PURPOSE This research examines the following hypothesis concerning clinical trials: different allocation ratios may cause different health outcomes due to the combined influence of placebo effects and compliance effects.

METHOD Recent work claims to show placebo effects in major clinical trials for anti-ulcer and cholesterol-lowering drugs. In a 2006 paper published in the Journal of Political Economy, Malani uses variation in the allocation proportion as a proxy for different patient expectations about the value of treatment. He finds that a higher allocation proportion is associated with a significantly higher healing rate in ulcer trials and significantly greater reduction in LDL levels in statin trials.

Placebo effects occur when outcomes are caused by patient expectations, as opposed to physiological treatment effects. In this paper, I distinguish between direct and indirect effects of patient expectations. The traditional notion of placebo effects is a direct effect of patient confidence in the value of treatment: better health outcomes may occur as a direct result of patient's thoughts, separate from any physiological treatment effect. It is inappropriate, however, to simply equate higher allocation ratio to higher patient expectations. An indirect effect of patient expectations may be better adherence to treatment. If this is the case, then better health outcomes may indirectly result from higher expectations, but the direct cause would indeed be a real physiological treatment effect. In the current paper, I replicate the Malani ulcer and cholesterol study with one important alteration: I have added compliance data collected from as many of the trials as possible.

RESULTS Regression analysis reveals a positive correlation between the probability of treatment and outcomes in the control group, but only when the control group is an active treatment (not a placebo control), namely H2 blockers in PPI trials. There is no positive correlation when the control is an inert pill. This result suggests that possible placebo effects differ in the treatment and control arms. A separate result shows that higher compliance is associated with higher allocation ratios.

CONCLUSIONS Taken together, the two primary findings indicate that placebo effects may be significant, but that compliance variables must be included for unbiased estimation of placebo effects.