There is a large literature criticising the use of significance tests of differences in baseline variables in randomised clinical trials [1], [2], [3], [4], [5], [6]. The usual argument is that, if groups are chosen randomly, the two groups are from the same population and any null hypothesis of a zero difference between their populations is true. Testing is superfluous, because any significant difference is a type I error, one of the 5% expected to occur. Even very small P values may happen occasionally when the null hypothesis is true. Not all agree and Berger [7] argues that such tests are a valid method to identify possible bias and subversion in allocation. Subversion can and does occur; Kennedy and Grant [8] reported a trial where some centres used a central randomisation system and others used sealed envelopes held locally. There was a significant difference in age between randomised groups in the local allocation centres, for three clinicians in particular, but not for the central allocation centres. Berger and Exner [9] present a test for imbalance which can be more powerful than baseline comparisons, being based on response data rather than baseline.
