Data is best understood when presented in visual format rather than text. So how do you choose the view that best captures what your marketing data is trying to say?
In this post I will cover the main ideas behind a good visualization option.
Your choice of visualization influences the story your data will tell
Data visualization captures any measured action in the customer journey. It means to organize the observations of a dimension or metric into a graph. But the right visualization choice isn’t always immediately apparent when analysts have to work in their data solutions. Solution menus and dashboards often have graphics representing the platforms they were meant to measure. Such options can work if you use the tool consistently.
Yet often analysts need to combine data found in one platform with other data or in calculated metrics. This will change their visualization options. They do not lack options, although the evolution of data has increased the number of visualization options for displaying results and incorporating real-time data.
All this makes choosing the right scene even more complicated.
Related article: How data visualization tools are making self-service analytics easier
To begin your visualization selection, ask Munzner questions
So where do you start choosing a good graph?
In my visualization post, What Makes a Good Data Visualization I mentioned two aspects of data to consider. You want a graph that conveys ideas from data that are too complex to interpret through word and that helps your audience quickly parse the information and act on the results.
To access that graph, ask a set of questions created by Tamara Munzner, a professor of computer studies at the University of British Columbia. Munjner is known for his extensive research into the development, evaluation and characterization of visualization systems and techniques. He highlighted this question framework in his presentation on Avoiding Visualization Analysis.
- Who are the end users? (It is the audience that needs the information.)
- What is being shown?
- Why is the user viewing this? (Questions 2 and 3 are meant to shed light on what the data is, how it is organized and its source.)
- How is this being shown? (That’s the important question – what type of graph best illustrates the data.)
Munzner’s answers to the questions help narrow down which graphs best represent the answers. Your graph choice should achieve one of the following objectives:
- To analyze distribution structure or change.
- To identify patterns or trends.
- Manifesting objective 1 and/or objective 2 within a subset of a given dataset.
Select a graph that displays a hierarchy in the data
Four categories of graphs are suitable for displaying hierarchies in data: structure, distribution, relationship, and comparison. Composition and distribution graphs both address the structure of your given dimension or metric as it relates through observations, while relationship and comparison graphs are meant to highlight contrasting differences through patterns and trends.
Composition graphs are meant to describe the makeup of a set of observations. Visualizations in this category include pie charts, treemaps, and stacked bar charts.
Distribution graphs represent a range of observations, making them perfect for data showing the quality of the dimensions and metrics containing those observations. Examples such as histograms or boxplots are chosen to address the statistical range.
Relationship graphs are about correlation trends between two or more dimensions or metrics. Scatterplots and bubble charts are good examples of this.
Comparison graphs are meant to highlight differences between two or more dimensions or metrics with respect to divergence, trends, or rankings. These are often a special variety of relation or composition graphs, such as regression charts, Pareto charts, bump charts and stacked column charts.
The best graph for your purpose is to organize the data to answer the question “why is the user viewing it?”
Each of these categories has more than one graph style that can be covered in one post. But when choosing a graph, you are looking for a graph that displays a hierarchy that clearly and accurately answers your questions.
RELATED ARTICLE: Stop Torturing Your Data & Other Tips to Reveal True Data Insights
Learn how your data and colors convey information
The success of a graph depends partly on whether it creates a cognitive load for the audience. Cognitive load refers to the amount of information a brain can process at any given time. That’s why you want to make sure that the graphical elements combine to tell a clear story with the least amount of effort on the part of the audience.
For example, bar charts and pie charts can show the structure of data similarly, but bar charts are better at displaying unit differences. Those differences are important to show the accuracy of the comparison. For example, instead of saying that organic traffic has increased by 20%, you need a bar chart that shows that there has been a 20% increase. With just a glance, your end user can easily understand the change.
In the chart below you can clearly see that there were few vehicles with rear-wheel drive (r) as compared to all wheel drive (4) or front wheel drive vehicles (f).
A good visualization focuses on accuracy when indicating measurements. Heat maps can show gradient changes, but can be a poor choice for accuracy when the viewer wants to understand the individual numerical differences between elements. For example, if a temperature change of a degree or two is important to your subject, you may want to choose a graph that highlights when that difference appears.
