We stop loving someone or something when we feel there is nothing more to discover, when we have no more questions, when we don’t care about the answers.
I started writing my Data Visualization for Excel Users series because I have an unhealthy number of open questions. I’d like to go beyond it depends but there are no right answers, only my answers (and, with your help, our answers). These are practical questions (from a business user perspective) that, once answered, will surely trigger new ones. Here is a sample, with a few starting points:
1. Reason or Emotion?
This is the big one. We’ve been discussing reason vs. emotion over the last 2,500 years, and we still disagree. The right answer would be reason and emotion (easier said than done). When applied to data visualization, this translates into charts that respect the data, attracts readers attention because they are beautiful and keeps it because they are interesting and insightful. And contain no cheap emotional tricks. This filters out 99,9% of all infographics published nowadays.
2. Pies or No Pies?
No self-respecting data visualization expert likes pie charts. And there seems to be a daily contest to find the “worst pie chart ever”. And people keep making them, relentlessly… Why?
3. Are there interchangeable charts?
Everyone tells you that you should use bar charts instead of pies. Well, I think you should. Not because pie charts are bad, but because you’ll have to ask better questions. But that’s not the point. The point is that there are no interchangeable charts. Different charts tell us different stories. You cannot tell a part-of-a-whole story with a bar chart. You have to have a pie.
4. Should I use logarithmic scales?
I’m still trying to find a simple way to implement a logarithmic perspective in my eyes… Logarithmic scales are important to look at growth from a different perspective, but if you can’t find a way to tell the readers how to read the chart you should consider not to use them.
5. Should origin start at zero?
Yes, by default. If you need to improve resolution set the scale to 20% below minimum and 20% above maximum values . Don’t do it with bar charts (origin should always be set to zero).
I would consider this rule: to improve resolution you must have more than one series (see this example). In a line chart, comparing slopes is often more interesting than the slope itself.
6. What kind of data visualization skills the information worker needs?
Above all, know your business. Learn how to use a database (table structures, basic SQL). Refresh what you know about descriptive statistics. Choose a tool and be an advanced user or hire someone who is. Learn the basics on how human perception works. Understand how perception impacts design choices and the other ways around. Learn what each chart type is used for. Find your charting style. Spread this knowledge across your organization. Know your business (again).
7. What is the role of design in data visualization?
We saw that a chart is a visual representation of distances between data points. Everything else is design. The first role of design in data visualization is to improve cognition. The second role is to provide aesthetic consistency. The third role is to grab users attention.
8. How to design a dashboard?
I think we owe Tufte and Bertin a consistent approach to data visualization, at the chart level. We still need something similar for dashboard design.
9. How to sell data visualization?
Find a sponsor and make a lot of bad chart/(good chart comparisons. Compress a 100-slide presentation into 50 slides. They will get it, sooner or later.
10. Should I use animation?
You probably shouldn’t. Try small multiples first. Try animation if you have too many series, if the animation defines clear patterns and if it is consistent with the law of common fate. But remember: are no interchangeable charts.
11. Do I need to code?
No, but you should at least be able to understand what a recorded macro is and how to improve it with simple changes.
12. What tools should I use?
Learn everything in Excel then switch to Tableau/Qlikview/Spotfire.
How would you respond to these questions? Please share your thoughts in the comments below…
14 thoughts on “12 Data Visualization Questions I Have No Answers For”
Yes! I completely agree that my interest in something, my curiosity about it, requires questions to answer, and wonder about why something is the way it is. Why does this look better than that? If you can figure out how to articulate it, you will deepen your mastery if it.
One small point, though. You write, “You cannot tell a part-of-a-whole story with a bar chart. You have to have a pie.” Stacked bar charts can do this pretty well and often allow for more elegant labeling too.
Stephanie: there is a lot to say about stacked bar charts, but I was considering a single series, so a pie chart vs. a single series bar chart.
I’m interested in the “respects the data” part. How do you judge how far is too far? What if I assert that every chart disrespects the data to some degree and what you call respecting the data is really minimizing the disrespect? If that’s true, then the degree of disrespect seems pretty subjective.
Dick: Good question. I would say that a chart respects the data when it is a good transcription of the underlying table. If the chart tells you that A = 50% B theoretically you could read the same result in the table, given enough time and processing power. I agree that “every chart disrespects the data to some degree”. As I said in this page, something is always lost when you move from one sign system to the other and yes, there is a dose of subjectivity. Who said charts are objective? 🙂
Hey! That’s a good question! Can a chart be objective?
I’m interested too.
I think a good alternative are the treemap.
Can a chart be objective? yes it is very good question 🙂
So 13 questions. One answer:
… Jon great answer! 😀
Jon, I wrote that above. But let’s assume that “it depends” is the lazy answer…
I was too lazy to read your intro, and went straight to the questions. I don’t think it hurts, though, to remind people that each situation, each data set, each audience requires a carefully thought out data presentation plan.
Jon: That’s the consultant’s dreamland! You are right of course, but let’s put it this way: we need good starting points, good defaults. Can we use this default data presentation plan in project x? What do we have to change? If a user needs to see market share and growth now, I’ll use a scatter plot by default, not two bar charts.
I decided to take a crack at Question 4: Should I use logarithmic scales? Since my answer is too long for a comment, you can read it at
Interesting post you wrote.
With regards to Question 4 I have an example that you might find useful: http://waukema.blogspot.com/2012/01/visualizing-47693-notes-users-on-74033.html
Please do let me know what you think.
With respect to question 3: Are there interchangeable charts?
I disagree with your last comment “You cannot tell a part-of-a-whole story with a bar chart. You have to have a pie.”
I recently wrote about the good and bad alternatives for a parts-to-whole relationship.
In the end, I would use a stacked bar when possible, as long as there aren’t too many categories or data points.
Andy: an interesting reading regarding pie charts and stacked bars is “An Information-Processing Analysis of Graphic Perception”, by David Simkin and Reid Hastie. They show that in some cases pie charts perform better than stacked bars.
We can compare multiple pie charts to a stacked bar chart, and in this case I do prefer a stacked bar chart. On the other hand, comparing single series (a pie chart vs. a stacked par chart with a single bar) I would choose the pie.
Comments are closed.