As part of an information architecture (IA) project, we have been looking into different ways of usefully analysing the results of card sorts. What is card sorting? Card sorting is a quick and informative way of finding out how users group information. It is often the first step to developing a site IA. Open or closed? Open card sorting works by giving users an unsorted number of cards, each with a user goal written on, and asking the user to organise the cards into groups that make sense to them. They then name the groups themselves. Some users go through the cards in a very structured way; others start by throwing them across the table. Quantitative or qualitative? In a quantitative card sort, the facilitator (usually us) takes quite a back seat, allowing the testers (often in groups of four) to get on with it. Quantitative sorts are to get the number of results we need to analyse. In a qualitative card sort the facilitator is more involved, asking for a running commentary on why the tester is grouping cards together and the thinking behind it. This helps us when we analyse the data and work on the IA as we have an idea of user thinking. We usually use a combination of these two methods to get our results, often with a rough 50/50 split. The cards and goals The goals on cards should always be research-based. Card sorting is a research-based exercise, and it’s results are dependent on the quality of the goals used to sort. On a recent Cogapp project, we conducted the following before undertaking card sorting:

• A heuristic site evaluation
• Review of existing research
• Stakeholder interviews which provided organisational goals
• User depth interviews which provided a number of user goals
• Strategy workshop to prioritise user goals and organisational goals

In this example, we held a strategy workshop with our client where we analysed the goals that had come out of the user research and prioritised the organisational need against the user need, shown below.

 This served as useful cross-referencing tool when we pulled the data from the card sort. Analysing the data We use Syncaps card sorting software. Syncaps prints cards for the sort to a template - with a scale to measure the fit in the group as perfect; good or fair. This adds an extra layer of accuracy to the results. Image from http://www.syntagm.co.uk/design/cardsort.shtml After sorting the cards and naming the groups, the testers are asked to mark on the card how well the card fits into the group. This is then scanned using a barcode scanner, and fed into the Syncaps software. The output is a dendogram. The principle behind the dendogram comes from biology and the classification of animals. Goals that users have closely aligned appear closer together in the list, with shorter connecting lines on the left. Syncaps allows us to change the number of groups until we feel the number is right.Image from http://www.syntagm.co.uk/design/cardsort.shtml Using the goal priorities from the strategy workshop, we can add an extra layer to the data. This show us which aspects of each group are important to our users, and thus which areas we should bring to the fore in the IA, drawing the users into the other content which is more organisationally driven. Surface map Syncaps also outputs a surface map, showing groupings in a more visual form. This is useful to see contextual links between goals, demonstrated below by the darker red sections towards the edges of the picture. Image from http://www.syntagm.co.uk/design/cardsort.shtml Group prioritisation Using the priorities assigned to each goal, we can calculate the organisational and user priorities associated with each group, plus an average of the two priorities to show which sections are most important. If a section has a low priority for both groups, we need to question if it should be included. Areas with a higher user priority are useful to know when publicising the site or orientating it as it gives another angle to draw in our users. Tag clouds When sorting cards, users were asked to give their groups names and to make notes on cards. By measuring word frequency by running all the user input for a particular group through a tag cloud generator (like tagcrowd.com), we can quickly see the type of words users associate with each group. These then give us a steer for either naming the groups, or for any qualifying text that accompanies the groups on landing pages. Analysis of this blog post image by http://tagcrowd.com What’s next? After analysing the data and groupings, we begin drafting the IA and developing a wireframe prototype. Developing the IA is usually done in a program like OmniOutliner. This gives us a quick and flexible way of sorting the IA. We have developed a program that can take the output from OmniOutliner and convert it seamlessly into a rough HTML prototype, making user testing the IA realistic for users as they are working with something that resembles a real website in both look and behaviour. We test the IA, we develop wireframes, we test them - it’s an iterative process of test and refine and then, that’s it.