Research statistics, with sources
By Olajuwon Aina, founder of Thesisroom · Updated
This page collects published figures on doctoral completion, citation and quotation accuracy, retractions and AI-generated references, each linked to the study it comes from.
Key figures
- In literature reviews generated by ChatGPT, 55% of the GPT-3.5 citations and 18% of the GPT-4 citations were fabricated.
Source: Walters and Wilder, Scientific Reports, 2023; accessed 2 October 2026.
- In biomedical literature, only 722 (5.4%) of 13,252 post-retraction citation contexts acknowledged the retraction.
Source: Hsiao and Schneider, Quantitative Science Studies, 2021; accessed 2 October 2026.
- A meta-analysis of quotation accuracy in medical journal articles estimated a total quotation error rate of 25.4%.
Source: Jergas and Baethge, PeerJ, 2015; accessed 2 October 2026.
- In the Council of Graduate Schools' PhD Completion Project sample, 57% of doctoral candidates completed their degree programs within a ten-year time span.
Source: Council of Graduate Schools, Ph.D. Completion and Attrition: Analysis of Baseline Demographic Data, 2008; accessed 2 October 2026.
- Of retracted biomedical and life-science research articles indexed by PubMed, 67.4% of retractions were attributable to misconduct.
Source: Fang, Steen and Casadevall, Proceedings of the National Academy of Sciences, 2012; accessed 2 October 2026.
From published research
Each figure describes the sample in the study named beside it.
Doctoral completion and time to degree
Among 2022 US doctorate recipients, education was the field with the longest median time to degree: 12.3 years from graduate school entry to doctorate.
Source: National Center for Science and Engineering Statistics, Doctorate Recipients from U.S. Universities: 2022, 2023; accessed 2 October 2026.
In US science and engineering fields, the longest median time from graduate school entry to a doctoral degree was nearly 9 years, in both health sciences and social sciences.
Source: National Center for Science and Engineering Statistics, Doctorate Recipients from U.S. Universities: 2022, 2023; accessed 2 October 2026.
In the Council of Graduate Schools' PhD Completion Project sample, 57% of doctoral candidates completed their degree programs within a ten-year time span.
Source: Council of Graduate Schools, Ph.D. Completion and Attrition: Analysis of Baseline Demographic Data, 2008; accessed 2 October 2026.
In that sample, ten-year completion ranged from 64% in engineering to 49% in the humanities.
Source: Council of Graduate Schools, Ph.D. Completion and Attrition: Analysis of Baseline Demographic Data, 2008; accessed 2 October 2026.
In a HEFCE study of UK PhD starters, by 2002-03, after seven years, completion rates were 71 per cent for full-time and 34 per cent for part-time starters.
Source: Higher Education Funding Council for England, PhD research degrees: Entry and completion, 2005; accessed 2 October 2026.
Citation and quotation accuracy
A meta-analysis of quotation accuracy in medical journal articles estimated a total quotation error rate of 25.4%.
Source: Jergas and Baethge, PeerJ, 2015; accessed 2 October 2026.
Recalculated per quotation, the rate of quotation errors in medical research articles was 14.5%.
Source: Mogull, PLOS ONE, 2017; accessed 2 October 2026.
In citations from high-impact orthopaedic and sports medicine journals, the overall quotation error rate was 13.6%.
Source: Gazendam and colleagues, JBJS Open Access, 2021; accessed 2 October 2026.
Thesisroom's free reference list check looks up every entry in a list of up to 200 references in the academic record. Check a reference list
Retractions
Of retracted biomedical and life-science research articles indexed by PubMed, 67.4% of retractions were attributable to misconduct.
Source: Fang, Steen and Casadevall, Proceedings of the National Academy of Sciences, 2012; accessed 2 October 2026.
Across retracted publications from the full spectrum of scholarly disciplines, retractions due to alleged publishing misconduct (47%) outnumbered those due to alleged research misconduct (20%).
Source: Grieneisen and Zhang, PLOS ONE, 2012; accessed 2 October 2026.
For retracted articles indexed in PubMed, the time from publication of the article to publication of its retraction averaged 32.91 months.
Source: Steen, Casadevall and Fang, PLOS ONE, 2013; accessed 2 October 2026.
In biomedical literature, only 722 (5.4%) of 13,252 post-retraction citation contexts acknowledged the retraction.
Source: Hsiao and Schneider, Quantitative Science Studies, 2021; accessed 2 October 2026.
Retracted articles in dental literature were cited 530 times after retraction, and 89.6% of those citations did not consider the existence of the retraction notice.
Source: Rapani and colleagues, Clinical and Experimental Dental Research, 2020; accessed 2 October 2026.
AI tools and references that do not exist
In literature reviews generated by ChatGPT, 55% of the GPT-3.5 citations and 18% of the GPT-4 citations were fabricated.
Source: Walters and Wilder, Scientific Reports, 2023; accessed 2 October 2026.
Of the real, non-fabricated citations in the same study, 43% from GPT-3.5 and 24% from GPT-4 included substantive citation errors.
Source: Walters and Wilder, Scientific Reports, 2023; accessed 2 October 2026.
Among references in medical articles generated by ChatGPT, 47% were fabricated, 46% were authentic but inaccurate, and only 7% were authentic and accurate.
Source: Bhattacharyya and colleagues, Cureus, 2023; accessed 2 October 2026.
Of 59 references evaluated in ChatGPT answers to medical questions, 41 (69%) were fabricated, although they appeared real.
Source: Gravel and colleagues, Mayo Clinic Proceedings: Digital Health, 2023; accessed 2 October 2026.
Thesisroom's free reference list check looks up every entry in a list of up to 200 references in the academic record. Check a reference list