Unsub bug fix: backfile estimates are now higher

[Heather Piwowar]

Hello all, 

We've just found and fixed a bug in the Unsub forecasts that will affect most dashboards.  

The bug caused us to  *underestimate* perpetual access usage -- your dashboard will now (correctly) show more uses are fulfilled via your backfile than it did before. The direction of the bug is such that unsubscribing will look slightly more appealing than it did before.

The calculated costs are unchanged -- this bug did not affect our subscription or ILL cost calculations (other than by a dollar or two due to rounding).

This fix has changed our cost-per-use (CPU) estimates, since our CPU calculations include uses that aren't available as OA or backfile. The CPU will be higher than it was before because the subscription price is the same but now there are fewer uses that require a subscription, so the "cost per uses-that-need-a-subscription" is higher than it used to be.

To get an estimate of the scope of the change, for the demo university the backfile percentage forecast has gone up from 26% to 34%, and the updated CPU has caused a slight reordering of cost effectiveness of journals, as can be seen here (before fix, after fix) .

We take Unsub's data accuracy very seriously, and do our utmost to avoid bugs.  We will continue our practices and continue to look for chances to improve.  We are very sorry for all inconvenience this bug has or will cause.

If you have any questions or feedback about this or anything else, please don't hesitate to get in touch.

Sincerely, 

Heather and Jason

te...@ourresearch.org

ps 

Extra detail for those who are interested:  there was a place in the code for backfile calculations where we were using just downloads but we should have been using downloads+ weighted citations+ weighted authorships.

---

Heather Piwowar, cofounder
Our Research: tools to make scholarly research more open, connected, and reusable
follow at @researchremix, @our_research, @unpaywall, and @unsub_org