Yukio Takahashi, CEO, BioStat Research Institute, Inc.
In most books on experimental design, you cannot find any description of various analytical models using JMP's "Model Fitting". One of the keywords is "random effect". In order to analyze nested experimental data for the purpose of estimating variance components, you need to change the attribute by selecting "Nested Factor" for the lower-level factor you want to make a random effect in "Configure Model Effects", and then selecting "Random Effects" from the "Attributes" structure.
Even more troublesome is the choice of analysis method, "EMS (conventional)" or "REML (recommended)." In conventional branched experiments, the degrees of freedom of the analysis of variance table expand as the branches become lower, resulting in inefficient experiments. To prevent the expansion of the degrees of freedom of the lower branches, a staggered branched experiment was devised and is included in "JIS Z8402-3 Accuracy of measurement methods and measurement results (trueness and precision) -- Part 3: Intermediate precision of standard measurement methods," but it is difficult to understand.
Here, we use artificial data based on human blood pressure measurements to compare a standard branching experiment with a staggered branching experiment, and carefully explain the analysis methods and how to interpret the results using EMS and REML.
