In this article by David Baldwin, the author of the book Mastering Tableau, we will cover how you need to create some effective dashboards.
(For more resources related to this topic, see here.)
Since that fateful week in Manhattan, I’ve read Edward Tufte, Stephen Few, and other thought leaders in the data visualization space. This knowledge has been very fruitful. For instance, quite recently a colleague told me that one of his clients thought a particular dashboard had too many bar charts and he wanted some variation. I shared the following two quotes:
Show data variation, not design variation.
–Edward Tufte in The Visual Display of Quantitative Information
Variety might be the spice of life, but, if it is introduced on a dashboard for its own sake, the display suffers.
–Stephen Few in Information Dashboard Design
Those quotes proved helpful for my colleague. Hopefully the following information will prove helpful to you.
Additionally I would also like to draw attention to Alberto Cairo—a relatively new voice providing new insight. Each of these authors should be considered a must-read for anyone working in data visualization.
- Visualization design theory
- Dashboard design
- Sheet selection
Visualization design theory
Any discussion on designing dashboards should begin with information about constructing well-designed content. The quality of the dashboard layout and the utilization of technical tips and tricks do not matter if the content is subpar. In other words we should consider the worksheets displayed on dashboards and ensure that those worksheets are well-designed. Therefore, our discussion will begin with a consideration of visualization design principles. Regarding these principles, it’s tempting to declare a set of rules such as:
- To plot change over time, use a line graph
- To show breakdowns of the whole, use a treemap
- To compare discrete elements, use a bar chart
- To visualize correlation, use a scatter plot
But of course even a cursory review of the preceding list brings to mind many variations and alternatives! Thus, we will consider various rules while always keeping in mind that rules (at least rules such as these) are meant to be broken.
Formatting rules
The following formatting rules encompass fonts, lines, and bands. Fonts are, of course, an obvious formatting consideration. Lines and bands, however, may not be something you typically think of when formatting, especially when considering formatting from the perspective of Microsoft Word. But if we broaden formatting considerations to think of Adobe Illustrator, InDesign, and other graphic design tools, then lines and bands are certainly considered. This illustrates that data visualization is closely related to graphic design and that formatting considers much more than just textual layout.
Rule – keep the font choice simple
Typically using one or two fonts on a dashboard is advisable. More fonts can create a confusing environment and interfere with readability. Fonts chosen for titles should be thick and solid while the body fonts should be easy to read. As of Tableau 10.0 choosing appropriate fonts is simple because of the new Tableau Font Family. Go to Format | Font to display the Format Font window to see and choose these new fonts:
Assuming your dashboard is primarily intended for the screen, sans serif fonts are best. On the rare occasions a dashboard is primarily intended for print, you may consider serif fonts; particularly if the print resolution is high.
Rule – Trend line > Fever line > Reference line > Drop line > Zero line > Grid line
The preceding pseudo formula is intended to communicate line visibility. For example, trend line visibility should be greater than fever line visibility. Visibility is usually enhanced by increasing line thickness but may be enhanced via color saturation or by choosing a dotted or dashed line over a solid line.
The trend line, if present, is usually the most visible line on the graph. Trend lines are displayed via the Analytics pane and can be adjusted via Format à Lines.
The fever line (for example, the line used on a time-series chart) should not be so heavy as to obscure twists and turns in the data. Although a fever line may be displayed as dotted or dashed by utilizing the Pages shelf, this is usually not advisable because it may obscure visibility. The thickness of a fever line can be adjusted by clicking on the Size shelf in the Marks View card.
Reference lines are usually less prevalent than either fever or trend lines and can be formatted by going to Format | Reference lines.
Drop lines are not frequently used. To deploy drop lines, right-click in a blank portion of the view and go to Drop lines | Show drop lines. Next, click on a point in the view to display a drop line. To format droplines, go to Format | Droplines. Drop lines are relevant only if at least one axis is utilized in the visualization.
Zero lines (sometimes referred to as base lines) display only if zero or negative values are included in the view or positive numerical values are relatively close to zero. Format zero lines by going to Format | Lines.
Grid lines should be the most muted lines on the view and may be dispensed with altogether. Format grid lines by going to Format | Lines.
Rule – band in groups of three to five
Visualizations comprised of a tall table of text or horizontal bars should segment dimension members in groups of three to five.
