Beginning in SAS 9.2, you can specify the NOMCAR option on the PROC SURVEYMEANS statement. This option requests that the procedure treat missing values in the variance computation as not missing completely at random (NOMCAR) for Taylor series variance estimation.

Prior to SAS 9.2, two approaches are available to include missing values (item nonrespondents) in the variance estimation and produce the same results as SUDAAN. For mean and total estimates and their variances, you can create a response indicator variable for a specific analysis variable and add the indicator variable to the DOMAIN statement. This will include the missing values in the variance estimation and effectively treat the non-missing values as a subgroup of item respondents. However, the degrees of freedom will be different between SAS and SUDAAN depending on the number of missing values. This difference will affect confidence limits. The number of missing values for the respondent population will always be reported as zero under the domain results in SAS, while SUDAAN will report the number of missing values for the analysis variable. The example titled "Analyze Survey Data with Missing Values" in the SURVEYMEANS chapter of the SAS/STAT User's Guide illustrates the indicator variable approach.

Operating System and Release Information