The Stanford Copyright Renewal Database has made its final report to the Hewlett Foundation available. It shows an accuracy of over 99%. From the Copyright Renewal Database site:
Stanford has performed two rounds of testing in order to assess the accuracy of this database. In each round, we pulled a minimum of 500 book titles published in the US between 1923 and 1963 from the Stanford library catalog. The works were checked manually in the CCE, and, in the first round, a subset of 100 records was also sent to the Copyright Office to be checked by their in-house staff. Each of these items was then separately searched by project staff in the Copyright Renewals Database. In each round, the error rate for the database was found to be less than 1%, although in practice there is significant opportunity for user error or other problems in searching. Details of these issues can be found in Stanford's final report to Hewlett on the project ( PDF), and in the two search results sets ( Search Set #1, Search Set #2).
Hat tip to Mimi Calter for letting me know.
Comments