Data Visualization Best Practices
Participant Guide
Section One: Introducing Data Visualization Overview The first introduces data visualization and describes its roots in the 17th century, with the invention of many of the techniques still used today. The role of vision in cognition, and the importance of data visualizations to leverage sensory memory for instant communication is discussed. This section also describes several very helpful basic elements such as preattentive attributes, gestalt principles, and color selection. Section one provides the foundational knowledge of how vision affects cognition and provides the base level toolkit for designing effective data visualizations.
Learning Objectives On completion of this section one, you will be able to: Describe the power inherent in data visualization techniques. Explain the concepts of meaning, clarity, and focus, in relation to data visualization. Explain the process of cognition relative to understanding data. Identify characteristics of data communication (preattentive attributes and gestalt principles) and explain their use. Select colors effectively, and understand the implications of color blindness.
Chapter 1. The Power of Data Visualization The History of Data Visualization Early History Since the earliest times, mankind has sought to explain things through visual medium. The earliest known attempt to portray mathematical values in a graphical format dates to circa 950 A.D. where an unknown author sought to portray the positions of the sun, moon, and planets throughout the year.
Funkhouser, H. Gray (1936). A note on a tenth century graph. Osiris, 1:260–262. URL http://tinyurl.com/2czmqc. 3, 5
Several examples of similar diagrammatic explanations exist from the 14th century, but it‘s in the 1600‘s that the foundations of modern day data visualization are found. Among the most important problems of the 17th century were those concerned with physical measurement— of time, distance, and space— for astronomy, surveying, map making, navigation and territorial expansion – and the data visualizations developed during that time reflect the need to express new thinking. During the 17th century, the key components of modern data visualization were invented; these include the relationship between a graphed line and an equation, the continuous distribution, and the use of iconic notations on maps.
The Golden Age of Data Visualization The 18th century continued the initial germination of the seeds of visualization which had been planted earlier. Map-makers began to try to show more than just geographical position on a map. As a result, new graphic forms (isolines and contours) were invented, and thematic mapping of physical quantities took root.
The Power of Data Visualization
3
Abstract graphs, and graphs of functions were introduced, along with the early beginnings of statistical theory (measurement error) and systematic collection of empirical data. As other (economic and political) data began to be collected, some novel visual forms were invented to portray them, so the data could ―speak to the eyes‖. Innovation and understanding grew out of the foundation laid in the previous century in innovations such as the test for statistical significance based on a deviation between observed data and a null hypothesis, (used to show that the guiding hand of a divine could be discerned in the nearly constant ratio of male to female births in London over 1629–1710)—John Arbuthnot. Other notable developments included literal and abstract line graphs (Hauksbee & Cruquius), the normal distribution (de Moivre), the first use of the term ‗statistik‘ (Achenwall), introduction of a notation which gives a name and address to every possible point in 3d space (x, y, z) (Euler). Key visualizations emerged during this time, such as additional curve fitting and interpolation from empirical data points, and the theory of measurement error by Johann Heinrich Lambert, who noted that ―a diagram does incomparably better service here than a table.‖ And the visualization of vibration patterns (by spreading a uniform layer of sand on a disk, and observing displacement when vibration is applied) by Ernest Florens Friedrich Chladni (now considered the father of acoustics).
William Playfair Perhaps the most notable single contributor to our modern day notion of data visualization is William Playfair (1759-1823). Playfair first published The Commercial and Political Atlas in London in 1786. It contained 43 time-series plots and one bar chart, a form introduced in this work. It has been described as the first major work to contain statistical graphs. Playfair's Statistical Breviary, published in London in 1801, contains what is generally credited as the first pie chart.
4
The Power of Data Visualization
Friendly, M. (2009) Milestones in the history of thematic cartography, statistical graphics, and data visualization. URL: http://www.math.yorku.ca/SCS/Gallery/milestone/milestone.pdf
Universal Acceptance From Playfair, the use of data visualization moved quickly and gained universal acceptance. Some milestones include John Snow‘s Use of a dot map to display epidemiological data, leads to discovery of the source of a cholera epidemic (1854), the first use of a statistical diagram in a school textbook by Emile Levasseur (1868), the periodic table of elements (Dmitri Mendeleev, 1869), the first-known use of a semi-graphic table to display a data table by shading levels (Toussaint Loua, 1873), the pictogram used to represent data by icons proportional to a number (1836), and Charles Booth‘s street maps of London, showing poverty and wealth by color coding (1889).
The Power of Data Visualization
5
Meaning, Clarity, & Focus in Data Visualization Meaning We are all sensitive to the subtleties of meaning. Consider the following sentence with the varied placement of just one word:
Only Linda ate the pizza today. Linda only ate the pizza today. Linda ate only the pizza today. Linda ate the only pizza today. Linda ate the pizza only today. Linda ate the pizza today only.
Six distinct meanings conveyed by the placement of a single word. The implication for data visualization is the challenge to provide precision in meaning. How do we explain precise concepts, relationships, and conclusions that will mean the same thing to all viewers?
Clarity The quest to visualize data has had its share of problems. Consider the following examples and add your comments on how they fail the clarity test.
6
The Power of Data Visualization
____________________________ ____________________________ ____________________________ ____________________________ ____________________________ ____________________________
____________________________ ____________________________ ____________________________ ____________________________ ____________________________ ____________________________
____________________________ ____________________________ ____________________________ ____________________________ ____________________________ ____________________________
The Power of Data Visualization
7
Focus The design of data visualizations can have significant impact on the focus and subsequent thought process of the viewer. Consider the following report cards:
Same grades, but the different focus probably leads to very different comments from the parents of this child!
The Purpose of Data Visualization Exercise 1: What do you consider are the purposes of data visualization? Be prepared to share a few ideas with the group. ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________
8
The Power of Data Visualization
Simplification At its core, the purpose of data visualization is to simplify volumes of data, thereby providing meaning, clarity, and focus, in an appropriate way. Perhaps the best explanation comes from Edward Tufte, a modern day data visualization expert, and the inventor of the ‗sparkline‘:
―What is sought in designs for the display of information is the clear portrayal of complexity.‖ Tufte, E. (2001) The Visual Display of Quantitative Information (2. Ed.) Graphics Press, Cheshire, CT.
Purpose In addition, most data visualizations have a purpose – to inform, alert, or perhaps to persuade. Within the dashboard context, each visualization is probably linked to a specific need, and an effective visualization design will use the principles of clarity and focus to provide the meaning that best meets the need.
Summary The way a data visualization is designed can have a significant impact on the way it is perceived. Good designers understand the purpose of a chart, graph, or diagram, and place emphasis in the areas that will drive effective decision making. Controlling focus is a power wielded by the data visualization designer and is a power that can be used or abused. Exercise 2: consider the following diagram and note any issues relative to the manipulation of meaning:
The Power of Data Visualization
9
Answer: You probably didn‘t find too much wrong with this cunningly designed graph. It appears to show rising costs of education, with minimal results. However, with some digging into the original data, you might find that some of the key problems include:
10
The graph on the left does not account for inflation or for the general increase in the number of students being educated in America from population increases. At first glance, the chart would appear to show that more money is being spent, when in reality the percentages of costs per pupil vary much less. The chart on the right shows the average student scores for a standardized test for 4th graders only. The scale goes from 0-500, with 500 being the absolute maximum score that relatively few students achieve. The average score on the same test when taken by high school seniors is just over 300. For fourth graders to be scoring in the 200′s seems pretty good!