Color is another element to consider. Sticking to a single color and using colors to indicate visual distinctions reduces cognitive load. Also consider accessibility concerns, such as color-blind users, when selecting your color scheme. A second color is acceptable for highlighting a specific dimension, so it stands out against other dimensions in the bar graph. The two colors are perfect for graphs that show two different extremes, such as a heatmap. You often see this in correlation charts, as shown below, for observations to indicate the strength of the correlation.
But there are limits to how many colors can be assigned to some composition graphs. Usually six to eight colors is a good ballpark for showing a meaningful difference in many dimensions or metrics. Any more than that introduces a lot of granularity. The resulting visualization gathers the graph visuals together and makes the distinctions hard to see.
If you need to show more than eight different dimensions with different colors, a treemap is a better option. A treemap is a diagram of nested rectangles displayed as a hierarchy according to the value of the given data. The area of each rectangle corresponds to the numerical value of its data. Shapes make the scale of each datapoint clear to see, color scales provide further contrast, all within a constrained display space.
In addition, advanced visualization platforms such as Tableau and Google Data Studio have options to query data subsets from data sources. It gives you additional color and visual options to tell your data story.
Related article: How to use Google Analytics treemap reports effectively
Choose visualizations that fit your timeline or location
The next visualization option deals with demonstrating how the data evolves over time. Relationship graphs usually work well, as do line charts that can show comparisons over time, or regression charts to chart data changes over a set period. But despite a slowly developing, trend, you may have to display a longer time frame to show a significant one.
This is where programming languages like R and Python can help. Libraries – scripts added for functionality – provide visualization options so that users can annotate graphs and create animations that demonstrate how data changes over time. Often the data is read into a program, then mapped to a visual graph using a library. Python users have a choice of libraries, such as Matplotlib and Seaborn, while R users have access to ggplot2, which provides the graphics concept of adding or removing each graph element as a layer, to provide customization options. There is a library based on the grammar of .
The advantage of these libraries is that you can create custom visuals to suit your needs using scripts that call real-time data via APIs. This allows the graph to stay up-to-date with the latest information.
These are also useful for spatial visualization such as geolocation graphs. The data is mapped to a place of interest, adding another view to display information. Libraries for both Python and R provide options for scene maps and graph combinations.
Ask how often graph updates are necessary
Does the graph need to be updated on a regular basis to monitor the ongoing performance or is it needed for a one-time analysis? The answer determines what type of workflow works best.
Real-time graphs are usually combined with cloud based dashboards to manage data and visuals. For example, in R programming, you can easily build a Shiny app, a simple web application that allows data, program results, and graphs to be displayed in a shared digital environment. A Shiny app can be hosted as a dashboard that quickly updates the visualization when the data is called. In addition, you can also add HTML features like buttons and sliders to allow your audience to adjust the display without touching the data or the underlying code.
Ultimately you should outline the reporting schedule that best addresses your audience from the data. Doing so will highlight the steps you need to take to distribute your graphs and see what impacts the decisions. Sometimes there are technical reasons for adjusting the timeline. If the graph is for printed material then many times people prefer a still image or are limited to a single image. Mapping the raw data to the visual raises the question of which data sources need access to feed the graph. If it is updated regularly, you need a simple means to update the data and associated annotations.
Related Articles: What is Tableau? How BI drives growth
Some Final Tips for Choosing Good Visualizations
Ultimately a good visualization selection will make your analysis clear. As I mentioned in 10 Mistakes to Avoid When Rethinking Your Analytics Strategy, you want to avoid broad questions that spiral into one long dull story about your data. This does not lead to any meaningful conclusions about your marketing efforts.
If you have a lot of important material but know stakeholders don’t have a lot of time, you can put those scenes in an appendix so recipients can review the details when convenient. You can find some more rules of thumb in my visualization post.
Choosing good visuals for storytelling keeps your marketing analysis in focus. A strong visualization will open up takeaway discussions in your audience that drive your customer experiences — and your organization’s — forward.
Pierre DeBois is the founder of Zimana, a small business digital analytics consultancy. He reviews data from web analytics and social media dashboard solutions, then provides recommendations and web development actions that improve marketing strategy and business profitability.