the same seeds — the author rows reuse the Block-1 sample order, the works samples use group seed + 77,000.

Field reference distributions

Values are percentiles within each field group. p10–p90 is the "typical band" the anomaly check compares against.

Peak-year citation share

Fieldnp10Medianp90
Medicine & Biology4,40821%48%95%
Life Sciences4,51720%45%89%
Chemistry4,59619%43%86%
Physics & Astronomy4,35519%45%91%
Mathematics7,09021%47%97%
Computer Science4,38725%50%99%
Engineering4,39922%46%92%
Economics & Finance4,70427%56%100%
Social Sciences4,51725%55%100%
Humanities5,28727%58%100%
Environmental & Earth5,22322%49%96%
All fields53,48322%50%100%

Largest annual growth jump (×)

Fieldnp10Medianp90
Medicine & Biology2,7130.6×3.4×17.1×
Life Sciences2,9890.6×3.3×17.3×
Chemistry3,3060.6×3.2×14.9×
Physics & Astronomy2,9330.7×3.8×21.0×
Mathematics4,0940.6×3.3×17.4×
Computer Science2,4580.5×2.9×14.0×
Engineering2,8420.6×3.1×15.4×
Economics & Finance1,9920.4×2.9×17.1×
Social Sciences1,8800.5×3.0×16.2×
Humanities1,4240.5×2.5×14.5×
Environmental & Earth3,0030.6×3.1×16.4×
All fields29,6340.6×3.2×16.6×

Coauthors per work

Fieldnp10Medianp90
Medicine & Biology1,000138
Life Sciences1,000127
Chemistry1,000137
Physics & Astronomy1,000127
Mathematics1,000124
Computer Science1,000125
Engineering1,000136
Economics & Finance1,000113
Social Sciences1,000113
Humanities1,000112
Environmental & Earth1,000126
All fields11,000126

Top-10 citation share

Per-author reference sampling for this metric is still running (target ≈300 authors per field from the same seeded sample). Rows are added to the CSV and this table as field groups complete — the check page reports the measured value meanwhile and marks the field comparison as pending rather than showing an estimated band.

Download & limitations

The full dataset (all metrics, p10–p90, per-field n) is available as CSV under CC BY 4.0 — cite "Calcuja Citation Pattern References 2026".

⬇ Download data.csv
  • OpenAlex is not the universe: coverage differs by field and era (pre-1990 fragmentary); author disambiguation errors distort individual records. Reference distributions describe the database, not the discipline.
  • Bands describe typicality, not normality: a value outside p10–p90 is unusual — not wrong, not fraudulent. The self-check applies exactly this band logic.
  • Drift: citation counts grow daily; percentiles are an October 2026 snapshot. Re-running the seeded samples today produces near-identical draws but slightly shifted values.
  • Companion dataset: h-index and i10 percentiles by field × career stage live in OpenAlex Benchmarks 2026.