|
Abstract:
|
Split plot designs have become recognized as a useful alternative to completely randomized designs in industrial experiments when there are practical difficulties in changing the levels of one or more factors due to cost or effort. These restrictions on the randomization lead to potential correlation structure that can affect inference procedures. In this talk, we examine estimation of the variance-covariance matrix, or standard errors, of the regression estimators. In many cases, the usual estimator is an underestimate of the true variance-covariance matrix of the regression parameter estimator since it does not account for the additional uncertainty attributed to the estimation of the correlation structure. We will discuss such cases as well as present commonly recommended adjustments to deal with this problem.
|