The Power of Data Visualization
Chapter 2. Visualization and Cognition Vision & Cognition Vision – The T1 to the Brain Of our five senses, vision is perhaps the most powerful. The phrase ‗a picture is worth a thousand words‘ hints at our ability to perceive vast amounts of information when processed visibly. Imagine the process of describing a great work of art to someone that had not seen it, and you get a sense of the amount of information processed through the eyes. Modern day data visualization expert, Colin Ware, described it this way, ―The eye and visual cortex of the brain form a massively parallel processor that provides the highest bandwidth channel into the human cognitive center.‖ Colin Ware, (2004) Information Visualization: Perception for Design (2nd ed.)
Rules at Work Within visual processing, there are clearly some rules at work. Consider the following images: The brain clearly perceives the triangle at the center of the image. In fact, it is perhaps the predominant element, even though it is not actually represented.
The word ‗data‘ is more easily perceptible when the overlapping lines are shown in front of the image.
For some reason the brain does a more efficient job of processing the information in the second image.
These rules suggest that it is possible to manipulate vision, and therefore cognition. In the next chapter, we will explain some ‗tricks of the trade‘ as they relate to rapid cognition through data visualization.
Visualization and Cognition
11
Vision & Memory Sensory Memory Sensory memory refers to items detected by the sensory receptors which are retained temporarily in the sensory registers and which have a large capacity for unprocessed information, but are only able to hold accurate images of sensory information momentarily. Since visual information is processed so quickly by the brain, we are capable of making judgments about sensory visual information very quickly, however, since the sensory receptors store this information for a very short time, visual images that are complex, or that have no adequate context, will not be processed by sensory memory.
Without the correct context (prior knowledge), this diagram is essentially meaningless. Those with the appropriate context will instantly recognize a simple light switch.
Even with the appropriate context, the following diagram is too complex to be processed by sensory memory:
12
Visualization and Cognition
Short-Term Memory Also known as ‗active‘ memory, short-term memory is the capacity for holding small amounts of information in an active, readily available state for a short period of time. Most scientists agree that short term memory can hold seven, plus or minus two, objects. Consider the process of copying a sentence from a book. You probably have to return to the text about every seven words. Extreme familiarity with elements can increase the ability of short term memory to process it. One common method for this is called ‗chunking‘. For example, in recalling a phone number, the person could chunk the digits into three groups: first, the area code (such as 801), then a three-digit chunk (805) and lastly a four-digit chunk (9400). This method of remembering phone numbers is far more effective than attempting to remember a string of 10 digits - 8018049400. Short term memory has limitations for data visualization. Consider the following diagram, where short term memory is used to hold the colorto-data relationship. You‘ll notice that with more than the ‘magic number seven‘ colors, this chart is relatively difficult to process.
Visualization and Cognition
13
Long-Term Memory With a few exceptions, long-term memory is not terribly helpful for data visualization. The experiential process required to move an item from short-term into long-term memory is beyond the scope of most data visualizations. There are some notable exceptions, for example, the exclamation point within the following table probably triggers some sort of alert, even to those seeing the chart for the first time. One reason is because the exclamation point as an icon has meaning in our longterm memory.
Another possible use of long-term memory within data visualization would be the use of a certain color to indicate a specific element. For example if product a is always represented as blue, and product b is always represented as orange, the repetition of these elements can secure that relationship into long term memory.
2.2 Data Ink & Chartjunk Data-ink Since it is clear that the best data visualizations will make use of sensory memory (that is they will be instantly understandable), it is important to look at the elements that contribute to instant cognition. The next chapter will deal with several techniques used to ensure instant cognition, but here are two concepts as an introduction. Academic Edward Tufte coined the phrase ‗data-ink‘ to indicate the amount of information within a data visualization that represents data, versus the amount of peripheral information (such as scales, gridlines, etc). He suggested a literal measurement of the amount of ink used on each element. For most purposes, however, an estimate is probably reasonable.
Chartjunk With a flair for the semantic, Edward Tufte also coined the phrase ‗chartjunk‘ to indicate elements that were completely unnecessary. ―Graphical decoration… comes cheaper than the hard work required to produce intriguing numbers and secure evidence. Sometimes the decoration is thought to reflect the artist‘s fundamental design contribution, capturing the essential spirit of the data and so on. Thus principles of artistic integrity and creativity are invoked to defend - even to advance – the cause
14
Visualization and Cognition
of chartjunk. Fortunately most chartjunk does not involve artistic considerations. It is simply conventional graphical paraphernalia routinely added... over-busy gridlines, and excess ticks, redundant representations of the simplest data, the debris of computer plotting, and many of the devices generating design variation.‖ Tufte, E. (2001) The Visual Display of Quantitative Information (2. Ed.), p.107.
Summary Vision is the sense with the largest capacity for cognition. Some estimate 70% of information processing is visual. As a result, the most effective data visualizations will communicate instantly (through sensory memory). Concepts such as data-ink and chartjunk can help us to identify superfluous content and ensure that our data visualizations are as effective as possible.
Exercise 3: Review the dashboard shown by your instructor. List the ways it breaks the rules for non-data ink and chartjunk. ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________
Visualization and Cognition
15
Chapter 3. Tools of the Trade Preattentive Attributes For many years vision researchers have been investigating how the human visual system analyses images. An important initial result was the discovery of a limited set of visual properties that are detected very rapidly and accurately by the low-level visual system. These properties were initially called preattentive, since their detection seemed to precede focused attention. We now know that attention plays a critical role in what we see, even at this early stage of vision. The term preattentive continues to be used, however, since it conveys an intuitive notion of the speed and ease with which these properties are identified. A simple example of a preattentive task is the detection of a red circle in a group of blue circles). The target object has a visual property "red" that the blue distractor objects do not (all non-target objects are considered distractors). A viewer can tell at a glance whether the target is present or absent.
From: Healey, C. Perception in Visualization. (2009). North Carolina State University. URL: http://www.csc.ncsu.edu/faculty/healey/PP/
16
Tools of the Trade
The Attributes Several lists of preattentive attributes exist, with a great deal of overlap, and some differences. The main preattentive attributes with applicability in data visualization are as follows:
Orientation ____________________________ ____________________________ ____________________________ ____________________________ ____________________________ ____________________________
Line Length
____________________________ ____________________________ ____________________________ ____________________________ ____________________________ ____________________________
Line Width ____________________________ ____________________________ ____________________________ ____________________________ ____________________________ ____________________________
Tools of the Trade
17
Size ____________________________ ____________________________ ____________________________ ____________________________ ____________________________ ____________________________
Shape
____________________________ ____________________________ ____________________________ ____________________________ ____________________________ ____________________________
Curvature
____________________________ ____________________________ ____________________________ ____________________________ ____________________________ ____________________________
18
Tools of the Trade
Added Marks ____________________________ ____________________________ ____________________________ ____________________________ ____________________________ ____________________________
Enclosure
____________________________ ____________________________ ____________________________ ____________________________ ____________________________ ____________________________
Color Intensity (Saturation)
____________________________ ____________________________ ____________________________ ____________________________ ____________________________ ____________________________
Tools of the Trade
19
Color Hue ____________________________ ____________________________ ____________________________ ____________________________ ____________________________ ____________________________
2D Position ____________________________ ____________________________ ____________________________ ____________________________ ____________________________ ____________________________ From: Stephen Few (2004). Tapping the Power of Visual Perception. URL: http://www.perceptualedge.com/articles/ie/visual_perception.pdf
Exercise 4: Describe the meaning of preattentive attributes, and how you might use them in dashboard design. ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________
20
Tools of the Trade
Gestalt Principles Similar to the concept of preattentive attributes are the principles identified in the early 20th century by the Gestalt School of Psychology. These principles reveal visual cues that cause us to associate groups.
