svycontrast: Linear and nonlinearconstrasts of survey statistics

Description

Computes linear or nonlinear contrasts of estimates produced by survey functions (or any object with coef and vcov methods).

Usage

svycontrast(stat, contrasts, add=FALSE, ...)

Value

Object of class svrepstat or svystat or svyvar

Arguments

stat: object of class svrepstat or svystat
contrasts: A vector or list of vectors of coefficients, or a call or list of calls
add: keep all the coefficients of the input in the output?
...: For future expansion

Details

If contrasts is a list, the element names are used as names for the returned statistics.

If an element of contrasts is shorter than coef(stat) and has names, the names are used to match up the vectors and the remaining elements of contrasts are assumed to be zero. If the names are not legal variable names (eg 0.1) they must be quoted (eg "0.1")

If contrasts is a "call" or list of "call"s, and stat is a svrepstat object including replicates, the replicates are transformed and used to compute the variance. If stat is a svystat object or a svrepstat object without replicates, the delta-method is used to compute variances, and the calls must use only functions that deriv knows how to differentiate. If the names are not legal variable names they must be quoted with backticks (eg `0.1`).

If stats is a svyvar object, the estimates are elements of a matrix and the names are the row and column names pasted together with a colon.

Examples

Run this code

data(api)
dclus1<-svydesign(id=~dnum, weights=~pw, data=apiclus1, fpc=~fpc)

a <- svytotal(~api00+enroll+api99, dclus1)
svycontrast(a, list(avg=c(0.5,0,0.5), diff=c(1,0,-1)))
## if contrast vectors have names, zeroes may be omitted
svycontrast(a, list(avg=c(api00=0.5,api99=0.5), diff=c(api00=1,api99=-1)))

## nonlinear contrasts
svycontrast(a, quote(api00/api99))
svyratio(~api00, ~api99, dclus1)

## Example: standardised skewness coefficient
moments<-svymean(~I(api00^3)+I(api00^2)+I(api00), dclus1)
svycontrast(moments, 
quote((`I(api00^3)`-3*`I(api00^2)`*`I(api00)`+ 3*`I(api00)`*`I(api00)`^2-`I(api00)`^3)/
      (`I(api00^2)`-`I(api00)`^2)^1.5))

## Example: geometric means
## using delta method
meanlogs <- svymean(~log(api00)+log(api99), dclus1)
svycontrast(meanlogs,
    list(api00=quote(exp(`log(api00)`)), api99=quote(exp(`log(api99)`))))

## using delta method
rclus1<-as.svrepdesign(dclus1)
meanlogs <- svymean(~log(api00)+log(api99), rclus1)
svycontrast(meanlogs,
    list(api00=quote(exp(`log(api00)`)),
api99=quote(exp(`log(api99)`))))

## why is add=TRUE useful?

(totals<-svyby(~enroll,~stype,design=dclus1,svytotal,covmat=TRUE))
totals1<-svycontrast(totals, list(total=c(1,1,1)), add=TRUE)

svycontrast(totals1, list(quote(E/total), quote(H/total), quote(M/total)))


totals2<-svycontrast(totals, list(total=quote(E+H+M)), add=TRUE)
all.equal(as.matrix(totals1),as.matrix(totals2))

## more complicated svyby
means <- svyby(~api00+api99, ~stype+sch.wide, design=dclus1, svymean,covmat=TRUE)
svycontrast(means, quote(`E.No:api00`-`E.No:api99`))
svycontrast(means, quote(`E.No:api00`/`E.No:api99`))

## transforming replicates
meanlogs_r <- svymean(~log(api00)+log(api99), rclus1, return.replicates=TRUE)
svycontrast(meanlogs_r,
    list(api00=quote(exp(`log(api00)`)), api99=quote(exp(`log(api99)`))))

## converting covariances to correlations
vmat <-svyvar(~api00+ell,dclus1)
print(vmat,cov=TRUE)
cov2cor(as.matrix(vmat))[1,2]
svycontrast(vmat, quote(`api00:ell`/sqrt(`api00:api00`*`ell:ell`)))