Control, volition, and the “experimental method”

Abstract

Our textbooks claim that the “experimental method,” in contrast to quasi-experimentation or mere observation, provides a special key for understanding causal relationships. I argue that this claim is based on several flawed assumptions. First, the postulated “experimental method” presumes that the effects of uncontrolled relevant variables can be stastically eliminated by random assignment of subjects to treatments. In most or all psychological settings, however, I argue that this presumption is false. Either because interactions between uncontrolled variables and independent variables are significant or because “control” treatments are “impure,” the effects of the uncontrolled variables cannot be statistically eliminated by random assignment, and conclusions about independent variables will be invalid. Thus, I argue that psychological research is, at best, quasi-experimental. Second, I argue that the postulated “experimental method” is logically incoherent; it presumes not only that natural laws determine the outcomes of experiments, but also, with equal importance, the contradictory premise that human freedom is necessary to design and conduct experiments. Because this postulate of human freedom for experiments is so central, it is puzzling why the power of the hypothesized “experimental method” would not be limited by comparable freedom available to subjects. Third, I argue that the textbook claim distorts historically important views. Presumptions that subjects do have such freedom have always been implicit, though their implications for causal analysis have not always been attended to, in many experimental procedures used to study perceptual, cognitive, and human motivational processes. These presumptions have played a central, and continuing, role in “actionist” theories of cognitive function and development, which reject more recent positivistic positions. Because they were so common, they also played a central role in the a priori rejection of the usefulness of cognitive psychology by behavioristic critics, like Skinner. Although our textbooks often slight the historical importance of Cartesian dualism, elements of it seem rooted in all of the methods for causal inference that psychologists have traditionally used. Finally, I argue that the problems faced by psychologists are more like those faced in long-range weather forecasting than in kinematics. Controlled experimentation may never be a particularly useful method for studying these problems, because of the very large number of causally effective independent variables. Access to information may always be more important for causal analysis than opportunities to manipulate putative controlling variables.