Your Data Science Degree Has No Accreditation Label. Stop Evaluating It Like One.

Bottom line up front: Computer science has a pan-European accreditation label (Euro-Inf); data science does not. It is an interdisciplinary field stitched from statistics, computing and domain knowledge, with heterogeneous curricula and no professional body that accredits its degrees. In that vacuum, the labour market judges graduates by a reproducible project portfolio — and the discipline's own reproducibility crisis defines what actually matters: can this graduate produce analysis another person can trust and rerun? A satisfaction Likert average answers none of that. Data science programmes should be evaluated for reproducible, ethical, domain-transferable capability, not for how much students enjoyed the module.

The accreditation asymmetry

Computing programmes can seek the Euro-Inf Quality Label, awarded by EQANIE (founded in Düsseldorf on 9 January 2009) against an outcomes-based framework developed by the Euro-Inf Project of 2006–2008; by 2021 programmes from 21 countries had been accredited under it. Data science has no equivalent. The closest artefacts are competence frameworks, not accreditation regimes:

• The EDISON Data Science Framework (EDSF) came out of the Horizon 2020 EDISON project (2015–2017). It has four parts — a competence framework (CF-DS), a body of knowledge (DS-BoK), a model curriculum (MC-DS) and professional profiles — and is now community-maintained via the University of Amsterdam. It is a reference model, not a quality label; nobody is "EDSF-accredited."
• In the UK, the Alliance for Data Science Professionals launched in 2021 (a coalition of the Royal Statistical Society, BCS, the Operational Research Society, the IMA, the Alan Turing Institute and the National Physical Laboratory), certifying individuals as "Data Science Professional" or "Advanced Data Science Professional" — the first cohort of 13 was recognised in July 2022. It certifies people, not programmes.
• Even in the US, ABET only approved data-science program criteria for public review in July 2020, piloting accreditation in 2021–2022 — the field standardised late everywhere.

The absence matters because it removes the external anchor that other verticals lean on. A nursing programme or a computing programme can point to a competence standard when a satisfaction number looks flattering. A data science programme cannot — which makes its choice of evaluation evidence entirely its own responsibility.

Curricula do not agree with each other

Because there is no accreditation, data science curricula vary widely. A study of US undergraduate data science degrees (Oliver & McNeil, PeerJ Computer Science, 2021) found programmes score high on statistics and computer science but fall short on domain-specific context, communication and ethics — with only around half requiring any ethics coursework. The same authors noted US undergraduate data science programmes grew from 13 in 2014 to at least 50 by September 2020, and that which academic unit runs the degree substantially shifts the statistics-versus-computing balance. The intended breadth is set out in the US National Academies' Data Science for Undergraduates (2018), but nothing enforces it.

The upshot: two graduates with the same degree title may have been taught profoundly different things. A programme mean cannot tell you which competences a cohort actually built — and averaging satisfaction across a heterogeneous curriculum is close to meaningless.

The reproducibility crisis is the real evaluation target

Data science exists to produce trustworthy analysis, and the field knows trustworthiness is under strain. Monya Baker's survey of 1,576 researchers (Nature 533, 452–454, 2016) found more than 70% had tried and failed to reproduce another scientist's experiments, more than half had failed to reproduce their own, and 52% agreed there was a significant reproducibility crisis. For a data-science graduate, reproducibility is not an abstract virtue — it is the core professional competence: version control, documented pipelines, tested analysis, transparent reporting. The community has even built open teaching resources for it, such as the Alan Turing Institute's The Turing Way, a handbook for reproducible, ethical and collaborative data science.

So the question a data-science evaluation should answer is: did this programme teach students to produce work another person can trust and rerun? Whether a student found the lectures engaging is, at best, weakly related — and the feeling of learning is a poor proxy for learning.

Why the portfolio is the market signal — and its limits

With no credential to trust, employers judge data-science graduates on a reproducible project portfolio: real datasets, a public repository, a defensible methodology, a clear write-up. This is a good market mechanism — but it is a lagging, self-selected signal. Only the students who choose to build and polish a portfolio get judged by it; the median graduate who does not is invisible to it. That is exactly the gap a programme-level evaluation should fill: a leading signal of whether reproducibility, ethics and domain application are being taught to everyone, not just the keen minority who already have GitHub profiles.