The six gestalt (meaning ‗pattern‘) principles vary in strength, and also in usefulness for data visualization. Each principle is shown below.
Proximity ________________________________ ________________________________ ________________________________ ________________________________ ________________________________
Similarity ________________________________ ________________________________ ________________________________ ________________________________ ________________________________
Tools of the Trade
21
Enclosure ________________________________ ________________________________ ________________________________ ________________________________ ________________________________
Closure
________________________________ ________________________________ ________________________________ ________________________________ ________________________________
Continuity
________________________________ ________________________________ ________________________________ ________________________________ ________________________________
Connection ________________________________ ________________________________ ________________________________ ________________________________ ________________________________
22
Tools of the Trade
The Visualization Toolkit The Basic Ingredients Most of your data visualizations will be comprised of the following basic components.
Points Lines Bars Area Text Icons
Points Points are probably the simplest technique for encoding data. They are typically used to draw attention to individual values, and less helpful for showing trends. Examples include dots, squares, triangles, etc.
Lines Lines include baselines, borders, as well as data representation. Within charts, lines tend to be effective for showing trends (either as data, or as ‗trendlines‘). Lines are often combined with points, and may be dotted or dashed.
Bars Technically speaking, bars are just thicker lines, and even though the width has no meaning, they are particularly effective at comparing discrete values. Bars can be displayed horizontally or vertically.
Area Data represented through area (such as a Pie Chart) is perhaps the least effective type of data encoding. This is because area is difficult to judge with the naked eye. In addition, superior alternatives exist for most common uses of area.
Text Text in the form of labels, date ranges, and legends is often used to answer questions such as what, when, and who. Consider readability issues such as font, size, and alignment for text. The need for excessive text may be a red flag that a particular visualization has been poorly designed.
Icons Many data visualizations make use of icons (specific shapes attributed with meaning). Icons are often used for alerts, and although many are culturally accepted, should be explained to allow for cultural diversity.
Tools of the Trade
23
Color Perhaps an element most troublesome to data visualization designers is color. Powerful, but mysterious in its use, we all know a little bit about color theory (green is soothing, right?) but perhaps not enough to feel comfortable in selecting colors confidently.
The Ingredients of Color Without getting too scientific, the term ‗color‘ is actually a combination of distinct elements, two of which it is helpful to be familiar with – hue and saturation. Hue is a measure of amount of red, yellow, and blue and is perhaps the main use of the generic term ‗color‘. Saturation is the degree of intensity (the amount of the color that is present).
The Color Wheel The easiest way to understand colors and how they work together is through The Color Wheel. The three primary colors are red, blue, and yellow; they cannot be made by mixing any other colors. Theoretically any color can be achieved by combining various mixtures of the three primaries, but certain bright colors are better obtained as such. The three secondary colors are made by combining two primary colors: red + yellow = orange; blue + yellow = green; and red + blue = violet. The six tertiary colors are obtained by mixing a secondary and a primary next to it.
Color Selection Since the color wheel is a mathematical representation of colors; it can be used to select colors that are ‗naturally‘ go together. There are several ways the color wheel can be used, and several matching types:
Complementary Colors located directly across from each other on the color wheel are mathematically complementary.
Analogous Colors located next to each other on the color wheel are ‗analogous‘.
24
Tools of the Trade
Split Complementary Colors equidistant from a complementary color are considered ‗split complementary‘.
Triad Three colors in an equilateral of isosceles triangle make a ‗triad‘. (Split complementary colors are a triad.)
Color & Culture Some colors have ‗meaning‘, but are culturally specific. This was perhaps more important in a historical context such as purple being reserved for royalty in Western Europe, but yellow being the royal equivalent in China. Besides a general awareness of cultural differences for a diverse audience, it is perhaps worth mentioning the red, yellow, green combination often found in data visualizations and dashboards. In western culture, the ‗traffic-light‘ colors are a simple mechanism to display bad, moderate, good, but this does not hold true in all countries. For example, many Asian countries do not consider red to be an ‗alert‘ color.
Color Blindness If you are considering using traffic-light colors, also consider that about 10% of males suffer from some form of color-blindness, of which Deuteranomoly (where red and green appear as shades of brown) is the most common.
Color Selection Tool Several free color selection tools are available. One we like is found at www.colorschemedesigner.com Exercise 5: Use an online color selector tool to identify two different color schemes you might use. Note the Hexadecimal values here: _________________________ _________________________ _________________________
_________________________
_________________________
_________________________
Use the ‗Colorblind‘ option to review your color selection and determine if it is color-blindness friendly.
Tools of the Trade
25
Section Two: Data Types Overview Section two describes some of the key issues within data visualization – the selection of the right visualization type (in other words, choosing which chart or graph to use). Types of data and the most appropriate visualizations are discussed in detail.
Learning Objectives On completion of this section two, you will be able to: Match your data to one of the six data types. Explain chart and graph types and their uses and limitations for each data type. Within each chart and graph type, utilize best practices to maximize data communication.
The Six Data Types Before identifying the appropriate visualization type, there are two key elements we must understand – first, the data type, and second the purpose of the visualization. Sometimes we put the cart before the horse. ―Let‘s have a bar chart of x‖, or ―let‘s add a gauge showing y.‖ This is backwards because our design might not end up fitting the need appropriately. The more effective method for selecting the right visualization is to fit the design to the need, not the other way around.
Data Types There are six main data types:
Nominal – discrete quantitative values. Time-Series – values over a sequential time period. Ranked – where order is related to value. Part-to-Whole – where the sum of parts equals a whole. Distribution – where one or multiple sets of values display frequency. Correlation – where the data shows the level of interaction between one or more elements.
Since chart selection is such an important topic, we will review each of the data types, and the possible chart types in the following chapters.
26
Tools of the Trade
Chapter 4. Nominal Data Nominal data shows discrete values that are comparative, but not connected. Examples include:
Sales by region. Volume of complaints per product type. Shipping expenses by department.
Because the data elements are not connected, but are instead comparative in nature, the most appropriate visualization type is a bar chart.
Sales by Region 5 4 3 2 1 0 West
Central
South
East
Bar Chart Best Practices A few simple rules apply for effective use of bar charts:
Nominal Data
Bar charts should be in 2-D with minimal distracting elements (such as shadows).
Bars should be the same width and be equally spaced. It is recommended that the bar space between bars be larger or smaller (but not the same) as the bar width.
The exception to spacing between bars is when a bar chart is used to show groups of nominal values.
Use of color within bars should be used to indicate a specific meaning that cannot be accomplished with the axis labels.
Horizontal grid lines should be used to facilitate comparison on values, but should be thin and light. Vertical gridlines are generally not helpful.
