Research

Data Visualisation

What is Data Visualisation

In simple terms, data visualisation is the representation of information in the form of a chart or diagram. Visualisations help tell stories by curating data and communicating information both clearly and efficiently. Learning to communicate your data in this way is both an art and a science.

Visualisations do not need to be boring to be functional, nor do they need to be sophisticated to look beautiful. A truly successful visualisation will help make complex data more accessible and enables readers to both analyse and reason with the underlying data.

The main purpose of a visualisation is to communicate information.

Different Types of Visualisation

Before you start to think visually, take a moment to consider the nature and purpose of your visualisation. It is helpful to consider your visualisation as existing across a spectrum of two gross domains. First, it is either conceptual or it is data driven. Conceptual visualisations express ideas whereas data driven visualisations express statistics. Second, your visualisation is either declarative or it is exploratory. Declarative visualisations communicate information while exploratory visualisations try to figure something out. You can plot your visualisation onto the diagram below once you have figured out where it lies within this spectrum. Use this diagram to categorise your visualisation into one of four categories: idea illustration, idea generation, visual discover and everyday data visualisation.

Visualisation and Deception

We’ve all read scientific articles before and I’m sure you have noticed how figures are fundamental components. This is because visualisations show how the gathered research data supports the key findings of the article. Unfortunately, visualisations are only as good as the information they portray and they will easily deceive users if they are based upon improper statistics or neglect key aspects of the data. After all, most people will believe any results if they are shown in a well-presented figure.

Typically, authors use figures to present summary statistics of their data. While this might seem like a good way to present data, it is not recommended when the sample size is small and there is not enough data for the summary statistics to be meaningful. This is because there is often not enough data to assess the sample distribution while small samples are easily distorted by outliers. The use of summary statistics is only suitable if performed appropriately. View our article on descriptive statistics for more information on how to perform summary statistics.

Visualisation Choice

The choice of visualisation is equally important. Many scientists choose to use bar and line graphs to represent continuous data. They will use them to show the mean alongside the standard error or the standard deviation of their results. This is a particularly bad idea because many different distributions of data can lead to the same bar chart or line graph. This is illustrated on the diagram below. Furthermore, use of paired data (i.e. side by side columns on a bar chart) could erroneously suggest that groups being compared are independent of one another.

Charting Error and Deviation

The use of the mean with standard error or standard deviation within visualisations can cause readers to wrongly infer that data is normally distributed with no outliers. It is also important to recognise how they are calculated because both measures can give very different visual impressions. Standard deviation measures variation in the sample and standard error measures the accuracy of the mean. The standard error is strongly dependent upon sample size and will decrease as the sample size increases. Thus, if two samples have the same standard error, the one with the larger sample size will have the larger standard deviation.

Honest Presentation

A better idea is to provide detailed information about the distribution of the data or to display the data in full. This enables readers to examine the data and detect any gross violations of the statistical assumptions. Additionally, readers are able to get a better grasp of the study design, and this will enhance their ability to understand the data. Visualisations that enable exploration of data distributions include univariate scatter plots, box plots and histograms.

If your data is from a small sample size and you cannot be certain that it is normally distributed, it is best to use non-parametric tests and show the median value as opposed to the mean. This is because a mean is easily misled by outliers or skewed distributions. When paired or matched data is used, authors should report the median difference rather than the median values for each condition. When creating your visualisations, remember to always consider the type of outcome variable (categorical vs. continuous), the sample size and the study design (independent vs. non-independent).

Selecting Your Visualisation

There are some fantastic websites out there to help you choose what kind of visualisation you can use to represent your data. Two well thought out examples include the datavizproject and the datavizcatalogue. Both websites allow you to search by function, summarise the different types of visualisation available and offer a brief explanation of each type. Below is a brief summary of the different types of visualisations available. Visit the websites above for a more in depth and interactive experience.