[R] survey package question

[Thomas Lumley]

On Thu, 18 Feb 2010, Richard Valliant wrote:

> Should the svyby function be able to work with svyquantile?  I get the
> error below ...

It works, but you need to either specify ci=TRUE or keep.var=FALSE.  The problem is that svyquantile() by default does not produce standard errors.

> svyby(~api00, ~stype, design=dclus1,svyquantile, quantile=c(0.25,0.5,.75),ci=TRUE)
   stype   0.25   0.5  0.75  se.0.25   se.0.5  se.0.75
E     E 553.00 652.0 729.0 27.74745 37.81025 17.12898
H     H 523.00 608.0 699.5 46.01881 65.82488 33.38789
M     M 532.75 636.5 696.5 60.78990 43.12310 55.27555
> svyby(~api00, ~stype, design=dclus1,svyquantile, quantile=c(0.25,0.5,.75), keep.var=FALSE)
   stype statistic1 statistic2 statistic3
E     E     553.00      652.0      729.0
H     H     523.00      608.0      699.5
M     M     532.75      636.5      696.5


> A more general question is: can quantiles and their SEs be computed for
> subgroups?
>

You can also use subset(), which is what svyby() does internally

> svyquantile(~api00, quantile=c(0.25,0.5,0.75), subset(dclus1, stype=="E"))
       0.25 0.5 0.75
api00  553 652  729


        -thomas


Thomas Lumley			Assoc. Professor, Biostatistics
tlumley at u.washington.edu	University of Washington, Seattle