Randomisation to protect against selection bias in healthcare trials. Cochrane Database of Methodology Reviews 4:MR000012
Regina Kunz, Gunn Vist and Andy Oxman showed in their 2002 Cochrane Review that comparisons of the results of randomised trials with those of non-randomised (observational) studies sometimes reveal concordance and sometimes discordance. Discordance seems likely to reflect inadequately controlled allocation biases in non-randomised studies. Unfortunately, it is not currently possible to predict which non-randomised (observational) studies will yield results that will be discordant with those of randomised trials.