[Maths-Education] From Dylan wiliam

[Gates Peter]

The crucial thing to point out to these purveyors of dubious practices
is
that if they are going to use measurements, they have to understand
measurement error. At key stage 2, the standard deviation of student
achievement is around 0.67 levels, and given that the reliability of the
tests that teachers use is around 0.7 (at best) then the standard error
of
measurement is around 0.37 levels. So when we say that a student is at
level
5.03, a 95% confidence interval for this would be three quarters of a
level
either side of this. In other words, even if the measurement were
perfectly
valid, the measurement error would mean that only the first figure is
significant.

Dylan Wiliam

Until 19 August:
Director, Learning and Teaching Research Center
ETS
Rosedale Road (ms 04-R)
Princeton, NJ 08541

Tel (609) 734 1307
Fax (609) 734 1755

After 19 August:
Deputy director, Institute of Education
20 Bedford Way
London WC1H 0AL
United Kingdom

Tel 020 7612 6000


http://web.mac.com/dylanwiliam/iWeb