Are Dashboards Rude?
This blog post accompanies a paper to be presented at IEEE VIS 2023, “Heuristics for Supporting Cooperative Dashboard Design” written by Vidya Setlur, Michael Correll, Arvind Satyanarayan, and Melanie Tory. For more details, read the paper.
Imagine you are an employee at a high-stakes Glengarry Glen Ross-style sales organization. If your sales numbers dip too low, that’s it, hit the bricks pal. But you don’t have direct access to this crucial performance data. Instead, there’s this guy, let’s call him Dash, and every day, before you sit down at your desk, he stops by every employee to give them his daily thoughts on the matter. He begins, as he always does, by telling you your employee number and what company you work for, and what their logo looks like. He opines on how the weather has been recently. He says that company sales have been up as a whole, and lists the performance of each employee, in alphabetical order. When he gets to you, he says that your sales are good, but your Foo index is way down. Before you can ask him what a Foo index is, or if it means you need to update your résumé, he’s already started gossiping about the other departments. You try to get him to wrap it up with increasingly exaggerated glances at your watch, but he continues talking. Eventually, he moves on without a word and, as you uneasily start your day, you can overhear him giving almost the exact same speech to the person in the next cubicle.
It’s my contention that Dash is kind of rude. Or, if rude is not precisely the right word, then he’s at the very least not particularly useful, and he’s not useful mostly because he’s very uncooperative. He’s giving you a lot of extraneous information that you didn’t ask for, does not give you the extraneous information you did ask for, and gives you information in an undifferentiated mass rather than emphasizing what’s important or cutting out the stuff you’ve already heard before and/or don’t care about. He ignores your input, is disrespectful of your time, and then leaves without giving you anything to show for it except for some conversational dead ends and a vague sense of unease.
My involvement with this paper began with thinking about the extent to which Dash’s daily monologue is or is not a good analogy for dashboards as tools for visual analytics. If a dashboard spends most of its time showing you the same obvious or stale data day after day, if it refuses to provide context or details about the information that is new to you, then maybe it is being impolite to you, in the same quasi-anthropomorphic way that a visualization that treats human death and misery the same way it treats hiring numbers is being cruel.
Central to whether or not this metaphor works for you is the extent to which you buy that a dashboard is part of what we call an “analytical conversation.” Some dashboards, particularly the emerging genre of complex interactive dashboards (with, hey, maybe even literal chatbots embedded in them with which you are meant to converse) provide lots of opportunities for the sort of turn taking and push and pull that are part of what we think of as conversation. But I think even old-school static dashboards are still implicitly or explicitly doing lots of things we think of as conversational, just by virtue of having textual elements, implicit reading orders, and differing amounts of visual saliency and emphasis in their chart elements. A dashboard doesn’t just dump information directly into your brain: it’s presented to you in a particular context, the viewer takes a visual path through the information and then (hopefully), this viewer goes off and does something about the information they saw. That’s a lot more like a conversation than you might think.
I don’t need you with me on this metaphor 100%, but I want you at least partially on board with thinking about how the ways of working with a good dashboard are like having a good conversation. One of the things a good conversation has is stages (we pick out five in the paper). Dash, if he were being more polite, might start out with an invitation (“can you talk right now?”), establish mutual context (“it’s nothing bad, but I did want to give you an update on your sales numbers”), leave some room for give and take rather than hogging all the time (“did you have any questions for me?”), work to get things back on track if there’s confusion or misunderstanding (“oh, the Foo index? It just a measurement of your relative ranking in the department: lower is good!”), and, finally, end the conversation in a mutually satisfying way (“so you’re doing great. I’ll talk to you later when I have more information”). I think that kind of conversation is way better than contextless monotone drone mixed with gossip where you don’t get a word in edgewise.
In the paper we instantiated this conversational thinking into a set of a few dozen design heuristics: principles like ensuring that key terms and concepts are defined or otherwise mutually understood (Heuristic 20), or making sure there’s a clear path (with breadcrumbs!) for the user to follow through the data (Heuristic 30). But, even though we spent a lot of time ideating and refining and workshopping these things, to me the actual heuristics we settled on are less important than this overall analytical conversational lens. What are people clicking on, hoping that it will lead them to more information or more insight, only to have their hopes dashed? Where are people confused by the brick wall of data you’ve thrown in their face? What useless information are you giving them over and over, and what useful information do they have to dig deep to uncover? Where are you hiding the explanations or definitions or provenance information that would make sure your users know what you’re showing them? Just this little anthropomorphic trick I think gets people thinking in a new way.
There’s a lot more in the paper: about the specific heuristics we developed, how they tie into philosophy of language concepts like Paul Grice’s cooperative principles and how to apply our heuristics in a classroom setting. And you can see my attempt to visually represent “furiously clicking” in a figure in a scientific paper, of course. And it’s at least worth a passing note that many of the dashboards people encountered lacked the ability to refine or answer follow up questions, repair misunderstandings, or otherwise engage in the sort of mutual initiative activities that we take for granted in our conversations with people who are trying to give us information — I think there’s a lot of potential future work in exploring what those interactions might look like in a dashboard.
But I really think this conversational lens is the key takeaway here. Not just for designers of specific dashboards out in the wild, but for visualization researchers as well. For instance, if all visualizations are to some extent conversational, we can analyze and evaluate them in those terms: not just what chart elements they use, but how they persuade or ground or function cooperatively. And if many dashboards are rude, why is that? What is going wrong in the way that we teach or practice or value visualization that we’re producing so much that’s annoying or useless, uncooperative or unhelpful?