July
14
2008
Or, let the beatings begin.
I started a new gig just about six months ago now and one of the goals I set for myself was to not fall into the trap of using bad visualizations. Have I been successful? Hard to tell yet, but it has taken an interesting turn – which could easily be defined as a failure 
. I’ve been proselytizing the use of Bullet Charts as a responsible replacement for gauges and dials. I’ve also been fostering the use of sparklines and clean charts in general. I’ve done really well with instituting sparklines and have raised the charting game of my co-workers.
However, Bullet charts have just not taken to the water.
I’ve said before that bullet charts faced a tough road. Basically, they are different enough to require end-user training which violates my the rules for effective chart design.
That being said, I’ve campaigned heavily for their use over the past 6 months but ultimately, the report users just weren’t groking the chart and continually made mistakes in interpreting them. That’s not good. Misinterpretation can lead to the wrong insight or answer which leads to the wrong action.
So my boss says “Look, just give them a gauge – how about a speedometer chart?”
-ABSOLUTELY NOT-
Then I explained all that was wrong with using speedometers in a business context and beyond that how hard they are to create and maintain.
That still left me with a problem though. the bullets weren’t hitting the target (har, har, har) and I needed a replacement. So, I grumbled a bit. Sighed and starting thinking about a gauge chart.
First, what are the typical problems with gauges as implemented in business dashboarding?
- They take up too much space (bad information-to-ink ratio)
- Low fidelity (data is not displayed with a high level of detail/resolution)
- A tendency to be visually imprecise – easy to tell if it’s bigger than a breadbox or smaller than a Volkswagen but assessing the actual value with any accuracy can be difficult
- May suffer from visual effects that create inaccuracies in interpretation (e.g. perspective, shadowing etc. create visual lies)
- All the other ones that I couldn’t think of – off the top of my head
OK, so to solve this correctly I needed to:
A. Maintain a (relatively) high information-to-ink ratio
B. Show the highest possible fidelity in the data
C. Show data precisely
D. Eliminate/Minimize distorting visual effects
Inspiraton: The picture in my head (hazy as it was) looked like some kind of decibel meter but this picture of an old pyrometer is also pretty close.
Photo Courtesy of RightBrainPhotography (c)
This old gauge is
- highly precise (to 1 degree Farenheit)
- relatively compact
- contains enough meta information to ward off imprecision in reading it
What’s missing?
- can’t get away from red, yellow & green in a business dashboard
- related marker values – minimum, acceptable, target and goal
Toolbox:
So what will we need to get the job done?
- Excel 2007
- Horizontal Bar Chart
- X/Y Scatter Chart
- XY Chart Labels Addon
There are many, many steps involved, which are covered in the xlsx download at the end of the post. The instructions assume a medium-to-expert level knowledge of charting in Excel. So, in order to keep this post relatively short, I won’t go through the steps here and leave that up to the download. Instead let’s take a look at the completed graph.
So here it is and it has only the vaguest relationship to the image above – right? About the only thing they have in common is the pointer.
BTW, I kinda like how Excel 2007 lets you use line ends (don’t remember that being in 2003/XP).
Alright, so what do we have going on here?
The chart is bounded by a box in a Dashboard. This box, by its title is “Page Views – Month to Date” so we know what we’re looking at. Immediately below the title we have the current count of page views (yeah, not a KPI [yet]) at 5 million in big, big text.
We can see that we’re in the red but closing on yellow. However, if we stopped here with no other cues it would look like page views are more than 50% along because our eye sees that the pointer has traversed most of the red.
But, we’re not done! There’s a custom X-Axis (courtesy of an XY Scatter) that the pointer points to and it shows that page views have traversed exactly 50% to the overall goal.
Of course, we could do the match in our heads (5 million is 50% of what?) but we’ve also got custom labels on the bar chart (our gauge background) so we know that our goal is 10 million page views, the minimum is 7 million and the target (minimum ‘acceptable’) is 9 million page views. Each of these markers in the gauge background are tied to a point on the x-axis so that we always know, with a high degree of accuracy, where we are, and where we want to get to. And notice how our non-KPI measurement of 5 million page views turns into a simple KPI as we measure it in terms of a goal – we’re 50% of the way there!
Lastly, this chart is no harder to maintain than a bullet chart is (actually, it’s just a bit easier/simpler).
So here are some questions for you:
Do you like the chart? Why/Why Not?
Is it a good chart? Why/Why Not?
