Strip-plot designs, mixed models, and comparisons between linear and non-linear models for microtitre plate bioassays.

Abstract

By paying careful attention to the experimental units and, randomizing where it is reasonable, we are able to devise designs for cell culture bioassay which are statistically legitimate and practical in the laboratory. These designs, strip-plot layouts, can accommodate linear or non-linear models, fixed or mixed models, and can provide good protection from location effects. Non-linear mixed models using these designs can be extended to address serial dilution error. A statistically based approach which includes randomization and proper design is a powerful tool for identifying subtle effects of factors within and outside bioassays; this attention to statistical detail is a necessary first step towards continuous improvement of biological assays.