Instructions on sampling are clearly set out in our inspection guidance. In primary schools, for example, observations must cover the subjects and areas of learning of the curriculum, all year groups and all teachers. Inspectors must spend at least 60 per cent of their time in school gathering first-hand evidence of teaching and learning.

Judgments about lessons are almost always based on at least 30 minutes of observation. Those on the quality of teaching across all schools rest normally on samples ranging from 50 to 200 lessons.

Our study of inspectors’ judgments was guided by researchers at the University of London Institute of Education and Dutch statisticians, subject to peer review, and published in a respected international journal. It found that there was exact agreement between pairs of inspectors in about two-thirds of lessons, and close - that is, acceptable - agreement (to within one grade on a seven-point scale) in one third. They were less reliable in only 3 per cent of lessons. We have begun a programme of further training designed to improve reliability still further.

As for value-added measures, all that were available to a few schools in 1983 were related to sixth-form subject performance. There are at present no reliable and comprehensive value-added measures through the key stages of the national curriculum, and researchers disagree about the validity of different approaches.

Other systems have their limitations as well as their uses. As useful and acceptable indicators become available, the inspection procedures incorporate them. Our PICSI (Pre-Inspection Context and School Indicator) reports are an example of this.

OFSTED has published, since 1994, a series of validated studies of inspection. So have other researchers, but Professor FitzGibbon is not thought to among them.

Peter Matthews

Office for Standards in Education

Head of Inspection Quality, Monitoring Development

London WC2