Restriction Errors for Linear Models (An Aid to Develop Models for Designed Experiments)

A random error component is ilntroduced into linear models for analyzing data from designed experiments. This error component is called a 'restriction error' because it is used whenever a restriction on randomization occurs in the design. Discussion of the usefulness of this restriction error in randomized complete block designs, nested factorials, and split plot designs is given. Accompanying each discussion is an example of an experiment on prosthetic cardiac valves. Further, the restrictioin error concept is used to look at the controversy regarding a pooled split plot error.