The headline finding for anyone building a business case on completion: undergraduate graduation is a mandatory statistical return in the systems that collect it; graduate completion largely is not. The best-known evidence on doctoral completion and attrition came from a time-limited grant project rather than a recurring series, and the definitive enrolment source for US graduate education is a membership survey. Neither of those is a criticism. Both change how you should cite.
What follows is a source map: who produces what, what each source actually covers, and where the joins are.
Why graduate completion is a harder measure than undergraduate graduation
Before blaming the statistical agencies, it is worth being precise about why the metric resists collection.
- The window is not fixed. Undergraduate graduation rates work because a normative duration exists and a multiple of it is a defensible cut-off. Doctoral completion times vary by field to a degree that makes any single window arbitrary.
- Interruption is normal. Parental leave, fieldwork, illness, funding gaps and employment all interrupt graduate study routinely rather than exceptionally. A cohort that is not continuously enrolled is not a cohort a fixed-window measure can follow.
- Exit awards blur the outcome. A doctoral candidate leaving with a master’s has not completed the programme they entered and has not left empty-handed. Whether that counts as completion changes the rate materially.
- Part-time and professional study dominate at master’s level. Aggregating professional master’s programmes with research master’s degrees produces an average of unlike things.
Every one of these is solvable within one institution, which is why local data is usable and national comparison generally is not.
The source map
| Source | What it covers | Type | What to watch |
|---|---|---|---|
| National statistical collections (US federal, UK, EU member-state) | Enrolment, awards conferred, demographics | Mandatory return | Awards conferred is a flow, not a completion rate — see below |
| CGS/ETS Survey of Graduate Enrollment and Degrees | Applications, first-time and total graduate enrolment, degrees and certificates conferred | Membership survey | Coverage is participating institutions, not the whole system |
| CGS Ph.D. Completion Project | Doctoral completion and attrition | Time-limited grant project | Not a live series; check the vintage before citing |
| CGS Doctoral Initiative on Minority Attrition and Completion | Completion and attrition among under-represented minority STEM students | Initiative dataset | Scoped population; do not generalise |
| Institutional annual reports and planning data | Local cohort outcomes | Internal | The most usable data you have, and the least comparable |
| Funder and research-council reporting | Outcomes for funded candidates | Administrative | Funded candidates are not representative of all candidates |
Two entries deserve expansion because they are the ones most often mis-cited.
The CGS/ETS survey. The Council of Graduate Schools describes it as “the only source for data on both first-time and total graduate enrollment across all fields of master’s and research doctorate programs”, covering “applications for admission to graduate school, graduate student enrollment, and graduate degrees and certificates conferred”. Its most recent publication is the Graduate Enrollment and Degrees 2025 Report, presenting data from the 2024 survey. Note the phrase “the only source” — it is the survey owner’s own characterisation, and if it is right, then the definitive series for US graduate enrolment is produced by an association from member responses rather than by a statistical agency from a compulsory return. That is a materially different provenance and belongs in your footnote.
The Ph.D. Completion Project. CGS describes it as “a seven-year, grant-funded project that addressed the issues surrounding Ph.D. completion and attrition”, producing publications on completion, attrition, demographic analysis and policies promoting student success. A seven-year project is not a recurring series. Figures from it can be cited — with their vintage — as evidence about the period studied. They cannot be cited as current national completion rates, and they routinely are.
The mistake that produces most wrong numbers
It is the same one that appears wherever enrolment and award data sit side by side: dividing degrees conferred in a year by students enrolled in that year and calling the result a completion rate.
That ratio fails four ways at once. It divides an annual flow by a point-in-time stock. It mixes reference periods, since award years and enrolment years rarely align. It assumes a closed cohort, when graduate populations gain and lose members continuously. And it aggregates programme types with different durations, so the denominator is dominated by whichever type is currently growing.
The result is a number that moves when programme mix changes and stands still when completion actually improves. If your paper contains a completion percentage and you cannot name the cohort it followed, it is probably this ratio.
What to do instead
- Use awards conferred, and call it that. “The institution conferred N doctorates in 2024” is defensible, sourceable and unambiguous. It is not a rate and should not be presented as one.
- Build one local cohort measure and keep it stable. Define entry, define the window, define how interruptions and exit awards are treated, then never change the definition without dating the change.
- Report the definition alongside the number, every time. A completion figure without its window and its treatment of interruption is not interpretable by anyone who did not build it.
- Treat project findings as period evidence. Cite them with their study period, and do not extrapolate them to the present.
- Ask peers for definitions before figures. If a consortium wants comparability, the definition negotiation is the whole task; the data collection is trivial by comparison.
Point three is the one that most often gets a paper through a scrutiny committee intact. Point five is the one that makes benchmarking possible at all.
How this interacts with data currency
Provenance and currency are different problems and they compound. A membership survey has a publication lag and a coverage limit; a completed project has a fixed study period and no update path. When both apply, the figure in your paper may be describing a cohort that entered study more than a decade ago.
The test to apply before a number goes into a board paper is in our guide to how current your higher education data actually is, and the parallel argument about integrity figures — that a count can measure your own processes rather than the world — is in why integrity case counts measure enforcement.
Why this matters for a business case
Completion and time-to-completion are the outcomes most often used to justify investment in graduate support, and they are among the weakest-sourced numbers in higher education. That is an argument for building a local instrument rather than for abandoning the case.
A defensible business case names a local cohort, a stable definition, a baseline measured before the intervention and an outcome measured after it. It does not rest on a national completion percentage, because the national completion percentage — in the sense the paper wants to use it — largely does not exist. Designing that instrument alongside the deployment rather than after it is the method in our guide to running a departmental pilot.
To discuss what a measurable graduate-support baseline looks like at your institution, request an institutional evaluation.
Frequently asked questions
Where does graduate completion data come from?
Chiefly membership surveys and time-limited research projects rather than mandatory statistical returns, plus institutions’ own cohort tracking.
What is the CGS/ETS survey?
A survey of graduate applications, first-time and total enrolment, and degrees and certificates conferred, described by its owner as the only source covering both first-time and total graduate enrolment across all fields of master’s and research doctorate programmes.
Is the Ph.D. Completion Project a current series?
No. CGS describes it as a seven-year, grant-funded project. Cite its findings with their study period.
Why is graduate completion harder to measure than undergraduate graduation?
Variable duration, routine interruption, exit awards, and the aggregation of professional with research programmes all defeat a fixed-window cohort measure.
Can I divide degrees conferred by students enrolled?
No. That divides a flow by a stock across mismatched reference periods and an open cohort. It is not a completion rate.
Can we benchmark completion against peers?
Only with their definitions in hand. Window length, interruption treatment and exit awards each move the figure more than the differences you are trying to detect.
What is the safest figure to publish?
Awards conferred in a stated year, labelled as such.
What should a local cohort measure define?
Entry point, window, treatment of interruption, treatment of exit awards, and the date the definition was last changed.
Are funder outcome statistics usable?
For funded candidates, yes. Funded candidates are not representative of the whole population, so do not generalise.
How old is too old for a completion figure?
There is no threshold, but a figure describing a cohort that entered before the current funding, supervision and assessment regime should be labelled as historical.
Does this differ by country?
The specific bodies differ; the structural problem does not. Undergraduate completion is collected more systematically than graduate completion in every system we have examined.
What should we do about the gap?
Build one stable local series and negotiate a common definition with a small number of comparable institutions. That produces the only genuinely comparable data likely to exist.