Summary: Statistical ANOVA makes a number of credence-stretching foundation assumptions. However, in reality much of the data collected in various research studies contains non-normality, heterogeneity, strong correlation between groups, large error variance, unknown parameter interactions, etc. The theoretically well-established technique of Ranking and Selection Methodology (RSM) can be applied successfully even when many of these problems are present. Rather than having the ANOVA goal of statistically “proving” that some group attribute is different from the same attribute of several other groups, RSM assumes at the outset that they are different and has the goal of ranking them per their differences.
This paper explains and compares RSM with \(F\)-test and multiple-comparison procedures, and examines its applicability to design-oriented research subjects in order to improve end results. Our emphasis is on practical aspects, with a view towards increasing applications of RSM, and with the statistically oriented practitioner in mind.
For the entire collection see [Zbl 0922.00040].