Statistics for Sustainable Development > Blog > 6 Principles for Data Visualisation

Technical Pieces

6 Principles for Data Visualisation

Data visualisations are everywhere. It seems they are becoming more prevalent in our day to day lives, and not just for us as statisticians, but for everyone. This seems especially true in the days of COVID-19. For months we could not go a day without seeing the latest graphs on cases, hospital admissions, and deaths being shown at the daily briefings. These graphs continue to dominate news coverage.

Data visualisations are arguably the best way to communicate clear messages to the public. But for these messages to be clearly understood and avoid misinterpretation, they need to be well designed and chosen carefully. Poor decisions in graphing data can result easily in wildly different and misleading messages. My mind thinks back to the infamous plotting of the Florida gun deaths from a few years ago, which inexplicably put the 0 at the top of the y-axis, making at first glance the increases look like decreases and vice versa.


Image 1. Famously misleading plot of gun deaths in Florida

To help everyone make better data visualisations, I have drawn up six guiding principles which should help in getting started. These are relevant whether you are making a table, graph, or map.

1. Effectiveness

Effectiveness concerns arguably the most important step - choosing the right type of visualisation. The visualisation you choose for the data and its message, can make a big difference to its effectiveness. For example, trying to plot anything other than proportions that total 100% onto a pie/donut chart just does not work.

A general guide would be that maps are for spatial data, tables are suited for structured numerical information and graphs are very multi-purpose - but are primarily used for indicating trends, making broad comparisons, and showing relationships. Tables can be used for similar purposes, the difference being the specific numbers are more important to a table while the broad patterns are the focus of a graph.

If you are choosing a graph, you need to then consider what type of graph to use. This is far too broad a consideration and could be multiple blog posts in of itself. Therefore, for more advice, please look at our resource on Presentation of Tables, Graphs and Maps.

For example, the pie chart below, while it looks pretty, it has too many categories and therefore it becomes increasingly difficult to read the differences between groups. Plus, it is incredibly difficult to get at any actual proportions because it is all based upon area. Moreover, for this type of data it is not really the proportions that matter. It is the numbers. This graph would have been much more effective as a table.

Image 2. Unnecessary pie chart of national contributions to the sample size of European Social Survey (ESS) 2018

2. Informative

Making sure your graphs are as informative as possible serves two purposes. Firstly, any data visualisation needs a clear message and a clear need. It must have a reason to exist. If there is no apparent message to the data, then the visualisation becomes just a bunch of white noise with nothing to say. It fails to inform because it has no useful information.

Once we have created a data visualisation and it has been publicly shared, it can be difficult to control exactly how people choose to share, present or even recontextualise. Realistically we should be able to provide/withdraw consent to certain uses of our visualisations, but we can’t keep track of every usage.  

This is the second purpose of being informative - providing enough information as possible to ensure that our visualisations can be understood when taken out of the context of its original publication. Any good visualisation should be able to be understood outside its original use and be able to serve as a stand-alone piece.

Therefore, be sure to provide source notes where required as this aids credibility. Additionally, you should be including relevant details such as measurement units, dates, analysis units, maybe footnotes to explain acronyms.


3. Readability

Readability is quite a simple concept that hopefully you should already be following. All elements of your visualisation should be legible, understandable, and coherent. This largely concerns the text elements of the graph such as titles, labels, notes etc.

Titles and headings should concisely explain the content and should not be needlessly long and complicated. The same goes for any axis markers, labels etc.

Of course, the elements will depend on the type of visualisation, but they should always be easy to read and understand. Avoid using language beyond the scope of your target audience. Using relatively broad language will also aid in making your visualisation informative to everyone not just your originally intended audience.

4. Tidiness

This is a similar idea to readability but focuses more on the positioning and spacing of elements and avoiding unnecessary clutter. The purpose is the same; making sure your visualisation can be read and understood.

This includes making sure that no elements are overlapping, there should be adequate spacing between them. Although not too much, otherwise the visualisation will just look empty. Make good use of the “white space”.

This also includes trying to avoid overplotting. Overplotting is when data points overlap making it difficult to read different points. Usually this is because there are too many data points with the same/similar values, or there are a limited number of unique values. This can be avoided by reducing the size of points, sub-setting the data, using transparent symbols, or jittering the points. See more examples in the resource guide. In the graphs below, there are few points because so many are lying on top of each other making the plot look quite empty. The second graph shows how this overplotting can be fixed with a little jittering of the points.

Also, tidiness means avoiding using “junk” features. E.g., shaded backgrounds, borders, patterns, textures, shadows, 3-D graphics etc. Anything that provides no real purpose other than to take up space.

Image 3. An example of overplotting

Image 4. Fixed overplotting by jittering the points

5. Consistency

This is largely relevant if you are intending to use multiple visualisations. It is important to maintain a level of internal consistency.

This involves many aspects, but I think there are two broad ideas at play. Firstly, be consistent with how you plot the data on your visualisation. This means paying attention to points such as order of categories. The order should be kept to a logical or ascending/descending order of a variable. Unless ranking is important to your message, try to keep the order the same throughout. Similarly, if you assign colours to categories, keep these colour schemes the same throughout. Do not change them up at random points as this is just confusing.

Secondly, there should be some level of “design” consistency. So, pay attention to keeping the finer details the same. These are things such as the font, size, face etc. Try to keep the formatting as consistent as possible.

6. Accessibility

Finally, accessibility is increasingly important to data visualisation. We want to make sure as many people as possible will be able to look at our visualisation and understand its message. This includes using non-technical language as mentioned previously, but also considers how to make the content accessible to those with impairments, especially those with difficulties with their vision.

For data visualisations, a lot of this comes down to accounting for colour blindness[1]. This affects about 1 in 12 men and 1 in 200 women. Most commonly “red-green” vision deficiency. Therefore, if we use colours which are inaccessible to people with these impairments, they will not be able to properly read our visualisations and leave them alienated from our research. The two graphs below show just how different someone else may see our graphs.

Image 5. Left - how the graph should look. Right - how it would look to someone with red-green colour blindness

There are endless more specific tips and tricks that I could detail here but many are largely dependent on your data, your messages, and the types of visualisations you choose to create. However, adhering to these 6 general principles of effectiveness, being informative, readability, tidiness, consistency, and accessibility, will hopefully help in getting started and enable you to think a bit more carefully when designing your own data visualisations.



Image 1: Gun death graph:

Image 2: Data for Pie Chart: European Social Survey 2018

Images 3 and 4: Exam Failure Graphs: Simulated data

Image 5: Colour blindness example: Created by Emily Nevitt (Stats4SD)

Alex Thomson
Author: Alex Thomson

Alex joined the team as a Statistics Intern in October 2019 following the completion of his Undergraduate and Master’s degrees in Population and Geography, and Social Research Methods at the University of Southampton. With a background in demographic and social science research, especially family demography - Alex hopes to extend both his skillset and knowledge base of issues affecting the developing world and building upon an undergraduate trip to Ghana. With experience in STATA and SPSS, Alex hopes to develop his skills in R, survey design and management while at Stats4SD.

0 comments for "6 Principles for Data Visualisation":

Add a comment:

We run an anonymous commenting system. If you are not logged in, we do not collect any information on who you are when you leave a comment. This means we manually confirm comments before they appear on the site.

If you want to have a comment you submitted deleted, please contact us, giving the date of the comment and name of the article.

This is the name that will be displayed along with your comment