Dear Ted,
You, Stephen Senn and I are mathematicians by background and natural
inclination. The majority of people in the EBH field are, hopefully,
clinicians or clinical epidemiologists. I believe they can
understand a difference between two proportions, and a CI around
this. But I'm sure they've little idea about singularities,
Fieller's theorem etc. As a profession we should be encouraging them
towards methods that are as transparent as possible as well as
performing well. A CI for a difference between two proportions, say
-5% to +25%, suggests that within the limits imposed by the limited
sample size, we are prepared to accept that the true or population
difference could be as large as 25%, in the direction of A being
better than B; there might be no difference; or conversely A could
be 5% poorer than B. Values for the difference outside this range
don't square well with the observed data. While there isn't a sudden
transition from "credible" to "incredible", this provides some sort
of yardstick that there is wide agreement to use. While the level
of statistical understanding required in order to grasp this sort of
presentation isn't trivial from a clinician's viewpoint, it's
attainable and probably the appropriate level.
Best wishes.
Robert.
..........................................
Robert G. Newcombe, PhD, CStat, Hon MFPHM
Senior Lecturer in Medical Statistics
University of Wales College of Medicine
Heath Park
Cardiff CF14 4XN, UK.
Phone 01222 742329 or 742311
Fax 01222 743664
Email [log in to unmask]
Macros for good methods for confidence intervals
for proportions and their differences available at
http://www.uwcm.ac.uk/uwcm/ms/Robert.html
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|