Confounding is an ever-present problem in scientific inference. The placebo effect in medicine is a classic example, in which a beneficial outcome can be produced by an inert treatment that has no effect per se. Whatever real medicinal value a treatment may have is thus overlaid by—confounded with—this placebo effect.
Confounding means that an observed effect could be due to more than one cause. In the placebo example, giving a medicine carries a suggestion that it should be beneficial; this suggestion is often beneficial of itself Medicine and suggestion are thus confounded. Accordingly, a positive outcome is not generally adequate evidence for a genuine effect of the medicine per se.
Four complementary, overlapping perspectives on confounding are useful. Stimulus confounding is the most common perspectivei in experimental analysis. Stimulus confounding arises from differences between the conceptual variable we wish to manipulate and the concrete physical situation we actually manipulate. The physical situation always involves other variables besides our intended variable. Our observed result may stem from these other variables, not from our intended variable.
Response confounding means that the observed response is, or may be, a compound of two or more response processes. Response confounding is a primary concern in observational studies, commonly discussed in terms of validity. Response confounding may be extrinsic, as with remediable ambiguity in a test question, or intrinsic, as when a one-dimensional measure is used to represent a multidimensional concept.