One of the most important elements of the library website design process is determining how information and resources will be organized on the site. It is crucial to get user feedback as early as possible to ensure that the final navigation scheme, which exposes the relationships between site resources and features, is practical and usable. Card sorting is a procedure used to develop an understanding of how users themselves perceive and categorize library collections and services, and can thus dramatically improve website taxonomies and ultimately the usability of the entire website. Card sorting is efficient, economical, and can be done quite quickly.

Card sorting studies range from informal exercises to highly structured and rigorously controlled research, but the basic design for all is quite similar. Subjects, who represent user groups, are recruited and asked to sort cards bearing labels and titles corresponding to library services, collections, and titles into groupings that seem logical and appropriate to them. Subjects are then asked to create their own labels for these groups. Analysis of these groupings and labels is often highly informative, and results in relevant and useful suggestions for improved website organization and navigation.

A card sorting study begins with creation of a set of cards that represents the information contained on or proposed for a web page or site. 3X5 index cards are fine, but slips of paper cut to size serve equally well. Names of services and resources are printed on individual cards, and include things like reference assistance, document delivery, library hours and contact information, along with important collections (catalog, theses, electronic resources, etc.), plus a sampling of titles of major reference works, journals, and databases in broad areas or specific disciplines. Large websites may require as many as 150 - 200 cards, which will take a typical user about an hour to sort and label. Digital simulations of card sorting exercises do exist, and are useful in large studies, but manual sorting of real cards is preferred in small scale studies because of the ease of set up and administration. In any size of study, analysis can be greatly aided by tools such as Microsoft Excel.

Like many studies, card sorting is as much a logistical as an analytical challenge, but one of the many appealing features of this approach to understanding user needs is that just about any size of study will yield good results. The first tasks are selecting the labels for the cards, and recruiting participants. Assuming a team is involved in conducting the study (though a team is not absolutely necessary for a small-scale exercise), these tasks can be split up and accomplished quickly. Most cards should reflect current resources and services -- if new ones are being considered for inclusion in the new or revised site, add these to test them as well. Participants are recruited in the usual ways, for example posters, class announcements, advertisements, and so on. The offer of an incentive (a small cash payment, perhaps a gift of some kind) is sometimes needed to get users involved. Participants are told that the activity is not a test, and that there are no right or wrong answers.

A card sorting session with a participant may be divided into three steps: initial sorting, grouping, and labeling. In the first stage, each participant is given a set of labeled cards and three "place mats". The mats (laminated 8.5 x 11 sheets of paper can be used) are labeled "Resources used at least once," "Resources recognized but not used," and "Resources not recognized." The subject is instructed to sort the cards into these categories, with the place mats serving to help them keep the three piles separate. After sorting the cards the participant is given blank cards and asked to add any resources he or she has used but which were not represented by one of the pre-labeled cards.

After finishing the initial sort, the participant is given a pen and asked to label the cards in their 'used at least once' pile according to frequency of usage. The frequency categories can be something like daily (use label 'd'), weekly (w), and occasionally (at least once--o). While the participant is doing this, the moderator marks the other two piles of cards as either known but not used (k) or not recognized (?). Having participants label their cards in this way will allow the team to detect any patterns in the resources that are or are not used. The pile marked with the '?', or not recognized, is set aside.

In the second step, the participant sorts the cards into what he or she considers logical groupings. All cards representing resources used by the subject must be included in this sort. Participants can be given the option to sort any or all of the 'Known but not used resources' cards; they may know, for example, that New England Journal of Medicine is an electronic journal, but have never had occasion to use it. The moderator must be careful not to comment on any particular selection; card sorting is designed to determine how the user perceives different resources (or at least the labels used to define them).

Finally, the participant is asked to label the groupings of cards they have created. Using fresh cards of a different color (to help keep category labels distinct from resource cards), the participant writes down a word or phrase that best represents the entire group of cards. The groupings themselves are necessarily completely subjective; this is indeed the point of the study and it cannot be stressed too much. Subjects should therefore feel completely free to make up whatever labels they think best describe the groups of cards. Again, the moderator should refrain from coaching or questioning, but it may be necessary at this point to seek clarification of individual labels if it is not clear what they mean.

As each session is completed, the team members record results in an Excel spread sheet which has one worksheet for each participant. All of the resources/services used on the cards are listed in rows, and there is a column for the user-created category the item was placed in or, if applicable, either '?' for an unknown resource or 'k' for a known but never used resource that was not included in the participant sort. There is also a column for recording frequency of use as indicated by d,w,o, or k (if the resource was never used but recognized and included in the sort). After the spreadsheet is filled in for each participant the team can sort resources into their user-created categories and look for points of comparison. Are certain resources tended to be grouped together? Do different users choose similar labels for groups of similar resources? There will likely be at least some very obvious general trends that can be used in the website design.

Excel can be used to cluster user responses to whether or not resources were 'known' (k), or 'not-known' (?) to gain insight into usefulness of current labels. Not surprisingly, commonly used resources like 'Library Hours' and 'University home page' will have broad recognition, while others like 'Proquest databases' or 'Nature (Online)' may be recognized by few, if any, participants. Information like this might lead designers to avoid placing specific titles and types of resources at the same hierarchical level in the website navigation scheme. Also, clustering can confirm trends revealed in the labels participants attach to different resources. For example, in a card sorting study conducted at a medium-sized university library, over half of the 26 participants grouped various writing-related resources together (e.g. 'Footnotes', 'How to Write Essays', 'Citing Sources in Your Writing') and gave them the label, 'Writing'. It had not occurred to designers to include such a category on the top level page before, but the study revealed that it would be useful, and it has since been included as a main page link.

Card sorting is an excellent method for testing the usefulness of links and label with real users, and can help minimize library jargon and other confusing language. Indeed, card sorting can even help keep web pages current, or at the very least ensure that definitions of resources are clear and unambiguous. In the same study mentioned above, one participant categorized the card labeled 'Cell' (a major journal title in health sciences) under 'Library Services'. When asked to explain, the subject replied that she was able to user her mobile phone anywhere in the library, and thus thought that the library had some kind of enhanced signal, which was a great service!