Rounding bug in variable info from RxSpssData

Question

I have discovered what I think is a bug in the way MicrosoftR handles metadata from .sav files from SPSS.

Here is a summary of the Variable View:

ColumnA: 1 - Yes, 2 - No
ColumnB: 0.33 - Yes, 0.5 - Maybe, 0.66 - No, 0.99 - Why not, 1.00 - Yes, for sure.
ColumnC: A - Yes, B - No

My code:

library(RevoScaleR)

df <- RxSpssData(
  "RoundingTest.sav", 
  stringsAsFactors = FALSE, 
  labelsAsInfo = TRUE, 
  labelsAsLevels = TRUE,
  mapMissingCodes = "none" 
)

test = rxImport(df)

The data reads in fine:

  ColumnA ColumnB ColumnC Var0001
1     Yes    0.33     Yes      NA
2      No    0.50     Yes      NA
3     Yes    0.66      No      NA

The valueInfoCodes, however, do not:

attr(test$ColumnA, ".rxValueInfoCodes") # NULL
attr(test$ColumnB, ".rxValueInfoCodes") # "0" "0" "0" "0" "1"
attr(test$ColumnC, ".rxValueInfoCodes") # NULL

It seems to be using some sort of floor function on the metadata in the numeric columns before converting them to character strings.

I have tried using options(scipen = 12) and rxOptions(numDigits = 12) with no success. Using rxDataStep instead of rxImport doesn't work. I believe the error to be somewhere in the RxSpssData() function.

Has anyone experienced this with RxSpssData, or any other file type?
Is there a workaround?
Is there a formal way to report it to Microsoft if it's a genuine bug?

Thanks!

Also, my session info:

R version 3.3.2 (2016-10-31)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows >= 8 x64 (build 9200)

EDIT: Added the sample SAV file to a GitHub repo here.

Can you use haven and do an equivalent process to see what the results are? — Steph Locke, Apr 19 '17 at 18:16
I've used a combination of libraries like `Rz`, `foreign` and `memisc` prior to this. I was hoping to get it working on MicrosoftR for a few reasons, but alas. — Oliver Frost, Apr 19 '17 at 18:31
You can - but it's good to confirm the issue is definitely with the microsoft bit — Steph Locke, Apr 19 '17 at 18:31
I've had no issues with other packages, if that's what you're asking — Oliver Frost, Apr 19 '17 at 18:41
In fact, I've tried `haven` tonight for the first time and found it to be a nicely put together package. I seem to be getting good results with everything except `RevoScaleR` — Oliver Frost, Apr 19 '17 at 22:30
Well that's good(ish)! Glad to see the github repro. Will look-see tomorrow — Steph Locke, Apr 19 '17 at 23:05
MS don't seem to be accepting bugs for Microsoft R right now, according to this: https://connect.microsoft.com/directory/accepting-bugs — Oliver Frost, Apr 23 '17 at 13:04
It might be worth posting to https://social.msdn.microsoft.com/Forums/en-US/home?forum=MicrosoftR as well — Bob Albright, Apr 23 '17 at 15:47
How `attr(dat$ColumnB, ".rxValueInfoCodes")` result should look like? — gonzalez.ivan90, Apr 24 '17 at 13:07

Rounding bug in variable info from RxSpssData

0 Answers0