File manager usability is a hot topic these days, particularly in the open source community. It began with John Siracusa’s 2003 Ars Technica article, About the Finder. In his critique of Apple’s Finder file manager he introduced a concept called the "spatial metaphor" as a better, more usable alternative. The spatial metaphor has since taken hold in the Linux community, most notably altering the direction of Nautilus, the GNOME file manager. And debate rages on between spatial advocates and their opponents (though most articles on the topic favor the former).

GNOME, KDE, Xfce, and other Linux GUI projects are building good user interfaces and file managers to compete with Microsoft Windows and Apple Mac OS X. It is incredibly exciting to see Linux coming into its own as an easy-to-install, easy-to-use system for non-technical users. But that makes the spatial debate more worrisome, because a core aspect of the computing experience is being determined by speculative, unsupported, oversimplified arguments on both sides.

The Players

I was concerned by John Siracusa’s article when I first read it, but I also thought it held some good suggestions for improving file management. Sadly, most of those details have been lost and the debate reduced to a simple one:

• Spatial proponents advocate a file manager similar to the old Mac OS 9 Finder. Each file or folder’s icon has a consistent location on the desktop or in its enclosing window, and each folder’s window remembers its location from use to use. There is a one-to-one mapping between folders and windows: You can’t have two windows open for the same folder. In this way, folders and their windows can be considered as unique, realistic objects. Spatial advocates sometimes propose the inclusion of a non-spatial interface as an advanced alternative, but often reject systems that implement spatial features selectively or partially as "breaking" the spatial metaphor.
• Spatial opponents usually advocate a "navigational" system that minimizes screen clutter (as the spatial approach can lead to numerous windows onscreen at once) and makes use of navigation similar to that found in a Web browser. When you open a folder, its contents appear in the current window instead of a new one. "Forward" and "Back" functions allow you to step through navigational history, while an "Up" button may be available to view the contents of the parent folder. Without one-to-one mapping between folders and windows other features are possible, like a sidebar showing the folder hierarchy. Most of today’s file managers - including the Mac OS X Finder, the Windows Explorer, and KDE’s Konquerer - fall into this category to some degree.

These approaches are not entirely mutually exclusive. For example, a "navigational" file manager can still maintain consistent positions for file and folder icons. The debate centers on those spatial features that preclude navigational tools and increase screen clutter; and those navigational features that "break" the spatial metaphor. Of particular importance are one-to-one mapping of folders to windows and its corollary, the opening of each folder in a separate, unique window.

The Problems

It is sometimes frustrating to be a user interface designer, because everyone has intuitions about how to design interfaces. I can’t even trust my own intuition at times: Just because I think a certain way doesn’t mean everyone else does, and that requires care. Everything I have read leads me to believe that the spatial debate, including its very existence, is based primarily on the intuitions of those involved. This is a poor way to design a user interface when the issues at stake are so complex.

Let me elaborate. Here are some quotes from Siracusa’s article:

"Although mentally visualizing abstract relationships and hierarchies is a skill familiar to computer programmers and other people who frequently use systems that require such mental gymnastics (e.g. the command line), it is not a skill that most people possess in any significant quantity."

"Visual/spatial cues are also a better catalyst for long-term memory. Remembering someone’s face is a lot easier than remembering someone’s name, for example."

These should worry you. He’s making broad statements about human behavior and cognition without supporting evidence. He may be right; he may be wrong. We just don’t know, but we certainly shouldn’t take him at his word. We might like to know how he reached his conclusions, and if he has evidence to support them. Regarding the quotes above:

• He says most people have trouble visualizing abstract relationships or hierarchies, and visual/spatial cues are a better catalyst for long-term memory. How does he know? What studies support those conclusions? In the absence of evidence, why should we find those conclusions more plausible than their alternatives?
• He uses the command line to exemplify hierarchical user interfaces. That seems like a scare tactic to me. There are other, familiar controls (like an expandable tree) that provide direct access to a hierarchy. One was built into the Mac OS 9 Finder.
• What "mental gymnastics" do we do to deal with hierarchies? Why should hierarchies require mental gymnastics but not spatial relationships? Is it safe to make such generalizations at all? We often hear that people think differently from one another. How much might this vary from person to person?
• Remembering someone’s face is probably easier than remembering someone’s name, but it’s a misleading example. It’s not simply a question of visual cues: We have an area of our brains devoted to face recognition, the Fusiform Face Area (Kanwisher et al, 1997). But that probably isn’t much help with icons and windows.

