Monday, February 25, 2008

The buzz

Every once in a while I look through the keyword referrals and do an informal assessment of how people find this blog -- one that's geared for a rather narrow audience. Here are the most popular keywords:

  • O'Brien-Fleming (especially doing this kind of design in SAS)
  • Bayesian statistics in R
  • noninferiority
  • NNT (number needed to treat)
  • confidence intervals in SAS
On occasion I will get hits from more clinical or scientific searches, such as TGN1412 (the monoclonal antibody that caused a cytokine storm in healthy volunteers, leading to gangrene, multiple organ failure, and for those lucky/unlucky enough to survive, cancer), Avandia, or CETP inhibitors.

At 2000 hits in a year, this is clearly a narrowly-targeted blog. :D

Tuesday, February 5, 2008

When you consult, be sure you know what your labels mean

In a recent project, I fit a varying slopes and intercepts model to data that was organized into replicated experiments, with several replications at each timepoint for several timepoints. Turns out, the only relationship between units labeled "Run 1" was the label. Nothing else. In this case, it turned out a simple linear regression was the more appropriate model.