I re-run script that previously worked with no errors about two months ago.
I used the haven package to upload an (non-public and proprietary) SPSS dataset and the survey package to analyze complex survey data.
Now, however, when I run even a simple logistic regression where both variables are dummies (coded 0 for no and 1 for yes)...something like this...
f <- read_sav("~/data.sav")
fsd <- svydesign(ids=~1, data=f, weights=~f$weight)
model <- svyglm(exclhlth~male,design=fsd,family=quasibinomial())
...I get the following errors:
Error: Must subset elements with a valid subscript vector.
x Subscript has the wrong type `omit`.
ℹ It must be logical, numeric, or character.
Run `rlang::last_error()` to see where the error occurred.
> rlang::last_error()
<error/vctrs_error_subscript_type>
Must subset elements with a valid subscript vector.
x Subscript has the wrong type `omit`.
ℹ It must be logical, numeric, or character.
Backtrace:
1. survey::svyglm(exclhlth ~ male, design = fsd, family = quasibinomial())
2. survey:::svyglm.survey.design(...)
4. survey:::`[.survey.design2`(design, -nas, )
5. base::`[.data.frame`(x$variables, i, ..1, drop = FALSE)
7. vctrs:::`[.vctrs_vctr`(xj, i)
8. vctrs:::vec_index(x, i, ...)
9. vctrs::vec_slice(x, i)
Run `rlang::last_trace()` to see the full context.
I've tried running it where I set male as a factor, and where both are set as factors. I get the same errors.
Since two months ago, I have updated R, Rstudio and both the haven and survey packages. So, I'm guessing that something changed but I am not sure what to do.
I only started transitioning from SPSS to R late last year, so I thank you in advance for any guidance and apologize in advance for newbie mistakes!
Ok, your problem seems to be that the RStudio data import functions are creating classes that hijack the subscript ([) operation. This has happened before, when RStudio switched from creating data.frame to tbl objects, but then it was sufficient to use as.data.frame() before calling svydesign().
Until a new version of the survey package is available, can you try using foreign::read.spss instead of haven::read_sav?
(Also, if you could come up with a less-confidential example and send it to the maintainer, I'm fairly sure he'd be grateful.)
Update: the issue is that the output of na.omit has class omit, and some of the variables have class haven_labelled, and the subsetting operator for haven_labelled is very fussy about the class of its arguments: it has to be plain integer or logical, without a class.
The help for the labelled class suggests using haven::as_factor or haven::zap_labels to coerce these labelled vectors to a standard R class.
Further update: I filed a github issue for the haven package, which was moved to the vctrs package, so this behaviour is likely to be changed.
Further further update: This has been fixed in the development version of vctrs