I don’t mean to single out John Siracusa. The same type of argument can be found throughout the discussion:

"I have seen, over and over again, that novice users (and often enough users who have used computers for several years!) don’t get the concept of a file hierarchy. They get lost as soon as they open Windows Explorer. They call me because they don’t find the attachment they just saved from an email." ("How to Deal with the Spatial Paradigm", 2004)

Indeed, hierarchies are a recurring theme in the spatial debate. The problem is that there are too many variables. Do users have trouble with file hierarchies, with the particular hierarchy in question (in this case, the Windows filesystem), or with the tool used to navigate it (Windows Explorer)? The difference in filesystem hierarchies between Mac OS X, Windows, and Linux is dramatic. Apple has tried to create an accessible filesystem hierarchy:

• On a Mac, you can find Internet Explorer at /Applications/Internet Explorer. In Windows it’s C:Program FilesInternet Exploreriexplore.
• On a Mac, shared libraries tend to be named things like CoreServices.framework and located in places like /System/Library/Frameworks. In Windows, they tend to be named things like mspmsnvs.dll and located in places like C:WindowsSystem32.

Windows and Linux have responded to a cryptic hierarchy by hiding it behind other hierarchies like the Start menu. Windows goes as far as to warn users against looking in system-related directories. Mac OS X doesn’t have a Start menu. It has Go > Applications, which results in a window showing your /Applications folder. The point is not to bash Windows or promote Mac, but to urge care in drawing conclusions about human aptitude for hierarchies. Bad Windows Explorer performance in itself isn’t evidence of poor hierarchy comprehension.

"I cannot resist myself so if you are looking for a insightful, well researched article you can cease reading here. Now back to the subject at hand." ("Navigational Man versus his Evil Archnemesis Dr. Spatial", 2004)

"A spatial file manager is closer to the way the people interact with the physical objects around them, such as desks, chairs and tables, and is easier for new computer users to understand and work with." ("Spatial file manager" entry in the Wikipedia)

The above quote from the Wikipedia is interesting since a subsequent paragraph provides support for a contradictory conclusion.

There are similar examples in the GNOME discussion lists, but I don’t want to single out people who didn’t publish their thoughts explicitly.

I agree, at least in part, with some of these statements. That doesn’t excuse their lack of support. The issues are complex, and if we are to address them we need real answers. If those making the arguments have support for them, they should cite it (and a few participants have); if not, there’s no good reason to believe them - particularly if the arguments don’t resonate well as you read.

You’ll notice that most of the examples above support the spatial metaphor. I don’t mean to single one side out: I simply can’t find many equivalent arguments on the other side. It may be because the spatial advocates are the ones challenging the status quo.

Complexities and Trade-Offs

I don’t think the spatial folks have it right. I also don’t think the navigational folks have it right, but the spatial folks are generally the ones insisting on an either/or, black-and-white decision, arguing that anything less than a complete implementation breaks the spatial metaphor. And, indeed, this is where I believe their error lies. But you don’t have to trust me or anyone else. Empirical methods exist. We can test how users respond to large numbers of overlapping windows, hierarchies, and spatial and navigational tools. We can test the trade-offs and learn which pros might outweigh which cons. This, more than anything, is the source of my frustration: Tools exist to gather real evidence about these questions, but no one seems to be using them. If some of the effort devoted to argument and speculation were instead used to employ such tools, we might have more answers by now.

The evidence that does exist paints a complex picture. A 1995 study by Microsoft found that:

• All but the most advanced users did not understand how to manage overlapping windows efficiently.
• Beginning users were bewildered by the hierarchical file system.

In other words, novices had problems with hierarchies and with too many windows onscreen. Choosing between spatial and navigational, we would be stuck with one of those drawbacks. And while a "pure" spatial metaphor precludes any hybrid solution, I am not convinced of that necessity. Again, though, there are other variables to consider. The study was limited to a single hierarchy and a single mode of interaction with windows. An alternative to either might have changed the results.

In addition, novices are not the only users. If a particular user interface requires a brief adjustment but is more efficient for long-term use, it might be an appropriate choice. As study’s authors note, "As we started working on new designs, we hoped to avoid the classic ‘easy to learn but hard to use’ paradox by always keeping in mind that the basic features of the UI must scale."

A 2004 study by Ravasio et al (brought to my attention by a participant in the GNOME discussion) observed users over a period of time to determine how they used their filesystems. The results were interesting: For temporary, short-term storage users tended to put files on their desktops and to group and arrange them spatially. For long-term, archival purposes they created hierarchies. The study included users of both Windows and Mac OS 9. Here, again, we see a combination of techniques: spatial, non-hierarchical navigation for short-term tasks and hierarchical navigation for long-term. Yet questions remain: How much of this behavior was dependent on the nature of the tools available to the user? What would these users do in an environment like XFCE, which has no desktop, or Mac OS X, which provides both a desktop and a dock?

