information from `label attribute` in R to `VARIAB

问题:

I'm working in R, but I need to deliver some data in SPSS format with both 'variable labels' and 'value labels' and I'm kinda stuck.

I've added variable labels to my data using the Hmisc's label function. This add the variable labels as a label attribute, which is handy when using describe() from the Hmisc package. The problem is that I cannot get the write.foreign() function, from the foreign package, to recognize these labels as variable labels. I imagine I need to modify write.foreign() to use the label attribute as variable label when writing the .sps file.

I looked at the R list and at stackoverflow, but I could only find a post from 2006 on the R list regarding exporting varibles labels to SPSS from R and it doesn't seem to answer my question.

Here is my working example,

# First I create a dummy dataset
df <- data.frame(id = c(1:6), p.code = c(1, 5, 4, NA, 0, 5),  
                 p.label = c('Optometrists', 'Nurses', 'Financial analysts',
                 '<NA>', '0', 'Nurses'), foo = LETTERS[1:6])

# Second, I add some variable labels using label from the Hmisc package
# install.packages('Hmisc', dependencies = TRUE)
library(Hmisc)
label(df) <- "Sweet sweet data"
label(df$id) <- "id !@#$%^" 
label(df$p.label) <- "Profession with human readable information"
label(df$p.code) <- "Profession code"
label(df$foo) <- "Variable label for variable x.var"
# modify the name of one varibes, just to see what happens when exported.
names(df)[4] <- "New crazy name for 'foo'"

# Third I export the data with write.foreign from the foreign package
# install.packages('foreign', dependencies = TRUE)
setwd('C:\\temp')
library(foreign)
write.foreign(df,"df.wf.txt","df.wf.sps",  package="SPSS")

list.files()
[1] "df.wf.sps" "df.wf.txt"

When I inspect the .sps file (see the content of 'df.wf.sps' below) my variable labels are identical to my variable names, except for foo that I renamed to "New crazy name for 'foo'." This variable has a new and seemly random name, but the correct variable label.

Does anyone know how to get the label attributes and the variable names exported as 'variable labels' and 'labels names' into a .sps file? Maybe there is a smarter way to store 'variable labels' then my current method?

Any help would be greatly appreciated.

Thanks, Eric

Content of 'df.wf.sps' export using write.foreign from the foreign package

DATA LIST FILE= "df.wf.txt"  free (",")
/ id p.code p.label Nwcnf.f.  .

VARIABLE LABELS
 id "id" 
 p.code "p.code" 
 p.label "p.label" 
 Nwcnf.f. "New crazy name for 'foo'" 
 .

VALUE LABELS
/
p.label  
 1 "0" 
 2 "Financial analysts" 
 3 "Nurses" 
 4 "Optometrists" 
/
Nwcnf.f.  
 1 "A" 
 2 "B" 
 3 "C" 
 4 "D" 
 5 "E" 
 6 "F" 
.

EXECUTE.

Update April 16 2012 at 15:54:24 PDT;

What I am looking for is a way to tweak write.foreign to write a .sps file where this part,

[…] 

VARIABLE LABELS
 id "id" 
 p.code "p.code" 
 p.label "p.label" 
 Nwcnf.f. "New crazy name for 'foo'" 

[…] 

looks like this,

[…] 

VARIABLE LABELS
 id "id !@#$%^" 
 p.code "Profession code" 
 p.label "Profession with human readable information" 
 "New crazy name for 'foo'" "New crazy name for 'foo'" 

[…]

The last line is a bit ambitious, I don't really need to have a variables with white spaces in the names, but I would like the label attributes to be transferred to the .spas file (that I produce with R).

回答1:

Try this function and see if it works for you. If not, add a comment and I can see what I can do as far as troubleshooting goes.

# Step 1: Make a backup of your data, just in case
df.orig = df
# Step 2: Load the following function
get.var.labels = function(data) {
  a = do.call(llist, data)
  tempout = vector("list", length(a))

  for (i in 1:length(a)) {
    tempout[[i]] = label(a[[i]])
  }
  b = unlist(tempout)
  structure(c(b), .Names = names(data))
}
# Step 3: Apply the variable.label attributes
attributes(df)$variable.labels = get.var.labels(df)
# Step 4: Load the write.SPSS function available from
# https://stat.ethz.ch/pipermail/r-help/2006-January/085941.html
# Step 5: Write your SPSS datafile and codefile
write.SPSS(df, "df.sav", "df.sps")

The above example is assuming that your data is named df, and you have used Hmisc to add labels, as you described in your question.

Update: A Self-Contained Function

If you do not want to alter your original file, as in the example above, and if you are connected to the internet while you are using this function, you can try this self-contained function:

write.Hmisc.SPSS = function(data, datafile, codefile) {
  a = do.call(llist, data)
  tempout = vector("list", length(a))

  for (i in 1:length(a)) {
    tempout[[i]] = label(a[[i]])
  }
  b = unlist(tempout)
  label.temp = structure(c(b), .Names = names(data))
  attributes(data)$variable.labels = label.temp
  source("http://dl.dropbox.com/u/2556524/R%20Functions/writeSPSS.R")
  write.SPSS(data, datafile, codefile)
}

Usage is simple:

write.Hmisc.SPSS(df, "df.sav", "df.sps")

回答2:

The function that you linked to (here) should work, but I think the problem is that your dataset doesn't actually have the variable.label and label.table attributes that would be needed to write the SPSS script file.

I don't have access to SPSS, but try the following and see if it at least points you in the right direction. Unfortunately, I don't see an easy way to do this other than editing the output of dput manually.

df = structure(list(id = 1:6, 
               p.code = c(1, 5, 4, NA, 0, 5), 
               p.label = structure(c(5L, 4L, 2L, 3L, 1L, 4L), 
                                   .Label = c("0", "Financial analysts",
                                              "<NA>", "Nurses", 
                                              "Optometrists"), 
                                   class = "factor"), 
               foo = structure(1:6, 
                               .Label = c("A", "B", "C", "D", "E", "F"), 
                               class = "factor")), 
               .Names = c("id", "p.code", "p.label", "foo"),
          label.table = structure(list(id = NULL,
                             p.code = NULL,
                             p.label = structure(c("1", "2", "3", "4", "5"),
                                      .Names = c("0", "Financial analysts", 
                                                 "<NA>", "Nurses", 
                                                 "Optometrists")),
                             foo = structure(1:6, 
                                  .Names = c("A", "B", "C", "D", "E", "F"))),
                             .Names = c("id", "p.code", "p.label", "foo")),
          variable.labels = structure(c("id !@#$%^",  "Profession code", 
                                 "Profession with human readable information",
                                 "New crazy name for 'foo'"), 
                            .Names = c("id", "p.code", "p.label", "foo")), 
          codepage = 65001L)

Compare the above with the output of dput for your sample dataset. Notice that label.table and variable.labels have been added, and a line that said something like row.names = c(NA, -6L), class = "data.frame" was removed.

Update

NOTE: This will not work with the default write.foreign function in R. To test this you first need to load the write.SPSS function shared here, and (of course), make sure that you have the foreign package loaded. Then, you write your files as follows:

write.SPSS(df, datafile="df.sav", codefile="df.sps")