Almost every grant application contains a version of the same claim: that a blended programme, combining a for example a podcast with a monograph, will outperform any one of those delivered alone. It is stated as though it were settled. It is rarely referenced, because it feels as though it does not need to be.
Blended costs more. Sometimes two or three times more, once you account for faculty time across multiple recordings, multiple production workflows, multiple accreditation submissions and multiple dissemination campaigns. If that premium is buying something, it is money well spent. If it is not, then a substantial share of everything the field spends is going on packaging.
So we went looking for the evidence. It is not what we expected.
I.What happens to the advantage under scrutiny
Blended learning does outperform single-format learning on knowledge outcomes, and at first glance the effect looks large. The most cited recent synthesis in medical education pooled 56 studies covering 9,943 participants and reported a standardised mean difference of 1.07. By conventional standards that is very large indeed. It is the number that ends up in grant applications.
Two things happen when you read further. Heterogeneity was 94.3% — the studies were measuring things so different from one another that pooling them is close to meaningless. And when the authors corrected for publication bias, the effect fell to 0.41.
The same pattern, from a different direction, appears in the US Department of Education's meta-analysis of online and blended learning. Its headline effect is a modest +0.20. Its subgroup analysis is the interesting part.
We have not been comparing formats. We have been comparing doses — and then crediting the format.
The Department of Education authors were unusually direct about it. Interpretations of the advantage, they wrote, should take into account that the conditions generally differ on multiple dimensions, including the amount of time that learners spend on task. The advantage may simply be the product of learners spending more time learning.
II.The question asked directly
Underneath all of this sits a question that is not really about education at all: do multifaceted interventions change professional behaviour more than single-component ones? That has been asked head-on.
An overview of systematic reviews in Implementation Science examined 25 moderate-to-high quality reviews and concluded that there is no compelling evidence that multifaceted interventions are more effective than single-component interventions in changing health professionals' behaviour.
- Of the reviews that ran proper statistical analyses, none found a relationship between the number of components and the size of the effect
- Of the direct comparisons, only about half favoured the multifaceted approach
- Where the two differed at all, single-component interventions were frequently comparable or better
The authors were careful about their own limits — only three reviews used robust statistical methods, vote-counting is a weak instrument, and there is no agreed way to classify what counts as a "component". Absence of evidence is not evidence of absence, and we should not overclaim in the other direction.
But this was the exact question, asked directly, across 25 reviews. The answer came back: we cannot show that more is better.
III.The number that reframes everything
Then there is the Cochrane review of continuing education meetings and workshops — 215 studies, more than 28,000 health professionals. The largest synthesis the field has.
So the format question that consumes most of our design energy and most of the budget has not actually been answered. And the ceiling on the entire category, measured where it was supposed to matter, is close to the floor.
IV.What other industries learned, and when
Learning science settled the media question in 1983
Richard Clark's argument was blunt: media are mere vehicles that deliver instruction but do not influence achievement, any more than the lorry delivering groceries changes your nutrition. What looks like a media effect is almost always a method effect, or a content effect, or a time effect, bundled in alongside the medium. His practical conclusion was that media comparison studies are structurally incapable of answering the question they pose, because the two arms never differ only in medium.
That was forty years ago. The field is still running media comparison studies, and still reporting them as format findings.
What does have large, replicated effects is not a format
Two mechanisms in learning science have survived a century of replication: spacing — the same total study time distributed across sessions rather than massed — and retrieval practice — being made to produce an answer rather than review one. The distributed-practice literature runs to hundreds of studies and the effect holds across materials, ages and retention intervals.
Neither of these is a format. Both are schedules. You can have them with one channel or with five.
Aviation and surgery reached for the job aid, not more training
The most-cited safety intervention in modern medicine is not an educational programme. It is a checklist. And the honest version of that story has two halves.
A 19-item surgical safety checklist. Death and inpatient complications each fell by more than a third. Nobody had been taught new surgery. The intervention was a piece of card, present at the moment of action. Haynes et al., NEJM, 2009.
Ontario required the checklist across its hospitals and measured the result: no significant reduction in mortality or complications. The tool had become a form signed on the way past. Urbach et al., NEJM, 2014.
That second finding is the most useful one here, and it cuts against our own interests. The checklist was not magic. It worked when it genuinely changed what happened in the room, and it did nothing when it became a compliance exercise. Anyone arguing — as we do — that tools beat content has to accept that discipline. A tool that becomes paperwork is just content with a checkbox.
V.So what is the active ingredient?
Putting it together, the plausible answer is that blended programmes work slightly better because they incidentally deliver three things, none of which is the blend itself.
VI.What this means if you fund education
A few questions that follow from the evidence rather than from custom.
- When you pay a premium for a blend, which of the three mechanisms are you receiving — and is the blend the most efficient way to obtain it?
- Would a single format delivered three times, spaced, with retrieval built in, outperform a four-channel programme delivered once? On the evidence above, there is no good reason to assume it would not.
- Does the programme put anything in front of a clinician at the moment of a decision, or does all of it arrive in advance and hope to be remembered?
- Is "we delivered across four formats" being reported as though it were an outcome? It is a description of spend.
VII.The honest counter-case
We would be doing exactly what we are criticising if we presented this one-sided.
- Reach is a legitimate reason for a blend
- An enduring asset reaches people who cannot attend live. That argument does not depend on the blend improving learning at all — it should simply be made honestly, as an access argument rather than an impact one.
- Tailoring may be the variable, not multiplicity
- Some multifaceted interventions are designed against identified barriers; others are simply several things at once. The reviews often cannot distinguish these, and it is plausible that tailored-multifaceted works while generic-multifaceted does not.
- The instruments are blunt
- Behaviour change is genuinely hard to capture, and a tenth of a percentage point at the patient level may partly reflect that we are measuring badly rather than that nothing happened.
- And absence of evidence is not evidence of absence
- The honest statement is not "blending does not work". It is: after a great deal of looking, we cannot demonstrate that the additional cost and complexity produce additional benefit.
VIII.What would change our mind
A clean test would not be difficult to run, and as far as we can find, nobody has run it.
Hold total exposure time constant. Hold content constant. Hold the spacing schedule constant. Vary only the number of formats. Measure behaviour, not recall.
If the four-format arm beats the one-format arm under those conditions, the premium is justified and we will say so. Until someone runs it, we should keep asking whether blending is worth the extra cost.
IX.The uncomfortable summary
We can demonstrate that blended programmes produce better test scores. We cannot demonstrate that they produce better patients. The largest synthesis we have, across 215 studies and 28,000 professionals, found a tenth of a percentage point at the patient level — and could not tell us whether the format mattered at all.
That is not an argument for doing less. It is an argument for being far more specific about what we are creating, and for putting the energies into the mechanisms that have evidence behind them: spacing, retrieval, and above all proximity to the decision.
The blend was never the active ingredient. It was the delivery vehicle — and, as Clark pointed out four decades ago, the lorry does not change the groceries.