surveysummary: Summary statistics for sample surveys

Description

Compute means, variances, ratios and totals for data from complex surveys.

Usage

# S3 method for survey.design
svymean(x, design, na.rm=FALSE,deff=FALSE,influence=FALSE,...) 
# S3 method for survey.design2
svymean(x, design, na.rm=FALSE,deff=FALSE,influence=FALSE,...) 
# S3 method for twophase
svymean(x, design, na.rm=FALSE,deff=FALSE,...) 
# S3 method for svyrep.design
svymean(x, design, na.rm=FALSE, rho=NULL,
  return.replicates=FALSE, deff=FALSE,...) 
# S3 method for survey.design
svyvar(x, design, na.rm=FALSE,...) 
# S3 method for svyrep.design
svyvar(x, design, na.rm=FALSE, rho=NULL,
   return.replicates=FALSE,...,estimate.only=FALSE) 
# S3 method for survey.design
svytotal(x, design, na.rm=FALSE,deff=FALSE,influence=FALSE,...) 
# S3 method for survey.design2
svytotal(x, design, na.rm=FALSE,deff=FALSE,influence=FALSE,...) 
# S3 method for twophase
svytotal(x, design, na.rm=FALSE,deff=FALSE,...) 
# S3 method for svyrep.design
svytotal(x, design, na.rm=FALSE, rho=NULL,
   return.replicates=FALSE, deff=FALSE,...)
# S3 method for svystat
coef(object,...)
# S3 method for svrepstat
coef(object,...)
# S3 method for svystat
vcov(object,...)
# S3 method for svrepstat
vcov(object,...)
# S3 method for svystat
confint(object,  parm, level = 0.95,df =Inf,...)
# S3 method for svrepstat
confint(object,  parm, level = 0.95,df =Inf,...)
cv(object,...)
deff(object, quietly=FALSE,...)
make.formula(names)

Value

Objects of class "svystat" or "svrepstat", which are vectors with a "var" attribute giving the variance and a "statistic" attribute giving the name of the statistic.

These objects have methods for vcov, SE, coef,

confint, svycontrast.

Arguments

x: A formula, vector or matrix
design: survey.design or svyrep.design object
na.rm: Should cases with missing values be dropped?
influence: Should a matrix of influence functions be returned (primarily to support svyby)
rho: parameter for Fay's variance estimator in a BRR design
return.replicates: Return the replicate means/totals?
deff: Return the design effect (see below)
object: The result of one of the other survey summary functions
quietly: Don't warn when there is no design effect computed
estimate.only: Don't compute standard errors (useful when svyvar is used to estimate the design effect)
parm: a specification of which parameters are to be given confidence intervals, either a vector of numbers or a vector of names. If missing, all parameters are considered.
level: the confidence level required.
df: degrees of freedom for t-distribution in confidence interval, use degf(design) for number of PSUs minus number of strata
...: additional arguments to methods,not currently used
names: vector of character strings

Author

Thomas Lumley

Details

These functions perform weighted estimation, with each observation being weighted by the inverse of its sampling probability. Except for the table functions, these also give precision estimates that incorporate the effects of stratification and clustering.

Factor variables are converted to sets of indicator variables for each category in computing means and totals. Combining this with the interaction function, allows crosstabulations. See ftable.svystat for formatting the output.

With na.rm=TRUE, all cases with missing data are removed. With na.rm=FALSE cases with missing data are not removed and so will produce missing results. When using replicate weights and na.rm=FALSE it may be useful to set options(na.action="na.pass"), otherwise all replicates with any missing results will be discarded.

The svytotal and svreptotal functions estimate a population total. Use predict on svyratio and svyglm, to get ratio or regression estimates of totals.

svyvar estimates the population variance. The object returned includes the full matrix of estimated population variances and covariances, but by default only the diagonal elements are printed. To display the whole matrix use as.matrix(v) or print(v, covariance=TRUE).

The design effect compares the variance of a mean or total to the variance from a study of the same size using simple random sampling without replacement. Note that the design effect will be incorrect if the weights have been rescaled so that they are not reciprocals of sampling probabilities. To obtain an estimate of the design effect comparing to simple random sampling with replacement, which does not have this requirement, use deff="replace". This with-replacement design effect is the square of Kish's "deft".

The design effect for a subset of a design conditions on the size of the subset. That is, it compares the variance of the estimate to the variance of an estimate based on a simple random sample of the same size as the subset, taken from the subpopulation. So, for example, under stratified random sampling the design effect in a subset consisting of a single stratum will be 1.0.

The cv function computes the coefficient of variation of a statistic such as ratio, mean or total. The default method is for any object with methods for SE and coef.

make.formula makes a formula from a vector of names. This is useful because formulas as the best way to specify variables to the survey functions.

Examples

Run this code


  data(api)

  ## one-stage cluster sample
  dclus1<-svydesign(id=~dnum, weights=~pw, data=apiclus1, fpc=~fpc)

  svymean(~api00, dclus1, deff=TRUE)
  svymean(~factor(stype),dclus1)
  svymean(~interaction(stype, comp.imp), dclus1)
  svyquantile(~api00, dclus1, c(.25,.5,.75))
  svytotal(~enroll, dclus1, deff=TRUE)
  svyratio(~api.stu, ~enroll, dclus1)

  v<-svyvar(~api00+api99, dclus1)
  v
  print(v, cov=TRUE)
  as.matrix(v)


  # replicate weights - jackknife (this is slower)
  dstrat<-svydesign(id=~1,strata=~stype, weights=~pw,
        data=apistrat, fpc=~fpc)
  jkstrat<-as.svrepdesign(dstrat)

  svymean(~api00, jkstrat)
  svymean(~factor(stype),jkstrat)
  svyvar(~api00+api99,jkstrat)

  svyquantile(~api00, jkstrat, c(.25,.5,.75))
  svytotal(~enroll, jkstrat)
  svyratio(~api.stu, ~enroll, jkstrat)

  # coefficients of variation
  cv(svytotal(~enroll,dstrat))
  cv(svyratio(~api.stu, ~enroll, jkstrat))

  # extracting information from the results
  coef(svytotal(~enroll,dstrat))
  vcov(svymean(~api00+api99,jkstrat))
  SE(svymean(~enroll, dstrat))
  confint(svymean(~api00+api00, dclus1))
  confint(svymean(~api00+api00, dclus1), df=degf(dclus1))

  # Design effect
  svymean(~api00, dstrat, deff=TRUE)
  svymean(~api00, dstrat, deff="replace")
  svymean(~api00, jkstrat, deff=TRUE)
  svymean(~api00, jkstrat, deff="replace")
 (a<-svytotal(~enroll, dclus1, deff=TRUE))
  deff(a)

## weights that are *already* calibrated to population size
sum(weights(dclus1))
nrow(apipop)
cdclus1<- svydesign(id=~dnum, weights=~pw, data=apiclus1,
fpc=~fpc,calibrate.formula=~1)
SE(svymean(~enroll, dclus1))
## not equal to SE(mean)
SE(svytotal(~enroll, dclus1))/nrow(apipop)
## equal to SE(mean)
SE(svytotal(~enroll, cdclus1))/nrow(apipop)

Run the code above in your browser using DataLab