Post #2785736
2025-04-02 13:53 UTC
Problem: when negotiating OA agreements, libraries and publishers need to the predict number of publications in future years. Current ways of averaging or trendlines can be inaccurate, causing credits to run out mid-year.
Our model takes 5 years of journal-level corresponding authored pub counts and predicts the distr of articles in year n+1 using Poisson, pooling info across journals. Article counts are by jnl-year, then summed to predict a total distr with uncertainty beyond a point estimate.
Replies (1)
-
@eschares@scholar.social 2025-04-02 13:54
We test on three real datasets of counts from ISU and show more accurate predictions for one of the case studies and similar for a second. For the 3rd, the actual count was significantly lower than any previous year, and no method predicts well.