UI Design Newsletter – April, 2005

Insights from Human Factors International

In This Issue:

Making it findable

Kath Straub, Ph.D., CUA, Chief Scientist, on the relative importance of user-centered labels and structure.

The Pragmatic Ergonomist

Dr. Eric Schaffer, Ph.D., CPE, founder and CEO of HFI offers practical advice.

It is frequently said that navigation is 80% of good usability. I've often wondered what that means. What are the parameters that make a site navigable? What specifically do I need to get right to automatically have it be 80% good?

User-centered Web designers answer this question reflexively. Good navigation means good information architecture. Good information architecture means having a hierarchical structure and the right labels. Having the right structure means deriving the hierarchy that reflects users' mental organization of the information. Using the right labels means ignoring the organization's (branded) terms for things and adopting the users' vocabulary for tokens and categories.

These two parameters – structure and labels – are asserted to be as independent and complimentary. Neither is individually sufficient to trigger that 80% usability threshold. You have to get both right.

Which comes first? The label or the structure?

According to Rosenfeld and Morville (1998), identifying the right structure depends on how well the target users know the taxonomy or classification of the content of the site. Exact schemes, such as alphabetical order or organizational structure, are best used when it is certain that users know the specific labels for the information they are seeking. Ambiguous schemes, such as organization by topic area, are preferred when users may not know keywords or specific content names, or when they may need to browse through the content to find what they need.

Organizational structures can be created organically, based on designer intuition or by adopting existing label sets; through experts performing professional indexing; or via user-centered methods in which user input / feedback is incorporated to construct and validate the information organization.

Large, not venti

Getting the labels right means matching the names for things to the words that the users use to refer to them. In creating effective labels, designers need to:

• consider how much target users will know about the system and domain,
• minimize jargon, particularly when speaking to a general population, and
• avoid ambiguity.

Labeling systems can come from existing labels sets, benchmarks across other Web sites, experts, and users. The best labels are typically generated directly by representative users.

When not informed by user-centered research strategies, both structural design and label creation can suffer from what Fleming (1998) refers to as the "disease of familiarity." That is, designers often assume that users know as much about the organization or topic as they do because they have a difficult time recalling what it's like to not know something. The result is an architecture that reflects the internal structure and labels of the organization. These architectures can often prove difficult for users who are not intimately familiar with the organization (Martin, 1999).

As such, it is critical to gather user data to motivate and externally validate any proposed architecture.

Flash back to reality, again. You have a short timeline and a limited budget. You can conduct some user research, but never as much as you would like to. You need to optimize your efforts.

You can focus your research on deriving a solid user-centered structure or clear labels reflecting the users' vocabulary. Which do you pick?

Resnick and Sanchez (2004) sought to answer exactly this question. They conducted a controlled experiment to explore the relative value of a user-defined structure versus user-generated labels in determining perceived ease-of-use and efficiency of a Health Product shopping Web site.

The study

To prepare the study materials, Resnick and Sanchez conducted a series of preliminary norming exercises.

First they conducted a survey of existing health food Web sites to identify existing schemes for product organization. They found two prevalent themes: by product (e.g., bars, pills, books) and by task (e.g., weight loss, reduce stress).

Then they conducted a card sorting task with ten participants to generate groupings and labels for the to-be-included elements. Users indicated a clear preference for the task-organization within the card sorting task.

Finally, they conducted a category label ranking task with 20 new participants to evaluate the goodness-of-fit between proposed content and category labels. They used these findings to derive three levels of labeling.

To test the relative contributions of user structure and good labels, Resnick and Sanchez created 6 versions of a fictitious health food Web site that systematically varied structure and label quality:

• Task-structure / Good Labels
• Task-structure / Medium Labels
• Task-structure / Poor Labels
• Product-Structure / Good Labels
• Product-Structure / Medium Labels
• Product Structure / Poor Labels

They then recruited 60 more participants to complete a shopping scenario on the fictitious site. Within the scenario each participant was instructed to collect six specific items on the site.

Resnick and Sanchez hypothesized that since the majority of the participants in the card sort had derived a task-based organization, a task-based structure would prove easier to navigate than the product-based structure.

They also hypothesized that users would be more efficient on the site with good category labels than on the site with medium or poor labels.

They measured efficiency via time to complete the task, number of clicks, number of errors, and number of products found.

Some things are more equal than others

Resnick and Sanchez' data shows that good labels had a robust effect on performance:

Time: Participants were 90% more efficient on the site with good labels than with poor labels.

Number of clicks: The site with good labels required 25% fewer clicks to complete the task.

Errors: Participants committed significantly less errors (strays from optimal path) on sites with better labels.

Number of items found: Participants found significantly more items.

Satisfaction: Participants' satisfaction ratings were significantly higher on sites with better labels.

The findings for structure were somewhat different.

First, users were more efficient using the product-based structure than the task-based structure. This enhanced performance on the product-centered organizations is in conflict with their tendency to organize the content by task in the preliminary card sorting study.

Second, structure had a significant impact on task efficiency for number of clicks and number of errors. However, efficiency differences by structural type were only meaningful for the test sites with bad labels. For sites with good labels, there was no benefit to having one site structure over the other.

Even Washington, DC is navigable if the street signs are good

These findings – that labels matter more than structure – are not surprising if you think about it.

When you land in a new city, if the street signs are visible you can get around without learning the structure of the street layout.

If the signs are hard to see or missing, a familiar – or at least predictable – structure helps. In a predictable city like New York, once you know where you are, you can predict where you are going based on the 44th street, then 45th street, then 46th street grid structure. Not so in cities like Washington, which is designed on a difficult-to-intuit diagonal spoke structure, or Mumbai, where streets emerged organically.

So, Resnick and Sanchez suggest that the key to good architecture, and by extension usable navigation, is good labels. If you get the labels right, you are most of the way there.

This structure vs. label study is interesting, but be careful. I have seen disasters when wonderful labels are placed in the wrong structure. For example, an alphabetic structure where the user does not know the word (is it "Hire Employee" or "New Employee"?). I am quite sure it is worthwhile continuing to design, refine, and validate the structure as well as the labels. In fact, it will cost little extra to attend fully to both as the research methods can be combined into a single test session.

References

Fleming, J. (1998). Web Navigation: designing the user experience, O'Reilly.

Martin, S. (1999). Cluster analysis for Web site organization. Internetworking 2(3).

Resnick, M. L. & Sanchez, J. (2004). Effects of organizational scheme and labeling on task performance in product-centered and user-centered retail websites. Human Factors, 46.

Rosenfeld, L. and Morville, P. (1998). Information architecture for the World Wide Web, O'Reilly, Sebastopol CA.