Using the similarity factor f2 in practice: A critical revision and suggestions for its standard error estimation

Abstract The purpose of this research was to develop new procedures with the aim of improving the usage of the similarity factor f 2 in dissolution data analysis, and to evaluate them jointly with preexisting ones. We introduce bias-correction and standard error estimation procedures based on the delta, the jackknife and the bootstrap methods. These methods, jointly with the rule of declaring similarity when f 2 exceeds 50 and some alternative testing procedures based on bootstrap confidence intervals, are evaluated on experimental data and studied by simulation. The results indicate that no method is strictly the best, but the following conclusions seem to appear: for estimation purposes the most reliable approach is to use the plain sample f 2 instead of any bias-corrected alternative, any of the standard error estimates may be used in practice and, most importantly, there are evidences against the validity of the procedure declaring similarity if the sample f 2 exceeds 50; a decision rule based on a confidence interval seems to be more adequate. In any case the question should be further investigated.