Exercise – banding
- Navigate to https://public.tableau.com/profile/david.baldwin#!/ to locate and download the workbook.
- Navigate to the worksheet titled Banding.
- Select the Superstore data source and place Product Name on the Rows shelf.
- Double-click on Discount, Profit, Quantity, and Sales.
- Navigate to Format | Shading and set Band Size under Row Banding so that three to five lines of text are encompassed by each band. Be sure to set an appropriate color for both Pane and Header:
Note that after completing the preceding five steps, Tableau defaulted to banding every other row. This default formatting is fine for a short table but is quite busy for a tall table. The band in groups of three to five rule is influenced by Dona W. Wong, who, in her book The Wall Street Journal Guide to Information Graphics, recommends separating long tables or bar charts with thin rules to separate the bars in groups of three to five to help the readers read across.
Color rules
It seems slightly ironic to discuss color rules in a black-and-white publication such as Mastering Tableau. Nonetheless, even in a monochromatic setting, a discussion of color is relevant. For example, exclusive use of black text communicates differently than using variations of gray. The following survey of color rules should be helpful to ensure that you use colors effectively in a variety of settings.
Rule – keep colors simple and limited
Stick to the basic hues and provide only a few (perhaps three to five) hue variations. Alberto Cairo, in his book The Functional Art: An Introduction to Information Graphics and Visualization, provides insights into why this is important. The limited capacity of our visual working memory helps explain why it’s not advisable to use more than four or five colors or pictograms to identify different phenomena on maps and charts.
Rule – respect the psychological implication of colors
In Western society, there is a color vocabulary so pervasive, it’s second nature. Exit signs marking stairwell locations are red. Traffic cones are orange. Baby boys are traditionally dressed in blue while baby girls wear pink. Similarly, in Tableau reds and oranges should usually be associated with negative performance while blues and greens should be associated with positive performance. Using colors counterintuitively can cause confusion.
Rule – be colorblind-friendly
Colorblindness is usually manifested as an inability to distinguish red and green or blue and yellow. Red/green and blue/yellow are on opposite sides of the color wheel. Consequently, the challenges these color combinations present for colorblind individuals can be easily recreated with image editing software such as Photoshop. If you are not colorblind, convert an image with these color combinations to grayscale and observe. The challenge presented to the 8.0% of the males and 0.5% of the females who are color blind becomes immediately obvious!
Rule – use pure colors sparingly
The resulting colors from the following exercise should be a very vibrant red, green, and blue. Depending on the monitor, you may even find it difficult to stare directly at the colors. These are known as pure colors and should be used sparingly; perhaps only to highlight particularly important items.
Exercise – using pure colors
- Open the workbook and navigate to the worksheet entitled Pure Colors.
- Select the Superstore data source and place Category on both the Rows shelf and the Color shelf.
- Set the Fit to Entire View.
- Click on the Color shelf and choose Edit Colors….
- In the Edit Colors dialog box, double-click on the color icons to the left of each dimension member; that is, Furniture, Office Supplies, and Technology:
- Within the resulting dialog box, set furniture to an HTML value of #0000ff, Office Supplies to #ff0000, and Technology to #00ff00.
Rule – color variations over symbol variation
Deciphering different symbols takes more mental energy for the end user than distinguishing color. Therefore color variation should be used over symbol variation. This rule can actually be observed in Tableau defaults. Create a scatter plot and place a dimension with many members on the Color shelf and Shape shelf respectively. Note that by default, the view will display 20 unique colors but only 10 unique shapes. Older versions of Tableau (such as Tableau 9.0) display warnings that include text such as “…the recommended maximum for this shelf is 10”:
Visualization type rules
We won’t spend time here to delve into a lengthy list of visualization type rules. However, it does seem appropriate to review at least a couple of rules. In the following exercise, we will consider keeping shapes simple and effectively using pie charts.
Rule – keep shapes simple
Too many shape details impede comprehension. This is because shape details draw the user’s focus away from the data. Consider the following exercise on using two different shopping cart images.
Exercise – shapes
- Open the workbook associated and navigate to the worksheet entitled Simple Shopping Cart.
- Note that the visualization is a scatterplot showing the top 10 selling Sub-Categories in terms of total sales and profits.
- On your computer, navigate to the Shapes directory located in the My Tableau Repository. On my computer, the path is C:UsersDavid BaldwinDocumentsMy Tableau RepositoryShapes.