27
Horizontal ticks are typically not necessary when horizontal gridlines are used. Vertical ticks are typically not needed.
Since the bar chart shows relative difference, it requires a zerobased scale, otherwise the differences between values will appear artificially inflated.
Bar charts can be displayed horizontally or vertically, however a horizontal display is most useful to show ranking (see the later section). When labels axis labels are long, a horizontal display can be more effective to provide space for the label.
As mentioned above, a bar chart requires a zero-based scale, if this is not desirable or possible a simple data-point plot is an adequate substitute:
Sales by Region $6,000 $5,750 $5,500 $5,250 $5,000 West
Central
South
East
Showing Targets and Goals There are several techniques for showing targets and goals comparative nominal values. Target lines can easily be added to standard bar graphs when the target is the same for each data series:
Sales by Store 5 4
TARGET
3 2 1 0 Austin
28
Houston
San Antonio
Houston
Nominal Data
The target line should always be labeled, and should be stronger than any gridlines that are present. Shading for ranges of good-moderate-bad can also be used, but this only works if the ranges apply to every series of data.
Sales by Product 5 4 3 2 1 0 Austin
Houston
San Antonio
Dallas
Shading for ranges should always appear in the background, and should be subtle enough that they do not compete with the data. When the target is the focus and overall number is less important, there is a variation of the bar chart that can provide very good visibility:
20
% Deviation from Target
10
0
-10
Austin
Houston
San Antonio
Dallas
This chart can easily be adjusted to spotlight either high or low performers (in this case low):
Nominal Data
29
% Deviation from Target 20
10
0
-10
Austin
Houston
San Antonio
Dallas
Single Nominal Values Sometimes nominal data consists of just a single value, such as Sales YTD. Typically a single metric is measured as deviation data – that is expressed relative to a target, goal or other standard. A bar chart doesn‘t really suit the needs of a single nominal value, but several other options exist:
Simple Counter – just showing the numeric value. Bulb Gauge – showing the status (but not the actual number). Combination – such as a counter value interposed on a bulb gauge. Other Gauge – such as radial, linear, or repeating gauges.
Simple Counter When no target, ranges, or tending exists a simple counter is all you need to express a single nominal value.
Bulb Gauge If a single nominal value has a small number of conditions, and the condition is more important than the actual value, a bulb gauge can effectively show status.
Radial Gauge Patterned after a common speedometer, a radial gauge consists of a scale and an arrow. Additional elements such as a target and good-moderate-bad ranges can also be displayed. For a single value, a radial gauge is often a poor use of dashboard real estate.
30
Nominal Data
Linear Gauge Following a thermometer theme, the linear gauge is essentially a bar graph with a single bar. Additional graphic elements can add character but not necessarily increase the value of the display.
Repeating Gauge When the actual value is not as important as which range it falls into, the repeating gauge can be effective in communicating the overall level of a single nominal value.
Gauge Best Practices Most gauges provide the ability to show a good-moderate-bad status in addition to the nominal value. One word of caution, there is a tendency for gauges to be overused since they are moderately slick in appearance, and can give a dashboard a sophisticated feel. The danger is that this sophistication comes at the expense of efficiency. Consider the following when using gauges:
Minimize the busy-ness on gauges to keep the view focused on the message, not the medium.
Show the actual value in a label.
Keep the data density as high as possible by using visible (but not distracting) indicators for good-moderate-bad ranges and hover effects for additional details-on-demand.
If red-yellow-green is used in gauges – use different saturation levels to ensure that the shades are distinguishable to colorblind viewers.
Comparing multiple nominal values with a similar scale is more effectively done in a bar chart than in a series of gauges, which tend to be space inefficient.
The Bullet Graph Berkley professor Stephen Few developed a specific gauge type that allows an efficient encoding of data. Called the ‗bullet graph‘, Few states, ―The bullet graph achieves the communication objectives without the problems that usually plague gauges and meters. It is designed to display a key measure, along with a comparative measure and qualitative ranges to instantly declare if the measure is good, bad, or in some other state.‖
Nominal Data
31
From: Stephen Few (2006). Information Dashboard Design: The Effective Visual Communication of Data. O‘Reilly Media, Sebastopol, CA.
In practice bullet graphs are typically used side by side to compare multiple nominal values that have unique targets and ranges:
Exercise 6: 1. Identify at least three nominal measures for your organization. 2. Sketch a best-practice bar graph for one of them. 3. Explain to a partner how you used preattentive attributes and gestalt principles to make your bar chart as effective as possible. 4. Sketch a gauge for a single nominal value. 5. Compare your gauge to your partners and briefly discuss the differences in your designs.
32
Nominal Data
Chapter 5. Time-Series Data Time-series data shows values over time, and is most often used to identify trends. Examples include: Year-to-date website visits by month. Daily ticket sales. YTD Profit by week.
Due to the connectedness of time-series data, line graphs are the first chart type to consider, and are often the best choice when the focus is showing trending.
Product Sales 5 Product A
4 Product B
3 2 1 0 Jan
Feb
Mar
Apr May
Jun
Jul
Aug
Sep
Line Graph Best Practices Line graphs tend to be fairly well done, but here are a few best practices for their effective use:
Time-Series Data
Time should be displayed along the x-axis, with equal time intervals.
More than three or four lines on a chart can make it unreadable. Using selection boxes to toggle each data set on or off can provide additional data-density.
Use both lines and points to provide visibility for actual values as well as the overall trend. The points should be clearly distinguishable.
Use hovers to display actual values on points.
33
Showing Targets and Goals There are several techniques for showing targets and goals for time series data, and in some cases these lead to use of a chart type other than the line graph. Target lines can be displayed fairly easily for line graphs, with a simple line placed on the chart to identify the target value. As with line graphs in general, showing targets for multiple series of data can quickly become confusing.
Sales by Product 5 4 A Target
3 B Target
2 Product A
1
Product B
0 Jan
Feb
Mar
Apr May
Jun
Jul
Aug
Sep
Shading for ranges of good-moderate-bad can also be used, but only works if the ranges apply to every series of data.
Sales by Product 5 4 3
Product A Product B
2 1 0 Jan
34
Feb
Mar
Apr May
Jun
Jul
Aug
Sep
Time-Series Data
When the target is the focus and overall trending of time-series data is less important, there is another chart type that can provide very good visibility:
% Deviation from Target 20
10
0
-10
Jan
Feb
Mar
Apr
May
Comparing Nominal Values Over Time When trending is less important than the comparison of nominal values over time, there are a couple of different options. The first is a grouped bar chart which uses the gestalt principle of proximity to clearly identify the time intervals:
Regional Sales YTD 5 4 3 2 1 0 Jan
Feb
West
Central
Mar
East
The impact of trending over time is lost, but the ability to directly compare values is gained. Identifying a target can be challenging here, unless there it can be accomplished by a single line. Another option is to use a stacked bar graph, which is essentially accomplished by placing the bars in the above graph on top of each other. Stacked bars can express a real value or a percentage (which is suited to part-to-whole data). When expressing a nominal value the
Time-Series Data
35
stacked bar also shows an overall trend:
Regional Sales YTD 15 10 5 0 Jan
Feb
West
Mar
Central
East
Stacked bars suffer from the problem that judging area by-eye is required. Effective use of hovers can provide additional data for each item comprising the stack, which is an efficient way to provide datadensity.
