In 1998, Paul Black and I published a pamphlet entitled Inside the Black Box. In it we presented the findings of a review of the research on whether assessment could actually support learning, rather than merely measure it.

We concluded that the use of assessment minute-by-minute and day-by-day during lessons generated some of the biggest improvements in learning that have ever been reported. Over the next two years, working with groups of teachers in Oxfordshire and Medway, we found that these effects generalised to real classrooms, even when achievement was measured in terms of scores on national curriculum tests and GCSE examinations.

Using assessment for learning increased student engagement and doubled the speed of learning. Students learned in six months what they would have taken a year to learn in other classrooms. And the costs of these improvements were modest - about pound;2,000 per classroom per year.

Achieving the same increase in student achievement by reducing class size would cost between five and 20 times as much, depending on the subject and whether we could find new teachers as good as those already in post.

Given such value for money, it is not surprising that governments in Scotland, England and elsewhere have adopted some version of assessment for learning as a key policy. But improved student achievement is unlikely to be realised unless one basic truth is acknowledged: that changing one’s teaching is very hard. For most teachers, changing their practice in this way will be the hardest thing they have ever done in their careers.

To understand why, it is necessary to understand the nature of teacher expertise and where it comes from. Expertise in teaching is surprisingly similar to expertise in other domains, such as medical practice, chess, or cricket. People become experts by repeating practices over and over again until they become almost automatic. The fluency that comes from such repetition allows teachers to get through the day by integrating routines into a seamless sequence.

This fluency is the experienced teacher’s greatest asset, but also the major impediment to getting even better. This is true in all domains, but there is an additional factor in teaching. Teachers learn most of what they know about teaching by the time they are 18 years old. In the same way that we learned most of what we know about parenting by being parented, most of what we know about teaching we internalised as students in the classroom.

This is why one cannot tell teachers what to do.

I say this not from a misguided sense of wanting to be nice to teachers, but because telling teachers what to do does not change their practice. For example, almost every teacher knows the research on “wait time”. They know that a teacher should wait at least three seconds at the end of a question, and at the end of a student’s answer, before saying or doing anything. And yet hardly any teacher does. The hard thing is not knowing what to do, but how to go about doing it.

For the past three years, my colleagues and I in the Learning and Teaching Research Center at the Educational Testing Service, in Princeton, New Jersey - the world’s largest not-for-profit educational research organisation - have been investigating how we might support teachers in changing their practice. As a result of exploring a range of alternatives, we have concluded that the best way to realise the potential of assessment for learning is through school-based teacher learning communities (TLCs).

We do not believe that these communities are a good thing in themselves. If you wanted to increase teachers’ subject knowledge, these communities would be a very bad way to go about it. Some form of direct teaching will be far more effective. And just putting teachers in a room together to chat may be enjoyable for them, but it is unlikely to change their practice. TLCs are therefore the means, not the end. Our experience is that TLCs are most effective when:

* teachers meet monthly for at least two hours;

* all participants have to report on what they have tried - and before the end of the meeting make a written commitment about what they will try next;

* teachers are accountable for changing something but are free to choose how much to change and how quickly; and

* teachers can customise assessment for learning to suit their personal style.

Improving the use of assessment for learning is the most powerful way to increase achievement, and TLCs appear to be the best way to support teachers in making the changes. Real sustained increases in student achievement are within our grasp as long as we do not get sidetracked by quick fixes. We need to get down to the hard part: investing in sustained, school-based professional development that is focused on assessment for learning.

Dylan Wiliam is director of the Learning and Teaching Research Center at the Educational Testing Service, in Princeton, New Jersey, United States.

He will take over as director of London university’s Institute of Education next month.See: www.uk.etseurope.org