st: Accounting for clusters with a log-binomial model

[Nicole Basta <nbasta@u.washington.edu>]

Hello STATA listers,

For my analysis, I would like to use a log-binomial model to estimate
the Relative Risk and 95% CI for the association between several
potential risk factors (for this example, "RISKFACTOR1") and my
outcome variable ("OUTCOME1").

The data were collected from participants in a cross-sectional,
cluster-randomized survey of households (cluster variable "HHID"), so
I will need to account for clustering in my analysis.

My question is: What is the difference between how STATA accounts for
clustering in the following two models?

MODEL 1:
glm OUTCOME1 RISKFACTOR1, fam(bin) link(log) eform vce(cluster HHID)

MODEL 2:
svyset HHID
svy: glm OUTCOME1 RISKFACTOR1, fam(bin) link(log) eform

The output gives slightly different estimates of the SEs and the CIs.
Model 1 gives "Robust Standard Errors" and Model 2 gives "Linearlized
Standard Errors."

I'm wondering if that is the only difference between these two models,
and if one estimate of the SEs is more appropriate than the other for
a cluster-randomized, cross-sectional survey.

I would appreciate any insight that anyone on the list has.


Nicole
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/