Skip to content
Go back

Online Certifications: The 3% Who Finish and the Recruiter Experiment

Illustration of a digital certificate and certification badge

Online certifications have two documented problems, and they are different in kind. First: almost nobody finishes them. Second, and more uncomfortable: even if you do finish, the effect on recruiter callbacks depends almost entirely on who issued it and which sector, not on the fact of having it.

Both problems have been measured. Worth looking at the numbers before you enrol in anything.

How many reach the end: the data on 12 million people

Reich and Ruipérez-Valiente (Science, 2019) analysed six years of HarvardX and MITx data on edX: roughly 12.67 million participants and 5.63 million course registrations.

The main findings:

Hansen and Reich (Science, 2015) had already documented the access point: enrollees came disproportionately from areas with higher income and education. The platform was open to everyone; the user base wasn’t.

The 3% figure doesn’t say online courses don’t work. It says enrolling costs nothing and therefore selects for nothing. The useful signal isn’t “I took a course”, it is “I completed something that had a cost of quitting”.

The recruiter experiment

The most important piece is by Deming, Yuchtman, Abulafi, Goldin and Katz (American Economic Review, 2016). They ran an experimental study with fictitious résumés sent to real job postings in the United States, randomly varying the type of institution that had issued the credential.

The result: résumés with a degree earned online at for-profit institutions received fewer responses than those with equivalent credentials from public or non-profit institutions. In some sectors, the for-profit credential produced no advantage over having no credential at all.

That is the result that closes the “is it worth it?” debate. There is no general answer for certifications: there is an answer per issuer.

Why they work extremely well in some sectors

A parallel literature gives the complementary picture, and it points the other way.

Xu and Trimble (Educational Evaluation and Policy Analysis, 2016) analysed administrative data from two US states on sub-baccalaureate certificates. The result is sharply heterogeneous: certificates in health and technical fields produce significant wage returns; those in general fields produce returns close to zero.

Jepsen, Troske and Coomes (Journal of Labor Economics, 2014) found the same pattern in Kentucky: returns to short certificates are real but much lower than to two-year degrees, and concentrated in specific fields.

Belfield and Bailey (2017), reviewing the literature, summarise it this way: variation by field is larger than variation by credential type.

If the certification…Documented effect
Is issued by a for-profit online providerLower recruiter callbacks (Deming 2016)
Is in health or a technical fieldSignificant wage returns (Xu & Trimble 2016)
Is in a general fieldReturns close to zero
Is legally required to practiseValue is categorical, not marginal

The theory that makes sense of all this

The model that makes these data readable is fifty years old. Spence (Quarterly Journal of Economics, 1973) formalised signalling theory: a credential works in the labour market not only for what it teaches, but because it is costly to obtain for those who lack the trait the employer wants.

The consequence is brutal, and it explains the 3% completion rate:

A certification anyone can get with little effort transmits no information. Not because it is low quality, but because it doesn’t separate candidates. Signal value is inversely proportional to ease.

This also explains why certifications with a proctored exam and a real failure rate (the major cloud vendors’ certifications, accounting credentials, information-security ones) retain value, while certificates of attendance never had any.

Micro-credentials, and what the official documents say

On the micro-credential strand, the OECD (2021) and the Council of the European Union (2022 Recommendation) produced documents worth citing for what they actually say. They establish transparency standards: what a micro-credential must contain, how workload must be described, how learning is verified.

They contain no evidence that micro-credentials increase employability. They are governance documents, not impact evaluations. It is a distinction much promotional material omits.

Four tests before paying for a course

1. Apply the cost-of-quitting test. Before enrolling, ask: can anyone get this certificate without passing anything? If so, signal value is zero by definition. That isn’t a judgement on content quality.

2. Look for the issuer in job ads, not in marketing. Take twenty postings for the role you want and count how often the certification’s name appears. It is the only market research you need, and it takes twenty minutes.

3. Distinguish three categories. Licensing certifications (without which you can’t practise) have categorical value. Vendor certifications have value where the vendor dominates. Generalist ones have, per the data, value close to zero in the market, though they may have value to you if the content is useful.

4. Put the thing you did on your CV, not the course you took. The course is an input, and recruiters treat it as one. How to turn it into verifiable output is in how to write your CV.

5. If the field is programming, check the path first. It is the area where the ratio between certifications sold and actual value is most skewed: we cover it in how to learn to code.

6. Check the skill is actually in demand. Before the credential comes the demand: surveys of the most in-demand skills are a better starting point than a course catalogue.

My take: the problem is incentives, not quality

I think the online certification market has an incentive problem more than a quality problem: whoever sells the course earns at enrolment, not at employment outcome. A 3% completion rate isn’t a failure of the business model. For the seller it is simply irrelevant.

My operating rule, unmeasured, is: if the certification has no exam you can fail, I treat it as a book. It may be an excellent book. But I don’t put it on a CV as though it were a qualification. That is an inference from signalling theory, not an experimental result.

Frequently asked questions

Are Google, AWS or Microsoft certifications worth it? The ones with proctored exams and real failure rates retain signal value, especially where the vendor dominates the sector. No independent experimental studies exist on individual certifications: the criterion to use is their presence in real job ads.

Why is the completion rate so low? Because enrolling costs nothing. The edX studies show completion rises when there is a financial commitment or an external deadline, exactly as the procrastination literature predicts.

A certification or a project? Depends on the sector. Where a recognised hard exam exists, the certification is faster for a hiring manager to assess. Where it doesn’t, a verifiable project is the stronger signal.

Are European micro-credentials recognised? There is an EU-level transparency framework defining what they must contain. Recognition doesn’t automatically mean labour-market value: those are two separate questions.

Is it worth paying for a certificate of attendance? By signalling logic, no: a certificate obtained by showing up distinguishes nothing. If you need the content, you need it anyway, and it is usually available without paying for the certificate.

Sources

Founder of UpSkillLab, with a Master's in Digital Marketing from the University of Florence. Every article starts from public sources cited in the text and is drafted with the help of AI tools, under my review and editorial responsibility: no purchased content, nothing published without human checking. Read the editorial method or report an error.


Share this post:

See all articles in Career & Work

Previous Post
Will AI Replace Your Job? The 47%, the 9%, and What They Really Measure
Next Post
Learning to Code: The "Geek Gene" Its Own Author Retracted