ISSN: 2641-9459
Authors: Emin Mehmet Yusuf*
In a randomized clinical trial (RCT), there are universally accepted rules of thumb for choice of alpha (α: 0.05) and power (1-β: 0.80). These choices require consideration and need to be anything but automatic. Next level of consideration should be given to what we actually chose to compare: the value / effect of the intervention per se or that of the intervention strategy. There is a subtle difference which hinges on which dataset we analyze. Possible data sets are intention to treat (ITT), per protocol or on treatment (PP), as treated (AT). And on the highest level, are we interested in how likely we are to observe the data (that suggests the intervention has some effect) in the event that the null hypothesis (that our intervention has no effect) is, in fact, true or as an entirely different construct, how likely is it that the intervention has some effect given our observed data? These are fundamental questions that we appear to decide on, when we chose alpha (α: 0.05) and power (1-β: 0.80) automatically, without giving them the thought process they deserve.
Keywords: Randomized; Trial; Meaning; Frequentist; Bayesian; Power
