Randomisation: magical cure for bias?

Introduction There is general consensus that randomised clinical trial (RCT) can provide the most valid conclusions about effects of different treatment as eligible patients are randomly allocated into two or more alternative treatments. Trials using non-randomised comparison groups like historical controls tend to yield more optimistic results than randomised trials. It is believed that random allocation will remove the “selection bias”, which is present when comparison groups are assembled in some other way. Another reason for randomisation is that the computation of sampling errors is based on random sampling. Hence, if the selected samples behave like random samples, the observations can be compared with what we would expect if there were no difference in treatment effects (null hypothesis). However, since most tests of statistical significance are fairly robust, randomisation is used primarily for the control of selection bias.