Decide which papers to keep

[tklebel]

In our sample, we have a sizeable share that has OA status "unknown", and also some that are closed access.
What to do? investigate, exclude?

[tklebel]

fractional_works %>% count(is_oa)
# # Source: spark<?> [?? x 2]
#   is_oa       n
#   <lgl>   <int>
# 1 FALSE   41532
# 2 TRUE  6694172
# 3 NA     369903

Potential reasons: those that are not OA might be errors in OpenAlex/crossref data in terms of publication date, e.g. the article having a publication date of 2020, when in fact it was originally published in 2005, when the journal was not OA yet.

Articles withi OA status NA might be those that are simply not covered by Unpaywall. However, they might still be OA (given that they are in an OA journal),

So there is likely a trade-off: only using the one's with OA flag == TRUE seems the most stable option, but it might introduce bias or lower precision, since we would be excluding content from certain publishers from certain regions.

One option would be to do small random samples from across the three categories and check some articles.

[tklebel]

For now, I'm keeping only the OA ones, since the potential for bias from excluding is likely smaller than the bias from including papers from wrong years or similar

[tklebel]

confirmed by Tony that this is the best approach, but will mention in discussion that the exclusion of NA might introcude bias