Car Talk Puzzler
There's a discussion on the National Association of Science Writers (NASW) listserv about the use of statistics for predicting events. This reminded me of the latest Car Talk puzzler I heard last Saturday. If I heard it correctly, the question is, given that a clinical test is 99% specific and sensitive, what is the probability that a positive result will indicate an actual case of disease (positive predictive value)?
I couldn't think of the equation into which I could plug in these values and get the answer, so I drew up a roman square and eventually derived an equation that would do the job. Below is a gif showing what I came up with.
The answer is 9%. In reality, I haven't seen a clinical test with a specificity as low as 99%. You'd expect to see at least 99.9% or 99.99%. You can see how this generates a large number of false positives given the low prevalence of 0.001.
For the Palm OS platform, I think MedMath 2.0 will do this calculation; I'll have to check.
9:38:31 AM
|