R help - Sep 2004 - Question on survey package

Is there a way to use one of the functions in the survey package to get
a table of estimated percentages (or proportions) and the standard error
for each?  For example, suppose that AGECODE AND SEX are two factors
with 5 and 2 levels. 
 
The 5x2 AGECODE x SEX table would have the estimated percentage of
persons in each cell, 
 
100*(sum of weights in the cell) / (sum of all weights)
 
and the std error of the percentage. The std error would have to be
computed by linearization or replication since the denominator of each
percentage is an estimate. I'd also like the option of getting row,
column, or total percentages.
 
Is this possible? A colleague and I have read the manuals closely and
tried many variations with svytable, svytotal with no success.
 
Version R 1.9.1 on Windows XP
 
TIA,
Richard Valliant
 
 

	[[alternative HTML version deleted]]

On Thu, 2 Sep 2004, Richard Valliant wrote:
> Is there a way to use one of the functions in the survey package to get
> a table of estimated percentages (or proportions) and the standard error
> for each?  For example, suppose that AGECODE AND SEX are two factors
> with 5 and 2 levels.
>
> The 5x2 AGECODE x SEX table would have the estimated percentage of
> persons in each cell,
>
The computations can be done easily enough with svymean or svrepmean, but
the layout isn't quite what you want

Example:

library(survey)
data(api)
dclus1<-svydesign(id=~dnum, weights=~pw, data=apiclus1, fpc=~fpc)
> svymean(~interaction(stype,comp.imp),design=dclus1)
                                      mean     SE
interaction(stype, comp.imp)E.No  0.174863 0.0260
interaction(stype, comp.imp)H.No  0.038251 0.0161
interaction(stype, comp.imp)M.No  0.060109 0.0246
interaction(stype, comp.imp)E.Yes 0.612022 0.0417
interaction(stype, comp.imp)H.Yes 0.038251 0.0161
interaction(stype, comp.imp)M.Yes 0.076503 0.0217


Turning this into a 3x2 table with a mean and SE in each cell would take a
bit of work since R doesn't have a very general table layout system.

One approach is

ftable.svystat<-function(x,rownames){

    m<-cbind(coef(x),sqrt(diag(vcov(x))))
    if (is.null(rownames))
        return(as.table(m))

    rowdim<-sapply(rownames,length)

    mm<-array(m,dim=c(rowdim,NCOL(m)),
             dimnames=c(as.list(rownames),
             list(c("coef","SE"))))


    ftable(mm,row.vars=length(rowdim)+0:1)

}

which gives

a <- svymean(~interaction(stype,comp.imp),design=dclus1)
b <-
ftable(a,list(c("E","H","M"),c("No","Yes")))

round(b*100,1)
             E    H    M

No  coef  17.5  3.8  6.0
    SE     2.6  1.6  2.5
Yes coef  61.2  3.8  7.7
    SE     4.2  1.6  2.2


or with named dimnames
>
b<-ftable(a,list(stype=c("E","H","M"),comp.imp=c("No","Yes")))
> round(b*100,1)
              stype    E    H    M
comp.imp
No       coef       17.5  3.8  6.0
         SE          2.6  1.6  2.5
Yes      coef       61.2  3.8  7.7
         SE          4.2  1.6  2.2



	-thomas