Combining Time Periods When the blocks of time are representative of many instances of that time, not necessarily a sequential period, a bar graph can be used to help to disassociate the time periods, which are not truly trends. An example would be data showing visitors to a website by day of the week – the data might combine all Mondays for the year-to-date into one data point for Monday. In this case, a vertical bar chart might be more applicable:
Web Visitors (YTD by Day of Week) 4000 3000 2000 1000 0 Sun
36
Mon
Tue
Wed
Thu
Fri
Sat
Time-Series Data
The Sparkline First used by Edward Tufte in the 1990‘s, the sparkline is a mini line graph used to show time-series data in a preview format. Described as a ―dataword‖ by Tufte, the sparkline does not have a scale, and is often inserted as an element within a table.
The sparkline can be used in tables, or associated with counters, etc to provide a preview of trending.
Best Practices for Sparklines
Start with the most basic design and add additional elements only as is necessary.
Indicate the time range if possible.
Two sparklines can be imposed over each other when comparison is important. Since there is no legend, each line should be color coded to match another element on the dashboard.
If you need a scale to make the sparkline meaningful, consider using a line chart instead.
Exercise 7: 1. Identify at least three time-series measures for your organization. 2. Consider the role of targets and ranges for these measures. 3. Compare the two time-series charts provided by the instructor. Which one is more effective? Why? ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________
Time-Series Data
37
Chapter 6. Ranked Data Ranked data organizes discrete elements in a worst-to-best (or bestto-worst) relationship. Examples include:
Sales by salesperson. Productivity per factory. Percent of target, by department.
Since ranking is essentially a nominal comparison, the bar chart tends to be the most effective visualization. However, the ranking element adds an additional dimension for consideration compared to a ‗normal‘ bar chart.
Issues by Region East South Central West 0
1
2
3
4
5
6
Ranking Best Practices We have already covered best practices for bar graphs, but let us add a few points when the purpose is ranking.
38
A horizontal layout is often more compelling for ranked data.
When using a horizontal layout, consider that the emphasis naturally appears on the object at the top, so you can essentially choose whether to highlight the highest or lowest value, depending on your goals.
Be aware that showing an incomplete selection of data values (for example the Top Five) can make the lowest look like a poor performer, and should be clearly labeled to avoid confusion.
Ranked Data
Targets and Ranges Targets can be shown as a standard for all ranked items, or can be displayed as individual values.
Sales by Region Target
East South Central West 0
1
2
3
4
5
6
5
6
Sales by Region East South Central West 0
1
2
3
4
Ranges can be displayed in the background, but must apply to each ranked item:
Sales by Region East South Central West 0
Ranked Data
1
2
3
4
5
6
39
Ranked items can also be displayed as a deviation from a target, when the target is the more important than reflecting the overall value:
% Deviation from Target -10
0
10
20
West Central South East
Exercise 8: 1. Identify one or more potential needs to show ranking within your organization. 2. What possible dangers are there in showing ranked data? Be prepared to share your thoughts with the group. ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________
40
Ranked Data
Chapter 7. Part-to-Whole Data Part-to-whole data relates individual measures within a larger whole. Examples include:
Sales by category (within the total sales). Units produced by facility (within the total units produced). Calls completed by customer service rep (within the total number of calls).
Part-to-whole data is often (but not always) expressed as a percentage. A common chart used to display this sort of data, is the Pie Chart, but because area is difficult to judge with the naked eye (especially when values are close), a pie chart is not best practice. Instead consider a bar chart which uses a percentage rather than a nominal value:
% of Sales by Region East South Central West
0%
10%
20%
30%
40%
50%
Part-to-Whole Best Practices The rules for bar charts in general apply to part-to-whole visualizations. There are few additional items to consider:
Part-toWhole Data
Ranking the data makes each bar easier to compare.
If using a percentage, be sure to mark the axis clearly to avoid confusion.
41
Pie Charts As mentioned previously, a pie chart is not best-practice visualization because it relies on the eye‘s ability to judge area.
% of Sales by Region West Central South East
When it is decided to use a pie chart, there are some rules to observe:
Sequence the sections from largest to smallest, with any ‗other‘ category being shown as the last item (regardless of size).
Values must add up to 100%.
A legend must be used.
Multiple Part-to-Whole Comparison A 100% stacked bar is often used to display the relative values of each element:
Regional Sales YTD 100%
50%
0% Jan
Feb
West
42
Central
Mar
East
Part-toWhole Data
But, like other charts that rely on area, 100% stacked bars are not very easy to interpret with any accuracy. Generally speaking the first element in each bar is easily comparable, but the others are not. An alternate chart type to overcome this is the double column:
Exercise 9: Compare the alternate visualizations provided by the instructor. They both show the same data. Note any remarks below. ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________
Part-toWhole Data
43
Chapter 8. Frequency Data Frequency data shows how often something occurs (usually the distribution within a defined set of ranges). Examples include:
Size of orders ($0-$10, $10-$20, $20-$50, $50+) Timeliness of deliveries (early, on-time, late) Lateness of accounts receivable (0-30, 30-60, 60-90, 90+)
Showing a single set of frequency data is fairly straightforward and is typically done with a histogram, which is essentially a bar chart or a line graph.
Orders by Size 70 60 50 40 30 20 10 0 $0-$10
$10-$20 $20-$30
$30-40
$40-$50
$50+
The histogram can also be represented with a line rather than bars (this is technically called a frequency polygon):
Orders by Size 70 60 50 40 30 20 10 0 $0-$10
44
$10-$20 $20-$30
$30-40
$40-$50
$50+
Frequency Data
By using lines, multiple data series can be plotted on a single visualization (remember that the ranges on the x axis must pertain to both data series):
Orders & Returns (By Size) 70 60 50 40 30 20 10 0 $0-$10
$10-$20
$20-$30
Orders
$30-$40
$40-$50
$50+
Returns
Histogram Best Practices Since a histogram is essentially a bar chart or a line graph, the best practices for those visualization types still apply. However, there are a few additional considerations specific to frequency data:
The ranges should be kept as equal as possible, or the data will appear skewed. The exception includes data sets where a grouping for less or more than a certain value makes sense (open ended).
Use a bar graph histogram when the actual value of each range is important, use the line version when the overall shape is the focus.
Comparing Multiple Frequency Data Series The line graph histogram can compare a few series of frequency data, but is limited by the fact that more than about four lines becomes difficult to read. A Princeton Professor named John Tukey developed a specific visualization type to accommodate this situation, the Box Plot. The Box plot has several names including the box-and-whisker plot, candlestick plot, and the open-high-low-close chart, however, for dashboard purposes all are similar enough in usage to be covered by the term ‗box plot‘. The box plot is a fairly data-dense visualization, and is made up of several elements, detailed below:
Frequency Data
45
From: Stephen Few (2005). Boxes of Insight Information Management Magazine. URL: http://www.information-management.com/issues/20050801/1033566-1.html
In practice, the box plot appears something like this:
46
Frequency Data
Box Plot Best Practices Some considerations for use of the box plot:
Keep bar widths and spacing between bars the same for each data series, otherwise there is an implied meaning.
Include a key if the box plot shows additional elements such as outliers.
Group data series next to each other for direct comparison.
Their function might not be instantly understandable by all audiences. Some explanation might be required. Their use in dashboards is somewhat limited – the box plot is more often used to display research data.