- Within the Shapes directory, create a folder named My Shapes.
- Reference the link included in the comment section of the worksheet to download the assets.
- In the downloaded material, find the images titled Shopping_Cart and Shopping_Cart_3D and copy those images into the My Shapes directory created previously.
- Within Tableau, access the Simple Shopping Cart worksheet.
- Click on the Shape shelf and then select More Shapes.
- Within the Edit Shape dialog box, click on the Reload Shapes button.
- Select the My Shapes palette and set the shape to the simple shopping cart.
- After closing the dialog box, click on the Size shelf and adjust as desired. Also adjust other aspects of the visualization as desired.
- Navigate to the 3D Shopping Cart worksheet and then repeat steps 8 to 11. Instead of using the simple shopping cart, use the 3D shopping cart:
Compare the two visualizations. Which version of the shopping cart is more attractive? Likely the cart with the 3D look was your choice. Why not choose the more attractive image? Making visualizations attractive is only of secondary concern. The primary goal is to display the data as clearly and efficiently as possible. A simple shape is grasped more quickly and intuitively than a complex shape. Besides, the cuteness of the 3D image will quickly wear off.
Rule – use pie charts sparingly
Edward Tufte makes an acrid (and somewhat humorous) comment against the use of pie charts in his book The Visual Display of Quantitative Information.
A table is nearly always better than a dumb pie chart; the only worse design than a pie chart is several of them. Given their low density and failure to order numbers along a visual dimension, pie charts should never be used.
The present sentiment in data visualization circles is largely sympathetic to Tufte’s criticism. There may, however, be some exceptions; that is, some circumstances where a pie chart is optimal. Consider the following visualization:
Which of the four visualizations best demonstrates that A accounts for 25% of the whole? Clearly it is the pie chart! Therefore, perhaps it is fairer to refer to pie charts as limited and to use them sparingly as opposed to considering them inherently evil.
Compromises
In this section, we will transition from more or less strict rules to compromises. Often, building visualizations is a balancing act. It’s common to encounter contradictory directions from books, blogs, consultants, and within organizations. One person may insist on utilizing every pixel of space while another urges simplicity and whitespace. One counsels a guided approach while another recommends building wide open dashboards that allow end users to discover their own path. Avant gardes may crave esoteric visualizations while those of a more conservative bent prefer to stay with the conventional. We now explore a few of the more common competing requests and suggests compromises.
Make the dashboard simple versus make the dashboard robust
Recently a colleague showed me a complex dashboard he had just completed. Although he was pleased that he had managed to get it working well, he felt the need to apologize by saying, “I know it’s dense and complex but it’s what the client wanted.” Occam’s Razor encourages the simplest possible solution for any problem. For my colleague’s dashboard, the simplest solution was rather complex. This is OK! Complexity in Tableau dashboarding need not be shunned. But a clear understanding of some basic guidelines can help the author intelligently determine how to compromise between demands for simplicity and demands for robustness.
More frequent data updates necessitate simpler design.
Some Tableau dashboards may be near-real-time. Third-party technology may be utilized to force a browser displaying a dashboard via Tableau Server to refresh every few minutes to ensure the absolute latest data displays. In such cases, the design should be quite simple. The end user must be able to see at a glance all pertinent data and should not use that dashboard for extensive analysis. Conversely, a dashboard that is refreshed monthly can support high complexity and thus may be used for deep exploration.
Greater end user expertise supports greater dashboard complexity.
Know thy users. If they want easy, at-a-glance visualizations, keep the dashboards simple. If they like deep dives, design accordingly.
Smaller audiences require more precise design.
If only a few people monitor a given dashboard, it may require a highly customized approach. In such cases, specifications may be detailed, complex, and difficult to execute and maintain because the small user base has expectations that may not be natively easy to produce in Tableau.
Screen resolution and visualization complexity are proportional.
Users with low-resolution devices will need to interact fairly simply with a dashboard. Thus the design of such a dashboard will likely be correspondingly uncomplicated. Conversely, high-resolution devices support greater complexity.
Greater distance from the screen requires larger dashboard elements.
If the dashboard is designed for conference room viewing, the elements on the dashboard may need to be fairly large to meet the viewing needs of those far from the screen. Thus the dashboard will likely be relatively simple. Conversely, a dashboard to be viewed primarily on end users desktops can be more complex.
