Thanks to Frank Harrell, Timothy J. Wade, Terry Therneau, Brad J.
Bickerstaff and David Parkhurst who all sent me extremely useful
answers. I'm much the wiser about exact and approximated
binomial confidence intervals now and have some references I may
find time and energy to chase. However, the simplest answer for
those with the Hmisc library is its binconf function which was clear
in the "Categorical Data" part of the help on the library.
Turned out the differences were small for my data, not problematical
for my purposes but very clear if you needed high precision.
However, it's lovely to know that definitively now. Interesting to see
that the method I implemented omitted a continuity correction in
some other "normal approximation" methods.
Best wishes all,
Chris
Chris Evans <chris@psyctc.org>
Consultant Psychiatrist in Psychotherapy,
Rampton Hospital; Associate R&D Director,
Tavistock & Portman NHS Trust;
Hon. SL Institute of Psychiatry
*** My views are my own and not representative
of those institutions ***
|