Earlier this year, I wrote about the large MetaSENse meta-analysis on interventions for students with special educational needs and disabilities (SEND). The conclusions were encouraging. Targeted interventions can make a substantial difference, averaging around five months of additional learning. At the same time, the review also showed that many popular interventions have not yet been rigorously evaluated.

During the summer, the researchers published a follow-up study. This time, they asked a different question. Rather than examining which interventions work, they looked at why they work. Which ingredients of successful interventions seem to matter most?

That may sound like a subtle distinction, but in such a diverse field it is arguably the more interesting question.

More is not necessarily better

The researchers examined more than 500 interventions and coded whether each included any of 18 instructional components, such as explicit instruction, feedback, scaffolding, metacognitive strategies, technology, collaboration and rewards.

Their first finding was striking. Interventions contained, on average, almost seven different components. Yet interventions with more components were not more effective. There was no relationship between the number of strategies used and the eventual effect size.

That fits nicely with something educational research has suggested for quite some time: teaching does not automatically improve simply by adding more ingredients.

Explicit instruction remains a winner

Did any components consistently stand out? Yes. Explicit instruction.

Across general outcomes in writing and mathematics, explicit instruction was consistently associated with better results. The machine-learning analyses, which enabled the researchers to examine complex interactions among components, also identified explicit instruction as one of the strongest predictors of success.

That will probably not surprise anyone who has followed the evidence on effective teaching over the past decade.

Feedback matters

Feedback also emerged as a particularly powerful ingredient. Especially in mathematics interventions, feedback consistently predicted better outcomes. Across the overall analyses, it remained one of the strongest instructional components.

Again, this is hardly a revolutionary finding. Rather, it reinforces what research has repeatedly shown: high-quality feedback is just as important for students with SEND as it is for other learners.

Not everything that sounds logical turns out to be effective

Up to this point, the findings may seem fairly predictable. However, some components that are currently very popular performed less well. For example, interventions that relied heavily on rewards tended to produce smaller effects across several outcome domains.

Even more surprising was the finding on adaptive teaching. On average, interventions coded as using adaptive teaching were associated with smaller effects.

Before drawing sweeping conclusions, however, the authors immediately add an important caveat. Their definition of adaptive teaching was much narrower than the way the concept is usually understood today. They mainly referred to continuously adjusting task difficulty during an intervention, often through digital systems. That is very different from the broader understanding of adaptive teaching in current educational policy and practice. That distinction seems crucial.

Technology is not a magic bullet

Another interesting finding was that technology, by itself, neither improved nor worsened outcomes.

This certainly does not mean that technology never works. It simply suggests that using a tablet, an app or a digital programme is no guarantee of better learning outcomes.

That conclusion also aligns well with previous research.

The combination matters too

One particularly interesting aspect of this study is that the researchers complemented traditional statistical analyses with machine learning. This allowed them to examine how instructional components interact with one another, something that I often find missing in research on approaches such as Universal Design for Learning (UDL).

And the results were revealing. Sometimes combining two strategies proved less effective than using either one on its own. Interventions that combined feedback with goal setting, for example, performed less well than interventions that focused primarily on one of these components. Retrieval practice also appeared less effective when combined with explicit goal setting.

This reminds us once again that teaching is not simply the sum of individual strategies.

I must admit that I cannot yet think of a convincing explanation for these findings. Interestingly, neither can the authors. They conclude that instructional components do not necessarily combine additively and suggest that well-implemented, simpler interventions may sometimes outperform more complex designs.

The overall message remains the same

Ultimately, the main conclusion has not really changed.

Successful support for students with SEND appears to rely largely on the same principles that underpin effective teaching for all students: clear explicit instruction, high-quality feedback and thoughtful implementation.

At the same time, the authors caution against treating instructional components in isolation. What works depends not only on the strategy itself, but also on the learning goal, the learner and the way the intervention is implemented.

In other words, teaching students with SEND looks remarkably similar to simply teaching well.