That the EU badly needs these graduates only raises the stakes. The Digital Decade targets 20 million ICT specialists in the EU by 2030, but only around 10.3 million were employed in such roles in 2024 — roughly half — and just 56% of EU adults had basic digital skills in 2023, 24 points below target (Cedefop). Producing data scientists who can be trusted is a public-interest question, not just an institutional one.

What to evaluate instead of satisfaction

1. Reproducible-analysis capability. Ask students what they can now do — build a documented, version-controlled, rerunnable pipeline — and triangulate with capstone and project assessment, not a Likert item.
2. Ethics and domain transfer. Given the documented ethics gap, evaluate whether the programme built ethical reasoning and the ability to apply methods in a real domain — the transfer that actually matters for employability.
3. Alignment to demand. Map programme outputs to live skills demand and digital-competence frameworks, so the heterogeneous curriculum is anchored to something external even without accreditation.

But doesn't the market already sort this out via portfolios?

The strongest objection is that portfolios and technical interviews already reveal who can do the work, so programme evaluation is redundant — let employers filter. This misreads what evaluation is for. The portfolio judges the finished, motivated individual after the fact; it gives the institution no early, representative signal about whether the curriculum is building reproducibility and ethics across the whole cohort — including the students who will never build a showcase repository. Evaluation is the programme's own instrument for improvement, and "the labour market will punish us eventually" is not an improvement strategy.

A second objection: students cannot reliably judge whether their own analysis is reproducible (the Dunning–Kruger problem). Correct — which is why student voice should be one strand, triangulated with assessed project artefacts, never the sole verdict. The aim is a richer evidence base, not a new single number to replace the old one.

Where Koji fits

Koji is an AI-native course-evaluation platform suited to exactly this problem. Rather than averaging a scale item, its AI-moderated conversational interviews probe what a student can actually do — walk me through how you would make this analysis reproducible — while its six structured question types capture both structured competence ratings and open narrative. Automatic thematic analysis surfaces recurring gaps (say, "we were never taught to document our pipelines") across a cohort, and quality scoring flags thin or low-effort responses. Programme-level reporting ties module feedback to the wider curriculum picture that a heterogeneous, unaccredited degree badly needs, formative mid-cycle collection catches a broken module before the capstone, and closing-the-loop action tracking records the fix. The moderation is standardized and bias-aware, and data handling is GDPR/AVG-compliant.

Koji does not replace assessed portfolios or manufacture reproducibility — it surfaces whether your programme is teaching the trustworthy-analysis capability the field runs on, and mitigates the distortion of a single averaged score. Teams running wider user or product research often use the same AI interview engine on the main Koji platform.

If your data science degree has no accreditation anchor, a satisfaction mean is the weakest possible substitute for one. Explore Koji for Education to evaluate reproducible, ethical capability instead.

Frequently asked questions

Is there an accreditation label for data science degrees like Euro-Inf for computer science? No unified one. Computer science has the Euro-Inf label from EQANIE, but data science relies on competence frameworks such as the EU's EDISON Data Science Framework and individual certifications like the UK Alliance for Data Science Professionals — neither of which accredits programmes.

Why does the reproducibility crisis matter for evaluating a data science degree? Because reproducibility is the field's core professional competence. A 2016 Nature survey of 1,576 researchers found more than 70% had failed to reproduce others' work and over half had failed to reproduce their own. The thing worth evaluating is whether graduates can produce trustworthy, rerunnable analysis — not whether they enjoyed the module.

If employers judge portfolios, why does course evaluation matter? A portfolio is a lagging, self-selected signal that only captures motivated students who build one. Course evaluation is the programme's leading, representative signal of whether reproducibility and ethics are being taught to the whole cohort, and its own instrument for improvement.

What should a data science programme actually measure? Reproducible-analysis capability, ethical reasoning and domain transfer — triangulated with assessed capstone and project artefacts and mapped to live skills demand — rather than a satisfaction average across a curriculum that varies widely between institutions.

Do data science curricula really differ that much? Yes. A 2021 PeerJ study of US undergraduate programmes found wide variation, strong statistics and computing content but weak ethics and domain coverage, with only about half requiring any ethics coursework — one reason a shared external competence reference is more useful than a satisfaction mean.