svyciprop_test.R

# setup
library("survey")
data("api")
apiclus1$proportion <- apiclus1$pcttest/100
dclus1<-svydesign(id=~dnum, fpc=~fpc, data=apiclus1)

# confint(svymean(...) == svyciprop(..., method = "me")
confint(svymean(~ proportion, dclus1))
##                2.5 %    97.5 %
## proportion 0.9852196 0.9943432
svyciprop(~ proportion, dclus1, method="me")
##                   2.5% 97.5%
## proportion 0.990 0.985  0.99

# but svyciprop(..., method = "lo") returns "logit" CIs
svyciprop(~ proportion, dclus1, method="lo")
##                   2.5% 97.5%
## proportion 0.990 0.983  0.99


# for subsets, we can use svyby()
svyby(~ proportion, ~ stype, dclus1, FUN = svymean, vartype = "ci")
##   stype proportion      ci_l      ci_u
## E     E  0.9921528 0.9891645 0.9951411
## H     H  0.9671429 0.9471201 0.9871656
## M     M  0.9888000 0.9802725 0.9973275

# which is the same as svyciprop() on a subset()
svyciprop(~ proportion, subset(dclus1, stype == "E"), method="me")
##                   2.5% 97.5%
## proportion 0.992 0.989     1
svyciprop(~ proportion, subset(dclus1, stype == "H"), method="me")
##                   2.5% 97.5%
## proportion 0.967 0.943  0.99
svyciprop(~ proportion, subset(dclus1, stype == "M"), method="me")
##                   2.5% 97.5%
## proportion 0.989 0.979     1

# but if I want svyciprop() confidence intervals on subsets, can I get them with svyby()?
# seems not. it's not this:
svyby(~ proportion, ~ stype, dclus1, FUN = svyciprop)
##   stype proportion se.as.numeric(proportion)
## E     E  0.9921528               0.001524666
## H     H  0.9671429               0.010215856
## M     M  0.9888000               0.004350837

I think it'll work like this:

svyby(~ proportion, ~ stype, dclus1, FUN = svyciprop, vartype="ci")

Cheers, maybe a little late.