Situation: The Hospital Administrator came to me recently and asked "how
many patients out of a group of sixty did he need to contact to ensure enough
were notified of a possible discrepancy in their medical records(?)".

Since this was non-numeric data, and most models are built around numeric
data, I was stumped. Google took me to www.surveysystem.com and a handy tool
for calculating such things. I was able to determine a sample size, but got
hung up on estimating the confidence interval. As I understand it (from
looking at textbooks and stretching my memory back to graduate school in
1985), the lower the confidence interval, the higher the level of confidence.
So, using the tool, if I set the confidence interval at 4, we would have to
notify most of the patients in order to be 95% confident of our "sample size".

Reality Check: In this case, in view of rhe possible repercussions, we
should notify ALL of the patients when something is suspected to be amiss in
their records. (The question as posed to me was based on an incorrect
assumption, in other words). And contacting sixty patients, while
time-consuming, is achieveable, too. So, perspective is important in
attempting to answer this question.

Comments?

DOUGLAS ECKERT