fsingletonthorn
commented
5 years ago Closed with commit 67e8daa, further updates maybe required as the specificity is not great ATM (i.e., it extracts a lot of things which are not the final sample size -- although given how differently folk tend to report n, this may be inevitable w/out machine learning etc.), just going to test using the largest extracted value as the approximate sample size - see how well that works
Extract all mentions of sample size (e.g., "sample size of x", "x participants", "N = x", and then figure out some way of approximating the actual sample size with acceptable error, e.g., using the largest reported N (clearly error prone), the first N, an average of all Ns... Will have to pilot carefully.