What are the most important principles to keep in mind when creating a data visualization?

I think this is extremely important. It's not what you find interesting, it's what's relevant to the question you're being asked. What does your audience need to know? If you discover something interesting and useful, but tangential then place it in am addendum after the main presentation. Keeping a good flow and narrative is important to explain complex relationships.

Knowing your audience is so helpful not only during data visualization. It can help you write and deliver your findings in effectively.

Beyond just knowing your audience - do they have any special requirements e.g., I was working for a client who was colourblind so we ensured we chose a pallet that enabled change to be perceived easily with the colours they could differentiate.

Absolutely important! Usually, I create more content than I expect to present. This allows me to go deeper if it feels appropriate for the audience. If the audience is new to me, and I’m using visualizations which are not familiar with folks in general (ie: box plots), I’ll offer to review how they are interpreted. I use boxplots often enough, and found myself reviewing their interpretation enough that I also keep a ‘How To Read A Boxplot’ slide on hand. Lastly, I believe it is safe to assume most audience members can read. As such, I shy away from reading bullet points from slides which folks are capable of doing on their own (more pictures, less words).

For those who are just starting out with data visualizations, I recommend visualizing the same data in every way possible - switch up the variables, the axes, and especially the type of visualization. Ask yourself, “which visualization makes the most sense for these variables? Can I somehow visualize a third variable? A fourth? At what point does this not make any sense?” Interacting with the data in this way helps new data practitioners understand the value of different visualization types and develops the creativity needed to tease out insights. It also provides ample opportunity for the analyst to understand their data better.

Always begin with plotting the raw data. What the lower dimensional data have to say ( is the distribution normal? non-symmetric? Single or multi modal?). When I jump to the higher complexity visualizations (say, heatmaps) without taking this first step, I tend to regret it. Use interactive plotting tools (ie: Rstudio). Write down the question you wish your visualization to answer. I often use these as plot titles. This simple act helps me assess if my plot is effective. I have found boxplots for grouped time series data to be very effective in seeing subtle emerging trends in data. I particularly like that boxplots assume no distribution. I avoid pie charts entirely. They seem a poor choice for presenting fractions of a population.

Consistency and simplicity are my guiding stars. If you need to explain many relationships make many charts. Understanding the relationship between two variables is easy even for a layman, each layer you add increases the complexity exponentially. If you need to show a complex series of relationships, build to it by showing the smaller steps and relationships that lead to the greater whole.

For me, this is the most important one, combining data storytelling with visualization often yields the most significant impact. Visualizations provide the evidence, while storytelling adds the context and emotional connection necessary for the audience to understand and act on the data. Together, they create more valuable and informative means of conveying insights. It’s like adding the finishing touches to a beautiful painting. While data visualization provides the vivid colors and shapes. ????

Dr. Scott please help me, I would like to know more about storytelling in data analysis. I understand the concept but I want to have a clear understanding of the methods. Thank you

I would argue a story is told less by one visualization and more by the combined visualizations shown during a presentation. It is important for each plot to answer a question, or have a message. And care should be taken to design a presentation that then tells a story through the series of visualizations.

Presenting raw data first is a great way to build trust. Checking in with colleagues and peers as I develop a presentation has been very valuable to get perspective on if I am accomplishing what I think I am (and what I hope to) with some plot or another. Remain on guard for falling prey to selecting data/plots which support a narrative and passing over data/plots which do not support a certain narrative. Another great way to gain trust? Share your data so others are able to explore the data and potentially reproduce your findings.