Unbiased medians for nhanes subgroups?

[ProfB]

I'm trying to generate median values for urine trace elements using NHANES data.  Proc surveymeans does not produce quantiles when a domain statement is used and employment of 'where' or 'class' disrupts the weights and does not produce population representative medians.  So....I'm stuck.  Any advice on how I can generate unbiased medians for age/gender subgroups?  Thanks much.

[ballardw]

I suspect the issue of the domain analysis and quantiles revolves around the confidencle limits and variability and such combined with the treatments of ties.

Since median is a central measure AND unless you need confidence limits I would start with Proc Means with the weights and your domain variable as a class variable. Generally the central measures don't vary between Means and SurveyMeans.

[ProfB]

Thanks very much, How would you suggest I handle the clustering?

[ballardw]

By clustering do you mean the Age group in your example? I would normally think of that as a domain variable as cluster with survey procs relates to sampling design. Does NHANES use a cluster sample design?

Any way, clustering is just factor in the calculations for variability, not the central measure. So even with a cluster design I don't think it will effect the actual median.