A Formal Test for the Existence of Placebo Effects in Clinical Trials

In this paper I criticize the methodology of existing studies that purport to find evidence of placebo effects in medical trials. Among other things, because these studies fail to define and properly model placebo effects, it is hard to determine the power of their results. I address the problem by proposing a model of health outcomes that formalizes the dominant medical theory of how placebo effects operate. (This theory posits that health outcomes rise in individuals' expectations about their beliefs about the probability that they are getting a beneficial treatment and their beliefs about how beneficial that treatment is.) I then model randomized, double-blind, parallel-arm, placebo-controlled trials and set forth conditions under which placebo effects alter observable outcomes in such trials given self-selection of subjects into the trials. I argue that a powerful method of testing for the existence of placebo effects is checking whether outcomes in trials where a higher fraction of subjects are randomized into active treatment are superior to outcomes in trials with a lower fraction given active treatment. I apply this test to data from over 60 trials of ulcer medication and find strong evidence of placebo effects in trials of H2-receptor antagonists such as Zantac and Tagamet, two widely-used ulcer medications