How old is a paper when somebody cites it? Every citation is a link between two papers, and each end of it carries a year. Subtract one from the other and you have the age of the cited work at the moment it was cited.
We ran that calculation over the Litlas corpus, across the 413,582,341 citations where we hold both the citing and the cited paper. The answer varies by a factor of 2.7 between fields. Arts and humanities papers are cited 13.3 years after publication and mathematics 12.5, while energy is cited at 5.0 years, engineering at 5.8 and computer science at 5.9. Here is the table for all 26 fields, followed by where that gap comes from.
Fields differ by a factor of 2.7 in how long they take to be cited
For each citation we take the publication year of the citing paper minus the publication year of the cited paper, and we call the difference the age of the citation. Group the citations by the subject field of the cited paper, take the median, and you have that field’s cited half-life. It is a standard quantity in bibliometrics — the statistic Clarivate publishes per journal in the Journal Citation Reports, defined there as “the median age of the citations received by a journal during the JCR year.” We compute the same statistic per field rather than per journal, and over every citation year at once rather than a single year.
| Field | Median age | Middle half | Citations counted |
|---|---|---|---|
| Energy | 5.0 | 2.7–9.0 | 1,223,442 |
| Engineering | 5.8 | 3.0–11.3 | 33,493,894 |
| Computer Science | 5.9 | 3.0–11.1 | 81,232,919 |
| Materials Science | 6.2 | 3.3–11.4 | 6,183,745 |
| Immunology & Microbiology | 6.2 | 3.2–11.5 | 11,021,388 |
| Medicine | 6.4 | 3.2–12.0 | 99,042,289 |
| Biochemistry, Genetics & Molecular Biology | 6.4 | 3.3–11.9 | 66,835,145 |
| Pharmacology & Toxicology | 6.4 | 3.3–12.3 | 2,032,180 |
| Dentistry | 7.0 | 3.5–13.4 | 1,583,220 |
| Nursing | 7.1 | 3.6–13.3 | 2,990,452 |
| Environmental Science | 7.1 | 3.6–13.3 | 8,694,202 |
| Health Professions | 7.2 | 3.6–13.3 | 4,946,999 |
| Agricultural & Biological Sciences | 7.4 | 3.7–13.9 | 10,723,207 |
| Chemical Engineering | 7.6 | 3.4–17.5 | 437,045 |
| Neuroscience | 7.6 | 3.9–13.7 | 22,636,835 |
| Social Sciences | 7.9 | 3.9–15.0 | 10,631,153 |
| Chemistry | 8.0 | 3.8–15.9 | 4,464,569 |
| Business & Management | 8.5 | 4.3–15.1 | 5,394,112 |
| Veterinary | 8.6 | 4.1–15.8 | 305,977 |
| Physics & Astronomy | 8.7 | 4.0–17.5 | 7,534,212 |
| Decision Sciences | 8.9 | 4.3–16.7 | 6,441,633 |
| Psychology | 8.9 | 4.3–16.5 | 11,391,772 |
| Earth & Planetary Sciences | 9.3 | 4.5–17.2 | 3,410,810 |
| Economics & Finance | 10.1 | 4.6–19.6 | 3,032,591 |
| Mathematics | 12.5 | 5.4–24.2 | 6,168,418 |
| Arts & Humanities | 13.3 | 6.0–25.1 | 1,730,132 |
“Middle half” is the 25th to the 75th percentile, taken over citations rather than over papers: a paper with 500 citations contributes 500 observations. Read the Mathematics row like this. Half of the citations that mathematics papers received arrived 12.5 or more years after publication, and a quarter of them arrived 24.2 or more years after.
The fast end sits at 5.0 years and the slow end at 13.3, a spread of 2.7×. JCR also caps the cited half-life it publishes at 10 years and displays anything longer as “>10” — three of these 26 fields are past that cap: Economics & Finance, Mathematics and Arts & Humanities.
