upa - home page JUS - Journal of usability studies
An international peer-reviewed journal

Card Sort Analysis Best Practices

Carol Righi, Janice James, Michael Beasley, Donald L. Day, Jean E. Fox, Jennifer Gieber, Chris Howe, and Laconya Ruby

Journal of Usability Studies, Volume 8, Issue 3, May 2013, pp. 69 - 89

Article Contents


Determine the Level of Agreement Between Items

Throughout the course of analyzing card sort data of an unmoderated/remote study, there are a number of data matrixes and diagrams created by either the automated tool or that you will most likely create manually. The first to consider is the item-by-item matrix. This matrix will help you quickly find key relationships between individual items in the card deck.

The item-by-item matrix (Figure 1) shows the number of times, as a percentage, participants have grouped each individual card with each other card in the set. The matrix provides useful insight into the strength of the relationship between each pair of individual content items, and thus, how strong a group the items form. In this example, the higher numbers (illustrated by the darker colors of the cells) indicate stronger item-to-item relationships. This tells you that those items should probably be placed within the same category in the IA.

Figure 1

Figure 1. Sample item-by-item matrix

The information provided in the item-by-item matrix offers an ideal opportunity to test any initial assumptions for the IA. In fact, before looking at the data, some analysts will write a list of their hypothesis and then later determine whether their assumptions are borne out in the results. These assumptions can be framed around the question, “I think X and Y belong together because [your reason],” or “There is a strong relationship between [a few specific cards], indicating the need for an IA based on [your guiding principle].” Later, if your hypotheses are not supported by the results, you may wish to redefine your assumptions and revisit the data to see if your initial hypotheses have any validity or if new hypotheses should take their place.

Strong Item-to-Item Connections

To start analyzing the information in the item-by-item matrix, look at the strongest relationships and ask yourself: “What is the connection between these items? What relationships are participants seeing when grouping these cards together? And, how does that affect items that are not connected that I expected to be?” Ideally, patterns of connections will begin to emerge.

Often, surprisingly strong connections between items may not fit neatly once the IA is derived. Rather than totally dismiss these insights, they can give valuable guidance into the cross-links and cross-promotions to use in the final IA. Note any of these surprising results for later.

Weak Item-to-Item Connections

Next, reverse the process and look at the cards that have the lowest correlations with one another. Determine whether there are any surprises or outliers that demand further investigation. Are there items you expected to be connected but weren’t? Again, note these results as you continue your analysis.

Subgroup Analysis

If you have a very heterogeneous group of participants’ data, you may wish to segment your participants by behavioral and demographic criteria. This will allow you to perform the item-by-item matrix review across a variety of user types. By doing so, you can determine whether and how the data patterns change across user audiences. You can then, for example, create an IA that is best suited to the highest-priority user audiences. For example, separating the data of the “coffee connoisseur” user type from the more general “coffee drinker” user type would likely give you a more technically accurate categorization and labeling of the different coffees. Most automated tools allow you to segment the data in some fashion or other.

What About the Labels?

You may be curious at this point about what the participants labeled the categories. Category labels will be addressed soon. For now, it is best to focus on how the items have been grouped.

 

Previous | Next