Exercise 10: 1. Review the brief scenarios below and identify which visualization type would be most appropriate: a. Average number of orders per day, by day of week (where the weekly pattern is the key element). b. Number of open cases, by priority 1-5 (where volume is the key consideration). c. Number of days taken for orders from vendors to arrive, by vendor (with twenty vendors).
Frequency Data
47
Chapter 9. Correlation Data Correlation data displays the relationship between two (or more) variables on a set of data points. Examples include:
Lateness and dollar value of accounts receivable. Test scores compared to hours spent doing homework. Real estate prices compared to proximity to freeway access.
In the simplest form, correlation is shown using data points on an x-y axis (known as a Scatterplot), often with a trend-line:
Scatterplots can show multiple data series, which can be revealing:
As seen above, color and/or shape can be used to differentiate data series. However, more than two (or perhaps three) data series on a single scatterplot can quickly become difficult to understand.
48
Correlation Data
Best Practices for Scatterplots Scatterplots conform to the best-practices of a number of other chart types:
Clearly label each axis to avoid confusion.
Only plot against values that have a correlative relationship. (Unless the goal of the chart is to identify whether correlation exists).
Use a scale that provides the best visibility for your data (not necessarily a zero-start).
Use hovers to provide additional information about specific data points.
When data points overlap, consider using circles (with a transparent center) rather than solidly filled dots. Another technique is to make the data point marker smaller.
Into the Third Dimension Some data has a third correlative dimension, which can be accommodated in a scatterplot by varying the size of the ‗dot‘. This visualization type is often called a Bubble Chart:
Correlation Data
49
Best Practices for Bubble Charts When using a bubble chart. Consider the following best practices:
Clearly identify all axes (including the third, which might require an additional label).
Use hovers to allow the view additional information about each data point.
If data points are close together, consider using an un-filled circle rather than a ‗dot‘ so that overlapping points can be seen.
If time is one of the dimensions, place it on the x-axis for maximum clarity.
Using size for the third dimension makes big differences easy to spot, but not small ones. If differences are small (but noteworthy) consider using color or shape instead.
Exercise 11: 1. Identify one or more correlative relationships in your organization. 2. Sketch out a scatterplot or bubble graph using best practices. 3. Share your visualization with a partner and have them explain what they think your chart communicates. Make any changes you deem necessary to more effectively communicate your intent.
50
Correlation Data
Chapter 10. Tables While it might be debated whether tables are really a visualization type or not, the fact remains that they have their place in dashboards. When appropriately used, tables have the following advantages:
Tables can show multiple data series over multiple dimensions.
Tables can adapt to the viewers needs through user-controlled sorting.
Tables can integrate other visualizations (particularly sparklines and icons).
However, if tables are not used appropriately and designed effectively, they can be overwhelming, difficult to read, and self-defeating.
When to Use Tables Tables have the unique ability to display multiple data series over multiple dimensions. They can be used to display nominal or time series data, can incorporate ranking, and can display alerts. With this list of credentials, why not use table all the time? Well, tables do not typically answer a single question (in fact they are particularly helpful when it is not clear what particular question the viewer will be trying to answer). As a result, tables do not provide the sort of instant communication to sensory memory that many other visualization types do. In fact, tables generally require the viewer to frame their own question and then use the table to find the answer. For this reason, they should be used sparingly, and should perhaps be limited to viewers who have the time, inclination, and need to frame their own questions.
Elements in Table Design Within the design of tables, there are several elements to be considered, these include:
Data organization Visual layout Use of icons and sparklines
We‘ll consider each of these areas in detail.
Tables
51
Data Organization Columns vs. Rows Since tables consist of rows and columns, an early design decision is which dimensions to organize along which axis. A few rules apply:
Time-series data should be displayed left-right for maximum readability.
If one set of dimensions has just a few divisions, while another has a lot, the larger number is best displayed as rows, with fewer columns.
Unless overridden by the two previous rules, use columns for dimensions whose values are in most need of comparison.
Sequence of Columns In addition to the rule listed above that time sequence takes priority in the sequence of columns; there are two rules that apply:
If there is a hierarchical relationship between elements, the column sequence (from left to right) should reflect that.
Any values that are calculated should be placed in the column to the right of the metric from which they are calculated.
Number Formatting Making numbers as readable as possible is an important goal. Here are some rules for accomplishing that:
Whenever numbers represent a specific unit (such as dollars or a percentage) the appropriate symbol should be used in each cell, not just in the header.
For large numbers, use commas to indicate thousands. If all numbers in a series are of the same minimum size, truncate the values (and note the truncation in the column header).
Display only as much precision as is necessary for the table to fulfill its purpose. (Do you really care about annual revenue down to the exact dollar?)
Visual Layout Visual layout is a key element in table design. Too many tables appear cluttered, and are difficult to read. In general, the data-ink and chartjunk principles apply to table design, but there are some specific best practices too.
52
Tables
Spacing As discussed in a previous section, spacing can alter the perception of alignment within tables. A good rule is that the space between rows should be equal (but no more than the row height). Spacing between columns will vary, based on the needs of the data, but the principle of proximity should always be considered.
Alignment A few simple rules apply to the alignment of elements within tables:
Numbers that represent quantitative values, as opposed to those that are merely identifiers (e.g. customer numbers) should always be right aligned.
Dates and text should be left aligned.
When negative values are expressed by using parentheses, the last digit (not the parentheses symbol) should be right aligned with the last digit of positive values.
Gridlines The gestalt principle of continuity tells us that items that are aligned will be perceived as a group. Gridlines can actually break this up, causing the brain to do extra work to associate elements that are otherwise aligned. This means that many fewer gridlines are necessary in tables than are commonly seen. A few best practices are:
Use gridlines only to separate elements that are dissimilar in nature (such as column headings and summary data, but not each data series).
Gridlines are more effective if displayed in a lighter saturation than the data color (i.e. use gray gridlines for black text).
Bolding & Shading Bolding and shading are similar in their usage in tables, and can be used to accomplish various things. In a non-formatted table, any bold or shaded areas will be preattentively highlighted, and so the most common use of these elements is to highlight exceptions. However, bolding and shading can also be used to group elements such as column headings and summary data such as totals. When vertical space is at a premium, shading can also be used to assist reading when the space between rows or columns is limited.
Icons & Sparklines While tables are effective at displaying multiple data series in multiple dimensions, they can also highlight anomalies, and to some extent trends.
Tables
53
Icons Icons are commonly used in tables to indicate some sort of alert status. This is most effective when an icon is only present when the data point is in an alert state, allowing the presence of the icon to be processed preattentively. The nature of an alert icon is subjective, since the icon is purely symbolic. However, building on culturally accepted norms such as an exclamation point, or the color red can aid in understanding and reduce the need for explanatory text on the dashboard. Icons are often used to give some indication of trending, such as the ubiquitous arrows used to symbolize an upwards, steady, or downward trend respectively. These are often combined with the use of color to add a dimension of good vs. bad. Overall icons can be helpful if the following rules are applied:
Icons are most effective when used sparingly. The rule of thumb is to start with the minimum requirement and add elements (such as size, color, and shape) only as necessary. The best icon is the one that has the lowest visual impact but still accomplishes its goal.
The meaning of icons must be consistent across the dashboard, and must be intuitively understood by ALL viewers, or explained through a legend or similar device.
