When It Doesn’t Work

Carousel Learning
8 min readMay 10, 2023


Imagine a student, who we will call S.

As teachers, we don’t just interact with students at one point in time. They might join us at the start of their schooling in general or a course in particular. We then teach them, and they leave sometime later. The student at the start of this journey can be called Si, where i stands for initial — their initial state as they first walk into our classrooms. By the time they leave us, we want them to be changed in some way, and we could call them Sf, where f stands for final.

How exactly Sf differs from Si will vary, but it will certainly involve increases in knowledge, understanding and skills. We also might hope for changes in more general characteristics and dispositions: we like to think that we have an impact on students’ motivation, their feelings towards the subject and a host of other character traits. The movement from Si to Sf is often therefore what we mean by the word “education”.

For most of us, the way this education is achieved is via teaching:

This is a simple model, based on the work of the late Frederick Reif. It’s a model I’ve written about before on a macro level regarding how we conceptualise a subject’s curriculum, which might be our summation and delineation of the gap between Si and Sf. More recently, I’ve thought about it on a more micro level, and looked at it over the course of just a lesson or two.

Imagine I’m teaching students three causes of deforestation: logging for timber, clearing for farmland and flooding from dams. Our model might look a bit like this:

Of course, as a teacher this is my intent. But it doesn’t always work out so cleanly:

We’ve taught a class of humans, not robots. And learning is messy and complicated and unpredictable. There was an Sf that we wanted, but it isn’t necessarily the Sf we got.

The better our teaching, the greater our chances of getting the Sf that we want. We certainly acknowledge reality and accept that we aren’t always going to get what we want, but we at least know what we are aiming for and can put together some best bets for achieving that goal.

Our simple model has already become more complicated, and there are plenty more complications we could add. One complication could be what exactly is meant by the word “know”? If a student “knows” that logging is a cause of deforestation, but they don’t know what “logging” means, then presumably our lesson hasn’t been successful. If a student “knows” what a dam is, and how they cause flooding and deforestation, but does not know why people would build dams, then has our lesson been successful? Maybe yes, maybe no. Maybe we need to modify exactly what our ideal Sf looks like.

Another complication is the fact that despite the multitude of Sf’s we end up with, there has still been teaching. We’ve done our bit, we’ve taught the content, but what that means in terms of the gap between Si and Sf is far from assured.

The purpose of all this is to highlight a simple problem:

Often, despite the fact that we have taught, the students do not come away from the lesson with exactly what we wanted them to learn.

The likelihood of this happening is ever present and is a constant battle. Sometimes they will learn less than what we wanted. Sometimes there will be gaps in their knowledge and understanding. Sometimes they may have picked up a misconception, or continue to harbour one from before. However exactly it plays out, it’s a rare lesson where all the students come out with exactly the knowledge we had intended for them to come out with.

There are elements of the problem above that are out of our control, but there are a great number that are within our control: how we assess prior knowledge, how we explain new content, how we check for understanding and how we have students practise. By manipulating these controllable elements, we increase or decrease the likelihood of the students learning what we wanted them to learn. Better explanations, greater likelihood of learning. Better checks for understanding: greater likelihood of learning. And so and so forth: better teaching means a greater likelihood of learning.

Let’s apply our models and problems to a different context: retrieval practice homework aimed at memorising specific content. We’ll imagine that we want students to remember:

Deforestation is the removal of trees from forest areas
Three causes of deforestation include: logging, farming, dam-building
Logging is chopping down trees to use the wood to make wood products like furniture
Trees need to be chopped down to make space to build farms
Dams can cause areas behind them to be flooded and for trees to die
Dams are built to provide power

To make the above a bit more amenable to retrieval practice and homework, we might rework it into
Core Questions:

What is deforestation?
Chopping down trees from forest areas

Give three causes of deforestation
Logging, farming, dam building

What is logging?
Chopping down trees to make wood products

Give an example of a wood product from logging

How does farming contribute to deforestation?
Land is needed to build farms on

Why are dams built?
For power

How do dams contribute to deforestation?
They flood the land and kill trees

My Si and Sf are now relatively well defined. Here, however, the medium by which the education is achieved is not “teaching” but “homework”:

As before, we won’t get every student as a perfect Sf. We will have some students who have learnt some answers, other students who have learnt others, some who have learnt a couple and some who have learnt most. The picture will be complicated:

Returning to our central problem above, given that we are dealing with homework and not teaching, let’s change from:

Often, despite the fact that we have taught, the students do not come away from the lesson with exactly what we wanted them to learn.


Often, despite the fact that they have done the homework, the students do not come away from it with exactly what we wanted them to learn.

This assumes that they did the homework, a fairly shaky assumption in many cases. Let’s add another problem then:

Often, they do not do the homework, and therefore do not learn any of what we wanted them to learn.

We now have two problems with homework:

The student who does not do the homework remains as Si. Students who have done the homework vary widely in how close they are to our ideal Sf.

We said earlier that when it comes to teaching, there are many variables within our control; how we explain, how we check for understanding and so on. When it comes to homework, many of those variables are removed from our control. We cannot stand over students whilst they are doing it. We cannot ensure they are completing it properly and speaking out or writing down their answers when using flashcards. We cannot be certain that they even used the flashcards before attempting the quiz. We cannot even guarantee they haven’t cheated from a friend or a search engine.

In reality, this means a student could have completed the homework and had no corresponding change in their knowledge. This would leave them as Si, a case where a student remains unchanged by your intervention. This might be fairly unlikely in a classroom, but when it comes to homework becomes a serious possibility:

We now have three problems:

Often, despite the fact that we have set the homework, they do not do the homework, and therefore do not learn any of what we wanted them to learn.
Often, despite the fact that they have done the homework, what they learn is not everything that we wanted them to learn.
Often, despite the fact that they have done the homework, they have learnt nothing.

One could be left pessimistic in the face of all this. Why bother with homework if it can lead to such varied undesirable outcomes?

I wouldn’t be writing this blog if I thought these issues were completely unsolvable. Whilst homework will always be harder to get right than classwork, there are ways to improve your chances, and we’ve spent a long time at Carousel trying to figure out the best way to go about issuing and administering homework. A large part of that is using the classroom (i.e., where you have control) to prepare students for homeworking (i.e., where you have no control). A couple of good places to start might be:

The Four Planks of an Effective Homework Policy: this article gives an overview and a general framework for thinking about homework implementation and the ingredients that will increase your chances of achieving Sf.
Retrieving Better: this document deals with a huge number of concepts surrounding retrieval practice, homework and classroom implementation.

Effective ways to use your classroom to leverage homework:

● Via the Do Now
● Via Whole Class Feedback
● Via Whiteboard quizzes
● Via in-class practice and modelling

It isn’t easy to get homework right. In fact, it’s probably harder to get homework right than to get classwork and teaching right. Despite that, there are tried and tested routes that will help you out and it’s a goal worth fighting for.

To enjoy a FREE 30-day trial of Carousel Gold or to purchase a Carousel Gold subscription (just £60 p.a.), log in to your account here and go to the MY ACCOUNT section.

To find out about prices for Carousel Platinum, click here.

To learn more about Carousel’s Gold and Platinum tiers, click here.



Carousel Learning

Carousel is a retrieval practice and online quizzing tool that helps students to embed knowledge in their long-term memory.