:Creativity / Equipment and tools

Visual encodings: Translating data into visual elements for everyday impact

woman at work studying woman at work studying

Even if you’re not immediately familiar with the term visual encoding, you’ve likely visualized data in one form or another. You might have used a line chart to better identify trends or color-coded different records in a spreadsheet based on a certain attribute. Whenever you do this, you’re simply translating data into different visual elements and creating rules on how the data is expressed visually.

For example: When you use a bar chart, you decide the length of a bar corresponds to the value of a certain metric. The larger the number, the longer the length of the bar. Therefore, when someone else looks at our bar chart, they learn your applied rules to decode the data. Hence the visual elements we use such as position, length, etc. are known as visual encodings.

The value of visualization: What is visual encoding?

Even though there are many resources with general guidelines on choosing the “right type” of data visualization, understanding visual encoding empowers you to make these determinations even as new and novel visualizations become mainstream.

Understanding visual encoding is also essential in the design and creation of new and custom visualizations. Regardless of the complexity of the final visualization, a successful visualization helps you gain insight about data that isn’t readily revealed by other means.

Encoding and decoding

A study done in 1984 measured how accurately we’re able to decode different forms of visual encodings.

visual enconding imagery visual enconding imagery
Source: Journal of the American Statistical Association

Depending on the type of data, it’s easier to accurately decode certain encodings than others. As we use different visual encoding examples to express different properties of our data, it’s important to understand this fact. Selecting poor encodings can also lead to inaccurate interpretation. A general strategy is to rank the different data dimensions in order of importance and assign a corresponding encoding.

types of data imagery types of data imagery
Source: Journal of the American Statistical Association

Putting it into practice: avatar design process

When I was working on the MRR (monthly recurring revenue) service at Teachable, I thought that a school with a high GMV (gross merchandise value) the previous year but no GMV in the last three months was at risk of churn. However, someone on my team pointed out to me that some schools operate seasonally. So, I decided to design avatars to help distinguish between seasonal schools and schools that enroll students year-round. 

I started the design process by writing down the goal and constraints for this visualization. I also identified different activities and metrics that may reflect seasonality. Writing how derived values are calculated can sometimes be helpful, too.

initial outline initial outline
Initial outline

With this initial outline, I created a shortlist of different metrics that I might use to understand seasonality. In my first iteration, I only used paid enrollment because that affected GMV; however, it didn’t account for another factor: a school may not need to generate revenue to get value from Teachable.

visual encoding first iteration visual encoding first iteration
First iteration
second outline second outline
Second outline

Using color to encode account length was limiting because a school with an account history of three months is at a different stage than a school that has been hosting their product on Teachable for 11 months.

visual encoding second iteration visual encoding second iteration
Second iteration
third outline third outline
Third outline

This version was more helpful in distinguishing schools with different account lengths, but I noticed there were still schools on paid plans that had no paid enrollment. Realizing that some schools may not need to be selling courses to get value from our platform, I decided I needed to encode all enrollments and distinguish paid/unpaid enrollments.

visual encoding final iteration visual encoding final iteration
Final iteration of visual encoding
final outline final outline
Final outline

Visually speaking

We now use the avatars I created in several internal dashboards at Teachable. Because the avatars are very information-dense, they allow the team to easily switch between getting a broad sense of trends, without losing interesting outliers. Without the avatars, some interesting behaviors might be flattened or hidden, preventing us from truly understanding our customers. 

Resources for general guidelines

Further reading

Editor’s note: As part of our editorial goal to expand creators’ knowledge and help you grow in your digital business journey, we’ll be sharing insight from in-house Teachable experts. The preceding post has been a guest post from Sandy Guberti Ng, a data visualization engineer at Teachable, who shares insight on the power of visual encoding and how it’s practically applied to creators at Teachable.



Author: Sandy Guberti Ng, Sandy Guberti Ng is a data visualization engineer on the data team. Her data visualization was shown at the Museum of the City of New York’s New York Responds exhibition.