Deeper Dive into Estimating BTAA Sociology Serials Holdings with WMS APIs, Z39.50, and Spreadsheets

Two years ago, my colleague Stephen Woods approached me about collaborating on an article¹ extending research he’d already performed about serials use in doctoral sociology work. He and another colleague, John Russell, had developed a methodology for determining “CDRank” based on the number of times a journal was citated across a dissertation dataset and the number/% of dissertations it was cited in.²

On his sabbatical, Stephen had mined citations in 518 sociology dissertations from Big Ten schools. He planned to perform a CDRank analysis and determine the most influential journals by school and see where they overlapped (or didn’t). He had a spreadsheet of titles and ISSNs for the 5,659 distinct journals cited and a related question: What did holdings for these look like across the Big Ten Libraries?

He was interested in whether the highest-ranked journals were more universally held, whether there were any noticeable gaps, basically would any patterns emerge if we looked at the Big Ten’s holdings for these journals. And then, at an institution-by-institution level, were any of the most-used journals for that institution not held by the library?

As anyone who works with it knows, holdings data is notoriously difficult. But I was interested in it as a challenge: could I combine a variety of resources to come up with a reasonable assessment of which libraries had some holdings of the serial in question?

Obtaining Library Holdings: The Summary

The journal article was focused on the outcome, so it wasn’t a place for me to write a deep dive of the process I used for identifying holdings. This is the summarized version from the article:

The 57,777 citations were condensed to a list of 5,659 distinct journal title/ISSN entries. Holdings data from across the BTAA was then queried to determine the extent to which these journals are held by BTAA institutions. It was first necessary to obtain all representative ISSNs for each journal. The WorldCat Metadata API 2.0 was queried by ISSN and results were processed to identify additional ISSNs. These additional ISSNs were used in subsequent queries. During this process, 25 titles were identified that did not include ISSNs and the list was reduced to 5,634 unique pairings.

Holdings data was obtained from the WorldCat Search API v.2 and Z39.50 services. First, the WorldCat Search API’s bibliographic holdings endpoint was queried by ISSN and results limited to a list of OCLC symbols representing the BTAA institutions. However, an institution’s WorldCat holdings may not be up-to-date and are unlikely to represent electronic-only items. In the second step, MarcEdit software was used to query each institution’s Z39.50 service by ISSN for any of the 5,634 entities not found at that institution during the WorldCat API phase. This combined holdings data was saved to a JSON database.

In limitations, I addressed some of the challenges I ran into:

Holdings represent those found in WorldCat and respective library ILSes during November 2023. Several factors limit the effectiveness of representing libraries’ journal holdings. Coverage is not recorded in ways which can be easily machine-parsed at scale to determine whether holdings represent part or all of a publication run. E-journal records are often updated on a monthly basis, resulting in varying results by month. Additionally, if a library does not have sufficient staffing to perform updates, their WorldCat holdings statements may not reflect recent weeding. The presence of WorldCat holdings or of a record in the library’s ILS (queried by Z39.50) indicates, at minimum, that the library has held some coverage of this journal at some point in time.

University of Nebraska-Lincoln’s Z39.50 documentation was not available online and email inquiries were not answered, so the Z39.50 phase could not be run. Gaps in Nebraska’s WorldCat holdings for the combined list of top 161 journals were manually queried by title and ISSNs using the library’s Primo discovery search. As indicated in Results, all but two of these journals were found.

Obtaining Library Holdings: The Whole Story

Even for our original intended audience at Serials Review,³ a full writeup would’ve been too deep a dive (buckle in, this is 2500 words), but I really enjoyed (and hated) the challenge of figuring out how to even tackle the project and solving problems along the way (except when I tore my hair out). So I thought I’d share it here.

I need to preface by noting again that my research question was not whether an institution had complete holdings or precisely which holdings they had. It’s challenging to do that at the scale of one’s own institution. My question was:

Does this institution appear to hold print or electronic versions of some segment of this serial?

