DESIGN TOPICS I
Reducing variability is a major concern in planning an experiment, discussed with the Experimental Pyramid in Chapter 1. One reason is substantive: Reducing variability means getting the behavior under better control; better control generally improves the quality of the response. The other reason is statistical: Lower variability means greater power and tighter confidence intervals.
The first main section considers two practical aspects of controlling variability that deserve more consideration than they often get: Screening extreme scores and writing instructions.
Three classes of designs that can help reduce variability are considered in the last three main sections. Block design includes an individual difference variable as a design factor, and thereby removes it from the error term. Latin square design extends repeated measures design by balancing treatments across serial position. Finally, some dangers of using repeated measures are considered in the discussion of Within Subjects and Between Subjects Design.
Increasing validity goes together with reducing variability. The techniques for reducing variability in the first two main sections tend also to increase validity. Latin squares in the third main section seek the same goals by using repeated measures. Latin square designs improve on the repeated measures designs of Chapter 6 by providing ways to deal with order effects.
SCREENING SUBJECTS AND WRITING INSTRUCTIONS
Two problems of design and procedure are considered here. Both address the problem of extreme scores, a major headache of experimental analysis.
One problem concerns eliminating subjects and/or data. The other problem concerns writing instructions. On these two intensely practical problems, the present discussion is limited and provisional.