Correlation and Causation

Abstract

Page and various collaborators (Keith, Page, & Robertson, 1984; Page, 1981; Page & Keith, 1981, 1982) have defended the use of data from investigations (correlational studies) to make causative conclusions. These methods are not defensible logically or statistically. They can only suggest hypotheses that then should be tested by proper experiments. At worst, as in the Coleman studies, they have been used to make social policy based on unjustified conclusions. The logical flaw of the methods of path analysis or structural equation modeling is shown. A proper evaluation of the role of investigations versus experiments is cited in the work of Cochran (as described by Rubin [1983]).