ABSTRACT
In this chapter, the authors discuss analysis from first principles rather than using the specialized results of the theory of confounded designs. There were 32 judges and the experiment was done on two days. For the data and several different analyses, see N. L. Johnson. There are some important special features about these data. A crucial point concerns the role of differences between judges. If these are relatively minor, they may regard the mean score given by a judge to a particular treatment as a derived reponse; because each judge looks only at two treatments, differences between judges are rather poorly determined. A rather complex experimental design has been used. The design used involves a particular scheme of confounding. A more cautious approach is to eliminate differences between judges by taking as derived response variable the difference in means between the two treatments for a particular judge.
