About us    Campaigns    Research    Support us    Publications    Media Room    Join Us    Contact us

Do teachers matter? Measuring the variation in teacher effectiveness in England

Quality, Teacher performance


Helen Slater, HM Treasury

Neil Davies,  Department of Social Medicine, University of Bristol

Simon Burgess, CMPO, University of Bristol

January 2009

Working Paper No. 09/212

Centre for Market and Public Organisation
Bristol Institute of Public Affairs
University of Bristol
2 Priory Road
Bristol BS8 1TX


It seems common sense that teachers matter, and that pupils will achieve more with an inspirational teacher than with an average or poor teacher. Anecdotes abound of the transformational effect of excellent teaching. Yet trying to quantify this is difficult, principally because of the data requirements. To a degree, social science research has emphasised family and home rather than teachers and school in the production of human capital. Disentangling the separate contributions of schools, teachers, classes, peers and pupils themselves needs extremely rich and full disaggregate data. Whilst a small number of papers have been able to make progress here, we do not yet have a settled view on the importance of teachers.
Using a unique primary dataset for the UK, we estimate the effect of individual teachers on student outcomes, and the variability in teacher quality. We show that teachers matter a great deal: being taught by a high quality (75th percentile) rather than low quality (25th percentile) teacher adds 0.425 of a GCSE point per subject to a given student, or 25% of the standard deviation of GCSE points. This shows the strong potential for improving educational standards by improving average teacher quality. However, implementing such a policy would not be straightforward, as we also corroborate recent US findings that good teachers are difficult to identify ex ante.
As Rockoff (2004) notes, most of the issues in this field relate to data quality. We use a unique primary dataset that matches a short panel of pupils to a short panel of teachers. We link over 7000 pupils, their exam results and prior attainment to the individual teachers who taught them, in each of their compulsory subjects in the crucial high-stakes exams at age 16. These exams provide access to higher education and are highly valued in the job market.
Our dataset complements and in some ways extends the current leading datasets in this field used by Aaronson, Barrow and Sander (2007) (ABS), Kane, Rockoff and Staiger (2007) (KRS), Rivkin, Hanushek and Kain (2005) RHK and Rockoff (2004) (R). Like ABS and R, but unlike RHK and KRS, we can match a student to her/his actual teacher, rather than to the school-grade average teacher. Unlike ABS, KRS, RHK and R, our context is one of students taking exams that are very important to them and to the school. Unlike ABS, KRS, RHK and R, we exploit the fact that we observe students taking three exams at the same date, allowing us to use a point-in-time student fixed effect, in addition to subject-specific prior attainment. We believe that this allows us to control well for variations in student ability that might otherwise corrupt our measures of teacher effectiveness if students are not randomly assigned to teachers (see Rothstein, 2008). Finally, and also unlike ABS, KRS and RHK, our student-teacher data are matched in and by the school, thus ensuring a high-quality match. Nevertheless, while our data have these advantages relative to existing datasets, there are other issues with our data, and we detail below these short-comings and what we can and cannot estimate.
We show that the standard deviation of teacher effectiveness is 32.6% of a GCSE point, or 18.9% of a standard deviation (1.722 GCSE points), from Table 5 column 1. The lowest bound estimate we have is 28.8% of a GCSE point or 16.7% of the standard deviation. These estimates are in line with those found in the US, which tend to be around a 10% impact on test scores of a unit standard deviation change in teacher quality. Using another metric, teacher effectiveness is about a quarter as variable as pupil effectiveness. However, a teacher’s effectiveness influences the GCSE outcomes of the entire class, and so the teacher’s effectiveness has greater leverage.
The next section reviews the current datasets used and highlights the advantages and disadvantages of ours; we also summarise the results from these studies. Section 3 discusses our own dataset, and section 4 the econometric approach. Section 5 presents the results. In the Conclusion, we discuss the implications of these results for policy on teacher effectiveness, teacher selection, and for the incentivisation of teachers.

To read more: http://www.bristol.ac.uk/cmpo/publications/papers/2009/wp212.pdf

Comments Off on Do teachers matter? Measuring the variation in teacher effectiveness in England

  Disclaimer: The copyright of the contents of this blog remains with the original author / publisher.