SELECTING THE SAMPLE SIZE

A typical sample size consultation often resembles a ritualistic dance. The investigator usually knows how many participants can be recruited and wants the statistician to justify this sample size by calculating the difference that is "detectable" for a given number of participants rather than the reverse. The statistician typically will not use less than 80% power in the calculations, mainly because, by convention, anything lower than that will be questioned. Because no guidelines exist about which "clinically important difference" should be used in the calculation, often anything can pass muster. The "detectable difference" that is calculated is typically larger than most investigators would consider important or even likely. But this "detectable difference" is usually not commented on by reviewing bodies and does not have to be quoted when the results of the study are analyzed, so researchers will often accept large differences for the purpose of getting a proposal approved. The result of this practice is that most clinical experiments are too small, and the journals are filled with a plethora of reports of clinically important but statistically nonsignificant effects, keeping persons who do meta-analysis in business [42].