A challenge by Brown (1984) of the Mantel-Haenszel variance for the log rank test is dismissed as due to the nonasymptotic situation considered by Brown. Where the censoring mechanism in a time-to-response study could be influenced by treatment, the permutational variance is inappropriate and the Mantel-Haenszel hypergeometric variance must be used. If treatment is known or postulated to have no effect on the censoring mechanism, then use of the Mantel-Haenszel variance removes as a source of variation the chance difference in censoring that might occur.