I bring up these examples to highlight the complexity of the situation. For every benefit of a particular method, there may be drawbacks. So it’s a trade-off: To compare methods, we must balance both the benefits and the drawbacks. The question isn’t, "Which way is right?" but, "Which way is righter?"

The Mac OS X Finder, source of Siracusa’s original ire, provides an intriguing hybrid that I suspect is the result of some empirical testing. Folders open in the current window, so there is no "pure" spatial metaphor. But window sizes, positions, and states are preserved whenever possible, and there is some one-to-one mapping of folders to windows: If you open a folder whose contents are already displayed, its window is brought to the front. For example, if you open your Applications folder and then try to open it again, the existing window is brought forward. If you open a sub-folder inside that window and then open Applications again, a new window is created for Applications (since the old window is showing something else now). This is not spatial, particularly in the most literal sense. But it preserves one key feature of spatial interfaces in a way that can coexist with navigational tools like Back and Forward buttons.

I’m not suggesting the OS X Finder is perfect. There are numerous details that could be refined, and there may be a better way altogether. But in considering spatial vs. non-spatial approaches a hybrid is worth examining, and the OS X Finder seems like an excellent one. It manages to achieve many of the benefits of both approaches while avoiding many of the drawbacks.

Hypotheses

I have my own suspicions about the answers. I’ll summarize them briefly because I believe they are centrist opinions in a fairly polarized discussion; empirical testing needs hypotheses to test; and I like my opinions and hope you will too. Bear in mind, though, that these are hypotheses only:

• Users probably comprehend some hierarchies well, but their success will vary based on the user interface and the hierarchy itself. Presumably they will be most comfortable in hierarchies they create themselves.
• Users should have multiple ways of getting to information. This is not only because users may think differently from one another, but because a user may think differently about a piece of information in different contexts. Thus there may be a place for a spatially-oriented desktop or temporary storage area; a dock or panel; a hierarchical tool; a desktop search tool; and other interfaces. These shouldn’t be divided into "basic" and "advanced" tools, but can be provided side by side as alternatives for all users.
• If the spatial questions are answered by good, thorough empirical testing, I suspect that neither camp will win: A hybrid will turn out to be best. Again, I believe Apple has tried to construct such a hybrid with the Mac OS X Finder. But I also think there’s room for improvement. In particular, I suspect the Windows Start menu provides a better first-time computing experience than the Finder. And of course, the spatial and navigational methods aren’t the only options: Innovators may find new and better techniques altogether.

Innovation

Revolutionary innovation is difficult and often ill-advised in this space, given strong de facto standards. But there’s plenty of room for incremental innovation and innovative new features.

Siracusa proposes some innovations in his article. Bruce Tognazzini suggests a few in "Apple Squandering the Advantage" (2000). Numerous innovations have been suggested on the GNOME usability list and elsewhere. In the heat of the debate, we sometimes forget about innovation. Yet if our goal is to improve the user experience, we should certainly look beyond established paradigms for the solution. Our best option in the long term may be neither spatial nor navigational.

Caveats

Empirical testing can identify good usability and poor usability. It can compare the usability of two systems. However, one cannot design by testing: Someone needs to take the results and synthesize a new design in response. Similarly, empirical testing may not be effective in creating an aesthetically pleasing product, and aesthetics are themselves critical to usability.

In criticizing the spatial discussion to date, I don’t mean we should stop speculating. Siracusa’s original article, the articles that follow it, and the ongoing discussion in the tech community raise questions, hypotheses, and new designs, and are essential to finding good solutions to fundamental problems. In the end, I have two basic criticisms:

• We must be careful in making or trusting statements about how people think and which user interaction paradigms work without adequate empirical support.
• We have very little empirical support for our hypotheses and could take steps to remedy that.

Conclusion

File management is central to the computing experience, increasingly so as more activites become computerized and the volume of information increases. The tools with which we manage and locate information have not changed substantially in years, though new tools (such as desktop search) are starting to change the landscape.

So reevaluating file management from the ground up seems like a good idea. The spatial debate and other discussions in the IT community are integral to that process, and I would never suggest otherwise. However, many of the arguments rest on generalizations about human behavior and usability that remain untested, despite the existence of established, empirical methods. If we seek good solutions, we should begin testing our hypotheses and assumptions. We should also look beyond comparisons of established methods to new, innovative techniques.

It’s easy to sit here and criticize, but of course the open source community is all about getting involved. If you’d like to help gather real data about these questions - by donating money, contributing time, or some other way - drop me a line. If there’s enough interest perhaps we can start making it happen, and I’d be delighted to help coordinate the effort.

This entry was posted on Thursday, May 5th, 2005 at 12:02 pm and is filed under General. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.