Dashboards: where they came from, why most aren’t fit for purpose and what to do about it

In this blog I’d like to discuss dashboards: where they come from, why most are seriously unfit for purpose and what to do about it.

Some history

The modern web based dashboard has its roots in tools known as executive information systems. EIS have been around in one form or another since the mid 1980s but although the early vision was good, the first attempts foundered because data storage and processing technology was just too expensive or not up to the task.

An early 1980s 1Gb drive

As a result, the few systems that were created remained the preserve of the mega corporation.

By the mid-nineties data warehousing and OLAP (online analytics and processing) were beginning to get some traction on the problems of storage and analysis, but little consideration was given to the presentation of data. Reporting required database experts and the output was usually in paper form.

Around the same time the business community latched on to the use of key performance indicators – an idea first developed by Prof Carol Fitz-Gibbon, an English academic working on the A-level information system. As use of KPIs became more widely adopted in the late 90s the EIS began to re-emerge, this time sporting near live data feeds and reinvented as the dashboard.

Regrettably, in the excitement of presenting ‘live’ data nobody stopped to think too much about the purpose of the dashboards themselves and the metaphor was carried too far. A fashion emerged, which still holds sway in many quarters today, of emulating physical dials and gauges from cars or factory plant. Higher resolution displays with greater color depth encouraged the production of photo realistic displays. Screens twinkled with polished chrome like 1960s car dashboards. Some even sported gnarled walnut backgrounds. Where the break with the car dashboard metaphor was made, the special effects trend persisted and walnut and chrome gave way to brightly coloured 3D charts displaying 2 dimensions of data.

The purpose of any information system should be to present to the user that which he or she needs to know as quickly and concisely as possible. That was the laudable goal of the early EIS systems, but dashboard bling does nothing to further it.

There are two classes of information – that which users know they want and that which they don’t know they want (these equate loosely to Donald Rumsfeld’s famous ‘known unknowns’ and ‘unknown unknowns’ ). Users seeks the former to answer particular questions that they have in mind, but their attention needs to be drawn to the latter because something unexpected has happened. This is exception reporting. So a dashboard must perform two quite different functions – regular information must be easily found and understood whereas exceptions must be bought to users’ attention even when they are not looking for them. This is where the fashion for photo realism starts to cause problems. Understanding of human visual perception has moved on quite a bit since physical instrument displays were initially designed and the reworks of these old displays do not take full advantage of this fact.

Strange truths about human vision

Before discussing what makes one dashboard good and another poor it’s worth spending a few moments exploring some rather surprising aspects of human vision. This might seem like a bit of an indulgent segue, but it’s actually fundamental to a proper understanding of what makes a good dashboard work.

Our everyday perception suggests that we have excellent high definition vision over a wide field of view. The reality could not be more different. In actuality our vision only has high resolution over a very narrow central portion covering an arc somewhere between one and two degrees wide. To make matters worse we perceive colours only in the center 15-20 degrees – our periphery vision is completely monochrome. By way of compensation the rod cells which account for all peripheral vision are extremely sensitive to lite, requiring as few as six photons to fire. This is the reason why a passengers on a ship approaching a dark shoreline at night will often first see land out to the side instead of dead ahead. The rod cells are also more sensitive to motion than the color sensitive cone cells, which is something we will come on to in a moment. There are about twenty times as many (black and white sensitive) rod cells in your eye as there are (color sensitive) cone cells.

We compensate for this tunnel vision with frequent sudden eye movements called saccades momentarily switching the high definition portion of our vision between different parts of the scene before us. The actual signal being fed to the brain is something like the picture you might obtain by sticking a six inch cardboard tube in front of the lens of a camcorder and then waving it around semi randomly in a series of high speed jerks. The brain somehow keeps track of all this mayhem, stitching the various image fragments together in a way that gives us the sensation of continuous wide angle high definition full color vision. You can see some parallels here with the MPEG codecs used in video streaming. The idea is to only send data about changes in the scene and ignore static parts of the image. This line of thinking suggested that the reason behind the evolved design was a limitation in the bandwidth of the optic nerve, however more recent experiments suggest that something else is at work here. Current estimates put the bandwidth of the optic nerve a little below 10Mbps – hardly fibre optic, but not too shoddy either. It now seems more likely that the real reason has more to do with power consumption. The brain is an extremely power hungry organ. It accounts for just around 2% of body mass but 20% of energy consumption. It is simply not desirable to continuously process large parts of an image that are not of interest and not changing.

