2

What I would like to achieve is get a 10% sample from each group (which is a combination of 2 factors - recency and frequency category). So far I have thought about package sampling and function strata(). Which looks promising but I am getting the following error and it is really hard to understand the error message and what is wrong or how to get around this.

Here is my code:

> d[1:10,]
        date id_email_op recency frequecy r_cat f_cat
1  29.8.2011       19393     294        1     A     G
2  29.8.2011       19394     230        4     A     D
3  29.8.2011       19395     238       12     A     B
4  29.8.2011       19396     294        1     A     G
5  29.8.2011       19397     223        9     A     C
6  29.8.2011       19398     185        7     A     C
7  29.8.2011       19399     273        2     A     F
8  29.8.2011       19400      16        4     C     D
9  29.8.2011       19401     294        1     A     G
10 29.8.2011       19402       3        5     F     C
> table(d$f_cat,d$r_cat)

         A      B      C      D      E      F
  A    176    203    289    228    335    983
  B   1044    966   1072    633    742   1398
  C   6623   3606   3020   1339   1534   2509
  D   4316   1790   1239    529    586    880
  E   8431   2798   2005    767    817   1151
  F  22140   5432   3937   1415   1361   1868
  G 100373  18316  11872   3760   3453   4778
> as.vector(table(d$f_cat,d$r_cat))
 [1]    176   1044   6623   4316   8431  22140 100373    203    966   3606   1790   2798   5432
[14]  18316    289   1072   3020   1239   2005   3937  11872    228    633   1339    529    767
[27]   1415   3760    335    742   1534    586    817   1361   3453    983   1398   2509    880
[40]   1151   1868   4778
> s <- strata(d,c("f_cat","r_cat"),size=as.vector(ceiling(0.1 * table(d$f_cat,d$r_cat))), method="srswor")
Error in strata(d, c("f_cat", "r_cat"), size = as.vector(table(d$f_cat,  : 
  not enough obervations for the stratum 6

I cant really see what is stratum 6. What is the condition the function checks in background? I am not sure I that I have the size param set up correctly. And yes I have checked the documentation of sampling package :)

Thanks everyone and

Radim Sevcik
  • 21
  • 1
  • 3
  • 1
    Just so we eliminate the issue of sample sizes having fractional values, can you replace `size=as.vector(table(d$f_cat,d$r_cat))*.1` with `size=as.vector(ceiling(0.1 * table(d$f_cat,d$r_cat)))`? – Iterator Sep 01 '11 at 14:15

2 Answers2

1

You could always do it yourself: 

stratified <- NULL
for(x in 1:6) {
  tmp1 <- sample(rownames(subset(d, r_cat == "A" & f_cat == LETTERS[x])),round(nrow(d[r_cat == "A")*0.1))
  tmp2 <- sample(rownames(subset(d, r_cat == "B" & f_cat == LETTERS[x])),round(nrow(d[r_cat == "B")*0.1))
  tmp3 <- sample(rownames(subset(d, r_cat == "C" & f_cat == LETTERS[x])),round(nrow(d[r_cat == "C")*0.1))
  tmp4 <- sample(rownames(subset(d, r_cat == "D" & f_cat == LETTERS[x])),round(nrow(d[r_cat == "D")*0.1))
  tmp5 <- sample(rownames(subset(d, r_cat == "E" & f_cat == LETTERS[x])),round(nrow(d[r_cat == "E")*0.1))
  tmp6 <- sample(rownames(subset(d, r_cat == "F" & f_cat == LETTERS[x])),round(nrow(d[r_cat == "F")*0.1))
  tmp7 <- sample(rownames(subset(d, r_cat == "G" & f_cat == LETTERS[x])),round(nrow(d[r_cat == "G")*0.1))
  stratified <- c(stratified,tmp1,tmp2,tmp3,tmp4,tmp5,tmp6,tmp7)
}

And then...

d[stratified,] would be your stratified sample.

Brandon Bertelsen
  • 43,807
  • 34
  • 160
  • 255
0

Problem SOLVED! enter image description here

This sentence "the sampling frame is stratified by region within state" helped me! If you use more than one variables for stratification, the you must pay attention to the "order" of these variables when you assign different sizes to the agument "size=". The more strata a variable has, the higher priority it has, therefore, the one which has the most strata should be on top of the list when you use "table()".

I hava 10 groups in GENDER, and 2 groups in age.group, so this WON'T work

 nnum <- as.vector(table(d.order$GENDER,d.order$age.group))

But this works

    d.order <- d.cut[order(d.cut$age.group,d.cut$GENDER),]
nnum <- as.vector(table(d.order$age.group, d.order$GENDER))
    n <- round(.05*nnum)
    testData <- strata(d.order, stratanames=c("age.group","GENDER"),size=n,method="srswor")

I highly recommend that you change the order of your variables accordingly throughout the whole script besides the function table() or ceilling(). It solved my problem, hopefully it will solve yours too. :)

Timomtt
  • 1