Dear SAS Community,
I am currently using PROC SURVEY procedures (such as SURVEYMEANS, SURVEYLOGISTIC) to analyze complex survey data and have some questions regarding how the NOMCAR option handles missing values in variance estimation.
According to the SAS documentation and related papers, when the NOMCAR option is specified, SAS treats observations with and without missing values for the analysis variable as two different domains. It then calculates the point estimates (e.g., mean, sum) only in the non-missing domain and performs a domain analysis for Taylor Series variance estimation.
My questions are:
1. During the Taylor Series linearization and variance estimation process, how exactly does SAS handle the observations that have missing values?
2. Although these missing observations do not contribute to the point estimates, how do they contribute to the final variance estimation (for example, their impact on strata, clusters, and weights)?
3. Are there any more detailed mathematical explanations or algorithmic descriptions available that can help me understand how NOMCAR works internally within the Taylor Series framework?
I have already reviewed the following materials but could not find more technical details on the exact computation:
SAS Official Documentation: The SURVEYMEANS Procedure (section on NOMCAR and Missing Values)
SAS Official Documentation: The SURVEYLOGISTIC Procedure (section on NOMCAR and Missing Values)
Gorrell, P. (2009). Survey Analysis: Options for Missing Data. NESUG 2009 Proceedings.
If anyone has come across more in-depth technical documents, SAS Help Center, or example code that illustrates this process, I would greatly appreciate your guidance.
Thank you in advance for your help!
