| Re: Zambia hiv dataset 2001-02 [message #3156 is a reply to message #3154] |
Fri, 24 October 2014 15:31   |
Reduced-For(u)m
Messages: 292
Registered: March 2013
|
Senior Member |
|
|
Trevor,
Given that, on its own, stratification should (weakly) reduce the size of standard errors, where the clustering on PSU (weakly) inflates them - isn't this likely to produce standard error estimates (and thus p-values and CIs) that are all too small?
I would think that clustering on some geographic level greater than PSU would be the more conservative way to do this. The AIS tabulation plan* seems to indicate that they keep region of residence. Clustering on region-X-urban/rural might provide enough clusters to use the cluster robust estimator, but there are small-cluster-number analogs that can be used as well.
*Maybe I'm looking at the wrong documentation: http://dhsprogram.com/pubs/pdf/AISM9/HIV_Testing_Tabplan.pdf
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
