Top banner ad available.

interaction-cost.jpg
Some of you might not have noticed the difference but there exist two big branches of visualizations: those used to communicate some concepts or findings to an audience and those used to explore data. For this second type user interaction is vital. We tend to believe visualization, for its name, is all about visual phenomena, like colors, shapes, textures, and so on. But who knew? A very large part of the power of visualization actually resides in the effectiveness of its interaction strategies.

Heidi Lam (formely at UBC now at Google) published a paper a couple of years ago at the IEEE InfoVis Conference titled: "A Framework of Interaction Costs in Information Visualization" (PDF). She realized that every time we interact with a visualization, interaction has a cost. If we want to develop effective tools, we have to make sure we know very well what these costs are, where they take place in the analysis process, and importantly, how we can minimize them.

The framework is based on the very well known Donald Norman's Seven Stages of Actions model, which describes how people interact with machines and their interfaces (think oven, refrigerator, car). This model in particular points out where interaction costs and problems can hide. Namely, in the execution of a task (Gulf of Execution) and in the interpretation of the results once the actions have been performed (Gulf of Evaluation).

The Interaction Costs

Cost of Decision. This is not included in the original Norman's model but is clearly relevant in information visualization. Complex visualizations often provide initial overviews without clues on what to do next. How do we form a goal? And how can the system help focus on some initial interesting data segments?

Cost of System Power. Once people have a goal in mind, they need to translate it into operations. If the interface is too complex or does not provide sufficient clues, the user might be stuck or lost. And, let's admit it, visualization interfaces can be very complex or have esotheric behaviors at times, so this is definitely a relevant cost to take into account.

Cost of Multiple Input Modes. I like this cost because it is very specific to information visualization. In order to deal with complex analysis tasks, designers and researchers had to invent complex interactions that require multiple modes or overloaded functions, but this, of course, often makes things too complicated. Think multiple-views or zoom & pan interfaces.

Cost of Motions. This is where the real action takes place, that is, when you move your finger and arm to point, click, drag, and so on. Here, the cost mainly refers to the ease with which certain targets can be reached (you might want to get acquainted with Fitt's Law if you do not know what it means). Sometimes we design our visualizations with targets that are too small to be reached, or easily occluded, or, even worse, they are moving. Other times an operation requires tens of clicks to be execute. All this add up to cost, and is of course bad.

Cost of Visual Clutter. Once an operation has been executed at the interface level you normally expect the visualization to react and you want to perceive the state of the system. Visual clutter can hinder this perception.

Cost of View Changes. Many interactions result in a view change in your visualization and this has of course a cost because each time the view changes you have to re-interpret it. Zoom & Pan is a very common example of interaction that triggers a view change.

Cost of State Changes. Often, when analyzing data, people go through a series of different states which require an intricated set of actions and operations. Every time the state changes there is a cost associated with being able to underatand the current state and especially being able to go back to previous ones.

Reducing the costs

Naturally, the paper discusses some strategies on how to reduce the costs. This is hopefully what you, dear reader, can use in practice in your own interactive visualization designs.

Again, they are based on the orginal Norman's model where he suggests to bridge two main gulfs: the Gulf of Execution and the Gulf of Evaluation.

Guidelines

(1) Less is More: You might have heard this advice a thousand times already, but surely it is never enough. I totally agree with this and I am often in struggle with myself when I know I have to remove features and not add more in a new design. The execution of any operation will always be easier and more memorable if the number of choices and modes is small. So, make your interface as minimal as possible and people will thank you.

(2) Keep the User in Control: As the visualization and its operations become more complex, as you add features (which by the way is against the first guideline), you might be tempted to let the interface adapt or perform automatic operations to provide the best views. This is of course desirable in many instances, but users tend to need to feel in full control and be able to always diagnose any inconsistency. Do not let your users distrust your interface.

(3) Do Not Trust Fancy Interaction Techniques: The Infovis literature is full of ingenious interaction techniques, and you might be tempted to employ them right away for their coolness factor. While some benefits might exist in using features like link and brush, fisheye views, complex animation schemes, you should be warned that people might get lost very easily. Or, at least you must account for some time before your audience get accustomed to it.

(4) Support Refinding: While people interact with an interface during data analysis, it is very common to flow from one state to another effortlessly. Then, at some point, it is necessary to store what you have accomplished, or come back to previous states, or change again for a new path. The design of the interface and the interaction capabilities must allow for this level of flexibility and help people reason on the process and navigate through it.

How you can use this framework?

Finally, let me spend few words to direct you on how you can use this framework in the context of your projects.

Preliminary assumption: you are designing an interactive sytem. While some of these guidelines might be helpful for static visualizations (less is more is one example) the best value you get from these is for interactive interfaces in particular.

The framework can be used in design and evaluation.

While designing a new interactive system do yourself a favor: print out a copy of the diagram that summarizes the costs and the gulfs and stick it on your wall.

When you think about a new feature or operation to add, refer back to the costs and the guidelines and make sure they influence your choices in a positive way. Try to think of a person going through all the stages of the model while solving a data analysis problem with your tool. Where are the traps? And how can you reduce the chances people fall into them? This will be extremely helpful.

If for any reason you are evaluating an existing product (maybe an old version of a software you have to re-design?) the process does not change much. Make sure you go through all the stages of the model and see where the traps can be. In this case do yourself another favor: do not just think about it, do it! Take some data, load it in the tool and try to solve a problem. Usually, problems become evident.

Conclusion

One thing I did not mention yet is that this research has been conducted by performing a thorough analysis of more than 480 papers (my compliments to Heidi for having done all this alone!), while the costs are distilled from a set of 32 papers. You can always refer back to them for more information at: http://www.cs.ubc.ca/~hllam/res_icost.htm

I hope you will find this model useful for your ideas and projects. You can of course always look into the details of the paper to get more inspiration.

This post has been written by Enrico Bertini. He is a researcher in the visualization and data analysis group at the University of Konstanz. He regularly posts his ideas, reviews, and experiments in his blog fellinlovewithdata.com, where he tries to bridge the gap between academia and the real world out there. If you have any doubts or question you can contact him on Twitter at @FILWD.