What solutions have you used in the stead of something like a bullet chart because users just couldn’t get it?
Oh and here’s the file for you.
Update: After multiple requests, here’s some side-by-side screen shots of the line-gauge and a bullet chart based on a bullet chart I have blogged about previously.
As a bonus here is Peter Walker’s quick mockup
Now you have visual comparisons…
Tags: bullet charts, data visualization
[...] blog.instantcognition.com wrote an interesting post today onHere’s a quick excerpt Or, let the beatings begin. I started a new gig just about six months ago now and one of the goals I set for myself was to not fall into the trap of using bad visualizations. Have I been successful? Hard to tell yet, but it has taken an interesting turn – which could easily be defined as a failure
. I’ve been proselytizing the use of Bullet Charts as a responsible replacement for gauges and dials. I’ve also been fostering the use of sparklines and clean charts in general. I’ve done really we [...]
[...] Original post by Clint [...]
Let me understand, “However, Bullet charts have just not taken to the water.”???? Wow, I just do not get that. As soon as I read about an Excel version of Bullet Charts from Charlie Kyd’s excellent http://www.exceluser.com/explore/bullet.htm website in July 2006 I was amazed at how easy they were to read and understand in Excel. Of course, Charlie referenced Stephen Few’s two excellent books and reviewed them in March 2006 http://www.exceluser.com/dash/chartbooks.htm [For those who do not know, Stephen Few is a data visualization expert who developed the Bullet Chart in response to the "speedometer" type dials in dashboards and is influenced by the works of Edward Tufte the developer of sparklines.]
I have been a student of Steven Few’s books and web site, http://www.PerceptualEdge.com ever since March 2006 and I was lucky to be 1 of 60 people who attended his 2008 East Coast Visual Business Intelligence Workshop where we spent three packed days not only going over his material but critiquing each others work. We reviewed many dials and gauges dashboards and shot them down left and right and replaced them with with much better visuals we reviewed in the workshop.
I just do not understand how your solution is not considered confusing. If your users found bullets confusing, why are they not finding your solution just as confusing?
Also, I think this blog post would have been better if you showed both your bullet solution and non bullet solution side by side. Or even better, shared all the raw data you were trying to encode and ask your readers to create a crowd source solution.
For those who would like to read Stephen Few’s revised 5-Page Bullet Graph Design Specification, use this link to Stephen Few’s website to read his Adobe Acrobat PDF document:
http://www.perceptualedge.com/articles/misc/Bullet_Graph_Design_Spec.pdf
@dmgerbino
In my post I missed typed in my second to last paragraph.
Original:
Also, I think this blog post would have been better if you showed both your bullet solution and non bullet solution side by side. Or even better, shared all the raw data you were trying to encode and ask your readers to create a crowd source solution.
Corrected:
Also, I think this blog post would have been better if you showed both your bullet solution and non bullet solution side by side. Or even better, I liked that you shared all the raw data you were trying to encode and asked your readers to create a crowd source solution.
Sorry for the confusion.
@dmgerbino
These charts aren’t too hard to maintain if you spend a lot of time setting them up. This is more a measure of one’s OCD than their Excel skills…
I’m not wild about the new gauge. I’ve sworn off gauges myself. You’ll notice that I’ve removed the speedometer tutorial from my web site, for at least a couple of reasons. First, it is a poor display device, for all the reasons people have been giving. Second, I received too many questions from people who don’t understand enough math to customize it. It’s algebra, a little trig: find a tenth grader to help, people!
On the plus side, your gauge is linear instead of circular. Also, the new Excel 2007 feature is nice that allows arrowheads on series lines, axes (I think), and error bars. One of the few nice things about Excel 2007 charting, and not worth putting up with the problems.
On the minus side, as you stated, the 50% is misaligned. And what if you go to 11? More important, there is no historical context. I can’t see how I’m doing lately, nor even how much of the month has elapsed. A line chart with X ranging from 1 to 31, and a stacked area background with your colored regions, would take up the same space, more or less, but provide a wealth of additional information.
Hmmm, good blog topic. I’ll get back to you.
While I agree that most charts should not require any end-user training, I some exceptions must be made when it comes to Bullet Charts. One must remember that the line chart and the bar chart, as we know them now, have been around since 1786 when they were invented by William Playfair. Anything that is that old and is that ubiquitous becomes common knowledge for people to use. Bullet charts do not have that kind of history and they should be treated as such. The business world and the Data Visualization world will have to go through a learning curve until everyone is used to them.
