The Top 2% Scientists List Explained: How It's Built and How to Read It

What the list is

The “Top 2% Scientists list” is the popular name for a dataset whose official title is Updated science-wide author databases of standardized citation indicators. It is produced by Prof. John P.A. Ioannidis of Stanford University's Meta-Research Innovation Center (METRICS) with colleagues, using citation data from Elsevier's Scopus database, and published openly on the Elsevier Data Repository.

The project began in 2019 with a paper in PLOS Biology. Its aim was to fix a long-standing problem with citation metrics: raw citation counts and h-indexes can't be compared fairly across fields, and they ignore how many people shared the credit for a paper. The database standardises these measures across every Scopus author and annotates each scientist with their field, so researchers can be compared with their peers rather than with everyone.

The authors are explicit about its purpose: the data are published “to facilitate transparency and research”, and all citation metrics “have limitations and their use should be tempered and judicious”. The list was never designed as a prize, even though it is often reported as one.

Who is included, and why it is called “top 2%”

Every author with a Scopus profile is scored. The published tables include a scientist if either of these is true:

• they are among the top 100,000 scientists by c-score (calculated both with and without self-citations), or
• they reach a percentile rank of 2% or above in their subfield, meaning they are in the top 2% of researchers working in the same subfield.

The second rule is where the “top 2%” name comes from, and it is what makes the list fairer across disciplines. A mathematician and a cardiologist are never compared directly; each is compared with others in their own subfield, where citation habits are similar. Percentiles are calculated for every scientist with at least five papers.

How the c-score works

The ranking is based on a composite indicator, the c-score, first described by Ioannidis and colleagues in 2016. Rather than relying on a single number, it combines six:

Each indicator is log-transformed and scaled before they are added together, so no single measure dominates. The result rewards impact rather than productivity: publishing many papers does not help unless they are cited. Three of the six indicators look at author position, so being a single, first or last author on a highly cited paper counts for more than being one of hundreds of co-authors.

Every indicator is reported both with and without self-citations. The dataset also reports the ratio of citations to citing papers and, since the 2024 edition, retracted papers and citations to and from them (using the Retraction Watch database). These columns are there so readers can spot unusual patterns. They are not penalties built into the score.

Career-long vs single-year

Each edition contains two separate rankings, and confusing them is the most common reason people think they are “missing” from the list.

When someone says they are “on the Top 2% list”, it is worth asking which one, and which year. Both are legitimate. They measure different things.

Fields and subfields

Scientists are classified using the Science-Metrix journal classification. Each scientist is assigned the field and subfields that appear most often among their publications, with the percentage for each. The 2026 dataset describes 20 fields and 174 subfields. Editions up to 2025 described 22 fields and 174 subfields, so field-level comparisons between 2026 and earlier years should be made with care.

Because the classification is based on journals, researchers who publish across several areas can find themselves assigned to a subfield they don't consider their main one. The percentage columns show how mixed someone's output is.

The 2026 edition

The 2026 edition was published on the Elsevier Data Repository on 7 October 2026 as Version 9 of the dataset (DOI 10.17632/btchxktzyw.9). In detail:

• Based on the 1 August 2026 Scopus snapshot, using all author profiles as of that date.
• Career-long data run to the end of 2025; single-year data cover citations received in calendar year 2025.
• Selection rule unchanged: top 100,000 by c-score, or top 2% of the subfield.
• Field classification now described as 20 fields and 174 subfields (previously 22 fields).
• Self-citation, citing-paper ratio and retraction columns continue.

It arrived about three weeks later than the 2024 and 2025 editions, which both appeared in mid-September, but in line with the October releases of 2020 to 2023.

Release history

Dates are taken from the Elsevier Data Repository's version record. Version 5 was a corrected re-issue of Version 4, published to improve subfield assignments.

How to check if you are on the list

There are two ways, depending on how much detail you want.

1. Search it on TOPSCINET (quickest)

TOPSCINET is DATAMETA LAB's searchable version of the list. Choose Single Year Data or Career Data, search by name or institution, and the profile shows every edition you appear in. If you are listed, you can claim the profile with your university email address to get a verified badge and a downloadable certificate. More about TOPSCINET.

2. Download the official data

The full tables can be downloaded free of charge from the Elsevier Data Repository. They are large spreadsheets, one for each ranking, and are the authoritative source to cite. Search them by surname, and check your Scopus author ID if your name is common.

If you think you should be on it but aren't

The most common cause is a split Scopus profile: your papers are spread across two or more author IDs, so neither one scores high enough. Merge them with Scopus's Author Feedback Wizard. The published list will not change, but the correction will be used in the next annual edition. The dataset authors ask people not to email them for Scopus corrections.

Five common misreadings

1. “Stanford ranked me in the top 2%.” The list is a research dataset led by a Stanford researcher, built on Elsevier data. Stanford University does not issue it as an award or endorsement.
2. “It's the top 2% of all scientists in the world.” It is the top 2% within each subfield of authors in Scopus with at least five papers, plus the overall top 100,000. Researchers whose work isn't indexed in Scopus aren't in the pool.
3. “A higher rank means better science.” The c-score measures citation impact. Citations reflect attention, which usually tracks importance, but not always, and not equally across fields or types of work.
4. “Not being on it means my work isn't good.” The dataset itself says so plainly: if an author is not on the list, “it is simply because the composite indicator value was not high enough to appear on the list. It does not mean that the author does not do good work.”
5. “The career and single-year lists are the same thing.” They are separate rankings with different populations. Always say which one, and which edition, when you cite it.

Limitations

The list is one of the most carefully constructed citation datasets available, and it is open about its weaknesses. The main ones:

• It is only as good as Scopus. Split or merged author profiles, wrong affiliations and missing journals all carry straight through into the rankings.
• Coverage is not neutral. Scopus indexes some publishers, languages and regions more thoroughly than others. Independent analyses published in Scientometrics in 2026 have examined publisher coverage bias and other anomalies in the database.
• Author-position conventions vary. Three of the six indicators reward single, first or last authorship. Fields that list authors alphabetically, such as mathematics and parts of economics, don't fit that pattern well.
• Books and non-journal outputs are under-counted, which affects the humanities and parts of the social sciences.
• Editions are frozen. Each version is archival and is never corrected after release, so errors stay until the next edition.

None of this makes the list useless. It means it should be used the way its authors intended: as one piece of evidence, alongside judgement, and in the spirit of the Leiden Manifesto for research metrics, which the dataset itself recommends.