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


Create Second-Level Categories

Depending on the number of items included in the original card sort study, and therefore in the resulting categories, it may be necessary to create subcategories from the top-level categories. The more content a website has, the more the need for subdividing categories so that the user can drill down through the content hierarchy as efficiently as possible4. The dendrogram included in most of the online card sorting tools makes this task relatively easy.

The other important point to realize before further subdividing the categories is that all of the rich data created by either manual methods or online tools to arrive at the top-level categories and labels will not be available for the subcategories. In other words, the participants did not create labels for the subcategories that you’re about to infer. But, at least your attempt will be based on the varying labels participants suggested for the top-level categories.

Once you decide to further break the large categories apart into subcategories, you have to first determine which categories are good candidates for such subcategorization. Characteristics of categories that are good candidates for subdividing include the following:

For this discussion, assuming that you settled on seven top-level categories for your IA, we’ll focus on a large category beginning with “Organic Guatemala Santa Isabel” that contains 26 content items (Figure 6).

Figure 6

Figure 6. Dendrogram of all seven categories

The large category does not begin to break apart until you move the slider to create 12 categories. The large category breaks into two subcategories: one containing a group of organic and fair-trade coffee and another group containing everything else (Figure 7).

Figure 7

Figure 7. Seventh category subdivided into two categories

As you move the slider further to create 16 categories, another break occurs to create three subcategories from the original one large category (Figure 8). You can perform this process not just once, but multiple times, with each subsequent iteration consisting of creating more categories containing fewer items.

Figure 8

Figure 8. Seventh category subdivided into three categories

As you continue to “play” with the categories to see if they break up easily into subcategories, it’s important to continue referring back to all of the content of the website to see how it will fit into the new subcategories you’re creating. That should definitely influence your decision about the number of subcategories to create by manipulating the dendrogram slider.

As mentioned previously, you may be able to derive labels for the new subcategories by reviewing the item-by-category matrix. In this example, if you decide on a final category of “Coffees,“ by breaking out the large group further, you can see that the users grouped the organic and fair trade coffees together in Subcategory 1. Based on the category labels participants assigned, some variation of “Fair Trade Coffees” and “Fair Trade Organic Coffees” appear multiple times in the list. Settling on a label for the second category above is more difficult because there are not any labels specific to these few items in all of those participants assigned. At this point, you may need to conduct some independent research to better understand what a “mocafe” coffee is. Because one appears to be a vanilla flavored coffee, you might consider breaking this small group apart and grouping the “Mocafe Tahitian Vanilla” with the other flavored coffee.

Figure 9

Figure 9. “Espresso” and “Blends” subcategories

In Figure 9, the espresso coffees and the blends and flavored coffees break apart further into subcategories. However, even in this blend and flavored category, some of the coffees, such as “French Roast” and “Columbia Supremo,” are not true blends or flavored coffees. You will probably need to remove them from this category and find a more appropriate group for them. 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. However, at this point, you will have to carefully consider the audience(s) for the website when attempting to determine subcategories and labels. If the audience does not largely consist of coffee connoisseurs, you will have to balance the categories and labels carefully to not alienate casual coffee drinkers, but also not to inaccurately categorize a coffee so that a connoisseur would be unable to find it on your site. Obviously, because the online tool used here as an example did not allow the users to create the subcategories and exact labels, you will be making subjective assumptions based on the data you do have.

If this exercise of further defining subcategories and labels seems too subjective, you could conduct subsequent card sorting studies for each of the top-level categories to arrive at second and third level categories and labels. Essentially, you’ll be conducting a card sort of the items within each of the top-level categories. Unfortunately, this is seldom feasible.

You have additional alternatives. Take a stab at defining the subcategories and labels and then do the following:

Creating subcategories and labels using data that is really only intended for top-level categories is a bit of an art, and perhaps, more trial and error than most might find comfortable. On the other hand, that data has been created by the actual users—always a better alternative than starting from your own best guesses based on your individual experiences and knowledge that might be very different from your user population’s.

 


4Deciding on breadth vs. width of categories is a topic covered extensively in the IA literature and will not be discussed in depth here.

 

Previous | Next