When combining colors to icons (such as red for bad, green for good), these colors must be culturally accepted by all viewers, and should be in different saturations levels to allow colorblind viewers to distinguish between them.
Sparklines An additional technique for indicating a trend is a sparkline (which is discussed in more detail on page XX). Within tables, sparklines can add additional detail. Exercise 12: 1. Identify any possible uses of tables in your organization. Be prepared to share them with the group. 2. What are the possible dangers of using tables in dashboards? ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________
54
Tables
Chapter 11. Maps Some data is tied to a geographic location, and displaying it on a map can aid understanding. Maps also tend to be fairly slick visualizations and can create excitement around a dashboard project.
When to Use Maps Because of their visual appeal (but not necessarily effectiveness). Maps are perhaps overused on dashboards. They are great visualizations for some specific situations:
To categorize regions into four or fewer groups. A US map showing states in red or blue for Republican vs. Democrat is a good example. More than four groups can be difficult to interpret.
When proximity of the geographic locations adds meaning to the data, such as a map showing outbreaks of a virus.
When the user understands the environment in primarily a geographic sense.
When the geographic data is multilayered, such as sales by state, and each store with the state.
Best Practices for Maps Perhaps the most important best-practice relating to maps is the decision whether or not to use one. Beyond that, there are several rules that apply only to maps:
Maps
Use different colors to identify different status (such as Republican vs. Democrat).
55
Use shades of the same color to indicate degree or size of the same value (such as population size). However, do not use more than three shades if accurate determination of the particular shade is necessary.
Use hovers to provide additional data-on-demand.
The overall size of a map will be determined by the readability of the smallest regions. If space doesn‘t allow that, consider a different visualization type.
Using Maps as Selectors Often, maps are used as a way to get an initial idea of status and then to allow the viewer to select a region to view in more detail (via a drill down). While purists like Stephen Few would frown on this approach, it is a somewhat common practice in corporate dashboards where the look of a dashboard can have a positive influence on adoption. Exercise 13: 1. Identify any possible uses of maps in your organization. Be prepared to share them with the group. 2. Review the example maps provided by the instructor. Make a list of any problems with each one. ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________
56
Maps
Section Three: Putting it all Together Overview This section brings together the principles and concepts taught in previous sections concerning data visualizations and adds an overall dashboard dimension to what has been essentially a discussion of individual KPI‘s. Certain principles for effective design are reviewed, as well as layout, drill-down, and the tension between aesthetics and effectiveness.
Learning Objectives On completion of this section three, you will be able to: Explain the characteristics of a well designed dashboard. Apply principles for effective dashboard layout, including multipage and drill-down design.
Section Three
57
Chapter 12. Characteristics of Effective Dashboards
When we speak of a dashboard, we are essentially talking about a collection of data visualizations brought together to form an insightful view of some area of organizational performance. Effective dashboards contribute to a viewer‘s ability to make decisions about their organization. In addition to the correct selection of visualization types for individual metrics, and the best practices used to make each visualization as meaningful as possible, an effective dashboard:
58
Acts as a spotlighting mechanism, de-emphasizing normal values and drawing attention to the exceptions.
Is based on a specific objective (or set of objectives), and provides insight which helps the owner to make decisions to reach those goals.
Is data-dense, packing information efficiently in the available space.
Provides details-on-demand, which allow the viewer to determine how much information they need about a particular topic.
Begins with summary information (relevant to the owner) and provides a hierarchical navigation that allows a view to view more specific information as it is needed.
Characterisitcs of Effective Dashboards
Balances aesthetics with efficient communication.
Is developed in phases and is updated to reflect the current business need.
Each of these characteristics will be considered in more detail.
Highlighting Exceptions A strength of dashboards is their ability to highlight exceptions. Exceptions are typically important in an organization because they warrant additional examination – either because they are over or under-performing.
Preattentive Attributes The knowledge and effective use of preattentive attributes is key in highlighting exceptions, and also in de-emphasizing values that fall into a normal range. Especially effective for communicating exceptions are the attributes of color, added marks, and size. Although these elements do not necessarily do an effective job at communicating the degree of difference. When that is needful, consider 2d position and line length as the most helpful. Sometimes these elements can be combined. Consider the following chart that uses color hue and line length to highlight the exception:
Sales by Region 5 4 3 2 1 0 West
Central
South
East
Goal Based Generally speaking a dashboard is created with a particular viewer in mind. This individual (also known as the dashboard owner) will use the dashboard as a tool to accomplish a specific task. That task might encompass all or just some of the owner‘s entire responsibilities within an organization.
Characteristics of Effective Dashboards
59
With that in mind, the dashboard has a rather specific scope – to provide tracking, insight, and enhance decision making relative to the goal.
Identifying the Goal The dashboard owner is the best person to identify the goals of a dashboard, but will frequently need help from the dashboard designer to express the goal in a succinct way. One effective technique to help a dashboard owner is to ask them to think ahead to a time when the dashboard is in place, what will they be able to do more effectively as a result? Their answers can illustrate their goals.
Embedding the Goal in the Dashboard Once the goal is understood, it can be used as a litmus test to review each proposed element on a dashboard. Will the proposed element increase the owner‘s ability to reach the goal? If not, it is probably not needed.
Details-on-Demand Effective dashboards recognize that the viewer doesn‘t need all of the information, all at once. Instead, effective dashboards provide detailson-demand, meaning that a viewer can get additional information as it is needed. One effective technique for providing details-on-demand is through the use of hovers. Closely related to the concept of detail-on-demand is the idea of datadensity. The term ‗data-density‘ has been used previously in earlier chapters, but a more detailed explanation is probably appropriate at this time.
Data-Density Data-density refers to the amount of information relative to the space it occupies, with the general rule that more information in a smaller space is better. However, this principle could be misconstrued to suggest that packing a table with 30 columns in six-point font is data-dense. But even a dashboard rookie could identify that it wouldn‘t be readable (and therefore not useful). So there is a balance between data-density and details-on-demand that suggests a dashboard is economic in its use of space, but also effective in its communication.
60
Characterisitcs of Effective Dashboards
Hierarchical Navigation While the notion of detail-on-demand refers to the use of hovers to provide additional information for a certain dashboard element, the principle of hierarchical navigation suggests that summary data will be provided first, with the ability to drill down into expanded detail based on the viewer‘s selection. For hierarchical navigation to be truly effective, it must be based on the types of questions that the dashboard view will be seeking to answer, rather than just being based on available data. For example a ranked bar graph showing sales by rep that drills into a view of product mix is only helpful if understanding the mix of products helps the dashboard owner to more effectively reach their goals. Perhaps a drill-down showing the number of sales calls made would be more helpful. An understanding of goals is critical here.
How Many Levels? Drilling through levels of data can become confusing for the viewer, although some assistance through bread-crumb navigation can be helpful. So how many levels of drill-down is the maximum? Actually that‘s really the wrong question. The right question to ask is ―is this drill down necessary to help the viewer reach their goal?‖ Too often drill down is provided because the data is available and it would be ‗cool‘ to see the extra dimension. Keeping drill-downs in line with the owners need for relevant insight will greatly enhance dashboard effectiveness.
