 |
 |
Dear Dr. White, An odds ratio (OR) is inherently asymmetric. It's logarithm is symmetric around zero under the null hypothesis that the OR is one. On the original scale, there is a symmetry around 1.0 with the interval from 0 to 1 being symmetric to the interval from 1 to infinity. That is, the same amount of probability is packed into the interval from 0 to 1 as is in the interval from 1 to infinity. For example, the point 1/2 is symmetric with 2, 1/4 is symmetric with 4, etc. Consequently, a confidence interval is usually computed on the log scale using the approximate standard error on that scale and then transformed back to the original scale to get an asymmetric confidence interval. A symmetric confidence interval (ie, the OR+/-zSE) on the original scale of the odds ratio would appear to include both sides evenly but would include different probabilities on either side of the observed OR. In principle, the asymmetric confidence interval evens out the probabilities on each side of the estimate, so this is the one usually used. One consequence of this common practice is that it may be difficult to find the se of the OR in most references. Cordially, David David W. Smith, Ph.D., M.P.H. Associate Professor Biostatistics and Epidemiology College of Public Health - CHB 323 University of Oklahoma Health Sciences Center PO Box 26901 Oklahoma City, OK 73190 Phone: (405) 271-2229, ext 48062 FAX: (405) 271-2068 [log in to unmask] > -----Original Message----- > From: Dr Allan White [SMTP:[log in to unmask]] > Sent: Tuesday, January 26, 1999 4:27 AM > To: [log in to unmask] > Cc: [log in to unmask] > Subject: QUERY: Standard Error of an Odds Ratio > > Members, > > Could someone please advise me how I calculate the standard error of an > odds > ratio (ad/bc) ? > > Thanks. > > > Allan %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%