Blog Post

Card Sorting: A Primer

Change your language and you change your thoughts.” –  Karl Albrecht

Card sorting is a specialized type of user test that is useful for assessing how people group related concepts and what common terminology they use.  At its simplest form, a card sort is the processes of writing the name of each item you want to test on a card, giving the cards to a user and asking him or her to group like items into piles. There are however, a number of advanced options and different types of card sorting techniques.

Open Card Sort – An “open” card sort is when you do not provide users with an initial structure. Users are given a stack of index cards, each with the name of an item or piece of content written on it. They are then asked to sort through and group the cards, putting them into piles of like-items on a table. Users are then asked to categorize each of the piles with the name that they think best represents that group/pile. Open card sorts are usually conducted early on in the design process because it tends to generate a large amount of wide ranging information about naming and categorization.

Given the high burden on participants and researchers, I personally find an open card sort to be the least attractive method for most contexts. It is, however, the most unbiased approach. As a general rule, I would reserve this method for testing users with a high degree of expertise in the area being evaluated, or for large scale exploratory studies when other methods have already been exhausted.

Closed Card Sort – The opposite of an open sort, a “closed” card sort is when the user is provided with an initial structure. Users are presented with a set of predefined categories (usually on a table) and given a stack of index cards of items. They are then asked to sort through the cards and place each item into the most appropriate category. A closed sort is best used later in a design process. Strictly speaking, participants in a closed sort are not expected to change, add, or remove categories. However, unless the context of your study prevents it, I would recommend allowing participants to suggest changes and have a mechanism for capturing this information.

Reverse Card Sort – Can also be called a “seeded” card sort. Users find information in an existing structure, such as a full site map laid out on index cards on a table. Users are then asked to review the structure and suggest changes. They are asked to move the cards around and re-name the items as they see fit. A reverse card sort has the highest potential for bias; however, it’s still a relatively effective means of validating (or invalidating) a taxonomic structure. The best structures to use are ones that were defined by an information architect, or someone with a high degree of subject matter expertise.

Modified Delphi Card Sort (Lyn Paul 2003) – Based on the Delphi Research Method[1], which in simple terms refers to a research method where you ask a respondent to modify information left by a previous respondent.  The idea is that over multiple test cycles, information will evolve into a near consensus with only the most contentious items remaining. A Modified Delphi Card Sort is where an initial user is asked to complete a card sort (open, closed, or reverse), and each subsequent user is asked to modify the card sort of their predecessor. This process is repeated until there is minimal fluctuation, indicating a general consensus. One significant benefit of this approach is ease of analysis. The researcher is left with one   final site structure and notes about any issue areas.

Online Card Sort – As the name implies, this refers to a card sort conducted online with a card sorting application. An online card sort allows for the possibility of gathering quantitative data from a large number of users. Most card sorting tools facilitate analysis by aggregating data and highlighting trends.

Paper Card Sort – A paper sort is done in person, usually on standard index cards or sticky notes. Unlike an online sort, there is opportunity to interact with participants and gain further insight into why they are categorizing things as they are.

Why Use Card Sorting?

  • Card sorting is a relatively quick, low cost, and low tech method of getting input from users.
  • Card sorting can be used to test the efficacy of a given taxonomic structure for a Web site or application. While commonly used for website navigation, the method can be applied to understand data structures for standalone applications as well.
  • When designing new products or major redesign efforts.
  • When creating a filtered or faceted search solution, or evaluating content tags
  • For troubleshooting, when other data sources that indicates users might be having a hard time finding content.

When is Card Sorting Most Useful?

Development Life-cycle

Card sorts are useful in the requirements gathering and design stages. Depending on where you are in the design process you may get more or less value from a given method (open, closed, reverse, etc).

Limitations of Card Sorting

  • The results of card sorting can be difficult and time consuming to analyze; results are rarely definitive and can reveal more questions than answers.
  • The results of a card sort will not provide you with a final architecture; it will only give you insight possible direction and problem areas.

How to Conduct a Card Sort

Card sorts are one of those things that are somewhat easier to conduct than to explain. Because there are so many variations, I’ve decided to illustrate the concept with a walkthrough of an actual project case study. I was recently brought into a card sorting study by a colleague of mine[2] who was working on a complex taxonomy evaluation. The project was for a top toy retailer’s e-commerce site. After weeks of evaluating traffic-patterns and other data, my colleague had developed what he hoped would be an improved new site structure. He wanted to use card sorting techniques to validate and refine what he had developed.

