Card sort analysis best practices

Abstract

Information architecture is the practice of effectively organizing, structuring, and labeling the content of a website or application into a structure that enables efficient navigation. Card sorting is a research method that employs users' input to help derive an effective navigation structure. Whether a card sorting study is conducted using manual methods and tools, or online automated tools, those tools only assist the User Experience (UX) professional in creating usable, intuitive information architecture. As with most user research, the UX professional still has to make sense of the collected data and take the data one (or several) steps further. How do you interpret user input to determine what categories to create? What should those categories be named? Which content should they contain? Should the categories contain subcategories, and which content should they contain? Should you duplicate links across categories, and where should these duplicate links reside? Although a wealth of articles, books, and blogs by experts in the information architecture field discuss the card sorting technique, most do not answer these questions, nor do they address the step-by-step details of how to analyze the data to create a navigation structure. Attendees at card sorting workshops taught by the two primary authors (Righi and James) of this article invariably report that of all the elements of card sorting, they have the most difficulty taking the leap from the data they've gathered to constructing that final navigation structure. This article presents a set of best practices for analyzing card sorting data to derive an effective information architecture and navigation structure. It addresses methods of interpreting cluster analysis data matrixes and dendrograms generated by automated card sorting tools. And, it focuses on the details of making a decision about final categories and category labels. In short, it helps the UX professional make informed judgments when multiple interpretations of the data are possible.