When we build a data visualization we are building vocabulary to translate our data into a message, and this vocabulary can be decomposed in a certain number of elements. We always do that, even if you may not conscious about it. Being aware of this helps us to think about how to better build this vocabulary and, ultimately, how to make your message more effective.

Here we will discuss these components, together with the fundamental design principles of data visualization, which are a set of guidelines to help us finding the most appropriate representation for our data.

Marks and channels

Marks and channels are the first two fundamental building blocks of a data visualization.

Marks

Marks are the geometric elements that we use to identify the items of our dataset 1. Since we are plotting on a two dimensional surface, we can use any geometric entity with dimensionality less or equal to two to represent our items, so we can use:

POINTS LINES AREAS
Someone also uses three-dimensional objects as markers. Please, don't!
Their encoding implies that we must model the perspective, and this causes a distortion into our perceived values, making them not suitable for data visualization.

Channels

We then have the channels, and they encode the values of the data associated to our items.

The most commonly channels used to encode quantitative information are

Channel Example
Position on aligned axis
Position on unaligned axis
Length
Width
Angle/Slope
Area
Color luminance
Color saturation

On the other hand, if we want to encode a categorical information, we have the following channels:

Channel Example
Spatial region
Color hue
Shape/Texture

The order of the items in the above tables is not random, but it reflects how easily we translate the visual information either into a quantity or into different categories, and this property is called effectiveness.

Other components

Marks and channels are the components which encode information about our data, but they are not the only components which constitute a data visualization. Other components are the ones which allow us to interpret, compare and give context to our data.

  • axis
  • grids
  • annotations
  • legends
  • labels
  • ticks
  • reference lines
Annotations may also be used to draw attention to patterns of interest.

Use this components only if they really help your reader. Remember that, most of the time, you want the reader to easily compare the values of your data, not to be able to assess the exact value of your data. You should keep your visualization as clean as possible, and in order to do so:

  • Avoid useless boxes
  • Don’t use grids if they don’t help understanding the data
  • Avoid too many ticks 2

Design principles

The above mentioned concept should be applied, as much as possible, according to the principles of data visualization design. These principles, that we will discuss in the rest of this post, will allow the reader to easily understand and decode the visualization, reducing the risk of a misinterpretation and so making the communication between you and your reader clearer.

Expressiveness principle

The expressiveness principle essentially states that we should make the message as clear as possible, without neglecting information and, probably most important, without adding, either implicitly or explicitly, information.

The visual representation should represent all and only the relations that exist in the data.

Some examples of violation to this principle are:

  • Line chart used to represent categorical data
  • Different color intensities to encode different categories
  • A diverging color map to represent a quantity that does not have an origin (a zero).
  • Using a channel that don’t encode any data.

Effectiveness principle

The effectiveness principle is another principle that helps us finding the most appropriate channel for each variable.

The relevance of the information should match the effectiveness of the channel.

The first obvious consequence of this principle is that the most important variable should always be encoded by using a spatial dimension. On the other hand, one should not rely on color to effectively communicate relevant quantities.

Conclusions

We have first seen what are the main components of any data visualization. We have then mentioned two important recommendations on how to translate our dataset into a graph.

You should always keep them in mind, but you should also balance them by taking into account your audience and your message.

Suggested readings

Munzner, T. (2015). Visualization Analysis and Design. CRC Press. ISBN: 9781498759717

  1. If we are representing a network rather than a tabular dataset, they are used to represents the links too. ↩

  2. The ticks are the graphical components that are used to mark the values of the axis. ↩