I have attempted to view the following experiment as a split plot design:
15 undergraduates serve as subjects for an experiment that compares two hypnosis methods. There are two phases to the experiment. In phase one, all subjects take an exam to recall 16 word-number pairs in a normal waking state. The experimenters record how many errors they make in reciting the lists. Their errors are recorded until they either go through the entire list without making an error or if they use up a maximum of 15 attempts.
In phase two, the subjects are randomly assigned one of the three conditions to memorize a new 16 word-pair list of equal difficulty: normal waking, hypnosis, or alternative hypnosis state. Next, they are given the second test and their errors are recorded in the same vein as the first test.
My argument is that it it's a split plot because the first test serves as a benchmark. The blocks are created by randomly assigning the between-block units, the subjects, a state (between-block factor). Next, all groups, the within-block unit, receive the within-block factor, the second test.
The response in this experiment is the # of errors made on the second test.
Does this reasoning make sense? Is it ok to view the first assessment as a benchmark? Please do not tell me the answer but do tell where my reasoning may have gone astray if that is the case?
My other inkling is that if I ignore, the first test, then this is just a completely randomized experiment where the units change to the three groups of five if this were to be balanced.