CODEWITHSUNDEEP
rsurveysampling
Roland Kofler
Roland Kofler

Reputation: 1332

drawing a stratified sample in R

Designing my stratified sample

library(survey)
design <- svydesign(id=~1,strata=~Category,  data=billa, fpc=~fpc)

So far so good, but how can I draw now a sample in the same way I was able for simple sampling?

set.seed(67359)  
samplerows <- sort(sample(x=1:N, size=n.pre$n))

Upvotes: 4

Views: 8533

Answers (5)

Thomas Lumley
Thomas Lumley

Reputation: 2765

While it's true that for more complicated sampling the sampling package is preferable, there's is actually a function stratsample in the survey package to do stratified sampling.

Upvotes: 0

mpalanco
mpalanco

Reputation: 13570

You can draw a stratified sample using dplyr. First we group by the column or columns in which we are interested in. In our example, 3 records of each Species.

library(dplyr)
set.seed(1)
iris %>%
  group_by (Species) %>%
  sample_n(., 3)

Output: 

Source: local data frame [9 x 5]
Groups: Species

  Sepal.Length Sepal.Width Petal.Length Petal.Width    Species
1          4.3         3.0          1.1         0.1     setosa
2          5.7         3.8          1.7         0.3     setosa
3          5.2         3.5          1.5         0.2     setosa
4          5.7         3.0          4.2         1.2 versicolor
5          5.2         2.7          3.9         1.4 versicolor
6          5.0         2.3          3.3         1.0 versicolor
7          6.5         3.0          5.2         2.0  virginica
8          6.4         2.8          5.6         2.2  virginica
9          7.4         2.8          6.1         1.9  virginica

Upvotes: 3

Anthony Damico
Anthony Damico

Reputation: 6104

here's a quick way to sample three records per distinct 'carb' value from the mtcars data frame without replacement

# choose how many records to sample per unique 'carb' value
records.per.carb.value <- 3

# draw the sample
your.sample <- 
    mtcars[ 
        unlist( 
            tapply( 
                1:nrow( mtcars ) , 
                mtcars$carb , 
                sample , 
                records.per.carb.value 
            ) 
        ) , ]

# print the results to the screen
your.sample

note that the survey package is mostly used for analyzing complex sample survey data, not creating it. @Iterator is right that you should check out the sampling package for more advanced ways to create complex sample survey data. :)

Upvotes: 2

Iterator
Iterator

Reputation: 20560

Take a look at the sampling package on CRAN (pdf here), and the strata function in particular.

This is a good package to know if you're doing surveys; there are several vignettes available from its page on CRAN.

The task view on "Official Statistics" includes several topics that are closely related to these issues of survey design and sampling - browsing through it and the packages recommended may also introduce other tools that you can use in your work.

Upvotes: 4

Andrie
Andrie

Reputation: 179398

If you have a stratified design, then I believe you can sample randomly within each stratum. Here is a short algorithm to do proportional sampling in each stratum, using ddply:

library(plyr)
set.seed(1)
dat <- data.frame(
    id = 1:100,
    Category = sample(LETTERS[1:3], 100, replace=TRUE, prob=c(0.2, 0.3, 0.5))
)

sampleOne <- function(id, fraction=0.1){
  sort(sample(id, round(length(id)*fraction)))
}

ddply(dat, .(Category), summarize, sampleID=sampleOne(id, fraction=0.2))

   Category sampleID
1         A       21
2         A       29
3         A       72
4         B       13
5         B       20
6         B       42
7         B       58
8         B       82
9         B      100
10        C        1
11        C       11
12        C       14
13        C       33
14        C       38
15        C       40
16        C       63
17        C       64
18        C       71
19        C       92

Upvotes: 4

Related Questions