Level: Intermediate
James Wisnowski, Principal Consultant, Adsurgo
Andrew Karl, Senior Consultant, Adsurgo
Standard hypothesis tests are set up to prove differences from a standard or between populations. The goal in many applications is to demonstrate with statistical confidence that, in fact, there is no practical difference. Common examples include performance between model and simulation results versus live events, generic versus brand drug safety and efficacy, material engineering properties between different treatments, and many other problems the JMP community faces daily. While several platforms have the very well-named option Test Equivalence that uses the Two One-Sided Tests (TOST), in practice we often require more than equal means to properly characterize the similarity. We will demonstrate the capabilities in JMP to help establish equivalence in means, variances and distributions as an introduction. The focus of the talk will be on the equality between models; that is, not only the output responses being approximately equal, but also the consistency in the parameters characterizing the process. We will extend the custom estimation in JMP for regression models with Hotelling's T2 test statistic and show how the Functional Data Explorer can be used to decompose complex curves, followed by implementing standard equivalence methods for comparison.