Although these points are all about simple versus complex, do not equate simple with easy. A simple and elegantly designed dashboard can be more difficult to create than a complex dashboard. In the words of Steve Jobs:
Simple can be harder than complex: You have to work hard to get your thinking clean to make it simple. But it’s worth it in the end because once you get there, you can move mountains.
Present dense information versus present sparse information
Normally, a line graph should have a maximum of four to five lines. However, there are times when you may wish to display many lines. A compromise can be achieved by presenting many lines and empowering the end user to highlight as desired. The following line graph displays the percentage of Internet usage by country from 2000 to 2012. Those countries with the largest increases have been highlighted. Assuming that Highlight Selected Items has been activated within the Color legend, the end user can select items (countries in this case) from the legend to highlight as desired. Or, even better, a worksheet can be created listing all countries and used in conjunction with a highlight action on a dashboard to focus attention on selected items on the line graph:
Tell a story versus allow a story to be discovered
Albert Cairo, in his excellent book The Functional Art: An Introduction to Information Graphics and Visualization, includes a section where he interviews prominent data visualization and information graphics professionals. Two of these interviews are remarkable for their opposing views.
I… feel that many visualization designers try to transform the user into an editor. They create these amazing interactive tools with tons of bubbles, lines, bars, filters, and scrubber bars, and expect readers to figure the story out by themselves, and draw conclusions from the data. That’s not an approach to information graphics I like. – Jim Grimwade
The most fascinating thing about the rise of data visualization is exactly that anyone can explore all those large data sets without anyone telling us what the key insight is. – Moritz Stefaner
Fortunately, the compromise position can be found in the Jim Grimwade interview:
[The New York Times presents] complex sets of data, and they let you go really deep into the figures and their connections. But beforehand, they give you some context, some pointers as to what you can do with those data. If you don’t do this… you will end up with a visualization that may look really beautiful and intricate, but that will leave readers wondering, What has this thing really told me? What is this useful for? – Jim Grimwade
Although the case scenarios considered in the preceding quotes are likely quite different from the Tableau work you are involved in, the underlying principles remain the same. You can choose to tell a story or build a platform that allows the discovery of numerous stories. Your choice will differ depending on the given dataset and audience. If you choose to create a platform for story discovery, be sure to take the New York Times approach suggested by Grimwade. Provide hints, pointers, and good documentation to lead your end user to successfully interact with the story you wish to tell or successfully discover their own story.
Document, Document, Document! But don’t use any space!
Immediately above we considered the suggestion Provide hints, pointers, and good documentation… but there’s an issue. These things take space. Dashboard space is precious. Often Tableau authors are asked to squeeze more and more stuff on a dashboard and are hence looking for ways to conserve space. Here are some suggestions for maximizing documentation on a dashboard while minimally impacting screen real estate.
Craft titles for clear communication
Titles are expected. Not just a title for a dashboard and worksheets on the dashboard, but also titles for legends, filters and other objects. These titles can be used for effective and efficient documentation. For instance a filter should not just read Market. Instead it should say something like Select a Market. Notice the imperative statement. The user is being told to do something and this is a helpful hint. Adding a couple of words to a title will usually not impact dashboard space.
Use subtitles to relay instructions
A subtitle will take some extra space but it does not have to be much. A small, italicized font immediately underneath a title is an obvious place a user will look at for guidance. Consider an example: red represents loss. This short sentence could be used as a subtitle that may eliminate the need for a legend and thus actually save space.
Use intuitive icons
Consider a use case of navigating from one dashboard to another. Of course you could associate an action with some hyperlinked text stating Click here to navigate to another dashboard. But this seems quite unnecessary when an action can be associated with a small, innocuous arrow, such as is natively used in PowerPoint, to communicate the same thing.
Store more extensive documentation in a tooltip associated with a help icon.
A small question mark in the top-right corner of an application is common. This clearly communicates where to go if additional help is required. As shown in the following exercise, it’s easy to create a similar feature on a Tableau dashboard.
Summary
Hence from this article we studied to create some effective dashboards that are very beneficial in corporate world as a statistical tool to calculate average growth in terms of revenue.
Resources for Article:
Further resources on this subject:
- Say Hi to Tableau [article]
- Tableau Data Extract Best Practices [article]
- Getting Started with Tableau Public [article]