  1. Define your research plan. Our research plan called for some online closed cards sorts to gather statistically relevant quantitative data, as well as the rather innovative idea to go to one of the retail locations and recruit shoppers to do card sorting onsite. The in-store tests would follow a reversed sort, using the modified Delphi method. I.e. Shoppers would be shown the full site structure and asked to make changes. Each shopper would build off of the modifications of the previous shopper until a reasonable consensus was achieved.
  2. Prepare your materials. In the case of in-store card sorts, we needed to take the newly defined top and second level navigation categories and put each on its own index card. The cards would be laid out on two long banquet tables so participants could see the structure in its entirety. Single page reference sheets of the starting navigation were printed up so we could take notes on each participant and track progressive changes. We had markers and blank index cards for modifications. A video camera would be used to record results, and a participant consent form was prepared.
  3. Recruit Participants. Unlike lab-based testing where you have to recruit participants in advance, the goal for the in-store testing was to directly approach shoppers. The idea was that not only would they be a highly targeted users group, but that we would be approaching them in a natural environment that closely approximated their mindset when on the e-commerce site i.e. shopping for toys. Because we would be asking shoppers to take about 10-20 minutes of their time, the client provided us with gift cards which we offered as an incentive/thank you. Recruitment was straightforward; we would approach shoppers, describe what we were doing and ask if they would like to participate. We attempted to screen for users who were familiar with the client’s website or at least had some online shopping experience.
  4. Conduct the Card Sort. After agreeing to participate and signing the consent form, we explained to the participant that the information on the table represented potential naming and structure for   content on the e-commerce site. Users were asked to look through the cards and call attention to anything they didn’t understand or things they would change. They could move items, rename them or even take them off the table. Initially we let the participant walk up and down the table looking at the cards. Some would immediately start editing the structure, while others we needed to encourage (while trying not to bias results) by asking them what they had come into the store for and where might they find that item, etc. After an initial pass, we would then point out to the participant some of the changes made by previous participants as well as describe any recurring patterns to elicit further opinions. After about 15 participants, the site structure stabilized and any grey areas were fairly clearly identified.
  5. Fig 3 Card Sort
     
    Figure 3: Sample Card Sort Display: Cut Index Cards on Table

  6. Prepare the Analysis. At the end of the study, there was a reference sheet with notes on each participant’s session, video of the full study, and a relatively stable final layout. With this data, it was fairly easy to identify a number of recurring themes, i.e. issues that stood out as confusing, contentious, or as a notable deviation from the original structure. As in any card sort, the results were not directly translatable to a final information structure. Rather, they provided insights that could be combined with other data (such as online sorting results) to create the final taxonomy.

Additional Resources

 


[1] The Delphi Research Method http://www.iit.edu/~it/delphi.html

[2] David Cooksey, Founder & Principal, saturdave, Philadelphia, PA

Blog Post

How to Prototype for User Testing

“If I have a thousand ideas and only one
turns out to be good, I am satisfied.”
– Alfred Bernhard Nobel

UXD prototyping is a robust topic and difficult to adequately cover in this type of overview guide. Therefore, I do so with the caveat that I am only providing the tip of the iceberg to get you started, with some references to where you can learn more.

Prototyping is not a research method, it is a research tool. An important thing to understand about UXD prototypes is that the term “prototype” itself can mean something slightly different to the UXD community than to the software development community at large. For example, one of the more common engineering usages of the term refers to an operational prototype, sometimes called a proof of concept. This is a fully or partially functional version of an application. UXD prototypes are, however, not usually operational. Most often they are simulations focused on how a user might interact with a system. In the case of the “paper prototype”, for example, there is no functionality whatsoever, just paper print-outs of software screens.

There are many different names for types of prototypes in software engineering, most of which describe the same (or very similar) concepts. Here are some of the most common terms:

  • Operational – Refers to a fully or partially functional prototype that may or may not be further developed into a production system. User testing prototypes can be operational, such as during late stage validation testing, though a program beta is more commonly used at this stage.
  • Evolutionary –As the name implies, an evolutionary prototype is developed iteratively with the idea that it will eventually become a production system. User testing prototypes are not “evolutionary” in the strictest sense if they do not become production systems. However, they can be iterative and evolve through different design and testing cycles.
  • Exploratory– Refers to a simulation that is intended to clarify requirements, user goals, interaction models, and/or alternate design concepts. An exploratory prototype is usually a “throw-away” prototype which means it will not be developed into a final system. User testing prototypes would usually be considered exploratory.

