A Common Class of Pseudo-Experiments
Julian C. Stanley · American Educational Research Journal · 1966
Suppose you wish to study the effect of a specific kind of brain damage on the maze-running ability of albino rats. You can damage the brains of a random half of your rats and compare their performance with that of the other random half whose brains are not damaged. Methodological niceties such as the analysis of or a levels-by-treatments design may be used to increase the precision of your comparison, but the essential element is randomized assignment of experimental units to experimental conditions before the experiment begins. If the analysis of or a matched-pairs arrangement is used, values of the covariable or matching variable (e.g., running time) must be secured prior to the brain damage in order not to be affected by it. One would not, for example, first inflict the brain damage on a random half of the rats and then secure maze-running times to serve as the antecedent variable. In this sense the word covariance in the analysis of has a misleading connotation.' Now suppose that you are a researcher in mental retardation and wish to ascertain the effect of brain damage on the arithmetic-reasoning ability of twelve-year-old boys. Humanitarian considerations rule out your taking a group of such boys and brain-damaging a random half of them. However, nature, in the form of congenital and other accidents, has performed an experiment for you. There exists a large number of twelveyear-old boys whose brains are damaged, and a much larger number who show no neurological or anatomical signs of brain damage. But nature,