Abstract:
|
Split-plot designs involve separate randomizations of two sets of factors, one at the whole-plot level and another at the split-plot level. Responses are typically measured following the application of both sets of factors. The linear model used to analyze these data includes error terms for the whole plot experiment and another for the split plot experiment with corresponding variance components denoted $\sigma_W^2$ and $\sigma_S^2$, respectively. It is sometimes possible to measure responses following application of the whole-plot factors and then again following application of the split-plot factors. For example, during the production of melt spun and drawn polymer fibers, polymer resin goes through an initial spinning process before the spun fibers are then drawn in a second separate operation. Multiple responses representing various properties are measured from the fibers before and after drawing. This talk explores multiple approaches to a combined analysis of this kind of split-plot data that improves estimation of the two variance components and leads to more reliable predictions.
|