Aesthetics vs. Communication Many organizations that undertake dashboard projects are initially attracted to the glitz of colorful visualizations compared to dull spreadsheet reports. The desire to make the dashboard ‗cool‘ or <insert your own adjective here> is often a very real pressure applied to the dashboard designer. However, long term effectiveness is based on the usefulness and insight provided by the dashboard, not the sleek look and feel.
Finding Balance So, how do we resolve the natural tension between cool and communication? While there is perhaps no complete answer to this question, we do have a few ideas to help you walk the tightrope:
Characteristics of Effective Dashboards
First seek to educate the dashboard owner about the existence of this natural tension between form and function. If they understand why you are pushing back against certain requests,
61
they will be more likely to agree.
Introducing the idea of goals as early in the project as possible will help to frame the dashboard conversation around communication and effectiveness, not look.
Recognize that user adoption is an important part of a successful project. Making your dashboard visually attractive will often entice a viewer to look at it, making the dashboard effective will keep them coming back.
Occasionally you might have to incorporate an element that you know will not be effective. Make a point to follow-up and discuss whether the dashboard is accomplishing all that they wanted. Often you‘ll have the chance to remove or change the element you originally objected to! We‘ll leave the decision whether or not to say “I told you so” up to you!
A Word on Branding Back in chapter two, we introduced the concept of ‗chart-junk‘ as unnecessary elements on a dashboard. It‘s possible that dashboard purists would look at corporate branding elements as chartjunk. However, we have found that most companies want (and even require) their corporate brand to distill through into their dashboards. Our experience has been that trying to resist this desire is often not helpful in the context of the overall project. The phrase ‗pick your battles‘ comes to mind. Instead consider ways to apply corporate branding in a de-emphasized way. Meet the needs of the organization without compromising the dashboard‘s ability to communicate effectively. Exercise 14: 1. Pick one of the characteristics of effective dashboards and explain it to your partner. Be sure to explain how to use it, and the possible consequences of not using it. 2. In a team, review the sample dashboard provided by your instructor. Discuss how well each of the characteristics were applied. What improvements could be made? Nominate a team member to present your points to the whole group. ______________________________________________________ ______________________________________________________ ______________________________________________________ ______________________________________________________
62
Characterisitcs of Effective Dashboards
Chapter 13. Dashboard Layout The organization of a dashboard is contingent on a number of factors. Also it‘s worth noting that there is no ultimate right-way to organize a dashboard. We‘ll review the relevant factors in this chapter.
Types of Metrics There are essentially two types of metrics found on dashboards, outcome metrics and driver metrics. We‘ll take a look at both types:
Outcome Metrics Outcome metrics measure overall results. They are often the end results of multiple sub-processes. Examples include sales, profit, # of widgets manufactured, customer satisfaction, # of calls processed, etc.
Driver Metrics In contrast, driver metrics are indicators of sub-processes that create the outcome. Driver metrics are usually tied to the outcome they ‗drive‘. For example, for the outcome metric ‗sales‘, drivers might include number of sales calls made, number of deals in the pipeline, etc.
Outcome vs. Driver The important thing regarding outcome and driver metrics is to understand them from the perspective of the dashboard owner. Metrics that are outcomes for the VP of Sales might only be drivers for the CFO (although both might want to see them on their own dashboard).
Keys of Dashboard Layout Once the type of dashboard is understood, there are a few key factors for consideration in dashboard layout.
Views Throughout this course it has been assumed that a dashboard has a specific owner. We don‘t really want to get bogged down in semantics, but it might be more appropriate to say that each dashboard viewer has a specific view that is important to them. Each view could be a separate dashboard, or could be a separate page within a multi-page dashboard. In this case, it‘s more correct to refer to individual views, rather than individual dashboards. The first step in organizing dashboard layout is to ensure that the right metrics are on the right views. If your dashboard truly has a single owner, identify whether the metrics create any ‗natural‘ groupings. For
Dashboard Layout
63
example, a financial dashboard for a CFO might have two natural sets of metrics about income and expenses. For dashboards with multiple viewers, or multiple views, these views should be organized into separate pages, with the pages being sequenced based on an estimate of the most used or most important (according to the dashboard owner).
Outcomes and Drivers For most dashboards it makes sense to show outcome metrics before the metrics that drives them. Depending on the number of metrics, this could be accomplished in several ways. Here are a few possible layouts (note the use of the gestalt principle of enclosure to show their relationship): Three outcome metrics with their associated driver metrics:
Two outcome metrics, both with two associated drivers:
64
Dashboard Layout
A single outcome metric with multiple drivers:
In many instances, it‘s not a clean as these three designs indicate, but wherever possible associating outcomes with their appropriate drivers makes sense.
Hierarchy and Drill-Downs Metrics that are sub-sets of other data displayed in the dashboard are often displayed as drill-down elements, allowing the viewer to access the additional information if they need it. There are three options for how a drill-down will function:
Dashboard Layout
The new chart pops-up in a smaller window in front of the current dashboard. When it is closed, the original dashboard window is still in place. This technique is used to show supporting information in more detail. Nested pop-ups (a popup within a pop-up) can be problematic to navigate.
The dashboard changes to a new page. This is often done to integrate multiple views, for example a high level outcome metric such as sales, drills down to a detailed page with several sales related metrics. The new ‗page‘ is also directly linkable, and might serve as the ‗home page‘ for a different dashboard owner.
The new chart appears in place of the original in the dashboard view. This works best if other metrics on the page update based on what is showing. This approach is often usedin conjunction with user controls such as drop-downs and radio buttons to allow the user to select their view.
65
Effective Comparisons Proximity and similarity in dashboards will often prompt comparisons. Dashboard designers need to use these as tools, and make sure they arenâ&#x20AC;&#x2DC;t mis-used by creating false comparisons. Here is an example:
Placing these two line graphs next to each other invites a false comparison since the scale for the chart on the left is in thousands. Sales for Product A are actually 100 times greater than for product B. Another component that can encourage comparisons is color. If red is used as an alert color on most graphs, it will be considered an alert on all graphs.
Priority In the Western world, we read left-to-right and top-to-bottom, which creates an implied priority for items that are located in the top-left vs those in the bottom right. As a result, metrics within a particular view should be organized based on their relative importance whenever possible, with the most important in the top-left position. Another important consideration regarding priority is to consider the display method, and understand what will be visible on the screen without the need for scrolling. This concept is called â&#x20AC;&#x2014;the foldâ&#x20AC;&#x2DC;, having first been applied to newspaper stories that appeared at the top of the page (above the fold). Metrics shown above the fold will naturally be considered more important than those below the fold. Placing more important measures above the fold, with secondary information below the fold will help the dashboard viewer understand the priority of the information presented.
Exercise 15: 1. Based on the case study info provided, organize the visualizations based on the rules for layout.
66
Dashboard Layout
Getting Layout Right The principles listed above will certainly provide a guide for dashboard layout. But as stated earlier, there is no ultimate answer to dashboard layout. However there are two techniques that can considerably help in finding the right layout.
Prototyping It could be as simple as a sketch on paper, or a screenshot style mock-up, but either way, delivering a prototype to a dashboard owner prior to actual development can certainly save time and contribute to a more effective deliverable.
Usability Testing Ultimately the â&#x20AC;&#x2014;rightâ&#x20AC;&#x2DC; layout will be the one that is most helpful to the dashboard viewer. One very insightful exercise is to observe the dashboard owner interacting with the dashboard.
Dashboard Layout
67