However, in order to survive, most animals need to be able to react quickly to motion at the periphery of their vision e.g. hungry predators. This is where the rod cells come in handy. They may not see color, but they are present in large numbers away from the fovea (center of the retina), can see in very low lite and better still have built in edge detection capability. This pre-processing means that signals of interest – moving edges or changes in lite can be singled out for onward transmission to the brain, filtering out the background clutter which is not of immediate importance.

Not only is the retina selective in the information that it sends to the brain, but the lower regions that receive it are selective in what they pass up to the neocortex which is where our consciousness resides. There is a large amount of what is referred to as ‘pre-attentive processing’ which blocks the majority of available input from reaching our consciousness and presents only fragments which are salient to our current thoughts or other goals like avoiding being eaten.

I started out by saying that there are two basic types of information presented in dashboards – ‘in gauge’ states (the regular numbers) and ‘out of gauge’ states (exceptions). Our little foray into the physiology of the human visual system suggests a couple of points to bear in mind when wishing to bring exceptions to a user’s attention in a display that may contain a large quantity of other information: If we want to catch attention, it is primarily the rod cells that we must cater to, because the dashboard may not be the current subject of attention when the alarm arises. Rod cells don’t see color, but they do see change in intensity and they do see changes in edges and movement. It’s therefore more important to change the luminance of a color and the weight (boldness) of the figure than the color itself. Rather ironically we have traditionally taken the color red to signify ‘alert’ but rod cells don’t even see red, so if a lite changes from green to red with the same intensity at the periphery of your vision, the only reason you notice it is that it will appear to have just gotten dimmer (our eyes are not as sensitive to red as they are to green) until that is, your gaze shifts directly to it and your cone cells get to do their stuff !

We can take this much further though. So far we have only considered effects that stem from the arrangement and properties of the different types of cells in our retinas.

We touched on pre-attentive processing. i.e. the subconscious processing that goes on in the background without our awareness. This happens in a massively parallel way over the entire visual field extracting such basic features as line ends, tilt, curvature, color, contrast, size and closure. Experiments using patterns with a large number of distractors and a handful of embedded target patterns show that processing time is almost independent of the number of distractors.

By taking some text that is not readily understandable – the classic lorum ipsum used in the design industry to provide dummy text for layout – we can readily illustrate this point.

See how many times the character sequence ‘eu’ appears in the text below….

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Fusce cursus dui id felis vehicula iaculis. Quisque molestie congue gravida. In eu tortor metus. Sed tincidunt magna imperdiet dolor porttitor, in iaculis enim tincidunt. Quisque vel ligula interdum, viverra ex ut, tempor est. Mauris scelerisque, arcu vel maximus eleifend, neque neque tincidunt ipsum, at feugiat tellus dolor eget ante. Morbi vulputate feugiat efficitur. Nullam tincidunt mi vel purus tempus, in viverra nulla hendrerit.

I bet that took more than a few seconds (irrespective of your political views)…

Now try again.

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Fusce cursus dui id felis vehicula iaculis. Quisque molestie congue gravida. In eu tortor metus. Sed tincidunt magna imperdiet dolor porttitor, in iaculis enim tincidunt. Quisque vel ligula interdum, viverra ex ut, tempor est. Mauris scelerisque, arcu vel maximus eleifend, neque neque tincidunt ipsum, at feugiat tellus dolor eget ante. Morbi vulputate feugiat efficitur. Nullam tincidunt mi vel purus tempus, in viverra nulla hendrerit.

