Wednesday, April 21, 2010

Weight of evidence

One of the annoying things in bio-medicine is that we often cannot replicate findings in a quick or efficient way. Unlike, for example, many areas of physics, an unusual finding can't be looked at quickly. So what do you do with a marginal finding?

Say for example, p=0.045 and conservative analytic approaches (such as BMA or LASSO regression) exclude this parameter?

If you publish it then you risk it being oversold and creating confusion.

If you don't publish it then you might join a long list of other people who ignore an important association.

Both alternatives seem unsatisfactory in the absence of a "hypothesis generation" tag for papers. I like the clinical pharmacology literature for these papers as the best of the alternatives but wish that it'd be cleaner.

No comments:

Post a Comment