A note on confidence limits for quartiles with right censored data.

Two new methods are proposed for constructing the confidence limits for quartiles. The small sample behaviour of these two methods is compared with the Jennison and Turnbull modified Brookmeyer-Crowley method at three quartiles. The simulation study indicates that one of the new methods, the smoothed modified reflected method, is preferred over the other methods when the censoring rate is less than 40 per cent while the Jennison and Turnbull method is preferred for higher censoring. The results for the upper quartile are similar to the median. For the lower quartile with small sample and high censoring, semi-infinite intervals may often occur. The correct practice is to permit the semi-infinite intervals without modifying them since it is more informative and gives closer to nominal level coverage.