As far as the chart you have created… To be brutally honest I do not see any advantage of this chart type over a thermometer type. After all, the thermometer has been around since 1617 and everyone knows how to read that. Your gauge portrays the same amount of visual information as a simple number or a simple percentage of Year To Date Goal.
I too was lucky to be one of the 60 people who attended the 2008 East Coast Visual Business Intelligence Workshop. There the gauge was brought up over and over again and each and every time they were shot down. And a bullet chart is no more difficult to maintain than any other chart in Excel. Anyone who uses Excel on a regular basis knows that there are quirks that one must deal with but with some ingenuity and a few more ideas one can make excel create some great charts.
@dmgerbino
David, thanks for stopping by and your comments.
Personally,and I think to anyone with even a minimal familiarity with information visualization that bullet charts make perfect sense. However, for the users in my business it just doesn’t work for them. They kept making mistakes in interpretation which could be disastrous. After 6 months of concerted effort I had to make a change.
The response from users to the gauge has been positive. Because they are familiar with gauges they can interpret it correctly without me walking them through it time and again.
BTW, I think you’re making my point for me in pointing out the 5 page spec for bullet charts. Data geeks appreciate that, business users generally don’t have the patience for it.
John,
as always, I appreciate your comments, insight and most of all your website! You’re right the chart itself isn’t really that hard to create and maintain but it does require attention to detail.
I’m not wild about the gauge either but had to find a decent comprimise that would work for my users who really wanted a circular dial/gauge.
It may be semantic, but it’s not that the marker line is mis-aligned (in fact it’s perfectly aligned to a traverse of 50%) it’s that the red background, without any other context creates the illusion of misalignment thus the x-axis where the pointer rests.
In this case, if the percentage to goal exceeds 100%, the pointer stays fixed to 100% while the label shows the actual percentage completed – again a comprimise – what’s the proper way to ’tilt!’ the chart?
There are several ways in which to address providing historical context. It would be a simple matter to change the minimum area to pace with the percentage of the month completed. In fact, I have used that with bullet charts in other solutions. However, in this case, the pacing gauge is an element within a dashboard and we are using sparklines plus some simple tables to show trending and W/W & M/M change. This particular visualization is not one-size fits all (I’m no Minard) and is designed to reflect one specific aspect of the data – pacing to a goal.
(I think I hear the rumblings of a new post in your comments, can’t wait to read it if that’s true)
Hubert, all true. I would only point out that as I commented above, after 6 months of effort with users still not understanding the bullet chart and being frustrated with it a change had to made. Otherwise they would ignore the dashboard as meaningless noise.
I’ll agree on the thermometer comment however. I think that visualizations of that type including gauges and (eek) pie charts tend to get a bad reputation more for over- and mis-use than for their lack of compliance to current visualization theory. Seth G was partially correct the other day when he said charts should be designed for impact. A properly used thermometer-style chart in the right place at the right time can have more impact than a line or a bar chart. IMHO, our job is to communicate a story that is buried in dry data. Impactful visualizations are our best tools for telling that story and communicating it correctly and quickly. If users are distracted by having to figure out how to read a visualization correctly we’ve lost the user and failed.
Graphing theory and rules are to visualization as grammar is to poetry, you need to know all the rules before you break them.
To all, I fully expect to end up on Stephen’s Wall of Shame ;~)
Like most here, I’m no fan of dial/circrlar gauges, but this bar gauge is a reasonable compromise, particularly if it communicates to your audience. Isn’t that the point: communicating actionable information. Otherwise, as Clint notes, it’s ignored b/c it can’t be clearly interpreted.
I find the bar gauge more similar to a bullet graph than not, and it contains the 3 primary criteria: labeling, a quantitative scale, and a featured measure. It wastes far less real estate than all the half-moon gauges.
Having also attended the East Coast Visual Business Intelligence Workshop, with Dave and Hubert, I’m not so certain Few would shoot down this gauge down outright. This one could arguably be viewed an alternate bar graph. But I’m no Stephen Few. ;^]
I agree with Dave, it would be nice to see a side-by-side comparison with the bullet graph solution. We could then view each on it’s merits and discern where your audience struggled reading bullet graph. There may be room for improvement — no thoughts at the moment in addition to what’s been offered here — but, all in all, your bar gauge is a decent compromise.