These numbers also tell you how far back a search for prior work has to reach. In mathematics or the arts and humanities, papers over ten years old are still being cited, so a search limited to the last few years drops the work everything else is built on. Once you have fixed that window, how to find prior research for a thesis covers how wide to cast and where to stop.
The gap is not an artefact of how the corpus was assembled
There is an obvious rival explanation. Litlas does not ingest every paper in existence: computer science and the biomedical fields enter the corpus from 10 citations upward, and every other field enters from 200 upward. Both thresholds are published on our data sources page. If most fields only get in once they have 200 citations, then the papers standing in for those fields are the well-cited ones, and well-cited papers keep attracting citations for longer. On that reading, the gap between fields would say nothing about the fields and everything about the corpus.
The table refutes it without any extra work. The two fastest fields, Energy at 5.0 years and Engineering at 5.8, both sit behind the 200-citation threshold, and so does Materials Science at 6.2. Mathematics at 12.5 and Arts & Humanities at 13.3 sit behind that same threshold. A threshold that manufactured long citation ages would have to manufacture them for every field behind it.
We ran the control directly as well, recomputing the entire table with every field held to the same 200-citation bar, so that computer science and medicine faced exactly what the other fields faced. Medians moved by at most 2 years, in both directions, and the distance between the fastest and the slowest field did not narrow. Whatever produces the gap, it is not the ingestion threshold.
The ranking survives a small sample; the number of years does not
Run the same calculation on a sample of roughly 4,100 papers and the years do not come close.
| Field | Sample of 4,100 | Full corpus | Overstated by |
|---|---|---|---|
| Computer Science | 9 | 5.9 | 1.5× |
| Medicine | 11 | 6.4 | 1.7× |
| Biochemistry, Genetics & Molecular Biology | 12 | 6.4 | 1.9× |
| Materials Science | 15 | 6.2 | 2.4× |
| Physics & Astronomy | 23 | 8.7 | 2.6× |
| Mathematics | 25 | 12.5 | 2.0× |
| Psychology | 25 | 8.9 | 2.8× |
| Engineering | 27 | 5.8 | 4.7× |
| Arts & Humanities | 32 | 13.3 | 2.4× |
Every field reads high, by a median factor of 2.4 and a worst case of 4.7. A few thousand papers cannot estimate this quantity, because the median age is set by a long tail of old citations and a sample that size does not hold enough of the tail to place it.
The ranking, on the other hand, mostly holds. Take the nine fields two at a time and you get 36 pairs; between the sample and the full corpus, 26 pairs kept their relative order, 9 reversed, and 1 was a tie — a Kendall’s tau-b of +0.48, where +1 means the two rankings agree everywhere and −1 means they are exact opposites. Seven of the nine reversals involve one field. The sample put Engineering near the slow end at 27 years; the full corpus puts it second fastest at 5.8. Set Engineering aside and only two of the remaining 28 pairs reverse (tau-b = +0.84), and both are pairs this method cannot really separate: Medicine and Biochemistry at 6.4 against Materials Science at 6.2, a gap of two tenths of a year. Every pair further apart than that kept its order.
So the ranking is the part you can check against a small sample of your own. The years are not.
What the table covers
Only citations with both ends in the corpus are counted. Each paper in the corpus reports how many citations it has received in total, and summed across the corpus that is 1,516,896,516. The number where both the citing and the cited paper are inside the corpus is 413,996,658, which puts the visible share at 27.3%. The other 72.7% point at papers we do not hold; those have no publication year and no subject field on the far end, so they cannot enter the calculation, and we cannot say which way they would move it if they could. Two checks confirm the visible part is internally consistent. Summing each paper’s own year-by-year record of the citations it received gives 413,952,032, which differs from the link count by 0.011% — the difference is exactly the citations coming from papers with no publication year. And of the 14,083,496 papers that carry such a record, 99.73% have a record whose total equals the number of links pointing at them.
