OpenAlex Funding Data Inconsistency

During my work for the KBOPENBIB project, I came across the following inconsistency in OpenAlex funding data which might be of interest to the wider community: the number of works per funder doesn’t always match up between works and funders object.

# query list of all funders, including works_count field
funders <- oa_request(query_url = "https://api.openalex.org/funders")
df <- oa2df(funders, entity = "funders")

# query works list grouped by funders
df2 <- oa_fetch(entity = "works", group_by = "grants.funder")

You would expect that the number in the works_count field of the funder object to match the count you get by counting unique works ids per funder id in the works table.

Agreement	Number of funders	p
❌	8025	24.74%
✔️	24412	75.26%

Table 1: Agreement of funded publications per funder between works_count and “manual” count.

For roughly a quarter of funders the publication counts do not match up. The mean difference between works_count and manual count is 10.26, so works_count field is missing 10 publications on average. However, if we remove the National Natural Science Foundation of China from the data — which has a whopping number of 343744 missing publications — the mean drops down to -0.34.

Table 2: Difference in count methods

Looking at funders with diverging publication counts in Table 2 we see that for a majority of funders the difference is only one publication.

I hope that this problem description contributes to the continuous improvement of OpenAlex. Until this inconsistency is addressed, I recommend to “manually” count work ids per funder instead of the works_count field.