The contrasting color allows us to effortlessly pick out the occurrences of ‘eu’ this time – which is exactly what we need for exception reporting. Note this only works when the ‘exceptions’ are few. If everything on the dashboard is lit up like a Christmas tree all the time, then we won’t be able to easily perceive changes and will back at square one again. The rule here is to steer away from strong color and emphasis – except when you need to bring something to the users attention.

While just about all pre-attentively perceived attributes are useful for making information stand out, only line length and position are good to accurately convey quantitive measure. Thus while pie charts may look pretty, they are not good for displaying differences in values – bar charts are way better for this.

Gestalt

Drawing on the findings of work done in early part of the last century, we can find more pointers to effective dashboard design.

In 1920’s Germany Max Wertheimer, Kurt Koffka and Wolfgang Köhler from the Berlin school of experimental psychology produced the Gestalt (literally ‘shape’ or ‘pattern’) theory of mind. The Gestalt theory aims to describe the ability of our brains to automatically generate groupings of forms. They showed that there are a number of simple principles by which the brain classifies visual patterns presented to it.

This has bearing in the presentation of information because it informs ways to effectively emphasise some data over other and to imply relationships.

There are a number of principles quoted which vary according to the source you choose (see e.g. http://en.wikipedia.org/wiki/Gestalt_psychology#Gestalt_laws_of_grouping) , but perhaps the most important from the point of view of the dashboard designer are:

Proximity

We perceive a horizontal grouping of three rows because the dots are closer to each other horizontally than they are vertically

Similarity

Here we see the lite and the dark dots as belonging to different groups

Enclosure

Here the background tint or an enclosing line suggest relationships between the elements

Information should be grouped by purpose. Group data that belong together by putting them a box with a lightly shaded background or a bounding line around it. Avoid strong color or 3D effects – these add nothing to the data and compete for attention. Edward Tufte’s famous maxim – ‘maximise data pixels, minimise all other pixels’ holds true. If you must use color in the base design, use pastels.

Putting it together

Combining Gestalt principles with judicious use of color to draw attention to out of gauge measures gives us the makings of a good design.

Here attention is drawn by use of a different hue. But we can do better.

Below is the same image completely desaturated – as seen by our rod cells in the periphery of our vision.

Note that we can still pick out the alerting cell – especially if we look at the image obliquely rather than straight on.

If the transition to an alerting state happened when the dashboard was only just within your field of view, you might well still notice that something had changed but it would be mainly due to the sudden change of luminance rather than the change of hue.

In order to be really noticed, we need to take advantage of some of the other capabilities of our pre-attentive processing system.

By increasing contrast in the border line when signaling alerting it’s possible to make it all the more obvious…

…even for peripheral viewing

Other ways of effectively claiming attention include small changes in size of elements relative to those around them and flashing (although the latter should be reserved for the strongest form of alert as it becomes seriously distracting).

What do you want from your dashboard?

There are two distinct types of dashboard: operational and executive. An Operational dashboard’s primary focus is normally on exception reporting whereas executive dashboards display KPIs.

This means that operational dashboards may display different sets of metrics at different times and benefit from real-time or near real-time updates whereas executive dashboards tend to have relatively static displays changing over hours or days with the same KPIs always on view. Of course, a fusion of the two is always possible.

Three rules

  1. Identify your audience. Careful consideration about who will use it and why will pay dividends. It is almost always better to aim for a number of simple dashboards targeted to different user roles than to try to be all things to all people.
  2. Keep it simple. Less really is more. Show top level indicators only, making use of Gestalt principles to group information and draw attention to out of gauge measures. Use drill down features to give access to the details where this might be useful rather than causing distraction by displaying everything all of the time.
  3. Make every pixel count but don’t cram the space.  Avoid over-prominent logos and eye candy such as 3-D charts displaying 2 dimensions of data. They add nothing to the usability of the system and generally only detract by the distraction that they cause.
Facebooktwittergoogle_plusredditpinterestlinkedintumblrmailby feather
Comments are closed.