ABSTRACT
The researcher designates the explanatory variables on the basis of substantive scientific theories. Statistical tests are unsatisfactory in non-experimental research for two fundamental reasons: It is almost impossible to design studies that meet the conditions for using the tests, and the situations in which the tests are employed make it difficult to draw correct inferences. The basic difficulty in design is that sociologists are unable to randomize their uncontrolled variables, so that the difference between “experimental” and “control” groups are a mixture of the effects of the variable being studied and the uncontrolled variables or correlated biases. Social scientists do so because, first, they are problems about which there is a great deal of misunderstanding, evident in current research. And second, they are statistical problems on which there is broad agreement among research statisticians—and on which these statisticians generally disagree with much in the current practice of research scientists.
