The next seminar of the South Wales local group of the RSS is on Wednesday 16th
November at 5:00pm (4:30pm for tea) at the Cinema, Welsh Assembly Building,
Cathays Park, Cardiff.
Linear regression with errors in both variables; a problem solved
J. W. Gillard, T. C. Iles (Cardiff University)
Abstract
Many experimental procedures involve fitting a straight line to a set of
bivariate data (x,y). Standard linear regression could fit this line, but it
assumes that measurement error is only associated with one variable - x or y. In
many situations, the scatter about the true line will be caused by measurement
error in both variables. Here, standard linear regression estimators for the
line are biased and should be avoided.
This talk will consider how to fit a linear regression when there are errors in
both variables. This is commonly known as errors in variables modelling. An
example using AFP data (used to detect Down's syndrome) will motivate the
talk. A maximum likelihood approach will be briefly outlined, but an alternative
approach using the method of moments will take centre stage. This alternative
approach has a number of distinct advantages, all of which will be outlined
during the talk. Details will be given during the talk of a method of approach
in fitting lines where both variables are subject to measurement error. These
guidelines have been specifically devised for practitioners.
The seminar will be preceded by an AGM.
The December meeting will be on Wednesday 14th at the Conference Room, Local
Government Data Unit-Wales, 8 Column Walk, Cardiff at 5pm (4.30 tea).
The speaker will be Andrew Garrett (Quintiles): "Switching between superiority
and non-inferiority in the clinical trial setting - ensuring a coherent
approach".
All are welcome to both meetings.
Rebecca Cannings-John
Statistician
Cardiff University
Department of General Practice
Room 2F 15
Cardigan House
Wales College of Medicine
Heath Park
Cardiff
CF14 4XN
Tel: 029 2074 5105
www.uwcm.ac.uk/study/medicine/general_practice
|