An update on breadth vs. depth

During the last 5 years a controversy has been brewing concerning the breadth vs. depth in menu design for Web sites. Which is best? A site that is broad and shallow, presenting a lot of choices to the user right away, but only requiring a few layers? Or is it best to have narrow and deep, which means presenting only a few choices at a time, but requiring many layers in? As is usually the case, the answer turns out not to be so simple. In this issue of the newsletter we explore the variables that are emerging as important in the debate.

The history of breadth vs. depth

In the March 1999 issue of this newsletter, Bob Bailey concluded that breadth was better than depth. He reported on two studies that also surveyed past literature and concluded that it was better to have lots of categories in Web menus at the highest level and therefore reduce the number of clicks needed to get to the end point. (Zaphiris and Mtei,1998; Larson and Czerwinski, 1998).

Earlier studies by Snowberry, Parkinson & Sission (1983) demonstrated that in addition to performance and preference declines, navigational error rates increased significantly as hypertext depth increased. This study further demonstrated the value of categorical grouping in shallow structures, showing that participants had an easier time finding resources in a 64 item list that was functionally organized than in a list with random presentation.

Follow-up studies (Kiger, 1984; Jacko & Slavendy 1996; Zaphiris and Mtei 1997) continued to demonstrate that:

• users found resources faster in broader, shallow sites than in narrow deep ones, and
• users understood broader, shallow sites better than they understood narrow deep ones.

However, these studies also began to demonstrate that although shallowness is a unique and defining parameter for success in learning hierarchies, sites can also be too shallow.

More variables surface

Recent research suggests that there is more to it. When we focus specifically on people browsing the Internet we change the question slightly. Instead of just asking "Which menu structure is best?" we really need to ask, "Which menu structures helps users quickly derive a conceptual model of the site hierarchy?" Creating an accurate representation of the structure and organization of the site results in a more successful and efficient navigation through the resource.

Several factors are thought to influence users' success in learning and traversing information hierarchies. These do include the breadth/depth of hierarchy, but additional critical factors are: the transparency of the category and sub-category labels, qualities of information scent, relative size of categories, and the shape of the hierarchy.

Good scent and moderation in everything

Close review of Snowberry et. al. suggests that an alternative explanation for participants' success with a shallow design may be that the category labels were discrete enough to support reliable decision making. It seems that users conceptually manage broader category sets when the names of the elements within the set are discrete and easy to compare. Clear category names provide road signs or 'scent' cues about what lurks behind the link. Good 'scent,' in turn, supports successful category selection. Larson and Czerwinski (1998) tested the hypothesis that there is an interaction between structure, scent and category soundness by measuring speed, click-stream accuracy and perceived 'lostness' while varying breadth/depth parameters of presentation in a well defined information space. They observed that the distinctness of category names was particularly helpful at the highest levels of the information hierarchy, since selecting an incorrect path at the first hierarchical level often resulted in multiple-click backtracks. Note, however, that even with clear and distinct category labels, Larson, et. al. concluded that moderate breadth affords optimal user performance.

The claim that moderate depth supports optimized performance is further buttressed by Zaphiris (2001) computational models of user performance in menu search. Using behavioral values from previous HCI studies, Zaphiris' model predicts that menu design on either extreme (very deep or very broad) will undermine learnability and usability for users. This is postulated to be particularly so for older individuals — the fastest growing segment of Internet users.

Initial and final selection menus should be broadest

An additional parameter of site structure design that is currently drawing more attention is site shape. Consensus is that the initial age should balance breadth with layout/white space to offer a moderate selection of navigational options. Work on the optimal shape of a hierarchical site suggests that concave designs are optimal. A concave shape presents a broad initial selection screen, followed by category decisions over small categories and then followed with a terminal option set that is again somewhat broad. Norman and Chin (1988), and more recently Bernard (2002), demonstrate that for browse-oriented tasks, concave designs take users less time to navigate, and evoke less wasted clicks. In contrast, they observed no differences in ability to successfully navigate the various menu structures for explicit, target-specific scenarios.

What now?

Research comparing navigation efficiency through sites of varying depths and breadths broadly converges on the findings that users find roughly 16 (ungrouped) top-level links leading into 2-3 subsequent menus the most efficient, learnable and least error prone. This knowledge is well and good, but what does that really mean for designers? Today? Now?

First, we can derive some broad (and largely intuitive) design guidelines from this work:

• Too deep is too deep: users have a more difficult time encoding, and consequently navigating, deep sites.
• Too broad is too broad: conversely extremely broad sites (which may encourage satisficing) also present a challenge to efficient navigation.
• Effective sub-grouping reduces perceived breadth: grouping navigation elements thematically improves performance for even the broadest structures.
• Clear labels improve navigation accuracy: creating clear and distinct labels for navigation elements enhances performance.

More interestingly, careful consideration of the specific tasks used in these studies suggest that the breadth/depth findings map directly to effective Intranet design. Consider the typical participant's task in these studies: find an explicitly named target (search navigation) or navigate to an implicit, user-selected token within a specified category (browse navigation). Since Internet users frequently come to the Web looking for information about a concept, but without a specific page in mind, it is commendable that researchers have begun to focus more extensively on browse-driven exploration. However, it also must be noted that both implicit and explicit search tasks are still essentially serial tasks. That is, the research participants in these studies completed well-defined, single tasks, returning to the home page before initiating the subsequent task. This approach maps directly to the serial task completion behavior patterns observed for the frequently executed Intranet tasks: find a phone number; download a form; check the stock price; change personal benefits information, find a policy.

Thus, the breadth/depth research speaks directly to optimal structure for Intranet design. Further, the cumulative findings of the research challenges the widely implemented approach that Intranets should focus corporate or institutional news wrapped in what is typically a tab-based (or tab plus left side navigation) functional navigation design. Instead, this research suggests that moderately broad site structures, consisting essentially of functionally grouped, transparently labeled link lists will provide the most effective navigation structure with the best perceived usability on Intranets. Anecdotal user-centered field analysis and prototype validation provides additional support for this approach to Intranet design.

Designers who want to know how to take advantage of the depth/breadth research need to think about the kinds of tasks that people do on their sites, and how people approach doing those tasks. Do users tend to do one task at a time? Do they finish that task before they start another? (Intranets are one example of a site type where this serial task completion model holds.) If so, then the task approach on your site is parallel to the task flow that was tested in the depth/breadth work. Therefore, these results apply and a broad, shallow menu architecture should provide users the most efficient and learnable access to resources on your site. In this case, however, the specific type of site is less important. In applying these results it is more important to think about what people do, and how they do it on your site.