LAUSR

There is something beguiling about a result where p<0.05... Something in our inadequate medical statistics training makes us believe that this is a symbol for the truth. We have to resist. There is a whole branch of statistics that deals with knowing on the basis of things we have known before (Bayesian statistics). We have discussed this in Archimedes in the past—the core of it we can all understand. A high potassium in a child with blocked dialysis tubing is much more likely to be real than a high potassium from a squeezed heel sample. While it is not quite as simple as that in reading research papers, there is a definite value in going into our reading with a sense of what we believe the truth to be—how likely or not we feel the potential therapy is to be beneficial—before we look at the analysis. This warning, like many others, applies particularly to observational studies. Traditional p values are brilliant when it comes to describing probabilities around well conducted studies with minimal risk of bias. They can never overcome the challenges of poor data or unthinking analysis, but our considered understanding and interpretation can certainly help us along that road.