In mabafaba/surveyweights: Calculate weights from sampling frames

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

General workflow

Prerequisites; You need to have:

• your data as a data.frame
  • one column should uniquely specify the stratum of the record
• your sampling frame as a data.frame
  • one row per stratum
  • one column with the stratum names (corresponding exactly to the values in the stratum name column in the data)
  • one column with the corresponding population numbers

library(surveyweights)
# example data:
mydata<-data.frame(somevar1=c(1:5),
                   somevar2=c("one","more","test","variable",NA),
                   strata_names=c("a","a","b","b","a"),
                   cluster_names=c("x","y","z","z","y"))

# example sampling frame:
mysf<-data.frame(strata=c("a","b"),pops=c(100,1000))
clustersf<-data.frame(strata=c("x","y","z"),pops=c(50,100,200))

1. Example data & sampling frame:

Example data:

mydata

Example sampling frame:

mysf

Note that the values in mydata$strata_names all appear exactly in mysf$strata.

Calculate weights

First we generate a weighting function that matches our dataset and samplingframe. Then we use that function to calculate weights for our dataset and/or different subsets of it.

Generate weight function

To generate a weight function, we need to provide at least:

• the sampling frame
• which column in the data denote the stratum
• which columns in the sampling frame are what

So we can do:

myweighting<-weighting_fun_from_samplingframe(sampling.frame = mysf,
                                            data.stratum.column = "strata_names",
                                            sampling.frame.population.column = "pops",
                                            sampling.frame.stratum.column = "strata")

weighting are not the weights itself, but a function to generate weights.

Calculate weights:

Now to get the actual weights:

myweighting(mydata)

This is useful because we can calculate weights for subsets on the fly:

# remove NA records:
mydata_subset<-mydata[!is.na(mydata$somevar2),]

# there are now only 2 records in stratum "a":
mydata_subset

# so the weights are different:
myweighting(mydata_subset)

Combining multiple sampling frames

Assume that in addition to the above, we have a second sampling frame with populations sizes for example for cluster inside the strata:

clustersf

As above, we generate a weighting function for the cluster:

myweighting_cluster<-weighting_fun_from_samplingframe(sampling.frame = clustersf,
                                            data.stratum.column = "cluster_names",
                                            sampling.frame.population.column = "pops",
                                            sampling.frame.stratum.column = "strata")

The weights we get with myweighting_cluster are incorrect, because the cluster between strata are not proportional. We can combine the two weighting functions like this:

myweighting_combined<-combine_weighting_functions(myweighting,
                                                  myweighting_cluster)

myweighting_combined(mydata)

# in comparison: the cluster weights only:
myweighting_cluster(mydata)

# This of course still works for subsets:
myweighting_cluster(mydata_subset)

mabafaba/surveyweights documentation built on Sept. 28, 2019, 8:18 p.m.