Longitudinal Analysis Through Multiple Process Steps, Mean and Variability

kathy_walker

Tony Cooper, PhD, Analytical Consultant, SAS

“Raw” parts might be randomly fed into an experimental subsequent process stage. How do you control for the initial value? How do you determine whether the experimental process stage seems better than the original version? What if there are many subsequent stages? What if the stages affect the variability, as well as the mean? Is a split-plot analysis the best option? What if the parts cannot be randomized, but must be fed into the next stage immediately? Industrial practitioners are familiar with the special cause/common cause model of causal instruction introduced by Shewhart and popularized by Deming. They may not be aware of the longitudinal repeated measures designs class of process studies. This presentation discusses the application of the special cause/common cause to longitudinal repeated measures designs. An immensely graphical tool is introduced that adheres to the Shewhart principles and, as a bonus, evaluates compound symmetry. An example of data collected on parts after each stage of the process will be used to demonstrate the synergistic use of control charts, ANOMV and mixed modeling.

Longitudinal Data.pdf

RepeatedMeasures.jmpaddin