The corpus holds very little old work to cite. Only 6.33% of its papers were published before 1990, and 53.71% were published in 2015 or later. A citation only enters the calculation when both ends are inside the corpus, so the reachable past is short by construction. The table above describes this corpus. It is not an estimate of a true half-life for the literature.
Each paper carries exactly one field, and the table uses the field of the cited paper. A physics paper cited by a medical paper counts once, as a physics citation, on the physics row. Interdisciplinary work is therefore filed under one heading and cannot show up as a category of its own. A further 1,913,334 papers, 9.87% of the corpus, carry no field at all, and their citations are absent from the table.
How to check the arithmetic
The numbers above come from a single pass over the corpus as it stood on 2026-08-22. The whole result rests on one count, so the count is worth stating precisely.
- The age of a citation is the publication year of the citing paper minus the publication year of the cited paper.
- One tally is kept for each pair of (field of the cited paper, age), over every citation with both ends in the corpus. That is 413,582,341 citations, or 99.90% of the visible ones; the remainder fell out because the cited paper carries no publication year or no subject field.
- Percentiles are taken over citations, not over papers, so a paper with 500 citations contributes 500 observations to its field’s row.
- A citation of age k is treated as falling anywhere in the twelve months from k to k+1, and every percentile is interpolated linearly inside the band that crosses the mark. This is how JCR computes a cited half-life, and it is why the table carries one decimal place instead of whole years. Without the interpolation, taking instead the first whole year in which the running total passes half, the two ends of the table come out at 4 and 13 rather than 5.0 and 13.3. Quoting one end from one rule and the other end from the other rule is how you get a range nobody can reproduce.
- Negative ages — where the citing paper is dated before the paper it cites, which is ordinary in a corpus holding preprints and versioned records — are kept and counted below every other age. They are 0.88% of the table.
- Nothing is weighted, normalised per year, or truncated to a citation window. Every citation year in the corpus is pooled into one table.
The two denominators — 19,379,997 papers and 413,996,658 citation links — are published on the data sources page, and both were read from the same build, so they describe one corpus rather than two.
Whether to drop the negative ages is the one step here a reasonable person could take differently. Drop them and recompute, and Arts & Humanities goes from 13.3 to 14.0, Mathematics from 12.5 to 12.9, and Economics & Finance from 10.1 to 10.5; the fast end does not move at one decimal place at all. It moves that way because negative ages are not spread evenly: they are 0.88% of the whole table but 4.30% of the citations to Arts & Humanities papers, the highest share of any field. Dropping them lifts the slowest field more than any other, which is why the table above keeps them. The ordering barely notices either way — compare the 26 fields pairwise under the two rules and 321 of the 325 pairs keep their order, with 4 reversals.
Following citations in Litlas
Litlas searches 16 academic sources with one query, opens the citation graph for any paper so you can walk backward into what it was built on and forward into what happened afterwards, and saves what you keep into a library with your own notes attached. When you start writing, you can export those citations as BibTeX or RIS.
The table above suggests where to spend the reading time. If your field sits at the slow end, work from a decade ago is still being cited around you, and the papers that qualified or extended it appeared years apart, so the forward direction is the one that repays a careful pass. If your field sits at the fast end, the backward direction closes quickly and the forward direction is where the corrections are.
The free plan needs no card and works right away.
Bibliography
- Clarivate. A Closer Look at Cited and Citing Half-Lives. 11 May 2017. https://clarivate.com/academia-government/blog/a-closer-look-at-cited-and-citing-half-lives/. Accessed 22 August 2026.
- Larivière, V., Archambault, É., Gingras, Y. Long-term variations in the aging of scientific literature: From exponential growth to steady-state science (1900-2004). Journal of the American Society for Information Science and Technology, 2008, 59(2), p. 288-296. https://doi.org/10.1002/asi.20744
- 棚橋佳子「情報の寿命を表す指標：被引用半減期」『情報の科学と技術』第 70 巻第 1 号、2020 年、30–33 頁。https://doi.org/10.18919/jkg.70.1_30