Yes, they both can do Taylor series linearization. It's all that SAS does, but SUDAAN has a bit more flexibility. So they get *similar* standard errors, which are approximations to the correct standard errors. It's easy enough to construct situations where they both over-estimate the true standard errors by a fair bit: the EMAP generalized random tesselation samples have a more complex error estimate (see papers by Donald Stevens of Oregon State U.) which is designed to be smaller than the linearization estimate under ideal conditions.

Probably. But I expect that the differences are small. I don't have SUDAAN to make any comparisons. If you get it, try it out and write back to SAS-L to let us all know how they compare.

David L. Cassell mathematical statistician Design Pathways 3115 NW Norwood Pl. Corvallis OR 97330

My colleagues, Arlene Siller and Linda Tompkins of NCHS presented a poster at NESUG which compares SAS, SPSS, Stata, and SUDAAN http://www.nesug.org/html/Proceedings/nesug05/pos/pos3.pdf

Ronald,

Attached paper is really help because we are evaluating the software to do survey analysis.

Thank you so much!

wensui