Just a few days ago, SalesForce purchased Tableau for more than $15 billion. This alone indicates how mainstream data visualization* has become.
*Note- For the purpose of this post, I am using Tableau’s definition of data visualization: Data visualization is the graphical representation of information and data. By using visual elements like charts, graphs, and maps, data visualization tools provide an accessible way to see and understand trends, outliers, and patterns in data.
It is no longer just a nice-to-have feature of software programs. Entire industries and professions are now dedicated to this field.
It is becoming easier by the day to throw your data into a software and a few clicks later you have a visual of your liking. There are thousands of resources available to learn the how-tos of data visualization: how to choose the right chart, how to create a visual in a certain software, how to build a dashboard within minutes and so on.
The Law of the Instrument
The abundance of resources and pervasiveness of visuals in all media and industries, is proof that data visualization is now the proverbial hammer. First proposed by Abraham Kaplan, the law of the instrument refers to a cognitive bias that suggests over-reliance on a familiar tool.
Give a small boy a hammer, and he will find that everything he encounters needs a pounding… What is objectionable is not that some techniques are pushed to the utmost, but that others, in consequence, are denied the name of science. The price of training is always a certain “trained incapacity”: the more we know how to do something, the harder it is to learn to do it differently.Abraham Kaplan, The Conduct of Inquiry
Over-reliance on data visualizations accompanied by trained incapacity results in dashboard graveyards that are full of pretty pictures but no actionable insight, visuals that intentionally or unintentionally distort perception to mislead and uninterpretable infographics that are nothing but visual clutter.
You really do not need to visualize every data finding for every viewer. Sometimes, all you need is a table!
How do we comprehend visual information?
To decide when and when not to visualize data for your audience, it is important to understand how visual information is processed by the mind. Steven Pinker proposes the following model:
- Visual array: Unprocessed neuronal firing in response of visualizations – visual stimulus
- Visual description: Viewer creates a mental representation of the visual stimulus
- Match: Viewers search their prior knowledge relevant for interpreting the visualization. E.g identify that the given graph is a bar graph.
- When a matching graph is found, it becomes instantiated.
- Message assembly: The viewer then perceives and interprets the visual
- Conceptual question: particular information that the reader wishes to extract from a graph
- Conceptual message: information that reader in fact takes away from the graph
Takeaway: Prior knowledge and perception are critical to the information that a viewer extracts from a visual.
Vessey’s cognitive fit theory suggests that if the visualization matches the viewer’s mental representations, the conceptual message answers the conceptual question. If the visualization does not match the mental representation, then the viewer performs cognitive transformations to make the two align.
Padilla compares this process to translating between languages:
If a visualization is communicated in your native tongue (in a way that matches how you naturally think about the data), then it is intuitive and doesn’t require any translation. If the visualization is communicated in a way that doesn’t match how you think about the data, then you have to translate the information to understand it. As with language, translating visual information is error-prone, time-consuming, and effortful.
Vessey describes two types of tasks (questions) that a viewer might perform (answer) with information conveyed in graphs or tables:
- Spatial tasks: Extract information about relationships in data.
- Example: In which month, is the difference between the referrals and conversions to visit the greatest?
- Symbolic tasks: Extract precise and discrete data values.
- Example: How many referrals were received in the Month of May 2018?
According to this qualification, data visualizations (graphs, maps) are spatial representations and tables are symbolic representations. Matching task type to representation type ensures cognitive fit.
In the simple example below, if all the viewer is looking for “What is the sum of A and B?”, he can get to a more precise answer faster using the table instead of a graph.
Back to our original question: When should you visualize data?
Visualize data for your audience only when it achieves cognitive fit. Visualize data only when the viewer wants to know about the relationships in data. Do not visualize, if the viewer is interested in precise values.
When NOT to Visualize Data
Don’t use a visual for symbolic tasks. Use tables for symbolic tasks.
Stephen Few agrees that tables are better than graphs when the answer to the viewer’s question lies in
- Individual values
- Comparing selective pairs of values and not entire series
- A precise value
- Both summary and detail values
- More than one unit of measure
OR the scale is broken: a lot of small numbers and a few very large numbers, or vice versa.
Nigel Holmes explains the scale as follows: Tables work especially well when numbers differ by orders of magnitude so that no scale suffices in plotting them. Consider this set of numbers: 20; 400; 160,000; and 25,600,000,000. A chart will lose the detail of low figures if it tries to reach the high ones without breaking the scale. If you must break the scale, then just use a table, because a chart with a broken scale is no longer a true picture of the numbers.
When to Visualize Data
Use visuals (graphs, maps) for spatial tasks, when the answer to the viewer’s question lies in
- the shape of the overall data (e.g. patterns, trends, exceptions)
- relationships among whole sets of values (distribution, part-to-whole, correlation)
AND the scale is not broken
Importance of Cognitive Fit
What if cognitive fit is missing? What happens when the task type doesn’t match the representation type? What happens when data is visualized in a graph and all the viewer wanted was a number?
The viewer must perform some mental transformations on the visual based on their prior knowledge and perception. These transformations may include: mental rotation, creating a mental image, modifying the mental image by adding or deleting features to/from, comparison between different views. This increases mental effort and reliance on visual perception. Both these factors can vary greatly among viewers.
Arien Mack’s paper on visual perception discusses how much more or less we see than would seem to be afforded by the retinal stimulation. Some of the phenomena that impacts what we think we see are: inattentional blindness (how you might miss a man dressed in gorilla suit walking through the court), attentional blink, change blindness, visual neglect, priming.
We cannot look squarely at either death or the sun.Francois de La Rochefoucauld
The complexity of visual perception and variation in prior knowledge among viewers further speaks to importance of cognitive fit to make sure your choice of graph vs table is informative, not misleading.
Lastly, only share a visual/table with your stakeholders, if it is strategy-centered: answers the business question by communicating information about a relevant measure of success or a measure of variation.
If it does not promote insight or coherent action, drop it.