When the CLUSTER and DOMAIN statements are specified in PROC SURVEYMEANS, if all the values of a continuous analysis variable are missing for a domain level within a cluster and that cluster contains nonmissing observations for other domain levels, then the variance estimation and related statistics will be incorrect in the Domain Analysis results for the domain(s) that have the missing values. The incorrect results for the affected domain(s) include the standard error and variance of the mean, the standard deviation and variance of the sum, the confidence limits for the mean and sum, the t statistic and its p-value, and the coefficient of variation.

To circumvent this problem, use the %SMSUB macro.

or you may contact Technical Support to obtain a copy of the macro.

We also have a fix available that can be applied to PROC SURVEYMEANS in Release 8.1 running on a Windows operating system. To obtain the fix, contact SAS Technical Support at (919) 677-8008 or via our web site at SAS Technical Support | SAS Support.