Small change to prevent converting literal "NA"s as missing.

[jmobrien]

Set type_convert() to use na = character() to avoid converting anything further to NA (all suitable NAs should have been identified by the read_csv() call just above.)

[juliasilge]

I wonder if it would be better to use the dots ... here for all arguments passed to type_convert(), and then we make ... on fetch_survey() for most of the arguments that pass through to read_survey() like col_types, etc.

[jmobrien]

Yeah, I had similar general thoughts, but I think a fixed argument works best for this specific issue. Looks like the endpoint only ever offers up "" for truly missing data (ignoring seen-but-missing). Unless I'm reading this wrong, setting those ""'s to NA is already covered by the read_csv() call on line 115. So, adding na = character() to type_convert just eliminates any extra missingness checks, which I think is what we want.

Considering the potential for something like ..., the potentially-flexible args for type_convert are col_types, locale, and trim_ws. We have col_types already, but only partly support locale through time_zone, and nothing for trim_ws. ... could offer consolidation and add support for whitespace removal. However, it's worth noting that the central role for the standalone time_zone arg in fetch_survey() is as a param for the API query.

So, I'm not sure. Though, I am thinking this PR is just a bugfix that can be considered separately.

[juliasilge]

Thank you so much @jmobrien! 🙌