View article

Figure 1 Conclusions for increasing values of the observed difference (sample size and variance being equal). For given sample size and variance, large differences can reach significance; small differences can (eventually) allow rejection of the hypothesis that treatments differ by the smallest relevant difference. In-between (eventually) may be a region of indifference (no conclusion possible). This process, mentioned by Feinstein (1975) as calculation of'PB'can be applied tocomparisons of two proportions with a methoddescribed by Dunnet & Gent (1977) or to a comparison oftwo means, as in a trial comparing two antidepressants by Coudray, Dufour, Garello, Mollo, Pascal, Poisson, Scotto, Simart, Sormani & Tourane