The Experimental design page has been updated with additional information about the considerable limitations of propensity matching in clinical studies. These studies should not be considered as a substitute for a properly conducted prospective trial, and results from them should only be considered as associations, not as implying causality. The increasing prevalence of these types of studies is due to the growth in health data collection (often in forms not suitable for clinical study) and because it is cheaper to employ statisticians than to conduct good clinical trials.