In May 2000, JSTOR made article-level linking tools available to its participating libraries, publishers, and scholars (http://makealink.jstor.org/). These tools are designed to facilitate linking from the various resources used by our constituents to full-text articles in the JSTOR archive. When we began to design these tools, little was known about how, or even whether, our participants would use them. There had been much discussion in the community about linking between resources, but little substantial activity. Now, three years later, it appears that there was indeed real interest and real need. Since we have made linking available, nearly 200 participating libraries and publishers, as well as nearly 1,000 individual faculty, researchers, and scholars, have registered to use the JSTOR linking tools to build links from resources such as online catalogs, course syllabi, and personal bibliographies. Increasingly, authorized users at JSTOR participating institutions are following the links to the full-text articles in the various JSTOR collections.

During the past three years we have also established linking relationships with a number of important database providers, including ABC-CLIO, EBSCO Publishing, Cambridge Scientific Abstracts, OCLC, and ProQuest. Library system vendors, such as ExLibris and Endeavor, have enabled JSTOR links in their linking software programs and services. We have partnered with several JSTOR participating publishers to provide article-level links from their sites to JSTOR. For a current list of linking partners, please visit our website at: http://www.jstor.org/resources/linking.partners.html.

Using SICIs?

Maintaining a high percentage of valid links into the JSTOR archive is a very important goal in our linking efforts. Using Serial Item and Contribution Identifiers (SICIs) to create direct URLs to our articles has been invaluable in maintaining a high percentage of quality links. SICI is an ANSI/NISO standard and was developed to uniquely identify serial items (issues) and their contributions (articles). It is also compatible with Digital Object Identifiers (DOI), which are used by many publishers to uniquely identify journal articles. SICIs are very useful for linking purposes because they are based upon citation information. Currently, a SICI can be used to link to any article, issue or volume within the JSTOR archive. We were also concerned about whether links to JSTOR created by outside resources would consistently work. The quality of links from these resources is dependent on how well their metadata matches JSTOR's. However, there is no single standard for configuring metadata and inconsistencies occur that can lead to the formation of invalid links. Inconsistent metadata can lead to a user not linking to the appropriate article. This is especially a concern given the idiosyncratic nature of the content in the JSTOR archive. Creating reliable and persistent links to articles that are sometimes hundreds of years old can be difficult - especially since some of those very old articles were not always published using the traditional serials nomenclature of volumes and issues.

SICIs help to mitigate the problem of invalid links. In order for outside organizations to successfully maintain links to JSTOR articles, the links must remain stable and consistent no matter how the JSTOR data is stored. Typically, URLs can and do change with the architecture of the serving site and therefore are not good identifiers of digital objects. By using SICIs, JSTOR creates permanent identifiers that are derivable from the citation information of the target articles.

Reviewing the Data: Early Trends?

With every link into the JSTOR archive, we capture important statistics. Where did the link originate? Was the link properly formed? Did the link resolve successfully? As more participating libraries, publishers, and scholars have initiated links into JSTOR, and as more external resource providers have begun linking their databases and content to the archive, the volume of linking has grown considerably since 2001. The following chart shows how many links originated from JSTOR's top five linking partners in 2002:

Resource	Links in 2002
ABC-CLIO America: History and Life Historical Abstracts	240,970
American Mathematical Society MathSciNet	107,812
History Cooperative	57,583
ExLibris SFX	51,350
EBSCO Publishing EBSCOhost	45,422

As we look at the statistical information we capture for links into the archive, five interesting trends have emerged which we will continue to follow over time

1. An increasing number of article views in JSTOR come directly from links from external resource providers.

In JSTOR, article views originate from keyword searches, table of contents searches, issue browsing, and external links. Thus far in 2003, links from participating institutions and linking partners account for nearly 11% of the total article views in JSTOR - an increase from nearly 10% in 2002 and nearly 3% in 2001.

	2001	2002	20031
Total Articles Viewed in JSTOR	50,279,178	80,082,633	33,691,778
Articles Viewed via Links	207,278	1,453,751	709,458
Percentage of Articles Viewed in JSTOR via Links	2.61%	9.93%	10.89%

2. The percentage of article views coming from these links varies by discipline.

In the discipline of mathematics, for instance, where JSTOR archives and provides access to 25 titles, almost 35% of the total article views originated from links external to JSTOR. This is a much higher percentage than the average, and is much higher than the next closest discipline within JSTOR (Economics - 20.5%). We think it would be useful to understand why these differences exist across disciplines and we will be researching what factors are involved.

Discipline	Percentage of Articles in JSTOR Viewed via Links2	Total Article Views
Mathematics	34.9%	209,439
Economics	20.5%	1,090,405
History	15.9%	1,381,786

3. The quality of links from JSTOR's linking partners has stayed fairly high, but has been declining gradually.

So far in 2003, 92% of links to JSTOR are successfully taking the user to the precise article they requested. For 2002, that percentage was 93%, and in 2001, 94%. Is that good? We do not really have a standard to make that determination. We do know that some of our partners are linking successfully at higher rates than others, and we plan to share these statistics with our partners so that we can improve the quality of the linking on an ongoing basis.

4. While the quality of the links, percentage-wise, has stayed fairly high, the raw numbers of invalid links is disconcerting.

In 2002, there were 1.4 million links into JSTOR, and of these almost 99,000 were unresponsive – that is, the link did not take the user either to the precise article they had requested or to a selection of articles that might fit the metadata provided. Even though those almost 99,000 links only accounted for 8% of the total links into JSTOR last year, each time a student, faculty, or librarian is not taken to the appropriate article, it is cause for concern to both JSTOR and our participating libraries. We need to work diligently to reduce the number of potentially unsatisfactory user experiences.

5. Linking "on the fly" results in significantly lower quality links than "pre-formulated" links.

There are two basic ways to link into the JSTOR archive. Some providers formulate their article links in advance, creating the SICI-based URLs, testing these links using tools provided by JSTOR, and then storing them in a database. Other providers use a method sometimes referred to as dynamic linking –– or linking "on the fly" –– which involves creating the link in real time. For instance, one might link to the full-text article in JSTOR from an OpenURL link provided by SFX or LinkFinderPlus. These links are constructed at the time the request is made. Our usage data has shown that dynamic linking does not produce the same high quality links as those produced by pre-formulated links. Dynamic links to JSTOR resolve accurately at a rate of only 82%. By comparison, providers who pre-formulate and test their links to JSTOR in advance have a success rate of over 98%.

We will continue to look closely at our linking data to see if these trends continue. We will also be sharing this data with our linking partners and continue to work with them to improve their linking algorithms and software. In addition, we hope that our participating libraries will continue to contact us at support@jstor.org with suggestions for future linking partners and to share with us their linking experiences in JSTOR.

_____________________________
¹Figures through March 31, 2003
²For usage from May - December 2002

Last updated on September 8, 2006