On the missing plot technique

Abstract

In using experimental designs the situation sometimes arises when one or more observations is "missing" when this happens the design loses its symmetry balance, so that the usual analysis of variance calculations on the remaining data are no longer as straightforward. One common method for handling this situation is to find appropriate estimates of the missing observations so that an analysis of variance carried out on the now "complete" data is equivalent to the analysis of the data actually available. When there are missing observations, the least sequares estimation and S.S. for testing a linear hypothesis are reviewed, and the methdo of maximum likelihood is considered about the model with uncorrelated errors, and with correlated errors. And when two-way classification model with mising cells is connected, the correct sums of squares for nine hypotheses usually tested are constructed.