Processing ISSNs for Siblings, Mostly

First, I evaluated my starting data. I had journal titles and ISSNs. A quick check confirmed my hypothesis that some were for print materials and some were for e-journals. I wanted to check for both kinds, of course.

Because library records don’t yet have rich cluster ISSNs and I didn’t have a API subscription to the ISSN Portal,⁴ I decided to use the next best thing – WorldCat. I searched brief bibs in the WorldCat Search API v.2 using the ISSN to obtain all records. I used a function to run through list of ISSNs in a brief bib, clean it up if needed, and append all ISSNs found to a new list. So my output was the title, original ISSN, and a list of all ISSNs found.

{ "title": "journal of loss and trauma",
"original_issn": "1532-5024",
"all_issns": ["1532-5024", "1532-5032"] }

Challenges

The first challenge I ran into was that I was relying on data which came from a field originally intended for recording misprints. However, it had become repurposed to do what I wanted – record the ISSN for the other (print / electronic) version of a serial. Frustration with this dual purpose sparked my MARC Misconceptions post re: the 022$y which explores the problem further. After some rabbit holes attempting to find ways to identify these and some spot checks to identify how often the problem was happening, I finally accepted that the holdings data was just going to be generally accurate. I also decided that I would allow for continuations if they showed up in the ISSN data because when a record had more than 2 ISSNs, my spot checking determined it was almost always for one or more continuations vs. another work entirely.

A more concrete problem was that sometimes ISSN data was recorded with hyphens and sometimes it wasn’t. Sometimes it even contained parentheticals. I developed some rather complex processing logic, including regular expressions and substring slices, to turn a field into just the ISSN, formatted as 1234-5678. Using Regex, I reviewed my data and manually corrected the few errors, most of which were caused by a cataloger typoing a 7-digit ISSN, e.g. 234-5678 and 1234-567.

I also used this phase to manually review a small handful of ISSNs which showed up in two records. In most cases, they were for the same serial. The original citations had used slight title variants (author error) and then the print and electronic ISSNs (natural variance), leading to a database duplication. A few were continuations. I resolved all of these, referring to the ISSN Portal when needed. I also informed Stephen so he could recalculate CD Rank of the merged records.

25 of the journals in the original dataset simply had no ISSNs. While I was running my big scripts to gather sibling ISSNs, I used the ISSN Portal to confirm that they really had no ISSN. Fortunately, all had extremely low CDRanks representing one-off citations.

Querying Holdings: WorldCat

Next, I needed to get actual holdings. The one place I could think of to get holdings in aggregate was, again, WorldCat. I used the bibs holdings API for this one.

First, I created a list of the institutional identifiers for each school. For each record in my database, I ran the list of its potential ISSNs (most often just a pair) through the API using the “heldBySymbol” limiter and grabbed a list of institutions with some holding for this ISSN. It output these to a JSON file/database of records consisting of: title, the original ISSN, the list of ISSNs, the list of holding institutions.

{ "title": "journal of loss and trauma",
"original_issn": "1532-5024",
"holdings": ["MNU", "UPM","IPL","UMC","OSU","GZM","NUI","LDL","IUL","EYM","EEM"],
"all_issns": ["1532-5024", "1532-5032"] }

However, my years of experience working in cataloging departments and with library data meant I also know that WorldCat holdings are unreliable. Worst case for this research, the institution had weeded the journal and not updated their holdings. But, conversely, they likely didn’t provide holdings information for their e-journal records.

Sampling the results I got at this phase, I knew I wasn’t getting the whole picture…

Querying Holdings: Z39.50

So far, I’d been able to work on the whole thing as a single batch – one big batch of ISSN sibling hunts, one big batch where I queried all the library codes at once.⁵ But now, it was time to get targeted.

