The last few days we've been looking for some patterns in how holdings change in WorldCat records. As libraries catalog materials on OCLC they can indicate that they 'hold' the item. The total number of these holdings just passed 1,000,000,000.
Part of the justification for this was to see if we could do a better job at selecting a particular manifestation of a work-set to represent the whole set. This is sort of a 'poor mans' FRBR, and what we are currently doing this for the records that Yahoo harvests from us--each one is the highest held record of one of the 3,000,000 most popular works in WorldCat. It's not perfect, but it was quick, and works surprisingly well. Once you find a record in Yahoo (or Google or anywhere else) the Open WorldCat pages have links to other records we have identified as being part of the same work-set.
We have counts for the number of holdings for the records we had seven years ago in 1998. 61% of the records now in the database have gained holdings since then, 4% have lost holdings, and 35% of the record holdings counts remained the same. Just looking at the changes in the records from 1998, excluding more recent records, the figures are: 42% gained holdings, 6% lost, and 52% remained constant.
Interesting to note that the highest OCLC number in 1998 was just under 40 million (as opposed to more than 60 million now). Over the last eight years, WorldCat has grown by 50%. Nearly 20% of WorldCat's 1,004,000,000+ holdings are on records entered in the last seven years and about one-third of all holdings have been added in that time (WorldCat went online 34 years ago).
One of the things we looked at are the records with the most holdings lost. Of the 100 records that lost the most holdings, only 2 were not the result of the record being merged with another (we do this regularly as part of our duplicate detection process).
We had the idea that the record in a work-set that has gained the most holdings recently might be a good record to point to. It turns out that most of these are electronic versions, typically from NetLibrary. This is great if your library has a NetLibrary account, but otherwise not particularly helpful.
Thanks to Ed O'Neill and Rick Bennett who saved the holdings data in an accessible form, and Jenny Toves who processed the data.