I mashed up a quick bullet graph based on the original gauge but following Few’s aesthetics. Thoughts/comments?
peter,
thanks for the comment and the quick mashup. I’ve updated the post to include your mockup. I’ve also included a bullet chart based on a design I had done previously. Now that you have an easy visual comparison, how does that change your thoughts?
Clint,
I appreciate the humor with which you presented your graph, knowing that the dataviz geeks would come out in full force. Allow me to pile on.
I don’t think your example is really easier to understand than a bullet graph, you’re just not presenting as much information as most simple bullet graphs contain. There is no indication of where the value is at the start of the month, and you don’t provide a comparative/target measure. In your example you just make the goal the same as the largest value in the range, but this seems like a cop-out, what happens if you achieve your goal mid-way through the month? What happens if you exceed it? How is this joyful information displayed in a meaningful way?
Is this really easier to learn? And if so is it just because you’re communicating less information? Both of the bullet graph counter-examples make assumptions at what you are trying to do with the gauge, starting the bar on the left side, and also putting in sperate comparative/target measures which you ommitted in your gauge.
Also consider if the salient point in ‘page views’ wasn’t to aim for higher numbers, but to count down the paid-for views one had contracted to an advertiser, or from a web services firm (i.e. the commercial version of the google maps api). Getting closer to (but maybe not too) zero might be the goal, and the arrow would travel from right to left. This is actually a real-world metric in my case so looking at your example I’m really not sure what I’m seeing. In the bullet graph versions, I’m given a clue (other than the much-maligned traffic light color scheme).
I agree with Hubert: a thermometer style chart does not seem that challenging. One of the things that I think attracts people to bullet graphs is that they are usually immediately graspable without education, at least this was my experience and seemingly the experience of most people I know who have seen them. I can’t help but ask the questions: they don’t understand bullet graphs after 6 months? for serious?
Kalin
[...] A Gauge Chart That Works?, Clint describes his efforts to come up with a dial-type chart that isn’t as terrible as [...]
I’ve posted the first part in my discussion of this chart type.
A Gauge that Works?
Like all gauge types (including bullet charts), this gauge suffers from the single-point-in-time limitation. I’ve shown how to make a line chart that shows all the information of Clint’s gauge (and Stephen Few’s bullet) but includes the time component as well, so you can track historical performance and extrapolate future results.
I didn’t include my usual how-to in this post, but I will add that soon, and comment here with a link. I welcome any feedback from Clint’s readers.
@kalin harvey: Hi Kalin, thanks for stopping by. Just to be sure, I went back to read Stephen’s design spec for the bullet chart. And frankly think I am encoding just as much information on the gauge as he encodes on the bullet:
1. Text Label: Got it, “Month-to-Date” Page Views
2. Quantitative Scale along a single axis: Check
3. Featured Measure: Page Views
4. One or two comparative measures (optional): yep, target value and goal value (9M and 10M page views respectively)
5. 2-5 ranges along the quantitative scale to declare the measure’s qualitative state (optional). I’ve got 3: 0 to Minimum acceptable (70%/7M), Minimum to target (7M-9M) and target to goal (9M-10M)
The traffic light scheme is trite but also useful because people understand it immediately – the connotations with those colors are so strong that it’s very difficult for users to misunderstand whereas shades of gray are less strongly correalated to a specific idea of feeling.
The problem I was trying to solve was pacing to a positive goal, not counting down, so this particular chart might not work in the scenario you describe, then again does that really matter? If I was building a “countdown” visualization I would probably approach it from a completely different angle.
I agree with John Peltier, the bullet chart is a gauge chart at heart and in fact, it’s a more violent version of a thermometer chart. All I ever see when I look at a bullet chart is a cross section of a rifle barrel.
And yes, it’s really true, my users had an extremely difficult time with the bullet chart. So lucky you for having such erudite users!
Thanks again for stopping by and for your feedback – I do appreciate it.
[...] Read up more on this at PTS blog and on a Gauge chart that actually works. [...]
[...] on Instant Cognition, Clint clearly agrees with me. In his article, A Gauge Chart that Works, he attempts to improve upon the gauge and succeeds. However, I still can’t think of an data [...]
[...] – bookmarked by 6 members originally found by rtwoods on 2008-08-13 A Gauge Chart That Works? http://blog.instantcognition.com/excel/2008/07/14/a-gauge-chart-that-works/ – bookmarked by 2 [...]