Estimating Treatment Response in Clinical Trials When Placebo Effects are Present

Abstract

In this paper I explore estimation of treatment response in blinded clinical trials when there exist placebo effects, i.e., when health outcomes are a function of not just the physiological effects of treatment, but also beliefs about treatment efficacy and treatment state. First, I argue that the appropriate parameter of interest in the short-run includes not only the physiological effect of treatment, but also individuals' beliefs about the efficacy of treatment. Second, the standard estimator of treatment response - differencing average outcomes in the treatment and placebo-control groups - is biased. The main component of this bias is a product of blinding. Individuals in the treatment arm of a blinded trial do not respond as well to treatment as individuals outside the trial context because they think there is a chance they are not being treated. Conversely, individuals in the placebo-control arm respond better than individuals outside the trial context because they think there is a chance they are being treated. Differencing average outcomes across the arms underestimates the impact of treatment. The second bias is due to selection. While selection may be a problem even in the absence of placebo effects, the problem is exacerbated by placebo effects because the beliefs that drive the participation decision now have an additional, direct impact on outcomes. The sign of this effect, however, is ambiguous. My third contribution is an experimental trial design that permits unbiased estimation of the proper parameter of interest in clinical trials.