Semantics aside, prototypes are probably the single most powerful tool for the researcher to understand user behavior in the context of the product being developed. Some commonly used prototypes in user testing are;

  • Wireframes – A wireframe is a static structural description of an interface without graphic design elements. Usually created in black, white and grey, a wireframe outlines where the content and functionality is on a screen. Annotated wireframes are wireframes with additional notes that further describe the screens’ content or interactivity.
  • Design Mockups – Similar to wireframes, a design mockup is a static representation of an interface screen. However, design mockups are full color descriptions that include the intended graphical look and feel of the design.
  • HTML Mockups –Used in web site design, HTML mockups refer to interface screens that are created in the Hypertext Markup Language and so can be viewed in a browser. Most often, HTML mockups simulate basic functionality such as navigation and workflows. HTML mockups are usually developed with wireframes or a simplified version of the intended look and feel.
  • Paper Prototype – A paper prototype is literally a paper print-out of the designed interface screen. A paper prototype could be of wireframes or design mockups. In addition, it could be one page to get feedback on a single screen, or a series of screens intended to represent a user task or workflow.
  • PDF Prototype – A PDF Prototype consists of a series of designed interface screens converted into the Adobe Acrobat (.pdf ) file type. Like HTML prototypes, PDF prototypes can simulate basic functionality such as navigation and workflows. However, they take less time to create than HTML prototypes and can be created by someone without web development skills.
  • Flash Prototype – A prototype developed using Adobe’s Flash technology. Flash prototypes are usually highly interactive, simulating not only buttons and workflows, but the system’s interaction design as well. In addition to being a quick prototype development method, Flash prototypes can be run via the web or as a desktop application making it very portable tool.

Why Use Prototyping

  • It saves money by allowing you to test and correct design flaws before a system is developed.
  • It allows for more freedom to explore risky, envelope-pushing ideas without the cost and complexity of developing it.
  • Since prototypes are simulations of actual functionality they theoretically bug-free. Test results are less likely to be altered or impeded by implementation issues.

When is a Prototype Most Useful?

  • When you are trying to articulate a new design or concept
  • When you want to test things in isolation (i.e. graphic design separate from information layout separate from interaction design)
  • To gather user feedback when requirements are still in a state of flux and/or can’t be resolved
  • To evaluate multiple approaches to the same user task or goal to see which users prefer

Limitations of Prototyping

  • A prototype will never be as accurate as testing on a live system; there is always some deviation between the real world and the simulation.
  • Depending on where you are in the iterative research process, there is a point of diminishing returns where the amount of effort to create the prototype is better put into building a beta.

Creating a UXD Prototype

Because the actual prototype creation process is highly specific to what you are creating and what tools you are using (paper napkins, layout tool, whiteboard, etc) I’ve included a few considerations for defining a prototype instead of detailing the mechanics of creating one.

  1. Consider your Testing Goals. – Are you looking to understand how users perform a specific set of tasks? Do you need to watch users interact as naturally as possible with the system? Or are you trying to get users’ responses to various experimental ideas and concepts? The answers to these questions will help you make some key decisions about the structure of your prototype.
  2. Decide on Degree of Fidelity. What level of fidelity will the prototype achieve? Here, the term fidelity refers to the degree to which a prototype accurately reproduces the intended final system. A low fidelity prototype might be a PowerPoint deck of wireframes. A high fidelity prototype could be an interactive simulation with active buttons and representative content.A good rule of thumb is to develop the lowest fidelity prototype possible to achieve the goals of your study. This ensures a lack of commitment to the ideas presented and allows more time and money for the recommended iterative process. If a significant amount of time is taken to create an initial prototype with all the bell and whistles, designers are less willing to see when the concept is not working, less likely to change their designs. In addition, the amount of time that goes into building one high fidelity prototype would have been better used building multiple lower fidelity versions that allowed for more testing in-between each revision.
  3. Scripted Tasks or Natural Exploration? – Another consideration when defining your testing prototype is what content and functionality should be included. Will a preset walkthrough of key screens be sufficient, or do the goals of your study require that users are able to find their own way around the system? On average, I tend to think that enough insight can be gleaned from a series of walkthrough tests and other research methods to warrant using these, leaving the open-ended user-directed tests to be conducted on a product beta or via A/B testing[1] on a live system. With a sufficiently complex system you can quickly hit a point of diminishing returns regarding the amount of effort it takes to simulate functionality vs. actually building it.

Additional Resources

  • Microsoft Visio (office.microsoft.com/en-us/visio) – The “old school” standby for wireframes and workflows.
  • Adobe InDesign (www.adobe.com/products/indesign) – Arguably the industry standard tool. Layout tool with key functionality conducive to prototype development.
  • Balsamiq – (www.balsamiq.com) – Good low-cost alternative when you only need the basics.
  • Further Reading – Rettig, Marc. “Prototyping for Tiny Fingers.” Communications of the ACM. Vol. 37, No.4. April, 1994.


[1] A/B Split Testing refers to a testing method where in a live system an alternate, experimental design is shown to a percentage of users while the baseline control design is shown to the rest. Analysis comes from observing any notable differences in user behavior between the experimental design and the control.