I wrote a script to check each each entry in the database for which institutions were not present. It wrote all the ISSNs from these entries to a line break-separated text file of ISSNs. I saved these by symbol, so UPM.txt, EEM.txt, etc. Some of these files were 3000 ISSNs long (but keep in mind that, in most cases, several ISSNs represent the same journal).

I then used MarcEdit to query each institution’s Z39.50 for the whole thing.

Now, in addition to writing MARC files, MarcEdit provides a handy log of your query and results:

Searching on: 1942-4620 using index: ISSN
0 records found in database University of Michigan
Searching on: 1942-4639 using index: ISSN
0 records found in database University of Michigan
Searching on: 1942-535X using index: ISSN
1 records found in database University of Michigan

I saved these as text files⁶ and then ran a Python script over them to process the search key and results. It read through each pair of lines, made a dict of the results {"1942-4620" : 0, "1942-4639" : 0, "1942-535X" : 1}, then opened the JSON database and updated the holdings. I used an “if value not in” check so that an entry’s holdings would only update once even if the Z39.50 output matched 3 sibling ISSNs from that entry.

…this was one of those moments in coding where you feel like an utter genius but worry that you might be a little unhinged as well.

Querying Holdings: Shared Print

In some cases, the reason an institution didn’t have a journal any more was that they’d contributed it to the BTAA Shared Print Repository. This project specifically targeted journals, so it was entirely possible that one of these institutions had eased its shelf space by sending a run to Shared Print.

Using my contacts at the BTAA, I got emails for the people at Indiana and Illinois who actually managed the projects. Fortunately, both had holdings spreadsheets, including ISSNs, and were willing to share them.

I wrote a Python script to take these (as CSVs) and check for the presence of an ISSN in each spreadsheet. If it found the ISSN, it would write the OCLC library code (UIUSP or IULSP) to the output database. I wrote and ran this while gathering Z39.50 data, since that took several weeks.

This turned out to be a non-issue for the overall project, since almost all of the top 161 journals were held at all the institutions. If contributions to shared print were partly judged on the basis of usage, this would make sense. Still, it might be interesting to look at shared print coverage of the database as a whole.

Minding the Gaps

There was one big gap in the whole thing – University of Nebraska-Lincoln. They had somewhat recently migrated to Alma, their systems librarian at left when they migrated, and they had not yet filled the position. I contacted several people there asking about Z39.50 access for a research project but didn’t hear anything. (Fortunately, they’ve now got a new and engaged systems librarian whom I met at ELUNA this year.)

Anyway, this posed a challenge. If they had Z39.50 turned on, it wasn’t showing up in any of the ways I could think of. I made several attempts, mimicking the many other Alma schools I had queried. Nothing worked.

By this point, we had a combined list of the top 161 journals. We also had partial holdings data for Nebraska from the WorldCat query. So I sat down and did this one manually. I think I searched ~30 journals by hand in their Primo front-end, using advanced search by ISSN and then by title/material type if the ISSN didn’t come up (and then double check to ensure it was the right journal). I marked all the ones I found and used this data to update the top 161.

Because there weren’t many, I decided to be as thorough as possible and manually check each institution’s catalog/discovery/e-journal finders for remaining gaps in the top 161.

Observations

In some ways, my findings were not very exciting: BTAA schools widely hold (at least some of) the journals most commonly used in sociology dissertations. Either that or the commonality of these holdings means that they’re the most widely used. (But many others were just as widely held and not as widely used, so I suspect the former, with internal usage data playing a role in retention.)

Ultimately, my process got far more data than we actually used. I could’ve just run the queries for the top 161. That would’ve been a much smaller project and I could’ve thoroughly validated my results. For example, I would’ve checked any instances where the ISSN list contained more than 2, using the ISSN Portal to be sure these were cases of a journal continuation vs. an actual incorrect ISSN. But when we started, Stephen was still working on his own analysis of the data. And while an enormous job, this yielded a really interesting database of results, something I might be able to revisit in the future. It was also a fascinating challenge.